Apply consistent code formatting across the repo. Add clang-format and pre-commit hooks.

.clang-format

@@ -0,0 +1,49 @@
+---
+AccessModifierOffset: -4
+AlignAfterOpenBracket: Align
+AlignConsecutiveAssignments: Consecutive
+AlignConsecutiveDeclarations: Consecutive
+AlignConsecutiveMacros: Consecutive
+AlignEscapedNewlines: Left
+AlignOperands: AlignAfterOperator
+AlignTrailingComments: true
+AllowAllParametersOfDeclarationOnNextLine: false
+BinPackArguments: false
+BinPackParameters: false
+BraceWrapping:
+    AfterClass: true
+    AfterControlStatement: false
+    AfterExternBlock: true
+    AfterFunction: true
+    AfterStruct: true
+    AfterUnion: true
+    BeforeCatch: true
+    BeforeElse: true
+    IndentBraces: false
+BreakBeforeBraces: Custom
+BreakBeforeConceptDeclarations: true
+BreakBeforeBinaryOperators: NonAssignment
+BreakBeforeTernaryOperators: true
+BreakConstructorInitializers: BeforeComma
+BreakInheritanceList: BeforeComma
+ColumnLimit: 120
+DerivePointerAlignment: false
+FixNamespaceComments: true
+IncludeCategories:
+  - Regex:           '^<.*>'
+    Priority:        1
+  - Regex:           '^".*"'
+    Priority:        2
+SortIncludes: true
+IncludeBlocks: Regroup
+IndentWidth: 4
+MaxEmptyLinesToKeep: 2
+PointerAlignment: Right
+SortUsingDeclarations: true
+SpaceAfterCStyleCast: false
+SpaceBeforeAssignmentOperators: true
+SpaceBeforeParens: ControlStatements
+Standard: c++17
+TabWidth: 4
+UseTab: Never
+...

.pre-commit-config.yaml

@@ -0,0 +1,100 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+ci:
+    autofix_commit_msg: |
+      [pre-commit.ci] auto code formatting
+    autofix_prs: false
+    autoupdate_branch: ''
+    autoupdate_commit_msg: '[pre-commit.ci] pre-commit autoupdate'
+    autoupdate_schedule: quarterly
+    skip: []
+    submodules: false
+
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v5.0.0
+    hooks:
+      - id: end-of-file-fixer
+        exclude: |
+          (?x)^(
+            .*\.raw$|
+            .*\.bin$|
+            .*\.dat$|
+            .*\.nv12$|
+            data/.*|
+            Common/.*
+          )
+        files: |
+          (?x)^(
+            .*\.txt$|
+            .*\.md$|
+            .*\.cpp$|
+            .*\.cxx$|
+            .*\.hpp$|
+            .*\.h$|
+            .*\.cu$|
+            .*\.cuh$
+          )
+      - id: mixed-line-ending
+        exclude: |
+          (?x)^(
+            .*\.raw$|
+            .*\.bin$|
+            .*\.dat$|
+            .*\.nv12$|
+            data/.*|
+            Common/.*
+          )
+        files: |
+          (?x)^(
+            .*\.txt$|
+            .*\.md$|
+            .*\.cpp$|
+            .*\.cxx$|
+            .*\.hpp$|
+            .*\.h$|
+            .*\.cu$|
+            .*\.cuh$
+          )
+      - id: trailing-whitespace
+        exclude: |
+          (?x)^(
+            .*\.raw$|
+            .*\.bin$|
+            .*\.dat$|
+            .*\.nv12$|
+            data/.*|
+            Common/.*
+          )
+        files: |
+          (?x)^(
+            .*\.txt$|
+            .*\.md$|
+            .*\.cpp$|
+            .*\.cxx$|
+            .*\.hpp$|
+            .*\.h$|
+            .*\.cu$|
+            .*\.cuh$
+          )
+  - repo: https://github.com/pre-commit/mirrors-clang-format
+    rev: v19.1.6
+    hooks:
+      - id: clang-format
+        types_or: [file]
+        files: |
+          (?x)^(
+            ^.*\.c$|
+            ^.*\.cpp$|
+            ^.*\.cu$|
+            ^.*\.cuh$|
+            ^.*\.cxx$|
+            ^.*\.h$|
+            ^.*\.hpp$|
+            ^.*\.inl$|
+            ^.*\.mm$
+          )
+        exclude: |
+          (?x)^(
+            Common/.*
+          )
+        args: ["-fallback-style=none", "-style=file", "-i"]

Samples/0_Introduction/UnifiedMemoryStreams/UnifiedMemoryStreams.cu

@@ -31,10 +31,10 @@
  */
 
 // system includes
+#include <algorithm>
 #include <cstdio>
 #include <ctime>
 #include <vector>
-#include <algorithm>
 #ifdef USE_PTHREADS
 #include <pthread.h>
 #else
@@ -58,15 +58,25 @@ double drand48() { return double(rand()) / RAND_MAX; }
 const char *sSDKname = "UnifiedMemoryStreams";
 
 // simple task
-template <typename T>
+template <typename T> struct Task
-struct Task {
+{
     unsigned int size, id;
     T           *data;
     T           *result;
     T           *vector;
 
-  Task() : size(0), id(0), data(NULL), result(NULL), vector(NULL){};
+    Task()
-  Task(unsigned int s) : size(s), id(0), data(NULL), result(NULL) {
+        : size(0)
+        , id(0)
+        , data(NULL)
+        , result(NULL)
+        , vector(NULL) {};
+    Task(unsigned int s)
+        : size(s)
+        , id(0)
+        , data(NULL)
+        , result(NULL)
+    {
         // allocate unified memory -- the operation performed in this example will
         // be a DGEMV
         checkCudaErrors(cudaMallocManaged(&data, sizeof(T) * size * size));
@@ -75,7 +85,8 @@ struct Task {
         checkCudaErrors(cudaDeviceSynchronize());
     }
 
-  ~Task() {
+    ~Task()
+    {
         // ensure all memory is deallocated
         checkCudaErrors(cudaDeviceSynchronize());
         checkCudaErrors(cudaFree(data));
@@ -83,7 +94,8 @@ struct Task {
         checkCudaErrors(cudaFree(vector));
     }
 
-  void allocate(const unsigned int s, const unsigned int unique_id) {
+    void allocate(const unsigned int s, const unsigned int unique_id)
+    {
         // allocate unified memory outside of constructor
         id   = unique_id;
         size = s;
@@ -105,7 +117,8 @@ struct Task {
 };
 
 #ifdef USE_PTHREADS
-struct threadData_t {
+struct threadData_t
+{
     int             tid;
     Task<double>   *TaskListPtr;
     cudaStream_t   *streams;
@@ -117,8 +130,8 @@ typedef struct threadData_t threadData;
 #endif
 
 // simple host dgemv: assume data is in row-major format and square
-template <typename T>
+template <typename T> void gemv(int m, int n, T alpha, T *A, T *x, T beta, T *result)
-void gemv(int m, int n, T alpha, T *A, T *x, T beta, T *result) {
+{
     // rows
     for (int i = 0; i < n; i++) {
         result[i] *= beta;
@@ -131,7 +144,8 @@ void gemv(int m, int n, T alpha, T *A, T *x, T beta, T *result) {
 
 // execute a single task on either host or device depending on size
 #ifdef USE_PTHREADS
-void *execute(void *inpArgs) {
+void *execute(void *inpArgs)
+{
     threadData     *dataPtr = (threadData *)inpArgs;
     cudaStream_t   *stream  = dataPtr->streams;
     cublasHandle_t *handle  = dataPtr->handles;
@@ -142,92 +156,75 @@ void *execute(void *inpArgs) {
 
         if (t.size < 100) {
             // perform on host
-      printf("Task [%d], thread [%d] executing on host (%d)\n", t.id, tid,
+            printf("Task [%d], thread [%d] executing on host (%d)\n", t.id, tid, t.size);
-             t.size);
 
             // attach managed memory to a (dummy) stream to allow host access while
             // the device is running
-      checkCudaErrors(
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.data, 0, cudaMemAttachHost));
-          cudaStreamAttachMemAsync(stream[0], t.data, 0, cudaMemAttachHost));
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.vector, 0, cudaMemAttachHost));
-      checkCudaErrors(
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.result, 0, cudaMemAttachHost));
-          cudaStreamAttachMemAsync(stream[0], t.vector, 0, cudaMemAttachHost));
-      checkCudaErrors(
-          cudaStreamAttachMemAsync(stream[0], t.result, 0, cudaMemAttachHost));
             // necessary to ensure Async cudaStreamAttachMemAsync calls have finished
             checkCudaErrors(cudaStreamSynchronize(stream[0]));
             // call the host operation
             gemv(t.size, t.size, 1.0, t.data, t.vector, 0.0, t.result);
-    } else {
+        }
+        else {
             // perform on device
-      printf("Task [%d], thread [%d] executing on device (%d)\n", t.id, tid,
+            printf("Task [%d], thread [%d] executing on device (%d)\n", t.id, tid, t.size);
-             t.size);
             double one  = 1.0;
             double zero = 0.0;
 
             // attach managed memory to my stream
             checkCudaErrors(cublasSetStream(handle[tid + 1], stream[tid + 1]));
-      checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.data, 0,
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.data, 0, cudaMemAttachSingle));
-                                               cudaMemAttachSingle));
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.vector, 0, cudaMemAttachSingle));
-      checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.vector, 0,
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.result, 0, cudaMemAttachSingle));
-                                               cudaMemAttachSingle));
-      checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.result, 0,
-                                               cudaMemAttachSingle));
             // call the device operation
-      checkCudaErrors(cublasDgemv(handle[tid + 1], CUBLAS_OP_N, t.size, t.size,
+            checkCudaErrors(cublasDgemv(
-                                  &one, t.data, t.size, t.vector, 1, &zero,
+                handle[tid + 1], CUBLAS_OP_N, t.size, t.size, &one, t.data, t.size, t.vector, 1, &zero, t.result, 1));
-                                  t.result, 1));
         }
     }
 
     pthread_exit(NULL);
 }
 #else
-template <typename T>
+template <typename T> void execute(Task<T> &t, cublasHandle_t *handle, cudaStream_t *stream, int tid)
-void execute(Task<T> &t, cublasHandle_t *handle, cudaStream_t *stream,
+{
-             int tid) {
     if (t.size < 100) {
         // perform on host
-    printf("Task [%d], thread [%d] executing on host (%d)\n", t.id, tid,
+        printf("Task [%d], thread [%d] executing on host (%d)\n", t.id, tid, t.size);
-           t.size);
 
         // attach managed memory to a (dummy) stream to allow host access while the
         // device is running
-    checkCudaErrors(
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.data, 0, cudaMemAttachHost));
-        cudaStreamAttachMemAsync(stream[0], t.data, 0, cudaMemAttachHost));
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.vector, 0, cudaMemAttachHost));
-    checkCudaErrors(
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.result, 0, cudaMemAttachHost));
-        cudaStreamAttachMemAsync(stream[0], t.vector, 0, cudaMemAttachHost));
-    checkCudaErrors(
-        cudaStreamAttachMemAsync(stream[0], t.result, 0, cudaMemAttachHost));
         // necessary to ensure Async cudaStreamAttachMemAsync calls have finished
         checkCudaErrors(cudaStreamSynchronize(stream[0]));
         // call the host operation
         gemv(t.size, t.size, 1.0, t.data, t.vector, 0.0, t.result);
-  } else {
+    }
+    else {
         // perform on device
-    printf("Task [%d], thread [%d] executing on device (%d)\n", t.id, tid,
+        printf("Task [%d], thread [%d] executing on device (%d)\n", t.id, tid, t.size);
-           t.size);
         double one  = 1.0;
         double zero = 0.0;
 
         // attach managed memory to my stream
         checkCudaErrors(cublasSetStream(handle[tid + 1], stream[tid + 1]));
-    checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.data, 0,
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.data, 0, cudaMemAttachSingle));
-                                             cudaMemAttachSingle));
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.vector, 0, cudaMemAttachSingle));
-    checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.vector, 0,
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.result, 0, cudaMemAttachSingle));
-                                             cudaMemAttachSingle));
-    checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.result, 0,
-                                             cudaMemAttachSingle));
         // call the device operation
-    checkCudaErrors(cublasDgemv(handle[tid + 1], CUBLAS_OP_N, t.size, t.size,
+        checkCudaErrors(cublasDgemv(
-                                &one, t.data, t.size, t.vector, 1, &zero,
+            handle[tid + 1], CUBLAS_OP_N, t.size, t.size, &one, t.data, t.size, t.vector, 1, &zero, t.result, 1));
-                                t.result, 1));
     }
 }
 #endif
 
 // populate a list of tasks with random sizes
-template <typename T>
+template <typename T> void initialise_tasks(std::vector<Task<T>> &TaskList)
-void initialise_tasks(std::vector<Task<T> > &TaskList) {
+{
     for (unsigned int i = 0; i < TaskList.size(); i++) {
         // generate random size
         int size;
@@ -236,7 +233,8 @@ void initialise_tasks(std::vector<Task<T> > &TaskList) {
     }
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     // set device
     cudaDeviceProp device_prop;
     int            dev_id = findCudaDevice(argc, (const char **)argv);
@@ -294,19 +292,17 @@ int main(int argc, char **argv) {
 
         if ((TaskList.size() / nthreads) == 0) {
             InputToThreads[i].taskSize    = (TaskList.size() / nthreads);
-      InputToThreads[i].TaskListPtr =
+            InputToThreads[i].TaskListPtr = &TaskList[i * (TaskList.size() / nthreads)];
-          &TaskList[i * (TaskList.size() / nthreads)];
+        }
-    } else {
+        else {
             if (i == nthreads - 1) {
-        InputToThreads[i].taskSize =
+                InputToThreads[i].taskSize = (TaskList.size() / nthreads) + (TaskList.size() % nthreads);
-            (TaskList.size() / nthreads) + (TaskList.size() % nthreads);
                 InputToThreads[i].TaskListPtr =
-            &TaskList[i * (TaskList.size() / nthreads) +
+                    &TaskList[i * (TaskList.size() / nthreads) + (TaskList.size() % nthreads)];
-                      (TaskList.size() % nthreads)];
+            }
-      } else {
+            else {
                 InputToThreads[i].taskSize    = (TaskList.size() / nthreads);
-        InputToThreads[i].TaskListPtr =
+                InputToThreads[i].TaskListPtr = &TaskList[i * (TaskList.size() / nthreads)];
-            &TaskList[i * (TaskList.size() / nthreads)];
             }
         }
 

Samples/0_Introduction/asyncAPI/asyncAPI.cu

@@ -38,19 +38,21 @@
 #include <stdio.h>
 
 // includes CUDA Runtime
-#include <cuda_runtime.h>
 #include <cuda_profiler_api.h>
+#include <cuda_runtime.h>
 
 // includes, project
 #include <helper_cuda.h>
 #include <helper_functions.h> // helper utility functions
 
-__global__ void increment_kernel(int *g_data, int inc_value) {
+__global__ void increment_kernel(int *g_data, int inc_value)
+{
     int idx     = blockIdx.x * blockDim.x + threadIdx.x;
     g_data[idx] = g_data[idx] + inc_value;
 }
 
-bool correct_output(int *data, const int n, const int x) {
+bool correct_output(int *data, const int n, const int x)
+{
     for (int i = 0; i < n; i++)
         if (data[i] != x) {
             printf("Error! data[%d] = %d, ref = %d\n", i, data[i], x);
@@ -60,7 +62,8 @@ bool correct_output(int *data, const int n, const int x) {
     return true;
 }
 
-int main(int argc, char *argv[]) {
+int main(int argc, char *argv[])
+{
     int            devID;
     cudaDeviceProp deviceProps;
 
@@ -126,8 +129,7 @@ int main(int argc, char *argv[]) {
     // print the cpu and gpu times
     printf("time spent executing by the GPU: %.2f\n", gpu_time);
     printf("time spent by CPU in CUDA calls: %.2f\n", sdkGetTimerValue(&timer));
-  printf("CPU executed %lu iterations while waiting for GPU to finish\n",
+    printf("CPU executed %lu iterations while waiting for GPU to finish\n", counter);
-         counter);
 
     // check the output for correctness
     bool bFinalResults = correct_output(a, n, value);

Samples/0_Introduction/clock/clock.cu

@@ -48,15 +48,16 @@
 // This kernel computes a standard parallel reduction and evaluates the
 // time it takes to do that for each block. The timing results are stored
 // in device memory.
-__global__ static void timedReduction(const float *input, float *output,
+__global__ static void timedReduction(const float *input, float *output, clock_t *timer)
-                                      clock_t *timer) {
+{
     // __shared__ float shared[2 * blockDim.x];
     extern __shared__ float shared[];
 
     const int tid = threadIdx.x;
     const int bid = blockIdx.x;
 
-  if (tid == 0) timer[bid] = clock();
+    if (tid == 0)
+        timer[bid] = clock();
 
     // Copy input.
     shared[tid]              = input[tid];
@@ -77,11 +78,13 @@ __global__ static void timedReduction(const float *input, float *output,
     }
 
     // Write result.
-  if (tid == 0) output[bid] = shared[0];
+    if (tid == 0)
+        output[bid] = shared[0];
 
     __syncthreads();
 
-  if (tid == 0) timer[bid + gridDim.x] = clock();
+    if (tid == 0)
+        timer[bid + gridDim.x] = clock();
 }
 
 #define NUM_BLOCKS  64
@@ -104,7 +107,8 @@ __global__ static void timedReduction(const float *input, float *output,
 // the memory. With more than 32 the speed scales linearly.
 
 // Start the main CUDA Sample here
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("CUDA Clock sample\n");
 
     // This will pick the best possible CUDA capable device
@@ -121,20 +125,15 @@ int main(int argc, char **argv) {
         input[i] = (float)i;
     }
 
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc((void **)&dinput, sizeof(float) * NUM_THREADS * 2));
-      cudaMalloc((void **)&dinput, sizeof(float) * NUM_THREADS * 2));
     checkCudaErrors(cudaMalloc((void **)&doutput, sizeof(float) * NUM_BLOCKS));
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc((void **)&dtimer, sizeof(clock_t) * NUM_BLOCKS * 2));
-      cudaMalloc((void **)&dtimer, sizeof(clock_t) * NUM_BLOCKS * 2));
 
-  checkCudaErrors(cudaMemcpy(dinput, input, sizeof(float) * NUM_THREADS * 2,
+    checkCudaErrors(cudaMemcpy(dinput, input, sizeof(float) * NUM_THREADS * 2, cudaMemcpyHostToDevice));
-                             cudaMemcpyHostToDevice));
 
-  timedReduction<<<NUM_BLOCKS, NUM_THREADS, sizeof(float) * 2 * NUM_THREADS>>>(
+    timedReduction<<<NUM_BLOCKS, NUM_THREADS, sizeof(float) * 2 * NUM_THREADS>>>(dinput, doutput, dtimer);
-      dinput, doutput, dtimer);
 
-  checkCudaErrors(cudaMemcpy(timer, dtimer, sizeof(clock_t) * NUM_BLOCKS * 2,
+    checkCudaErrors(cudaMemcpy(timer, dtimer, sizeof(clock_t) * NUM_BLOCKS * 2, cudaMemcpyDeviceToHost));
-                             cudaMemcpyDeviceToHost));
 
     checkCudaErrors(cudaFree(dinput));
     checkCudaErrors(cudaFree(doutput));

Samples/0_Introduction/clock_nvrtc/clock.cpp

@@ -34,12 +34,11 @@
  */
 
 // System includes
-#include <stdio.h>
-#include <stdint.h>
 #include <assert.h>
-
 #include <cuda_runtime.h>
 #include <nvrtc_helper.h>
+#include <stdint.h>
+#include <stdio.h>
 
 // helper functions and utilities to work with CUDA
 #include <helper_functions.h>
@@ -71,7 +70,8 @@
 
 // Start the main CUDA Sample here
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("CUDA Clock sample\n");
 
     typedef long clock_t;
@@ -106,17 +106,20 @@ int main(int argc, char **argv) {
 
     void *arr[] = {(void *)&dinput, (void *)&doutput, (void *)&dtimer};
 
-  checkCudaErrors(cuLaunchKernel(
+    checkCudaErrors(cuLaunchKernel(kernel_addr,
-      kernel_addr, cudaGridSize.x, cudaGridSize.y,
+                                   cudaGridSize.x,
+                                   cudaGridSize.y,
                                    cudaGridSize.z, /* grid dim */
-      cudaBlockSize.x, cudaBlockSize.y, cudaBlockSize.z, /* block dim */
+                                   cudaBlockSize.x,
-      sizeof(float) * 2 * NUM_THREADS, 0, /* shared mem, stream */
+                                   cudaBlockSize.y,
+                                   cudaBlockSize.z, /* block dim */
+                                   sizeof(float) * 2 * NUM_THREADS,
+                                   0,       /* shared mem, stream */
                                    &arr[0], /* arguments */
                                    0));
 
     checkCudaErrors(cuCtxSynchronize());
-  checkCudaErrors(
+    checkCudaErrors(cuMemcpyDtoH(timer, dtimer, sizeof(clock_t) * NUM_BLOCKS * 2));
-      cuMemcpyDtoH(timer, dtimer, sizeof(clock_t) * NUM_BLOCKS * 2));
     checkCudaErrors(cuMemFree(dinput));
     checkCudaErrors(cuMemFree(doutput));
     checkCudaErrors(cuMemFree(dtimer));

Samples/0_Introduction/clock_nvrtc/clock_kernel.cu

@@ -37,15 +37,16 @@
 // time it takes to do that for each block. The timing results are stored
 // in device memory.
 
-extern "C" __global__ void timedReduction(const float *input, float *output,
+extern "C" __global__ void timedReduction(const float *input, float *output, clock_t *timer)
-                                          clock_t *timer) {
+{
     // __shared__ float shared[2 * blockDim.x];
     extern __shared__ float shared[];
 
     const int tid = threadIdx.x;
     const int bid = blockIdx.x;
 
-  if (tid == 0) timer[bid] = clock();
+    if (tid == 0)
+        timer[bid] = clock();
 
     // Copy input.
     shared[tid]              = input[tid];
@@ -66,9 +67,11 @@ extern "C" __global__ void timedReduction(const float *input, float *output,
     }
 
     // Write result.
-  if (tid == 0) output[bid] = shared[0];
+    if (tid == 0)
+        output[bid] = shared[0];
 
     __syncthreads();
 
-  if (tid == 0) timer[bid + gridDim.x] = clock();
+    if (tid == 0)
+        timer[bid + gridDim.x] = clock();
 }

Samples/0_Introduction/cudaOpenMP/cudaOpenMP.cu

@@ -37,20 +37,24 @@
 using namespace std;
 
 // a simple kernel that simply increments each array element by b
-__global__ void kernelAddConstant(int *g_a, const int b) {
+__global__ void kernelAddConstant(int *g_a, const int b)
+{
     int idx = blockIdx.x * blockDim.x + threadIdx.x;
     g_a[idx] += b;
 }
 
 // a predicate that checks whether each array element is set to its index plus b
-int correctResult(int *data, const int n, const int b) {
+int correctResult(int *data, const int n, const int b)
+{
     for (int i = 0; i < n; i++)
-    if (data[i] != i + b) return 0;
+        if (data[i] != i + b)
+            return 0;
 
     return 1;
 }
 
-int main(int argc, char *argv[]) {
+int main(int argc, char *argv[])
+{
     int num_gpus = 0; // number of CUDA GPUs
 
     printf("%s Starting...\n\n", argv[0]);
@@ -93,7 +97,8 @@ int main(int argc, char *argv[]) {
         return 1;
     }
 
-  for (unsigned int i = 0; i < n; i++) a[i] = i;
+    for (unsigned int i = 0; i < n; i++)
+        a[i] = i;
 
     ////////////////////////////////////////////////////////////////
     // run as many CPU threads as there are CUDA devices
@@ -105,8 +110,7 @@ int main(int argc, char *argv[]) {
     //   Recall that all variables declared inside an "omp parallel" scope are
     //   local to each CPU thread
     //
-  omp_set_num_threads(
+    omp_set_num_threads(num_gpus); // create as many CPU threads as there are CUDA devices
-      num_gpus);  // create as many CPU threads as there are CUDA devices
 // omp_set_num_threads(2*num_gpus);// create twice as many CPU threads as there
 // are CUDA devices
 #pragma omp parallel
@@ -116,31 +120,23 @@ int main(int argc, char *argv[]) {
 
         // set and check the CUDA device for this CPU thread
         int gpu_id = -1;
-    checkCudaErrors(cudaSetDevice(
+        checkCudaErrors(
-        cpu_thread_id %
+            cudaSetDevice(cpu_thread_id % num_gpus)); // "% num_gpus" allows more CPU threads than GPU devices
-        num_gpus));  // "% num_gpus" allows more CPU threads than GPU devices
         checkCudaErrors(cudaGetDevice(&gpu_id));
-    printf("CPU thread %d (of %d) uses CUDA device %d\n", cpu_thread_id,
+        printf("CPU thread %d (of %d) uses CUDA device %d\n", cpu_thread_id, num_cpu_threads, gpu_id);
-           num_cpu_threads, gpu_id);
+
-
+        int         *d_a   = 0; // pointer to memory on the device associated with this CPU thread
-    int *d_a =
+        int         *sub_a = a + cpu_thread_id * n / num_cpu_threads; // pointer to this CPU thread's portion of data
-        0;  // pointer to memory on the device associated with this CPU thread
-    int *sub_a =
-        a +
-        cpu_thread_id * n /
-            num_cpu_threads;  // pointer to this CPU thread's portion of data
         unsigned int nbytes_per_kernel = nbytes / num_cpu_threads;
         dim3         gpu_threads(128); // 128 threads per block
         dim3         gpu_blocks(n / (gpu_threads.x * num_cpu_threads));
 
         checkCudaErrors(cudaMalloc((void **)&d_a, nbytes_per_kernel));
         checkCudaErrors(cudaMemset(d_a, 0, nbytes_per_kernel));
-    checkCudaErrors(
+        checkCudaErrors(cudaMemcpy(d_a, sub_a, nbytes_per_kernel, cudaMemcpyHostToDevice));
-        cudaMemcpy(d_a, sub_a, nbytes_per_kernel, cudaMemcpyHostToDevice));
         kernelAddConstant<<<gpu_blocks, gpu_threads>>>(d_a, b);
 
-    checkCudaErrors(
+        checkCudaErrors(cudaMemcpy(sub_a, d_a, nbytes_per_kernel, cudaMemcpyDeviceToHost));
-        cudaMemcpy(sub_a, d_a, nbytes_per_kernel, cudaMemcpyDeviceToHost));
         checkCudaErrors(cudaFree(d_a));
     }
     printf("---------------------------\n");
@@ -153,7 +149,8 @@ int main(int argc, char *argv[]) {
     //
     bool bResult = correctResult(a, n, b);
 
-  if (a) free(a);  // free CPU memory
+    if (a)
+        free(a); // free CPU memory
 
     exit(bResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }

Samples/0_Introduction/fp16ScalarProduct/fp16ScalarProduct.cu

@@ -25,17 +25,18 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include "cuda_fp16.h"
-#include "helper_cuda.h"
-
 #include <cstdio>
 #include <cstdlib>
 #include <ctime>
 
+#include "cuda_fp16.h"
+#include "helper_cuda.h"
+
 #define NUM_OF_BLOCKS  128
 #define NUM_OF_THREADS 128
 
-__forceinline__ __device__ void reduceInShared_intrinsics(half2 *const v) {
+__forceinline__ __device__ void reduceInShared_intrinsics(half2 *const v)
+{
     if (threadIdx.x < 64)
         v[threadIdx.x] = __hadd2(v[threadIdx.x], v[threadIdx.x + 64]);
     __syncthreads();
@@ -59,27 +60,34 @@ __forceinline__ __device__ void reduceInShared_intrinsics(half2 *const v) {
     __syncthreads();
 }
 
-__forceinline__ __device__ void reduceInShared_native(half2 *const v) {
+__forceinline__ __device__ void reduceInShared_native(half2 *const v)
-  if (threadIdx.x < 64) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 64];
+{
+    if (threadIdx.x < 64)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 64];
     __syncthreads();
-  if (threadIdx.x < 32) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 32];
+    if (threadIdx.x < 32)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 32];
     __syncthreads();
-  if (threadIdx.x < 16) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 16];
+    if (threadIdx.x < 16)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 16];
     __syncthreads();
-  if (threadIdx.x < 8) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 8];
+    if (threadIdx.x < 8)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 8];
     __syncthreads();
-  if (threadIdx.x < 4) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 4];
+    if (threadIdx.x < 4)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 4];
     __syncthreads();
-  if (threadIdx.x < 2) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 2];
+    if (threadIdx.x < 2)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 2];
     __syncthreads();
-  if (threadIdx.x < 1) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 1];
+    if (threadIdx.x < 1)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 1];
     __syncthreads();
 }
 
-__global__ void scalarProductKernel_intrinsics(half2 const *const a,
+__global__ void
-                                               half2 const *const b,
+scalarProductKernel_intrinsics(half2 const *const a, half2 const *const b, float *const results, size_t const size)
-                                               float *const results,
+{
-                                               size_t const size) {
     const int        stride = gridDim.x * blockDim.x;
     __shared__ half2 shArray[NUM_OF_THREADS];
 
@@ -101,10 +109,9 @@ __global__ void scalarProductKernel_intrinsics(half2 const *const a,
     }
 }
 
-__global__ void scalarProductKernel_native(half2 const *const a,
+__global__ void
-                                           half2 const *const b,
+scalarProductKernel_native(half2 const *const a, half2 const *const b, float *const results, size_t const size)
-                                           float *const results,
+{
-                                           size_t const size) {
     const int        stride = gridDim.x * blockDim.x;
     __shared__ half2 shArray[NUM_OF_THREADS];
 
@@ -126,7 +133,8 @@ __global__ void scalarProductKernel_native(half2 const *const a,
     }
 }
 
-void generateInput(half2 *a, size_t size) {
+void generateInput(half2 *a, size_t size)
+{
     for (size_t i = 0; i < size; ++i) {
         half2 temp;
         temp.x = static_cast<float>(rand() % 4);
@@ -135,7 +143,8 @@ void generateInput(half2 *a, size_t size) {
     }
 }
 
-int main(int argc, char *argv[]) {
+int main(int argc, char *argv[])
+{
     srand((unsigned int)time(NULL));
     size_t size = NUM_OF_BLOCKS * NUM_OF_THREADS * 16;
 
@@ -151,8 +160,7 @@ int main(int argc, char *argv[]) {
     checkCudaErrors(cudaGetDeviceProperties(&devProp, devID));
 
     if (devProp.major < 5 || (devProp.major == 5 && devProp.minor < 3)) {
-    printf(
+        printf("ERROR: fp16ScalarProduct requires GPU devices with compute SM 5.3 or "
-        "ERROR: fp16ScalarProduct requires GPU devices with compute SM 5.3 or "
                "higher.\n");
         return EXIT_WAIVED;
     }
@@ -162,23 +170,17 @@ int main(int argc, char *argv[]) {
         checkCudaErrors(cudaMalloc((void **)&devVec[i], size * sizeof *devVec[i]));
     }
 
-  checkCudaErrors(
+    checkCudaErrors(cudaMallocHost((void **)&results, NUM_OF_BLOCKS * sizeof *results));
-      cudaMallocHost((void **)&results, NUM_OF_BLOCKS * sizeof *results));
+    checkCudaErrors(cudaMalloc((void **)&devResults, NUM_OF_BLOCKS * sizeof *devResults));
-  checkCudaErrors(
-      cudaMalloc((void **)&devResults, NUM_OF_BLOCKS * sizeof *devResults));
 
     for (int i = 0; i < 2; ++i) {
         generateInput(vec[i], size);
-    checkCudaErrors(cudaMemcpy(devVec[i], vec[i], size * sizeof *vec[i],
+        checkCudaErrors(cudaMemcpy(devVec[i], vec[i], size * sizeof *vec[i], cudaMemcpyHostToDevice));
-                               cudaMemcpyHostToDevice));
     }
 
-  scalarProductKernel_native<<<NUM_OF_BLOCKS, NUM_OF_THREADS>>>(
+    scalarProductKernel_native<<<NUM_OF_BLOCKS, NUM_OF_THREADS>>>(devVec[0], devVec[1], devResults, size);
-      devVec[0], devVec[1], devResults, size);
 
-  checkCudaErrors(cudaMemcpy(results, devResults,
+    checkCudaErrors(cudaMemcpy(results, devResults, NUM_OF_BLOCKS * sizeof *results, cudaMemcpyDeviceToHost));
-                             NUM_OF_BLOCKS * sizeof *results,
-                             cudaMemcpyDeviceToHost));
 
     float result_native = 0;
     for (int i = 0; i < NUM_OF_BLOCKS; ++i) {
@@ -186,12 +188,9 @@ int main(int argc, char *argv[]) {
     }
     printf("Result native operators\t: %f \n", result_native);
 
-  scalarProductKernel_intrinsics<<<NUM_OF_BLOCKS, NUM_OF_THREADS>>>(
+    scalarProductKernel_intrinsics<<<NUM_OF_BLOCKS, NUM_OF_THREADS>>>(devVec[0], devVec[1], devResults, size);
-      devVec[0], devVec[1], devResults, size);
 
-  checkCudaErrors(cudaMemcpy(results, devResults,
+    checkCudaErrors(cudaMemcpy(results, devResults, NUM_OF_BLOCKS * sizeof *results, cudaMemcpyDeviceToHost));
-                             NUM_OF_BLOCKS * sizeof *results,
-                             cudaMemcpyDeviceToHost));
 
     float result_intrinsics = 0;
     for (int i = 0; i < NUM_OF_BLOCKS; ++i) {
@@ -199,9 +198,7 @@ int main(int argc, char *argv[]) {
     }
     printf("Result intrinsics\t: %f \n", result_intrinsics);
 
-  printf("&&&& fp16ScalarProduct %s\n",
+    printf("&&&& fp16ScalarProduct %s\n", (fabs(result_intrinsics - result_native) < 0.00001) ? "PASSED" : "FAILED");
-         (fabs(result_intrinsics - result_native) < 0.00001) ? "PASSED"
-                                                             : "FAILED");
 
     for (int i = 0; i < 2; ++i) {
         checkCudaErrors(cudaFree(devVec[i]));

Samples/0_Introduction/matrixMul/matrixMul.cu

@@ -40,24 +40,23 @@
  */
 
 // System includes
-#include <stdio.h>
 #include <assert.h>
+#include <stdio.h>
 
 // CUDA runtime
-#include <cuda_runtime.h>
 #include <cuda_profiler_api.h>
+#include <cuda_runtime.h>
 
 // Helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 /**
  * Matrix multiplication (CUDA Kernel) on the device: C = A * B
  * wA is A's width and wB is B's width
  */
-template <int BLOCK_SIZE> __global__ void MatrixMulCUDA(float *C, float *A,
+template <int BLOCK_SIZE> __global__ void MatrixMulCUDA(float *C, float *A, float *B, int wA, int wB)
-    float *B, int wA,
+{
-    int wB) {
     // Block index
     int bx = blockIdx.x;
     int by = blockIdx.y;
@@ -87,9 +86,7 @@ template <int BLOCK_SIZE> __global__ void MatrixMulCUDA(float *C, float *A,
 
     // Loop over all the sub-matrices of A and B
     // required to compute the block sub-matrix
-  for (int a = aBegin, b = bBegin;
+    for (int a = aBegin, b = bBegin; a <= aEnd; a += aStep, b += bStep) {
-       a <= aEnd;
-       a += aStep, b += bStep) {
         // Declaration of the shared memory array As used to
         // store the sub-matrix of A
         __shared__ float As[BLOCK_SIZE][BLOCK_SIZE];
@@ -128,7 +125,8 @@ template <int BLOCK_SIZE> __global__ void MatrixMulCUDA(float *C, float *A,
     C[c + wB * ty + tx] = Csub;
 }
 
-void ConstantInit(float *data, int size, float val) {
+void ConstantInit(float *data, int size, float val)
+{
     for (int i = 0; i < size; ++i) {
         data[i] = val;
     }
@@ -137,9 +135,8 @@ void ConstantInit(float *data, int size, float val) {
 /**
  * Run a simple test of matrix multiplication using CUDA
  */
-int MatrixMultiply(int argc, char **argv,
+int MatrixMultiply(int argc, char **argv, int block_size, const dim3 &dimsA, const dim3 &dimsB)
-                   int block_size, const dim3 &dimsA,
+{
-                   const dim3 &dimsB) {
     // Allocate host memory for matrices A and B
     unsigned int size_A     = dimsA.x * dimsA.y;
     unsigned int mem_size_A = sizeof(float) * size_A;
@@ -181,10 +178,8 @@ int MatrixMultiply(int argc, char **argv,
     checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
 
     // copy host memory to device
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(d_A, h_A, mem_size_A, cudaMemcpyHostToDevice, stream));
-      cudaMemcpyAsync(d_A, h_A, mem_size_A, cudaMemcpyHostToDevice, stream));
+    checkCudaErrors(cudaMemcpyAsync(d_B, h_B, mem_size_B, cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(
-      cudaMemcpyAsync(d_B, h_B, mem_size_B, cudaMemcpyHostToDevice, stream));
 
     // Setup execution parameters
     dim3 threads(block_size, block_size);
@@ -195,11 +190,10 @@ int MatrixMultiply(int argc, char **argv,
 
     // Performs warmup operation using matrixMul CUDA kernel
     if (block_size == 16) {
-    MatrixMulCUDA<16>
+        MatrixMulCUDA<16><<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
-        <<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
+    }
-  } else {
+    else {
-    MatrixMulCUDA<32>
+        MatrixMulCUDA<32><<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
-        <<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
     }
 
     printf("done\n");
@@ -213,11 +207,10 @@ int MatrixMultiply(int argc, char **argv,
 
     for (int j = 0; j < nIter; j++) {
         if (block_size == 16) {
-      MatrixMulCUDA<16>
+            MatrixMulCUDA<16><<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
-          <<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
+        }
-    } else {
+        else {
-      MatrixMulCUDA<32>
+            MatrixMulCUDA<32><<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
-          <<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
         }
     }
 
@@ -232,19 +225,18 @@ int MatrixMultiply(int argc, char **argv,
 
     // Compute and print the performance
     float  msecPerMatrixMul = msecTotal / nIter;
-  double flopsPerMatrixMul = 2.0 * static_cast<double>(dimsA.x) *
+    double flopsPerMatrixMul =
-                             static_cast<double>(dimsA.y) *
+        2.0 * static_cast<double>(dimsA.x) * static_cast<double>(dimsA.y) * static_cast<double>(dimsB.x);
-                             static_cast<double>(dimsB.x);
+    double gigaFlops = (flopsPerMatrixMul * 1.0e-9f) / (msecPerMatrixMul / 1000.0f);
-  double gigaFlops =
+    printf("Performance= %.2f GFlop/s, Time= %.3f msec, Size= %.0f Ops,"
-      (flopsPerMatrixMul * 1.0e-9f) / (msecPerMatrixMul / 1000.0f);
-  printf(
-      "Performance= %.2f GFlop/s, Time= %.3f msec, Size= %.0f Ops,"
            " WorkgroupSize= %u threads/block\n",
-      gigaFlops, msecPerMatrixMul, flopsPerMatrixMul, threads.x * threads.y);
+           gigaFlops,
+           msecPerMatrixMul,
+           flopsPerMatrixMul,
+           threads.x * threads.y);
 
     // Copy result from device to host
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(h_C, d_C, mem_size_C, cudaMemcpyDeviceToHost, stream));
-      cudaMemcpyAsync(h_C, d_C, mem_size_C, cudaMemcpyDeviceToHost, stream));
     checkCudaErrors(cudaStreamSynchronize(stream));
 
     printf("Checking computed result for correctness: ");
@@ -261,8 +253,7 @@ int MatrixMultiply(int argc, char **argv,
         double rel_err    = abs_err / abs_val / dot_length;
 
         if (rel_err > eps) {
-      printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n",
+            printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n", i, h_C[i], dimsA.x * valB, eps);
-             i, h_C[i], dimsA.x * valB, eps);
             correct = false;
         }
     }
@@ -278,13 +269,13 @@ int MatrixMultiply(int argc, char **argv,
     checkCudaErrors(cudaFree(d_C));
     checkCudaErrors(cudaEventDestroy(start));
     checkCudaErrors(cudaEventDestroy(stop));
-  printf(
+    printf("\nNOTE: The CUDA Samples are not meant for performance "
-      "\nNOTE: The CUDA Samples are not meant for performance "
            "measurements. Results may vary when GPU Boost is enabled.\n");
 
     if (correct) {
         return EXIT_SUCCESS;
-  } else {
+    }
+    else {
         return EXIT_FAILURE;
     }
 }
@@ -293,15 +284,15 @@ int MatrixMultiply(int argc, char **argv,
 /**
  * Program main
  */
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("[Matrix Multiply Using CUDA] - Starting...\n");
 
-  if (checkCmdLineFlag(argc, (const char **)argv, "help") ||
+    if (checkCmdLineFlag(argc, (const char **)argv, "help") || checkCmdLineFlag(argc, (const char **)argv, "?")) {
-      checkCmdLineFlag(argc, (const char **)argv, "?")) {
         printf("Usage -device=n (n >= 0 for deviceID)\n");
         printf("      -wA=WidthA -hA=HeightA (Width x Height of Matrix A)\n");
         printf("      -wB=WidthB -hB=HeightB (Width x Height of Matrix B)\n");
-    printf("  Note: Outer matrix dimensions of A & B matrices" \
+        printf("  Note: Outer matrix dimensions of A & B matrices"
                " must be equal.\n");
 
         exit(EXIT_SUCCESS);
@@ -337,13 +328,11 @@ int main(int argc, char **argv) {
     }
 
     if (dimsA.x != dimsB.y) {
-    printf("Error: outer matrix dimensions must be equal. (%d != %d)\n",
+        printf("Error: outer matrix dimensions must be equal. (%d != %d)\n", dimsA.x, dimsB.y);
-           dimsA.x, dimsB.y);
         exit(EXIT_FAILURE);
     }
 
-  printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y,
+    printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y, dimsB.x, dimsB.y);
-         dimsB.x, dimsB.y);
 
     checkCudaErrors(cudaProfilerStart());
     int matrix_result = MatrixMultiply(argc, argv, block_size, dimsA, dimsB);

Samples/0_Introduction/matrixMulDrv/matrixMulDrv.cpp

@@ -46,23 +46,23 @@
 
 // includes, system
 #include <builtin_types.h>
+#include <cstring>
+#include <iostream>
 #include <math.h>
-#include <stdlib.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <iostream>
-#include <cstring>
 
 // includes, project, CUDA
+#include <cstring>
 #include <cuda.h>
 #include <helper_cuda_drvapi.h>
 #include <helper_image.h>
 #include <helper_string.h>
 #include <helper_timer.h>
-
-#include <cstring>
 #include <iostream>
 #include <string>
+
 #include "matrixMul.h"
 
 
@@ -71,11 +71,9 @@
 void runTest(int argc, char **argv);
 void randomInit(float *, int);
 
-extern "C" void computeGold(float *, const float *, const float *, unsigned int,
+extern "C" void computeGold(float *, const float *, const float *, unsigned int, unsigned int, unsigned int);
-                            unsigned int, unsigned int);
 
-static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
+static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *blk_size);
-                    int *blk_size);
 
 #ifndef FATBIN_FILE
 #define FATBIN_FILE "matrixMul_kernel64.fatbin"
@@ -91,7 +89,8 @@ size_t totalGlobalMem;
 
 const char *sSDKsample = "matrixMulDrv (Driver API)";
 
-void constantInit(float *data, int size, float val) {
+void constantInit(float *data, int size, float val)
+{
     for (int i = 0; i < size; ++i) {
         data[i] = val;
     }
@@ -100,7 +99,8 @@ void constantInit(float *data, int size, float val) {
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("[ %s ]\n", sSDKsample);
 
     runTest(argc, argv);
@@ -109,7 +109,8 @@ int main(int argc, char **argv) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     // initialize CUDA
     CUfunction matrixMul  = NULL;
     int        block_size = 0;
@@ -172,10 +173,19 @@ void runTest(int argc, char **argv) {
         size_t Matrix_Width_B = (size_t)WB;
         void  *args[5]        = {&d_C, &d_A, &d_B, &Matrix_Width_A, &Matrix_Width_B};
         // new CUDA 4.0 Driver API Kernel launch call
-    checkCudaErrors(cuLaunchKernel(
+        checkCudaErrors(cuLaunchKernel(matrixMul,
-        matrixMul, grid.x, grid.y, grid.z, block.x, block.y, block.z,
+                                       grid.x,
-        2 * block_size * block_size * sizeof(float), NULL, args, NULL));
+                                       grid.y,
-  } else {
+                                       grid.z,
+                                       block.x,
+                                       block.y,
+                                       block.z,
+                                       2 * block_size * block_size * sizeof(float),
+                                       NULL,
+                                       args,
+                                       NULL));
+    }
+    else {
         // This is the new CUDA 4.0 API for Kernel Parameter passing and Kernel
         // Launching (advanced method)
         int  offset = 0;
@@ -198,14 +208,20 @@ void runTest(int argc, char **argv) {
         *(reinterpret_cast<CUdeviceptr *>(&argBuffer[offset])) = Matrix_Width_B;
         offset += sizeof(Matrix_Width_B);
 
-    void *kernel_launch_config[5] = {CU_LAUNCH_PARAM_BUFFER_POINTER, argBuffer,
+        void *kernel_launch_config[5] = {
-                                     CU_LAUNCH_PARAM_BUFFER_SIZE, &offset,
+            CU_LAUNCH_PARAM_BUFFER_POINTER, argBuffer, CU_LAUNCH_PARAM_BUFFER_SIZE, &offset, CU_LAUNCH_PARAM_END};
-                                     CU_LAUNCH_PARAM_END};
 
         // new CUDA 4.0 Driver API Kernel launch call
-    checkCudaErrors(cuLaunchKernel(
+        checkCudaErrors(cuLaunchKernel(matrixMul,
-        matrixMul, grid.x, grid.y, grid.z, block.x, block.y, block.z,
+                                       grid.x,
-        2 * block_size * block_size * sizeof(float), NULL, NULL,
+                                       grid.y,
+                                       grid.z,
+                                       block.x,
+                                       block.y,
+                                       block.z,
+                                       2 * block_size * block_size * sizeof(float),
+                                       NULL,
+                                       NULL,
                                        reinterpret_cast<void **>(&kernel_launch_config)));
     }
 
@@ -222,8 +238,7 @@ void runTest(int argc, char **argv) {
 
     for (int i = 0; i < static_cast<int>(WC * HC); i++) {
         if (fabs(h_C[i] - (WA * valB)) > 1e-5) {
-      printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > 1e-5\n", i,
+            printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > 1e-5\n", i, h_C[i], WA * valB);
-             h_C[i], WA * valB);
             correct = false;
         }
     }
@@ -244,14 +259,15 @@ void runTest(int argc, char **argv) {
 }
 
 // Allocates a matrix with random float entries.
-void randomInit(float *data, int size) {
+void randomInit(float *data, int size)
+{
     for (int i = 0; i < size; ++i) {
         data[i] = rand() / static_cast<float>(RAND_MAX);
     }
 }
 
-static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
+static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *blk_size)
-                    int *blk_size) {
+{
     CUfunction cuFunction = 0;
     int        major = 0, minor = 0;
     char       deviceName[100];
@@ -259,16 +275,13 @@ static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
     cuDevice = findCudaDeviceDRV(argc, (const char **)argv);
 
     // get compute capabilities and the devicename
-  checkCudaErrors(cuDeviceGetAttribute(
+    checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
-      &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
+    checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
-  checkCudaErrors(cuDeviceGetAttribute(
-      &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
     checkCudaErrors(cuDeviceGetName(deviceName, sizeof(deviceName), cuDevice));
     printf("> GPU Device has SM %d.%d compute capability\n", major, minor);
 
     checkCudaErrors(cuDeviceTotalMem(&totalGlobalMem, cuDevice));
-  printf("  Total amount of global memory:     %llu bytes\n",
+    printf("  Total amount of global memory:     %llu bytes\n", (long long unsigned int)totalGlobalMem);
-         (long long unsigned int)totalGlobalMem);
 
     checkCudaErrors(cuCtxCreate(&cuContext, 0, cuDevice));
 
@@ -278,7 +291,8 @@ static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
 
     if (!findFatbinPath(FATBIN_FILE, module_path, argv, fatbin)) {
         exit(EXIT_FAILURE);
-  } else {
+    }
+    else {
         printf("> initCUDA loading module: <%s>\n", module_path.c_str());
     }
 
@@ -291,8 +305,7 @@ static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
     checkCudaErrors(cuModuleLoadData(&cuModule, fatbin.str().c_str()));
 
     // select the suitable kernel function
-  const char *kernels[] = {"matrixMul_bs32_64bit", "matrixMul_bs16_64bit",
+    const char *kernels[] = {"matrixMul_bs32_64bit", "matrixMul_bs16_64bit", "matrixMul_bs8_64bit"};
-                           "matrixMul_bs8_64bit"};
 
     int idx        = 0;
     int block_size = 32;
@@ -302,12 +315,12 @@ static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
 
         checkCudaErrors(cuModuleGetFunction(&cuFunction, cuModule, kernels[idx]));
         checkCudaErrors(cuOccupancyMaxPotentialBlockSize(
-        &blocksPerGrid, &threadsPerBlock, cuFunction, 0,
+            &blocksPerGrid, &threadsPerBlock, cuFunction, 0, 2 * block_size * block_size * sizeof(float), 0));
-        2 * block_size * block_size * sizeof(float), 0));
         if (block_size * block_size <= threadsPerBlock) {
             printf("> %d block size selected\n", block_size);
             break;
-    } else {
+        }
+        else {
             block_size /= 2;
         }
         idx++;

Samples/0_Introduction/matrixMulDrv/matrixMul_kernel.cu

@@ -42,8 +42,8 @@
 //! wA is A's width and wB is B's width
 ////////////////////////////////////////////////////////////////////////////////
 template <int block_size, typename size_type>
-__device__ void matrixMul(float *C, float *A, float *B, size_type wA,
+__device__ void matrixMul(float *C, float *A, float *B, size_type wA, size_type wB)
-                          size_type wB) {
+{
     // Block index
     size_type bx = blockIdx.x;
     size_type by = blockIdx.y;
@@ -96,7 +96,8 @@ __device__ void matrixMul(float *C, float *A, float *B, size_type wA,
         // of the block sub-matrix
 #pragma unroll
 
-    for (size_type k = 0; k < block_size; ++k) Csub += AS(ty, k) * BS(k, tx);
+        for (size_type k = 0; k < block_size; ++k)
+            Csub += AS(ty, k) * BS(k, tx);
 
         // Synchronize to make sure that the preceding
         // computation is done before loading two new
@@ -111,16 +112,16 @@ __device__ void matrixMul(float *C, float *A, float *B, size_type wA,
 }
 
 // C wrappers around our template kernel
-extern "C" __global__ void matrixMul_bs8_64bit(float *C, float *A, float *B,
+extern "C" __global__ void matrixMul_bs8_64bit(float *C, float *A, float *B, size_t wA, size_t wB)
-                                               size_t wA, size_t wB) {
+{
     matrixMul<8, size_t>(C, A, B, wA, wB);
 }
-extern "C" __global__ void matrixMul_bs16_64bit(float *C, float *A, float *B,
+extern "C" __global__ void matrixMul_bs16_64bit(float *C, float *A, float *B, size_t wA, size_t wB)
-                                                size_t wA, size_t wB) {
+{
     matrixMul<16, size_t>(C, A, B, wA, wB);
 }
-extern "C" __global__ void matrixMul_bs32_64bit(float *C, float *A, float *B,
+extern "C" __global__ void matrixMul_bs32_64bit(float *C, float *A, float *B, size_t wA, size_t wB)
-                                                size_t wA, size_t wB) {
+{
     matrixMul<32, size_t>(C, A, B, wA, wB);
 }
 

Samples/0_Introduction/matrixMulDynlinkJIT/cuda_drvapi_dynlink.c

@@ -20,9 +20,10 @@
 // #define CUDA_INIT_D3D11
 // #define CUDA_INIT_OPENGL
 
-#include <stdio.h>
 #include "cuda_drvapi_dynlink.h"
 
+#include <stdio.h>
+
 tcuInit                    *_cuInit;
 tcuDriverGetVersion        *cuDriverGetVersion;
 tcuDeviceGet               *cuDeviceGet;
@@ -239,8 +240,7 @@ static CUresult LOAD_LIBRARY(CUDADRIVER *pInstance)
 {
     *pInstance = LoadLibrary(__CudaLibName);
 
-    if (*pInstance == NULL)
+    if (*pInstance == NULL) {
-    {
         printf("LoadLibrary \"%s\" failed!\n", __CudaLibName);
         return CUDA_ERROR_UNKNOWN;
     }
@@ -251,24 +251,21 @@ static CUresult LOAD_LIBRARY(CUDADRIVER *pInstance)
 #define GET_PROC_EX(name, alias, required)                                               \
     alias = (t##name *)GetProcAddress(CudaDrvLib, #name);                                \
     if (alias == NULL && required) {                                                     \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", #name, __CudaLibName); \
-               #name, __CudaLibName);                                  \
         return CUDA_ERROR_UNKNOWN;                                                       \
     }
 
 #define GET_PROC_EX_V2(name, alias, required)                                                           \
     alias = (t##name *)GetProcAddress(CudaDrvLib, STRINGIFY(name##_v2));                                \
     if (alias == NULL && required) {                                                                    \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", STRINGIFY(name##_v2), __CudaLibName); \
-               STRINGIFY(name##_v2), __CudaLibName);                       \
         return CUDA_ERROR_UNKNOWN;                                                                      \
     }
 
 #define GET_PROC_EX_V3(name, alias, required)                                                           \
     alias = (t##name *)GetProcAddress(CudaDrvLib, STRINGIFY(name##_v3));                                \
     if (alias == NULL && required) {                                                                    \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", STRINGIFY(name##_v3), __CudaLibName); \
-               STRINGIFY(name##_v3), __CudaLibName);                       \
         return CUDA_ERROR_UNKNOWN;                                                                      \
     }
 
@@ -294,8 +291,7 @@ static CUresult LOAD_LIBRARY(CUDADRIVER *pInstance)
 {
     *pInstance = dlopen(__CudaLibName, RTLD_NOW);
 
-    if (*pInstance == NULL)
+    if (*pInstance == NULL) {
-    {
         printf("dlopen \"%s\" failed!\n", __CudaLibName);
         return CUDA_ERROR_UNKNOWN;
     }
@@ -306,24 +302,21 @@ static CUresult LOAD_LIBRARY(CUDADRIVER *pInstance)
 #define GET_PROC_EX(name, alias, required)                                               \
     alias = (t##name *)dlsym(CudaDrvLib, #name);                                         \
     if (alias == NULL && required) {                                                     \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", #name, __CudaLibName); \
-               #name, __CudaLibName);                                  \
         return CUDA_ERROR_UNKNOWN;                                                       \
     }
 
 #define GET_PROC_EX_V2(name, alias, required)                                                           \
     alias = (t##name *)dlsym(CudaDrvLib, STRINGIFY(name##_v2));                                         \
     if (alias == NULL && required) {                                                                    \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", STRINGIFY(name##_v2), __CudaLibName); \
-               STRINGIFY(name##_v2), __CudaLibName);                    \
         return CUDA_ERROR_UNKNOWN;                                                                      \
     }
 
 #define GET_PROC_EX_V3(name, alias, required)                                                           \
     alias = (t##name *)dlsym(CudaDrvLib, STRINGIFY(name##_v3));                                         \
     if (alias == NULL && required) {                                                                    \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", STRINGIFY(name##_v3), __CudaLibName); \
-               STRINGIFY(name##_v3), __CudaLibName);                    \
         return CUDA_ERROR_UNKNOWN;                                                                      \
     }
 
@@ -359,8 +352,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
     // available since 2.2. if not present, version 1.0 is assumed
     GET_PROC_OPTIONAL(cuDriverGetVersion);
 
-    if (cuDriverGetVersion)
+    if (cuDriverGetVersion) {
-    {
         CHECKED_CALL(cuDriverGetVersion(&driverVer));
     }
 
@@ -428,24 +420,21 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
     GET_PROC(cuStreamDestroy);
 
     // These are CUDA 5.0 new functions
-    if (driverVer >= 5000)
+    if (driverVer >= 5000) {
-    {
         GET_PROC(cuMipmappedArrayCreate);
         GET_PROC(cuMipmappedArrayDestroy);
         GET_PROC(cuMipmappedArrayGetLevel);
     }
 
     // These are CUDA 4.2 new functions
-    if (driverVer >= 4020)
+    if (driverVer >= 4020) {
-    {
         GET_PROC(cuFuncSetSharedMemConfig);
         GET_PROC(cuCtxGetSharedMemConfig);
         GET_PROC(cuCtxSetSharedMemConfig);
     }
 
     // These are CUDA 4.1 new functions
-    if (cudaVersion >= 4010 && __CUDA_API_VERSION >= 4010)
+    if (cudaVersion >= 4010 && __CUDA_API_VERSION >= 4010) {
-    {
         GET_PROC(cuDeviceGetByPCIBusId);
         GET_PROC(cuDeviceGetPCIBusId);
         GET_PROC(cuIpcGetEventHandle);
@@ -456,8 +445,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
     }
 
     // These could be _v2 interfaces
-    if (cudaVersion >= 4000 && __CUDA_API_VERSION >= 4000)
+    if (cudaVersion >= 4000 && __CUDA_API_VERSION >= 4000) {
-    {
         GET_PROC_V2(cuCtxDestroy);
         GET_PROC_V2(cuCtxPopCurrent);
         GET_PROC_V2(cuCtxPushCurrent);
@@ -465,8 +453,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC_V2(cuEventDestroy);
     }
 
-    if (cudaVersion >= 3020 && __CUDA_API_VERSION >= 3020)
+    if (cudaVersion >= 3020 && __CUDA_API_VERSION >= 3020) {
-    {
         GET_PROC_V2(cuDeviceTotalMem);
         GET_PROC_V2(cuCtxCreate);
         GET_PROC_V2(cuModuleGetGlobal);
@@ -507,17 +494,14 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC_V2(cuTexRefSetAddress);
         GET_PROC_V2(cuTexRefGetAddress);
 
-        if (cudaVersion >= 4010 && __CUDA_API_VERSION >= 4010)
+        if (cudaVersion >= 4010 && __CUDA_API_VERSION >= 4010) {
-        {
             GET_PROC_V3(cuTexRefSetAddress2D);
         }
-        else
+        else {
-        {
             GET_PROC_V2(cuTexRefSetAddress2D);
         }
     }
-    else
+    else {
-    {
         // versions earlier than 3020
         GET_PROC(cuDeviceTotalMem);
         GET_PROC(cuCtxCreate);
@@ -562,8 +546,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
     }
 
     // The following functions are specific to CUDA versions
-    if (driverVer >= 4000)
+    if (driverVer >= 4000) {
-    {
         GET_PROC(cuCtxSetCurrent);
         GET_PROC(cuCtxGetCurrent);
         GET_PROC(cuMemHostRegister);
@@ -574,8 +557,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC(cuProfilerStop);
     }
 
-    if (driverVer >= 3010)
+    if (driverVer >= 3010) {
-    {
         GET_PROC(cuModuleGetSurfRef);
         GET_PROC(cuSurfRefSetArray);
         GET_PROC(cuSurfRefGetArray);
@@ -583,8 +565,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC(cuCtxGetLimit);
     }
 
-    if (driverVer >= 3000)
+    if (driverVer >= 3000) {
-    {
         GET_PROC(cuMemcpyDtoDAsync);
         GET_PROC(cuFuncSetCacheConfig);
 #ifdef CUDA_INIT_D3D11
@@ -595,12 +576,10 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC(cuGraphicsUnregisterResource);
         GET_PROC(cuGraphicsSubResourceGetMappedArray);
 
-        if (cudaVersion >= 3020 && __CUDA_API_VERSION >= 3020)
+        if (cudaVersion >= 3020 && __CUDA_API_VERSION >= 3020) {
-        {
             GET_PROC_V2(cuGraphicsResourceGetMappedPointer);
         }
-        else
+        else {
-        {
             GET_PROC(cuGraphicsResourceGetMappedPointer);
         }
 
@@ -610,8 +589,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC(cuGetExportTable);
     }
 
-    if (driverVer >= 2030)
+    if (driverVer >= 2030) {
-    {
         GET_PROC(cuMemHostGetFlags);
 #ifdef CUDA_INIT_D3D10
         GET_PROC(cuD3D10GetDevice);
@@ -624,8 +602,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
 #endif
     }
 
-    if (driverVer >= 2010)
+    if (driverVer >= 2010) {
-    {
         GET_PROC(cuModuleLoadDataEx);
         GET_PROC(cuModuleLoadFatBinary);
 #ifdef CUDA_INIT_OPENGL

Samples/0_Introduction/matrixMulDynlinkJIT/cuda_drvapi_dynlink_cuda.h

@@ -43,7 +43,8 @@
 #define CUDA_VERSION 3020 /* 3.2 */
 
 #ifdef __cplusplus
-extern "C" {
+extern "C"
+{
 #endif
 
 /**
@@ -81,8 +82,7 @@ typedef struct CUuuid_st                                  /**< CUDA definition o
     /**
      * Context creation flags
      */
-typedef enum CUctx_flags_enum
+    typedef enum CUctx_flags_enum {
-{
         CU_CTX_SCHED_AUTO          = 0x00, /**< Automatic scheduling */
         CU_CTX_SCHED_SPIN          = 0x01, /**< Set spin as default scheduling */
         CU_CTX_SCHED_YIELD         = 0x02, /**< Set yield as default scheduling */
@@ -103,8 +103,7 @@ typedef enum CUctx_flags_enum
     /**
      * Event creation flags
      */
-typedef enum CUevent_flags_enum
+    typedef enum CUevent_flags_enum {
-{
         CU_EVENT_DEFAULT        = 0, /**< Default event flag */
         CU_EVENT_BLOCKING_SYNC  = 1, /**< Event uses blocking synchronization */
         CU_EVENT_DISABLE_TIMING = 2  /**< Event will not record timing data */
@@ -113,8 +112,7 @@ typedef enum CUevent_flags_enum
     /**
      * Array formats
      */
-typedef enum CUarray_format_enum
+    typedef enum CUarray_format_enum {
-{
         CU_AD_FORMAT_UNSIGNED_INT8  = 0x01, /**< Unsigned 8-bit integers */
         CU_AD_FORMAT_UNSIGNED_INT16 = 0x02, /**< Unsigned 16-bit integers */
         CU_AD_FORMAT_UNSIGNED_INT32 = 0x03, /**< Unsigned 32-bit integers */
@@ -128,8 +126,7 @@ typedef enum CUarray_format_enum
     /**
      * Texture reference addressing modes
      */
-typedef enum CUaddress_mode_enum
+    typedef enum CUaddress_mode_enum {
-{
         CU_TR_ADDRESS_MODE_WRAP   = 0, /**< Wrapping address mode */
         CU_TR_ADDRESS_MODE_CLAMP  = 1, /**< Clamp to edge address mode */
         CU_TR_ADDRESS_MODE_MIRROR = 2, /**< Mirror address mode */
@@ -139,8 +136,7 @@ typedef enum CUaddress_mode_enum
     /**
      * Texture reference filtering modes
      */
-typedef enum CUfilter_mode_enum
+    typedef enum CUfilter_mode_enum {
-{
         CU_TR_FILTER_MODE_POINT  = 0, /**< Point filter mode */
         CU_TR_FILTER_MODE_LINEAR = 1  /**< Linear filter mode */
     } CUfilter_mode;
@@ -148,8 +144,7 @@ typedef enum CUfilter_mode_enum
     /**
      * Device properties
      */
-typedef enum CUdevice_attribute_enum
+    typedef enum CUdevice_attribute_enum {
-{
         CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK       = 1, /**< Maximum number of threads per block */
         CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X             = 2, /**< Maximum block dimension X */
         CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y             = 3, /**< Maximum block dimension Y */
@@ -158,12 +153,15 @@ typedef enum CUdevice_attribute_enum
         CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y              = 6, /**< Maximum grid dimension Y */
         CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z              = 7, /**< Maximum grid dimension Z */
         CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK = 8, /**< Maximum shared memory available per block in bytes */
-    CU_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK = 8,            /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK */
+        CU_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK =
-    CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY = 9,              /**< Memory available on device for __constant__ variables in a CUDA C kernel in bytes */
+            8, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK */
+        CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY =
+            9, /**< Memory available on device for __constant__ variables in a CUDA C kernel in bytes */
         CU_DEVICE_ATTRIBUTE_WARP_SIZE               = 10, /**< Warp size in threads */
         CU_DEVICE_ATTRIBUTE_MAX_PITCH               = 11, /**< Maximum pitch in bytes allowed by memory copies */
         CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK = 12, /**< Maximum number of 32-bit registers available per block */
-    CU_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK = 12,               /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK */
+        CU_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK =
+            12,                                     /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK */
         CU_DEVICE_ATTRIBUTE_CLOCK_RATE        = 13, /**< Peak clock frequency in kilohertz */
         CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT = 14, /**< Alignment requirement for textures */
         CU_DEVICE_ATTRIBUTE_GPU_OVERLAP = 15, /**< Device can possibly copy memory and execute a kernel concurrently */
@@ -190,7 +188,8 @@ typedef enum CUdevice_attribute_enum
         CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR = 75, /**< Major compute capability version number */
         CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR = 76  /**< Minor compute capability version number */
 #if __CUDA_API_VERSION >= 4000
-                                     , CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE = 36,                 /**< Peak memory clock frequency in kilohertz */
+        ,
+        CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE              = 36, /**< Peak memory clock frequency in kilohertz */
         CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH        = 37, /**< Global memory bus width in bits */
         CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE                  = 38, /**< Size of L2 cache in bytes */
         CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR = 39, /**< Maximum resident threads per multiprocessor */
@@ -221,8 +220,7 @@ typedef struct CUdevprop_st
     /**
      * Function properties
      */
-typedef enum CUfunction_attribute_enum
+    typedef enum CUfunction_attribute_enum {
-{
         /**
          * The maximum number of threads per block, beyond which a launch of the
          * function would fail. This number depends on both the function and the
@@ -277,8 +275,7 @@ typedef enum CUfunction_attribute_enum
     /**
      * Function cache configurations
      */
-typedef enum CUfunc_cache_enum
+    typedef enum CUfunc_cache_enum {
-{
         CU_FUNC_CACHE_PREFER_NONE   = 0x00, /**< no preference for shared memory or L1 (default) */
         CU_FUNC_CACHE_PREFER_SHARED = 0x01, /**< prefer larger shared memory and smaller L1 cache */
         CU_FUNC_CACHE_PREFER_L1     = 0x02  /**< prefer larger L1 cache and smaller shared memory */
@@ -287,8 +284,7 @@ typedef enum CUfunc_cache_enum
     /**
      * Shared memory configurations
      */
-typedef enum CUsharedconfig_enum
+    typedef enum CUsharedconfig_enum {
-{
         CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE    = 0x00, /**< set default shared memory bank size */
         CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE  = 0x01, /**< set shared memory bank width to four bytes */
         CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE = 0x02  /**< set shared memory bank width to eight bytes */
@@ -297,33 +293,34 @@ typedef enum CUsharedconfig_enum
     /**
      * Memory types
      */
-typedef enum CUmemorytype_enum
+    typedef enum CUmemorytype_enum {
-{
         CU_MEMORYTYPE_HOST   = 0x01, /**< Host memory */
         CU_MEMORYTYPE_DEVICE = 0x02, /**< Device memory */
         CU_MEMORYTYPE_ARRAY  = 0x03  /**< Array memory */
 #if __CUDA_API_VERSION >= 4000
-                            , CU_MEMORYTYPE_UNIFIED = 0x04     /**< Unified device or host memory */
+        ,
+        CU_MEMORYTYPE_UNIFIED = 0x04 /**< Unified device or host memory */
 #endif
     } CUmemorytype;
 
     /**
      * Compute Modes
      */
-typedef enum CUcomputemode_enum
+    typedef enum CUcomputemode_enum {
-{
         CU_COMPUTEMODE_DEFAULT = 0, /**< Default compute mode (Multiple contexts allowed per device) */
-    CU_COMPUTEMODE_PROHIBITED        = 2  /**< Compute-prohibited mode (No contexts can be created on this device at this time) */
+        CU_COMPUTEMODE_PROHIBITED =
+            2 /**< Compute-prohibited mode (No contexts can be created on this device at this time) */
 #if __CUDA_API_VERSION >= 4000
-                                       , CU_COMPUTEMODE_EXCLUSIVE_PROCESS = 3  /**< Compute-exclusive-process mode (Only one context used by a single process can be present on this device at a time) */
+        ,
+        CU_COMPUTEMODE_EXCLUSIVE_PROCESS = 3 /**< Compute-exclusive-process mode (Only one context used by a single
+                                                process can be present on this device at a time) */
 #endif
     } CUcomputemode;
 
     /**
      * Online compiler options
      */
-typedef enum CUjit_option_enum
+    typedef enum CUjit_option_enum {
-{
         /**
          * Max number of registers that a thread may use.\n
          * Option type: unsigned int
@@ -414,8 +411,7 @@ typedef enum CUjit_option_enum
     /**
      * Online compilation targets
      */
-typedef enum CUjit_target_enum
+    typedef enum CUjit_target_enum {
-{
         CU_TARGET_COMPUTE_20 = 20, /**< Compute device class 2.0 */
         CU_TARGET_COMPUTE_21 = 21, /**< Compute device class 2.1 */
         CU_TARGET_COMPUTE_30 = 30, /**< Compute device class 3.0 */
@@ -434,8 +430,7 @@ typedef enum CUjit_target_enum
     /**
      * Cubin matching fallback strategies
      */
-typedef enum CUjit_fallback_enum
+    typedef enum CUjit_fallback_enum {
-{
         CU_PREFER_PTX = 0, /**< Prefer to compile ptx */
         CU_PREFER_BINARY   /**< Prefer to fall back to compatible binary code */
     } CUjit_fallback;
@@ -443,8 +438,7 @@ typedef enum CUjit_fallback_enum
     /**
      * Flags to register a graphics resource
      */
-typedef enum CUgraphicsRegisterFlags_enum
+    typedef enum CUgraphicsRegisterFlags_enum {
-{
         CU_GRAPHICS_REGISTER_FLAGS_NONE          = 0x00,
         CU_GRAPHICS_REGISTER_FLAGS_READ_ONLY     = 0x01,
         CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD = 0x02,
@@ -454,8 +448,7 @@ typedef enum CUgraphicsRegisterFlags_enum
     /**
      * Flags for mapping and unmapping interop resources
      */
-typedef enum CUgraphicsMapResourceFlags_enum
+    typedef enum CUgraphicsMapResourceFlags_enum {
-{
         CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE          = 0x00,
         CU_GRAPHICS_MAP_RESOURCE_FLAGS_READ_ONLY     = 0x01,
         CU_GRAPHICS_MAP_RESOURCE_FLAGS_WRITE_DISCARD = 0x02
@@ -464,8 +457,7 @@ typedef enum CUgraphicsMapResourceFlags_enum
     /**
      * Array indices for cube faces
      */
-typedef enum CUarray_cubemap_face_enum
+    typedef enum CUarray_cubemap_face_enum {
-{
         CU_CUBEMAP_FACE_POSITIVE_X = 0x00, /**< Positive X face of cubemap */
         CU_CUBEMAP_FACE_NEGATIVE_X = 0x01, /**< Negative X face of cubemap */
         CU_CUBEMAP_FACE_POSITIVE_Y = 0x02, /**< Positive Y face of cubemap */
@@ -477,8 +469,7 @@ typedef enum CUarray_cubemap_face_enum
     /**
      * Limits
      */
-typedef enum CUlimit_enum
+    typedef enum CUlimit_enum {
-{
         CU_LIMIT_STACK_SIZE       = 0x00, /**< GPU thread stack size */
         CU_LIMIT_PRINTF_FIFO_SIZE = 0x01, /**< GPU printf FIFO size */
         CU_LIMIT_MALLOC_HEAP_SIZE = 0x02  /**< GPU malloc heap size */
@@ -487,8 +478,7 @@ typedef enum CUlimit_enum
     /**
      * Resource types
      */
-typedef enum CUresourcetype_enum
+    typedef enum CUresourcetype_enum {
-{
         CU_RESOURCE_TYPE_ARRAY           = 0x00, /**< Array resoure */
         CU_RESOURCE_TYPE_MIPMAPPED_ARRAY = 0x01, /**< Mipmapped array resource */
         CU_RESOURCE_TYPE_LINEAR          = 0x02, /**< Linear resource */
@@ -498,8 +488,7 @@ typedef enum CUresourcetype_enum
     /**
      * Error codes
      */
-typedef enum cudaError_enum
+    typedef enum cudaError_enum {
-{
         /**
          * The API call returned with no errors. In the case of query calls, this
          * can also mean that the operation being queried is complete (see
@@ -1064,8 +1053,7 @@ typedef struct CUDA_TEXTURE_DESC_st
     /**
      * Resource view format
      */
-typedef enum CUresourceViewFormat_enum
+    typedef enum CUresourceViewFormat_enum {
-{
         CU_RES_VIEW_FORMAT_NONE          = 0x00, /**< No resource view format (use underlying resource format) */
         CU_RES_VIEW_FORMAT_UINT_1X8      = 0x01, /**< 1 channel unsigned 8-bit integers */
         CU_RES_VIEW_FORMAT_UINT_2X8      = 0x02, /**< 2 channel unsigned 8-bit integers */
@@ -1130,7 +1118,6 @@ typedef struct CUDA_POINTER_ATTRIBUTE_P2P_TOKENS_st
 #endif
 
 
-
 /**
  * If set, the CUDA array is a collection of layers, where each layer is either a 1D
  * or a 2D array and the Depth member of CUDA_ARRAY3D_DESCRIPTOR specifies the number
@@ -1420,7 +1407,11 @@ typedef CUresult  CUDAAPI tcuCtxSynchronize(void);
 
     typedef CUresult CUDAAPI tcuModuleLoad(CUmodule *module, const char *fname);
     typedef CUresult CUDAAPI tcuModuleLoadData(CUmodule *module, const void *image);
-typedef CUresult  CUDAAPI tcuModuleLoadDataEx(CUmodule *module, const void *image, unsigned int numOptions, CUjit_option *options, void **optionValues);
+    typedef CUresult CUDAAPI tcuModuleLoadDataEx(CUmodule     *module,
+                                                 const void   *image,
+                                                 unsigned int  numOptions,
+                                                 CUjit_option *options,
+                                                 void        **optionValues);
     typedef CUresult CUDAAPI tcuModuleLoadFatBinary(CUmodule *module, const void *fatCubin);
     typedef CUresult CUDAAPI tcuModuleUnload(CUmodule hmod);
     typedef CUresult CUDAAPI tcuModuleGetFunction(CUfunction *hfunc, CUmodule hmod, const char *name);
@@ -1449,8 +1440,7 @@ typedef CUresult CUDAAPI tcuMemAllocPitch(CUdeviceptr *dptr,
                                               size_t       Height,
                                               // size of biggest r/w to be performed by kernels on this memory
                                               // 4, 8 or 16 bytes
-                                          unsigned int ElementSizeBytes
+                                              unsigned int ElementSizeBytes);
-                                         );
 #else
 typedef CUresult CUDAAPI tcuMemGetInfo(unsigned int *free, unsigned int *total);
 typedef CUresult CUDAAPI tcuMemAlloc(CUdeviceptr *dptr, unsigned int bytesize);
@@ -1461,8 +1451,7 @@ typedef CUresult CUDAAPI tcuMemAllocPitch(CUdeviceptr *dptr,
                                           unsigned int  Height,
                                           // size of biggest r/w to be performed by kernels on this memory
                                           // 4, 8 or 16 bytes
-                                          unsigned int ElementSizeBytes
+                                          unsigned int ElementSizeBytes);
-                                         );
 #endif
 
     typedef CUresult CUDAAPI tcuMemFree(CUdeviceptr dptr);
@@ -1495,9 +1484,9 @@ typedef struct CUipcMemHandle_st
         char reserved[CU_IPC_HANDLE_SIZE];
     } CUipcMemHandle;
 
-typedef enum CUipcMem_flags_enum
+    typedef enum CUipcMem_flags_enum {
-{
+        CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS =
-    CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS = 0x1 /**< Automatically enable peer access between remote devices as needed */
+            0x1 /**< Automatically enable peer access between remote devices as needed */
     } CUipcMem_flags;
 
     typedef CUresult CUDAAPI tcuDeviceGetByPCIBusId(CUdevice *dev, char *pciBusId);
@@ -1510,9 +1499,14 @@ typedef CUresult CUDAAPI tcuIpcCloseMemHandle(CUdeviceptr dptr);
 #endif
 
     typedef CUresult CUDAAPI tcuMemHostRegister(void *p, size_t bytesize, unsigned int Flags);
-typedef CUresult CUDAAPI tcuMemHostUnregister(void *p);;
+    typedef CUresult CUDAAPI tcuMemHostUnregister(void *p);
+    ;
     typedef CUresult CUDAAPI tcuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount);
-typedef CUresult CUDAAPI tcuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount);
+    typedef CUresult CUDAAPI tcuMemcpyPeer(CUdeviceptr dstDevice,
+                                           CUcontext   dstContext,
+                                           CUdeviceptr srcDevice,
+                                           CUcontext   srcContext,
+                                           size_t      ByteCount);
 
 /************************************
  **
@@ -1541,7 +1535,8 @@ typedef CUresult  CUDAAPI tcuMemcpyHtoA(CUarray dstArray, size_t dstOffset, cons
     typedef CUresult CUDAAPI tcuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount);
 
     // array <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount);
+    typedef CUresult CUDAAPI
+    tcuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount);
 #else
 // system <-> device memory
 typedef CUresult CUDAAPI tcuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, unsigned int ByteCount);
@@ -1551,15 +1546,28 @@ typedef CUresult  CUDAAPI tcuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, un
 typedef CUresult CUDAAPI tcuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, unsigned int ByteCount);
 
 // device <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyDtoA(CUarray dstArray, unsigned int dstOffset, CUdeviceptr srcDevice, unsigned int ByteCount);
+typedef CUresult CUDAAPI tcuMemcpyDtoA(CUarray      dstArray,
-typedef CUresult  CUDAAPI tcuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount);
+                                       unsigned int dstOffset,
+                                       CUdeviceptr  srcDevice,
+                                       unsigned int ByteCount);
+typedef CUresult CUDAAPI tcuMemcpyAtoD(CUdeviceptr  dstDevice,
+                                       CUarray      srcArray,
+                                       unsigned int srcOffset,
+                                       unsigned int ByteCount);
 
 // system <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyHtoA(CUarray dstArray, unsigned int dstOffset, const void *srcHost, unsigned int ByteCount);
+typedef CUresult CUDAAPI tcuMemcpyHtoA(CUarray      dstArray,
+                                       unsigned int dstOffset,
+                                       const void  *srcHost,
+                                       unsigned int ByteCount);
 typedef CUresult CUDAAPI tcuMemcpyAtoH(void *dstHost, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount);
 
 // array <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyAtoA(CUarray dstArray, unsigned int dstOffset, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount);
+typedef CUresult CUDAAPI tcuMemcpyAtoA(CUarray      dstArray,
+                                       unsigned int dstOffset,
+                                       CUarray      srcArray,
+                                       unsigned int srcOffset,
+                                       unsigned int ByteCount);
 #endif
 
     // 2D memcpy
@@ -1586,36 +1594,51 @@ typedef CUresult  CUDAAPI tcuMemcpy3D(const CUDA_MEMCPY3D *pCopy);
 #if __CUDA_API_VERSION >= 3020
     // system <-> device memory
     typedef CUresult CUDAAPI tcuMemcpyHtoDAsync(CUdeviceptr dstDevice,
-                                             const void *srcHost, size_t ByteCount, CUstream hStream);
+                                                const void *srcHost,
+                                                size_t      ByteCount,
+                                                CUstream    hStream);
     typedef CUresult CUDAAPI tcuMemcpyDtoHAsync(void       *dstHost,
-                                             CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
+                                                CUdeviceptr srcDevice,
+                                                size_t      ByteCount,
+                                                CUstream    hStream);
 
     // device <-> device memory
     typedef CUresult CUDAAPI tcuMemcpyDtoDAsync(CUdeviceptr dstDevice,
-                                             CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
+                                                CUdeviceptr srcDevice,
+                                                size_t      ByteCount,
+                                                CUstream    hStream);
 
     // system <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset,
+    typedef CUresult CUDAAPI
-                                             const void *srcHost, size_t ByteCount, CUstream hStream);
+    tcuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount, CUstream hStream);
-typedef CUresult  CUDAAPI tcuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOffset,
+    typedef CUresult CUDAAPI
-                                             size_t ByteCount, CUstream hStream);
+    tcuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount, CUstream hStream);
 
 #else
 // system <-> device memory
 typedef CUresult CUDAAPI tcuMemcpyHtoDAsync(CUdeviceptr  dstDevice,
-                                             const void *srcHost, unsigned int ByteCount, CUstream hStream);
+                                            const void  *srcHost,
+                                            unsigned int ByteCount,
+                                            CUstream     hStream);
 typedef CUresult CUDAAPI tcuMemcpyDtoHAsync(void        *dstHost,
-                                             CUdeviceptr srcDevice, unsigned int ByteCount, CUstream hStream);
+                                            CUdeviceptr  srcDevice,
+                                            unsigned int ByteCount,
+                                            CUstream     hStream);
 
 // device <-> device memory
 typedef CUresult CUDAAPI tcuMemcpyDtoDAsync(CUdeviceptr  dstDevice,
-                                             CUdeviceptr srcDevice, unsigned int ByteCount, CUstream hStream);
+                                            CUdeviceptr  srcDevice,
+                                            unsigned int ByteCount,
+                                            CUstream     hStream);
 
 // system <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyHtoAAsync(CUarray dstArray, unsigned int dstOffset,
+typedef CUresult CUDAAPI tcuMemcpyHtoAAsync(CUarray      dstArray,
-                                             const void *srcHost, unsigned int ByteCount, CUstream hStream);
+                                            unsigned int dstOffset,
-typedef CUresult  CUDAAPI tcuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, unsigned int srcOffset,
+                                            const void  *srcHost,
-                                             unsigned int ByteCount, CUstream hStream);
+                                            unsigned int ByteCount,
+                                            CUstream     hStream);
+typedef CUresult CUDAAPI
+tcuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount, CUstream hStream);
 #endif
 
     // 2D memcpy
@@ -1634,13 +1657,22 @@ typedef CUresult  CUDAAPI tcuMemsetD16(CUdeviceptr dstDevice, unsigned short us,
     typedef CUresult CUDAAPI tcuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, unsigned int N);
 
 #if __CUDA_API_VERSION >= 3020
-typedef CUresult  CUDAAPI tcuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, size_t Width, size_t Height);
+    typedef CUresult CUDAAPI
-typedef CUresult  CUDAAPI tcuMemsetD2D16(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned short us, size_t Width, size_t Height);
+    tcuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, size_t Width, size_t Height);
-typedef CUresult  CUDAAPI tcuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, size_t Width, size_t Height);
+    typedef CUresult CUDAAPI
+    tcuMemsetD2D16(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned short us, size_t Width, size_t Height);
+    typedef CUresult CUDAAPI
+    tcuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, size_t Width, size_t Height);
 #else
-typedef CUresult  CUDAAPI tcuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, unsigned int Width, unsigned int Height);
+typedef CUresult CUDAAPI
-typedef CUresult  CUDAAPI tcuMemsetD2D16(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned short us, unsigned int Width, unsigned int Height);
+tcuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, unsigned int Width, unsigned int Height);
-typedef CUresult  CUDAAPI tcuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, unsigned int Width, unsigned int Height);
+typedef CUresult CUDAAPI tcuMemsetD2D16(CUdeviceptr    dstDevice,
+                                        unsigned int   dstPitch,
+                                        unsigned short us,
+                                        unsigned int   Width,
+                                        unsigned int   Height);
+typedef CUresult CUDAAPI
+tcuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, unsigned int Width, unsigned int Height);
 #endif
 
     /************************************
@@ -1657,10 +1689,16 @@ typedef CUresult CUDAAPI tcuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache co
     typedef CUresult CUDAAPI tcuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig config);
 
     typedef CUresult CUDAAPI tcuLaunchKernel(CUfunction   f,
-                                         unsigned int gridDimX,  unsigned int gridDimY,  unsigned int gridDimZ,
+                                             unsigned int gridDimX,
-                                         unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ,
+                                             unsigned int gridDimY,
+                                             unsigned int gridDimZ,
+                                             unsigned int blockDimX,
+                                             unsigned int blockDimY,
+                                             unsigned int blockDimZ,
                                              unsigned int sharedMemBytes,
-                                         CUstream hStream, void **kernelParams, void **extra);
+                                             CUstream     hStream,
+                                             void       **kernelParams,
+                                             void       **extra);
 
     /************************************
      **
@@ -1676,8 +1714,12 @@ typedef CUresult  CUDAAPI tcuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_
     typedef CUresult CUDAAPI tcuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray);
 
 #if __CUDA_API_VERSION >= 5000
-typedef CUresult CUDAAPI tcuMipmappedArrayCreate(CUmipmappedArray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc, unsigned int numMipmapLevels);
+    typedef CUresult CUDAAPI tcuMipmappedArrayCreate(CUmipmappedArray              *pHandle,
-typedef CUresult CUDAAPI tcuMipmappedArrayGetLevel(CUarray *pLevelArray, CUmipmappedArray hMipmappedArray, unsigned int level);
+                                                     const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc,
+                                                     unsigned int                   numMipmapLevels);
+    typedef CUresult CUDAAPI tcuMipmappedArrayGetLevel(CUarray         *pLevelArray,
+                                                       CUmipmappedArray hMipmappedArray,
+                                                       unsigned int     level);
     typedef CUresult CUDAAPI tcuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray);
 #endif
 
@@ -1694,10 +1736,19 @@ typedef CUresult  CUDAAPI tcuTexRefSetArray(CUtexref hTexRef, CUarray hArray, un
 
 #if __CUDA_API_VERSION >= 3020
     typedef CUresult CUDAAPI tcuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, size_t bytes);
-typedef CUresult  CUDAAPI tcuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, size_t Pitch);
+    typedef CUresult CUDAAPI tcuTexRefSetAddress2D(CUtexref                     hTexRef,
+                                                   const CUDA_ARRAY_DESCRIPTOR *desc,
+                                                   CUdeviceptr                  dptr,
+                                                   size_t                       Pitch);
 #else
-typedef CUresult  CUDAAPI tcuTexRefSetAddress(unsigned int *ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, unsigned int bytes);
+typedef CUresult CUDAAPI tcuTexRefSetAddress(unsigned int *ByteOffset,
-typedef CUresult  CUDAAPI tcuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, unsigned int Pitch);
+                                             CUtexref      hTexRef,
+                                             CUdeviceptr   dptr,
+                                             unsigned int  bytes);
+typedef CUresult CUDAAPI tcuTexRefSetAddress2D(CUtexref                     hTexRef,
+                                               const CUDA_ARRAY_DESCRIPTOR *desc,
+                                               CUdeviceptr                  dptr,
+                                               unsigned int                 Pitch);
 #endif
 
     typedef CUresult CUDAAPI tcuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int NumPackedComponents);
@@ -1763,7 +1814,10 @@ typedef CUresult CUDAAPI tcuEventElapsedTime(float *pMilliseconds, CUevent hStar
      ***********************************/
     typedef CUresult CUDAAPI tcuStreamCreate(CUstream *phStream, unsigned int Flags);
     typedef CUresult CUDAAPI tcuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags);
-typedef CUresult CUDAAPI tcuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags);
+    typedef CUresult CUDAAPI tcuStreamAddCallback(CUstream         hStream,
+                                                  CUstreamCallback callback,
+                                                  void            *userData,
+                                                  unsigned int     flags);
 
     typedef CUresult CUDAAPI tcuStreamQuery(CUstream hStream);
     typedef CUresult CUDAAPI tcuStreamSynchronize(CUstream hStream);
@@ -1775,17 +1829,28 @@ typedef CUresult CUDAAPI tcuStreamDestroy(CUstream hStream);
      **
      ***********************************/
     typedef CUresult CUDAAPI tcuGraphicsUnregisterResource(CUgraphicsResource resource);
-typedef CUresult CUDAAPI tcuGraphicsSubResourceGetMappedArray(CUarray *pArray, CUgraphicsResource resource, unsigned int arrayIndex, unsigned int mipLevel);
+    typedef CUresult CUDAAPI tcuGraphicsSubResourceGetMappedArray(CUarray           *pArray,
+                                                                  CUgraphicsResource resource,
+                                                                  unsigned int       arrayIndex,
+                                                                  unsigned int       mipLevel);
 
 #if __CUDA_API_VERSION >= 3020
-typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t *pSize, CUgraphicsResource resource);
+    typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr       *pDevPtr,
+                                                                 size_t            *pSize,
+                                                                 CUgraphicsResource resource);
 #else
-typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, unsigned int *pSize, CUgraphicsResource resource);
+typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr       *pDevPtr,
+                                                             unsigned int      *pSize,
+                                                             CUgraphicsResource resource);
 #endif
 
     typedef CUresult CUDAAPI tcuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsigned int flags);
-typedef CUresult CUDAAPI tcuGraphicsMapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream);
+    typedef CUresult CUDAAPI tcuGraphicsMapResources(unsigned int        count,
-typedef CUresult CUDAAPI tcuGraphicsUnmapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream);
+                                                     CUgraphicsResource *resources,
+                                                     CUstream            hStream);
+    typedef CUresult CUDAAPI tcuGraphicsUnmapResources(unsigned int        count,
+                                                       CUgraphicsResource *resources,
+                                                       CUstream            hStream);
 
     /************************************
      **

Samples/0_Introduction/matrixMulDynlinkJIT/helper_cuda_drvapi.h

@@ -14,21 +14,17 @@
 #ifndef HELPER_CUDA_DRVAPI_H
 #define HELPER_CUDA_DRVAPI_H
 
+#include <helper_string.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 
-#include <helper_string.h>
-
 #ifndef MAX
 #define MAX(a, b) (a > b ? a : b)
 #endif
 
 #ifndef HELPER_CUDA_DRVAPI_H
-inline int ftoi(float value) {
+inline int ftoi(float value) { return (value >= 0 ? static_cast<int>(value + 0.5) : static_cast<int>(value - 0.5)); }
-  return (value >= 0 ? static_cast<int>(value + 0.5)
-                     : static_cast<int>(value - 0.5));
-}
 #endif
 
 #ifndef EXIT_WAIVED
@@ -47,39 +43,43 @@ inline int ftoi(float value) {
 #define checkCudaErrors(err) __checkCudaErrors(err, __FILE__, __LINE__)
 
 // These are the inline versions for all of the SDK helper functions
-inline void __checkCudaErrors(CUresult err, const char *file, const int line) {
+inline void __checkCudaErrors(CUresult err, const char *file, const int line)
+{
     if (CUDA_SUCCESS != err) {
         const char *errorStr = NULL;
         cuGetErrorString(err, &errorStr);
         fprintf(stderr,
                 "checkCudaErrors() Driver API error = %04d \"%s\" from file <%s>, "
                 "line %i.\n",
-            err, errorStr, file, line);
+                err,
+                errorStr,
+                file,
+                line);
         exit(EXIT_FAILURE);
     }
 }
 #endif
 
 // This function wraps the CUDA Driver API into a template function
-template <class T>
+template <class T> inline void getCudaAttribute(T *attribute, CUdevice_attribute device_attribute, int device)
-inline void getCudaAttribute(T *attribute, CUdevice_attribute device_attribute,
+{
-                             int device) {
     checkCudaErrors(cuDeviceGetAttribute(attribute, device_attribute, device));
 }
 #endif
 
 // Beginning of GPU Architecture definitions
-inline int _ConvertSMVer2CoresDRV(int major, int minor) {
+inline int _ConvertSMVer2CoresDRV(int major, int minor)
+{
     // Defines for GPU Architecture types (using the SM version to determine the #
     // of cores per SM
-  typedef struct {
+    typedef struct
+    {
         int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM
                 // minor version
         int Cores;
     } sSMtoCores;
 
-  sSMtoCores nGpuArchCoresPerSM[] = {
+    sSMtoCores nGpuArchCoresPerSM[] = {{0x30, 192},
-      {0x30, 192},
                                        {0x32, 192},
                                        {0x35, 192},
                                        {0x37, 192},
@@ -110,16 +110,18 @@ inline int _ConvertSMVer2CoresDRV(int major, int minor) {
 
     // If we don't find the values, we default use the previous one to run
     // properly
-  printf(
+    printf("MapSMtoCores for SM %d.%d is undefined.  Default to use %d Cores/SM\n",
-      "MapSMtoCores for SM %d.%d is undefined.  Default to use %d Cores/SM\n",
+           major,
-      major, minor, nGpuArchCoresPerSM[index - 1].Cores);
+           minor,
+           nGpuArchCoresPerSM[index - 1].Cores);
     return nGpuArchCoresPerSM[index - 1].Cores;
 }
 // end of GPU Architecture definitions
 
 #ifdef __cuda_cuda_h__
 // General GPU Device CUDA Initialization
-inline int gpuDeviceInitDRV(int ARGC, const char **ARGV) {
+inline int gpuDeviceInitDRV(int ARGC, const char **ARGV)
+{
     int cuDevice    = 0;
     int deviceCount = 0;
     checkCudaErrors(cuInit(0, __CUDA_API_VERSION));
@@ -140,11 +142,8 @@ inline int gpuDeviceInitDRV(int ARGC, const char **ARGV) {
 
     if (dev > deviceCount - 1) {
         fprintf(stderr, "\n");
-    fprintf(stderr, ">> %d CUDA capable GPU device(s) detected. <<\n",
+        fprintf(stderr, ">> %d CUDA capable GPU device(s) detected. <<\n", deviceCount);
-            deviceCount);
+        fprintf(stderr, ">> cudaDeviceInit (-device=%d) is not a valid GPU device. <<\n", dev);
-    fprintf(stderr,
-            ">> cudaDeviceInit (-device=%d) is not a valid GPU device. <<\n",
-            dev);
         fprintf(stderr, "\n");
         return -dev;
     }
@@ -171,7 +170,8 @@ inline int gpuDeviceInitDRV(int ARGC, const char **ARGV) {
 }
 
 // This function returns the best GPU based on performance
-inline int gpuGetMaxGflopsDeviceIdDRV() {
+inline int gpuGetMaxGflopsDeviceIdDRV()
+{
     CUdevice           current_device   = 0;
     CUdevice           max_perf_device  = 0;
     int                device_count     = 0;
@@ -187,8 +187,7 @@ inline int gpuGetMaxGflopsDeviceIdDRV() {
     checkCudaErrors(cuDeviceGetCount(&device_count));
 
     if (device_count == 0) {
-    fprintf(stderr,
+        fprintf(stderr, "gpuGetMaxGflopsDeviceIdDRV error: no devices supporting CUDA\n");
-            "gpuGetMaxGflopsDeviceIdDRV error: no devices supporting CUDA\n");
         exit(EXIT_FAILURE);
     }
 
@@ -196,36 +195,31 @@ inline int gpuGetMaxGflopsDeviceIdDRV() {
     current_device = 0;
 
     while (current_device < device_count) {
-    checkCudaErrors(cuDeviceGetAttribute(
+        checkCudaErrors(
-        &multiProcessorCount, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
+            cuDeviceGetAttribute(&multiProcessorCount, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, current_device));
-        current_device));
+        checkCudaErrors(cuDeviceGetAttribute(&clockRate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, current_device));
-    checkCudaErrors(cuDeviceGetAttribute(
+        checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, current_device));
-        &clockRate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, current_device));
+        checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, current_device));
-    checkCudaErrors(cuDeviceGetAttribute(
-        &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, current_device));
-    checkCudaErrors(cuDeviceGetAttribute(
-        &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, current_device));
 
         int computeMode;
-    getCudaAttribute<int>(&computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE,
+        getCudaAttribute<int>(&computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, current_device);
-                          current_device);
 
         if (computeMode != CU_COMPUTEMODE_PROHIBITED) {
             if (major == 9999 && minor == 9999) {
                 sm_per_multiproc = 1;
-      } else {
+            }
+            else {
                 sm_per_multiproc = _ConvertSMVer2CoresDRV(major, minor);
             }
 
-      unsigned long long compute_perf =
+            unsigned long long compute_perf = (unsigned long long)(multiProcessorCount * sm_per_multiproc * clockRate);
-          (unsigned long long)(multiProcessorCount * sm_per_multiproc *
-                               clockRate);
 
             if (compute_perf > max_compute_perf) {
                 max_compute_perf = compute_perf;
                 max_perf_device  = current_device;
             }
-    } else {
+        }
+        else {
             devices_prohibited++;
         }
 
@@ -243,7 +237,8 @@ inline int gpuGetMaxGflopsDeviceIdDRV() {
 }
 
 // General initialization call to pick the best CUDA Device
-inline CUdevice findCudaDeviceDRV(int argc, const char **argv) {
+inline CUdevice findCudaDeviceDRV(int argc, const char **argv)
+{
     CUdevice cuDevice;
     int      devID = 0;
 
@@ -255,7 +250,8 @@ inline CUdevice findCudaDeviceDRV(int argc, const char **argv) {
             printf("exiting...\n");
             exit(EXIT_SUCCESS);
         }
-  } else {
+    }
+    else {
         // Otherwise pick the device with highest Gflops/s
         char name[100];
         devID = gpuGetMaxGflopsDeviceIdDRV();
@@ -269,7 +265,8 @@ inline CUdevice findCudaDeviceDRV(int argc, const char **argv) {
     return cuDevice;
 }
 
-inline CUdevice findIntegratedGPUDrv() {
+inline CUdevice findIntegratedGPUDrv()
+{
     CUdevice current_device     = 0;
     int      device_count       = 0;
     int      devices_prohibited = 0;
@@ -286,28 +283,22 @@ inline CUdevice findIntegratedGPUDrv() {
     // Find the integrated GPU which is compute capable
     while (current_device < device_count) {
         int computeMode = -1;
-    checkCudaErrors(cuDeviceGetAttribute(
+        checkCudaErrors(cuDeviceGetAttribute(&isIntegrated, CU_DEVICE_ATTRIBUTE_INTEGRATED, current_device));
-        &isIntegrated, CU_DEVICE_ATTRIBUTE_INTEGRATED, current_device));
+        checkCudaErrors(cuDeviceGetAttribute(&computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, current_device));
-    checkCudaErrors(cuDeviceGetAttribute(
-        &computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, current_device));
 
         // If GPU is integrated and is not running on Compute Mode prohibited use
         // that
         if (isIntegrated && (computeMode != CU_COMPUTEMODE_PROHIBITED)) {
             int  major = 0, minor = 0;
             char deviceName[256];
-      checkCudaErrors(cuDeviceGetAttribute(
+            checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, current_device));
-          &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR,
+            checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, current_device));
-          current_device));
-      checkCudaErrors(cuDeviceGetAttribute(
-          &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR,
-          current_device));
             checkCudaErrors(cuDeviceGetName(deviceName, 256, current_device));
-      printf("GPU Device %d: \"%s\" with compute capability %d.%d\n\n",
+            printf("GPU Device %d: \"%s\" with compute capability %d.%d\n\n", current_device, deviceName, major, minor);
-             current_device, deviceName, major, minor);
 
             return current_device;
-    } else {
+        }
+        else {
             devices_prohibited++;
         }
 
@@ -323,29 +314,26 @@ inline CUdevice findIntegratedGPUDrv() {
 }
 
 // General check for CUDA GPU SM Capabilities
-inline bool checkCudaCapabilitiesDRV(int major_version, int minor_version,
+inline bool checkCudaCapabilitiesDRV(int major_version, int minor_version, int devID)
-                                     int devID) {
+{
     CUdevice cuDevice;
     char     name[256];
     int      major = 0, minor = 0;
 
     checkCudaErrors(cuDeviceGet(&cuDevice, devID));
     checkCudaErrors(cuDeviceGetName(name, 100, cuDevice));
-  checkCudaErrors(cuDeviceGetAttribute(
+    checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
-      &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
+    checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
-  checkCudaErrors(cuDeviceGetAttribute(
+
-      &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
+    if ((major > major_version) || (major == major_version && minor >= minor_version)) {
-
+        printf("> Device %d: <%16s >, Compute SM %d.%d detected\n", devID, name, major, minor);
-  if ((major > major_version) ||
-      (major == major_version && minor >= minor_version)) {
-    printf("> Device %d: <%16s >, Compute SM %d.%d detected\n", devID, name,
-           major, minor);
         return true;
-  } else {
+    }
-    printf(
+    else {
-        "No GPU device was found that can support CUDA compute capability "
+        printf("No GPU device was found that can support CUDA compute capability "
                "%d.%d.\n",
-        major_version, minor_version);
+               major_version,
+               minor_version);
         return false;
     }
 }
@@ -354,4 +342,3 @@ inline bool checkCudaCapabilitiesDRV(int major_version, int minor_version,
 // end of CUDA Helper Functions
 
 #endif // HELPER_CUDA_DRVAPI_H
-

Samples/0_Introduction/matrixMulDynlinkJIT/matrixMulDynlinkJIT.cpp

@@ -43,10 +43,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 // includes, CUDA
 #include "cuda_drvapi_dynlink.h"
@@ -78,8 +78,7 @@ static const char *sSDKsample = "matrixMulDynlinkJIT (CUDA dynamic linking)";
 ////////////////////////////////////////////////////////////////////////////////
 void randomInit(float *data, size_t size)
 {
-    for (size_t i = 0; i < size; ++i)
+    for (size_t i = 0; i < size; ++i) {
-    {
         data[i] = rand() / (float)RAND_MAX;
     }
 }
@@ -100,18 +99,14 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
     checkCudaErrors(cuInit(0, __CUDA_API_VERSION));
 
     // This assumes that the user is attempting to specify a explicit device -device=n
-    if (argc > 1)
+    if (argc > 1) {
-    {
         bool bFound = false;
 
-        for (int param=0; param < argc; param++)
+        for (int param = 0; param < argc; param++) {
-        {
+            if (!strncmp(argv[param], "-device", 7)) {
-            if (!strncmp(argv[param], "-device", 7))
-            {
                 int i = (int)strlen(argv[1]);
 
-                while (argv[1][i] != '=')
+                while (argv[1][i] != '=') {
-                {
                     i--;
                 }
 
@@ -128,16 +123,15 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
     int deviceCount = 0;
     checkCudaErrors(cuDeviceGetCount(&deviceCount));
 
-    if (deviceCount == 0)
+    if (deviceCount == 0) {
-    {
         fprintf(stderr, "No devices supporting CUDA detected, exiting...\n");
         exit(EXIT_SUCCESS);
     }
 
-    if (devID < 0) devID = 0;
+    if (devID < 0)
+        devID = 0;
 
-    if (devID > deviceCount -1)
+    if (devID > deviceCount - 1) {
-    {
         fprintf(stderr, "initCUDA (Device=%d) invalid GPU device.  %d GPU device(s) detected.\n\n", devID, deviceCount);
         status = CUDA_ERROR_NOT_FOUND;
 
@@ -159,8 +153,7 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
     // create context for picked device
     status = cuCtxCreate(&g_cuContext, 0, cuDevice);
 
-    if (CUDA_SUCCESS != status)
+    if (CUDA_SUCCESS != status) {
-    {
         cuCtxDestroy(g_cuContext);
         exit(EXIT_SUCCESS);
     }
@@ -191,9 +184,11 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
         printf("> Compiling CUDA module\n");
 
 #if defined(_WIN64) || defined(__LP64__)
-        status = cuModuleLoadDataEx(&cuModule, matrixMul_kernel_64_ptxdump, jitNumOptions, jitOptions, (void **)jitOptVals);
+        status =
+            cuModuleLoadDataEx(&cuModule, matrixMul_kernel_64_ptxdump, jitNumOptions, jitOptions, (void **)jitOptVals);
 #else
-        status = cuModuleLoadDataEx(&cuModule, matrixMul_kernel_32_ptxdump, jitNumOptions, jitOptions, (void **)jitOptVals);
+        status =
+            cuModuleLoadDataEx(&cuModule, matrixMul_kernel_32_ptxdump, jitNumOptions, jitOptions, (void **)jitOptVals);
 #endif
 
         printf("> PTX JIT log:\n%s\n", jitLogBuffer);
@@ -203,19 +198,17 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
         delete[] jitLogBuffer;
     }
 
-    if (CUDA_SUCCESS != status)
+    if (CUDA_SUCCESS != status) {
-    {
         printf("Error while compiling PTX\n");
         cuCtxDestroy(g_cuContext);
         exit(EXIT_FAILURE);
     }
 
     // retrieve CUDA function from the compiled module
-    status = cuModuleGetFunction(&cuFunction, cuModule,
+    status = cuModuleGetFunction(
-                                 (block_size == 16) ? "matrixMul_bs16_32bit" : "matrixMul_bs32_32bit");
+        &cuFunction, cuModule, (block_size == 16) ? "matrixMul_bs16_32bit" : "matrixMul_bs32_32bit");
 
-    if (CUDA_SUCCESS != status)
+    if (CUDA_SUCCESS != status) {
-    {
         cuCtxDestroy(g_cuContext);
         exit(EXIT_FAILURE);
     }
@@ -280,10 +273,8 @@ int main(int argc, char **argv)
         int   Matrix_Width_B = WB;
         void *args[5]        = {&d_C, &d_A, &d_B, &Matrix_Width_A, &Matrix_Width_B};
 
-        checkCudaErrors(cuLaunchKernel(matrixMul, (WC/block_size), (HC/block_size), 1,
+        checkCudaErrors(cuLaunchKernel(
-                                       block_size     , block_size     , 1,
+            matrixMul, (WC / block_size), (HC / block_size), 1, block_size, block_size, 1, 0, NULL, args, NULL));
-                                       0,
-                                       NULL, args, NULL));
     }
 #else // __CUDA_API_VERSION <= 3020
     {
@@ -331,8 +322,7 @@ int main(int argc, char **argv)
     // check result
     float diff = 0.0f;
 
-    for (unsigned int i=0; i<size_C; i++)
+    for (unsigned int i = 0; i < size_C; i++) {
-    {
         float tmp = reference[i] - h_C[i];
         diff += tmp * tmp;
     }

Samples/0_Introduction/matrixMulDynlinkJIT/matrixMul_gold.cpp

@@ -28,8 +28,7 @@
 
 ////////////////////////////////////////////////////////////////////////////////
 // export C interface
-extern "C"
+extern "C" void computeGold(float *, const float *, const float *, unsigned int, unsigned int, unsigned int);
-void computeGold(float *, const float *, const float *, unsigned int, unsigned int, unsigned int);
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Compute reference data set
@@ -40,16 +39,13 @@ void computeGold(float *, const float *, const float *, unsigned int, unsigned i
 //! @param hA         height of matrix A
 //! @param wB         width of matrix B
 ////////////////////////////////////////////////////////////////////////////////
-void
+void computeGold(float *C, const float *A, const float *B, unsigned int hA, unsigned int wA, unsigned int wB)
-computeGold(float *C, const float *A, const float *B, unsigned int hA, unsigned int wA, unsigned int wB)
 {
     for (unsigned int i = 0; i < hA; ++i)
-        for (unsigned int j = 0; j < wB; ++j)
+        for (unsigned int j = 0; j < wB; ++j) {
-        {
             double sum = 0;
 
-            for (unsigned int k = 0; k < wA; ++k)
+            for (unsigned int k = 0; k < wA; ++k) {
-            {
                 double a = A[i * wA + k];
                 double b = B[k * wB + j];
                 sum += a * b;

Samples/0_Introduction/matrixMulDynlinkJIT/matrixMul_kernel_32_ptxdump.h

@@ -32,7 +32,8 @@
 #define __matrixMul_kernel_32_ptxdump_h__
 
 #if defined __cplusplus
-extern "C" {
+extern "C"
+{
 #endif
 
     extern unsigned char matrixMul_kernel_32_ptxdump[25784];

Samples/0_Introduction/matrixMulDynlinkJIT/matrixMul_kernel_64_ptxdump.h

@@ -32,7 +32,8 @@
 #define __matrixMul_kernel_64_ptxdump_h__
 
 #if defined __cplusplus
-extern "C" {
+extern "C"
+{
 #endif
 
     extern unsigned char matrixMul_kernel_64_ptxdump[26489];

Samples/0_Introduction/matrixMul_nvrtc/matrixMul.cpp

@@ -42,17 +42,19 @@
  */
 
 // System includes
-#include <stdio.h>
 #include <assert.h>
+#include <stdio.h>
 
 // CUDA runtime
 #include <cuda_runtime.h>
+
 #include "nvrtc_helper.h"
 
 // Helper functions and utilities to work with CUDA
 #include <helper_functions.h>
 
-void constantInit(float *data, int size, float val) {
+void constantInit(float *data, int size, float val)
+{
     for (int i = 0; i < size; ++i) {
         data[i] = val;
     }
@@ -61,8 +63,8 @@ void constantInit(float *data, int size, float val) {
 /**
  * Run a simple test of matrix multiplication using CUDA
  */
-int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
+int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA, dim3 &dimsB)
-                   dim3 &dimsB) {
+{
     // Allocate host memory for matrices A and B
     unsigned int size_A     = dimsA.x * dimsA.y;
     unsigned int mem_size_A = sizeof(float) * size_A;
@@ -114,24 +116,27 @@ int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
 
     CUfunction kernel_addr;
     if (block_size == 16) {
-    checkCudaErrors(
+        checkCudaErrors(cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block16"));
-        cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block16"));
+    }
-  } else {
+    else {
-    checkCudaErrors(
+        checkCudaErrors(cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block32"));
-        cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block32"));
     }
 
-  void *arr[] = {(void *)&d_C, (void *)&d_A, (void *)&d_B, (void *)&dimsA.x,
+    void *arr[] = {(void *)&d_C, (void *)&d_A, (void *)&d_B, (void *)&dimsA.x, (void *)&dimsB.x};
-                 (void *)&dimsB.x};
 
     // Execute the kernel
     int nIter = 300;
 
     for (int j = 0; j < nIter; j++) {
-    checkCudaErrors(
+        checkCudaErrors(cuLaunchKernel(kernel_addr,
-        cuLaunchKernel(kernel_addr, grid.x, grid.y, grid.z, /* grid dim */
+                                       grid.x,
-                       threads.x, threads.y, threads.z,     /* block dim */
+                                       grid.y,
-                       0, 0,    /* shared mem, stream */
+                                       grid.z, /* grid dim */
+                                       threads.x,
+                                       threads.y,
+                                       threads.z, /* block dim */
+                                       0,
+                                       0,       /* shared mem, stream */
                                        &arr[0], /* arguments */
                                        0));
 
@@ -157,16 +162,14 @@ int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
         double rel_err    = abs_err / abs_val / dot_length;
 
         if (rel_err > eps) {
-      printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n", i,
+            printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n", i, h_C[i], dimsA.x * valB, eps);
-             h_C[i], dimsA.x * valB, eps);
             correct = false;
         }
     }
 
     printf("%s\n", correct ? "Result = PASS" : "Result = FAIL");
 
-  printf(
+    printf("\nNOTE: The CUDA Samples are not meant for performance measurements. "
-      "\nNOTE: The CUDA Samples are not meant for performance measurements. "
            "Results may vary when GPU Boost is enabled.\n");
 
     // Clean up memory
@@ -180,7 +183,8 @@ int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
 
     if (correct) {
         return EXIT_SUCCESS;
-  } else {
+    }
+    else {
         return EXIT_FAILURE;
     }
 }
@@ -189,16 +193,15 @@ int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
  * Program main
  */
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("[Matrix Multiply Using CUDA] - Starting...\n");
 
-  if (checkCmdLineFlag(argc, (const char **)argv, "help") ||
+    if (checkCmdLineFlag(argc, (const char **)argv, "help") || checkCmdLineFlag(argc, (const char **)argv, "?")) {
-      checkCmdLineFlag(argc, (const char **)argv, "?")) {
         printf("Usage -device=n (n >= 0 for deviceID)\n");
         printf("      -wA=WidthA -hA=HeightA (Width x Height of Matrix A)\n");
         printf("      -wB=WidthB -hB=HeightB (Width x Height of Matrix B)\n");
-    printf(
+        printf("  Note: Outer matrix dimensions of A & B matrices must be equal.\n");
-        "  Note: Outer matrix dimensions of A & B matrices must be equal.\n");
 
         exit(EXIT_SUCCESS);
     }
@@ -234,13 +237,11 @@ int main(int argc, char **argv) {
     }
 
     if (dimsA.x != dimsB.y) {
-    printf("Error: outer matrix dimensions must be equal. (%d != %d)\n",
+        printf("Error: outer matrix dimensions must be equal. (%d != %d)\n", dimsA.x, dimsB.y);
-           dimsA.x, dimsB.y);
         exit(EXIT_FAILURE);
     }
 
-  printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y, dimsB.x,
+    printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y, dimsB.x, dimsB.y);
-         dimsB.y);
 
     int matrix_result = matrixMultiply(argc, argv, block_size, dimsA, dimsB);
 

Samples/0_Introduction/matrixMul_nvrtc/matrixMul_kernel.cu

@@ -48,11 +48,10 @@
 
 #include <cooperative_groups.h>
 
-template <int BLOCK_SIZE>
+template <int BLOCK_SIZE> __device__ void matrixMulCUDA(float *C, float *A, float *B, int wA, int wB)
-__device__ void matrixMulCUDA(float *C, float *A, float *B, int wA, int wB) {
+{
     // Handle to thread block group
-  cooperative_groups::thread_block cta =
+    cooperative_groups::thread_block cta = cooperative_groups::this_thread_block();
-      cooperative_groups::this_thread_block();
     // Block index
     int bx = blockIdx.x;
     int by = blockIdx.y;
@@ -120,12 +119,12 @@ __device__ void matrixMulCUDA(float *C, float *A, float *B, int wA, int wB) {
     C[c + wB * ty + tx] = Csub;
 }
 
-extern "C" __global__ void matrixMulCUDA_block16(float *C, float *A, float *B,
+extern "C" __global__ void matrixMulCUDA_block16(float *C, float *A, float *B, int wA, int wB)
-                                                 int wA, int wB) {
+{
     matrixMulCUDA<16>(C, A, B, wA, wB);
 }
 
-extern "C" __global__ void matrixMulCUDA_block32(float *C, float *A, float *B,
+extern "C" __global__ void matrixMulCUDA_block32(float *C, float *A, float *B, int wA, int wB)
-                                                 int wA, int wB) {
+{
     matrixMulCUDA<32>(C, A, B, wA, wB);
 }

Samples/0_Introduction/mergeSort/bitonic.cu

@@ -28,12 +28,13 @@
 #include <cooperative_groups.h>
 
 namespace cg = cooperative_groups;
-#include <helper_cuda.h>
 #include <assert.h>
+#include <helper_cuda.h>
+
 #include "mergeSort_common.h"
 
-inline __device__ void Comparator(uint &keyA, uint &valA, uint &keyB,
+inline __device__ void Comparator(uint &keyA, uint &valA, uint &keyB, uint &valB, uint arrowDir)
-                                  uint &valB, uint arrowDir) {
+{
     uint t;
 
     if ((keyA > keyB) == arrowDir) {
@@ -46,9 +47,9 @@ inline __device__ void Comparator(uint &keyA, uint &valA, uint &keyB,
     }
 }
 
-__global__ void bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
+__global__ void
-                                        uint *d_SrcKey, uint *d_SrcVal,
+bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey, uint *d_SrcVal, uint arrayLength, uint sortDir)
-                                        uint arrayLength, uint sortDir) {
+{
     // Handle to thread block group
     cg::thread_block cta = cg::this_thread_block();
     // Shared memory storage for one or more short vectors
@@ -62,10 +63,8 @@ __global__ void bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
     d_DstVal += blockIdx.x * SHARED_SIZE_LIMIT + threadIdx.x;
     s_key[threadIdx.x + 0]                       = d_SrcKey[0];
     s_val[threadIdx.x + 0]                       = d_SrcVal[0];
-  s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] =
+    s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] = d_SrcKey[(SHARED_SIZE_LIMIT / 2)];
-      d_SrcKey[(SHARED_SIZE_LIMIT / 2)];
+    s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] = d_SrcVal[(SHARED_SIZE_LIMIT / 2)];
-  s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] =
-      d_SrcVal[(SHARED_SIZE_LIMIT / 2)];
 
     for (uint size = 2; size < arrayLength; size <<= 1) {
         // Bitonic merge
@@ -74,8 +73,7 @@ __global__ void bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
         for (uint stride = size / 2; stride > 0; stride >>= 1) {
             cg::sync(cta);
             uint pos = 2 * threadIdx.x - (threadIdx.x & (stride - 1));
-      Comparator(s_key[pos + 0], s_val[pos + 0], s_key[pos + stride],
+            Comparator(s_key[pos + 0], s_val[pos + 0], s_key[pos + stride], s_val[pos + stride], dir);
-                 s_val[pos + stride], dir);
         }
     }
 
@@ -84,26 +82,25 @@ __global__ void bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
         for (uint stride = arrayLength / 2; stride > 0; stride >>= 1) {
             cg::sync(cta);
             uint pos = 2 * threadIdx.x - (threadIdx.x & (stride - 1));
-      Comparator(s_key[pos + 0], s_val[pos + 0], s_key[pos + stride],
+            Comparator(s_key[pos + 0], s_val[pos + 0], s_key[pos + stride], s_val[pos + stride], sortDir);
-                 s_val[pos + stride], sortDir);
         }
     }
 
     cg::sync(cta);
     d_DstKey[0]                       = s_key[threadIdx.x + 0];
     d_DstVal[0]                       = s_val[threadIdx.x + 0];
-  d_DstKey[(SHARED_SIZE_LIMIT / 2)] =
+    d_DstKey[(SHARED_SIZE_LIMIT / 2)] = s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
-      s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
+    d_DstVal[(SHARED_SIZE_LIMIT / 2)] = s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
-  d_DstVal[(SHARED_SIZE_LIMIT / 2)] =
-      s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
 }
 
 // Helper function (also used by odd-even merge sort)
-extern "C" uint factorRadix2(uint *log2L, uint L) {
+extern "C" uint factorRadix2(uint *log2L, uint L)
+{
     if (!L) {
         *log2L = 0;
         return 0;
-  } else {
+    }
+    else {
         for (*log2L = 0; (L & 1) == 0; L >>= 1, *log2L++)
             ;
 
@@ -111,10 +108,14 @@ extern "C" uint factorRadix2(uint *log2L, uint L) {
     }
 }
 
-extern "C" void bitonicSortShared(uint *d_DstKey, uint *d_DstVal,
+extern "C" void bitonicSortShared(uint *d_DstKey,
-                                  uint *d_SrcKey, uint *d_SrcVal,
+                                  uint *d_DstVal,
-                                  uint batchSize, uint arrayLength,
+                                  uint *d_SrcKey,
-                                  uint sortDir) {
+                                  uint *d_SrcVal,
+                                  uint  batchSize,
+                                  uint  arrayLength,
+                                  uint  sortDir)
+{
     // Nothing to sort
     if (arrayLength < 2) {
         return;
@@ -131,32 +132,25 @@ extern "C" void bitonicSortShared(uint *d_DstKey, uint *d_DstVal,
     assert(arrayLength <= SHARED_SIZE_LIMIT);
     assert((batchSize * arrayLength) % SHARED_SIZE_LIMIT == 0);
 
-  bitonicSortSharedKernel<<<blockCount, threadCount>>>(
+    bitonicSortSharedKernel<<<blockCount, threadCount>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength, sortDir);
-      d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength, sortDir);
     getLastCudaError("bitonicSortSharedKernel<<<>>> failed!\n");
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 3: merge elementary intervals
 ////////////////////////////////////////////////////////////////////////////////
-static inline __host__ __device__ uint iDivUp(uint a, uint b) {
+static inline __host__ __device__ uint iDivUp(uint a, uint b) { return ((a % b) == 0) ? (a / b) : (a / b + 1); }
-  return ((a % b) == 0) ? (a / b) : (a / b + 1);
-}
 
-static inline __host__ __device__ uint getSampleCount(uint dividend) {
+static inline __host__ __device__ uint getSampleCount(uint dividend) { return iDivUp(dividend, SAMPLE_STRIDE); }
-  return iDivUp(dividend, SAMPLE_STRIDE);
-}
 
 template <uint sortDir>
-static inline __device__ void ComparatorExtended(uint &keyA, uint &valA,
+static inline __device__ void
-                                                 uint &flagA, uint &keyB,
+ComparatorExtended(uint &keyA, uint &valA, uint &flagA, uint &keyB, uint &valB, uint &flagB, uint arrowDir)
-                                                 uint &valB, uint &flagB,
+{
-                                                 uint arrowDir) {
     uint t;
 
-  if ((!(flagA || flagB) && ((keyA > keyB) == arrowDir)) ||
+    if ((!(flagA || flagB) && ((keyA > keyB) == arrowDir)) || ((arrowDir == sortDir) && (flagA == 1))
-      ((arrowDir == sortDir) && (flagA == 1)) ||
+        || ((arrowDir != sortDir) && (flagB == 1))) {
-      ((arrowDir != sortDir) && (flagB == 1))) {
         t     = keyA;
         keyA  = keyB;
         keyB  = t;
@@ -170,9 +164,15 @@ static inline __device__ void ComparatorExtended(uint &keyA, uint &valA,
 }
 
 template <uint sortDir>
-__global__ void bitonicMergeElementaryIntervalsKernel(
+__global__ void bitonicMergeElementaryIntervalsKernel(uint *d_DstKey,
-    uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey, uint *d_SrcVal,
+                                                      uint *d_DstVal,
-    uint *d_LimitsA, uint *d_LimitsB, uint stride, uint N) {
+                                                      uint *d_SrcKey,
+                                                      uint *d_SrcVal,
+                                                      uint *d_LimitsA,
+                                                      uint *d_LimitsB,
+                                                      uint  stride,
+                                                      uint  N)
+{
     // Handle to thread block group
     cg::thread_block cta = cg::this_thread_block();
     __shared__ uint  s_key[2 * SAMPLE_STRIDE];
@@ -200,10 +200,8 @@ __global__ void bitonicMergeElementaryIntervalsKernel(
         startSrcB = d_LimitsB[blockIdx.x];
         startDst  = startSrcA + startSrcB;
 
-    uint endSrcA = (intervalI + 1 < segmentSamples) ? d_LimitsA[blockIdx.x + 1]
+        uint endSrcA = (intervalI + 1 < segmentSamples) ? d_LimitsA[blockIdx.x + 1] : segmentElementsA;
-                                                    : segmentElementsA;
+        uint endSrcB = (intervalI + 1 < segmentSamples) ? d_LimitsB[blockIdx.x + 1] : segmentElementsB;
-    uint endSrcB = (intervalI + 1 < segmentSamples) ? d_LimitsB[blockIdx.x + 1]
-                                                    : segmentElementsB;
         lenSrcA      = endSrcA - startSrcA;
         lenSrcB      = endSrcB - startSrcB;
     }
@@ -222,10 +220,8 @@ __global__ void bitonicMergeElementaryIntervalsKernel(
 
     // Prepare for bitonic merge by inversing the ordering
     if (threadIdx.x < lenSrcB) {
-    s_key[2 * SAMPLE_STRIDE - 1 - threadIdx.x] =
+        s_key[2 * SAMPLE_STRIDE - 1 - threadIdx.x] = d_SrcKey[stride + startSrcB + threadIdx.x];
-        d_SrcKey[stride + startSrcB + threadIdx.x];
+        s_val[2 * SAMPLE_STRIDE - 1 - threadIdx.x] = d_SrcVal[stride + startSrcB + threadIdx.x];
-    s_val[2 * SAMPLE_STRIDE - 1 - threadIdx.x] =
-        d_SrcVal[stride + startSrcB + threadIdx.x];
         s_inf[2 * SAMPLE_STRIDE - 1 - threadIdx.x] = 0;
     }
 
@@ -233,9 +229,13 @@ __global__ void bitonicMergeElementaryIntervalsKernel(
     for (uint stride = SAMPLE_STRIDE; stride > 0; stride >>= 1) {
         cg::sync(cta);
         uint pos = 2 * threadIdx.x - (threadIdx.x & (stride - 1));
-    ComparatorExtended<sortDir>(s_key[pos + 0], s_val[pos + 0], s_inf[pos + 0],
+        ComparatorExtended<sortDir>(s_key[pos + 0],
-                                s_key[pos + stride], s_val[pos + stride],
+                                    s_val[pos + 0],
-                                s_inf[pos + stride], sortDir);
+                                    s_inf[pos + 0],
+                                    s_key[pos + stride],
+                                    s_val[pos + stride],
+                                    s_inf[pos + stride],
+                                    sortDir);
     }
 
     // Store sorted data
@@ -254,26 +254,28 @@ __global__ void bitonicMergeElementaryIntervalsKernel(
     }
 }
 
-extern "C" void bitonicMergeElementaryIntervals(uint *d_DstKey, uint *d_DstVal,
+extern "C" void bitonicMergeElementaryIntervals(uint *d_DstKey,
-                                                uint *d_SrcKey, uint *d_SrcVal,
+                                                uint *d_DstVal,
+                                                uint *d_SrcKey,
+                                                uint *d_SrcVal,
                                                 uint *d_LimitsA,
-                                                uint *d_LimitsB, uint stride,
+                                                uint *d_LimitsB,
-                                                uint N, uint sortDir) {
+                                                uint  stride,
+                                                uint  N,
+                                                uint  sortDir)
+{
     uint lastSegmentElements = N % (2 * stride);
 
-  uint mergePairs = (lastSegmentElements > stride)
+    uint mergePairs = (lastSegmentElements > stride) ? getSampleCount(N) : (N - lastSegmentElements) / SAMPLE_STRIDE;
-                        ? getSampleCount(N)
-                        : (N - lastSegmentElements) / SAMPLE_STRIDE;
 
     if (sortDir) {
-    bitonicMergeElementaryIntervalsKernel<1U><<<mergePairs, SAMPLE_STRIDE>>>(
+        bitonicMergeElementaryIntervalsKernel<1U>
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride,
+            <<<mergePairs, SAMPLE_STRIDE>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride, N);
-        N);
         getLastCudaError("mergeElementaryIntervalsKernel<1> failed\n");
-  } else {
+    }
-    bitonicMergeElementaryIntervalsKernel<0U><<<mergePairs, SAMPLE_STRIDE>>>(
+    else {
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride,
+        bitonicMergeElementaryIntervalsKernel<0U>
-        N);
+            <<<mergePairs, SAMPLE_STRIDE>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride, N);
         getLastCudaError("mergeElementaryIntervalsKernel<0> failed\n");
     }
 }

Samples/0_Introduction/mergeSort/main.cpp

@@ -26,17 +26,19 @@
  */
 
 #include <assert.h>
-#include <stdio.h>
-#include <stdlib.h>
 #include <cuda_runtime.h>
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
+#include <stdio.h>
+#include <stdlib.h>
+
 #include "mergeSort_common.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Test driver
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     uint               *h_SrcKey, *h_SrcVal, *h_DstKey, *h_DstVal;
     uint               *d_SrcKey, *d_SrcVal, *d_BufKey, *d_BufVal, *d_DstKey, *d_DstVal;
     StopWatchInterface *hTimer = NULL;
@@ -75,10 +77,8 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaMalloc((void **)&d_BufVal, N * sizeof(uint)));
     checkCudaErrors(cudaMalloc((void **)&d_SrcKey, N * sizeof(uint)));
     checkCudaErrors(cudaMalloc((void **)&d_SrcVal, N * sizeof(uint)));
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpy(d_SrcKey, h_SrcKey, N * sizeof(uint), cudaMemcpyHostToDevice));
-      cudaMemcpy(d_SrcKey, h_SrcKey, N * sizeof(uint), cudaMemcpyHostToDevice));
+    checkCudaErrors(cudaMemcpy(d_SrcVal, h_SrcVal, N * sizeof(uint), cudaMemcpyHostToDevice));
-  checkCudaErrors(
-      cudaMemcpy(d_SrcVal, h_SrcVal, N * sizeof(uint), cudaMemcpyHostToDevice));
 
     printf("Initializing GPU merge sort...\n");
     initMergeSort();
@@ -93,10 +93,8 @@ int main(int argc, char **argv) {
     printf("Time: %f ms\n", sdkGetTimerValue(&hTimer));
 
     printf("Reading back GPU merge sort results...\n");
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpy(h_DstKey, d_DstKey, N * sizeof(uint), cudaMemcpyDeviceToHost));
-      cudaMemcpy(h_DstKey, d_DstKey, N * sizeof(uint), cudaMemcpyDeviceToHost));
+    checkCudaErrors(cudaMemcpy(h_DstVal, d_DstVal, N * sizeof(uint), cudaMemcpyDeviceToHost));
-  checkCudaErrors(
-      cudaMemcpy(h_DstVal, d_DstVal, N * sizeof(uint), cudaMemcpyDeviceToHost));
 
     printf("Inspecting the results...\n");
     uint keysFlag = validateSortedKeys(h_DstKey, h_SrcKey, 1, N, numValues, DIR);

Samples/0_Introduction/mergeSort/mergeSort.cu

@@ -39,21 +39,19 @@
 namespace cg = cooperative_groups;
 
 #include <helper_cuda.h>
+
 #include "mergeSort_common.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Helper functions
 ////////////////////////////////////////////////////////////////////////////////
-static inline __host__ __device__ uint iDivUp(uint a, uint b) {
+static inline __host__ __device__ uint iDivUp(uint a, uint b) { return ((a % b) == 0) ? (a / b) : (a / b + 1); }
-  return ((a % b) == 0) ? (a / b) : (a / b + 1);
-}
 
-static inline __host__ __device__ uint getSampleCount(uint dividend) {
+static inline __host__ __device__ uint getSampleCount(uint dividend) { return iDivUp(dividend, SAMPLE_STRIDE); }
-  return iDivUp(dividend, SAMPLE_STRIDE);
-}
 
 #define W (sizeof(uint) * 8)
-static inline __device__ uint nextPowerOfTwo(uint x) {
+static inline __device__ uint nextPowerOfTwo(uint x)
+{
     /*
         --x;
         x |= x >> 1;
@@ -66,9 +64,8 @@ static inline __device__ uint nextPowerOfTwo(uint x) {
     return 1U << (W - __clz(x - 1));
 }
 
-template <uint sortDir>
+template <uint sortDir> static inline __device__ uint binarySearchInclusive(uint val, uint *data, uint L, uint stride)
-static inline __device__ uint binarySearchInclusive(uint val, uint *data,
+{
-                                                    uint L, uint stride) {
     if (L == 0) {
         return 0;
     }
@@ -78,8 +75,7 @@ static inline __device__ uint binarySearchInclusive(uint val, uint *data,
     for (; stride > 0; stride >>= 1) {
         uint newPos = umin(pos + stride, L);
 
-    if ((sortDir && (data[newPos - 1] <= val)) ||
+        if ((sortDir && (data[newPos - 1] <= val)) || (!sortDir && (data[newPos - 1] >= val))) {
-        (!sortDir && (data[newPos - 1] >= val))) {
             pos = newPos;
         }
     }
@@ -87,9 +83,8 @@ static inline __device__ uint binarySearchInclusive(uint val, uint *data,
     return pos;
 }
 
-template <uint sortDir>
+template <uint sortDir> static inline __device__ uint binarySearchExclusive(uint val, uint *data, uint L, uint stride)
-static inline __device__ uint binarySearchExclusive(uint val, uint *data,
+{
-                                                    uint L, uint stride) {
     if (L == 0) {
         return 0;
     }
@@ -99,8 +94,7 @@ static inline __device__ uint binarySearchExclusive(uint val, uint *data,
     for (; stride > 0; stride >>= 1) {
         uint newPos = umin(pos + stride, L);
 
-    if ((sortDir && (data[newPos - 1] < val)) ||
+        if ((sortDir && (data[newPos - 1] < val)) || (!sortDir && (data[newPos - 1] > val))) {
-        (!sortDir && (data[newPos - 1] > val))) {
             pos = newPos;
         }
     }
@@ -112,9 +106,8 @@ static inline __device__ uint binarySearchExclusive(uint val, uint *data,
 // Bottom-level merge sort (binary search-based)
 ////////////////////////////////////////////////////////////////////////////////
 template <uint sortDir>
-__global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
+__global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey, uint *d_SrcVal, uint arrayLength)
-                                      uint *d_SrcKey, uint *d_SrcVal,
+{
-                                      uint arrayLength) {
     // Handle to thread block group
     cg::thread_block cta = cg::this_thread_block();
     __shared__ uint  s_key[SHARED_SIZE_LIMIT];
@@ -126,10 +119,8 @@ __global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
     d_DstVal += blockIdx.x * SHARED_SIZE_LIMIT + threadIdx.x;
     s_key[threadIdx.x + 0]                       = d_SrcKey[0];
     s_val[threadIdx.x + 0]                       = d_SrcVal[0];
-  s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] =
+    s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] = d_SrcKey[(SHARED_SIZE_LIMIT / 2)];
-      d_SrcKey[(SHARED_SIZE_LIMIT / 2)];
+    s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] = d_SrcVal[(SHARED_SIZE_LIMIT / 2)];
-  s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] =
-      d_SrcVal[(SHARED_SIZE_LIMIT / 2)];
 
     for (uint stride = 1; stride < arrayLength; stride <<= 1) {
         uint  lPos    = threadIdx.x & (stride - 1);
@@ -141,12 +132,8 @@ __global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
         uint valA = baseVal[lPos + 0];
         uint keyB = baseKey[lPos + stride];
         uint valB = baseVal[lPos + stride];
-    uint posA =
+        uint posA = binarySearchExclusive<sortDir>(keyA, baseKey + stride, stride, stride) + lPos;
-        binarySearchExclusive<sortDir>(keyA, baseKey + stride, stride, stride) +
+        uint posB = binarySearchInclusive<sortDir>(keyB, baseKey + 0, stride, stride) + lPos;
-        lPos;
-    uint posB =
-        binarySearchInclusive<sortDir>(keyB, baseKey + 0, stride, stride) +
-        lPos;
 
         cg::sync(cta);
         baseKey[posA] = keyA;
@@ -158,15 +145,18 @@ __global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
     cg::sync(cta);
     d_DstKey[0]                       = s_key[threadIdx.x + 0];
     d_DstVal[0]                       = s_val[threadIdx.x + 0];
-  d_DstKey[(SHARED_SIZE_LIMIT / 2)] =
+    d_DstKey[(SHARED_SIZE_LIMIT / 2)] = s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
-      s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
+    d_DstVal[(SHARED_SIZE_LIMIT / 2)] = s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
-  d_DstVal[(SHARED_SIZE_LIMIT / 2)] =
+}
-      s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
+
-}
+static void mergeSortShared(uint *d_DstKey,
-
+                            uint *d_DstVal,
-static void mergeSortShared(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey,
+                            uint *d_SrcKey,
-                            uint *d_SrcVal, uint batchSize, uint arrayLength,
+                            uint *d_SrcVal,
-                            uint sortDir) {
+                            uint  batchSize,
+                            uint  arrayLength,
+                            uint  sortDir)
+{
     if (arrayLength < 2) {
         return;
     }
@@ -177,12 +167,11 @@ static void mergeSortShared(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey,
     uint threadCount = SHARED_SIZE_LIMIT / 2;
 
     if (sortDir) {
-    mergeSortSharedKernel<1U><<<blockCount, threadCount>>>(
+        mergeSortSharedKernel<1U><<<blockCount, threadCount>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength);
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength);
         getLastCudaError("mergeSortShared<1><<<>>> failed\n");
-  } else {
+    }
-    mergeSortSharedKernel<0U><<<blockCount, threadCount>>>(
+    else {
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength);
+        mergeSortSharedKernel<0U><<<blockCount, threadCount>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength);
         getLastCudaError("mergeSortShared<0><<<>>> failed\n");
     }
 }
@@ -191,9 +180,9 @@ static void mergeSortShared(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey,
 // Merge step 1: generate sample ranks
 ////////////////////////////////////////////////////////////////////////////////
 template <uint sortDir>
-__global__ void generateSampleRanksKernel(uint *d_RanksA, uint *d_RanksB,
+__global__ void
-                                          uint *d_SrcKey, uint stride, uint N,
+generateSampleRanksKernel(uint *d_RanksA, uint *d_RanksB, uint *d_SrcKey, uint stride, uint N, uint threadCount)
-                                          uint threadCount) {
+{
     uint pos = blockIdx.x * blockDim.x + threadIdx.x;
 
     if (pos >= threadCount) {
@@ -214,33 +203,30 @@ __global__ void generateSampleRanksKernel(uint *d_RanksA, uint *d_RanksB,
     if (i < segmentSamplesA) {
         d_RanksA[i] = i * SAMPLE_STRIDE;
         d_RanksB[i] = binarySearchExclusive<sortDir>(
-        d_SrcKey[i * SAMPLE_STRIDE], d_SrcKey + stride, segmentElementsB,
+            d_SrcKey[i * SAMPLE_STRIDE], d_SrcKey + stride, segmentElementsB, nextPowerOfTwo(segmentElementsB));
-        nextPowerOfTwo(segmentElementsB));
     }
 
     if (i < segmentSamplesB) {
         d_RanksB[(stride / SAMPLE_STRIDE) + i] = i * SAMPLE_STRIDE;
         d_RanksA[(stride / SAMPLE_STRIDE) + i] = binarySearchInclusive<sortDir>(
-        d_SrcKey[stride + i * SAMPLE_STRIDE], d_SrcKey + 0, segmentElementsA,
+            d_SrcKey[stride + i * SAMPLE_STRIDE], d_SrcKey + 0, segmentElementsA, nextPowerOfTwo(segmentElementsA));
-        nextPowerOfTwo(segmentElementsA));
     }
 }
 
-static void generateSampleRanks(uint *d_RanksA, uint *d_RanksB, uint *d_SrcKey,
+static void generateSampleRanks(uint *d_RanksA, uint *d_RanksB, uint *d_SrcKey, uint stride, uint N, uint sortDir)
-                                uint stride, uint N, uint sortDir) {
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint threadCount =
+    uint threadCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
-      (lastSegmentElements > stride)
-          ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
                                                       : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
 
     if (sortDir) {
-    generateSampleRanksKernel<1U><<<iDivUp(threadCount, 256), 256>>>(
+        generateSampleRanksKernel<1U>
-        d_RanksA, d_RanksB, d_SrcKey, stride, N, threadCount);
+            <<<iDivUp(threadCount, 256), 256>>>(d_RanksA, d_RanksB, d_SrcKey, stride, N, threadCount);
         getLastCudaError("generateSampleRanksKernel<1U><<<>>> failed\n");
-  } else {
+    }
-    generateSampleRanksKernel<0U><<<iDivUp(threadCount, 256), 256>>>(
+    else {
-        d_RanksA, d_RanksB, d_SrcKey, stride, N, threadCount);
+        generateSampleRanksKernel<0U>
+            <<<iDivUp(threadCount, 256), 256>>>(d_RanksA, d_RanksB, d_SrcKey, stride, N, threadCount);
         getLastCudaError("generateSampleRanksKernel<0U><<<>>> failed\n");
     }
 }
@@ -248,9 +234,8 @@ static void generateSampleRanks(uint *d_RanksA, uint *d_RanksB, uint *d_SrcKey,
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 2: generate sample ranks and indices
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void mergeRanksAndIndicesKernel(uint *d_Limits, uint *d_Ranks,
+__global__ void mergeRanksAndIndicesKernel(uint *d_Limits, uint *d_Ranks, uint stride, uint N, uint threadCount)
-                                           uint stride, uint N,
+{
-                                           uint threadCount) {
     uint pos = blockIdx.x * blockDim.x + threadIdx.x;
 
     if (pos >= threadCount) {
@@ -269,36 +254,29 @@ __global__ void mergeRanksAndIndicesKernel(uint *d_Limits, uint *d_Ranks,
 
     if (i < segmentSamplesA) {
         uint dstPos = binarySearchExclusive<1U>(
-                      d_Ranks[i], d_Ranks + segmentSamplesA, segmentSamplesB,
+                          d_Ranks[i], d_Ranks + segmentSamplesA, segmentSamplesB, nextPowerOfTwo(segmentSamplesB))
-                      nextPowerOfTwo(segmentSamplesB)) +
+                    + i;
-                  i;
         d_Limits[dstPos] = d_Ranks[i];
     }
 
     if (i < segmentSamplesB) {
-    uint dstPos = binarySearchInclusive<1U>(d_Ranks[segmentSamplesA + i],
+        uint dstPos = binarySearchInclusive<1U>(
-                                            d_Ranks, segmentSamplesA,
+                          d_Ranks[segmentSamplesA + i], d_Ranks, segmentSamplesA, nextPowerOfTwo(segmentSamplesA))
-                                            nextPowerOfTwo(segmentSamplesA)) +
+                    + i;
-                  i;
         d_Limits[dstPos] = d_Ranks[segmentSamplesA + i];
     }
 }
 
-static void mergeRanksAndIndices(uint *d_LimitsA, uint *d_LimitsB,
+static void mergeRanksAndIndices(uint *d_LimitsA, uint *d_LimitsB, uint *d_RanksA, uint *d_RanksB, uint stride, uint N)
-                                 uint *d_RanksA, uint *d_RanksB, uint stride,
+{
-                                 uint N) {
     uint lastSegmentElements = N % (2 * stride);
-  uint threadCount =
+    uint threadCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
-      (lastSegmentElements > stride)
-          ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
                                                       : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
 
-  mergeRanksAndIndicesKernel<<<iDivUp(threadCount, 256), 256>>>(
+    mergeRanksAndIndicesKernel<<<iDivUp(threadCount, 256), 256>>>(d_LimitsA, d_RanksA, stride, N, threadCount);
-      d_LimitsA, d_RanksA, stride, N, threadCount);
     getLastCudaError("mergeRanksAndIndicesKernel(A)<<<>>> failed\n");
 
-  mergeRanksAndIndicesKernel<<<iDivUp(threadCount, 256), 256>>>(
+    mergeRanksAndIndicesKernel<<<iDivUp(threadCount, 256), 256>>>(d_LimitsB, d_RanksB, stride, N, threadCount);
-      d_LimitsB, d_RanksB, stride, N, threadCount);
     getLastCudaError("mergeRanksAndIndicesKernel(B)<<<>>> failed\n");
 }
 
@@ -306,24 +284,30 @@ static void mergeRanksAndIndices(uint *d_LimitsA, uint *d_LimitsB,
 // Merge step 3: merge elementary intervals
 ////////////////////////////////////////////////////////////////////////////////
 template <uint sortDir>
-inline __device__ void merge(uint *dstKey, uint *dstVal, uint *srcAKey,
+inline __device__ void merge(uint            *dstKey,
-                             uint *srcAVal, uint *srcBKey, uint *srcBVal,
+                             uint            *dstVal,
-                             uint lenA, uint nPowTwoLenA, uint lenB,
+                             uint            *srcAKey,
-                             uint nPowTwoLenB, cg::thread_block cta) {
+                             uint            *srcAVal,
+                             uint            *srcBKey,
+                             uint            *srcBVal,
+                             uint             lenA,
+                             uint             nPowTwoLenA,
+                             uint             lenB,
+                             uint             nPowTwoLenB,
+                             cg::thread_block cta)
+{
     uint keyA, valA, keyB, valB, dstPosA, dstPosB;
 
     if (threadIdx.x < lenA) {
         keyA    = srcAKey[threadIdx.x];
         valA    = srcAVal[threadIdx.x];
-    dstPosA = binarySearchExclusive<sortDir>(keyA, srcBKey, lenB, nPowTwoLenB) +
+        dstPosA = binarySearchExclusive<sortDir>(keyA, srcBKey, lenB, nPowTwoLenB) + threadIdx.x;
-              threadIdx.x;
     }
 
     if (threadIdx.x < lenB) {
         keyB    = srcBKey[threadIdx.x];
         valB    = srcBVal[threadIdx.x];
-    dstPosB = binarySearchInclusive<sortDir>(keyB, srcAKey, lenA, nPowTwoLenA) +
+        dstPosB = binarySearchInclusive<sortDir>(keyB, srcAKey, lenA, nPowTwoLenA) + threadIdx.x;
-              threadIdx.x;
     }
 
     cg::sync(cta);
@@ -340,10 +324,15 @@ inline __device__ void merge(uint *dstKey, uint *dstVal, uint *srcAKey,
 }
 
 template <uint sortDir>
-__global__ void mergeElementaryIntervalsKernel(uint *d_DstKey, uint *d_DstVal,
+__global__ void mergeElementaryIntervalsKernel(uint *d_DstKey,
-                                               uint *d_SrcKey, uint *d_SrcVal,
+                                               uint *d_DstVal,
-                                               uint *d_LimitsA, uint *d_LimitsB,
+                                               uint *d_SrcKey,
-                                               uint stride, uint N) {
+                                               uint *d_SrcVal,
+                                               uint *d_LimitsA,
+                                               uint *d_LimitsB,
+                                               uint  stride,
+                                               uint  N)
+{
     // Handle to thread block group
     cg::thread_block cta = cg::this_thread_block();
     __shared__ uint  s_key[2 * SAMPLE_STRIDE];
@@ -368,10 +357,8 @@ __global__ void mergeElementaryIntervalsKernel(uint *d_DstKey, uint *d_DstVal,
 
         startSrcA    = d_LimitsA[blockIdx.x];
         startSrcB    = d_LimitsB[blockIdx.x];
-    uint endSrcA = (intervalI + 1 < segmentSamples) ? d_LimitsA[blockIdx.x + 1]
+        uint endSrcA = (intervalI + 1 < segmentSamples) ? d_LimitsA[blockIdx.x + 1] : segmentElementsA;
-                                                    : segmentElementsA;
+        uint endSrcB = (intervalI + 1 < segmentSamples) ? d_LimitsB[blockIdx.x + 1] : segmentElementsB;
-    uint endSrcB = (intervalI + 1 < segmentSamples) ? d_LimitsB[blockIdx.x + 1]
-                                                    : segmentElementsB;
         lenSrcA      = endSrcA - startSrcA;
         lenSrcB      = endSrcB - startSrcB;
         startDstA    = startSrcA + startSrcB;
@@ -387,17 +374,23 @@ __global__ void mergeElementaryIntervalsKernel(uint *d_DstKey, uint *d_DstVal,
     }
 
     if (threadIdx.x < lenSrcB) {
-    s_key[threadIdx.x + SAMPLE_STRIDE] =
+        s_key[threadIdx.x + SAMPLE_STRIDE] = d_SrcKey[stride + startSrcB + threadIdx.x];
-        d_SrcKey[stride + startSrcB + threadIdx.x];
+        s_val[threadIdx.x + SAMPLE_STRIDE] = d_SrcVal[stride + startSrcB + threadIdx.x];
-    s_val[threadIdx.x + SAMPLE_STRIDE] =
-        d_SrcVal[stride + startSrcB + threadIdx.x];
     }
 
     // Merge data in shared memory
     cg::sync(cta);
-  merge<sortDir>(s_key, s_val, s_key + 0, s_val + 0, s_key + SAMPLE_STRIDE,
+    merge<sortDir>(s_key,
-                 s_val + SAMPLE_STRIDE, lenSrcA, SAMPLE_STRIDE, lenSrcB,
+                   s_val,
-                 SAMPLE_STRIDE, cta);
+                   s_key + 0,
+                   s_val + 0,
+                   s_key + SAMPLE_STRIDE,
+                   s_val + SAMPLE_STRIDE,
+                   lenSrcA,
+                   SAMPLE_STRIDE,
+                   lenSrcB,
+                   SAMPLE_STRIDE,
+                   cta);
 
     // Store merged data
     cg::sync(cta);
@@ -413,63 +406,77 @@ __global__ void mergeElementaryIntervalsKernel(uint *d_DstKey, uint *d_DstVal,
     }
 }
 
-static void mergeElementaryIntervals(uint *d_DstKey, uint *d_DstVal,
+static void mergeElementaryIntervals(uint *d_DstKey,
-                                     uint *d_SrcKey, uint *d_SrcVal,
+                                     uint *d_DstVal,
-                                     uint *d_LimitsA, uint *d_LimitsB,
+                                     uint *d_SrcKey,
-                                     uint stride, uint N, uint sortDir) {
+                                     uint *d_SrcVal,
+                                     uint *d_LimitsA,
+                                     uint *d_LimitsB,
+                                     uint  stride,
+                                     uint  N,
+                                     uint  sortDir)
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint mergePairs = (lastSegmentElements > stride)
+    uint mergePairs = (lastSegmentElements > stride) ? getSampleCount(N) : (N - lastSegmentElements) / SAMPLE_STRIDE;
-                        ? getSampleCount(N)
-                        : (N - lastSegmentElements) / SAMPLE_STRIDE;
 
     if (sortDir) {
-    mergeElementaryIntervalsKernel<1U><<<mergePairs, SAMPLE_STRIDE>>>(
+        mergeElementaryIntervalsKernel<1U>
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride,
+            <<<mergePairs, SAMPLE_STRIDE>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride, N);
-        N);
         getLastCudaError("mergeElementaryIntervalsKernel<1> failed\n");
-  } else {
+    }
-    mergeElementaryIntervalsKernel<0U><<<mergePairs, SAMPLE_STRIDE>>>(
+    else {
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride,
+        mergeElementaryIntervalsKernel<0U>
-        N);
+            <<<mergePairs, SAMPLE_STRIDE>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride, N);
         getLastCudaError("mergeElementaryIntervalsKernel<0> failed\n");
     }
 }
 
-extern "C" void bitonicSortShared(uint *d_DstKey, uint *d_DstVal,
+extern "C" void bitonicSortShared(uint *d_DstKey,
-                                  uint *d_SrcKey, uint *d_SrcVal,
+                                  uint *d_DstVal,
-                                  uint batchSize, uint arrayLength,
+                                  uint *d_SrcKey,
+                                  uint *d_SrcVal,
+                                  uint  batchSize,
+                                  uint  arrayLength,
                                   uint  sortDir);
 
-extern "C" void bitonicMergeElementaryIntervals(uint *d_DstKey, uint *d_DstVal,
+extern "C" void bitonicMergeElementaryIntervals(uint *d_DstKey,
-                                                uint *d_SrcKey, uint *d_SrcVal,
+                                                uint *d_DstVal,
+                                                uint *d_SrcKey,
+                                                uint *d_SrcVal,
                                                 uint *d_LimitsA,
-                                                uint *d_LimitsB, uint stride,
+                                                uint *d_LimitsB,
-                                                uint N, uint sortDir);
+                                                uint  stride,
+                                                uint  N,
+                                                uint  sortDir);
 
 static uint      *d_RanksA, *d_RanksB, *d_LimitsA, *d_LimitsB;
 static const uint MAX_SAMPLE_COUNT = 32768;
 
-extern "C" void initMergeSort(void) {
+extern "C" void initMergeSort(void)
-  checkCudaErrors(
+{
-      cudaMalloc((void **)&d_RanksA, MAX_SAMPLE_COUNT * sizeof(uint)));
+    checkCudaErrors(cudaMalloc((void **)&d_RanksA, MAX_SAMPLE_COUNT * sizeof(uint)));
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc((void **)&d_RanksB, MAX_SAMPLE_COUNT * sizeof(uint)));
-      cudaMalloc((void **)&d_RanksB, MAX_SAMPLE_COUNT * sizeof(uint)));
+    checkCudaErrors(cudaMalloc((void **)&d_LimitsA, MAX_SAMPLE_COUNT * sizeof(uint)));
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc((void **)&d_LimitsB, MAX_SAMPLE_COUNT * sizeof(uint)));
-      cudaMalloc((void **)&d_LimitsA, MAX_SAMPLE_COUNT * sizeof(uint)));
-  checkCudaErrors(
-      cudaMalloc((void **)&d_LimitsB, MAX_SAMPLE_COUNT * sizeof(uint)));
 }
 
-extern "C" void closeMergeSort(void) {
+extern "C" void closeMergeSort(void)
+{
     checkCudaErrors(cudaFree(d_RanksA));
     checkCudaErrors(cudaFree(d_RanksB));
     checkCudaErrors(cudaFree(d_LimitsB));
     checkCudaErrors(cudaFree(d_LimitsA));
 }
 
-extern "C" void mergeSort(uint *d_DstKey, uint *d_DstVal, uint *d_BufKey,
+extern "C" void mergeSort(uint *d_DstKey,
-                          uint *d_BufVal, uint *d_SrcKey, uint *d_SrcVal,
+                          uint *d_DstVal,
-                          uint N, uint sortDir) {
+                          uint *d_BufKey,
+                          uint *d_BufVal,
+                          uint *d_SrcKey,
+                          uint *d_SrcVal,
+                          uint  N,
+                          uint  sortDir)
+{
     uint stageCount = 0;
 
     for (uint stride = SHARED_SIZE_LIMIT; stride < N; stride <<= 1, stageCount++)
@@ -482,7 +489,8 @@ extern "C" void mergeSort(uint *d_DstKey, uint *d_DstVal, uint *d_BufKey,
         ival = d_BufVal;
         okey = d_DstKey;
         oval = d_DstVal;
-  } else {
+    }
+    else {
         ikey = d_DstKey;
         ival = d_DstVal;
         okey = d_BufKey;
@@ -491,8 +499,7 @@ extern "C" void mergeSort(uint *d_DstKey, uint *d_DstVal, uint *d_BufKey,
 
     assert(N <= (SAMPLE_STRIDE * MAX_SAMPLE_COUNT));
     assert(N % SHARED_SIZE_LIMIT == 0);
-  mergeSortShared(ikey, ival, d_SrcKey, d_SrcVal, N / SHARED_SIZE_LIMIT,
+    mergeSortShared(ikey, ival, d_SrcKey, d_SrcVal, N / SHARED_SIZE_LIMIT, SHARED_SIZE_LIMIT, sortDir);
-                  SHARED_SIZE_LIMIT, sortDir);
 
     for (uint stride = SHARED_SIZE_LIMIT; stride < N; stride <<= 1) {
         uint lastSegmentElements = N % (2 * stride);
@@ -504,18 +511,19 @@ extern "C" void mergeSort(uint *d_DstKey, uint *d_DstVal, uint *d_BufKey,
         mergeRanksAndIndices(d_LimitsA, d_LimitsB, d_RanksA, d_RanksB, stride, N);
 
         // Merge elementary intervals
-    mergeElementaryIntervals(okey, oval, ikey, ival, d_LimitsA, d_LimitsB,
+        mergeElementaryIntervals(okey, oval, ikey, ival, d_LimitsA, d_LimitsB, stride, N, sortDir);
-                             stride, N, sortDir);
 
         if (lastSegmentElements <= stride) {
             // Last merge segment consists of a single array which just needs to be
             // passed through
-      checkCudaErrors(cudaMemcpy(
+            checkCudaErrors(cudaMemcpy(okey + (N - lastSegmentElements),
-          okey + (N - lastSegmentElements), ikey + (N - lastSegmentElements),
+                                       ikey + (N - lastSegmentElements),
-          lastSegmentElements * sizeof(uint), cudaMemcpyDeviceToDevice));
+                                       lastSegmentElements * sizeof(uint),
-      checkCudaErrors(cudaMemcpy(
+                                       cudaMemcpyDeviceToDevice));
-          oval + (N - lastSegmentElements), ival + (N - lastSegmentElements),
+            checkCudaErrors(cudaMemcpy(oval + (N - lastSegmentElements),
-          lastSegmentElements * sizeof(uint), cudaMemcpyDeviceToDevice));
+                                       ival + (N - lastSegmentElements),
+                                       lastSegmentElements * sizeof(uint),
+                                       cudaMemcpyDeviceToDevice));
         }
 
         uint *t;

Samples/0_Introduction/mergeSort/mergeSort_common.h

@@ -36,14 +36,12 @@ typedef unsigned int uint;
 ////////////////////////////////////////////////////////////////////////////////
 // Extensive sort validation routine
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
+extern "C" uint
-                                   uint arrayLength, uint numValues,
+validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize, uint arrayLength, uint numValues, uint sortDir);
-                                   uint sortDir);
 
 extern "C" void fillValues(uint *val, uint N);
 
-extern "C" int validateSortedValues(uint *resKey, uint *resVal, uint *srcKey,
+extern "C" int validateSortedValues(uint *resKey, uint *resVal, uint *srcKey, uint batchSize, uint arrayLength);
-                                    uint batchSize, uint arrayLength);
 
 ////////////////////////////////////////////////////////////////////////////////
 // CUDA merge sort
@@ -52,13 +50,11 @@ extern "C" void initMergeSort(void);
 
 extern "C" void closeMergeSort(void);
 
-extern "C" void mergeSort(uint *dstKey, uint *dstVal, uint *bufKey,
+extern "C" void
-                          uint *bufVal, uint *srcKey, uint *srcVal, uint N,
+mergeSort(uint *dstKey, uint *dstVal, uint *bufKey, uint *bufVal, uint *srcKey, uint *srcVal, uint N, uint sortDir);
-                          uint sortDir);
 
 ////////////////////////////////////////////////////////////////////////////////
 // CPU "emulation"
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
+extern "C" void
-                              uint *bufVal, uint *srcKey, uint *srcVal, uint N,
+mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey, uint *bufVal, uint *srcKey, uint *srcVal, uint N, uint sortDir);
-                              uint sortDir);

Samples/0_Introduction/mergeSort/mergeSort_host.cpp

@@ -29,19 +29,20 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+
 #include "mergeSort_common.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Helper functions
 ////////////////////////////////////////////////////////////////////////////////
-static void checkOrder(uint *data, uint N, uint sortDir) {
+static void checkOrder(uint *data, uint N, uint sortDir)
+{
     if (N <= 1) {
         return;
     }
 
     for (uint i = 0; i < N - 1; i++)
-    if ((sortDir && (data[i] > data[i + 1])) ||
+        if ((sortDir && (data[i] > data[i + 1])) || (!sortDir && (data[i] < data[i + 1]))) {
-        (!sortDir && (data[i] < data[i + 1]))) {
             fprintf(stderr, "checkOrder() failed!!!\n");
             exit(EXIT_FAILURE);
         }
@@ -49,12 +50,13 @@ static void checkOrder(uint *data, uint N, uint sortDir) {
 
 static uint umin(uint a, uint b) { return (a <= b) ? a : b; }
 
-static uint getSampleCount(uint dividend) {
+static uint getSampleCount(uint dividend)
-  return ((dividend % SAMPLE_STRIDE) != 0) ? (dividend / SAMPLE_STRIDE + 1)
+{
-                                           : (dividend / SAMPLE_STRIDE);
+    return ((dividend % SAMPLE_STRIDE) != 0) ? (dividend / SAMPLE_STRIDE + 1) : (dividend / SAMPLE_STRIDE);
 }
 
-static uint nextPowerOfTwo(uint x) {
+static uint nextPowerOfTwo(uint x)
+{
     --x;
     x |= x >> 1;
     x |= x >> 2;
@@ -64,7 +66,8 @@ static uint nextPowerOfTwo(uint x) {
     return ++x;
 }
 
-static uint binarySearchInclusive(uint val, uint *data, uint L, uint sortDir) {
+static uint binarySearchInclusive(uint val, uint *data, uint L, uint sortDir)
+{
     if (L == 0) {
         return 0;
     }
@@ -74,8 +77,7 @@ static uint binarySearchInclusive(uint val, uint *data, uint L, uint sortDir) {
     for (uint stride = nextPowerOfTwo(L); stride > 0; stride >>= 1) {
         uint newPos = umin(pos + stride, L);
 
-    if ((sortDir && (data[newPos - 1] <= val)) ||
+        if ((sortDir && (data[newPos - 1] <= val)) || (!sortDir && (data[newPos - 1] >= val))) {
-        (!sortDir && (data[newPos - 1] >= val))) {
             pos = newPos;
         }
     }
@@ -83,7 +85,8 @@ static uint binarySearchInclusive(uint val, uint *data, uint L, uint sortDir) {
     return pos;
 }
 
-static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir) {
+static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir)
+{
     if (L == 0) {
         return 0;
     }
@@ -93,8 +96,7 @@ static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir) {
     for (uint stride = nextPowerOfTwo(L); stride > 0; stride >>= 1) {
         uint newPos = umin(pos + stride, L);
 
-    if ((sortDir && (data[newPos - 1] < val)) ||
+        if ((sortDir && (data[newPos - 1] < val)) || (!sortDir && (data[newPos - 1] > val))) {
-        (!sortDir && (data[newPos - 1] > val))) {
             pos = newPos;
         }
     }
@@ -105,12 +107,10 @@ static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir) {
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 1: find sample ranks in each segment
 ////////////////////////////////////////////////////////////////////////////////
-static void generateSampleRanks(uint *ranksA, uint *ranksB, uint *srcKey,
+static void generateSampleRanks(uint *ranksA, uint *ranksB, uint *srcKey, uint stride, uint N, uint sortDir)
-                                uint stride, uint N, uint sortDir) {
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint sampleCount =
+    uint sampleCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
-      (lastSegmentElements > stride)
-          ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
                                                       : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
 
     for (uint pos = 0; pos < sampleCount; pos++) {
@@ -124,17 +124,14 @@ static void generateSampleRanks(uint *ranksA, uint *ranksB, uint *srcKey,
 
         if (i < nA) {
             ranksA[(segmentBase + 0) / SAMPLE_STRIDE + i] = i * SAMPLE_STRIDE;
-      ranksB[(segmentBase + 0) / SAMPLE_STRIDE + i] =
+            ranksB[(segmentBase + 0) / SAMPLE_STRIDE + i] = binarySearchExclusive(
-          binarySearchExclusive(srcKey[segmentBase + i * SAMPLE_STRIDE],
+                srcKey[segmentBase + i * SAMPLE_STRIDE], srcKey + segmentBase + stride, lenB, sortDir);
-                                srcKey + segmentBase + stride, lenB, sortDir);
         }
 
         if (i < nB) {
             ranksB[(segmentBase + stride) / SAMPLE_STRIDE + i] = i * SAMPLE_STRIDE;
-      ranksA[(segmentBase + stride) / SAMPLE_STRIDE + i] =
+            ranksA[(segmentBase + stride) / SAMPLE_STRIDE + i] = binarySearchInclusive(
-          binarySearchInclusive(
+                srcKey[segmentBase + stride + i * SAMPLE_STRIDE], srcKey + segmentBase, lenA, sortDir);
-              srcKey[segmentBase + stride + i * SAMPLE_STRIDE],
-              srcKey + segmentBase, lenA, sortDir);
         }
     }
 }
@@ -142,12 +139,10 @@ static void generateSampleRanks(uint *ranksA, uint *ranksB, uint *srcKey,
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 2: merge ranks and indices to derive elementary intervals
 ////////////////////////////////////////////////////////////////////////////////
-static void mergeRanksAndIndices(uint *limits, uint *ranks, uint stride,
+static void mergeRanksAndIndices(uint *limits, uint *ranks, uint stride, uint N)
-                                 uint N) {
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint sampleCount =
+    uint sampleCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
-      (lastSegmentElements > stride)
-          ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
                                                       : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
 
     for (uint pos = 0; pos < sampleCount; pos++) {
@@ -161,23 +156,20 @@ static void mergeRanksAndIndices(uint *limits, uint *ranks, uint stride,
 
         if (i < nA) {
             uint dstPosA =
-          binarySearchExclusive(ranks[(segmentBase + 0) / SAMPLE_STRIDE + i],
+                binarySearchExclusive(
-                                ranks + (segmentBase + stride) / SAMPLE_STRIDE,
+                    ranks[(segmentBase + 0) / SAMPLE_STRIDE + i], ranks + (segmentBase + stride) / SAMPLE_STRIDE, nB, 1)
-                                nB, 1) +
+                + i;
-          i;
             assert(dstPosA < nA + nB);
-      limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] =
+            limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] = ranks[(segmentBase + 0) / SAMPLE_STRIDE + i];
-          ranks[(segmentBase + 0) / SAMPLE_STRIDE + i];
         }
 
         if (i < nB) {
-      uint dstPosA = binarySearchInclusive(
+            uint dstPosA =
-                         ranks[(segmentBase + stride) / SAMPLE_STRIDE + i],
+                binarySearchInclusive(
-                         ranks + (segmentBase + 0) / SAMPLE_STRIDE, nA, 1) +
+                    ranks[(segmentBase + stride) / SAMPLE_STRIDE + i], ranks + (segmentBase + 0) / SAMPLE_STRIDE, nA, 1)
-                     i;
+                + i;
             assert(dstPosA < nA + nB);
-      limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] =
+            limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] = ranks[(segmentBase + stride) / SAMPLE_STRIDE + i];
-          ranks[(segmentBase + stride) / SAMPLE_STRIDE + i];
         }
     }
 }
@@ -185,9 +177,16 @@ static void mergeRanksAndIndices(uint *limits, uint *ranks, uint stride,
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 3: merge elementary intervals (each interval is <= SAMPLE_STRIDE)
 ////////////////////////////////////////////////////////////////////////////////
-static void merge(uint *dstKey, uint *dstVal, uint *srcAKey, uint *srcAVal,
+static void merge(uint *dstKey,
-                  uint *srcBKey, uint *srcBVal, uint lenA, uint lenB,
+                  uint *dstVal,
-                  uint sortDir) {
+                  uint *srcAKey,
+                  uint *srcAVal,
+                  uint *srcBKey,
+                  uint *srcBVal,
+                  uint  lenA,
+                  uint  lenB,
+                  uint  sortDir)
+{
     checkOrder(srcAKey, lenA, sortDir);
     checkOrder(srcBKey, lenB, sortDir);
 
@@ -206,13 +205,18 @@ static void merge(uint *dstKey, uint *dstVal, uint *srcAKey, uint *srcAVal,
     }
 }
 
-static void mergeElementaryIntervals(uint *dstKey, uint *dstVal, uint *srcKey,
+static void mergeElementaryIntervals(uint *dstKey,
-                                     uint *srcVal, uint *limitsA, uint *limitsB,
+                                     uint *dstVal,
-                                     uint stride, uint N, uint sortDir) {
+                                     uint *srcKey,
+                                     uint *srcVal,
+                                     uint *limitsA,
+                                     uint *limitsB,
+                                     uint  stride,
+                                     uint  N,
+                                     uint  sortDir)
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint mergePairs = (lastSegmentElements > stride)
+    uint mergePairs = (lastSegmentElements > stride) ? getSampleCount(N) : (N - lastSegmentElements) / SAMPLE_STRIDE;
-                        ? getSampleCount(N)
-                        : (N - lastSegmentElements) / SAMPLE_STRIDE;
 
     for (uint pos = 0; pos < mergePairs; pos++) {
         uint i           = pos & ((2 * stride) / SAMPLE_STRIDE - 1);
@@ -240,15 +244,18 @@ static void mergeElementaryIntervals(uint *dstKey, uint *dstVal, uint *srcKey,
               (srcKey + segmentBase + 0) + startPosA,
               (srcVal + segmentBase + 0) + startPosA,
               (srcKey + segmentBase + stride) + startPosB,
-          (srcVal + segmentBase + stride) + startPosB, endPosA - startPosA,
+              (srcVal + segmentBase + stride) + startPosB,
-          endPosB - startPosB, sortDir);
+              endPosA - startPosA,
+              endPosB - startPosB,
+              sortDir);
     }
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Retarded bubble sort
 ////////////////////////////////////////////////////////////////////////////////
-static void bubbleSort(uint *key, uint *val, uint N, uint sortDir) {
+static void bubbleSort(uint *key, uint *val, uint N, uint sortDir)
+{
     if (N <= 1) {
         return;
     }
@@ -278,9 +285,9 @@ static void bubbleSort(uint *key, uint *val, uint N, uint sortDir) {
 ////////////////////////////////////////////////////////////////////////////////
 // Interface function
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
+extern "C" void
-                              uint *bufVal, uint *srcKey, uint *srcVal, uint N,
+mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey, uint *bufVal, uint *srcKey, uint *srcVal, uint N, uint sortDir)
-                              uint sortDir) {
+{
     uint *ikey, *ival, *okey, *oval;
     uint  stageCount = 0;
 
@@ -292,7 +299,8 @@ extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
         ival = bufVal;
         okey = dstKey;
         oval = dstVal;
-  } else {
+    }
+    else {
         ikey = dstKey;
         ival = dstVal;
         okey = bufKey;
@@ -304,8 +312,7 @@ extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
     memcpy(ival, srcVal, N * sizeof(uint));
 
     for (uint pos = 0; pos < N; pos += SHARED_SIZE_LIMIT) {
-    bubbleSort(ikey + pos, ival + pos, umin(SHARED_SIZE_LIMIT, N - pos),
+        bubbleSort(ikey + pos, ival + pos, umin(SHARED_SIZE_LIMIT, N - pos), sortDir);
-               sortDir);
     }
 
     printf("Merge...\n");
@@ -329,16 +336,15 @@ extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
         mergeRanksAndIndices(limitsB, ranksB, stride, N);
 
         // Merge elementary intervals
-    mergeElementaryIntervals(okey, oval, ikey, ival, limitsA, limitsB, stride,
+        mergeElementaryIntervals(okey, oval, ikey, ival, limitsA, limitsB, stride, N, sortDir);
-                             N, sortDir);
 
         if (lastSegmentElements <= stride) {
             // Last merge segment consists of a single array which just needs to be
             // passed through
-      memcpy(okey + (N - lastSegmentElements), ikey + (N - lastSegmentElements),
+            memcpy(
-             lastSegmentElements * sizeof(uint));
+                okey + (N - lastSegmentElements), ikey + (N - lastSegmentElements), lastSegmentElements * sizeof(uint));
-      memcpy(oval + (N - lastSegmentElements), ival + (N - lastSegmentElements),
+            memcpy(
-             lastSegmentElements * sizeof(uint));
+                oval + (N - lastSegmentElements), ival + (N - lastSegmentElements), lastSegmentElements * sizeof(uint));
         }
 
         uint *t;

Samples/0_Introduction/mergeSort/mergeSort_validate.cpp

@@ -29,14 +29,15 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+
 #include "mergeSort_common.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Validate sorted keys array (check for integrity and proper order)
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
+extern "C" uint
-                                   uint arrayLength, uint numValues,
+validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize, uint arrayLength, uint numValues, uint sortDir)
-                                   uint sortDir) {
+{
     uint *srcHist;
     uint *resHist;
 
@@ -51,8 +52,7 @@ extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
 
     int flag = 1;
 
-  for (uint j = 0; j < batchSize;
+    for (uint j = 0; j < batchSize; j++, srcKey += arrayLength, resKey += arrayLength) {
-       j++, srcKey += arrayLength, resKey += arrayLength) {
         // Build histograms for keys arrays
         memset(srcHist, 0, numValues * sizeof(uint));
         memset(resHist, 0, numValues * sizeof(uint));
@@ -61,11 +61,9 @@ extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
             if ((srcKey[i] < numValues) && (resKey[i] < numValues)) {
                 srcHist[srcKey[i]]++;
                 resHist[resKey[i]]++;
-      } else {
+            }
-        fprintf(
+            else {
-            stderr,
+                fprintf(stderr, "***Set %u source/result key arrays are not limited properly***\n", j);
-            "***Set %u source/result key arrays are not limited properly***\n",
-            j);
                 flag = 0;
                 goto brk;
             }
@@ -74,18 +72,15 @@ extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
         // Compare the histograms
         for (uint i = 0; i < numValues; i++)
             if (srcHist[i] != resHist[i]) {
-        fprintf(stderr,
+                fprintf(stderr, "***Set %u source/result keys histograms do not match***\n", j);
-                "***Set %u source/result keys histograms do not match***\n", j);
                 flag = 0;
                 goto brk;
             }
 
         // Finally check the ordering
         for (uint i = 0; i < arrayLength - 1; i++)
-      if ((sortDir && (resKey[i] > resKey[i + 1])) ||
+            if ((sortDir && (resKey[i] > resKey[i + 1])) || (!sortDir && (resKey[i] < resKey[i + 1]))) {
-          (!sortDir && (resKey[i] < resKey[i + 1]))) {
+                fprintf(stderr, "***Set %u result key array is not ordered properly***\n", j);
-        fprintf(stderr,
-                "***Set %u result key array is not ordered properly***\n", j);
                 flag = 0;
                 goto brk;
             }
@@ -95,7 +90,8 @@ brk:
     free(resHist);
     free(srcHist);
 
-  if (flag) printf("OK\n");
+    if (flag)
+        printf("OK\n");
 
     return flag;
 }
@@ -103,30 +99,30 @@ brk:
 ////////////////////////////////////////////////////////////////////////////////
 // Value validation / stability check routines
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" void fillValues(uint *val, uint N) {
+extern "C" void fillValues(uint *val, uint N)
-  for (uint i = 0; i < N; i++) val[i] = i;
+{
+    for (uint i = 0; i < N; i++)
+        val[i] = i;
 }
 
-extern "C" int validateSortedValues(uint *resKey, uint *resVal, uint *srcKey,
+extern "C" int validateSortedValues(uint *resKey, uint *resVal, uint *srcKey, uint batchSize, uint arrayLength)
-                                    uint batchSize, uint arrayLength) {
+{
     int correctFlag = 1, stableFlag = 1;
 
     printf("...inspecting keys and values array: ");
 
-  for (uint i = 0; i < batchSize;
+    for (uint i = 0; i < batchSize; i++, resKey += arrayLength, resVal += arrayLength) {
-       i++, resKey += arrayLength, resVal += arrayLength) {
         for (uint j = 0; j < arrayLength; j++) {
-      if (resKey[j] != srcKey[resVal[j]]) correctFlag = 0;
+            if (resKey[j] != srcKey[resVal[j]])
+                correctFlag = 0;
 
-      if ((j < arrayLength - 1) && (resKey[j] == resKey[j + 1]) &&
+            if ((j < arrayLength - 1) && (resKey[j] == resKey[j + 1]) && (resVal[j] > resVal[j + 1]))
-          (resVal[j] > resVal[j + 1]))
                 stableFlag = 0;
         }
     }
 
     printf(correctFlag ? "OK\n" : "***corrupted!!!***\n");
-  printf(stableFlag ? "...stability property: stable!\n"
+    printf(stableFlag ? "...stability property: stable!\n" : "...stability property: NOT stable\n");
-                    : "...stability property: NOT stable\n");
 
     return correctFlag;
 }

Samples/0_Introduction/simpleAWBarrier/simpleAWBarrier.cu

@@ -29,9 +29,9 @@
 #include <stdio.h>
 
 // Includes CUDA
-#include <cuda_runtime.h>
-#include <cuda/barrier>
 #include <cooperative_groups.h>
+#include <cuda/barrier>
+#include <cuda_runtime.h>
 
 // Utilities and timing functions
 #include <helper_functions.h> // includes cuda.h and cuda_runtime_api.h
@@ -43,9 +43,11 @@ namespace cg = cooperative_groups;
 
 #if __CUDA_ARCH__ >= 700
 template <bool writeSquareRoot>
-__device__ void reduceBlockData(
+__device__ void reduceBlockData(cuda::barrier<cuda::thread_scope_block> &barrier,
-    cuda::barrier<cuda::thread_scope_block> &barrier,
+                                cg::thread_block_tile<32>               &tile32,
-    cg::thread_block_tile<32> &tile32, double &threadSum, double *result) {
+                                double                                  &threadSum,
+                                double                                  *result)
+{
     extern __shared__ double tmp[];
 
 #pragma unroll
@@ -62,9 +64,7 @@ __device__ void reduceBlockData(
 
     // The warp 0 will perform last round of reduction
     if (tile32.meta_group_rank() == 0) {
-    double beta = tile32.thread_rank() < tile32.meta_group_size()
+        double beta = tile32.thread_rank() < tile32.meta_group_size() ? tmp[tile32.thread_rank()] : 0.0;
-                      ? tmp[tile32.thread_rank()]
-                      : 0.0;
 
 #pragma unroll
         for (int offset = tile32.size() / 2; offset > 0; offset /= 2) {
@@ -81,8 +81,8 @@ __device__ void reduceBlockData(
 }
 #endif
 
-__global__ void normVecByDotProductAWBarrier(float *vecA, float *vecB,
+__global__ void normVecByDotProductAWBarrier(float *vecA, float *vecB, double *partialResults, int size)
-                                             double *partialResults, int size) {
+{
 #if __CUDA_ARCH__ >= 700
 #pragma diag_suppress static_var_with_dynamic_init
     cg::thread_block cta  = cg::this_thread_block();
@@ -105,8 +105,7 @@ __global__ void normVecByDotProductAWBarrier(float *vecA, float *vecB,
 
     // Each thread block performs reduction of partial dotProducts and writes to
     // global mem.
-  reduceBlockData<false>(barrier, tile32, threadSum,
+    reduceBlockData<false>(barrier, tile32, threadSum, &partialResults[blockIdx.x]);
-                         &partialResults[blockIdx.x]);
 
     cg::sync(grid);
 
@@ -137,15 +136,15 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId);
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", argv[0]);
 
     // This will pick the best possible CUDA capable device
     int dev = findCudaDevice(argc, (const char **)argv);
 
     int major = 0;
-  checkCudaErrors(
+    checkCudaErrors(cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, dev));
-      cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, dev));
 
     // Arrive-Wait Barrier require a GPU of Volta (SM7X) architecture or higher.
     if (major < 7) {
@@ -154,12 +153,10 @@ int main(int argc, char **argv) {
     }
 
     int supportsCooperativeLaunch = 0;
-  checkCudaErrors(cudaDeviceGetAttribute(&supportsCooperativeLaunch,
+    checkCudaErrors(cudaDeviceGetAttribute(&supportsCooperativeLaunch, cudaDevAttrCooperativeLaunch, dev));
-                                         cudaDevAttrCooperativeLaunch, dev));
 
     if (!supportsCooperativeLaunch) {
-    printf(
+        printf("\nSelected GPU (%d) does not support Cooperative Kernel Launch, "
-        "\nSelected GPU (%d) does not support Cooperative Kernel Launch, "
                "Waiving the run\n",
                dev);
         exit(EXIT_WAIVED);
@@ -171,7 +168,8 @@ int main(int argc, char **argv) {
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId) {
+int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId)
+{
     float  *vecA, *d_vecA;
     float  *vecB, *d_vecB;
     double *d_partialResults;
@@ -191,16 +189,14 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId) {
     cudaStream_t stream;
     checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
 
-  checkCudaErrors(cudaMemcpyAsync(d_vecA, vecA, sizeof(float) * size,
+    checkCudaErrors(cudaMemcpyAsync(d_vecA, vecA, sizeof(float) * size, cudaMemcpyHostToDevice, stream));
-                                  cudaMemcpyHostToDevice, stream));
+    checkCudaErrors(cudaMemcpyAsync(d_vecB, vecB, sizeof(float) * size, cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(cudaMemcpyAsync(d_vecB, vecB, sizeof(float) * size,
-                                  cudaMemcpyHostToDevice, stream));
 
     // Kernel configuration, where a one-dimensional
     // grid and one-dimensional blocks are configured.
     int minGridSize = 0, blockSize = 0;
-  checkCudaErrors(cudaOccupancyMaxPotentialBlockSize(
+    checkCudaErrors(
-      &minGridSize, &blockSize, (void *)normVecByDotProductAWBarrier, 0, size));
+        cudaOccupancyMaxPotentialBlockSize(&minGridSize, &blockSize, (void *)normVecByDotProductAWBarrier, 0, size));
 
     int smemSize = ((blockSize / 32) + 1) * sizeof(double);
 
@@ -209,28 +205,24 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId) {
         &numBlocksPerSm, normVecByDotProductAWBarrier, blockSize, smemSize));
 
     int multiProcessorCount = 0;
-  checkCudaErrors(cudaDeviceGetAttribute(
+    checkCudaErrors(cudaDeviceGetAttribute(&multiProcessorCount, cudaDevAttrMultiProcessorCount, deviceId));
-      &multiProcessorCount, cudaDevAttrMultiProcessorCount, deviceId));
 
     minGridSize = multiProcessorCount * numBlocksPerSm;
     checkCudaErrors(cudaMalloc(&d_partialResults, minGridSize * sizeof(double)));
 
-  printf(
+    printf("Launching normVecByDotProductAWBarrier kernel with numBlocks = %d "
-      "Launching normVecByDotProductAWBarrier kernel with numBlocks = %d "
            "blockSize = %d\n",
-      minGridSize, blockSize);
+           minGridSize,
+           blockSize);
 
     dim3 dimGrid(minGridSize, 1, 1), dimBlock(blockSize, 1, 1);
 
-  void *kernelArgs[] = {(void *)&d_vecA, (void *)&d_vecB,
+    void *kernelArgs[] = {(void *)&d_vecA, (void *)&d_vecB, (void *)&d_partialResults, (void *)&size};
-                        (void *)&d_partialResults, (void *)&size};
 
-  checkCudaErrors(
+    checkCudaErrors(cudaLaunchCooperativeKernel(
-      cudaLaunchCooperativeKernel((void *)normVecByDotProductAWBarrier, dimGrid,
+        (void *)normVecByDotProductAWBarrier, dimGrid, dimBlock, kernelArgs, smemSize, stream));
-                                  dimBlock, kernelArgs, smemSize, stream));
 
-  checkCudaErrors(cudaMemcpyAsync(vecA, d_vecA, sizeof(float) * size,
+    checkCudaErrors(cudaMemcpyAsync(vecA, d_vecA, sizeof(float) * size, cudaMemcpyDeviceToHost, stream));
-                                  cudaMemcpyDeviceToHost, stream));
     checkCudaErrors(cudaStreamSynchronize(stream));
 
     float        expectedResult = (baseVal / sqrt(size * baseVal * baseVal));
@@ -239,7 +231,8 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId) {
         if ((vecA[i] - expectedResult) > 0.00001) {
             printf("mismatch at i = %d\n", i);
             break;
-    } else {
+        }
+        else {
             matches++;
         }
     }

Samples/0_Introduction/simpleAssert/simpleAssert.cu

@@ -34,8 +34,8 @@
 #endif
 
 // Includes, system
-#include <stdio.h>
 #include <cassert>
+#include <stdio.h>
 
 // Includes CUDA
 #include <cuda_runtime.h>
@@ -58,7 +58,8 @@ bool testResult = true;
 //! Tests assert function.
 //! Thread whose id > N will print assertion failed error message.
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void testKernel(int N) {
+__global__ void testKernel(int N)
+{
     int gtid = blockIdx.x * blockDim.x + threadIdx.x;
     assert(gtid < N);
 }
@@ -70,17 +71,18 @@ void runTest(int argc, char **argv);
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sampleName,
+    printf("%s completed, returned %s\n", sampleName, testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     int         Nblocks  = 2;
     int         Nthreads = 32;
     cudaError_t error;
@@ -94,7 +96,8 @@ void runTest(int argc, char **argv) {
     if (!strcasecmp(OS_System_Type.sysname, "Darwin")) {
         printf("simpleAssert is not current supported on Mac OSX\n\n");
         exit(EXIT_SUCCESS);
-  } else {
+    }
+    else {
         printf("OS Info: <%s>\n\n", OS_System_Type.version);
     }
 
@@ -118,8 +121,7 @@ void runTest(int argc, char **argv) {
 
     // Check for errors and failed asserts in asynchronous kernel launch.
     if (error == cudaErrorAssert) {
-    printf(
+        printf("Device assert failed as expected, "
-        "Device assert failed as expected, "
                "CUDA error message is: %s\n\n",
                cudaGetErrorString(error));
     }

Samples/0_Introduction/simpleAssert_nvrtc/simpleAssert.cpp

@@ -34,11 +34,12 @@
 #endif
 
 // Includes, system
-#include <stdio.h>
 #include <cassert>
+#include <stdio.h>
 
 // Includes CUDA
 #include <cuda_runtime.h>
+
 #include "nvrtc_helper.h"
 
 // Utilities and timing functions
@@ -58,7 +59,8 @@ void runTest(int argc, char **argv);
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     runTest(argc, argv);
@@ -66,7 +68,8 @@ int main(int argc, char **argv) {
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     int Nblocks  = 2;
     int Nthreads = 32;
 
@@ -91,10 +94,15 @@ void runTest(int argc, char **argv) {
     int   count  = 60;
     void *args[] = {(void *)&count};
 
-  checkCudaErrors(cuLaunchKernel(
+    checkCudaErrors(cuLaunchKernel(kernel_addr,
-      kernel_addr, dimGrid.x, dimGrid.y, dimGrid.z, /* grid dim */
+                                   dimGrid.x,
-      dimBlock.x, dimBlock.y, dimBlock.z,           /* block dim */
+                                   dimGrid.y,
-      0, 0,                                         /* shared mem, stream */
+                                   dimGrid.z, /* grid dim */
+                                   dimBlock.x,
+                                   dimBlock.y,
+                                   dimBlock.z, /* block dim */
+                                   0,
+                                   0,        /* shared mem, stream */
                                    &args[0], /* arguments */
                                    0));
 

Samples/0_Introduction/simpleAssert_nvrtc/simpleAssert_kernel.cu

@@ -32,7 +32,8 @@
 //! Thread whose id > N will print assertion failed error message.
 ////////////////////////////////////////////////////////////////////////////////
 
-extern "C" __global__ void testKernel(int N) {
+extern "C" __global__ void testKernel(int N)
+{
     int gtid = blockIdx.x * blockDim.x + threadIdx.x;
     assert(gtid < N);
 }

Samples/0_Introduction/simpleAtomicIntrinsics/simpleAtomicIntrinsics.cu

@@ -30,10 +30,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 #ifdef _WIN32
 #define WINDOWS_LEAN_AND_MEAN
@@ -68,20 +68,21 @@ extern "C" bool computeGold(int *gpuData, const int len);
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sampleName,
+    printf("%s completed, returned %s\n", sampleName, testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     cudaStream_t stream;
     // This will pick the best possible CUDA capable device
     findCudaDevice(argc, (const char **)argv);
@@ -100,7 +101,8 @@ void runTest(int argc, char **argv) {
     checkCudaErrors(cudaMallocHost(&hOData, memSize));
 
     // initialize the memory
-  for (unsigned int i = 0; i < numData; i++) hOData[i] = 0;
+    for (unsigned int i = 0; i < numData; i++)
+        hOData[i] = 0;
 
     // To make the AND and XOR tests generate something other than 0...
     hOData[8] = hOData[10] = 0xff;
@@ -110,15 +112,13 @@ void runTest(int argc, char **argv) {
     int *dOData;
     checkCudaErrors(cudaMalloc((void **)&dOData, memSize));
     // copy host memory to device to initialize to zero
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(dOData, hOData, memSize, cudaMemcpyHostToDevice, stream));
-      cudaMemcpyAsync(dOData, hOData, memSize, cudaMemcpyHostToDevice, stream));
 
     // execute the kernel
     testKernel<<<numBlocks, numThreads, 0, stream>>>(dOData);
 
     // Copy result from device to host
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(hOData, dOData, memSize, cudaMemcpyDeviceToHost, stream));
-      cudaMemcpyAsync(hOData, dOData, memSize, cudaMemcpyDeviceToHost, stream));
     checkCudaErrors(cudaStreamSynchronize(stream));
 
     sdkStopTimer(&timer);

Samples/0_Introduction/simpleAtomicIntrinsics/simpleAtomicIntrinsics_cpu.cpp

@@ -42,7 +42,8 @@ extern "C" int computeGold(int *gpuData, const int len);
 //! @param idata      input data as provided to device
 //! @param len        number of elements in reference / idata
 ////////////////////////////////////////////////////////////////////////////////
-int computeGold(int *gpuData, const int len) {
+int computeGold(int *gpuData, const int len)
+{
     int val = 0;
 
     for (int i = 0; i < len; ++i) {

Samples/0_Introduction/simpleAtomicIntrinsics/simpleAtomicIntrinsics_kernel.cuh

@@ -35,7 +35,8 @@
 //! @param g_idata  input data in global memory
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void testKernel(int *g_odata) {
+__global__ void testKernel(int *g_odata)
+{
     // access thread id
     const unsigned int tid = blockDim.x * blockIdx.x + threadIdx.x;
 

Samples/0_Introduction/simpleAtomicIntrinsics_nvrtc/simpleAtomicIntrinsics.cpp

@@ -30,10 +30,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 #ifdef _WIN32
 #define WINDOWS_LEAN_AND_MEAN
@@ -64,13 +64,13 @@ extern "C" bool computeGold(int *gpuData, const int len);
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sampleName,
+    printf("%s completed, returned %s\n", sampleName, testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
 
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
@@ -79,7 +79,8 @@ int main(int argc, char **argv) {
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
 
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     int dev = 0;
 
     char  *cubin, *kernel_file;
@@ -106,7 +107,8 @@ void runTest(int argc, char **argv) {
     int *hOData = (int *)malloc(memSize);
 
     // initialize the memory
-  for (unsigned int i = 0; i < numData; i++) hOData[i] = 0;
+    for (unsigned int i = 0; i < numData; i++)
+        hOData[i] = 0;
 
     // To make the AND and XOR tests generate something other than 0...
     hOData[8] = hOData[10] = 0xff;
@@ -121,11 +123,15 @@ void runTest(int argc, char **argv) {
     dim3 cudaGridSize(numBlocks, 1, 1);
 
     void *arr[] = {(void *)&dOData};
-  checkCudaErrors(cuLaunchKernel(kernel_addr, cudaGridSize.x, cudaGridSize.y,
+    checkCudaErrors(cuLaunchKernel(kernel_addr,
+                                   cudaGridSize.x,
+                                   cudaGridSize.y,
                                    cudaGridSize.z, /* grid dim */
-                                 cudaBlockSize.x, cudaBlockSize.y,
+                                   cudaBlockSize.x,
+                                   cudaBlockSize.y,
                                    cudaBlockSize.z, /* block dim */
-                                 0, 0,            /* shared mem, stream */
+                                   0,
+                                   0,       /* shared mem, stream */
                                    &arr[0], /* arguments */
                                    0));
 

Samples/0_Introduction/simpleAtomicIntrinsics_nvrtc/simpleAtomicIntrinsics_cpu.cpp

@@ -43,7 +43,8 @@ extern "C" int computeGold(int *gpuData, const int len);
 //! @param len        number of elements in reference / idata
 ////////////////////////////////////////////////////////////////////////////////
 
-int computeGold(int *gpuData, const int len) {
+int computeGold(int *gpuData, const int len)
+{
     int val = 0;
 
     for (int i = 0; i < len; ++i) {

Samples/0_Introduction/simpleAtomicIntrinsics_nvrtc/simpleAtomicIntrinsics_kernel.cuh

@@ -36,7 +36,8 @@
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
 
-extern "C" __global__ void testKernel(int *g_odata) {
+extern "C" __global__ void testKernel(int *g_odata)
+{
     // access thread id
     const unsigned int tid = blockDim.x * blockIdx.x + threadIdx.x;
 

Samples/0_Introduction/simpleAttributes/simpleAttributes.cu

@@ -26,10 +26,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 // includes CUDA
 #include <cuda_runtime.h>
@@ -42,7 +42,8 @@
 // declaration, forward
 void runTest(int argc, char **argv);
 
-cudaAccessPolicyWindow initAccessPolicyWindow(void) {
+cudaAccessPolicyWindow initAccessPolicyWindow(void)
+{
     cudaAccessPolicyWindow accessPolicyWindow = {0};
     accessPolicyWindow.base_ptr               = (void *)0;
     accessPolicyWindow.num_bytes              = 0;
@@ -60,8 +61,8 @@ cudaAccessPolicyWindow initAccessPolicyWindow(void) {
 //! @param bigDataSize  input bigData size
 //! @param hitcount how many data access are done within block
 ////////////////////////////////////////////////////////////////////////////////
-static __global__ void kernCacheSegmentTest(int *data, int dataSize, int *trash,
+static __global__ void kernCacheSegmentTest(int *data, int dataSize, int *trash, int bigDataSize, int hitCount)
-                                            int bigDataSize, int hitCount) {
+{
     __shared__ unsigned int hit;
     int                     row    = blockIdx.y * blockDim.y + threadIdx.y;
     int                     col    = blockIdx.x * blockDim.x + threadIdx.x;
@@ -82,9 +83,9 @@ static __global__ void kernCacheSegmentTest(int *data, int dataSize, int *trash,
 
         if ((tID % 2) == 0) {
             data[psRand % dataSize] = data[psRand % dataSize] + data[idx % dataSize];
-    } else {
+        }
-      trash[psRand % bigDataSize] =
+        else {
-          trash[psRand % bigDataSize] + trash[idx % bigDataSize];
+            trash[psRand % bigDataSize] = trash[psRand % bigDataSize] + trash[idx % bigDataSize];
         }
 
         atomicAdd(&hit, 1);
@@ -98,7 +99,8 @@ int main(int argc, char **argv) { runTest(argc, argv); }
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     bool                   bTestResult = true;
     cudaAccessPolicyWindow accessPolicyWindow;
     cudaDeviceProp         deviceProp;
@@ -127,8 +129,7 @@ void runTest(int argc, char **argv) {
 
     // Make sure device the l2 optimization
     if (deviceProp.persistingL2CacheMaxSize == 0) {
-    printf(
+        printf("Waiving execution as device %d does not support persisting L2 "
-        "Waiving execution as device %d does not support persisting L2 "
                "Caching\n",
                devID);
         exit(EXIT_WAIVED);
@@ -139,8 +140,7 @@ void runTest(int argc, char **argv) {
 
     // Set the amount of l2 cache that will be persisting to maximum the device
     // can support
-  checkCudaErrors(cudaDeviceSetLimit(cudaLimitPersistingL2CacheSize,
+    checkCudaErrors(cudaDeviceSetLimit(cudaLimitPersistingL2CacheSize, deviceProp.persistingL2CacheMaxSize));
-                                     deviceProp.persistingL2CacheMaxSize));
 
     // Stream attribute to set
     streamAttrID = cudaStreamAttributeAccessPolicyWindow;
@@ -155,8 +155,7 @@ void runTest(int argc, char **argv) {
 
     // Allocate data
     checkCudaErrors(cudaMallocHost(&dataHostPointer, dataSize * sizeof(int)));
-  checkCudaErrors(
+    checkCudaErrors(cudaMallocHost(&bigDataHostPointer, bigDataSize * sizeof(int)));
-      cudaMallocHost(&bigDataHostPointer, bigDataSize * sizeof(int)));
 
     for (int i = 0; i < bigDataSize; ++i) {
         if (i < dataSize) {
@@ -166,16 +165,12 @@ void runTest(int argc, char **argv) {
         bigDataHostPointer[bigDataSize - i - 1] = i;
     }
 
+    checkCudaErrors(cudaMalloc((void **)&dataDevicePointer, dataSize * sizeof(int)));
+    checkCudaErrors(cudaMalloc((void **)&bigDataDevicePointer, bigDataSize * sizeof(int)));
     checkCudaErrors(
-      cudaMalloc((void **)&dataDevicePointer, dataSize * sizeof(int)));
+        cudaMemcpyAsync(dataDevicePointer, dataHostPointer, dataSize * sizeof(int), cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(
-      cudaMalloc((void **)&bigDataDevicePointer, bigDataSize * sizeof(int)));
+        bigDataDevicePointer, bigDataHostPointer, bigDataSize * sizeof(int), cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(cudaMemcpyAsync(dataDevicePointer, dataHostPointer,
-                                  dataSize * sizeof(int),
-                                  cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(cudaMemcpyAsync(bigDataDevicePointer, bigDataHostPointer,
-                                  bigDataSize * sizeof(int),
-                                  cudaMemcpyHostToDevice, stream));
 
     // Make a window for the buffer of interest
     accessPolicyWindow.base_ptr        = (void *)dataDevicePointer;
@@ -186,8 +181,7 @@ void runTest(int argc, char **argv) {
     streamAttrValue.accessPolicyWindow = accessPolicyWindow;
 
     // Assign window to stream
-  checkCudaErrors(
+    checkCudaErrors(cudaStreamSetAttribute(stream, streamAttrID, &streamAttrValue));
-      cudaStreamSetAttribute(stream, streamAttrID, &streamAttrValue));
 
     // Demote any previous persisting lines
     checkCudaErrors(cudaCtxResetPersistingL2Cache());

Samples/0_Introduction/simpleCUDA2GL/README.md

@@ -34,4 +34,3 @@ Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-d
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)
-

Samples/0_Introduction/simpleCUDA2GL/main.cpp

@@ -50,8 +50,8 @@
 #endif
 
 // CUDA includes
-#include <cuda_runtime.h>
 #include <cuda_gl_interop.h>
+#include <cuda_runtime.h>
 
 // CUDA utilities and system includes
 #include <helper_cuda.h>
@@ -124,8 +124,7 @@ StopWatchInterface *timer = NULL;
 GLuint shDraw;
 
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" void launch_cudaProcess(dim3 grid, dim3 block, int sbytes,
+extern "C" void launch_cudaProcess(dim3 grid, dim3 block, int sbytes, unsigned int *g_odata, int imgw);
-                                   unsigned int *g_odata, int imgw);
 
 // Forward declarations
 void runStdProgram(int argc, char **argv);
@@ -140,8 +139,7 @@ void createPBO(GLuint *pbo, struct cudaGraphicsResource **pbo_resource);
 void deletePBO(GLuint *pbo);
 #endif
 
-void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x,
+void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x, unsigned int size_y);
-                      unsigned int size_y);
 void deleteTexture(GLuint *tex);
 
 // rendering callbacks
@@ -155,7 +153,8 @@ void mainMenu(int i);
 ////////////////////////////////////////////////////////////////////////////////
 //! Create PBO
 ////////////////////////////////////////////////////////////////////////////////
-void createPBO(GLuint *pbo, struct cudaGraphicsResource **pbo_resource) {
+void createPBO(GLuint *pbo, struct cudaGraphicsResource **pbo_resource)
+{
     // set up vertex data parameter
     num_texels    = image_width * image_height;
     num_values    = num_texels * 4;
@@ -171,33 +170,32 @@ void createPBO(GLuint *pbo, struct cudaGraphicsResource **pbo_resource) {
     glBindBuffer(GL_ARRAY_BUFFER, 0);
 
     // register this buffer object with CUDA
-  checkCudaErrors(cudaGraphicsGLRegisterBuffer(pbo_resource, *pbo,
+    checkCudaErrors(cudaGraphicsGLRegisterBuffer(pbo_resource, *pbo, cudaGraphicsMapFlagsNone));
-                                               cudaGraphicsMapFlagsNone));
 
     SDK_CHECK_ERROR_GL();
 }
 
-void deletePBO(GLuint *pbo) {
+void deletePBO(GLuint *pbo)
+{
     glDeleteBuffers(1, pbo);
     SDK_CHECK_ERROR_GL();
     *pbo = 0;
 }
 #endif
 
-const GLenum fbo_targets[] = {
+const GLenum fbo_targets[] = {GL_COLOR_ATTACHMENT0_EXT,
-    GL_COLOR_ATTACHMENT0_EXT, GL_COLOR_ATTACHMENT1_EXT,
+                              GL_COLOR_ATTACHMENT1_EXT,
-    GL_COLOR_ATTACHMENT2_EXT, GL_COLOR_ATTACHMENT3_EXT};
+                              GL_COLOR_ATTACHMENT2_EXT,
+                              GL_COLOR_ATTACHMENT3_EXT};
 
 #ifndef USE_TEXSUBIMAGE2D
-static const char *glsl_drawtex_vertshader_src =
+static const char *glsl_drawtex_vertshader_src = "void main(void)\n"
-    "void main(void)\n"
                                                  "{\n"
                                                  "	gl_Position = gl_Vertex;\n"
                                                  "	gl_TexCoord[0].xy = gl_MultiTexCoord0.xy;\n"
                                                  "}\n";
 
-static const char *glsl_drawtex_fragshader_src =
+static const char *glsl_drawtex_fragshader_src = "#version 130\n"
-    "#version 130\n"
                                                  "uniform usampler2D texImage;\n"
                                                  "void main()\n"
                                                  "{\n"
@@ -227,15 +225,15 @@ static const char *glsl_draw_fragshader_src =
 #endif
 
 // copy image and process using CUDA
-void generateCUDAImage() {
+void generateCUDAImage()
+{
     // run the Cuda kernel
     unsigned int *out_data;
 
 #ifdef USE_TEXSUBIMAGE2D
     checkCudaErrors(cudaGraphicsMapResources(1, &cuda_pbo_dest_resource, 0));
     size_t num_bytes;
-  checkCudaErrors(cudaGraphicsResourceGetMappedPointer(
+    checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&out_data, &num_bytes, cuda_pbo_dest_resource));
-      (void **)&out_data, &num_bytes, cuda_pbo_dest_resource));
 // printf("CUDA mapped pointer of pbo_out: May access %ld bytes, expected %d\n",
 // num_bytes, size_tex_data);
 #else
@@ -258,8 +256,7 @@ void generateCUDAImage() {
     glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo_dest);
 
     glBindTexture(GL_TEXTURE_2D, tex_cudaResult);
-  glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, image_width, image_height, GL_RGBA,
+    glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, image_width, image_height, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
-                  GL_UNSIGNED_BYTE, NULL);
     SDK_CHECK_ERROR_GL();
     glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, 0);
     glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
@@ -268,21 +265,20 @@ void generateCUDAImage() {
     // map buffer objects to get CUDA device pointers
     cudaArray *texture_ptr;
     checkCudaErrors(cudaGraphicsMapResources(1, &cuda_tex_result_resource, 0));
-  checkCudaErrors(cudaGraphicsSubResourceGetMappedArray(
+    checkCudaErrors(cudaGraphicsSubResourceGetMappedArray(&texture_ptr, cuda_tex_result_resource, 0, 0));
-      &texture_ptr, cuda_tex_result_resource, 0, 0));
 
     int num_texels    = image_width * image_height;
     int num_values    = num_texels * 4;
     int size_tex_data = sizeof(GLubyte) * num_values;
-  checkCudaErrors(cudaMemcpyToArray(texture_ptr, 0, 0, cuda_dest_resource,
+    checkCudaErrors(cudaMemcpyToArray(texture_ptr, 0, 0, cuda_dest_resource, size_tex_data, cudaMemcpyDeviceToDevice));
-                                    size_tex_data, cudaMemcpyDeviceToDevice));
 
     checkCudaErrors(cudaGraphicsUnmapResources(1, &cuda_tex_result_resource, 0));
 #endif
 }
 
 // display image to the screen as textured quad
-void displayImage(GLuint texture) {
+void displayImage(GLuint texture)
+{
     glBindTexture(GL_TEXTURE_2D, texture);
     glEnable(GL_TEXTURE_2D);
     glDisable(GL_DEPTH_TEST);
@@ -332,7 +328,8 @@ void displayImage(GLuint texture) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Display callback
 ////////////////////////////////////////////////////////////////////////////////
-void display() {
+void display()
+{
     sdkStartTimer(&timer);
 
     if (enable_cuda) {
@@ -358,9 +355,7 @@ void display() {
             sprintf(currentOutputPPM, "kilt.ppm");
             g_CheckRender->savePPM(currentOutputPPM, true, NULL);
 
-      if (!g_CheckRender->PPMvsPPM(currentOutputPPM,
+            if (!g_CheckRender->PPMvsPPM(currentOutputPPM, sdkFindFilePath(ref_file, pArgv[0]), MAX_EPSILON, 0.30f)) {
-                                   sdkFindFilePath(ref_file, pArgv[0]),
-                                   MAX_EPSILON, 0.30f)) {
                 g_TotalErrors++;
             }
 
@@ -374,8 +369,7 @@ void display() {
     if (++fpsCount == fpsLimit) {
         char  cTitle[256];
         float fps = 1000.0f / sdkGetAverageTimerValue(&timer);
-    sprintf(cTitle, "CUDA GL Post Processing (%d x %d): %.1f fps", window_width,
+        sprintf(cTitle, "CUDA GL Post Processing (%d x %d): %.1f fps", window_width, window_height, fps);
-            window_height, fps);
         glutSetWindowTitle(cTitle);
         // printf("%s\n", cTitle);
         fpsCount = 0;
@@ -384,7 +378,8 @@ void display() {
     }
 }
 
-void timerEvent(int value) {
+void timerEvent(int value)
+{
     glutPostRedisplay();
     glutTimerFunc(REFRESH_DELAY, timerEvent, 0);
 }
@@ -392,7 +387,8 @@ void timerEvent(int value) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Keyboard events handler
 ////////////////////////////////////////////////////////////////////////////////
-void keyboard(unsigned char key, int /*x*/, int /*y*/) {
+void keyboard(unsigned char key, int /*x*/, int /*y*/)
+{
     switch (key) {
     case (27):
         Cleanup(EXIT_SUCCESS);
@@ -404,7 +400,8 @@ void keyboard(unsigned char key, int /*x*/, int /*y*/) {
 
         if (enable_cuda) {
             glClearColorIuiEXT(128, 128, 128, 255);
-      } else {
+        }
+        else {
             glClearColor(0.5, 0.5, 0.5, 1.0);
         }
 
@@ -413,7 +410,8 @@ void keyboard(unsigned char key, int /*x*/, int /*y*/) {
     }
 }
 
-void reshape(int w, int h) {
+void reshape(int w, int h)
+{
     window_width  = w;
     window_height = h;
 }
@@ -423,8 +421,8 @@ void mainMenu(int i) { keyboard((unsigned char)i, 0, 0); }
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x,
+void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x, unsigned int size_y)
-                      unsigned int size_y) {
+{
     // create a texture
     glGenTextures(1, tex_cudaResult);
     glBindTexture(GL_TEXTURE_2D, *tex_cudaResult);
@@ -436,24 +434,22 @@ void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x,
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
 
 #ifdef USE_TEXSUBIMAGE2D
-  glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, size_x, size_y, 0, GL_RGBA,
+    glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, size_x, size_y, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
-               GL_UNSIGNED_BYTE, NULL);
     SDK_CHECK_ERROR_GL();
 #else
-  glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8UI_EXT, size_x, size_y, 0,
+    glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8UI_EXT, size_x, size_y, 0, GL_RGBA_INTEGER_EXT, GL_UNSIGNED_BYTE, NULL);
-               GL_RGBA_INTEGER_EXT, GL_UNSIGNED_BYTE, NULL);
     SDK_CHECK_ERROR_GL();
     // register this texture with CUDA
     checkCudaErrors(cudaGraphicsGLRegisterImage(
-      &cuda_tex_result_resource, *tex_cudaResult, GL_TEXTURE_2D,
+        &cuda_tex_result_resource, *tex_cudaResult, GL_TEXTURE_2D, cudaGraphicsMapFlagsWriteDiscard));
-      cudaGraphicsMapFlagsWriteDiscard));
 #endif
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-void deleteTexture(GLuint *tex) {
+void deleteTexture(GLuint *tex)
+{
     glDeleteTextures(1, tex);
     SDK_CHECK_ERROR_GL();
 
@@ -463,7 +459,8 @@ void deleteTexture(GLuint *tex) {
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
 #if defined(__linux__)
     char *Xstatus = getenv("DISPLAY");
     if (Xstatus == NULL) {
@@ -487,8 +484,7 @@ int main(int argc, char **argv) {
     if (checkCmdLineFlag(argc, (const char **)argv, "device")) {
         printf("[%s]\n", argv[0]);
         printf("   Does not explicitly support -device=n\n");
-    printf(
+        printf("   This sample requires OpenGL.  Only -file=<reference> are "
-        "   This sample requires OpenGL.  Only -file=<reference> are "
                "supported\n");
         printf("exiting...\n");
         exit(EXIT_WAIVED);
@@ -497,7 +493,8 @@ int main(int argc, char **argv) {
     if (ref_file) {
         printf("(Test with OpenGL verification)\n");
         runStdProgram(argc, argv);
-  } else {
+    }
+    else {
         printf("(Interactive OpenGL Demo)\n");
         runStdProgram(argc, argv);
     }
@@ -508,7 +505,8 @@ int main(int argc, char **argv) {
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-void FreeResource() {
+void FreeResource()
+{
     sdkDeleteTimer(&timer);
 
 // unregister this buffer object with CUDA
@@ -530,18 +528,18 @@ void FreeResource() {
     printf("simpleCUDA2GL Exiting...\n");
 }
 
-void Cleanup(int iExitCode) {
+void Cleanup(int iExitCode)
+{
     FreeResource();
-  printf("PPM Images are %s\n",
+    printf("PPM Images are %s\n", (iExitCode == EXIT_SUCCESS) ? "Matching" : "Not Matching");
-         (iExitCode == EXIT_SUCCESS) ? "Matching" : "Not Matching");
     exit(iExitCode);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-GLuint compileGLSLprogram(const char *vertex_shader_src,
+GLuint compileGLSLprogram(const char *vertex_shader_src, const char *fragment_shader_src)
-                          const char *fragment_shader_src) {
+{
     GLuint v, f, p = 0;
 
     p = glCreateProgram();
@@ -563,7 +561,8 @@ GLuint compileGLSLprogram(const char *vertex_shader_src,
             // #endif
             glDeleteShader(v);
             return 0;
-    } else {
+        }
+        else {
             glAttachShader(p, v);
         }
     }
@@ -585,7 +584,8 @@ GLuint compileGLSLprogram(const char *vertex_shader_src,
             // #endif
             glDeleteShader(f);
             return 0;
-    } else {
+        }
+        else {
             glAttachShader(p, f);
         }
     }
@@ -611,7 +611,8 @@ GLuint compileGLSLprogram(const char *vertex_shader_src,
 //! Allocate the "render target" of CUDA
 ////////////////////////////////////////////////////////////////////////////////
 #ifndef USE_TEXSUBIMAGE2D
-void initCUDABuffers() {
+void initCUDABuffers()
+{
     // set up vertex data parameter
     num_texels    = image_width * image_height;
     num_values    = num_texels * 4;
@@ -625,7 +626,8 @@ void initCUDABuffers() {
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-void initGLBuffers() {
+void initGLBuffers()
+{
 // create pbo
 #ifdef USE_TEXSUBIMAGE2D
     createPBO(&pbo_dest, &cuda_pbo_dest_resource);
@@ -636,8 +638,7 @@ void initGLBuffers() {
     shDraw = compileGLSLprogram(NULL, glsl_draw_fragshader_src);
 
 #ifndef USE_TEXSUBIMAGE2D
-  shDrawTex = compileGLSLprogram(glsl_drawtex_vertshader_src,
+    shDrawTex = compileGLSLprogram(glsl_drawtex_vertshader_src, glsl_drawtex_fragshader_src);
-                                 glsl_drawtex_fragshader_src);
 #endif
     SDK_CHECK_ERROR_GL();
 }
@@ -645,7 +646,8 @@ void initGLBuffers() {
 ////////////////////////////////////////////////////////////////////////////////
 //! Run standard demo loop with or without GL verification
 ////////////////////////////////////////////////////////////////////////////////
-void runStdProgram(int argc, char **argv) {
+void runStdProgram(int argc, char **argv)
+{
     // First initialize OpenGL context, so we can properly set the GL for CUDA.
     // This is necessary in order to achieve optimal performance with OpenGL/CUDA
     // interop.
@@ -683,8 +685,7 @@ void runStdProgram(int argc, char **argv) {
         g_CheckRender->EnableQAReadback(true);
     }
 
-  printf(
+    printf("\n"
-      "\n"
            "\tControls\n"
            "\t(right click mouse button for Menu)\n"
            "\t[esc] - Quit\n\n");
@@ -699,7 +700,8 @@ void runStdProgram(int argc, char **argv) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Initialize GL
 ////////////////////////////////////////////////////////////////////////////////
-bool initGL(int *argc, char **argv) {
+bool initGL(int *argc, char **argv)
+{
     // Create GL context
     glutInit(argc, argv);
     glutInitDisplayMode(GLUT_RGBA | GLUT_ALPHA | GLUT_DOUBLE | GLUT_DEPTH);
@@ -707,8 +709,8 @@ bool initGL(int *argc, char **argv) {
     iGLUTWindowHandle = glutCreateWindow("CUDA OpenGL post-processing");
 
     // initialize necessary OpenGL extensions
-  if (!isGLVersionSupported(2, 0) ||
+    if (!isGLVersionSupported(2, 0)
-      !areGLExtensionsSupported("GL_ARB_pixel_buffer_object "
+        || !areGLExtensionsSupported("GL_ARB_pixel_buffer_object "
                                      "GL_EXT_framebuffer_object")) {
         printf("ERROR: Support for necessary OpenGL extensions missing.");
         fflush(stderr);
@@ -729,8 +731,7 @@ bool initGL(int *argc, char **argv) {
     // projection
     glMatrixMode(GL_PROJECTION);
     glLoadIdentity();
-  gluPerspective(60.0, (GLfloat)window_width / (GLfloat)window_height, 0.1f,
+    gluPerspective(60.0, (GLfloat)window_width / (GLfloat)window_height, 0.1f, 10.0f);
-                 10.0f);
 
     glPolygonMode(GL_FRONT_AND_BACK, GL_FILL);
 

Samples/0_Introduction/simpleCUDA2GL/simpleCUDA2GL.cu

@@ -35,14 +35,16 @@ __device__ float clamp(float x, float a, float b) { return max(a, min(b, x)); }
 __device__ int clamp(int x, int a, int b) { return max(a, min(b, x)); }
 
 // convert floating point rgb color to 8-bit integer
-__device__ int rgbToInt(float r, float g, float b) {
+__device__ int rgbToInt(float r, float g, float b)
+{
     r = clamp(r, 0.0f, 255.0f);
     g = clamp(g, 0.0f, 255.0f);
     b = clamp(b, 0.0f, 255.0f);
     return (int(b) << 16) | (int(g) << 8) | int(r);
 }
 
-__global__ void cudaProcess(unsigned int *g_odata, int imgw) {
+__global__ void cudaProcess(unsigned int *g_odata, int imgw)
+{
     extern __shared__ uchar4 sdata[];
 
     int tx = threadIdx.x;
@@ -56,7 +58,7 @@ __global__ void cudaProcess(unsigned int *g_odata, int imgw) {
     g_odata[y * imgw + x] = rgbToInt(c4.z, c4.y, c4.x);
 }
 
-extern "C" void launch_cudaProcess(dim3 grid, dim3 block, int sbytes,
+extern "C" void launch_cudaProcess(dim3 grid, dim3 block, int sbytes, unsigned int *g_odata, int imgw)
-                                   unsigned int *g_odata, int imgw) {
+{
     cudaProcess<<<grid, block, sbytes>>>(g_odata, imgw);
 }

Samples/0_Introduction/simpleCallback/multithreading.cpp

@@ -29,18 +29,21 @@
 
 #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
 // Create thread
-CUTThread cutStartThread(CUT_THREADROUTINE func, void *data) {
+CUTThread cutStartThread(CUT_THREADROUTINE func, void *data)
+{
     return CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)func, data, 0, NULL);
 }
 
 // Wait for thread to finish
-void cutEndThread(CUTThread thread) {
+void cutEndThread(CUTThread thread)
+{
     WaitForSingleObject(thread, INFINITE);
     CloseHandle(thread);
 }
 
 // Wait for multiple threads
-void cutWaitForThreads(const CUTThread *threads, int num) {
+void cutWaitForThreads(const CUTThread *threads, int num)
+{
     WaitForMultipleObjects(num, threads, true, INFINITE);
 
     for (int i = 0; i < num; i++) {
@@ -49,7 +52,8 @@ void cutWaitForThreads(const CUTThread *threads, int num) {
 }
 
 // Create barrier.
-CUTBarrier cutCreateBarrier(int releaseCount) {
+CUTBarrier cutCreateBarrier(int releaseCount)
+{
     CUTBarrier barrier;
 
     InitializeCriticalSection(&barrier.criticalSection);
@@ -61,7 +65,8 @@ CUTBarrier cutCreateBarrier(int releaseCount) {
 }
 
 // Increment barrier. (execution continues)
-void cutIncrementBarrier(CUTBarrier *barrier) {
+void cutIncrementBarrier(CUTBarrier *barrier)
+{
     int myBarrierCount;
     EnterCriticalSection(&barrier->criticalSection);
     myBarrierCount = ++barrier->count;
@@ -73,16 +78,15 @@ void cutIncrementBarrier(CUTBarrier *barrier) {
 }
 
 // Wait for barrier release.
-void cutWaitForBarrier(CUTBarrier *barrier) {
+void cutWaitForBarrier(CUTBarrier *barrier) { WaitForSingleObject(barrier->barrierEvent, INFINITE); }
-  WaitForSingleObject(barrier->barrierEvent, INFINITE);
-}
 
 // Destroy barrier
 void cutDestroyBarrier(CUTBarrier *barrier) {}
 
 #else
 // Create thread
-CUTThread cutStartThread(CUT_THREADROUTINE func, void *data) {
+CUTThread cutStartThread(CUT_THREADROUTINE func, void *data)
+{
     pthread_t thread;
     pthread_create(&thread, NULL, func, data);
     return thread;
@@ -92,14 +96,16 @@ CUTThread cutStartThread(CUT_THREADROUTINE func, void *data) {
 void cutEndThread(CUTThread thread) { pthread_join(thread, NULL); }
 
 // Wait for multiple threads
-void cutWaitForThreads(const CUTThread *threads, int num) {
+void cutWaitForThreads(const CUTThread *threads, int num)
+{
     for (int i = 0; i < num; i++) {
         cutEndThread(threads[i]);
     }
 }
 
 // Create barrier.
-CUTBarrier cutCreateBarrier(int releaseCount) {
+CUTBarrier cutCreateBarrier(int releaseCount)
+{
     CUTBarrier barrier;
 
     barrier.count        = 0;
@@ -112,7 +118,8 @@ CUTBarrier cutCreateBarrier(int releaseCount) {
 }
 
 // Increment barrier. (execution continues)
-void cutIncrementBarrier(CUTBarrier *barrier) {
+void cutIncrementBarrier(CUTBarrier *barrier)
+{
     int myBarrierCount;
     pthread_mutex_lock(&barrier->mutex);
     myBarrierCount = ++barrier->count;
@@ -124,7 +131,8 @@ void cutIncrementBarrier(CUTBarrier *barrier) {
 }
 
 // Wait for barrier release.
-void cutWaitForBarrier(CUTBarrier *barrier) {
+void cutWaitForBarrier(CUTBarrier *barrier)
+{
     pthread_mutex_lock(&barrier->mutex);
 
     while (barrier->count < barrier->releaseCount) {
@@ -135,7 +143,8 @@ void cutWaitForBarrier(CUTBarrier *barrier) {
 }
 
 // Destroy barrier
-void cutDestroyBarrier(CUTBarrier *barrier) {
+void cutDestroyBarrier(CUTBarrier *barrier)
+{
     pthread_mutex_destroy(&barrier->mutex);
     pthread_cond_destroy(&barrier->conditionVariable);
 }

Samples/0_Introduction/simpleCallback/multithreading.h

@@ -37,7 +37,8 @@
 typedef HANDLE CUTThread;
 typedef unsigned(WINAPI *CUT_THREADROUTINE)(void *);
 
-struct CUTBarrier {
+struct CUTBarrier
+{
     CRITICAL_SECTION criticalSection;
     HANDLE           barrierEvent;
     int              releaseCount;
@@ -57,7 +58,8 @@ typedef void *(*CUT_THREADROUTINE)(void *);
 #define CUT_THREADPROC void *
 #define CUT_THREADEND  return 0
 
-struct CUTBarrier {
+struct CUTBarrier
+{
     pthread_mutex_t mutex;
     pthread_cond_t  conditionVariable;
     int             releaseCount;
@@ -67,7 +69,8 @@ struct CUTBarrier {
 #endif
 
 #ifdef __cplusplus
-extern "C" {
+extern "C"
+{
 #endif
 
     // Create thread.

Samples/0_Introduction/simpleCallback/simpleCallback.cu

@@ -43,8 +43,8 @@
 #include <stdio.h>
 
 // helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 #include "multithreading.h"
 
@@ -53,10 +53,10 @@ const int N_elements_per_workload = 100000;
 
 CUTBarrier thread_barrier;
 
-void CUDART_CB myStreamCallback(cudaStream_t event, cudaError_t status,
+void CUDART_CB myStreamCallback(cudaStream_t event, cudaError_t status, void *data);
-                                void *data);
 
-struct heterogeneous_workload {
+struct heterogeneous_workload
+{
     int id;
     int cudaDeviceID;
 
@@ -67,13 +67,16 @@ struct heterogeneous_workload {
     bool success;
 };
 
-__global__ void incKernel(int *data, int N) {
+__global__ void incKernel(int *data, int N)
+{
     int i = blockIdx.x * blockDim.x + threadIdx.x;
 
-  if (i < N) data[i]++;
+    if (i < N)
+        data[i]++;
 }
 
-CUT_THREADPROC launch(void *void_arg) {
+CUT_THREADPROC launch(void *void_arg)
+{
     heterogeneous_workload *workload = (heterogeneous_workload *)void_arg;
 
     // Select GPU for this CPU thread
@@ -81,11 +84,8 @@ CUT_THREADPROC launch(void *void_arg) {
 
     // Allocate Resources
     checkCudaErrors(cudaStreamCreate(&workload->stream));
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc(&workload->d_data, N_elements_per_workload * sizeof(int)));
-      cudaMalloc(&workload->d_data, N_elements_per_workload * sizeof(int)));
+    checkCudaErrors(cudaHostAlloc(&workload->h_data, N_elements_per_workload * sizeof(int), cudaHostAllocPortable));
-  checkCudaErrors(cudaHostAlloc(&workload->h_data,
-                                N_elements_per_workload * sizeof(int),
-                                cudaHostAllocPortable));
 
     // CPU thread generates data
     for (int i = 0; i < N_elements_per_workload; ++i) {
@@ -97,25 +97,28 @@ CUT_THREADPROC launch(void *void_arg) {
     dim3 block(512);
     dim3 grid((N_elements_per_workload + block.x - 1) / block.x);
 
-  checkCudaErrors(cudaMemcpyAsync(workload->d_data, workload->h_data,
+    checkCudaErrors(cudaMemcpyAsync(workload->d_data,
+                                    workload->h_data,
                                     N_elements_per_workload * sizeof(int),
-                                  cudaMemcpyHostToDevice, workload->stream));
+                                    cudaMemcpyHostToDevice,
-  incKernel<<<grid, block, 0, workload->stream>>>(workload->d_data,
+                                    workload->stream));
-                                                  N_elements_per_workload);
+    incKernel<<<grid, block, 0, workload->stream>>>(workload->d_data, N_elements_per_workload);
-  checkCudaErrors(cudaMemcpyAsync(workload->h_data, workload->d_data,
+    checkCudaErrors(cudaMemcpyAsync(workload->h_data,
+                                    workload->d_data,
                                     N_elements_per_workload * sizeof(int),
-                                  cudaMemcpyDeviceToHost, workload->stream));
+                                    cudaMemcpyDeviceToHost,
+                                    workload->stream));
 
     // New in CUDA 5.0: Add a CPU callback which is called once all currently
     // pending operations in the CUDA stream have finished
-  checkCudaErrors(
+    checkCudaErrors(cudaStreamAddCallback(workload->stream, myStreamCallback, workload, 0));
-      cudaStreamAddCallback(workload->stream, myStreamCallback, workload, 0));
 
     CUT_THREADEND;
     // CPU thread end of life, GPU continues to process data...
 }
 
-CUT_THREADPROC postprocess(void *void_arg) {
+CUT_THREADPROC postprocess(void *void_arg)
+{
     heterogeneous_workload *workload = (heterogeneous_workload *)void_arg;
     // ... GPU is done with processing, continue on new CPU thread...
 
@@ -140,8 +143,8 @@ CUT_THREADPROC postprocess(void *void_arg) {
     CUT_THREADEND;
 }
 
-void CUDART_CB myStreamCallback(cudaStream_t stream, cudaError_t status,
+void CUDART_CB myStreamCallback(cudaStream_t stream, cudaError_t status, void *data)
-                                void *data) {
+{
     // Check status of GPU after stream operations are done
     checkCudaErrors(status);
 
@@ -149,7 +152,8 @@ void CUDART_CB myStreamCallback(cudaStream_t stream, cudaError_t status,
     cutStartThread(postprocess, data);
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     int N_gpus, max_gpus = 0;
     int gpuInfo[32]; // assume a maximum of 32 GPUs in a system configuration
 
@@ -168,10 +172,8 @@ int main(int argc, char **argv) {
         cudaSetDevice(devid);
         cudaGetDeviceProperties(&deviceProp, devid);
         SMversion = deviceProp.major << 4 + deviceProp.minor;
-    printf("GPU[%d] %s supports SM %d.%d", devid, deviceProp.name,
+        printf("GPU[%d] %s supports SM %d.%d", devid, deviceProp.name, deviceProp.major, deviceProp.minor);
-           deviceProp.major, deviceProp.minor);
+        printf(", %s GPU Callback Functions\n", (SMversion >= 0x11) ? "capable" : "NOT capable");
-    printf(", %s GPU Callback Functions\n",
-           (SMversion >= 0x11) ? "capable" : "NOT capable");
 
         if (SMversion >= 0x11) {
             gpuInfo[max_gpus++] = devid;
@@ -181,8 +183,7 @@ int main(int argc, char **argv) {
     printf("%d GPUs available to run Callback Functions\n", max_gpus);
 
     heterogeneous_workload *workloads;
-  workloads = (heterogeneous_workload *)malloc(N_workloads *
+    workloads = (heterogeneous_workload *)malloc(N_workloads * sizeof(heterogeneous_workload));
-                                               sizeof(heterogeneous_workload));
     ;
     thread_barrier = cutCreateBarrier(N_workloads);
 

Samples/0_Introduction/simpleCooperativeGroups/simpleCooperativeGroups.cu

@@ -38,8 +38,8 @@
  *
  */
 
-#include <stdio.h>
 #include <cooperative_groups.h>
+#include <stdio.h>
 
 using namespace cooperative_groups;
 
@@ -49,7 +49,8 @@ using namespace cooperative_groups;
  * calculates the sum of val across the group g. The workspace array, x,
  * must be large enough to contain g.size() integers.
  */
-__device__ int sumReduction(thread_group g, int *x, int val) {
+__device__ int sumReduction(thread_group g, int *x, int val)
+{
     // rank of this thread in the group
     int lane = g.thread_rank();
 
@@ -85,7 +86,8 @@ __device__ int sumReduction(thread_group g, int *x, int val) {
  *
  * Creates cooperative groups and performs reductions
  */
-__global__ void cgkernel() {
+__global__ void cgkernel()
+{
     // threadBlockGroup includes all threads in the block
     thread_block threadBlockGroup     = this_thread_block();
     int          threadBlockGroupSize = threadBlockGroup.size();
@@ -107,24 +109,22 @@ __global__ void cgkernel() {
 
     // master thread in group prints out result
     if (threadBlockGroup.thread_rank() == 0) {
-    printf(
+        printf(" Sum of all ranks 0..%d in threadBlockGroup is %d (expected %d)\n\n",
-        " Sum of all ranks 0..%d in threadBlockGroup is %d (expected %d)\n\n",
+               (int)threadBlockGroup.size() - 1,
-        (int)threadBlockGroup.size() - 1, output, expectedOutput);
+               output,
+               expectedOutput);
 
-    printf(" Now creating %d groups, each of size 16 threads:\n\n",
+        printf(" Now creating %d groups, each of size 16 threads:\n\n", (int)threadBlockGroup.size() / 16);
-           (int)threadBlockGroup.size() / 16);
     }
 
     threadBlockGroup.sync();
 
     // each tiledPartition16 group includes 16 threads
-  thread_block_tile<16> tiledPartition16 =
+    thread_block_tile<16> tiledPartition16 = tiled_partition<16>(threadBlockGroup);
-      tiled_partition<16>(threadBlockGroup);
 
     // This offset allows each group to have its own unique area in the workspace
     // array
-  int workspaceOffset =
+    int workspaceOffset = threadBlockGroup.thread_rank() - tiledPartition16.thread_rank();
-      threadBlockGroup.thread_rank() - tiledPartition16.thread_rank();
 
     // input to reduction, for each thread, is its' rank in the group
     input = tiledPartition16.thread_rank();
@@ -138,10 +138,10 @@ __global__ void cgkernel() {
 
     // each master thread prints out result
     if (tiledPartition16.thread_rank() == 0)
-    printf(
+        printf("   Sum of all ranks 0..15 in this tiledPartition16 group is %d "
-        "   Sum of all ranks 0..15 in this tiledPartition16 group is %d "
                "(expected %d)\n",
-        output, expectedOutput);
+               output,
+               expectedOutput);
 
     return;
 }
@@ -149,7 +149,8 @@ __global__ void cgkernel() {
 /**
  * Host main routine
  */
-int main() {
+int main()
+{
     // Error code to check return values for CUDA calls
     cudaError_t err;
 
@@ -166,8 +167,7 @@ int main() {
     err = cudaDeviceSynchronize();
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to launch kernel (error code %s)!\n",
+        fprintf(stderr, "Failed to launch kernel (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 

Samples/0_Introduction/simpleCubemapTexture/simpleCubemapTexture.cu

@@ -36,17 +36,17 @@
  */
 
 // includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 // includes CUDA
 #include <cuda_runtime.h>
 
 // helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 static const char *sSDKname = "simpleCubemapTexture";
 
@@ -56,8 +56,8 @@ static const char *sSDKname = "simpleCubemapTexture";
 //! Transform a cubemap face of a linear buffe using cubemap texture lookups
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void transformKernel(float *g_odata, int width,
+__global__ void transformKernel(float *g_odata, int width, cudaTextureObject_t tex)
-                                cudaTextureObject_t tex) {
+{
     // calculate this thread's data point
     unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
     unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
@@ -110,15 +110,15 @@ __global__ void transformKernel(float *g_odata, int width,
         }
 
         // read from texture, do expected transformation and write to global memory
-    g_odata[face * width * width + y * width + x] =
+        g_odata[face * width * width + y * width + x] = -texCubemap<float>(tex, cx, cy, cz);
-        -texCubemap<float>(tex, cx, cy, cz);
     }
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     // use command-line specified CUDA device, otherwise use device with highest
     // Gflops/s
     int devID = findCudaDevice(argc, (const char **)argv);
@@ -129,13 +129,11 @@ int main(int argc, char **argv) {
     cudaDeviceProp deviceProps;
 
     checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
-  printf("CUDA device [%s] has %d Multi-Processors ", deviceProps.name,
+    printf("CUDA device [%s] has %d Multi-Processors ", deviceProps.name, deviceProps.multiProcessorCount);
-         deviceProps.multiProcessorCount);
     printf("SM %d.%d\n", deviceProps.major, deviceProps.minor);
 
     if (deviceProps.major < 2) {
-    printf(
+        printf("%s requires SM 2.0 or higher for support of Texture Arrays.  Test "
-        "%s requires SM 2.0 or higher for support of Texture Arrays.  Test "
                "will exit... \n",
                sSDKname);
 
@@ -157,8 +155,7 @@ int main(int argc, char **argv) {
 
     for (unsigned int layer = 0; layer < num_layers; layer++) {
         for (int i = 0; i < (int)(cubemap_size); i++) {
-      h_data_ref[layer * cubemap_size + i] =
+            h_data_ref[layer * cubemap_size + i] = -h_data[layer * cubemap_size + i] + layer;
-          -h_data[layer * cubemap_size + i] + layer;
         }
     }
 
@@ -167,19 +164,16 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaMalloc((void **)&d_data, size));
 
     // allocate array and copy image data
-  cudaChannelFormatDesc channelDesc =
+    cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
-      cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
     cudaArray            *cu_3darray;
     //    checkCudaErrors(cudaMalloc3DArray( &cu_3darray, &channelDesc,
     //    make_cudaExtent(width, height, num_layers), cudaArrayLayered ));
-  checkCudaErrors(cudaMalloc3DArray(&cu_3darray, &channelDesc,
+    checkCudaErrors(
-                                    make_cudaExtent(width, width, num_faces),
+        cudaMalloc3DArray(&cu_3darray, &channelDesc, make_cudaExtent(width, width, num_faces), cudaArrayCubemap));
-                                    cudaArrayCubemap));
     cudaMemcpy3DParms myparms = {0};
     myparms.srcPos            = make_cudaPos(0, 0, 0);
     myparms.dstPos            = make_cudaPos(0, 0, 0);
-  myparms.srcPtr =
+    myparms.srcPtr            = make_cudaPitchedPtr(h_data, width * sizeof(float), width, width);
-      make_cudaPitchedPtr(h_data, width * sizeof(float), width, width);
     myparms.dstArray          = cu_3darray;
     myparms.extent            = make_cudaExtent(width, width, num_faces);
     myparms.kind              = cudaMemcpyHostToDevice;
@@ -207,10 +201,12 @@ int main(int argc, char **argv) {
     dim3 dimBlock(8, 8, 1);
     dim3 dimGrid(width / dimBlock.x, width / dimBlock.y, 1);
 
-  printf(
+    printf("Covering Cubemap data array of %d~3 x %d: Grid size is %d x %d, each "
-      "Covering Cubemap data array of %d~3 x %d: Grid size is %d x %d, each "
            "block has 8 x 8 threads\n",
-      width, num_layers, dimGrid.x, dimGrid.y);
+           width,
+           num_layers,
+           dimGrid.x,
+           dimGrid.y);
 
     transformKernel<<<dimGrid, dimBlock>>>(d_data, width,
                                            tex); // warmup (for better timing)
@@ -233,8 +229,7 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaDeviceSynchronize());
     sdkStopTimer(&timer);
     printf("Processing time: %.3f msec\n", sdkGetTimerValue(&timer));
-  printf("%.2f Mtexlookups/sec\n",
+    printf("%.2f Mtexlookups/sec\n", (cubemap_size / (sdkGetTimerValue(&timer) / 1000.0f) / 1e6));
-         (cubemap_size / (sdkGetTimerValue(&timer) / 1000.0f) / 1e6));
     sdkDeleteTimer(&timer);
 
     // allocate mem for the result on host side
@@ -245,14 +240,13 @@ int main(int argc, char **argv) {
     // write regression file if necessary
     if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
         // write file for regression test
-    sdkWriteFile<float>("./data/regression.dat", h_odata, width * width, 0.0f,
+        sdkWriteFile<float>("./data/regression.dat", h_odata, width * width, 0.0f, false);
-                        false);
+    }
-  } else {
+    else {
         printf("Comparing kernel output to expected data\n");
 
 #define MIN_EPSILON_ERROR 5e-3f
-    bResult =
+        bResult = compareData(h_odata, h_data_ref, cubemap_size, MIN_EPSILON_ERROR, 0.0f);
-        compareData(h_odata, h_data_ref, cubemap_size, MIN_EPSILON_ERROR, 0.0f);
     }
 
     // cleanup memory

Samples/0_Introduction/simpleDrvRuntime/simpleDrvRuntime.cpp

@@ -33,12 +33,12 @@
  */
 
 // Includes
+#include <cstring>
 #include <cuda.h>
 #include <cuda_runtime.h>
+#include <iostream>
 #include <stdio.h>
 #include <string.h>
-#include <cstring>
-#include <iostream>
 
 // includes, project
 #include <helper_cuda.h>
@@ -66,11 +66,10 @@ int CleanupNoFailure(CUcontext &cuContext);
 void RandomInit(float *, int);
 bool findModulePath(const char *, string &, char **, ostringstream &);
 
-static void check(CUresult result, char const *const func,
+static void check(CUresult result, char const *const func, const char *const file, int const line)
-                  const char *const file, int const line) {
+{
     if (result) {
-    fprintf(stderr, "CUDA error at %s:%d code=%d \"%s\" \n", file, line,
+        fprintf(stderr, "CUDA error at %s:%d code=%d \"%s\" \n", file, line, static_cast<unsigned int>(result), func);
-            static_cast<unsigned int>(result), func);
         exit(EXIT_FAILURE);
     }
 }
@@ -78,7 +77,8 @@ static void check(CUresult result, char const *const func,
 #define checkCudaDrvErrors(val) check((val), #val, __FILE__, __LINE__)
 
 // Host code
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("simpleDrvRuntime..\n");
     int        N = 50000, devID = 0;
     size_t     size = N * sizeof(float);
@@ -100,7 +100,8 @@ int main(int argc, char **argv) {
 
     if (!findModulePath(FATBIN_FILE, module_path, argv, fatbin)) {
         exit(EXIT_FAILURE);
-  } else {
+    }
+    else {
         printf("> initCUDA loading module: <%s>\n", module_path.c_str());
     }
 
@@ -113,8 +114,7 @@ int main(int argc, char **argv) {
     checkCudaDrvErrors(cuModuleLoadData(&cuModule, fatbin.str().c_str()));
 
     // Get function handle from module
-  checkCudaDrvErrors(
+    checkCudaDrvErrors(cuModuleGetFunction(&vecAdd_kernel, cuModule, "VecAdd_kernel"));
-      cuModuleGetFunction(&vecAdd_kernel, cuModule, "VecAdd_kernel"));
 
     // Allocate input vectors h_A and h_B in host memory
     checkCudaErrors(cudaMallocHost(&h_A, size));
@@ -133,10 +133,8 @@ int main(int argc, char **argv) {
     cudaStream_t stream;
     checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
     // Copy vectors from host memory to device memory
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(d_A, h_A, size, cudaMemcpyHostToDevice, stream));
-      cudaMemcpyAsync(d_A, h_A, size, cudaMemcpyHostToDevice, stream));
+    checkCudaErrors(cudaMemcpyAsync(d_B, h_B, size, cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(
-      cudaMemcpyAsync(d_B, h_B, size, cudaMemcpyHostToDevice, stream));
 
     int threadsPerBlock = 256;
     int blocksPerGrid   = (N + threadsPerBlock - 1) / threadsPerBlock;
@@ -144,14 +142,12 @@ int main(int argc, char **argv) {
     void *args[] = {&d_A, &d_B, &d_C, &N};
 
     // Launch the CUDA kernel
-  checkCudaDrvErrors(cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1,
+    checkCudaDrvErrors(
-                                    threadsPerBlock, 1, 1, 0, stream, args,
+        cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1, threadsPerBlock, 1, 1, 0, stream, args, NULL));
-                                    NULL));
 
     // Copy result from device memory to host memory
     // h_C contains the result in host memory
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(h_C, d_C, size, cudaMemcpyDeviceToHost, stream));
-      cudaMemcpyAsync(h_C, d_C, size, cudaMemcpyDeviceToHost, stream));
     checkCudaErrors(cudaStreamSynchronize(stream));
     // Verify result
     int i;
@@ -171,7 +167,8 @@ int main(int argc, char **argv) {
     exit((i == N) ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-int CleanupNoFailure(CUcontext &cuContext) {
+int CleanupNoFailure(CUcontext &cuContext)
+{
     // Free device memory
     checkCudaErrors(cudaFree(d_A));
     checkCudaErrors(cudaFree(d_B));
@@ -195,19 +192,21 @@ int CleanupNoFailure(CUcontext &cuContext) {
     return EXIT_SUCCESS;
 }
 // Allocates an array with random float entries.
-void RandomInit(float *data, int n) {
+void RandomInit(float *data, int n)
+{
     for (int i = 0; i < n; ++i) {
         data[i] = rand() / (float)RAND_MAX;
     }
 }
 
-bool inline findModulePath(const char *module_file, string &module_path,
+bool inline findModulePath(const char *module_file, string &module_path, char **argv, ostringstream &ostrm)
-                           char **argv, ostringstream &ostrm) {
+{
     char *actual_path = sdkFindFilePath(module_file, argv[0]);
 
     if (actual_path) {
         module_path = actual_path;
-  } else {
+    }
+    else {
         printf("> findModulePath file not found: <%s> \n", module_file);
         return false;
     }
@@ -215,7 +214,8 @@ bool inline findModulePath(const char *module_file, string &module_path,
     if (module_path.empty()) {
         printf("> findModulePath could not find file: <%s> \n", module_file);
         return false;
-  } else {
+    }
+    else {
         printf("> findModulePath found file at <%s>\n", module_path.c_str());
         if (module_path.rfind("fatbin") != string::npos) {
             ifstream fileIn(module_path.c_str(), ios::binary);

Samples/0_Introduction/simpleDrvRuntime/vectorAdd_kernel.cu

@@ -34,9 +34,10 @@
  */
 
 // Device code
-extern "C" __global__ void VecAdd_kernel(const float *A, const float *B,
+extern "C" __global__ void VecAdd_kernel(const float *A, const float *B, float *C, int N)
-                                         float *C, int N) {
+{
     int i = blockDim.x * blockIdx.x + threadIdx.x;
 
-  if (i < N) C[i] = A[i] + B[i];
+    if (i < N)
+        C[i] = A[i] + B[i];
 }

Samples/0_Introduction/simpleHyperQ/simpleHyperQ.cu

@@ -44,7 +44,8 @@ const char *sSDKsample = "hyperQ";
 
 // This subroutine does no real work but runs for at least the specified number
 // of clock ticks.
-__device__ void clock_block(clock_t *d_o, clock_t clock_count) {
+__device__ void clock_block(clock_t *d_o, clock_t clock_count)
+{
     unsigned int start_clock = (unsigned int)clock();
 
     clock_t clock_offset = 0;
@@ -71,15 +72,12 @@ __device__ void clock_block(clock_t *d_o, clock_t clock_count) {
 // We create two identical kernels calling clock_block(), we create two so that
 // we can identify dependencies in the profile timeline ("kernel_B" is always
 // dependent on "kernel_A" in the same stream).
-__global__ void kernel_A(clock_t *d_o, clock_t clock_count) {
+__global__ void kernel_A(clock_t *d_o, clock_t clock_count) { clock_block(d_o, clock_count); }
-  clock_block(d_o, clock_count);
+__global__ void kernel_B(clock_t *d_o, clock_t clock_count) { clock_block(d_o, clock_count); }
-}
-__global__ void kernel_B(clock_t *d_o, clock_t clock_count) {
-  clock_block(d_o, clock_count);
-}
 
 // Single-warp reduction kernel (note: this is not optimized for simplicity)
-__global__ void sum(clock_t *d_clocks, int N) {
+__global__ void sum(clock_t *d_clocks, int N)
+{
     // Handle to thread block group
     cg::thread_block   cta = cg::this_thread_block();
     __shared__ clock_t s_clocks[32];
@@ -106,7 +104,8 @@ __global__ void sum(clock_t *d_clocks, int N) {
     }
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     int   nstreams    = 32; // One stream for each pair of kernels
     float kernel_time = 10; // Time each kernel should run in ms
     float elapsed_time;
@@ -131,18 +130,20 @@ int main(int argc, char **argv) {
     // HyperQ is available in devices of Compute Capability 3.5 and higher
     if (deviceProp.major < 3 || (deviceProp.major == 3 && deviceProp.minor < 5)) {
         if (deviceProp.concurrentKernels == 0) {
-      printf(
+            printf("> GPU does not support concurrent kernel execution (SM 3.5 or "
-          "> GPU does not support concurrent kernel execution (SM 3.5 or "
                    "higher required)\n");
             printf("  CUDA kernel runs will be serialized\n");
-    } else {
+        }
+        else {
             printf("> GPU does not support HyperQ\n");
             printf("  CUDA kernel runs will have limited concurrency\n");
         }
     }
 
     printf("> Detected Compute SM %d.%d hardware with %d multi-processors\n",
-         deviceProp.major, deviceProp.minor, deviceProp.multiProcessorCount);
+           deviceProp.major,
+           deviceProp.minor,
+           deviceProp.multiProcessorCount);
 
     // Allocate host memory for the output (reduced to a single value)
     clock_t *a = 0;
@@ -153,8 +154,7 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaMalloc((void **)&d_a, 2 * nstreams * sizeof(clock_t)));
 
     // Allocate and initialize an array of stream handles
-  cudaStream_t *streams =
+    cudaStream_t *streams = (cudaStream_t *)malloc(nstreams * sizeof(cudaStream_t));
-      (cudaStream_t *)malloc(nstreams * sizeof(cudaStream_t));
 
     for (int i = 0; i < nstreams; i++) {
         checkCudaErrors(cudaStreamCreate(&(streams[i])));
@@ -203,15 +203,15 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaEventSynchronize(stop_event));
     checkCudaErrors(cudaEventElapsedTime(&elapsed_time, start_event, stop_event));
 
-  printf(
+    printf("Expected time for serial execution of %d sets of kernels is between "
-      "Expected time for serial execution of %d sets of kernels is between "
            "approx. %.3fs and %.3fs\n",
-      nstreams, (nstreams + 1) * kernel_time / 1000.0f,
+           nstreams,
+           (nstreams + 1) * kernel_time / 1000.0f,
            2 * nstreams * kernel_time / 1000.0f);
-  printf(
+    printf("Expected time for fully concurrent execution of %d sets of kernels is "
-      "Expected time for fully concurrent execution of %d sets of kernels is "
            "approx. %.3fs\n",
-      nstreams, 2 * kernel_time / 1000.0f);
+           nstreams,
+           2 * kernel_time / 1000.0f);
     printf("Measured time for sample = %.3fs\n", elapsed_time / 1000.0f);
 
     bool bTestResult = (a[0] >= total_clocks);

Samples/0_Introduction/simpleIPC/README.md

@@ -34,4 +34,3 @@ Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-d
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)
-

Samples/0_Introduction/simpleIPC/simpleIPC.cu

@@ -32,6 +32,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <vector>
+
 #include "helper_cuda.h"
 #include "helper_multiprocess.h"
 static const char shmName[] = "simpleIPCshm";
@@ -49,7 +50,8 @@ static const char shmName[] = "simpleIPCshm";
 #error Unsupported system
 #endif
 
-typedef struct shmStruct_st {
+typedef struct shmStruct_st
+{
     size_t               nprocesses;
     int                  barrier;
     int                  sense;
@@ -58,15 +60,16 @@ typedef struct shmStruct_st {
     cudaIpcEventHandle_t eventHandle[MAX_DEVICES];
 } shmStruct;
 
-__global__ void simpleKernel(char *ptr, int sz, char val) {
+__global__ void simpleKernel(char *ptr, int sz, char val)
+{
     int idx = blockIdx.x * blockDim.x + threadIdx.x;
     for (; idx < sz; idx += (gridDim.x * blockDim.x)) {
         ptr[idx] = val;
     }
 }
 
-static void barrierWait(volatile int *barrier, volatile int *sense,
+static void barrierWait(volatile int *barrier, volatile int *sense, unsigned int n)
-                        unsigned int n) {
+{
     int count;
 
     // Check-in
@@ -84,7 +87,8 @@ static void barrierWait(volatile int *barrier, volatile int *sense,
         ;
 }
 
-static void childProcess(int id) {
+static void childProcess(int id)
+{
     volatile shmStruct      *shm = NULL;
     cudaStream_t             stream;
     sharedMemoryInfo         info;
@@ -108,8 +112,7 @@ static void childProcess(int id) {
     checkCudaErrors(cudaSetDevice(shm->devices[id]));
     checkCudaErrors(cudaGetDeviceProperties(&prop, shm->devices[id]));
     checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
-  checkCudaErrors(cudaOccupancyMaxActiveBlocksPerMultiprocessor(
+    checkCudaErrors(cudaOccupancyMaxActiveBlocksPerMultiprocessor(&blocks, simpleKernel, threads, 0));
-      &blocks, simpleKernel, threads, 0));
     blocks *= prop.multiProcessorCount;
 
     // Open and track all the allocations and events created in the master
@@ -121,10 +124,8 @@ static void childProcess(int id) {
         // Notice, we don't need to explicitly enable peer access for
         // allocations on other devices.
         checkCudaErrors(
-        cudaIpcOpenMemHandle(&ptr, *(cudaIpcMemHandle_t *)&shm->memHandle[i],
+            cudaIpcOpenMemHandle(&ptr, *(cudaIpcMemHandle_t *)&shm->memHandle[i], cudaIpcMemLazyEnablePeerAccess));
-                             cudaIpcMemLazyEnablePeerAccess));
+        checkCudaErrors(cudaIpcOpenEventHandle(&event, *(cudaIpcEventHandle_t *)&shm->eventHandle[i]));
-    checkCudaErrors(cudaIpcOpenEventHandle(
-        &event, *(cudaIpcEventHandle_t *)&shm->eventHandle[i]));
 
         ptrs.push_back(ptr);
         events.push_back(event);
@@ -141,8 +142,7 @@ static void childProcess(int id) {
         // Wait for the buffer to be accessed to be ready
         checkCudaErrors(cudaStreamWaitEvent(stream, events[bufferId], 0));
         // Push a simple kernel on it
-    simpleKernel<<<blocks, threads, 0, stream>>>((char *)ptrs[bufferId],
+        simpleKernel<<<blocks, threads, 0, stream>>>((char *)ptrs[bufferId], DATA_SIZE, id);
-                                                 DATA_SIZE, id);
         checkCudaErrors(cudaGetLastError());
         // Signal that this buffer is ready for the next consumer
         checkCudaErrors(cudaEventRecord(events[bufferId], stream));
@@ -158,8 +158,7 @@ static void childProcess(int id) {
 
     // Now wait for my buffer to be ready so I can copy it locally and verify it
     checkCudaErrors(cudaStreamWaitEvent(stream, events[id], 0));
-  checkCudaErrors(cudaMemcpyAsync(&verification_buffer[0], ptrs[id], DATA_SIZE,
+    checkCudaErrors(cudaMemcpyAsync(&verification_buffer[0], ptrs[id], DATA_SIZE, cudaMemcpyDeviceToHost, stream));
-                                  cudaMemcpyDeviceToHost, stream));
     // And wait for all the queued up work to complete
     checkCudaErrors(cudaStreamSynchronize(stream));
 
@@ -169,8 +168,11 @@ static void childProcess(int id) {
     char compareId = (char)((id + 1) % procCount);
     for (unsigned long long j = 0; j < DATA_SIZE; j++) {
         if (verification_buffer[j] != compareId) {
-      printf("Process %d: Verification mismatch at %lld: %d != %d\n", id, j,
+            printf("Process %d: Verification mismatch at %lld: %d != %d\n",
-             (int)verification_buffer[j], (int)compareId);
+                   id,
+                   j,
+                   (int)verification_buffer[j],
+                   (int)compareId);
         }
     }
 
@@ -185,7 +187,8 @@ static void childProcess(int id) {
     printf("Process %d complete!\n", id);
 }
 
-static void parentProcess(char *app) {
+static void parentProcess(char *app)
+{
     sharedMemoryInfo         info;
     int                      devCount, i;
     volatile shmStruct      *shm = NULL;
@@ -219,17 +222,14 @@ static void parentProcess(char *app) {
         // This sample requires two processes accessing each device, so we need
         // to ensure exclusive or prohibited mode is not set
         if (prop.computeMode != cudaComputeModeDefault) {
-      printf("Device %d is in an unsupported compute mode for this sample\n",
+            printf("Device %d is in an unsupported compute mode for this sample\n", i);
-             i);
             continue;
         }
 
         for (int j = 0; j < shm->nprocesses; j++) {
             int canAccessPeerIJ, canAccessPeerJI;
-      checkCudaErrors(
+            checkCudaErrors(cudaDeviceCanAccessPeer(&canAccessPeerJI, shm->devices[j], i));
-          cudaDeviceCanAccessPeer(&canAccessPeerJI, shm->devices[j], i));
+            checkCudaErrors(cudaDeviceCanAccessPeer(&canAccessPeerIJ, i, shm->devices[j]));
-      checkCudaErrors(
-          cudaDeviceCanAccessPeer(&canAccessPeerIJ, i, shm->devices[j]));
             if (!canAccessPeerIJ || !canAccessPeerJI) {
                 allPeers = false;
                 break;
@@ -246,10 +246,11 @@ static void parentProcess(char *app) {
                 checkCudaErrors(cudaDeviceEnablePeerAccess(i, 0));
             }
             shm->devices[shm->nprocesses++] = i;
-      if (shm->nprocesses >= MAX_DEVICES) break;
+            if (shm->nprocesses >= MAX_DEVICES)
-    } else {
+                break;
-      printf(
+        }
-          "Device %d is not peer capable with some other selected peers, "
+        else {
+            printf("Device %d is not peer capable with some other selected peers, "
                    "skipping\n",
                    i);
         }
@@ -268,12 +269,9 @@ static void parentProcess(char *app) {
 
         checkCudaErrors(cudaSetDevice(shm->devices[i]));
         checkCudaErrors(cudaMalloc(&ptr, DATA_SIZE));
-    checkCudaErrors(
+        checkCudaErrors(cudaIpcGetMemHandle((cudaIpcMemHandle_t *)&shm->memHandle[i], ptr));
-        cudaIpcGetMemHandle((cudaIpcMemHandle_t *)&shm->memHandle[i], ptr));
+        checkCudaErrors(cudaEventCreate(&event, cudaEventDisableTiming | cudaEventInterprocess));
-    checkCudaErrors(cudaEventCreate(
+        checkCudaErrors(cudaIpcGetEventHandle((cudaIpcEventHandle_t *)&shm->eventHandle[i], event));
-        &event, cudaEventDisableTiming | cudaEventInterprocess));
-    checkCudaErrors(cudaIpcGetEventHandle(
-        (cudaIpcEventHandle_t *)&shm->eventHandle[i], event));
 
         ptrs.push_back(ptr);
         events.push_back(event);
@@ -314,14 +312,16 @@ static void parentProcess(char *app) {
     sharedMemoryClose(&info);
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
 #if defined(__arm__) || defined(__aarch64__)
     printf("Not supported on ARM\n");
     return EXIT_WAIVED;
 #else
     if (argc == 1) {
         parentProcess(argv[0]);
-  } else {
+    }
+    else {
         childProcess(atoi(argv[1]));
     }
     return EXIT_SUCCESS;

Samples/0_Introduction/simpleLayeredTexture/simpleLayeredTexture.cu

@@ -36,10 +36,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 // includes, kernels
 #include <cuda_runtime.h>
@@ -54,8 +54,8 @@ static const char *sSDKname = "simpleLayeredTexture";
 //! Transform a layer of a layered 2D texture using texture lookups
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void transformKernel(float *g_odata, int width, int height,
+__global__ void transformKernel(float *g_odata, int width, int height, int layer, cudaTextureObject_t tex)
-                                int layer, cudaTextureObject_t tex) {
+{
     // calculate this thread's data point
     unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
     unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
@@ -67,14 +67,14 @@ __global__ void transformKernel(float *g_odata, int width, int height,
     float v = (y + 0.5f) / (float)height;
 
     // read from texture, do expected transformation and write to global memory
-  g_odata[layer * width * height + y * width + x] =
+    g_odata[layer * width * height + y * width + x] = -tex2DLayered<float>(tex, u, v, layer) + layer;
-      -tex2DLayered<float>(tex, u, v, layer) + layer;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("[%s] - Starting...\n", sSDKname);
 
     // use command-line specified CUDA device, otherwise use device with highest
@@ -87,8 +87,7 @@ int main(int argc, char **argv) {
     cudaDeviceProp deviceProps;
 
     checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
-  printf("CUDA device [%s] has %d Multi-Processors ", deviceProps.name,
+    printf("CUDA device [%s] has %d Multi-Processors ", deviceProps.name, deviceProps.multiProcessorCount);
-         deviceProps.multiProcessorCount);
     printf("SM %d.%d\n", deviceProps.major, deviceProps.minor);
 
     // generate input data for layered texture
@@ -106,8 +105,7 @@ int main(int argc, char **argv) {
 
     for (unsigned int layer = 0; layer < num_layers; layer++)
         for (int i = 0; i < (int)(width * height); i++) {
-      h_data_ref[layer * width * height + i] =
+            h_data_ref[layer * width * height + i] = -h_data[layer * width * height + i] + layer;
-          -h_data[layer * width * height + i] + layer;
         }
 
     // allocate device memory for result
@@ -115,17 +113,14 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaMalloc((void **)&d_data, size));
 
     // allocate array and copy image data
-  cudaChannelFormatDesc channelDesc =
+    cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
-      cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
     cudaArray            *cu_3darray;
-  checkCudaErrors(cudaMalloc3DArray(&cu_3darray, &channelDesc,
+    checkCudaErrors(
-                                    make_cudaExtent(width, height, num_layers),
+        cudaMalloc3DArray(&cu_3darray, &channelDesc, make_cudaExtent(width, height, num_layers), cudaArrayLayered));
-                                    cudaArrayLayered));
     cudaMemcpy3DParms myparms = {0};
     myparms.srcPos            = make_cudaPos(0, 0, 0);
     myparms.dstPos            = make_cudaPos(0, 0, 0);
-  myparms.srcPtr =
+    myparms.srcPtr            = make_cudaPitchedPtr(h_data, width * sizeof(float), width, height);
-      make_cudaPitchedPtr(h_data, width * sizeof(float), width, height);
     myparms.dstArray          = cu_3darray;
     myparms.extent            = make_cudaExtent(width, height, num_layers);
     myparms.kind              = cudaMemcpyHostToDevice;
@@ -152,10 +147,12 @@ int main(int argc, char **argv) {
     dim3 dimBlock(8, 8, 1);
     dim3 dimGrid(width / dimBlock.x, height / dimBlock.y, 1);
 
-  printf(
+    printf("Covering 2D data array of %d x %d: Grid size is %d x %d, each block has "
-      "Covering 2D data array of %d x %d: Grid size is %d x %d, each block has "
            "8 x 8 threads\n",
-      width, height, dimGrid.x, dimGrid.y);
+           width,
+           height,
+           dimGrid.x,
+           dimGrid.y);
 
     transformKernel<<<dimGrid, dimBlock>>>(d_data, width, height, 0,
                                            tex); // warmup (for better timing)
@@ -171,8 +168,7 @@ int main(int argc, char **argv) {
 
     // execute the kernel
     for (unsigned int layer = 0; layer < num_layers; layer++)
-    transformKernel<<<dimGrid, dimBlock, 0>>>(d_data, width, height, layer,
+        transformKernel<<<dimGrid, dimBlock, 0>>>(d_data, width, height, layer, tex);
-                                              tex);
 
     // check if kernel execution generated an error
     getLastCudaError("Kernel execution failed");
@@ -180,9 +176,7 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaDeviceSynchronize());
     sdkStopTimer(&timer);
     printf("Processing time: %.3f msec\n", sdkGetTimerValue(&timer));
-  printf("%.2f Mtexlookups/sec\n",
+    printf("%.2f Mtexlookups/sec\n", (width * height * num_layers / (sdkGetTimerValue(&timer) / 1000.0f) / 1e6));
-         (width * height * num_layers / (sdkGetTimerValue(&timer) / 1000.0f) /
-          1e6));
     sdkDeleteTimer(&timer);
 
     // allocate mem for the result on host side
@@ -193,14 +187,13 @@ int main(int argc, char **argv) {
     // write regression file if necessary
     if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
         // write file for regression test
-    sdkWriteFile<float>("./data/regression.dat", h_odata, width * height, 0.0f,
+        sdkWriteFile<float>("./data/regression.dat", h_odata, width * height, 0.0f, false);
-                        false);
+    }
-  } else {
+    else {
         printf("Comparing kernel output to expected data\n");
 
 #define MIN_EPSILON_ERROR 5e-3f
-    bResult = compareData(h_odata, h_data_ref, width * height * num_layers,
+        bResult = compareData(h_odata, h_data_ref, width * height * num_layers, MIN_EPSILON_ERROR, 0.0f);
-                          MIN_EPSILON_ERROR, 0.0f);
     }
 
     // cleanup memory

Samples/0_Introduction/simpleMPI/simpleMPI.cpp

@@ -42,8 +42,8 @@
 // System includes
 #include <iostream>
 
-using std::cout;
 using std::cerr;
+using std::cout;
 using std::endl;
 
 // User include
@@ -58,7 +58,8 @@ using std::endl;
 
 // Host code
 // No CUDA here, only MPI
-int main(int argc, char *argv[]) {
+int main(int argc, char *argv[])
+{
     // Dimensions of the dataset
     int blockSize       = 256;
     int gridSize        = 10000;
@@ -87,8 +88,8 @@ int main(int argc, char *argv[]) {
     float *dataNode = new float[dataSizePerNode];
 
     // Dispatch a portion of the input data to each node
-  MPI_CHECK(MPI_Scatter(dataRoot, dataSizePerNode, MPI_FLOAT, dataNode,
+    MPI_CHECK(
-                        dataSizePerNode, MPI_FLOAT, 0, MPI_COMM_WORLD));
+        MPI_Scatter(dataRoot, dataSizePerNode, MPI_FLOAT, dataNode, dataSizePerNode, MPI_FLOAT, 0, MPI_COMM_WORLD));
 
     if (commRank == 0) {
         // No need for root data any more
@@ -102,8 +103,7 @@ int main(int argc, char *argv[]) {
     float sumNode = sum(dataNode, dataSizePerNode);
     float sumRoot;
 
-  MPI_CHECK(
+    MPI_CHECK(MPI_Reduce(&sumNode, &sumRoot, 1, MPI_FLOAT, MPI_SUM, 0, MPI_COMM_WORLD));
-      MPI_Reduce(&sumNode, &sumRoot, 1, MPI_FLOAT, MPI_SUM, 0, MPI_COMM_WORLD));
 
     if (commRank == 0) {
         float average = sumRoot / dataSizeTotal;
@@ -122,7 +122,8 @@ int main(int argc, char *argv[]) {
 }
 
 // Shut down MPI cleanly if something goes wrong
-void my_abort(int err) {
+void my_abort(int err)
+{
     cout << "Test FAILED\n";
     MPI_Abort(MPI_COMM_WORLD, err);
 }

Samples/0_Introduction/simpleMPI/simpleMPI.cu

@@ -51,13 +51,15 @@ using std::endl;
 
 // Device code
 // Very simple GPU Kernel that computes square roots of input numbers
-__global__ void simpleMPIKernel(float *input, float *output) {
+__global__ void simpleMPIKernel(float *input, float *output)
+{
     int tid     = blockIdx.x * blockDim.x + threadIdx.x;
     output[tid] = sqrt(input[tid]);
 }
 
 // Initialize an array with random data (between 0 and 1)
-void initData(float *data, int dataSize) {
+void initData(float *data, int dataSize)
+{
     for (int i = 0; i < dataSize; i++) {
         data[i] = (float)rand() / RAND_MAX;
     }
@@ -65,7 +67,8 @@ void initData(float *data, int dataSize) {
 
 // CUDA computation on each node
 // No MPI here, only CUDA
-void computeGPU(float *hostData, int blockSize, int gridSize) {
+void computeGPU(float *hostData, int blockSize, int gridSize)
+{
     int dataSize = blockSize * gridSize;
 
     // Allocate data on GPU memory
@@ -76,22 +79,21 @@ void computeGPU(float *hostData, int blockSize, int gridSize) {
     CUDA_CHECK(cudaMalloc((void **)&deviceOutputData, dataSize * sizeof(float)));
 
     // Copy to GPU memory
-  CUDA_CHECK(cudaMemcpy(deviceInputData, hostData, dataSize * sizeof(float),
+    CUDA_CHECK(cudaMemcpy(deviceInputData, hostData, dataSize * sizeof(float), cudaMemcpyHostToDevice));
-                        cudaMemcpyHostToDevice));
 
     // Run kernel
     simpleMPIKernel<<<gridSize, blockSize>>>(deviceInputData, deviceOutputData);
 
     // Copy data back to CPU memory
-  CUDA_CHECK(cudaMemcpy(hostData, deviceOutputData, dataSize * sizeof(float),
+    CUDA_CHECK(cudaMemcpy(hostData, deviceOutputData, dataSize * sizeof(float), cudaMemcpyDeviceToHost));
-                        cudaMemcpyDeviceToHost));
 
     // Free GPU memory
     CUDA_CHECK(cudaFree(deviceInputData));
     CUDA_CHECK(cudaFree(deviceOutputData));
 }
 
-float sum(float *data, int size) {
+float sum(float *data, int size)
+{
     float accum = 0.f;
 
     for (int i = 0; i < size; i++) {

Samples/0_Introduction/simpleMPI/simpleMPI.h

@@ -36,7 +36,8 @@
  */
 
 // Forward declarations
-extern "C" {
+extern "C"
+{
     void  initData(float *data, int dataSize);
     void  computeGPU(float *hostData, int blockSize, int gridSize);
     float sum(float *data, int size);

Samples/0_Introduction/simpleMultiCopy/simpleMultiCopy.cu

@@ -55,7 +55,8 @@ const char *sSDKname = "simpleMultiCopy";
 // includes, kernels
 // Declare the CUDA kernels here and main() code that is needed to launch
 // Compute workload on the system
-__global__ void incKernel(int *g_out, int *g_in, int N, int inner_reps) {
+__global__ void incKernel(int *g_out, int *g_in, int N, int inner_reps)
+{
     int idx = blockIdx.x * blockDim.x + threadIdx.x;
 
     if (idx < N) {
@@ -102,7 +103,8 @@ bool test();
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char *argv[]) {
+int main(int argc, char *argv[])
+{
     int            cuda_device = 0;
     float          scale_factor;
     cudaDeviceProp deviceProp;
@@ -115,7 +117,8 @@ int main(int argc, char *argv[]) {
         if (cuda_device < 0) {
             printf("Invalid command line parameters\n");
             exit(EXIT_FAILURE);
-    } else {
+        }
+        else {
             printf("cuda_device = %d\n", cuda_device);
             cuda_device = gpuDeviceInit(cuda_device);
 
@@ -124,7 +127,8 @@ int main(int argc, char *argv[]) {
                 exit(EXIT_SUCCESS);
             }
         }
-  } else {
+    }
+    else {
         // Otherwise pick the device with the highest Gflops/s
         cuda_device = gpuGetMaxGflopsDeviceId();
         checkCudaErrors(cudaSetDevice(cuda_device));
@@ -133,22 +137,23 @@ int main(int argc, char *argv[]) {
     }
 
     checkCudaErrors(cudaGetDeviceProperties(&deviceProp, cuda_device));
-  printf("[%s] has %d MP(s) x %d (Cores/MP) = %d (Cores)\n", deviceProp.name,
+    printf("[%s] has %d MP(s) x %d (Cores/MP) = %d (Cores)\n",
+           deviceProp.name,
            deviceProp.multiProcessorCount,
            _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor),
-         _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) *
+           _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) * deviceProp.multiProcessorCount);
-             deviceProp.multiProcessorCount);
 
     // Anything that is less than 32 Cores will have scaled down workload
     scale_factor =
-      max((32.0f / (_ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) *
+        max((32.0f / (_ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) * (float)deviceProp.multiProcessorCount)),
-                    (float)deviceProp.multiProcessorCount)),
             1.0f);
     N = (int)((float)N / scale_factor);
 
     printf("> Device name: %s\n", deviceProp.name);
     printf("> CUDA Capability %d.%d hardware with %d multi-processors\n",
-         deviceProp.major, deviceProp.minor, deviceProp.multiProcessorCount);
+           deviceProp.major,
+           deviceProp.minor,
+           deviceProp.multiProcessorCount);
     printf("> scale_factor = %.2f\n", 1.0f / scale_factor);
     printf("> array_size   = %d\n\n", N);
 
@@ -165,13 +170,11 @@ int main(int argc, char *argv[]) {
     h_data_sink   = (int *)malloc(memsize);
 
     for (int i = 0; i < STREAM_COUNT; ++i) {
-    checkCudaErrors(
+        checkCudaErrors(cudaHostAlloc(&h_data_in[i], memsize, cudaHostAllocDefault));
-        cudaHostAlloc(&h_data_in[i], memsize, cudaHostAllocDefault));
         checkCudaErrors(cudaMalloc(&d_data_in[i], memsize));
         checkCudaErrors(cudaMemset(d_data_in[i], 0, memsize));
 
-    checkCudaErrors(
+        checkCudaErrors(cudaHostAlloc(&h_data_out[i], memsize, cudaHostAllocDefault));
-        cudaHostAlloc(&h_data_out[i], memsize, cudaHostAllocDefault));
         checkCudaErrors(cudaMalloc(&d_data_out[i], memsize));
 
         checkCudaErrors(cudaStreamCreate(&stream[i]));
@@ -190,8 +193,7 @@ int main(int argc, char *argv[]) {
 
     // Time copies and kernel
     cudaEventRecord(start, 0);
-  checkCudaErrors(cudaMemcpyAsync(d_data_in[0], h_data_in[0], memsize,
+    checkCudaErrors(cudaMemcpyAsync(d_data_in[0], h_data_in[0], memsize, cudaMemcpyHostToDevice, 0));
-                                  cudaMemcpyHostToDevice, 0));
     cudaEventRecord(stop, 0);
     cudaEventSynchronize(stop);
 
@@ -199,8 +201,7 @@ int main(int argc, char *argv[]) {
     cudaEventElapsedTime(&memcpy_h2d_time, start, stop);
 
     cudaEventRecord(start, 0);
-  checkCudaErrors(cudaMemcpyAsync(h_data_out[0], d_data_out[0], memsize,
+    checkCudaErrors(cudaMemcpyAsync(h_data_out[0], d_data_out[0], memsize, cudaMemcpyDeviceToHost, 0));
-                                  cudaMemcpyDeviceToHost, 0));
     cudaEventRecord(stop, 0);
     cudaEventSynchronize(stop);
 
@@ -217,35 +218,27 @@ int main(int argc, char *argv[]) {
 
     printf("\n");
     printf("Relevant properties of this CUDA device\n");
-  printf(
+    printf("(%s) Can overlap one CPU<>GPU data transfer with GPU kernel execution "
-      "(%s) Can overlap one CPU<>GPU data transfer with GPU kernel execution "
            "(device property \"deviceOverlap\")\n",
            deviceProp.deviceOverlap ? "X" : " ");
     // printf("(%s) Can execute several GPU kernels simultaneously (compute
     // capability >= 2.0)\n", deviceProp.major >= 2 ? "X": " ");
-  printf(
+    printf("(%s) Can overlap two CPU<>GPU data transfers with GPU kernel execution\n"
-      "(%s) Can overlap two CPU<>GPU data transfers with GPU kernel execution\n"
            "    (Compute Capability >= 2.0 AND (Tesla product OR Quadro "
            "4000/5000/6000/K5000)\n",
            (deviceProp.major >= 2 && deviceProp.asyncEngineCount > 1) ? "X" : " ");
 
     printf("\n");
     printf("Measured timings (throughput):\n");
-  printf(" Memcpy host to device\t: %f ms (%f GB/s)\n", memcpy_h2d_time,
+    printf(" Memcpy host to device\t: %f ms (%f GB/s)\n", memcpy_h2d_time, (memsize * 1e-6) / memcpy_h2d_time);
-         (memsize * 1e-6) / memcpy_h2d_time);
+    printf(" Memcpy device to host\t: %f ms (%f GB/s)\n", memcpy_d2h_time, (memsize * 1e-6) / memcpy_d2h_time);
-  printf(" Memcpy device to host\t: %f ms (%f GB/s)\n", memcpy_d2h_time,
+    printf(" Kernel\t\t\t: %f ms (%f GB/s)\n", kernel_time, (inner_reps * memsize * 2e-6) / kernel_time);
-         (memsize * 1e-6) / memcpy_d2h_time);
-  printf(" Kernel\t\t\t: %f ms (%f GB/s)\n", kernel_time,
-         (inner_reps * memsize * 2e-6) / kernel_time);
 
     printf("\n");
-  printf(
+    printf("Theoretical limits for speedup gained from overlapped data "
-      "Theoretical limits for speedup gained from overlapped data "
            "transfers:\n");
-  printf("No overlap at all (transfer-kernel-transfer): %f ms \n",
+    printf("No overlap at all (transfer-kernel-transfer): %f ms \n", memcpy_h2d_time + memcpy_d2h_time + kernel_time);
-         memcpy_h2d_time + memcpy_d2h_time + kernel_time);
+    printf("Compute can overlap with one transfer: %f ms\n", max((memcpy_h2d_time + memcpy_d2h_time), kernel_time));
-  printf("Compute can overlap with one transfer: %f ms\n",
-         max((memcpy_h2d_time + memcpy_d2h_time), kernel_time));
     printf("Compute can overlap with both data transfers: %f ms\n",
            max(max(memcpy_h2d_time, memcpy_d2h_time), kernel_time));
 
@@ -254,18 +247,13 @@ int main(int argc, char *argv[]) {
     float overlap_time = processWithStreams(STREAM_COUNT);
 
     printf("\nAverage measured timings over %d repetitions:\n", nreps);
-  printf(" Avg. time when execution fully serialized\t: %f ms\n",
+    printf(" Avg. time when execution fully serialized\t: %f ms\n", serial_time / nreps);
-         serial_time / nreps);
+    printf(" Avg. time when overlapped using %d streams\t: %f ms\n", STREAM_COUNT, overlap_time / nreps);
-  printf(" Avg. time when overlapped using %d streams\t: %f ms\n", STREAM_COUNT,
+    printf(" Avg. speedup gained (serialized - overlapped)\t: %f ms\n", (serial_time - overlap_time) / nreps);
-         overlap_time / nreps);
-  printf(" Avg. speedup gained (serialized - overlapped)\t: %f ms\n",
-         (serial_time - overlap_time) / nreps);
 
     printf("\nMeasured throughput:\n");
-  printf(" Fully serialized execution\t\t: %f GB/s\n",
+    printf(" Fully serialized execution\t\t: %f GB/s\n", (nreps * (memsize * 2e-6)) / serial_time);
-         (nreps * (memsize * 2e-6)) / serial_time);
+    printf(" Overlapped using %d streams\t\t: %f GB/s\n", STREAM_COUNT, (nreps * (memsize * 2e-6)) / overlap_time);
-  printf(" Overlapped using %d streams\t\t: %f GB/s\n", STREAM_COUNT,
-         (nreps * (memsize * 2e-6)) / overlap_time);
 
     // Verify the results, we will use the results for final output
     bool bResults = test();
@@ -293,7 +281,8 @@ int main(int argc, char *argv[]) {
     exit(bResults ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-float processWithStreams(int streams_used) {
+float processWithStreams(int streams_used)
+{
     int current_stream = 0;
 
     float time;
@@ -326,17 +315,17 @@ float processWithStreams(int streams_used) {
             d_data_out[current_stream], d_data_in[current_stream], N, inner_reps);
 
         // Upload next frame
-    checkCudaErrors(
+        checkCudaErrors(cudaMemcpyAsync(
-        cudaMemcpyAsync(d_data_in[next_stream], h_data_in[next_stream], memsize,
+            d_data_in[next_stream], h_data_in[next_stream], memsize, cudaMemcpyHostToDevice, stream[next_stream]));
-                        cudaMemcpyHostToDevice, stream[next_stream]));
 
         // Download current frame
-    checkCudaErrors(cudaMemcpyAsync(
+        checkCudaErrors(cudaMemcpyAsync(h_data_out[current_stream],
-        h_data_out[current_stream], d_data_out[current_stream], memsize,
+                                        d_data_out[current_stream],
-        cudaMemcpyDeviceToHost, stream[current_stream]));
+                                        memsize,
+                                        cudaMemcpyDeviceToHost,
+                                        stream[current_stream]));
 
-    checkCudaErrors(
+        checkCudaErrors(cudaEventRecord(cycleDone[current_stream], stream[current_stream]));
-        cudaEventRecord(cycleDone[current_stream], stream[current_stream]));
 
         current_stream = next_stream;
     }
@@ -350,7 +339,8 @@ float processWithStreams(int streams_used) {
     return time;
 }
 
-void init() {
+void init()
+{
     for (int i = 0; i < N; ++i) {
         h_data_source[i] = 0;
     }
@@ -360,7 +350,8 @@ void init() {
     }
 }
 
-bool test() {
+bool test()
+{
     bool passed = true;
 
     for (int j = 0; j < STREAM_COUNT; ++j) {

Samples/0_Introduction/simpleMultiGPU/simpleMultiGPU.cu

@@ -37,15 +37,15 @@
  */
 
 // System includes
-#include <stdio.h>
 #include <assert.h>
+#include <stdio.h>
 
 // CUDA runtime
 #include <cuda_runtime.h>
 
 // helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 #ifndef MAX
 #define MAX(a, b) (a > b ? a : b)
@@ -64,12 +64,14 @@ const int DATA_N = 1048576 * 32;
 // Refer to the 'reduction' CUDA Sample describing
 // reduction optimization strategies
 ////////////////////////////////////////////////////////////////////////////////
-__global__ static void reduceKernel(float *d_Result, float *d_Input, int N) {
+__global__ static void reduceKernel(float *d_Result, float *d_Input, int N)
+{
     const int tid     = blockIdx.x * blockDim.x + threadIdx.x;
     const int threadN = gridDim.x * blockDim.x;
     float     sum     = 0;
 
-  for (int pos = tid; pos < N; pos += threadN) sum += d_Input[pos];
+    for (int pos = tid; pos < N; pos += threadN)
+        sum += d_Input[pos];
 
     d_Result[tid] = sum;
 }
@@ -77,7 +79,8 @@ __global__ static void reduceKernel(float *d_Result, float *d_Input, int N) {
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     // Solver config
     TGPUplan plan[MAX_GPU_COUNT];
 
@@ -129,14 +132,10 @@ int main(int argc, char **argv) {
         checkCudaErrors(cudaSetDevice(i));
         checkCudaErrors(cudaStreamCreate(&plan[i].stream));
         // Allocate memory
-    checkCudaErrors(
+        checkCudaErrors(cudaMalloc((void **)&plan[i].d_Data, plan[i].dataN * sizeof(float)));
-        cudaMalloc((void **)&plan[i].d_Data, plan[i].dataN * sizeof(float)));
+        checkCudaErrors(cudaMalloc((void **)&plan[i].d_Sum, ACCUM_N * sizeof(float)));
-    checkCudaErrors(
+        checkCudaErrors(cudaMallocHost((void **)&plan[i].h_Sum_from_device, ACCUM_N * sizeof(float)));
-        cudaMalloc((void **)&plan[i].d_Sum, ACCUM_N * sizeof(float)));
+        checkCudaErrors(cudaMallocHost((void **)&plan[i].h_Data, plan[i].dataN * sizeof(float)));
-    checkCudaErrors(cudaMallocHost((void **)&plan[i].h_Sum_from_device,
-                                   ACCUM_N * sizeof(float)));
-    checkCudaErrors(cudaMallocHost((void **)&plan[i].h_Data,
-                                   plan[i].dataN * sizeof(float)));
 
         for (j = 0; j < plan[i].dataN; j++) {
             plan[i].h_Data[j] = (float)rand() / (float)RAND_MAX;
@@ -158,19 +157,16 @@ int main(int argc, char **argv) {
         checkCudaErrors(cudaSetDevice(i));
 
         // Copy input data from CPU
-    checkCudaErrors(cudaMemcpyAsync(plan[i].d_Data, plan[i].h_Data,
+        checkCudaErrors(cudaMemcpyAsync(
-                                    plan[i].dataN * sizeof(float),
+            plan[i].d_Data, plan[i].h_Data, plan[i].dataN * sizeof(float), cudaMemcpyHostToDevice, plan[i].stream));
-                                    cudaMemcpyHostToDevice, plan[i].stream));
 
         // Perform GPU computations
-    reduceKernel<<<BLOCK_N, THREAD_N, 0, plan[i].stream>>>(
+        reduceKernel<<<BLOCK_N, THREAD_N, 0, plan[i].stream>>>(plan[i].d_Sum, plan[i].d_Data, plan[i].dataN);
-        plan[i].d_Sum, plan[i].d_Data, plan[i].dataN);
         getLastCudaError("reduceKernel() execution failed.\n");
 
         // Read back GPU results
-    checkCudaErrors(cudaMemcpyAsync(plan[i].h_Sum_from_device, plan[i].d_Sum,
+        checkCudaErrors(cudaMemcpyAsync(
-                                    ACCUM_N * sizeof(float),
+            plan[i].h_Sum_from_device, plan[i].d_Sum, ACCUM_N * sizeof(float), cudaMemcpyDeviceToHost, plan[i].stream));
-                                    cudaMemcpyDeviceToHost, plan[i].stream));
     }
 
     // Process GPU results

Samples/0_Introduction/simpleMultiGPU/simpleMultiGPU.h

@@ -37,7 +37,8 @@
 #ifndef SIMPLEMULTIGPU_H
 #define SIMPLEMULTIGPU_H
 
-typedef struct {
+typedef struct
+{
     // Host-side input data
     int    dataN;
     float *h_Data;
@@ -56,7 +57,6 @@ typedef struct {
 
 } TGPUplan;
 
-extern "C" void launch_reduceKernel(float *d_Result, float *d_Input, int N,
+extern "C" void launch_reduceKernel(float *d_Result, float *d_Input, int N, int BLOCK_N, int THREAD_N, cudaStream_t &s);
-                                    int BLOCK_N, int THREAD_N, cudaStream_t &s);
 
 #endif

Samples/0_Introduction/simpleOccupancy/simpleOccupancy.cu

@@ -25,8 +25,8 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <iostream>
 #include <helper_cuda.h> // helper functions for CUDA error check
+#include <iostream>
 
 const int manualBlockSize = 32;
 
@@ -38,7 +38,8 @@ const int manualBlockSize = 32;
 // execution configuration, including anything the launch configurator
 // API suggests.
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void square(int *array, int arrayCount) {
+__global__ void square(int *array, int arrayCount)
+{
     extern __shared__ int dynamicSmem[];
     int                   idx = threadIdx.x + blockIdx.x * blockDim.x;
 
@@ -58,8 +59,8 @@ __global__ void square(int *array, int arrayCount) {
 // This wrapper routine computes the occupancy of kernel, and reports
 // it in terms of active warps / maximum warps per SM.
 ////////////////////////////////////////////////////////////////////////////////
-static double reportPotentialOccupancy(void *kernel, int blockSize,
+static double reportPotentialOccupancy(void *kernel, int blockSize, size_t dynamicSMem)
-                                       size_t dynamicSMem) {
+{
     int            device;
     cudaDeviceProp prop;
 
@@ -72,8 +73,7 @@ static double reportPotentialOccupancy(void *kernel, int blockSize,
     checkCudaErrors(cudaGetDevice(&device));
     checkCudaErrors(cudaGetDeviceProperties(&prop, device));
 
-  checkCudaErrors(cudaOccupancyMaxActiveBlocksPerMultiprocessor(
+    checkCudaErrors(cudaOccupancyMaxActiveBlocksPerMultiprocessor(&numBlocks, kernel, blockSize, dynamicSMem));
-      &numBlocks, kernel, blockSize, dynamicSMem));
 
     activeWarps = numBlocks * blockSize / prop.warpSize;
     maxWarps    = prop.maxThreadsPerMultiProcessor / prop.warpSize;
@@ -99,7 +99,8 @@ static double reportPotentialOccupancy(void *kernel, int blockSize,
 // This function configures the launch based on the "automatic"
 // argument, records the runtime, and reports occupancy and runtime.
 ////////////////////////////////////////////////////////////////////////////////
-static int launchConfig(int *array, int arrayCount, bool automatic) {
+static int launchConfig(int *array, int arrayCount, bool automatic)
+{
     int    blockSize;
     int    minGridSize;
     int    gridSize;
@@ -116,14 +117,13 @@ static int launchConfig(int *array, int arrayCount, bool automatic) {
     checkCudaErrors(cudaEventCreate(&end));
 
     if (automatic) {
-    checkCudaErrors(cudaOccupancyMaxPotentialBlockSize(
+        checkCudaErrors(
-        &minGridSize, &blockSize, (void *)square, dynamicSMemUsage,
+            cudaOccupancyMaxPotentialBlockSize(&minGridSize, &blockSize, (void *)square, dynamicSMemUsage, arrayCount));
-        arrayCount));
 
         std::cout << "Suggested block size: " << blockSize << std::endl
-              << "Minimum grid size for maximum occupancy: " << minGridSize
+                  << "Minimum grid size for maximum occupancy: " << minGridSize << std::endl;
-              << std::endl;
+    }
-  } else {
+    else {
         // This block size is too small. Given limited number of
         // active blocks per multiprocessor, the number of active
         // threads will be limited, and thus unable to achieve maximum
@@ -146,11 +146,9 @@ static int launchConfig(int *array, int arrayCount, bool automatic) {
 
     // Calculate occupancy
     //
-  potentialOccupancy =
+    potentialOccupancy = reportPotentialOccupancy((void *)square, blockSize, dynamicSMemUsage);
-      reportPotentialOccupancy((void *)square, blockSize, dynamicSMemUsage);
 
-  std::cout << "Potential occupancy: " << potentialOccupancy * 100 << "%"
+    std::cout << "Potential occupancy: " << potentialOccupancy * 100 << "%" << std::endl;
-            << std::endl;
 
     // Report elapsed time
     //
@@ -166,7 +164,8 @@ static int launchConfig(int *array, int arrayCount, bool automatic) {
 // The test generates an array and squares it with a CUDA kernel, then
 // verifies the result.
 ////////////////////////////////////////////////////////////////////////////////
-static int test(bool automaticLaunchConfig, const int count = 1000000) {
+static int test(bool automaticLaunchConfig, const int count = 1000000)
+{
     int *array;
     int *dArray;
     int  size = count * sizeof(int);
@@ -193,8 +192,7 @@ static int test(bool automaticLaunchConfig, const int count = 1000000) {
     //
     for (int i = 0; i < count; i += 1) {
         if (array[i] != i * i) {
-      std::cout << "element " << i << " expected " << i * i << " actual "
+            std::cout << "element " << i << " expected " << i * i << " actual " << array[i] << std::endl;
-                << array[i] << std::endl;
             return 1;
         }
     }
@@ -210,13 +208,13 @@ static int test(bool automaticLaunchConfig, const int count = 1000000) {
 // automatically configured launch, and reports the occupancy and
 // performance.
 ////////////////////////////////////////////////////////////////////////////////
-int main() {
+int main()
+{
     int status;
 
     std::cout << "starting Simple Occupancy" << std::endl << std::endl;
 
-  std::cout << "[ Manual configuration with " << manualBlockSize
+    std::cout << "[ Manual configuration with " << manualBlockSize << " threads per block ]" << std::endl;
-            << " threads per block ]" << std::endl;
 
     status = test(false);
     if (status) {

Samples/0_Introduction/simpleP2P/README.md

@@ -34,4 +34,3 @@ Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-d
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)
-

Samples/0_Introduction/simpleP2P/simpleP2P.cu

@@ -31,8 +31,8 @@
  */
 
 // includes, system
-#include <stdlib.h>
 #include <stdio.h>
+#include <stdlib.h>
 
 // CUDA includes
 #include <cuda_runtime.h>
@@ -41,7 +41,8 @@
 #include <helper_cuda.h>
 #include <helper_functions.h> // helper for shared that are common to CUDA Samples
 
-__global__ void SimpleKernel(float *src, float *dst) {
+__global__ void SimpleKernel(float *src, float *dst)
+{
     // Just a dummy kernel, doing enough for us to verify that everything
     // worked
     const int idx = blockIdx.x * blockDim.x + threadIdx.x;
@@ -50,12 +51,12 @@ __global__ void SimpleKernel(float *src, float *dst) {
 
 inline bool IsAppBuiltAs64() { return sizeof(void *) == 8; }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("[%s] - Starting...\n", argv[0]);
 
     if (!IsAppBuiltAs64()) {
-    printf(
+        printf("%s is only supported with on 64-bit OSs and the application must be "
-        "%s is only supported with on 64-bit OSs and the application must be "
                "built as a 64-bit target.  Test is being waived.\n",
                argv[0]);
         exit(EXIT_WAIVED);
@@ -68,8 +69,7 @@ int main(int argc, char **argv) {
     printf("CUDA-capable device count: %i\n", gpu_n);
 
     if (gpu_n < 2) {
-    printf(
+        printf("Two or more GPUs with Peer-to-Peer access capability are required for "
-        "Two or more GPUs with Peer-to-Peer access capability are required for "
                "%s.\n",
                argv[0]);
         printf("Waiving test.\n");
@@ -97,8 +97,12 @@ int main(int argc, char **argv) {
                 continue;
             }
             checkCudaErrors(cudaDeviceCanAccessPeer(&can_access_peer, i, j));
-      printf("> Peer access from %s (GPU%d) -> %s (GPU%d) : %s\n", prop[i].name,
+            printf("> Peer access from %s (GPU%d) -> %s (GPU%d) : %s\n",
-             i, prop[j].name, j, can_access_peer ? "Yes" : "No");
+                   prop[i].name,
+                   i,
+                   prop[j].name,
+                   j,
+                   can_access_peer ? "Yes" : "No");
             if (can_access_peer && p2pCapableGPUs[0] == -1) {
                 p2pCapableGPUs[0] = i;
                 p2pCapableGPUs[1] = j;
@@ -107,12 +111,10 @@ int main(int argc, char **argv) {
     }
 
     if (p2pCapableGPUs[0] == -1 || p2pCapableGPUs[1] == -1) {
-    printf(
+        printf("Two or more GPUs with Peer-to-Peer access capability are required for "
-        "Two or more GPUs with Peer-to-Peer access capability are required for "
                "%s.\n",
                argv[0]);
-    printf(
+        printf("Peer to Peer access is not available amongst GPUs in the system, "
-        "Peer to Peer access is not available amongst GPUs in the system, "
                "waiving test.\n");
 
         exit(EXIT_WAIVED);
@@ -123,8 +125,7 @@ int main(int argc, char **argv) {
     gpuid[1] = p2pCapableGPUs[1];
 
     // Enable peer access
-  printf("Enabling peer access between GPU%d and GPU%d...\n", gpuid[0],
+    printf("Enabling peer access between GPU%d and GPU%d...\n", gpuid[0], gpuid[1]);
-         gpuid[1]);
     checkCudaErrors(cudaSetDevice(gpuid[0]));
     checkCudaErrors(cudaDeviceEnablePeerAccess(gpuid[1], 0));
     checkCudaErrors(cudaSetDevice(gpuid[1]));
@@ -132,8 +133,8 @@ int main(int argc, char **argv) {
 
     // Allocate buffers
     const size_t buf_size = 1024 * 1024 * 16 * sizeof(float);
-  printf("Allocating buffers (%iMB on GPU%d, GPU%d and CPU Host)...\n",
+    printf(
-         int(buf_size / 1024 / 1024), gpuid[0], gpuid[1]);
+        "Allocating buffers (%iMB on GPU%d, GPU%d and CPU Host)...\n", int(buf_size / 1024 / 1024), gpuid[0], gpuid[1]);
     checkCudaErrors(cudaSetDevice(gpuid[0]));
     float *g0;
     checkCudaErrors(cudaMalloc(&g0, buf_size));
@@ -141,8 +142,7 @@ int main(int argc, char **argv) {
     float *g1;
     checkCudaErrors(cudaMalloc(&g1, buf_size));
     float *h0;
-  checkCudaErrors(
+    checkCudaErrors(cudaMallocHost(&h0, buf_size)); // Automatically portable with UVA
-      cudaMallocHost(&h0, buf_size));  // Automatically portable with UVA
 
     // Create CUDA event handles
     printf("Creating event handles...\n");
@@ -161,7 +161,8 @@ int main(int argc, char **argv) {
         // Ping-pong copy between GPUs
         if (i % 2 == 0) {
             checkCudaErrors(cudaMemcpy(g1, g0, buf_size, cudaMemcpyDefault));
-    } else {
+        }
+        else {
             checkCudaErrors(cudaMemcpy(g0, g1, buf_size, cudaMemcpyDefault));
         }
     }
@@ -170,9 +171,9 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaEventSynchronize(stop_event));
     checkCudaErrors(cudaEventElapsedTime(&time_memcpy, start_event, stop_event));
     printf("cudaMemcpyPeer / cudaMemcpy between GPU%d and GPU%d: %.2fGB/s\n",
-         gpuid[0], gpuid[1],
+           gpuid[0],
-         (1.0f / (time_memcpy / 1000.0f)) * ((100.0f * buf_size)) / 1024.0f /
+           gpuid[1],
-             1024.0f / 1024.0f);
+           (1.0f / (time_memcpy / 1000.0f)) * ((100.0f * buf_size)) / 1024.0f / 1024.0f / 1024.0f);
 
     // Prepare host buffer and copy to GPU 0
     printf("Preparing host buffer and memcpy to GPU%d...\n", gpuid[0]);
@@ -190,10 +191,11 @@ int main(int argc, char **argv) {
 
     // Run kernel on GPU 1, reading input from the GPU 0 buffer, writing
     // output to the GPU 1 buffer
-  printf(
+    printf("Run kernel on GPU%d, taking source data from GPU%d and writing to "
-      "Run kernel on GPU%d, taking source data from GPU%d and writing to "
            "GPU%d...\n",
-      gpuid[1], gpuid[0], gpuid[1]);
+           gpuid[1],
+           gpuid[0],
+           gpuid[1]);
     checkCudaErrors(cudaSetDevice(gpuid[1]));
     SimpleKernel<<<blocks, threads>>>(g0, g1);
 
@@ -201,10 +203,11 @@ int main(int argc, char **argv) {
 
     // Run kernel on GPU 0, reading input from the GPU 1 buffer, writing
     // output to the GPU 0 buffer
-  printf(
+    printf("Run kernel on GPU%d, taking source data from GPU%d and writing to "
-      "Run kernel on GPU%d, taking source data from GPU%d and writing to "
            "GPU%d...\n",
-      gpuid[0], gpuid[1], gpuid[0]);
+           gpuid[0],
+           gpuid[1],
+           gpuid[0]);
     checkCudaErrors(cudaSetDevice(gpuid[0]));
     SimpleKernel<<<blocks, threads>>>(g1, g0);
 
@@ -220,8 +223,7 @@ int main(int argc, char **argv) {
         // Re-generate input data and apply 2x '* 2.0f' computation of both
         // kernel runs
         if (h0[i] != float(i % 4096) * 2.0f * 2.0f) {
-      printf("Verification error @ element %i: val = %f, ref = %f\n", i, h0[i],
+            printf("Verification error @ element %i: val = %f, ref = %f\n", i, h0[i], (float(i % 4096) * 2.0f * 2.0f));
-             (float(i % 4096) * 2.0f * 2.0f));
 
             if (error_count++ > 10) {
                 break;
@@ -253,7 +255,8 @@ int main(int argc, char **argv) {
     if (error_count != 0) {
         printf("Test failed!\n");
         exit(EXIT_FAILURE);
-  } else {
+    }
+    else {
         printf("Test passed\n");
         exit(EXIT_SUCCESS);
     }

Samples/0_Introduction/simplePitchLinearTexture/simplePitchLinearTexture.cu

@@ -70,29 +70,26 @@ bool bTestResult = true;
 //! Shifts matrix elements using pitch linear array
 //! @param odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void shiftPitchLinear(float *odata, int pitch, int width, int height,
+__global__ void
-                                 int shiftX, int shiftY,
+shiftPitchLinear(float *odata, int pitch, int width, int height, int shiftX, int shiftY, cudaTextureObject_t texRefPL)
-                                 cudaTextureObject_t texRefPL) {
+{
     int xid = blockIdx.x * blockDim.x + threadIdx.x;
     int yid = blockIdx.y * blockDim.y + threadIdx.y;
 
-  odata[yid * pitch + xid] = tex2D<float>(
+    odata[yid * pitch + xid] = tex2D<float>(texRefPL, (xid + shiftX) / (float)width, (yid + shiftY) / (float)height);
-      texRefPL, (xid + shiftX) / (float)width, (yid + shiftY) / (float)height);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Shifts matrix elements using regular array
 //! @param odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void shiftArray(float *odata, int pitch, int width, int height,
+__global__ void
-                           int shiftX, int shiftY,
+shiftArray(float *odata, int pitch, int width, int height, int shiftX, int shiftY, cudaTextureObject_t texRefArray)
-                           cudaTextureObject_t texRefArray) {
+{
     int xid = blockIdx.x * blockDim.x + threadIdx.x;
     int yid = blockIdx.y * blockDim.y + threadIdx.y;
 
-  odata[yid * pitch + xid] =
+    odata[yid * pitch + xid] = tex2D<float>(texRefArray, (xid + shiftX) / (float)width, (yid + shiftY) / (float)height);
-      tex2D<float>(texRefArray, (xid + shiftX) / (float)width,
-                   (yid + shiftY) / (float)height);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
@@ -102,20 +99,21 @@ void runTest(int argc, char **argv);
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n\n", sSDKsample);
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sSDKsample,
+    printf("%s completed, returned %s\n", sSDKsample, bTestResult ? "OK" : "ERROR!");
-         bTestResult ? "OK" : "ERROR!");
     exit(bTestResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     // Set array size
     const int nx = 2048;
     const int ny = 2048;
@@ -154,8 +152,7 @@ void runTest(int argc, char **argv) {
     float *d_idataPL;
     size_t d_pitchBytes;
 
-  checkCudaErrors(cudaMallocPitch((void **)&d_idataPL, &d_pitchBytes,
+    checkCudaErrors(cudaMallocPitch((void **)&d_idataPL, &d_pitchBytes, nx * sizeof(float), ny));
-                                  nx * sizeof(float), ny));
 
     // Array input data
     cudaArray            *d_idataArray;
@@ -165,20 +162,17 @@ void runTest(int argc, char **argv) {
 
     // Pitch linear output data
     float *d_odata;
-  checkCudaErrors(cudaMallocPitch((void **)&d_odata, &d_pitchBytes,
+    checkCudaErrors(cudaMallocPitch((void **)&d_odata, &d_pitchBytes, nx * sizeof(float), ny));
-                                  nx * sizeof(float), ny));
 
     // Copy host data to device
     // Pitch linear
     size_t h_pitchBytes = nx * sizeof(float);
 
-  checkCudaErrors(cudaMemcpy2D(d_idataPL, d_pitchBytes, h_idata, h_pitchBytes,
+    checkCudaErrors(
-                               nx * sizeof(float), ny, cudaMemcpyHostToDevice));
+        cudaMemcpy2D(d_idataPL, d_pitchBytes, h_idata, h_pitchBytes, nx * sizeof(float), ny, cudaMemcpyHostToDevice));
 
     // Array
-  checkCudaErrors(cudaMemcpyToArray(d_idataArray, 0, 0, h_idata,
+    checkCudaErrors(cudaMemcpyToArray(d_idataArray, 0, 0, h_idata, nx * ny * sizeof(float), cudaMemcpyHostToDevice));
-                                    nx * ny * sizeof(float),
-                                    cudaMemcpyHostToDevice));
 
     cudaTextureObject_t texRefPL;
     cudaTextureObject_t texRefArray;
@@ -210,8 +204,7 @@ void runTest(int argc, char **argv) {
     texDescr.addressMode[0]   = cudaAddressModeWrap;
     texDescr.addressMode[1]   = cudaAddressModeWrap;
     texDescr.readMode         = cudaReadModeElementType;
-  checkCudaErrors(
+    checkCudaErrors(cudaCreateTextureObject(&texRefArray, &texRes, &texDescr, NULL));
-      cudaCreateTextureObject(&texRefArray, &texRes, &texDescr, NULL));
 
     // Reference calculation
     for (int j = 0; j < ny; ++j) {
@@ -224,15 +217,13 @@ void runTest(int argc, char **argv) {
     }
 
     // Run ShiftPitchLinear kernel
-  checkCudaErrors(
+    checkCudaErrors(cudaMemset2D(d_odata, d_pitchBytes, 0, nx * sizeof(float), ny));
-      cudaMemset2D(d_odata, d_pitchBytes, 0, nx * sizeof(float), ny));
 
     checkCudaErrors(cudaEventRecord(start, 0));
 
     for (int i = 0; i < NUM_REPS; ++i) {
-    shiftPitchLinear<<<dimGrid, dimBlock>>>(d_odata,
+        shiftPitchLinear<<<dimGrid, dimBlock>>>(
-                                            (int)(d_pitchBytes / sizeof(float)),
+            d_odata, (int)(d_pitchBytes / sizeof(float)), nx, ny, x_shift, y_shift, texRefPL);
-                                            nx, ny, x_shift, y_shift, texRefPL);
     }
 
     checkCudaErrors(cudaEventRecord(stop, 0));
@@ -241,8 +232,8 @@ void runTest(int argc, char **argv) {
     checkCudaErrors(cudaEventElapsedTime(&timePL, start, stop));
 
     // Check results
-  checkCudaErrors(cudaMemcpy2D(h_odata, h_pitchBytes, d_odata, d_pitchBytes,
+    checkCudaErrors(
-                               nx * sizeof(float), ny, cudaMemcpyDeviceToHost));
+        cudaMemcpy2D(h_odata, h_pitchBytes, d_odata, d_pitchBytes, nx * sizeof(float), ny, cudaMemcpyDeviceToHost));
 
     bool res = compareData(gold, h_odata, nx * ny, 0.0f, 0.15f);
 
@@ -254,14 +245,12 @@ void runTest(int argc, char **argv) {
     }
 
     // Run ShiftArray kernel
-  checkCudaErrors(
+    checkCudaErrors(cudaMemset2D(d_odata, d_pitchBytes, 0, nx * sizeof(float), ny));
-      cudaMemset2D(d_odata, d_pitchBytes, 0, nx * sizeof(float), ny));
     checkCudaErrors(cudaEventRecord(start, 0));
 
     for (int i = 0; i < NUM_REPS; ++i) {
-    shiftArray<<<dimGrid, dimBlock>>>(d_odata,
+        shiftArray<<<dimGrid, dimBlock>>>(
-                                      (int)(d_pitchBytes / sizeof(float)), nx,
+            d_odata, (int)(d_pitchBytes / sizeof(float)), nx, ny, x_shift, y_shift, texRefArray);
-                                      ny, x_shift, y_shift, texRefArray);
     }
 
     checkCudaErrors(cudaEventRecord(stop, 0));
@@ -270,8 +259,8 @@ void runTest(int argc, char **argv) {
     checkCudaErrors(cudaEventElapsedTime(&timeArray, start, stop));
 
     // Check results
-  checkCudaErrors(cudaMemcpy2D(h_odata, h_pitchBytes, d_odata, d_pitchBytes,
+    checkCudaErrors(
-                               nx * sizeof(float), ny, cudaMemcpyDeviceToHost));
+        cudaMemcpy2D(h_odata, h_pitchBytes, d_odata, d_pitchBytes, nx * sizeof(float), ny, cudaMemcpyDeviceToHost));
     res = compareData(gold, h_odata, nx * ny, 0.0f, 0.15f);
 
     if (res == false) {
@@ -279,21 +268,18 @@ void runTest(int argc, char **argv) {
         bTestResult = false;
     }
 
-  float bandwidthPL =
+    float bandwidthPL    = 2.f * 1000.f * nx * ny * sizeof(float) / (1.e+9f) / (timePL / NUM_REPS);
-      2.f * 1000.f * nx * ny * sizeof(float) / (1.e+9f) / (timePL / NUM_REPS);
+    float bandwidthArray = 2.f * 1000.f * nx * ny * sizeof(float) / (1.e+9f) / (timeArray / NUM_REPS);
-  float bandwidthArray = 2.f * 1000.f * nx * ny * sizeof(float) / (1.e+9f) /
-                         (timeArray / NUM_REPS);
 
-  printf("\nBandwidth (GB/s) for pitch linear: %.2e; for array: %.2e\n",
+    printf("\nBandwidth (GB/s) for pitch linear: %.2e; for array: %.2e\n", bandwidthPL, bandwidthArray);
-         bandwidthPL, bandwidthArray);
 
     float fetchRatePL    = nx * ny / 1.e+6f / (timePL / (1000.0f * NUM_REPS));
     float fetchRateArray = nx * ny / 1.e+6f / (timeArray / (1000.0f * NUM_REPS));
 
-  printf(
+    printf("\nTexture fetch rate (Mpix/s) for pitch linear: "
-      "\nTexture fetch rate (Mpix/s) for pitch linear: "
            "%.2e; for array: %.2e\n\n",
-      fetchRatePL, fetchRateArray);
+           fetchRatePL,
+           fetchRateArray);
 
     // Cleanup
     free(h_idata);

Samples/0_Introduction/simplePrintf/simplePrintf.cu

@@ -26,28 +26,30 @@
  */
 
 // System includes
-#include <stdio.h>
 #include <assert.h>
+#include <stdio.h>
 
 // CUDA runtime
 #include <cuda_runtime.h>
 
 // helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 #ifndef MAX
 #define MAX(a, b) (a > b ? a : b)
 #endif
 
-__global__ void testKernel(int val) {
+__global__ void testKernel(int val)
-  printf("[%d, %d]:\t\tValue is:%d\n", blockIdx.y * gridDim.x + blockIdx.x,
+{
-         threadIdx.z * blockDim.x * blockDim.y + threadIdx.y * blockDim.x +
+    printf("[%d, %d]:\t\tValue is:%d\n",
-             threadIdx.x,
+           blockIdx.y * gridDim.x + blockIdx.x,
+           threadIdx.z * blockDim.x * blockDim.y + threadIdx.y * blockDim.x + threadIdx.x,
            val);
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     int            devID;
     cudaDeviceProp props;
 
@@ -57,8 +59,7 @@ int main(int argc, char **argv) {
     // Get GPU information
     checkCudaErrors(cudaGetDevice(&devID));
     checkCudaErrors(cudaGetDeviceProperties(&props, devID));
-  printf("Device %d: \"%s\" with Compute capability %d.%d\n", devID, props.name,
+    printf("Device %d: \"%s\" with Compute capability %d.%d\n", devID, props.name, props.major, props.minor);
-         props.major, props.minor);
 
     printf("printf() is called. Output:\n\n");
 

Samples/0_Introduction/simpleStreams/simpleStreams.cu

@@ -48,24 +48,25 @@
 
 const char *sSDKsample = "simpleStreams";
 
-const char *sEventSyncMethod[] = {"cudaEventDefault", "cudaEventBlockingSync",
+const char *sEventSyncMethod[] = {"cudaEventDefault", "cudaEventBlockingSync", "cudaEventDisableTiming", NULL};
-                                  "cudaEventDisableTiming", NULL};
 
-const char *sDeviceSyncMethod[] = {
+const char *sDeviceSyncMethod[] = {"cudaDeviceScheduleAuto",
-    "cudaDeviceScheduleAuto",         "cudaDeviceScheduleSpin",
+                                   "cudaDeviceScheduleSpin",
-    "cudaDeviceScheduleYield",        "INVALID",
+                                   "cudaDeviceScheduleYield",
-    "cudaDeviceScheduleBlockingSync", NULL};
+                                   "INVALID",
+                                   "cudaDeviceScheduleBlockingSync",
+                                   NULL};
 
 // System includes
-#include <stdio.h>
 #include <assert.h>
+#include <stdio.h>
 
 // CUDA runtime
 #include <cuda_runtime.h>
 
 // helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 #ifndef WIN32
 #include <sys/mman.h> // for mmap() / munmap()
@@ -75,7 +76,8 @@ const char *sDeviceSyncMethod[] = {
 #define MEMORY_ALIGNMENT  4096
 #define ALIGN_UP(x, size) (((size_t)x + (size - 1)) & (~(size - 1)))
 
-__global__ void init_array(int *g_data, int *factor, int num_iterations) {
+__global__ void init_array(int *g_data, int *factor, int num_iterations)
+{
     int idx = blockIdx.x * blockDim.x + threadIdx.x;
 
     for (int i = 0; i < num_iterations; i++) {
@@ -83,7 +85,8 @@ __global__ void init_array(int *g_data, int *factor, int num_iterations) {
     }
 }
 
-bool correct_data(int *a, const int n, const int c) {
+bool correct_data(int *a, const int n, const int c)
+{
     for (int i = 0; i < n; i++) {
         if (a[i] != c) {
             printf("%d: %d %d\n", i, a[i], c);
@@ -94,51 +97,45 @@ bool correct_data(int *a, const int n, const int c) {
     return true;
 }
 
-inline void AllocateHostMemory(bool bPinGenericMemory, int **pp_a,
+inline void AllocateHostMemory(bool bPinGenericMemory, int **pp_a, int **ppAligned_a, int nbytes)
-                               int **ppAligned_a, int nbytes) {
+{
 #if CUDART_VERSION >= 4000
 #if !defined(__arm__) && !defined(__aarch64__)
     if (bPinGenericMemory) {
 // allocate a generic page-aligned chunk of system memory
 #ifdef WIN32
-    printf(
+        printf("> VirtualAlloc() allocating %4.2f Mbytes of (generic page-aligned "
-        "> VirtualAlloc() allocating %4.2f Mbytes of (generic page-aligned "
                "system memory)\n",
                (float)nbytes / 1048576.0f);
-    *pp_a = (int *)VirtualAlloc(NULL, (nbytes + MEMORY_ALIGNMENT),
+        *pp_a = (int *)VirtualAlloc(NULL, (nbytes + MEMORY_ALIGNMENT), MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE);
-                                MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE);
 #else
-    printf(
+        printf("> mmap() allocating %4.2f Mbytes (generic page-aligned system "
-        "> mmap() allocating %4.2f Mbytes (generic page-aligned system "
                "memory)\n",
                (float)nbytes / 1048576.0f);
-    *pp_a = (int *)mmap(NULL, (nbytes + MEMORY_ALIGNMENT),
+        *pp_a = (int *)mmap(NULL, (nbytes + MEMORY_ALIGNMENT), PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, -1, 0);
-                        PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, -1, 0);
 #endif
 
         *ppAligned_a = (int *)ALIGN_UP(*pp_a, MEMORY_ALIGNMENT);
 
-    printf(
+        printf("> cudaHostRegister() registering %4.2f Mbytes of generic allocated "
-        "> cudaHostRegister() registering %4.2f Mbytes of generic allocated "
                "system memory\n",
                (float)nbytes / 1048576.0f);
         // pin allocate memory
-    checkCudaErrors(
+        checkCudaErrors(cudaHostRegister(*ppAligned_a, nbytes, cudaHostRegisterMapped));
-        cudaHostRegister(*ppAligned_a, nbytes, cudaHostRegisterMapped));
+    }
-  } else
+    else
 #endif
 #endif
     {
-    printf("> cudaMallocHost() allocating %4.2f Mbytes of system memory\n",
+        printf("> cudaMallocHost() allocating %4.2f Mbytes of system memory\n", (float)nbytes / 1048576.0f);
-           (float)nbytes / 1048576.0f);
         // allocate host memory (pinned is required for achieve asynchronicity)
         checkCudaErrors(cudaMallocHost((void **)pp_a, nbytes));
         *ppAligned_a = *pp_a;
     }
 }
 
-inline void FreeHostMemory(bool bPinGenericMemory, int **pp_a,
+inline void FreeHostMemory(bool bPinGenericMemory, int **pp_a, int **ppAligned_a, int nbytes)
-                           int **ppAligned_a, int nbytes) {
+{
 #if CUDART_VERSION >= 4000
 #if !defined(__arm__) && !defined(__aarch64__)
     // CUDA 4.0 support pinning of generic host memory
@@ -150,7 +147,8 @@ inline void FreeHostMemory(bool bPinGenericMemory, int **pp_a,
 #else
         munmap(*pp_a, nbytes);
 #endif
-  } else
+    }
+    else
 #endif
 #endif
     {
@@ -158,26 +156,24 @@ inline void FreeHostMemory(bool bPinGenericMemory, int **pp_a,
     }
 }
 
-static const char *sSyncMethod[] = {
+static const char *sSyncMethod[] = {"0 (Automatic Blocking)",
-    "0 (Automatic Blocking)",
                                     "1 (Spin Blocking)",
                                     "2 (Yield Blocking)",
                                     "3 (Undefined Blocking Method)",
                                     "4 (Blocking Sync Event) = low CPU utilization",
                                     NULL};
 
-void printHelp() {
+void printHelp()
+{
     printf("Usage: %s [options below]\n", sSDKsample);
     printf("\t--sync_method=n for CPU/GPU synchronization\n");
     printf("\t             n=%s\n", sSyncMethod[0]);
     printf("\t             n=%s\n", sSyncMethod[1]);
     printf("\t             n=%s\n", sSyncMethod[2]);
     printf("\t   <Default> n=%s\n", sSyncMethod[4]);
-  printf(
+    printf("\t--use_generic_memory (default) use generic page-aligned for system "
-      "\t--use_generic_memory (default) use generic page-aligned for system "
            "memory\n");
-  printf(
+    printf("\t--use_cuda_malloc_host (optional) use cudaMallocHost to allocate "
-      "\t--use_cuda_malloc_host (optional) use cudaMallocHost to allocate "
            "system memory\n");
 }
 
@@ -187,7 +183,8 @@ void printHelp() {
 #define DEFAULT_PINNED_GENERIC_MEMORY true
 #endif
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     int   cuda_device = 0;
     int   nstreams    = 4;                        // number of streams for CUDA calls
     int   nreps       = 10;                       // number of times each experiment is repeated
@@ -199,10 +196,8 @@ int main(int argc, char **argv) {
 
     // allocate generic memory and pin it laster instead of using cudaHostAlloc()
 
-  bool bPinGenericMemory =
+    bool bPinGenericMemory  = DEFAULT_PINNED_GENERIC_MEMORY; // we want this to be the default behavior
-      DEFAULT_PINNED_GENERIC_MEMORY;  // we want this to be the default behavior
+    int  device_sync_method = cudaDeviceBlockingSync;        // by default we use BlockingSync
-  int device_sync_method =
-      cudaDeviceBlockingSync;  // by default we use BlockingSync
 
     int niterations; // number of iterations for the loop inside the kernel
 
@@ -213,20 +208,18 @@ int main(int argc, char **argv) {
         return EXIT_SUCCESS;
     }
 
-  if ((device_sync_method = getCmdLineArgumentInt(argc, (const char **)argv,
+    if ((device_sync_method = getCmdLineArgumentInt(argc, (const char **)argv, "sync_method")) >= 0) {
-                                                  "sync_method")) >= 0) {
+        if (device_sync_method == 0 || device_sync_method == 1 || device_sync_method == 2 || device_sync_method == 4) {
-    if (device_sync_method == 0 || device_sync_method == 1 ||
+            printf("Device synchronization method set to = %s\n", sSyncMethod[device_sync_method]);
-        device_sync_method == 2 || device_sync_method == 4) {
-      printf("Device synchronization method set to = %s\n",
-             sSyncMethod[device_sync_method]);
             printf("Setting reps to 100 to demonstrate steady state\n");
             nreps = 100;
-    } else {
+        }
-      printf("Invalid command line option sync_method=\"%d\"\n",
+        else {
-             device_sync_method);
+            printf("Invalid command line option sync_method=\"%d\"\n", device_sync_method);
             return EXIT_FAILURE;
         }
-  } else {
+    }
+    else {
         printHelp();
         return EXIT_SUCCESS;
     }
@@ -252,16 +245,13 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaGetDeviceCount(&num_devices));
 
     if (0 == num_devices) {
-    printf(
+        printf("your system does not have a CUDA capable device, waiving test...\n");
-        "your system does not have a CUDA capable device, waiving test...\n");
         return EXIT_WAIVED;
     }
 
     // check if the command-line chosen device ID is within range, exit if not
     if (cuda_device >= num_devices) {
-    printf(
+        printf("cuda_device=%d is invalid, must choose device ID between 0 and %d\n", cuda_device, num_devices - 1);
-        "cuda_device=%d is invalid, must choose device ID between 0 and %d\n",
-        cuda_device, num_devices - 1);
         return EXIT_FAILURE;
     }
 
@@ -276,12 +266,10 @@ int main(int argc, char **argv) {
     // Check if GPU can map host memory (Generic Method), if not then we override
     // bPinGenericMemory to be false
     if (bPinGenericMemory) {
-    printf("Device: <%s> canMapHostMemory: %s\n", deviceProp.name,
+        printf("Device: <%s> canMapHostMemory: %s\n", deviceProp.name, deviceProp.canMapHostMemory ? "Yes" : "No");
-           deviceProp.canMapHostMemory ? "Yes" : "No");
 
         if (deviceProp.canMapHostMemory == 0) {
-      printf(
+            printf("Using cudaMallocHost, CUDA device does not support mapping of "
-          "Using cudaMallocHost, CUDA device does not support mapping of "
                    "generic host memory\n");
             bPinGenericMemory = false;
         }
@@ -289,27 +277,22 @@ int main(int argc, char **argv) {
 
     // Anything that is less than 32 Cores will have scaled down workload
     scale_factor =
-      max((32.0f / (_ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) *
+        max((32.0f / (_ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) * (float)deviceProp.multiProcessorCount)),
-                    (float)deviceProp.multiProcessorCount)),
             1.0f);
     n = (int)rint((float)n / scale_factor);
 
-  printf("> CUDA Capable: SM %d.%d hardware\n", deviceProp.major,
+    printf("> CUDA Capable: SM %d.%d hardware\n", deviceProp.major, deviceProp.minor);
-         deviceProp.minor);
     printf("> %d Multiprocessor(s) x %d (Cores/Multiprocessor) = %d (Cores)\n",
            deviceProp.multiProcessorCount,
            _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor),
-         _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) *
+           _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) * deviceProp.multiProcessorCount);
-             deviceProp.multiProcessorCount);
 
     printf("> scale_factor = %1.4f\n", 1.0f / scale_factor);
     printf("> array_size   = %d\n\n", n);
 
     // enable use of blocking sync, to reduce CPU usage
-  printf("> Using CPU/GPU Device Synchronization method (%s)\n",
+    printf("> Using CPU/GPU Device Synchronization method (%s)\n", sDeviceSyncMethod[device_sync_method]);
-         sDeviceSyncMethod[device_sync_method]);
+    checkCudaErrors(cudaSetDeviceFlags(device_sync_method | (bPinGenericMemory ? cudaDeviceMapHost : 0)));
-  checkCudaErrors(cudaSetDeviceFlags(
-      device_sync_method | (bPinGenericMemory ? cudaDeviceMapHost : 0)));
 
     // allocate host memory
     int  c          = 5; // value to which the array will be initialized
@@ -332,8 +315,7 @@ int main(int argc, char **argv) {
     printf("\nStarting Test\n");
 
     // allocate and initialize an array of stream handles
-  cudaStream_t *streams =
+    cudaStream_t *streams = (cudaStream_t *)malloc(nstreams * sizeof(cudaStream_t));
-      (cudaStream_t *)malloc(nstreams * sizeof(cudaStream_t));
 
     for (int i = 0; i < nstreams; i++) {
         checkCudaErrors(cudaStreamCreate(&(streams[i])));
@@ -342,9 +324,7 @@ int main(int argc, char **argv) {
     // create CUDA event handles
     // use blocking sync
     cudaEvent_t start_event, stop_event;
-  int eventflags =
+    int eventflags = ((device_sync_method == cudaDeviceBlockingSync) ? cudaEventBlockingSync : cudaEventDefault);
-      ((device_sync_method == cudaDeviceBlockingSync) ? cudaEventBlockingSync
-                                                      : cudaEventDefault);
 
     checkCudaErrors(cudaEventCreateWithFlags(&start_event, eventflags));
     checkCudaErrors(cudaEventCreateWithFlags(&stop_event, eventflags));
@@ -354,11 +334,9 @@ int main(int argc, char **argv) {
                                                       // ensure that all previous
                                                       // CUDA calls have
                                                       // completed
-  checkCudaErrors(cudaMemcpyAsync(hAligned_a, d_a, nbytes,
+    checkCudaErrors(cudaMemcpyAsync(hAligned_a, d_a, nbytes, cudaMemcpyDeviceToHost, streams[0]));
-                                  cudaMemcpyDeviceToHost, streams[0]));
     checkCudaErrors(cudaEventRecord(stop_event, 0));
-  checkCudaErrors(cudaEventSynchronize(
+    checkCudaErrors(cudaEventSynchronize(stop_event)); // block until the event is actually recorded
-      stop_event));  // block until the event is actually recorded
     checkCudaErrors(cudaEventElapsedTime(&time_memcpy, start_event, stop_event));
     printf("memcopy:\t%.2f\n", time_memcpy);
 
@@ -380,8 +358,7 @@ int main(int argc, char **argv) {
 
     for (int k = 0; k < nreps; k++) {
         init_array<<<blocks, threads>>>(d_a, d_c, niterations);
-    checkCudaErrors(
+        checkCudaErrors(cudaMemcpy(hAligned_a, d_a, nbytes, cudaMemcpyDeviceToHost));
-        cudaMemcpy(hAligned_a, d_a, nbytes, cudaMemcpyDeviceToHost));
     }
 
     checkCudaErrors(cudaEventRecord(stop_event, 0));
@@ -395,16 +372,14 @@ int main(int argc, char **argv) {
     blocks  = dim3(n / (nstreams * threads.x), 1);
     memset(hAligned_a, 255,
            nbytes);                              // set host memory bits to all 1s, for testing correctness
-  checkCudaErrors(cudaMemset(
+    checkCudaErrors(cudaMemset(d_a, 0, nbytes)); // set device memory to all 0s, for testing correctness
-      d_a, 0, nbytes));  // set device memory to all 0s, for testing correctness
     checkCudaErrors(cudaEventRecord(start_event, 0));
 
     for (int k = 0; k < nreps; k++) {
         // asynchronously launch nstreams kernels, each operating on its own portion
         // of data
         for (int i = 0; i < nstreams; i++) {
-      init_array<<<blocks, threads, 0, streams[i]>>>(d_a + i * n / nstreams,
+            init_array<<<blocks, threads, 0, streams[i]>>>(d_a + i * n / nstreams, d_c, niterations);
-                                                     d_c, niterations);
         }
 
         // asynchronously launch nstreams memcopies.  Note that memcopy in stream x
@@ -413,8 +388,10 @@ int main(int argc, char **argv) {
         //   completed
         for (int i = 0; i < nstreams; i++) {
             checkCudaErrors(cudaMemcpyAsync(hAligned_a + i * n / nstreams,
-                                      d_a + i * n / nstreams, nbytes / nstreams,
+                                            d_a + i * n / nstreams,
-                                      cudaMemcpyDeviceToHost, streams[i]));
+                                            nbytes / nstreams,
+                                            cudaMemcpyDeviceToHost,
+                                            streams[i]));
         }
     }
 

Samples/0_Introduction/simpleSurfaceWrite/simpleSurfaceWrite.cu

@@ -34,10 +34,10 @@
  */
 
 // Includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 #ifdef _WIN32
 #define WINDOWS_LEAN_AND_MEAN
@@ -73,23 +73,22 @@ static const char *sampleName = "simpleSurfaceWrite";
 //! Write to a cuArray (texture data source) using surface writes
 //! @param gIData input data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void surfaceWriteKernel(float *gIData, int width, int height,
+__global__ void surfaceWriteKernel(float *gIData, int width, int height, cudaSurfaceObject_t outputSurface)
-                                   cudaSurfaceObject_t outputSurface) {
+{
     // calculate surface coordinates
     unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
     unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
 
     // read from global memory and write to cuarray (via surface reference)
-  surf2Dwrite(gIData[y * width + x], outputSurface, x * 4, y,
+    surf2Dwrite(gIData[y * width + x], outputSurface, x * 4, y, cudaBoundaryModeTrap);
-              cudaBoundaryModeTrap);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Transform an image using texture lookups
 //! @param gOData  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void transformKernel(float *gOData, int width, int height,
+__global__ void transformKernel(float *gOData, int width, int height, float theta, cudaTextureObject_t tex)
-                                float theta, cudaTextureObject_t tex) {
+{
     // calculate normalized texture coordinates
     unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
     unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
@@ -111,29 +110,29 @@ __global__ void transformKernel(float *gOData, int width, int height,
 // Declaration, forward
 void runTest(int argc, char **argv);
 
-extern "C" void computeGold(float *reference, float *idata,
+extern "C" void computeGold(float *reference, float *idata, const unsigned int len);
-                            const unsigned int len);
 
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     // Process command-line arguments
     if (argc > 1) {
         if (checkCmdLineFlag(argc, (const char **)argv, "input")) {
-      getCmdLineArgumentString(argc, (const char **)argv, "input",
+            getCmdLineArgumentString(argc, (const char **)argv, "input", (char **)&imageFilename);
-                               (char **)&imageFilename);
 
             if (checkCmdLineFlag(argc, (const char **)argv, "reference")) {
-        getCmdLineArgumentString(argc, (const char **)argv, "reference",
+                getCmdLineArgumentString(argc, (const char **)argv, "reference", (char **)&refFilename);
-                                 (char **)&refFilename);
+            }
-      } else {
+            else {
                 printf("-input flag should be used with -reference flag");
                 exit(EXIT_FAILURE);
             }
-    } else if (checkCmdLineFlag(argc, (const char **)argv, "reference")) {
+        }
+        else if (checkCmdLineFlag(argc, (const char **)argv, "reference")) {
             printf("-reference flag should be used with -input flag");
             exit(EXIT_FAILURE);
         }
@@ -141,15 +140,15 @@ int main(int argc, char **argv) {
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sampleName,
+    printf("%s completed, returned %s\n", sampleName, testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     // Use command-line specified CUDA device,
     // otherwise use device with highest Gflops/s
     int devID = findCudaDevice(argc, (const char **)argv);
@@ -159,7 +158,9 @@ void runTest(int argc, char **argv) {
 
     checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
     printf("CUDA device [%s] has %d Multi-Processors, SM %d.%d\n",
-         deviceProps.name, deviceProps.multiProcessorCount, deviceProps.major,
+           deviceProps.name,
+           deviceProps.multiProcessorCount,
+           deviceProps.major,
            deviceProps.minor);
 
     // Load image from disk
@@ -193,11 +194,9 @@ void runTest(int argc, char **argv) {
     checkCudaErrors(cudaMalloc((void **)&dData, size));
 
     // Allocate array and copy image data
-  cudaChannelFormatDesc channelDesc =
+    cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
-      cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
     cudaArray            *cuArray;
-  checkCudaErrors(cudaMallocArray(&cuArray, &channelDesc, width, height,
+    checkCudaErrors(cudaMallocArray(&cuArray, &channelDesc, width, height, cudaArraySurfaceLoadStore));
-                                  cudaArraySurfaceLoadStore));
 
     dim3 dimBlock(8, 8, 1);
     dim3 dimGrid(width / dimBlock.x, height / dimBlock.y, 1);
@@ -211,11 +210,9 @@ void runTest(int argc, char **argv) {
     checkCudaErrors(cudaCreateSurfaceObject(&outputSurface, &surfRes));
 #if 1
     checkCudaErrors(cudaMemcpy(dData, hData, size, cudaMemcpyHostToDevice));
-  surfaceWriteKernel<<<dimGrid, dimBlock>>>(dData, width, height,
+    surfaceWriteKernel<<<dimGrid, dimBlock>>>(dData, width, height, outputSurface);
-                                            outputSurface);
 #else // This is what differs from the example simpleTexture
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyToArray(cuArray, 0, 0, hData, size, cudaMemcpyHostToDevice));
-      cudaMemcpyToArray(cuArray, 0, 0, hData, size, cudaMemcpyHostToDevice));
 #endif
 
     cudaTextureObject_t tex;
@@ -254,8 +251,7 @@ void runTest(int argc, char **argv) {
     cudaDeviceSynchronize();
     sdkStopTimer(&timer);
     printf("Processing time: %f (ms)\n", sdkGetTimerValue(&timer));
-  printf("%.2f Mpixels/sec\n",
+    printf("%.2f Mpixels/sec\n", (width * height / (sdkGetTimerValue(&timer) / 1000.0f)) / 1e6);
-         (width * height / (sdkGetTimerValue(&timer) / 1000.0f)) / 1e6);
     sdkDeleteTimer(&timer);
 
     // Allocate mem for the result on host side
@@ -272,9 +268,9 @@ void runTest(int argc, char **argv) {
     // Write regression file if necessary
     if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
         // Write file for regression test
-    sdkWriteFile<float>("./data/regression.dat", hOData, width * height, 0.0f,
+        sdkWriteFile<float>("./data/regression.dat", hOData, width * height, 0.0f, false);
-                        false);
+    }
-  } else {
+    else {
         // We need to reload the data from disk,
         // because it is inverted upon output
         sdkLoadPGM(outputFilename, &hOData, &width, &height);
@@ -282,8 +278,7 @@ void runTest(int argc, char **argv) {
         printf("Comparing files\n");
         printf("\toutput:    <%s>\n", outputFilename);
         printf("\treference: <%s>\n", refPath);
-    testResult =
+        testResult = compareData(hOData, hDataRef, width * height, MIN_EPSILON_ERROR, 0.0f);
-        compareData(hOData, hDataRef, width * height, MIN_EPSILON_ERROR, 0.0f);
     }
 
     checkCudaErrors(cudaDestroySurfaceObject(outputSurface));

Samples/0_Introduction/simpleTemplates/sharedmem.cuh

@@ -68,10 +68,11 @@
 // this
 // struct by putting an undefined symbol in the function body so it won't
 // compile.
-template <typename T>
+template <typename T> struct SharedMemory
-struct SharedMemory {
+{
     // Ensure that we won't compile any un-specialized types
-  __device__ T *getPointer() {
+    __device__ T *getPointer()
+    {
         extern __device__ void error(void);
         error();
         return NULL;
@@ -82,89 +83,100 @@ struct SharedMemory {
 // int, uint, char, uchar, short, ushort, long, ulong, bool, float, and double
 // One could also specialize it for user-defined types.
 
-template <>
+template <> struct SharedMemory<int>
-struct SharedMemory<int> {
+{
-  __device__ int *getPointer() {
+    __device__ int *getPointer()
+    {
         extern __shared__ int s_int[];
         return s_int;
     }
 };
 
-template <>
+template <> struct SharedMemory<unsigned int>
-struct SharedMemory<unsigned int> {
+{
-  __device__ unsigned int *getPointer() {
+    __device__ unsigned int *getPointer()
+    {
         extern __shared__ unsigned int s_uint[];
         return s_uint;
     }
 };
 
-template <>
+template <> struct SharedMemory<char>
-struct SharedMemory<char> {
+{
-  __device__ char *getPointer() {
+    __device__ char *getPointer()
+    {
         extern __shared__ char s_char[];
         return s_char;
     }
 };
 
-template <>
+template <> struct SharedMemory<unsigned char>
-struct SharedMemory<unsigned char> {
+{
-  __device__ unsigned char *getPointer() {
+    __device__ unsigned char *getPointer()
+    {
         extern __shared__ unsigned char s_uchar[];
         return s_uchar;
     }
 };
 
-template <>
+template <> struct SharedMemory<short>
-struct SharedMemory<short> {
+{
-  __device__ short *getPointer() {
+    __device__ short *getPointer()
+    {
         extern __shared__ short s_short[];
         return s_short;
     }
 };
 
-template <>
+template <> struct SharedMemory<unsigned short>
-struct SharedMemory<unsigned short> {
+{
-  __device__ unsigned short *getPointer() {
+    __device__ unsigned short *getPointer()
+    {
         extern __shared__ unsigned short s_ushort[];
         return s_ushort;
     }
 };
 
-template <>
+template <> struct SharedMemory<long>
-struct SharedMemory<long> {
+{
-  __device__ long *getPointer() {
+    __device__ long *getPointer()
+    {
         extern __shared__ long s_long[];
         return s_long;
     }
 };
 
-template <>
+template <> struct SharedMemory<unsigned long>
-struct SharedMemory<unsigned long> {
+{
-  __device__ unsigned long *getPointer() {
+    __device__ unsigned long *getPointer()
+    {
         extern __shared__ unsigned long s_ulong[];
         return s_ulong;
     }
 };
 
-template <>
+template <> struct SharedMemory<bool>
-struct SharedMemory<bool> {
+{
-  __device__ bool *getPointer() {
+    __device__ bool *getPointer()
+    {
         extern __shared__ bool s_bool[];
         return s_bool;
     }
 };
 
-template <>
+template <> struct SharedMemory<float>
-struct SharedMemory<float> {
+{
-  __device__ float *getPointer() {
+    __device__ float *getPointer()
+    {
         extern __shared__ float s_float[];
         return s_float;
     }
 };
 
-template <>
+template <> struct SharedMemory<double>
-struct SharedMemory<double> {
+{
-  __device__ double *getPointer() {
+    __device__ double *getPointer()
+    {
         extern __shared__ double s_double[];
         return s_double;
     }

Samples/0_Introduction/simpleTemplates/simpleTemplates.cu

@@ -32,17 +32,17 @@
  */
 
 // System includes
-#include <stdio.h>
 #include <assert.h>
-#include <string.h>
 #include <math.h>
+#include <stdio.h>
+#include <string.h>
 
 // CUDA runtime
 #include <cuda_runtime.h>
 
 // helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 #ifndef MAX
 #define MAX(a, b) (a > b ? a : b)
@@ -58,8 +58,8 @@ int g_TotalFailures = 0;
 //! @param g_idata  input data in global memory
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-template <class T>
+template <class T> __global__ void testKernel(T *g_idata, T *g_odata)
-__global__ void testKernel(T *g_idata, T *g_odata) {
+{
     // Shared mem size is determined by the host app at run time
     SharedMemory<T> smem;
     T              *sdata = smem.getPointer();
@@ -83,11 +83,10 @@ __global__ void testKernel(T *g_idata, T *g_odata) {
 
 ////////////////////////////////////////////////////////////////////////////////
 // declaration, forward
-template <class T>
+template <class T> void runTest(int argc, char **argv, int len);
-void runTest(int argc, char **argv, int len);
 
-template <class T>
+template <class T> void computeGold(T *reference, T *idata, const unsigned int len)
-void computeGold(T *reference, T *idata, const unsigned int len) {
+{
     const T T_len = static_cast<T>(len);
 
     for (unsigned int i = 0; i < len; ++i) {
@@ -98,7 +97,8 @@ void computeGold(T *reference, T *idata, const unsigned int len) {
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("> runTest<float,32>\n");
     runTest<float>(argc, argv, 32);
     printf("> runTest<int,64>\n");
@@ -114,60 +114,63 @@ int main(int argc, char **argv) {
 // functions for different types.
 
 // Here's the generic wrapper for cutCompare*
-template <class T>
+template <class T> class ArrayComparator
-class ArrayComparator {
+{
 public:
-  bool compare(const T *reference, T *data, unsigned int len) {
+    bool compare(const T *reference, T *data, unsigned int len)
-    fprintf(stderr,
+    {
-            "Error: no comparison function implemented for this type\n");
+        fprintf(stderr, "Error: no comparison function implemented for this type\n");
         return false;
     }
 };
 
 // Here's the specialization for ints:
-template <>
+template <> class ArrayComparator<int>
-class ArrayComparator<int> {
+{
 public:
-  bool compare(const int *reference, int *data, unsigned int len) {
+    bool compare(const int *reference, int *data, unsigned int len)
+    {
         return compareData(reference, data, len, 0.15f, 0.0f);
     }
 };
 
 // Here's the specialization for floats:
-template <>
+template <> class ArrayComparator<float>
-class ArrayComparator<float> {
+{
 public:
-  bool compare(const float *reference, float *data, unsigned int len) {
+    bool compare(const float *reference, float *data, unsigned int len)
+    {
         return compareData(reference, data, len, 0.15f, 0.15f);
     }
 };
 
 // Here's the generic wrapper for cutWriteFile*
-template <class T>
+template <class T> class ArrayFileWriter
-class ArrayFileWriter {
+{
 public:
-  bool write(const char *filename, T *data, unsigned int len, float epsilon) {
+    bool write(const char *filename, T *data, unsigned int len, float epsilon)
-    fprintf(stderr,
+    {
-            "Error: no file write function implemented for this type\n");
+        fprintf(stderr, "Error: no file write function implemented for this type\n");
         return false;
     }
 };
 
 // Here's the specialization for ints:
-template <>
+template <> class ArrayFileWriter<int>
-class ArrayFileWriter<int> {
+{
 public:
-  bool write(const char *filename, int *data, unsigned int len, float epsilon) {
+    bool write(const char *filename, int *data, unsigned int len, float epsilon)
+    {
         return sdkWriteFile(filename, data, len, epsilon, false);
     }
 };
 
 // Here's the specialization for floats:
-template <>
+template <> class ArrayFileWriter<float>
-class ArrayFileWriter<float> {
+{
 public:
-  bool write(const char *filename, float *data, unsigned int len,
+    bool write(const char *filename, float *data, unsigned int len, float epsilon)
-             float epsilon) {
+    {
         return sdkWriteFile(filename, data, len, epsilon, false);
     }
 };
@@ -175,8 +178,8 @@ class ArrayFileWriter<float> {
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-template <class T>
+template <class T> void runTest(int argc, char **argv, int len)
-void runTest(int argc, char **argv, int len) {
+{
     int            devID;
     cudaDeviceProp deviceProps;
 
@@ -184,8 +187,7 @@ void runTest(int argc, char **argv, int len) {
 
     // get number of SMs on this GPU
     checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
-  printf("CUDA device [%s] has %d Multi-Processors\n", deviceProps.name,
+    printf("CUDA device [%s] has %d Multi-Processors\n", deviceProps.name, deviceProps.multiProcessorCount);
-         deviceProps.multiProcessorCount);
 
     // create and start timer
     StopWatchInterface *timer = NULL;
@@ -209,8 +211,7 @@ void runTest(int argc, char **argv, int len) {
     T *d_idata;
     checkCudaErrors(cudaMalloc((void **)&d_idata, mem_size));
     // copy host memory to device
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpy(d_idata, h_idata, mem_size, cudaMemcpyHostToDevice));
-      cudaMemcpy(d_idata, h_idata, mem_size, cudaMemcpyHostToDevice));
 
     // allocate device memory for result
     T *d_odata;
@@ -229,8 +230,7 @@ void runTest(int argc, char **argv, int len) {
     // allocate mem for the result on host side
     T *h_odata = (T *)malloc(mem_size);
     // copy result from device to host
-  checkCudaErrors(cudaMemcpy(h_odata, d_odata, sizeof(T) * num_threads,
+    checkCudaErrors(cudaMemcpy(h_odata, d_odata, sizeof(T) * num_threads, cudaMemcpyDeviceToHost));
-                             cudaMemcpyDeviceToHost));
 
     sdkStopTimer(&timer);
     printf("Processing time: %f (ms)\n", sdkGetTimerValue(&timer));
@@ -247,7 +247,8 @@ void runTest(int argc, char **argv, int len) {
     if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
         // write file for regression test
         writer.write("./data/regression.dat", h_odata, num_threads, 0.0f);
-  } else {
+    }
+    else {
         // custom output handling when no regression test running
         // in this case check if the result is equivalent to the expected solution
         bool res = comparator.compare(reference, h_odata, num_threads);

Samples/0_Introduction/simpleTexture/simpleTexture.cu

@@ -34,10 +34,10 @@
  */
 
 // Includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 #ifdef _WIN32
 #define WINDOWS_LEAN_AND_MEAN
@@ -73,8 +73,8 @@ bool testResult = true;
 //! Transform an image using texture lookups
 //! @param outputData  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void transformKernel(float *outputData, int width, int height,
+__global__ void transformKernel(float *outputData, int width, int height, float theta, cudaTextureObject_t tex)
-                                float theta, cudaTextureObject_t tex) {
+{
     // calculate normalized texture coordinates
     unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
     unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
@@ -98,23 +98,24 @@ void runTest(int argc, char **argv);
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     // Process command-line arguments
     if (argc > 1) {
         if (checkCmdLineFlag(argc, (const char **)argv, "input")) {
-      getCmdLineArgumentString(argc, (const char **)argv, "input",
+            getCmdLineArgumentString(argc, (const char **)argv, "input", (char **)&imageFilename);
-                               (char **)&imageFilename);
 
             if (checkCmdLineFlag(argc, (const char **)argv, "reference")) {
-        getCmdLineArgumentString(argc, (const char **)argv, "reference",
+                getCmdLineArgumentString(argc, (const char **)argv, "reference", (char **)&refFilename);
-                                 (char **)&refFilename);
+            }
-      } else {
+            else {
                 printf("-input flag should be used with -reference flag");
                 exit(EXIT_FAILURE);
             }
-    } else if (checkCmdLineFlag(argc, (const char **)argv, "reference")) {
+        }
+        else if (checkCmdLineFlag(argc, (const char **)argv, "reference")) {
             printf("-reference flag should be used with -input flag");
             exit(EXIT_FAILURE);
         }
@@ -122,15 +123,15 @@ int main(int argc, char **argv) {
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sampleName,
+    printf("%s completed, returned %s\n", sampleName, testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     int devID = findCudaDevice(argc, (const char **)argv);
 
     // load image from disk
@@ -164,12 +165,10 @@ void runTest(int argc, char **argv) {
     checkCudaErrors(cudaMalloc((void **)&dData, size));
 
     // Allocate array and copy image data
-  cudaChannelFormatDesc channelDesc =
+    cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
-      cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
     cudaArray            *cuArray;
     checkCudaErrors(cudaMallocArray(&cuArray, &channelDesc, width, height));
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyToArray(cuArray, 0, 0, hData, size, cudaMemcpyHostToDevice));
-      cudaMemcpyToArray(cuArray, 0, 0, hData, size, cudaMemcpyHostToDevice));
 
     cudaTextureObject_t tex;
     cudaResourceDesc    texRes;
@@ -209,8 +208,7 @@ void runTest(int argc, char **argv) {
     checkCudaErrors(cudaDeviceSynchronize());
     sdkStopTimer(&timer);
     printf("Processing time: %f (ms)\n", sdkGetTimerValue(&timer));
-  printf("%.2f Mpixels/sec\n",
+    printf("%.2f Mpixels/sec\n", (width * height / (sdkGetTimerValue(&timer) / 1000.0f)) / 1e6);
-         (width * height / (sdkGetTimerValue(&timer) / 1000.0f)) / 1e6);
     sdkDeleteTimer(&timer);
 
     // Allocate mem for the result on host side
@@ -228,9 +226,9 @@ void runTest(int argc, char **argv) {
     // Write regression file if necessary
     if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
         // Write file for regression test
-    sdkWriteFile<float>("./data/regression.dat", hOutputData, width * height,
+        sdkWriteFile<float>("./data/regression.dat", hOutputData, width * height, 0.0f, false);
-                        0.0f, false);
+    }
-  } else {
+    else {
         // We need to reload the data from disk,
         // because it is inverted upon output
         sdkLoadPGM(outputFilename, &hOutputData, &width, &height);
@@ -239,8 +237,7 @@ void runTest(int argc, char **argv) {
         printf("\toutput:    <%s>\n", outputFilename);
         printf("\treference: <%s>\n", refPath);
 
-    testResult = compareData(hOutputData, hDataRef, width * height,
+        testResult = compareData(hOutputData, hDataRef, width * height, MAX_EPSILON_ERROR, 0.15f);
-                             MAX_EPSILON_ERROR, 0.15f);
     }
 
     checkCudaErrors(cudaDestroyTextureObject(tex));

Samples/0_Introduction/simpleTexture3D/README.md

@@ -34,4 +34,3 @@ Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-d
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)
-

Samples/0_Introduction/simpleTexture3D/simpleTexture3D.cpp

@@ -32,11 +32,11 @@
   using 3D texture lookups.
 */
 
-#include <stdlib.h>
+#include <helper_gl.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
-#include <helper_gl.h>
 
 #if defined(__APPLE__) || defined(MACOSX)
 #pragma clang diagnostic ignored "-Wdeprecated-declarations"
@@ -49,9 +49,9 @@
 #endif
 
 // includes, cuda
-#include <vector_types.h>
-#include <cuda_runtime.h>
 #include <cuda_gl_interop.h>
+#include <cuda_runtime.h>
+#include <vector_types.h>
 
 // CUDA utilities and system includes
 #include <helper_cuda.h>
@@ -76,8 +76,7 @@ const dim3 gridSize(width / blockSize.x, height / blockSize.y);
 float w = 0.5; // texture coordinate in z
 
 GLuint                       pbo;               // OpenGL pixel buffer object
-struct cudaGraphicsResource
+struct cudaGraphicsResource *cuda_pbo_resource; // CUDA Graphics Resource (to transfer PBO)
-    *cuda_pbo_resource;  // CUDA Graphics Resource (to transfer PBO)
 
 bool linearFiltering = true;
 bool animate         = true;
@@ -105,13 +104,13 @@ char **pArgv = NULL;
 extern "C" void cleanup();
 extern "C" void setTextureFilterMode(bool bLinearFilter);
 extern "C" void initCuda(const uchar *h_volume, cudaExtent volumeSize);
-extern "C" void render_kernel(dim3 gridSize, dim3 blockSize, uint *d_output,
+extern "C" void render_kernel(dim3 gridSize, dim3 blockSize, uint *d_output, uint imageW, uint imageH, float w);
-                              uint imageW, uint imageH, float w);
 extern void     cleanupCuda();
 
 void loadVolumeData(char *exec_path);
 
-void computeFPS() {
+void computeFPS()
+{
     frameCount++;
     fpsCount++;
 
@@ -129,13 +128,13 @@ void computeFPS() {
 }
 
 // render image using CUDA
-void render() {
+void render()
+{
     // map PBO to get CUDA device pointer
     g_GraphicsMapFlag++;
     checkCudaErrors(cudaGraphicsMapResources(1, &cuda_pbo_resource, 0));
     size_t num_bytes;
-  checkCudaErrors(cudaGraphicsResourceGetMappedPointer(
+    checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&d_output, &num_bytes, cuda_pbo_resource));
-      (void **)&d_output, &num_bytes, cuda_pbo_resource));
     // printf("CUDA mapped PBO: May access %ld bytes\n", num_bytes);
 
     // call CUDA kernel, writing results to PBO
@@ -150,7 +149,8 @@ void render() {
 }
 
 // display results using OpenGL (called by GLUT)
-void display() {
+void display()
+{
     sdkStartTimer(&timer);
 
     render();
@@ -172,14 +172,16 @@ void display() {
     computeFPS();
 }
 
-void idle() {
+void idle()
+{
     if (animate) {
         w += 0.01f;
         glutPostRedisplay();
     }
 }
 
-void keyboard(unsigned char key, int x, int y) {
+void keyboard(unsigned char key, int x, int y)
+{
     switch (key) {
     case 27:
 #if defined(__APPLE__) || defined(MACOSX)
@@ -216,7 +218,8 @@ void keyboard(unsigned char key, int x, int y) {
     glutPostRedisplay();
 }
 
-void reshape(int x, int y) {
+void reshape(int x, int y)
+{
     glViewport(0, 0, x, y);
 
     glMatrixMode(GL_MODELVIEW);
@@ -227,7 +230,8 @@ void reshape(int x, int y) {
     glOrtho(0.0, 1.0, 0.0, 1.0, 0.0, 1.0);
 }
 
-void cleanup() {
+void cleanup()
+{
     sdkDeleteTimer(&timer);
 
     // add extra check to unmap the resource before unregistering it
@@ -242,21 +246,21 @@ void cleanup() {
     cleanupCuda();
 }
 
-void initGLBuffers() {
+void initGLBuffers()
+{
     // create pixel buffer object
     glGenBuffers(1, &pbo);
     glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo);
-  glBufferData(GL_PIXEL_UNPACK_BUFFER_ARB, width * height * sizeof(GLubyte) * 4,
+    glBufferData(GL_PIXEL_UNPACK_BUFFER_ARB, width * height * sizeof(GLubyte) * 4, 0, GL_STREAM_DRAW_ARB);
-               0, GL_STREAM_DRAW_ARB);
     glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
 
     // register this buffer object with CUDA
-  checkCudaErrors(cudaGraphicsGLRegisterBuffer(
+    checkCudaErrors(cudaGraphicsGLRegisterBuffer(&cuda_pbo_resource, pbo, cudaGraphicsMapFlagsWriteDiscard));
-      &cuda_pbo_resource, pbo, cudaGraphicsMapFlagsWriteDiscard));
 }
 
 // Load raw data from disk
-uchar *loadRawFile(const char *filename, size_t size) {
+uchar *loadRawFile(const char *filename, size_t size)
+{
     FILE *fp = fopen(filename, "rb");
 
     if (!fp) {
@@ -273,7 +277,8 @@ uchar *loadRawFile(const char *filename, size_t size) {
     return data;
 }
 
-void initGL(int *argc, char **argv) {
+void initGL(int *argc, char **argv)
+{
     // initialize GLUT callback functions
     glutInit(argc, argv);
     glutInitDisplayMode(GLUT_RGB | GLUT_DOUBLE);
@@ -284,16 +289,15 @@ void initGL(int *argc, char **argv) {
     glutReshapeFunc(reshape);
     glutIdleFunc(idle);
 
-  if (!isGLVersionSupported(2, 0) ||
+    if (!isGLVersionSupported(2, 0) || !areGLExtensionsSupported("GL_ARB_pixel_buffer_object")) {
-      !areGLExtensionsSupported("GL_ARB_pixel_buffer_object")) {
         fprintf(stderr, "Required OpenGL extensions are missing.");
         exit(EXIT_FAILURE);
     }
 }
 
-void runAutoTest(const char *ref_file, char *exec_path) {
+void runAutoTest(const char *ref_file, char *exec_path)
-  checkCudaErrors(
+{
-      cudaMalloc((void **)&d_output, width * height * sizeof(GLubyte) * 4));
+    checkCudaErrors(cudaMalloc((void **)&d_output, width * height * sizeof(GLubyte) * 4));
 
     // render the volumeData
     render_kernel(gridSize, blockSize, d_output, width, height, w);
@@ -302,15 +306,15 @@ void runAutoTest(const char *ref_file, char *exec_path) {
     getLastCudaError("render_kernel failed");
 
     void *h_output = malloc(width * height * sizeof(GLubyte) * 4);
-  checkCudaErrors(cudaMemcpy(h_output, d_output,
+    checkCudaErrors(cudaMemcpy(h_output, d_output, width * height * sizeof(GLubyte) * 4, cudaMemcpyDeviceToHost));
-                             width * height * sizeof(GLubyte) * 4,
+    sdkDumpBin(h_output, width * height * sizeof(GLubyte) * 4, "simpleTexture3D.bin");
-                             cudaMemcpyDeviceToHost));
-  sdkDumpBin(h_output, width * height * sizeof(GLubyte) * 4,
-             "simpleTexture3D.bin");
 
-  bool bTestResult = sdkCompareBin2BinFloat(
+    bool bTestResult = sdkCompareBin2BinFloat("simpleTexture3D.bin",
-      "simpleTexture3D.bin", sdkFindFilePath(ref_file, exec_path),
+                                              sdkFindFilePath(ref_file, exec_path),
-      width * height, MAX_EPSILON_ERROR, THRESHOLD, exec_path);
+                                              width * height,
+                                              MAX_EPSILON_ERROR,
+                                              THRESHOLD,
+                                              exec_path);
 
     checkCudaErrors(cudaFree(d_output));
     free(h_output);
@@ -321,13 +325,13 @@ void runAutoTest(const char *ref_file, char *exec_path) {
     exit(bTestResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-void loadVolumeData(char *exec_path) {
+void loadVolumeData(char *exec_path)
+{
     // load volume data
     const char *path = sdkFindFilePath(volumeFilename, exec_path);
 
     if (path == NULL) {
-    fprintf(stderr, "Error unable to find 3D Volume file: '%s'\n",
+        fprintf(stderr, "Error unable to find 3D Volume file: '%s'\n", volumeFilename);
-            volumeFilename);
         exit(EXIT_FAILURE);
     }
 
@@ -343,7 +347,8 @@ void loadVolumeData(char *exec_path) {
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     pArgc = &argc;
     pArgv = argv;
 
@@ -367,7 +372,8 @@ int main(int argc, char **argv) {
     if (ref_file) {
         loadVolumeData(argv[0]);
         runAutoTest(ref_file, argv[0]);
-  } else {
+    }
+    else {
         initGL(&argc, argv);
 
         // OpenGL buffers
@@ -376,8 +382,7 @@ int main(int argc, char **argv) {
         loadVolumeData(argv[0]);
     }
 
-  printf(
+    printf("Press space to toggle animation\n"
-      "Press space to toggle animation\n"
            "Press '+' and '-' to change displayed slice\n");
 
 #if defined(__APPLE__) || defined(MACOSX)

Samples/0_Introduction/simpleTexture3D/simpleTexture3D_kernel.cu

@@ -28,13 +28,12 @@
 #ifndef _SIMPLETEXTURE3D_KERNEL_CU_
 #define _SIMPLETEXTURE3D_KERNEL_CU_
 
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include <math.h>
-
 #include <helper_cuda.h>
 #include <helper_math.h>
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 typedef unsigned int  uint;
 typedef unsigned char uchar;
@@ -42,8 +41,8 @@ typedef unsigned char uchar;
 cudaArray          *d_volumeArray = 0;
 cudaTextureObject_t tex; // 3D texture
 
-__global__ void d_render(uint *d_output, uint imageW, uint imageH, float w,
+__global__ void d_render(uint *d_output, uint imageW, uint imageH, float w, cudaTextureObject_t texObj)
-                         cudaTextureObject_t texObj) {
+{
     uint x = __umul24(blockIdx.x, blockDim.x) + threadIdx.x;
     uint y = __umul24(blockIdx.y, blockDim.y) + threadIdx.y;
 
@@ -59,7 +58,8 @@ __global__ void d_render(uint *d_output, uint imageW, uint imageH, float w,
     }
 }
 
-extern "C" void setTextureFilterMode(bool bLinearFilter) {
+extern "C" void setTextureFilterMode(bool bLinearFilter)
+{
     if (tex) {
         checkCudaErrors(cudaDestroyTextureObject(tex));
     }
@@ -73,8 +73,7 @@ extern "C" void setTextureFilterMode(bool bLinearFilter) {
     memset(&texDescr, 0, sizeof(cudaTextureDesc));
 
     texDescr.normalizedCoords = true;
-  texDescr.filterMode =
+    texDescr.filterMode       = bLinearFilter ? cudaFilterModeLinear : cudaFilterModePoint;
-      bLinearFilter ? cudaFilterModeLinear : cudaFilterModePoint;
     ;
     texDescr.addressMode[0] = cudaAddressModeWrap;
     texDescr.addressMode[1] = cudaAddressModeWrap;
@@ -84,7 +83,8 @@ extern "C" void setTextureFilterMode(bool bLinearFilter) {
     checkCudaErrors(cudaCreateTextureObject(&tex, &texRes, &texDescr, NULL));
 }
 
-extern "C" void initCuda(const uchar *h_volume, cudaExtent volumeSize) {
+extern "C" void initCuda(const uchar *h_volume, cudaExtent volumeSize)
+{
     // create 3D array
     cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc<uchar>();
     checkCudaErrors(cudaMalloc3DArray(&d_volumeArray, &channelDesc, volumeSize));
@@ -92,8 +92,7 @@ extern "C" void initCuda(const uchar *h_volume, cudaExtent volumeSize) {
     // copy data to 3D array
     cudaMemcpy3DParms copyParams = {0};
     copyParams.srcPtr =
-      make_cudaPitchedPtr((void *)h_volume, volumeSize.width * sizeof(uchar),
+        make_cudaPitchedPtr((void *)h_volume, volumeSize.width * sizeof(uchar), volumeSize.width, volumeSize.height);
-                          volumeSize.width, volumeSize.height);
     copyParams.dstArray = d_volumeArray;
     copyParams.extent   = volumeSize;
     copyParams.kind     = cudaMemcpyHostToDevice;
@@ -121,12 +120,13 @@ extern "C" void initCuda(const uchar *h_volume, cudaExtent volumeSize) {
     checkCudaErrors(cudaCreateTextureObject(&tex, &texRes, &texDescr, NULL));
 }
 
-extern "C" void render_kernel(dim3 gridSize, dim3 blockSize, uint *d_output,
+extern "C" void render_kernel(dim3 gridSize, dim3 blockSize, uint *d_output, uint imageW, uint imageH, float w)
-                              uint imageW, uint imageH, float w) {
+{
     d_render<<<gridSize, blockSize>>>(d_output, imageW, imageH, w, tex);
 }
 
-void cleanupCuda() {
+void cleanupCuda()
+{
     if (tex) {
         checkCudaErrors(cudaDestroyTextureObject(tex));
     }

Samples/0_Introduction/simpleTextureDrv/simpleTextureDrv.cpp

@@ -39,16 +39,16 @@
  */
 
 // includes, system
-#include <stdlib.h>
+#include <cstring>
+#include <iostream>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
-#include <iostream>
-#include <cstring>
 
 // includes, CUDA
-#include <cuda.h>
 #include <builtin_types.h>
+#include <cuda.h>
 // includes, project
 #include <helper_cuda_drvapi.h>
 #include <helper_functions.h>
@@ -65,8 +65,7 @@ float angle = 0.5f;  // angle to rotate image by (in radians)
 // declaration, forward
 void runTest(int argc, char **argv);
 
-extern "C" void computeGold(float *reference, float *idata,
+extern "C" void computeGold(float *reference, float *idata, const unsigned int len);
-                            const unsigned int len);
 
 static CUresult initCUDA(int argc, char **argv, CUfunction *);
 
@@ -84,7 +83,8 @@ CUdevice cuDevice;
 CUcontext cuContext;
 CUmodule  cuModule;
 
-void showHelp() {
+void showHelp()
+{
     printf("\n> [%s] Command line options\n", sSDKsample);
     printf("\t-device=n          (where n=0,1,2.... for the GPU device)\n\n");
 }
@@ -92,7 +92,8 @@ void showHelp() {
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     if (checkCmdLineFlag(argc, (const char **)argv, "help")) {
         showHelp();
         return 0;
@@ -104,7 +105,8 @@ int main(int argc, char **argv) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     bool bTestResults = true;
 
     // initialize CUDA
@@ -191,18 +193,17 @@ void runTest(int argc, char **argv) {
         // Launching (simpler method)
         void *args[5] = {&d_data, &width, &height, &angle, &TexObject};
 
-    checkCudaErrors(cuLaunchKernel(transform, (width / block_size),
+        checkCudaErrors(cuLaunchKernel(
-                                   (height / block_size), 1, block_size,
+            transform, (width / block_size), (height / block_size), 1, block_size, block_size, 1, 0, NULL, args, NULL));
-                                   block_size, 1, 0, NULL, args, NULL));
         checkCudaErrors(cuCtxSynchronize());
         sdkCreateTimer(&timer);
         sdkStartTimer(&timer);
 
         // launch kernel again for performance measurement
-    checkCudaErrors(cuLaunchKernel(transform, (width / block_size),
+        checkCudaErrors(cuLaunchKernel(
-                                   (height / block_size), 1, block_size,
+            transform, (width / block_size), (height / block_size), 1, block_size, block_size, 1, 0, NULL, args, NULL));
-                                   block_size, 1, 0, NULL, args, NULL));
+    }
-  } else {
+    else {
         // This is the new CUDA 4.0 API for Kernel Parameter passing and Kernel
         // Launching (advanced method)
         int  offset = 0;
@@ -222,29 +223,43 @@ void runTest(int argc, char **argv) {
         *((CUtexObject *)&argBuffer[offset]) = TexObject;
         offset += sizeof(TexObject);
 
-    void *kernel_launch_config[5] = {CU_LAUNCH_PARAM_BUFFER_POINTER, argBuffer,
+        void *kernel_launch_config[5] = {
-                                     CU_LAUNCH_PARAM_BUFFER_SIZE, &offset,
+            CU_LAUNCH_PARAM_BUFFER_POINTER, argBuffer, CU_LAUNCH_PARAM_BUFFER_SIZE, &offset, CU_LAUNCH_PARAM_END};
-                                     CU_LAUNCH_PARAM_END};
 
         // new CUDA 4.0 Driver API Kernel launch call (warmup)
-    checkCudaErrors(cuLaunchKernel(
+        checkCudaErrors(cuLaunchKernel(transform,
-        transform, (width / block_size), (height / block_size), 1, block_size,
+                                       (width / block_size),
-        block_size, 1, 0, NULL, NULL, (void **)&kernel_launch_config));
+                                       (height / block_size),
+                                       1,
+                                       block_size,
+                                       block_size,
+                                       1,
+                                       0,
+                                       NULL,
+                                       NULL,
+                                       (void **)&kernel_launch_config));
         checkCudaErrors(cuCtxSynchronize());
         sdkCreateTimer(&timer);
         sdkStartTimer(&timer);
 
         // launch kernel again for performance measurement
-    checkCudaErrors(cuLaunchKernel(
+        checkCudaErrors(cuLaunchKernel(transform,
-        transform, (width / block_size), (height / block_size), 1, block_size,
+                                       (width / block_size),
-        block_size, 1, 0, 0, NULL, (void **)&kernel_launch_config));
+                                       (height / block_size),
+                                       1,
+                                       block_size,
+                                       block_size,
+                                       1,
+                                       0,
+                                       0,
+                                       NULL,
+                                       (void **)&kernel_launch_config));
     }
 
     checkCudaErrors(cuCtxSynchronize());
     sdkStopTimer(&timer);
     printf("Processing time: %f (ms)\n", sdkGetTimerValue(&timer));
-  printf("%.2f Mpixels/sec\n",
+    printf("%.2f Mpixels/sec\n", (width * height / (sdkGetTimerValue(&timer) / 1000.0f)) / 1e6);
-         (width * height / (sdkGetTimerValue(&timer) / 1000.0f)) / 1e6);
     sdkDeleteTimer(&timer);
 
     // allocate mem for the result on host side
@@ -262,17 +277,16 @@ void runTest(int argc, char **argv) {
     // write regression file if necessary
     if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
         // write file for regression test
-    sdkWriteFile<float>("./data/regression.dat", h_odata, width * height, 0.0f,
+        sdkWriteFile<float>("./data/regression.dat", h_odata, width * height, 0.0f, false);
-                        false);
+    }
-  } else {
+    else {
         // We need to reload the data from disk, because it is inverted upon output
         sdkLoadPGM(output_filename, &h_odata, &width, &height);
 
         printf("Comparing files\n");
         printf("\toutput:    <%s>\n", output_filename);
         printf("\treference: <%s>\n", ref_path);
-    bTestResults = compareData(h_odata, h_data_ref, width * height,
+        bTestResults = compareData(h_odata, h_data_ref, width * height, MIN_EPSILON_ERROR, 0.15f);
-                               MIN_EPSILON_ERROR, 0.15f);
     }
 
     // cleanup memory
@@ -293,7 +307,8 @@ void runTest(int argc, char **argv) {
 //! kernel function.  After the module is loaded, cuModuleGetFunction
 //! retrieves the CUDA function pointer "cuFunction"
 ////////////////////////////////////////////////////////////////////////////////
-static CUresult initCUDA(int argc, char **argv, CUfunction *transform) {
+static CUresult initCUDA(int argc, char **argv, CUfunction *transform)
+{
     CUfunction cuFunction = 0;
     int        major = 0, minor = 0, devID = 0;
     char       deviceName[100];
@@ -302,10 +317,8 @@ static CUresult initCUDA(int argc, char **argv, CUfunction *transform) {
     cuDevice = findCudaDeviceDRV(argc, (const char **)argv);
 
     // get compute capabilities and the devicename
-  checkCudaErrors(cuDeviceGetAttribute(
+    checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
-      &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
+    checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
-  checkCudaErrors(cuDeviceGetAttribute(
-      &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
     checkCudaErrors(cuDeviceGetName(deviceName, sizeof(deviceName), cuDevice));
     printf("> GPU Device has SM %d.%d compute capability\n", major, minor);
 
@@ -316,7 +329,8 @@ static CUresult initCUDA(int argc, char **argv, CUfunction *transform) {
 
     if (!findFatbinPath(FATBIN_FILE, module_path, argv, fatbin)) {
         exit(EXIT_FAILURE);
-  } else {
+    }
+    else {
         printf("> initCUDA loading module: <%s>\n", module_path.c_str());
     }
 
@@ -328,8 +342,7 @@ static CUresult initCUDA(int argc, char **argv, CUfunction *transform) {
     // Create module from binary file (FATBIN)
     checkCudaErrors(cuModuleLoadData(&cuModule, fatbin.str().c_str()));
 
-  checkCudaErrors(
+    checkCudaErrors(cuModuleGetFunction(&cuFunction, cuModule, "transformKernel"));
-      cuModuleGetFunction(&cuFunction, cuModule, "transformKernel"));
 
     *transform = cuFunction;
 

Samples/0_Introduction/simpleTextureDrv/simpleTexture_kernel.cu

@@ -33,9 +33,8 @@
 //! Transform an image using texture lookups
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" __global__ void transformKernel(float *g_odata, int width,
+extern "C" __global__ void transformKernel(float *g_odata, int width, int height, float theta, CUtexObject tex)
-                                           int height, float theta,
+{
-                                           CUtexObject tex) {
     // calculate normalized texture coordinates
     unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
     unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;

Samples/0_Introduction/simpleVoteIntrinsics/simpleVoteIntrinsics.cu

@@ -53,7 +53,8 @@ static const char *sSDKsample = "[simpleVoteIntrinsics]\0";
 #include "simpleVote_kernel.cuh"
 
 // Generate the test pattern for Tests 1 and 2
-void genVoteTestPattern(unsigned int *VOTE_PATTERN, int size) {
+void genVoteTestPattern(unsigned int *VOTE_PATTERN, int size)
+{
     // For testing VOTE.Any (all of these threads will return 0)
     for (int i = 0; i < size / 4; i++) {
         VOTE_PATTERN[i] = 0x00000000;
@@ -75,8 +76,8 @@ void genVoteTestPattern(unsigned int *VOTE_PATTERN, int size) {
     }
 }
 
-int checkErrors1(unsigned int *h_result, int start, int end, int warp_size,
+int checkErrors1(unsigned int *h_result, int start, int end, int warp_size, const char *voteType)
-                 const char *voteType) {
+{
     int i, sum = 0;
 
     for (sum = 0, i = start; i < end; i++) {
@@ -96,8 +97,8 @@ int checkErrors1(unsigned int *h_result, int start, int end, int warp_size,
     return (sum > 0);
 }
 
-int checkErrors2(unsigned int *h_result, int start, int end, int warp_size,
+int checkErrors2(unsigned int *h_result, int start, int end, int warp_size, const char *voteType)
-                 const char *voteType) {
+{
     int i, sum = 0;
 
     for (sum = 0, i = start; i < end; i++) {
@@ -118,49 +119,42 @@ int checkErrors2(unsigned int *h_result, int start, int end, int warp_size,
 }
 
 // Verification code for Kernel #1
-int checkResultsVoteAnyKernel1(unsigned int *h_result, int size,
+int checkResultsVoteAnyKernel1(unsigned int *h_result, int size, int warp_size)
-                               int warp_size) {
+{
     int error_count = 0;
 
-  error_count += checkErrors1(h_result, 0, VOTE_DATA_GROUP * warp_size / 4,
+    error_count += checkErrors1(h_result, 0, VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.Any");
-                              warp_size, "Vote.Any");
+    error_count += checkErrors2(
-  error_count +=
+        h_result, VOTE_DATA_GROUP * warp_size / 4, 2 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.Any");
-      checkErrors2(h_result, VOTE_DATA_GROUP * warp_size / 4,
+    error_count += checkErrors2(
-                   2 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.Any");
+        h_result, 2 * VOTE_DATA_GROUP * warp_size / 4, 3 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.Any");
-  error_count +=
+    error_count += checkErrors2(
-      checkErrors2(h_result, 2 * VOTE_DATA_GROUP * warp_size / 4,
+        h_result, 3 * VOTE_DATA_GROUP * warp_size / 4, 4 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.Any");
-                   3 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.Any");
-  error_count +=
-      checkErrors2(h_result, 3 * VOTE_DATA_GROUP * warp_size / 4,
-                   4 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.Any");
 
     printf((error_count == 0) ? "\tOK\n" : "\tERROR\n");
     return error_count;
 }
 
 // Verification code for Kernel #2
-int checkResultsVoteAllKernel2(unsigned int *h_result, int size,
+int checkResultsVoteAllKernel2(unsigned int *h_result, int size, int warp_size)
-                               int warp_size) {
+{
     int error_count = 0;
 
-  error_count += checkErrors1(h_result, 0, VOTE_DATA_GROUP * warp_size / 4,
+    error_count += checkErrors1(h_result, 0, VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.All");
-                              warp_size, "Vote.All");
+    error_count += checkErrors1(
-  error_count +=
+        h_result, VOTE_DATA_GROUP * warp_size / 4, 2 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.All");
-      checkErrors1(h_result, VOTE_DATA_GROUP * warp_size / 4,
+    error_count += checkErrors1(
-                   2 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.All");
+        h_result, 2 * VOTE_DATA_GROUP * warp_size / 4, 3 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.All");
-  error_count +=
+    error_count += checkErrors2(
-      checkErrors1(h_result, 2 * VOTE_DATA_GROUP * warp_size / 4,
+        h_result, 3 * VOTE_DATA_GROUP * warp_size / 4, 4 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.All");
-                   3 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.All");
-  error_count +=
-      checkErrors2(h_result, 3 * VOTE_DATA_GROUP * warp_size / 4,
-                   4 * VOTE_DATA_GROUP * warp_size / 4, warp_size, "Vote.All");
 
     printf((error_count == 0) ? "\tOK\n" : "\tERROR\n");
     return error_count;
 }
 
 // Verification code for Kernel #3
-int checkResultsVoteAnyKernel3(bool *hinfo, int size) {
+int checkResultsVoteAnyKernel3(bool *hinfo, int size)
+{
     int i, error_count = 0;
 
     for (i = 0; i < size * 3; i++) {
@@ -198,7 +192,8 @@ int checkResultsVoteAnyKernel3(bool *hinfo, int size) {
     return error_count;
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     unsigned int *h_input, *h_result;
     unsigned int *d_input, *d_result;
 
@@ -216,24 +211,20 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaGetDeviceProperties(&deviceProp, devID));
 
     // Statistics about the GPU device
-  printf(
+    printf("> GPU device has %d Multi-Processors, SM %d.%d compute capabilities\n\n",
-      "> GPU device has %d Multi-Processors, SM %d.%d compute capabilities\n\n",
+           deviceProp.multiProcessorCount,
-      deviceProp.multiProcessorCount, deviceProp.major, deviceProp.minor);
+           deviceProp.major,
-
+           deviceProp.minor);
-  h_input = (unsigned int *)malloc(VOTE_DATA_GROUP * warp_size *
+
-                                   sizeof(unsigned int));
+    h_input  = (unsigned int *)malloc(VOTE_DATA_GROUP * warp_size * sizeof(unsigned int));
-  h_result = (unsigned int *)malloc(VOTE_DATA_GROUP * warp_size *
+    h_result = (unsigned int *)malloc(VOTE_DATA_GROUP * warp_size * sizeof(unsigned int));
-                                    sizeof(unsigned int));
     checkCudaErrors(
-      cudaMalloc(reinterpret_cast<void **>(&d_input),
+        cudaMalloc(reinterpret_cast<void **>(&d_input), VOTE_DATA_GROUP * warp_size * sizeof(unsigned int)));
-                 VOTE_DATA_GROUP * warp_size * sizeof(unsigned int)));
     checkCudaErrors(
-      cudaMalloc(reinterpret_cast<void **>(&d_result),
+        cudaMalloc(reinterpret_cast<void **>(&d_result), VOTE_DATA_GROUP * warp_size * sizeof(unsigned int)));
-                 VOTE_DATA_GROUP * warp_size * sizeof(unsigned int)));
     genVoteTestPattern(h_input, VOTE_DATA_GROUP * warp_size);
-  checkCudaErrors(cudaMemcpy(d_input, h_input,
+    checkCudaErrors(
-                             VOTE_DATA_GROUP * warp_size * sizeof(unsigned int),
+        cudaMemcpy(d_input, h_input, VOTE_DATA_GROUP * warp_size * sizeof(unsigned int), cudaMemcpyHostToDevice));
-                             cudaMemcpyHostToDevice));
 
     // Start of Vote Any Test Kernel #1
     printf("[VOTE Kernel Test 1/3]\n");
@@ -242,16 +233,13 @@ int main(int argc, char **argv) {
         checkCudaErrors(cudaDeviceSynchronize());
         dim3 gridBlock(1, 1);
         dim3 threadBlock(VOTE_DATA_GROUP * warp_size, 1);
-    VoteAnyKernel1<<<gridBlock, threadBlock>>>(d_input, d_result,
+        VoteAnyKernel1<<<gridBlock, threadBlock>>>(d_input, d_result, VOTE_DATA_GROUP * warp_size);
-                                               VOTE_DATA_GROUP * warp_size);
         getLastCudaError("VoteAnyKernel() execution failed\n");
         checkCudaErrors(cudaDeviceSynchronize());
     }
-  checkCudaErrors(cudaMemcpy(h_result, d_result,
+    checkCudaErrors(
-                             VOTE_DATA_GROUP * warp_size * sizeof(unsigned int),
+        cudaMemcpy(h_result, d_result, VOTE_DATA_GROUP * warp_size * sizeof(unsigned int), cudaMemcpyDeviceToHost));
-                             cudaMemcpyDeviceToHost));
+    error_count[0] += checkResultsVoteAnyKernel1(h_result, VOTE_DATA_GROUP * warp_size, warp_size);
-  error_count[0] += checkResultsVoteAnyKernel1(
-      h_result, VOTE_DATA_GROUP * warp_size, warp_size);
 
     // Start of Vote All Test Kernel #2
     printf("\n[VOTE Kernel Test 2/3]\n");
@@ -260,23 +248,18 @@ int main(int argc, char **argv) {
         checkCudaErrors(cudaDeviceSynchronize());
         dim3 gridBlock(1, 1);
         dim3 threadBlock(VOTE_DATA_GROUP * warp_size, 1);
-    VoteAllKernel2<<<gridBlock, threadBlock>>>(d_input, d_result,
+        VoteAllKernel2<<<gridBlock, threadBlock>>>(d_input, d_result, VOTE_DATA_GROUP * warp_size);
-                                               VOTE_DATA_GROUP * warp_size);
         getLastCudaError("VoteAllKernel() execution failed\n");
         checkCudaErrors(cudaDeviceSynchronize());
     }
-  checkCudaErrors(cudaMemcpy(h_result, d_result,
+    checkCudaErrors(
-                             VOTE_DATA_GROUP * warp_size * sizeof(unsigned int),
+        cudaMemcpy(h_result, d_result, VOTE_DATA_GROUP * warp_size * sizeof(unsigned int), cudaMemcpyDeviceToHost));
-                             cudaMemcpyDeviceToHost));
+    error_count[1] += checkResultsVoteAllKernel2(h_result, VOTE_DATA_GROUP * warp_size, warp_size);
-  error_count[1] += checkResultsVoteAllKernel2(
-      h_result, VOTE_DATA_GROUP * warp_size, warp_size);
 
     // Second Vote Kernel Test #3 (both Any/All)
     hinfo = reinterpret_cast<bool *>(calloc(warp_size * 3 * 3, sizeof(bool)));
-  cudaMalloc(reinterpret_cast<void **>(&dinfo),
+    cudaMalloc(reinterpret_cast<void **>(&dinfo), warp_size * 3 * 3 * sizeof(bool));
-             warp_size * 3 * 3 * sizeof(bool));
+    cudaMemcpy(dinfo, hinfo, warp_size * 3 * 3 * sizeof(bool), cudaMemcpyHostToDevice);
-  cudaMemcpy(dinfo, hinfo, warp_size * 3 * 3 * sizeof(bool),
-             cudaMemcpyHostToDevice);
 
     printf("\n[VOTE Kernel Test 3/3]\n");
     printf("\tRunning <<Vote.Any>> kernel3 ...\n");
@@ -286,8 +269,7 @@ int main(int argc, char **argv) {
         checkCudaErrors(cudaDeviceSynchronize());
     }
 
-  cudaMemcpy(hinfo, dinfo, warp_size * 3 * 3 * sizeof(bool),
+    cudaMemcpy(hinfo, dinfo, warp_size * 3 * 3 * sizeof(bool), cudaMemcpyDeviceToHost);
-             cudaMemcpyDeviceToHost);
 
     error_count[2] = checkResultsVoteAnyKernel3(hinfo, warp_size * 3);
 
@@ -303,7 +285,5 @@ int main(int argc, char **argv) {
 
     printf("\tShutting down...\n");
 
-  return (error_count[0] == 0 && error_count[1] == 0 && error_count[2] == 0)
+    return (error_count[0] == 0 && error_count[1] == 0 && error_count[2] == 0) ? EXIT_SUCCESS : EXIT_FAILURE;
-             ? EXIT_SUCCESS
-             : EXIT_FAILURE;
 }

Samples/0_Introduction/simpleVoteIntrinsics/simpleVote_kernel.cuh

@@ -38,8 +38,8 @@
 // If ANY one of the threads (within the warp) of the predicated condition
 // returns a non-zero value, then all threads within this warp will return a
 // non-zero value
-__global__ void VoteAnyKernel1(unsigned int *input, unsigned int *result,
+__global__ void VoteAnyKernel1(unsigned int *input, unsigned int *result, int size)
-                               int size) {
+{
     int tx = threadIdx.x;
 
     int mask   = 0xffffffff;
@@ -50,8 +50,8 @@ __global__ void VoteAnyKernel1(unsigned int *input, unsigned int *result,
 // If ALL of the threads (within the warp) of the predicated condition returns
 // a non-zero value, then all threads within this warp will return a non-zero
 // value
-__global__ void VoteAllKernel2(unsigned int *input, unsigned int *result,
+__global__ void VoteAllKernel2(unsigned int *input, unsigned int *result, int size)
-                               int size) {
+{
     int tx = threadIdx.x;
 
     int mask   = 0xffffffff;
@@ -60,7 +60,8 @@ __global__ void VoteAllKernel2(unsigned int *input, unsigned int *result,
 
 // Kernel #3 is a directed test for the across-the-warp vote(all) intrinsic.
 // This kernel will test for conditions across warps, and within half warps
-__global__ void VoteAnyKernel3(bool *info, int warp_size) {
+__global__ void VoteAnyKernel3(bool *info, int warp_size)
+{
     int          tx   = threadIdx.x;
     unsigned int mask = 0xffffffff;
     bool        *offs = info + (tx * 3);

Samples/0_Introduction/simpleZeroCopy/simpleZeroCopy.cu

@@ -41,7 +41,8 @@
 #endif
 
 /* Add two vectors on the GPU */
-__global__ void vectorAddGPU(float *a, float *b, float *c, int N) {
+__global__ void vectorAddGPU(float *a, float *b, float *c, int N)
+{
     int idx = blockIdx.x * blockDim.x + threadIdx.x;
 
     if (idx < N) {
@@ -57,7 +58,8 @@ bool bPinGenericMemory = false;
 #define MEMORY_ALIGNMENT  4096
 #define ALIGN_UP(x, size) (((size_t)x + (size - 1)) & (~(size - 1)))
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     int            n, nelem, deviceCount;
     int            idev   = 0; // use default device 0
     char          *device = NULL;
@@ -73,8 +75,7 @@ int main(int argc, char **argv) {
         printf("Usage:  simpleZeroCopy [OPTION]\n\n");
         printf("Options:\n");
         printf("  --device=[device #]  Specify the device to be used\n");
-    printf(
+        printf("  --use_generic_memory (optional) use generic page-aligned for system "
-        "  --use_generic_memory (optional) use generic page-aligned for system "
                "memory\n");
         return EXIT_SUCCESS;
     }
@@ -85,9 +86,7 @@ int main(int argc, char **argv) {
         idev = atoi(device);
 
         if (idev >= deviceCount || idev < 0) {
-      fprintf(stderr,
+            fprintf(stderr, "Device number %d is invalid, will use default CUDA device 0.\n", idev);
-              "Device number %d is invalid, will use default CUDA device 0.\n",
-              idev);
             idev = 0;
         }
     }
@@ -108,7 +107,8 @@ int main(int argc, char **argv) {
 
     if (bPinGenericMemory) {
         printf("> Using Generic System Paged Memory (malloc)\n");
-  } else {
+    }
+    else {
         printf("> Using CUDA Host Allocated (cudaHostAlloc)\n");
     }
 
@@ -122,8 +122,7 @@ int main(int argc, char **argv) {
 #if CUDART_VERSION >= 2020
 
     if (!deviceProp.canMapHostMemory) {
-    fprintf(stderr, "Device %d does not support mapping CPU host memory!\n",
+        fprintf(stderr, "Device %d does not support mapping CPU host memory!\n", idev);
-            idev);
 
         exit(EXIT_SUCCESS);
     }
@@ -133,7 +132,9 @@ int main(int argc, char **argv) {
     fprintf(stderr,
             "CUDART version %d.%d does not support "
             "<cudaDeviceProp.canMapHostMemory> field\n",
-          , CUDART_VERSION / 1000, (CUDART_VERSION % 100) / 10);
+            ,
+            CUDART_VERSION / 1000,
+            (CUDART_VERSION % 100) / 10);
 
     exit(EXIT_SUCCESS);
 #endif
@@ -141,10 +142,10 @@ int main(int argc, char **argv) {
 #if CUDART_VERSION < 4000
 
     if (bPinGenericMemory) {
-    fprintf(
+        fprintf(stderr,
-        stderr,
                 "CUDART version %d.%d does not support <cudaHostRegister> function\n",
-        CUDART_VERSION / 1000, (CUDART_VERSION % 100) / 10);
+                CUDART_VERSION / 1000,
+                (CUDART_VERSION % 100) / 10);
 
         exit(EXIT_SUCCESS);
     }
@@ -172,7 +173,8 @@ int main(int argc, char **argv) {
         checkCudaErrors(cudaHostRegister(b, bytes, cudaHostRegisterMapped));
         checkCudaErrors(cudaHostRegister(c, bytes, cudaHostRegisterMapped));
 #endif
-  } else {
+    }
+    else {
 #if CUDART_VERSION >= 2020
         flags = cudaHostAllocMapped;
         checkCudaErrors(cudaHostAlloc((void **)&a, bytes, flags));
@@ -235,7 +237,8 @@ int main(int argc, char **argv) {
         free(b_UA);
         free(c_UA);
 #endif
-  } else {
+    }
+    else {
 #if CUDART_VERSION >= 2020
         checkCudaErrors(cudaFreeHost(a));
         checkCudaErrors(cudaFreeHost(b));

Samples/0_Introduction/systemWideAtomics/README.md

@@ -34,4 +34,3 @@ Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-d
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)
-

Samples/0_Introduction/systemWideAtomics/systemWideAtomics.cu

@@ -29,19 +29,20 @@
  * memory.
  */
 
+#include <cstdio>
+#include <ctime>
 #include <cuda_runtime.h>
 #include <helper_cuda.h>
 #include <math.h>
 #include <stdint.h>
-#include <cstdio>
-#include <ctime>
 
 #define min(a, b) (a) < (b) ? (a) : (b)
 #define max(a, b) (a) > (b) ? (a) : (b)
 
 #define LOOP_NUM 50
 
-__global__ void atomicKernel(int *atom_arr) {
+__global__ void atomicKernel(int *atom_arr)
+{
     unsigned int tid = blockDim.x * blockIdx.x + threadIdx.x;
 
     for (int i = 0; i < LOOP_NUM; i++) {
@@ -79,7 +80,8 @@ __global__ void atomicKernel(int *atom_arr) {
     }
 }
 
-void atomicKernel_CPU(int *atom_arr, int no_of_threads) {
+void atomicKernel_CPU(int *atom_arr, int no_of_threads)
+{
     for (int i = no_of_threads; i < 2 * no_of_threads; i++) {
         for (int j = 0; j < LOOP_NUM; j++) {
             // Atomic addition
@@ -92,23 +94,20 @@ void atomicKernel_CPU(int *atom_arr, int no_of_threads) {
             int old, expected;
             do {
                 expected = atom_arr[2];
-        old = __sync_val_compare_and_swap(&atom_arr[2], expected,
+                old      = __sync_val_compare_and_swap(&atom_arr[2], expected, max(expected, i));
-                                          max(expected, i));
             } while (old != expected);
 
             // Atomic minimum
             do {
                 expected = atom_arr[3];
-        old = __sync_val_compare_and_swap(&atom_arr[3], expected,
+                old      = __sync_val_compare_and_swap(&atom_arr[3], expected, min(expected, i));
-                                          min(expected, i));
             } while (old != expected);
 
             // Atomic increment (modulo 17+1)
             int limit = 17;
             do {
                 expected = atom_arr[4];
-        old = __sync_val_compare_and_swap(
+                old      = __sync_val_compare_and_swap(&atom_arr[4], expected, (expected >= limit) ? 0 : expected + 1);
-            &atom_arr[4], expected, (expected >= limit) ? 0 : expected + 1);
             } while (old != expected);
 
             // Atomic decrement
@@ -116,8 +115,7 @@ void atomicKernel_CPU(int *atom_arr, int no_of_threads) {
             do {
                 expected = atom_arr[5];
                 old      = __sync_val_compare_and_swap(
-            &atom_arr[5], expected,
+                    &atom_arr[5], expected, ((expected == 0) || (expected > limit)) ? limit : expected - 1);
-            ((expected == 0) || (expected > limit)) ? limit : expected - 1);
             } while (old != expected);
 
             // Atomic compare-and-swap
@@ -145,7 +143,8 @@ void atomicKernel_CPU(int *atom_arr, int no_of_threads) {
 //! @param idata      input data as provided to device
 //! @param len        number of elements in reference / idata
 ////////////////////////////////////////////////////////////////////////////////
-int verify(int *testData, const int len) {
+int verify(int *testData, const int len)
+{
     int val = 0;
 
     for (int i = 0; i < len * LOOP_NUM; ++i) {
@@ -275,7 +274,8 @@ int verify(int *testData, const int len) {
     return true;
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     // set device
     cudaDeviceProp device_prop;
     int            dev_id = findCudaDevice(argc, (const char **)argv);
@@ -296,8 +296,7 @@ int main(int argc, char **argv) {
     }
 
     if (device_prop.major < 6) {
-    printf(
+        printf("%s: requires a minimum CUDA compute 6.0 capability, waiving "
-        "%s: requires a minimum CUDA compute 6.0 capability, waiving "
                "testing.\n",
                argv[0]);
         exit(EXIT_WAIVED);
@@ -312,12 +311,14 @@ int main(int argc, char **argv) {
     if (device_prop.pageableMemoryAccess) {
         printf("CAN access pageable memory\n");
         atom_arr = (int *)malloc(sizeof(int) * numData);
-  } else {
+    }
+    else {
         printf("CANNOT access pageable memory\n");
         checkCudaErrors(cudaMallocManaged(&atom_arr, sizeof(int) * numData));
     }
 
-  for (unsigned int i = 0; i < numData; i++) atom_arr[i] = 0;
+    for (unsigned int i = 0; i < numData; i++)
+        atom_arr[i] = 0;
 
     // To make the AND and XOR tests generate something other than 0...
     atom_arr[7] = atom_arr[9] = 0xff;
@@ -332,11 +333,11 @@ int main(int argc, char **argv) {
 
     if (device_prop.pageableMemoryAccess) {
         free(atom_arr);
-  } else {
+    }
+    else {
         cudaFree(atom_arr);
     }
 
-  printf("systemWideAtomics completed, returned %s \n",
+    printf("systemWideAtomics completed, returned %s \n", testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }

Samples/0_Introduction/template/template.cu

@@ -31,10 +31,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
+#include <math.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <math.h>
 
 // includes CUDA
 #include <cuda_runtime.h>
@@ -47,15 +47,15 @@
 // declaration, forward
 void runTest(int argc, char **argv);
 
-extern "C" void computeGold(float *reference, float *idata,
+extern "C" void computeGold(float *reference, float *idata, const unsigned int len);
-                            const unsigned int len);
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Simple test kernel for device functionality
 //! @param g_idata  input data in global memory
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void testKernel(float *g_idata, float *g_odata) {
+__global__ void testKernel(float *g_idata, float *g_odata)
+{
     // shared memory
     // the size is determined by the host application
     extern __shared__ float sdata[];
@@ -85,7 +85,8 @@ int main(int argc, char **argv) { runTest(argc, argv); }
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     bool bTestResult = true;
 
     printf("%s Starting...\n\n", argv[0]);
@@ -113,8 +114,7 @@ void runTest(int argc, char **argv) {
     float *d_idata;
     checkCudaErrors(cudaMalloc((void **)&d_idata, mem_size));
     // copy host memory to device
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpy(d_idata, h_idata, mem_size, cudaMemcpyHostToDevice));
-      cudaMemcpy(d_idata, h_idata, mem_size, cudaMemcpyHostToDevice));
 
     // allocate device memory for result
     float *d_odata;
@@ -133,8 +133,7 @@ void runTest(int argc, char **argv) {
     // allocate mem for the result on host side
     float *h_odata = (float *)malloc(mem_size);
     // copy result from device to host
-  checkCudaErrors(cudaMemcpy(h_odata, d_odata, sizeof(float) * num_threads,
+    checkCudaErrors(cudaMemcpy(h_odata, d_odata, sizeof(float) * num_threads, cudaMemcpyDeviceToHost));
-                             cudaMemcpyDeviceToHost));
 
     sdkStopTimer(&timer);
     printf("Processing time: %f (ms)\n", sdkGetTimerValue(&timer));
@@ -148,7 +147,8 @@ void runTest(int argc, char **argv) {
     if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
         // write file for regression test
         sdkWriteFile("./data/regression.dat", h_odata, num_threads, 0.0f, false);
-  } else {
+    }
+    else {
         // custom output handling when no regression test running
         // in this case check if the result is equivalent to the expected solution
         bTestResult = compareData(reference, h_odata, num_threads, 0.0f, 0.0f);

Samples/0_Introduction/template/template_cpu.cpp

@@ -26,8 +26,7 @@
  */
 
 // export C interface
-extern "C" void computeGold(float *reference, float *idata,
+extern "C" void computeGold(float *reference, float *idata, const unsigned int len);
-                            const unsigned int len);
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Compute reference data set
@@ -36,7 +35,8 @@ extern "C" void computeGold(float *reference, float *idata,
 //! @param idata      input data as provided to device
 //! @param len        number of elements in reference / idata
 ////////////////////////////////////////////////////////////////////////////////
-void computeGold(float *reference, float *idata, const unsigned int len) {
+void computeGold(float *reference, float *idata, const unsigned int len)
+{
     const float f_len = static_cast<float>(len);
 
     for (unsigned int i = 0; i < len; ++i) {

Samples/0_Introduction/vectorAdd/vectorAdd.cu

@@ -37,7 +37,6 @@
 
 // For the CUDA runtime routines (prefixed with "cuda_")
 #include <cuda_runtime.h>
-
 #include <helper_cuda.h>
 /**
  * CUDA Kernel Device code
@@ -45,8 +44,8 @@
  * Computes the vector addition of A and B into C. The 3 vectors have the same
  * number of elements numElements.
  */
-__global__ void vectorAdd(const float *A, const float *B, float *C,
+__global__ void vectorAdd(const float *A, const float *B, float *C, int numElements)
-                          int numElements) {
+{
     int i = blockDim.x * blockIdx.x + threadIdx.x;
 
     if (i < numElements) {
@@ -57,7 +56,8 @@ __global__ void vectorAdd(const float *A, const float *B, float *C,
 /**
  * Host main routine
  */
-int main(void) {
+int main(void)
+{
     // Error code to check return values for CUDA calls
     cudaError_t err = cudaSuccess;
 
@@ -92,8 +92,7 @@ int main(void) {
     err        = cudaMalloc((void **)&d_A, size);
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to allocate device vector A (error code %s)!\n",
+        fprintf(stderr, "Failed to allocate device vector A (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
@@ -102,8 +101,7 @@ int main(void) {
     err        = cudaMalloc((void **)&d_B, size);
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to allocate device vector B (error code %s)!\n",
+        fprintf(stderr, "Failed to allocate device vector B (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
@@ -112,8 +110,7 @@ int main(void) {
     err        = cudaMalloc((void **)&d_C, size);
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to allocate device vector C (error code %s)!\n",
+        fprintf(stderr, "Failed to allocate device vector C (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
@@ -124,32 +121,26 @@ int main(void) {
     err = cudaMemcpy(d_A, h_A, size, cudaMemcpyHostToDevice);
 
     if (err != cudaSuccess) {
-    fprintf(stderr,
+        fprintf(stderr, "Failed to copy vector A from host to device (error code %s)!\n", cudaGetErrorString(err));
-            "Failed to copy vector A from host to device (error code %s)!\n",
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
     err = cudaMemcpy(d_B, h_B, size, cudaMemcpyHostToDevice);
 
     if (err != cudaSuccess) {
-    fprintf(stderr,
+        fprintf(stderr, "Failed to copy vector B from host to device (error code %s)!\n", cudaGetErrorString(err));
-            "Failed to copy vector B from host to device (error code %s)!\n",
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
     // Launch the Vector Add CUDA Kernel
     int threadsPerBlock = 256;
     int blocksPerGrid   = (numElements + threadsPerBlock - 1) / threadsPerBlock;
-  printf("CUDA kernel launch with %d blocks of %d threads\n", blocksPerGrid,
+    printf("CUDA kernel launch with %d blocks of %d threads\n", blocksPerGrid, threadsPerBlock);
-         threadsPerBlock);
     vectorAdd<<<blocksPerGrid, threadsPerBlock>>>(d_A, d_B, d_C, numElements);
     err = cudaGetLastError();
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to launch vectorAdd kernel (error code %s)!\n",
+        fprintf(stderr, "Failed to launch vectorAdd kernel (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
@@ -159,9 +150,7 @@ int main(void) {
     err = cudaMemcpy(h_C, d_C, size, cudaMemcpyDeviceToHost);
 
     if (err != cudaSuccess) {
-    fprintf(stderr,
+        fprintf(stderr, "Failed to copy vector C from device to host (error code %s)!\n", cudaGetErrorString(err));
-            "Failed to copy vector C from device to host (error code %s)!\n",
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
@@ -179,24 +168,21 @@ int main(void) {
     err = cudaFree(d_A);
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to free device vector A (error code %s)!\n",
+        fprintf(stderr, "Failed to free device vector A (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
     err = cudaFree(d_B);
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to free device vector B (error code %s)!\n",
+        fprintf(stderr, "Failed to free device vector B (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 
     err = cudaFree(d_C);
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to free device vector C (error code %s)!\n",
+        fprintf(stderr, "Failed to free device vector C (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 

Samples/0_Introduction/vectorAddDrv/vectorAddDrv.cpp

@@ -34,11 +34,11 @@
  */
 
 // Includes
-#include <stdio.h>
-#include <string.h>
-#include <iostream>
 #include <cstring>
 #include <cuda.h>
+#include <iostream>
+#include <stdio.h>
+#include <string.h>
 
 // includes, project
 #include <helper_cuda_drvapi.h>
@@ -72,7 +72,8 @@ bool findModulePath(const char *, string &, char **, string &);
 #endif
 
 // Host code
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("Vector Addition (Driver API)\n");
     int    N = 50000, devID = 0;
     size_t size = N * sizeof(float);
@@ -91,7 +92,8 @@ int main(int argc, char **argv) {
 
     if (!findFatbinPath(FATBIN_FILE, module_path, argv, fatbin)) {
         exit(EXIT_FAILURE);
-  } else {
+    }
+    else {
         printf("> initCUDA loading module: <%s>\n", module_path.c_str());
     }
 
@@ -104,8 +106,7 @@ int main(int argc, char **argv) {
     checkCudaErrors(cuModuleLoadData(&cuModule, fatbin.str().c_str()));
 
     // Get function handle from module
-  checkCudaErrors(
+    checkCudaErrors(cuModuleGetFunction(&vecAdd_kernel, cuModule, "VecAdd_kernel"));
-      cuModuleGetFunction(&vecAdd_kernel, cuModule, "VecAdd_kernel"));
 
     // Allocate input vectors h_A and h_B in host memory
     h_A = (float *)malloc(size);
@@ -139,9 +140,9 @@ int main(int argc, char **argv) {
         void *args[] = {&d_A, &d_B, &d_C, &N};
 
         // Launch the CUDA kernel
-    checkCudaErrors(cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1,
+        checkCudaErrors(cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1, threadsPerBlock, 1, 1, 0, NULL, args, NULL));
-                                   threadsPerBlock, 1, 1, 0, NULL, args, NULL));
+    }
-  } else {
+    else {
         // This is the new CUDA 4.0 API for Kernel Parameter Passing and Kernel
         // Launch (advanced method)
         int   offset = 0;
@@ -160,9 +161,8 @@ int main(int argc, char **argv) {
         int blocksPerGrid   = (N + threadsPerBlock - 1) / threadsPerBlock;
 
         // Launch the CUDA kernel
-    checkCudaErrors(cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1,
+        checkCudaErrors(
-                                   threadsPerBlock, 1, 1, 0, NULL, NULL,
+            cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1, threadsPerBlock, 1, 1, 0, NULL, NULL, argBuffer));
-                                   argBuffer));
     }
 
 #ifdef _DEBUG
@@ -190,7 +190,8 @@ int main(int argc, char **argv) {
     exit((i == N) ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-int CleanupNoFailure() {
+int CleanupNoFailure()
+{
     // Free device memory
     checkCudaErrors(cuMemFree(d_A));
     checkCudaErrors(cuMemFree(d_B));
@@ -214,7 +215,8 @@ int CleanupNoFailure() {
     return EXIT_SUCCESS;
 }
 // Allocates an array with random float entries.
-void RandomInit(float *data, int n) {
+void RandomInit(float *data, int n)
+{
     for (int i = 0; i < n; ++i) {
         data[i] = rand() / (float)RAND_MAX;
     }

Samples/0_Introduction/vectorAddDrv/vectorAdd_kernel.cu

@@ -33,9 +33,10 @@
  */
 
 // Device code
-extern "C" __global__ void VecAdd_kernel(const float *A, const float *B,
+extern "C" __global__ void VecAdd_kernel(const float *A, const float *B, float *C, int N)
-                                         float *C, int N) {
+{
     int i = blockDim.x * blockIdx.x + threadIdx.x;
 
-  if (i < N) C[i] = A[i] + B[i];
+    if (i < N)
+        C[i] = A[i] + B[i];
 }

Samples/0_Introduction/vectorAddMMAP/README.md

@@ -30,4 +30,3 @@ cuMemcpyDtoH, cuDeviceCanAccessPeer, cuModuleGetFunction, cuMemSetAccess, cuMemR
 Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)
-

Samples/0_Introduction/vectorAddMMAP/multidevicealloc_memmap.cpp

@@ -29,10 +29,13 @@
 
 static size_t round_up(size_t x, size_t y) { return ((x + y - 1) / y) * y; }
 
-CUresult simpleMallocMultiDeviceMmap(
+CUresult simpleMallocMultiDeviceMmap(CUdeviceptr                 *dptr,
-    CUdeviceptr *dptr, size_t *allocationSize, size_t size,
+                                     size_t                      *allocationSize,
+                                     size_t                       size,
                                      const std::vector<CUdevice> &residentDevices,
-    const std::vector<CUdevice> &mappingDevices, size_t align) {
+                                     const std::vector<CUdevice> &mappingDevices,
+                                     size_t                       align)
+{
     CUresult status          = CUDA_SUCCESS;
     size_t   min_granularity = 0;
     size_t   stripeSize;
@@ -53,8 +56,7 @@ CUresult simpleMallocMultiDeviceMmap(
 
         // get the minnimum granularity for residentDevices[idx]
         prop.location.id = residentDevices[idx];
-    status = cuMemGetAllocationGranularity(&granularity, &prop,
+        status           = cuMemGetAllocationGranularity(&granularity, &prop, CU_MEM_ALLOC_GRANULARITY_MINIMUM);
-                                           CU_MEM_ALLOC_GRANULARITY_MINIMUM);
         if (status != CUDA_SUCCESS) {
             goto done;
         }
@@ -70,8 +72,7 @@ CUresult simpleMallocMultiDeviceMmap(
 
         // get the minnimum granularity for mappingDevices[idx]
         prop.location.id = mappingDevices[idx];
-    status = cuMemGetAllocationGranularity(&granularity, &prop,
+        status           = cuMemGetAllocationGranularity(&granularity, &prop, CU_MEM_ALLOC_GRANULARITY_MINIMUM);
-                                           CU_MEM_ALLOC_GRANULARITY_MINIMUM);
         if (status != CUDA_SUCCESS) {
             goto done;
         }
@@ -121,8 +122,7 @@ CUresult simpleMallocMultiDeviceMmap(
         // Since we do not need to make any other mappings of this memory or export
         // it, we no longer need and can release the allocationHandle. The
         // allocation will be kept live until it is unmapped.
-    status = cuMemMap(*dptr + (stripeSize * idx), stripeSize, 0,
+        status = cuMemMap(*dptr + (stripeSize * idx), stripeSize, 0, allocationHandle, 0);
-                      allocationHandle, 0);
 
         // the handle needs to be released even if the mapping failed.
         status2 = cuMemRelease(allocationHandle);
@@ -157,8 +157,7 @@ CUresult simpleMallocMultiDeviceMmap(
         }
 
         // Apply the access descriptors to the whole VA range.
-    status = cuMemSetAccess(*dptr, size, &accessDescriptors[0],
+        status = cuMemSetAccess(*dptr, size, &accessDescriptors[0], accessDescriptors.size());
-                            accessDescriptors.size());
         if (status != CUDA_SUCCESS) {
             goto done;
         }
@@ -174,7 +173,8 @@ done:
     return status;
 }
 
-CUresult simpleFreeMultiDeviceMmap(CUdeviceptr dptr, size_t size) {
+CUresult simpleFreeMultiDeviceMmap(CUdeviceptr dptr, size_t size)
+{
     CUresult status = CUDA_SUCCESS;
 
     // Unmap the mapped virtual memory region

Samples/0_Introduction/vectorAddMMAP/multidevicealloc_memmap.hpp

@@ -63,10 +63,12 @@
 //! handle
 //!   is not needed after its mappings are set up.
 ////////////////////////////////////////////////////////////////////////////
-CUresult simpleMallocMultiDeviceMmap(
+CUresult simpleMallocMultiDeviceMmap(CUdeviceptr                 *dptr,
-    CUdeviceptr *dptr, size_t *allocationSize, size_t size,
+                                     size_t                      *allocationSize,
+                                     size_t                       size,
                                      const std::vector<CUdevice> &residentDevices,
-    const std::vector<CUdevice> &mappingDevices, size_t align = 0);
+                                     const std::vector<CUdevice> &mappingDevices,
+                                     size_t                       align = 0);
 
 ////////////////////////////////////////////////////////////////////////////
 //! Frees resources allocated by simpleMallocMultiDeviceMmap

Samples/0_Introduction/vectorAddMMAP/vectorAddMMAP.cpp

@@ -36,11 +36,11 @@
  */
 
 // Includes
+#include <cstring>
 #include <cuda.h>
+#include <iostream>
 #include <stdio.h>
 #include <string.h>
-#include <cstring>
-#include <iostream>
 
 // includes, project
 #include <helper_cuda_drvapi.h>
@@ -76,7 +76,8 @@ void RandomInit(float *, int);
 #endif
 
 // collect all of the devices whose memory can be mapped from cuDevice.
-vector<CUdevice> getBackingDevices(CUdevice cuDevice) {
+vector<CUdevice> getBackingDevices(CUdevice cuDevice)
+{
     int num_devices;
 
     checkCudaErrors(cuDeviceGetCount(&num_devices));
@@ -100,9 +101,8 @@ vector<CUdevice> getBackingDevices(CUdevice cuDevice) {
 
         // The device needs to support virtual address management for the required
         // apis to work
-    checkCudaErrors(cuDeviceGetAttribute(
+        checkCudaErrors(
-        &attributeVal, CU_DEVICE_ATTRIBUTE_VIRTUAL_ADDRESS_MANAGEMENT_SUPPORTED,
+            cuDeviceGetAttribute(&attributeVal, CU_DEVICE_ATTRIBUTE_VIRTUAL_ADDRESS_MANAGEMENT_SUPPORTED, cuDevice));
-        cuDevice));
         if (attributeVal == 0) {
             continue;
         }
@@ -113,7 +113,8 @@ vector<CUdevice> getBackingDevices(CUdevice cuDevice) {
 }
 
 // Host code
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("Vector Addition (Driver API)\n");
     int    N            = 50000;
     size_t size         = N * sizeof(float);
@@ -125,11 +126,9 @@ int main(int argc, char **argv) {
     cuDevice = findCudaDeviceDRV(argc, (const char **)argv);
 
     // Check that the selected device supports virtual address management
-  checkCudaErrors(cuDeviceGetAttribute(
+    checkCudaErrors(
-      &attributeVal, CU_DEVICE_ATTRIBUTE_VIRTUAL_ADDRESS_MANAGEMENT_SUPPORTED,
+        cuDeviceGetAttribute(&attributeVal, CU_DEVICE_ATTRIBUTE_VIRTUAL_ADDRESS_MANAGEMENT_SUPPORTED, cuDevice));
-      cuDevice));
+    printf("Device %d VIRTUAL ADDRESS MANAGEMENT SUPPORTED = %d.\n", cuDevice, attributeVal);
-  printf("Device %d VIRTUAL ADDRESS MANAGEMENT SUPPORTED = %d.\n", cuDevice,
-         attributeVal);
     if (attributeVal == 0) {
         printf("Device %d doesn't support VIRTUAL ADDRESS MANAGEMENT.\n", cuDevice);
         exit(EXIT_WAIVED);
@@ -152,17 +151,14 @@ int main(int argc, char **argv) {
 
     std::ostringstream fatbin;
 
-    if (!findFatbinPath(FATBIN_FILE, module_path, argv, fatbin))
+    if (!findFatbinPath(FATBIN_FILE, module_path, argv, fatbin)) {
-    {
         exit(EXIT_FAILURE);
     }
-    else
+    else {
-    {
         printf("> initCUDA loading module: <%s>\n", module_path.c_str());
     }
 
-    if (!fatbin.str().size())
+    if (!fatbin.str().size()) {
-    {
         printf("fatbin file empty. exiting..\n");
         exit(EXIT_FAILURE);
     }
@@ -207,10 +203,7 @@ int main(int argc, char **argv) {
     void *args[] = {&d_A, &d_B, &d_C, &N};
 
     // Launch the CUDA kernel
-    checkCudaErrors(cuLaunchKernel(vecAdd_kernel,  blocksPerGrid, 1, 1,
+    checkCudaErrors(cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1, threadsPerBlock, 1, 1, 0, NULL, args, NULL));
-                               threadsPerBlock, 1, 1,
-                               0,
-                               NULL, args, NULL));
 
     // Copy result from device memory to host memory
     // h_C contains the result in host memory
@@ -219,12 +212,10 @@ int main(int argc, char **argv) {
     // Verify result
     int i;
 
-    for (i = 0; i < N; ++i)
+    for (i = 0; i < N; ++i) {
-    {
         float sum = h_A[i] + h_B[i];
 
-        if (fabs(h_C[i] - sum) > 1e-7f)
+        if (fabs(h_C[i] - sum) > 1e-7f) {
-        {
             break;
         }
     }
@@ -243,18 +234,15 @@ int CleanupNoFailure()
     checkCudaErrors(simpleFreeMultiDeviceMmap(d_C, allocationSize));
 
     // Free host memory
-    if (h_A)
+    if (h_A) {
-    {
         free(h_A);
     }
 
-    if (h_B)
+    if (h_B) {
-    {
         free(h_B);
     }
 
-    if (h_C)
+    if (h_C) {
-    {
         free(h_C);
     }
 
@@ -265,8 +253,7 @@ int CleanupNoFailure()
 // Allocates an array with random float entries.
 void RandomInit(float *data, int n)
 {
-    for (int i = 0; i < n; ++i)
+    for (int i = 0; i < n; ++i) {
-    {
         data[i] = rand() / (float)RAND_MAX;
     }
 }

Samples/0_Introduction/vectorAddMMAP/vectorAdd_kernel.cu

@@ -34,9 +34,10 @@
  */
 
 // Device code
-extern "C" __global__ void VecAdd_kernel(const float *A, const float *B,
+extern "C" __global__ void VecAdd_kernel(const float *A, const float *B, float *C, int N)
-                                         float *C, int N) {
+{
     int i = blockDim.x * blockIdx.x + threadIdx.x;
 
-  if (i < N) C[i] = A[i] + B[i];
+    if (i < N)
+        C[i] = A[i] + B[i];
 }

Samples/0_Introduction/vectorAdd_nvrtc/vectorAdd.cpp

@@ -33,8 +33,8 @@
  * of the programming guide with some additions like error checking.
  */
 
-#include <stdio.h>
 #include <cmath>
+#include <stdio.h>
 
 // For the CUDA runtime routines (prefixed with "cuda_")
 #include <cuda.h>
@@ -42,13 +42,13 @@
 
 // helper functions and utilities to work with CUDA
 #include <helper_functions.h>
-
 #include <nvrtc_helper.h>
 
 /**
  * Host main routine
  */
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     char  *cubin, *kernel_file;
     size_t cubinSize;
     kernel_file = sdkFindFilePath("vectorAdd_kernel.cu", argv[0]);
@@ -105,19 +105,23 @@ int main(int argc, char **argv) {
     // Launch the Vector Add CUDA Kernel
     int threadsPerBlock = 256;
     int blocksPerGrid   = (numElements + threadsPerBlock - 1) / threadsPerBlock;
-  printf("CUDA kernel launch with %d blocks of %d threads\n", blocksPerGrid,
+    printf("CUDA kernel launch with %d blocks of %d threads\n", blocksPerGrid, threadsPerBlock);
-         threadsPerBlock);
     dim3 cudaBlockSize(threadsPerBlock, 1, 1);
     dim3 cudaGridSize(blocksPerGrid, 1, 1);
 
-  void *arr[] = {reinterpret_cast<void *>(&d_A), reinterpret_cast<void *>(&d_B),
+    void *arr[] = {reinterpret_cast<void *>(&d_A),
+                   reinterpret_cast<void *>(&d_B),
                    reinterpret_cast<void *>(&d_C),
                    reinterpret_cast<void *>(&numElements)};
-  checkCudaErrors(cuLaunchKernel(kernel_addr, cudaGridSize.x, cudaGridSize.y,
+    checkCudaErrors(cuLaunchKernel(kernel_addr,
+                                   cudaGridSize.x,
+                                   cudaGridSize.y,
                                    cudaGridSize.z, /* grid dim */
-                                 cudaBlockSize.x, cudaBlockSize.y,
+                                   cudaBlockSize.x,
+                                   cudaBlockSize.y,
                                    cudaBlockSize.z, /* block dim */
-                                 0, 0,            /* shared mem, stream */
+                                   0,
+                                   0,       /* shared mem, stream */
                                    &arr[0], /* arguments */
                                    0));
     checkCudaErrors(cuCtxSynchronize());

Samples/0_Introduction/vectorAdd_nvrtc/vectorAdd_kernel.cu

@@ -32,8 +32,8 @@
  * number of elements numElements.
  */
 
-extern "C" __global__ void vectorAdd(const float *A, const float *B, float *C,
+extern "C" __global__ void vectorAdd(const float *A, const float *B, float *C, int numElements)
-                                     int numElements) {
+{
     int i = blockDim.x * blockIdx.x + threadIdx.x;
 
     if (i < numElements) {

Samples/1_Utilities/bandwidthTest/bandwidthTest.cu

@@ -39,12 +39,10 @@
 #include <cuda_runtime.h>
 
 // includes
+#include <cassert>
+#include <cuda.h>
 #include <helper_cuda.h>      // helper functions for CUDA error checking and initialization
 #include <helper_functions.h> // helper for shared functions common to CUDA Samples
-
-#include <cuda.h>
-
-#include <cassert>
 #include <iostream>
 #include <memory>
 
@@ -83,8 +81,7 @@ enum memcpyKind { DEVICE_TO_HOST, HOST_TO_DEVICE, DEVICE_TO_DEVICE };
 enum printMode { USER_READABLE, CSV };
 enum memoryMode { PINNED, PAGEABLE };
 
-const char *sMemoryCopyKind[] = {"Device to Host", "Host to Device",
+const char *sMemoryCopyKind[] = {"Device to Host", "Host to Device", "Device to Device", NULL};
-                                 "Device to Device", NULL};
 
 const char *sMemoryMode[] = {"PINNED", "PAGEABLE", NULL};
 
@@ -97,36 +94,62 @@ char **pArgv = NULL;
 ////////////////////////////////////////////////////////////////////////////////
 // declaration, forward
 int   runTest(const int argc, const char **argv);
-void testBandwidth(unsigned int start, unsigned int end, unsigned int increment,
+void  testBandwidth(unsigned int start,
-                   testMode mode, memcpyKind kind, printMode printmode,
+                    unsigned int end,
-                   memoryMode memMode, int startDevice, int endDevice, bool wc);
+                    unsigned int increment,
-void testBandwidthQuick(unsigned int size, memcpyKind kind, printMode printmode,
+                    testMode     mode,
-                        memoryMode memMode, int startDevice, int endDevice,
+                    memcpyKind   kind,
+                    printMode    printmode,
+                    memoryMode   memMode,
+                    int          startDevice,
+                    int          endDevice,
                     bool         wc);
-void testBandwidthRange(unsigned int start, unsigned int end,
+void  testBandwidthQuick(unsigned int size,
-                        unsigned int increment, memcpyKind kind,
+                         memcpyKind   kind,
-                        printMode printmode, memoryMode memMode,
+                         printMode    printmode,
-                        int startDevice, int endDevice, bool wc);
+                         memoryMode   memMode,
-void testBandwidthShmoo(memcpyKind kind, printMode printmode,
+                         int          startDevice,
-                        memoryMode memMode, int startDevice, int endDevice,
+                         int          endDevice,
                          bool         wc);
-float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode,
+void  testBandwidthRange(unsigned int start,
+                         unsigned int end,
+                         unsigned int increment,
+                         memcpyKind   kind,
+                         printMode    printmode,
+                         memoryMode   memMode,
+                         int          startDevice,
+                         int          endDevice,
                          bool         wc);
-float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode,
+void  testBandwidthShmoo(memcpyKind kind,
+                         printMode  printmode,
+                         memoryMode memMode,
+                         int        startDevice,
+                         int        endDevice,
                          bool       wc);
+float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode, bool wc);
+float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode, bool wc);
 float testDeviceToDeviceTransfer(unsigned int memSize);
-void printResultsReadable(unsigned int *memSizes, double *bandwidths,
+void  printResultsReadable(unsigned int *memSizes,
-                          unsigned int count, memcpyKind kind,
+                           double       *bandwidths,
-                          memoryMode memMode, int iNumDevs, bool wc);
+                           unsigned int  count,
-void printResultsCSV(unsigned int *memSizes, double *bandwidths,
+                           memcpyKind    kind,
-                     unsigned int count, memcpyKind kind, memoryMode memMode,
+                           memoryMode    memMode,
-                     int iNumDevs, bool wc);
+                           int           iNumDevs,
+                           bool          wc);
+void  printResultsCSV(unsigned int *memSizes,
+                      double       *bandwidths,
+                      unsigned int  count,
+                      memcpyKind    kind,
+                      memoryMode    memMode,
+                      int           iNumDevs,
+                      bool          wc);
 void  printHelp(void);
 
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     pArgc = &argc;
     pArgv = argv;
 
@@ -144,8 +167,7 @@ int main(int argc, char **argv) {
     // finish
     printf("%s\n", (iRetVal == 0) ? "Result = PASS" : "Result = FAIL");
 
-  printf(
+    printf("\nNOTE: The CUDA Samples are not meant for performance measurements. "
-      "\nNOTE: The CUDA Samples are not meant for performance measurements. "
            "Results may vary when GPU Boost is enabled.\n");
 
     free(flush_buf);
@@ -156,7 +178,8 @@ int main(int argc, char **argv) {
 ///////////////////////////////////////////////////////////////////////////////
 // Parse args, run the appropriate tests
 ///////////////////////////////////////////////////////////////////////////////
-int runTest(const int argc, const char **argv) {
+int runTest(const int argc, const char **argv)
+{
     int        start       = DEFAULT_SIZE;
     int        end         = DEFAULT_SIZE;
     int        startDevice = 0;
@@ -186,14 +209,17 @@ int runTest(const int argc, const char **argv) {
     if (getCmdLineArgumentString(argc, argv, "memory", &memModeStr)) {
         if (strcmp(memModeStr, "pageable") == 0) {
             memMode = PAGEABLE;
-    } else if (strcmp(memModeStr, "pinned") == 0) {
+        }
+        else if (strcmp(memModeStr, "pinned") == 0) {
             memMode = PINNED;
-    } else {
+        }
+        else {
             printf("Invalid memory mode - valid modes are pageable or pinned\n");
             printf("See --help for more information\n");
             return -1000;
         }
-  } else {
+    }
+    else {
         // default - pinned memory
         memMode = PINNED;
     }
@@ -203,8 +229,7 @@ int runTest(const int argc, const char **argv) {
         cudaError_t error_id = cudaGetDeviceCount(&deviceCount);
 
         if (error_id != cudaSuccess) {
-      printf("cudaGetDeviceCount returned %d\n-> %s\n", (int)error_id,
+            printf("cudaGetDeviceCount returned %d\n-> %s\n", (int)error_id, cudaGetErrorString(error_id));
-             cudaGetErrorString(error_id));
             exit(EXIT_FAILURE);
         }
 
@@ -214,19 +239,19 @@ int runTest(const int argc, const char **argv) {
         }
 
         if (strcmp(device, "all") == 0) {
-      printf(
+            printf("\n!!!!!Cumulative Bandwidth to be computed from all the devices "
-          "\n!!!!!Cumulative Bandwidth to be computed from all the devices "
                    "!!!!!!\n\n");
             startDevice = 0;
             endDevice   = deviceCount - 1;
-    } else {
+        }
+        else {
             startDevice = endDevice = atoi(device);
 
             if (startDevice >= deviceCount || startDevice < 0) {
-        printf(
+                printf("\n!!!!!Invalid GPU number %d given hence default gpu %d will be "
-            "\n!!!!!Invalid GPU number %d given hence default gpu %d will be "
                        "used !!!!!\n",
-            startDevice, 0);
+                       startDevice,
+                       0);
                 startDevice = endDevice = 0;
             }
         }
@@ -234,8 +259,7 @@ int runTest(const int argc, const char **argv) {
 
     printf("Running on...\n\n");
 
-  for (int currentDevice = startDevice; currentDevice <= endDevice;
+    for (int currentDevice = startDevice; currentDevice <= endDevice; currentDevice++) {
-       currentDevice++) {
         cudaDeviceProp deviceProp;
         cudaError_t    error_id = cudaGetDeviceProperties(&deviceProp, currentDevice);
 
@@ -250,9 +274,9 @@ int runTest(const int argc, const char **argv) {
 
                 exit(EXIT_FAILURE);
             }
-    } else {
+        }
-      printf("cudaGetDeviceProperties returned %d\n-> %s\n", (int)error_id,
+        else {
-             cudaGetErrorString(error_id));
+            printf("cudaGetDeviceProperties returned %d\n-> %s\n", (int)error_id, cudaGetErrorString(error_id));
             checkCudaErrors(cudaSetDevice(currentDevice));
 
             exit(EXIT_FAILURE);
@@ -264,18 +288,22 @@ int runTest(const int argc, const char **argv) {
         if (strcmp(modeStr, "quick") == 0) {
             printf(" Quick Mode\n\n");
             mode = QUICK_MODE;
-    } else if (strcmp(modeStr, "shmoo") == 0) {
+        }
+        else if (strcmp(modeStr, "shmoo") == 0) {
             printf(" Shmoo Mode\n\n");
             mode = SHMOO_MODE;
-    } else if (strcmp(modeStr, "range") == 0) {
+        }
+        else if (strcmp(modeStr, "range") == 0) {
             printf(" Range Mode\n\n");
             mode = RANGE_MODE;
-    } else {
+        }
+        else {
             printf("Invalid mode - valid modes are quick, range, or shmoo\n");
             printf("See --help for more information\n");
             return -3000;
         }
-  } else {
+    }
+    else {
         // default mode - quick
         printf(" Quick Mode\n\n");
         mode = QUICK_MODE;
@@ -320,7 +348,8 @@ int runTest(const int argc, const char **argv) {
                 printf("Illegal argument - start must be greater than zero\n");
                 return -4000;
             }
-    } else {
+        }
+        else {
             printf("Must specify a starting size in range mode\n");
             printf("See --help for more information\n");
             return -5000;
@@ -338,7 +367,8 @@ int runTest(const int argc, const char **argv) {
                 printf("Illegal argument - start is greater than end\n");
                 return -7000;
             }
-    } else {
+        }
+        else {
             printf("Must specify an end size in range mode.\n");
             printf("See --help for more information\n");
             return -8000;
@@ -351,7 +381,8 @@ int runTest(const int argc, const char **argv) {
                 printf("Illegal argument - increment must be greater than zero\n");
                 return -9000;
             }
-    } else {
+        }
+        else {
             printf("Must specify an increment in user mode\n");
             printf("See --help for more information\n");
             return -10000;
@@ -359,21 +390,42 @@ int runTest(const int argc, const char **argv) {
     }
 
     if (htod) {
-    testBandwidth((unsigned int)start, (unsigned int)end,
+        testBandwidth((unsigned int)start,
-                  (unsigned int)increment, mode, HOST_TO_DEVICE, printmode,
+                      (unsigned int)end,
-                  memMode, startDevice, endDevice, wc);
+                      (unsigned int)increment,
+                      mode,
+                      HOST_TO_DEVICE,
+                      printmode,
+                      memMode,
+                      startDevice,
+                      endDevice,
+                      wc);
     }
 
     if (dtoh) {
-    testBandwidth((unsigned int)start, (unsigned int)end,
+        testBandwidth((unsigned int)start,
-                  (unsigned int)increment, mode, DEVICE_TO_HOST, printmode,
+                      (unsigned int)end,
-                  memMode, startDevice, endDevice, wc);
+                      (unsigned int)increment,
+                      mode,
+                      DEVICE_TO_HOST,
+                      printmode,
+                      memMode,
+                      startDevice,
+                      endDevice,
+                      wc);
     }
 
     if (dtod) {
-    testBandwidth((unsigned int)start, (unsigned int)end,
+        testBandwidth((unsigned int)start,
-                  (unsigned int)increment, mode, DEVICE_TO_DEVICE, printmode,
+                      (unsigned int)end,
-                  memMode, startDevice, endDevice, wc);
+                      (unsigned int)increment,
+                      mode,
+                      DEVICE_TO_DEVICE,
+                      printmode,
+                      memMode,
+                      startDevice,
+                      endDevice,
+                      wc);
     }
 
     // Ensure that we reset all CUDA Devices in question
@@ -387,19 +439,24 @@ int runTest(const int argc, const char **argv) {
 ///////////////////////////////////////////////////////////////////////////////
 //  Run a bandwidth test
 ///////////////////////////////////////////////////////////////////////////////
-void testBandwidth(unsigned int start, unsigned int end, unsigned int increment,
+void testBandwidth(unsigned int start,
-                   testMode mode, memcpyKind kind, printMode printmode,
+                   unsigned int end,
-                   memoryMode memMode, int startDevice, int endDevice,
+                   unsigned int increment,
-                   bool wc) {
+                   testMode     mode,
+                   memcpyKind   kind,
+                   printMode    printmode,
+                   memoryMode   memMode,
+                   int          startDevice,
+                   int          endDevice,
+                   bool         wc)
+{
     switch (mode) {
     case QUICK_MODE:
-      testBandwidthQuick(DEFAULT_SIZE, kind, printmode, memMode, startDevice,
+        testBandwidthQuick(DEFAULT_SIZE, kind, printmode, memMode, startDevice, endDevice, wc);
-                         endDevice, wc);
         break;
 
     case RANGE_MODE:
-      testBandwidthRange(start, end, increment, kind, printmode, memMode,
+        testBandwidthRange(start, end, increment, kind, printmode, memMode, startDevice, endDevice, wc);
-                         startDevice, endDevice, wc);
         break;
 
     case SHMOO_MODE:
@@ -414,20 +471,30 @@ void testBandwidth(unsigned int start, unsigned int end, unsigned int increment,
 //////////////////////////////////////////////////////////////////////
 //  Run a quick mode bandwidth test
 //////////////////////////////////////////////////////////////////////
-void testBandwidthQuick(unsigned int size, memcpyKind kind, printMode printmode,
+void testBandwidthQuick(unsigned int size,
-                        memoryMode memMode, int startDevice, int endDevice,
+                        memcpyKind   kind,
-                        bool wc) {
+                        printMode    printmode,
-  testBandwidthRange(size, size, DEFAULT_INCREMENT, kind, printmode, memMode,
+                        memoryMode   memMode,
-                     startDevice, endDevice, wc);
+                        int          startDevice,
+                        int          endDevice,
+                        bool         wc)
+{
+    testBandwidthRange(size, size, DEFAULT_INCREMENT, kind, printmode, memMode, startDevice, endDevice, wc);
 }
 
 ///////////////////////////////////////////////////////////////////////
 //  Run a range mode bandwidth test
 //////////////////////////////////////////////////////////////////////
-void testBandwidthRange(unsigned int start, unsigned int end,
+void testBandwidthRange(unsigned int start,
-                        unsigned int increment, memcpyKind kind,
+                        unsigned int end,
-                        printMode printmode, memoryMode memMode,
+                        unsigned int increment,
-                        int startDevice, int endDevice, bool wc) {
+                        memcpyKind   kind,
+                        printMode    printmode,
+                        memoryMode   memMode,
+                        int          startDevice,
+                        int          endDevice,
+                        bool         wc)
+{
     // count the number of copies we're going to run
     unsigned int count = 1 + ((end - start) / increment);
 
@@ -441,8 +508,7 @@ void testBandwidthRange(unsigned int start, unsigned int end,
     }
 
     // Use the device asked by the user
-  for (int currentDevice = startDevice; currentDevice <= endDevice;
+    for (int currentDevice = startDevice; currentDevice <= endDevice; currentDevice++) {
-       currentDevice++) {
         cudaSetDevice(currentDevice);
 
         // run each of the copies
@@ -467,11 +533,10 @@ void testBandwidthRange(unsigned int start, unsigned int end,
 
     // print results
     if (printmode == CSV) {
-    printResultsCSV(memSizes, bandwidths, count, kind, memMode,
+        printResultsCSV(memSizes, bandwidths, count, kind, memMode, (1 + endDevice - startDevice), wc);
-                    (1 + endDevice - startDevice), wc);
+    }
-  } else {
+    else {
-    printResultsReadable(memSizes, bandwidths, count, kind, memMode,
+        printResultsReadable(memSizes, bandwidths, count, kind, memMode, (1 + endDevice - startDevice), wc);
-                         (1 + endDevice - startDevice), wc);
     }
 
     // clean up
@@ -482,18 +547,21 @@ void testBandwidthRange(unsigned int start, unsigned int end,
 //////////////////////////////////////////////////////////////////////////////
 // Intense shmoo mode - covers a large range of values with varying increments
 //////////////////////////////////////////////////////////////////////////////
-void testBandwidthShmoo(memcpyKind kind, printMode printmode,
+void testBandwidthShmoo(memcpyKind kind,
-                        memoryMode memMode, int startDevice, int endDevice,
+                        printMode  printmode,
-                        bool wc) {
+                        memoryMode memMode,
+                        int        startDevice,
+                        int        endDevice,
+                        bool       wc)
+{
     // count the number of copies to make
-  unsigned int count =
+    unsigned int count = 1 + (SHMOO_LIMIT_20KB / SHMOO_INCREMENT_1KB)
-      1 + (SHMOO_LIMIT_20KB / SHMOO_INCREMENT_1KB) +
+                       + ((SHMOO_LIMIT_50KB - SHMOO_LIMIT_20KB) / SHMOO_INCREMENT_2KB)
-      ((SHMOO_LIMIT_50KB - SHMOO_LIMIT_20KB) / SHMOO_INCREMENT_2KB) +
+                       + ((SHMOO_LIMIT_100KB - SHMOO_LIMIT_50KB) / SHMOO_INCREMENT_10KB)
-      ((SHMOO_LIMIT_100KB - SHMOO_LIMIT_50KB) / SHMOO_INCREMENT_10KB) +
+                       + ((SHMOO_LIMIT_1MB - SHMOO_LIMIT_100KB) / SHMOO_INCREMENT_100KB)
-      ((SHMOO_LIMIT_1MB - SHMOO_LIMIT_100KB) / SHMOO_INCREMENT_100KB) +
+                       + ((SHMOO_LIMIT_16MB - SHMOO_LIMIT_1MB) / SHMOO_INCREMENT_1MB)
-      ((SHMOO_LIMIT_16MB - SHMOO_LIMIT_1MB) / SHMOO_INCREMENT_1MB) +
+                       + ((SHMOO_LIMIT_32MB - SHMOO_LIMIT_16MB) / SHMOO_INCREMENT_2MB)
-      ((SHMOO_LIMIT_32MB - SHMOO_LIMIT_16MB) / SHMOO_INCREMENT_2MB) +
+                       + ((SHMOO_MEMSIZE_MAX - SHMOO_LIMIT_32MB) / SHMOO_INCREMENT_4MB);
-      ((SHMOO_MEMSIZE_MAX - SHMOO_LIMIT_32MB) / SHMOO_INCREMENT_4MB);
 
     unsigned int *memSizes   = (unsigned int *)malloc(count * sizeof(unsigned int));
     double       *bandwidths = (double *)malloc(count * sizeof(double));
@@ -505,8 +573,7 @@ void testBandwidthShmoo(memcpyKind kind, printMode printmode,
     }
 
     // Use the device asked by the user
-  for (int currentDevice = startDevice; currentDevice <= endDevice;
+    for (int currentDevice = startDevice; currentDevice <= endDevice; currentDevice++) {
-       currentDevice++) {
         cudaSetDevice(currentDevice);
         // Run the shmoo
         int          iteration = 0;
@@ -515,17 +582,23 @@ void testBandwidthShmoo(memcpyKind kind, printMode printmode,
         while (memSize <= SHMOO_MEMSIZE_MAX) {
             if (memSize < SHMOO_LIMIT_20KB) {
                 memSize += SHMOO_INCREMENT_1KB;
-      } else if (memSize < SHMOO_LIMIT_50KB) {
+            }
+            else if (memSize < SHMOO_LIMIT_50KB) {
                 memSize += SHMOO_INCREMENT_2KB;
-      } else if (memSize < SHMOO_LIMIT_100KB) {
+            }
+            else if (memSize < SHMOO_LIMIT_100KB) {
                 memSize += SHMOO_INCREMENT_10KB;
-      } else if (memSize < SHMOO_LIMIT_1MB) {
+            }
+            else if (memSize < SHMOO_LIMIT_1MB) {
                 memSize += SHMOO_INCREMENT_100KB;
-      } else if (memSize < SHMOO_LIMIT_16MB) {
+            }
+            else if (memSize < SHMOO_LIMIT_16MB) {
                 memSize += SHMOO_INCREMENT_1MB;
-      } else if (memSize < SHMOO_LIMIT_32MB) {
+            }
+            else if (memSize < SHMOO_LIMIT_32MB) {
                 memSize += SHMOO_INCREMENT_2MB;
-      } else {
+            }
+            else {
                 memSize += SHMOO_INCREMENT_4MB;
             }
 
@@ -533,18 +606,15 @@ void testBandwidthShmoo(memcpyKind kind, printMode printmode,
 
             switch (kind) {
             case DEVICE_TO_HOST:
-          bandwidths[iteration] +=
+                bandwidths[iteration] += testDeviceToHostTransfer(memSizes[iteration], memMode, wc);
-              testDeviceToHostTransfer(memSizes[iteration], memMode, wc);
                 break;
 
             case HOST_TO_DEVICE:
-          bandwidths[iteration] +=
+                bandwidths[iteration] += testHostToDeviceTransfer(memSizes[iteration], memMode, wc);
-              testHostToDeviceTransfer(memSizes[iteration], memMode, wc);
                 break;
 
             case DEVICE_TO_DEVICE:
-          bandwidths[iteration] +=
+                bandwidths[iteration] += testDeviceToDeviceTransfer(memSizes[iteration]);
-              testDeviceToDeviceTransfer(memSizes[iteration]);
                 break;
             }
 
@@ -558,11 +628,10 @@ void testBandwidthShmoo(memcpyKind kind, printMode printmode,
     printf("\n");
 
     if (CSV == printmode) {
-    printResultsCSV(memSizes, bandwidths, count, kind, memMode,
+        printResultsCSV(memSizes, bandwidths, count, kind, memMode, (1 + endDevice - startDevice), wc);
-                    (1 + endDevice - startDevice), wc);
+    }
-  } else {
+    else {
-    printResultsReadable(memSizes, bandwidths, count, kind, memMode,
+        printResultsReadable(memSizes, bandwidths, count, kind, memMode, (1 + endDevice - startDevice), wc);
-                         (1 + endDevice - startDevice), wc);
     }
 
     // clean up
@@ -573,8 +642,8 @@ void testBandwidthShmoo(memcpyKind kind, printMode printmode,
 ///////////////////////////////////////////////////////////////////////////////
 //  test the bandwidth of a device to host memcopy of a specific size
 ///////////////////////////////////////////////////////////////////////////////
-float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode,
+float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode, bool wc)
-                               bool wc) {
+{
     StopWatchInterface *timer           = NULL;
     float               elapsedTimeInMs = 0.0f;
     float               bandwidthInGBs  = 0.0f;
@@ -590,15 +659,14 @@ float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode,
     if (PINNED == memMode) {
         // pinned memory mode - use special function to get OS-pinned memory
 #if CUDART_VERSION >= 2020
-    checkCudaErrors(cudaHostAlloc((void **)&h_idata, memSize,
+        checkCudaErrors(cudaHostAlloc((void **)&h_idata, memSize, (wc) ? cudaHostAllocWriteCombined : 0));
-                                  (wc) ? cudaHostAllocWriteCombined : 0));
+        checkCudaErrors(cudaHostAlloc((void **)&h_odata, memSize, (wc) ? cudaHostAllocWriteCombined : 0));
-    checkCudaErrors(cudaHostAlloc((void **)&h_odata, memSize,
-                                  (wc) ? cudaHostAllocWriteCombined : 0));
 #else
         checkCudaErrors(cudaMallocHost((void **)&h_idata, memSize));
         checkCudaErrors(cudaMallocHost((void **)&h_odata, memSize));
 #endif
-  } else {
+    }
+    else {
         // pageable memory mode - use malloc
         h_idata = (unsigned char *)malloc(memSize);
         h_odata = (unsigned char *)malloc(memSize);
@@ -619,16 +687,15 @@ float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode,
     checkCudaErrors(cudaMalloc((void **)&d_idata, memSize));
 
     // initialize the device memory
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpy(d_idata, h_idata, memSize, cudaMemcpyHostToDevice));
-      cudaMemcpy(d_idata, h_idata, memSize, cudaMemcpyHostToDevice));
 
     // copy data from GPU to Host
     if (PINNED == memMode) {
-    if (bDontUseGPUTiming) sdkStartTimer(&timer);
+        if (bDontUseGPUTiming)
+            sdkStartTimer(&timer);
         checkCudaErrors(cudaEventRecord(start, 0));
         for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
-      checkCudaErrors(cudaMemcpyAsync(h_odata, d_idata, memSize,
+            checkCudaErrors(cudaMemcpyAsync(h_odata, d_idata, memSize, cudaMemcpyDeviceToHost, 0));
-                                      cudaMemcpyDeviceToHost, 0));
         }
         checkCudaErrors(cudaEventRecord(stop, 0));
         checkCudaErrors(cudaDeviceSynchronize());
@@ -638,12 +705,12 @@ float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode,
             elapsedTimeInMs = sdkGetTimerValue(&timer);
             sdkResetTimer(&timer);
         }
-  } else {
+    }
+    else {
         elapsedTimeInMs = 0;
         for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
             sdkStartTimer(&timer);
-      checkCudaErrors(
+            checkCudaErrors(cudaMemcpy(h_odata, d_idata, memSize, cudaMemcpyDeviceToHost));
-          cudaMemcpy(h_odata, d_idata, memSize, cudaMemcpyDeviceToHost));
             sdkStopTimer(&timer);
             elapsedTimeInMs += sdkGetTimerValue(&timer);
             sdkResetTimer(&timer);
@@ -663,7 +730,8 @@ float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode,
     if (PINNED == memMode) {
         checkCudaErrors(cudaFreeHost(h_idata));
         checkCudaErrors(cudaFreeHost(h_odata));
-  } else {
+    }
+    else {
         free(h_idata);
         free(h_odata);
     }
@@ -676,8 +744,8 @@ float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode,
 ///////////////////////////////////////////////////////////////////////////////
 //! test the bandwidth of a host to device memcopy of a specific size
 ///////////////////////////////////////////////////////////////////////////////
-float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode,
+float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode, bool wc)
-                               bool wc) {
+{
     StopWatchInterface *timer           = NULL;
     float               elapsedTimeInMs = 0.0f;
     float               bandwidthInGBs  = 0.0f;
@@ -692,13 +760,13 @@ float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode,
     if (PINNED == memMode) {
 #if CUDART_VERSION >= 2020
         // pinned memory mode - use special function to get OS-pinned memory
-    checkCudaErrors(cudaHostAlloc((void **)&h_odata, memSize,
+        checkCudaErrors(cudaHostAlloc((void **)&h_odata, memSize, (wc) ? cudaHostAllocWriteCombined : 0));
-                                  (wc) ? cudaHostAllocWriteCombined : 0));
 #else
         // pinned memory mode - use special function to get OS-pinned memory
         checkCudaErrors(cudaMallocHost((void **)&h_odata, memSize));
 #endif
-  } else {
+    }
+    else {
         // pageable memory mode - use malloc
         h_odata = (unsigned char *)malloc(memSize);
 
@@ -732,11 +800,11 @@ float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode,
 
     // copy host memory to device memory
     if (PINNED == memMode) {
-    if (bDontUseGPUTiming) sdkStartTimer(&timer);
+        if (bDontUseGPUTiming)
+            sdkStartTimer(&timer);
         checkCudaErrors(cudaEventRecord(start, 0));
         for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
-      checkCudaErrors(cudaMemcpyAsync(d_idata, h_odata, memSize,
+            checkCudaErrors(cudaMemcpyAsync(d_idata, h_odata, memSize, cudaMemcpyHostToDevice, 0));
-                                      cudaMemcpyHostToDevice, 0));
         }
         checkCudaErrors(cudaEventRecord(stop, 0));
         checkCudaErrors(cudaDeviceSynchronize());
@@ -746,12 +814,12 @@ float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode,
             elapsedTimeInMs = sdkGetTimerValue(&timer);
             sdkResetTimer(&timer);
         }
-  } else {
+    }
+    else {
         elapsedTimeInMs = 0;
         for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
             sdkStartTimer(&timer);
-      checkCudaErrors(
+            checkCudaErrors(cudaMemcpy(d_idata, h_odata, memSize, cudaMemcpyHostToDevice));
-          cudaMemcpy(d_idata, h_odata, memSize, cudaMemcpyHostToDevice));
             sdkStopTimer(&timer);
             elapsedTimeInMs += sdkGetTimerValue(&timer);
             sdkResetTimer(&timer);
@@ -770,7 +838,8 @@ float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode,
 
     if (PINNED == memMode) {
         checkCudaErrors(cudaFreeHost(h_odata));
-  } else {
+    }
+    else {
         free(h_odata);
     }
 
@@ -784,7 +853,8 @@ float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode,
 ///////////////////////////////////////////////////////////////////////////////
 //! test the bandwidth of a device to device memcopy of a specific size
 ///////////////////////////////////////////////////////////////////////////////
-float testDeviceToDeviceTransfer(unsigned int memSize) {
+float testDeviceToDeviceTransfer(unsigned int memSize)
+{
     StopWatchInterface *timer           = NULL;
     float               elapsedTimeInMs = 0.0f;
     float               bandwidthInGBs  = 0.0f;
@@ -814,16 +884,14 @@ float testDeviceToDeviceTransfer(unsigned int memSize) {
     checkCudaErrors(cudaMalloc((void **)&d_odata, memSize));
 
     // initialize memory
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpy(d_idata, h_idata, memSize, cudaMemcpyHostToDevice));
-      cudaMemcpy(d_idata, h_idata, memSize, cudaMemcpyHostToDevice));
 
     // run the memcopy
     sdkStartTimer(&timer);
     checkCudaErrors(cudaEventRecord(start, 0));
 
     for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
-    checkCudaErrors(
+        checkCudaErrors(cudaMemcpy(d_odata, d_idata, memSize, cudaMemcpyDeviceToDevice));
-        cudaMemcpy(d_odata, d_idata, memSize, cudaMemcpyDeviceToDevice));
     }
 
     checkCudaErrors(cudaEventRecord(stop, 0));
@@ -860,9 +928,14 @@ float testDeviceToDeviceTransfer(unsigned int memSize) {
 /////////////////////////////////////////////////////////
 // print results in an easily read format
 ////////////////////////////////////////////////////////
-void printResultsReadable(unsigned int *memSizes, double *bandwidths,
+void printResultsReadable(unsigned int *memSizes,
-                          unsigned int count, memcpyKind kind,
+                          double       *bandwidths,
-                          memoryMode memMode, int iNumDevs, bool wc) {
+                          unsigned int  count,
+                          memcpyKind    kind,
+                          memoryMode    memMode,
+                          int           iNumDevs,
+                          bool          wc)
+{
     printf(" %s Bandwidth, %i Device(s)\n", sMemoryCopyKind[kind], iNumDevs);
     printf(" %s Memory Transfers\n", sMemoryMode[memMode]);
 
@@ -874,35 +947,41 @@ void printResultsReadable(unsigned int *memSizes, double *bandwidths,
     unsigned int i;
 
     for (i = 0; i < (count - 1); i++) {
-    printf("   %u\t\t\t%s%.1f\n", memSizes[i],
+        printf("   %u\t\t\t%s%.1f\n", memSizes[i], (memSizes[i] < 10000) ? "\t" : "", bandwidths[i]);
-           (memSizes[i] < 10000) ? "\t" : "", bandwidths[i]);
     }
 
-  printf("   %u\t\t\t%s%.1f\n\n", memSizes[i],
+    printf("   %u\t\t\t%s%.1f\n\n", memSizes[i], (memSizes[i] < 10000) ? "\t" : "", bandwidths[i]);
-         (memSizes[i] < 10000) ? "\t" : "", bandwidths[i]);
 }
 
 ///////////////////////////////////////////////////////////////////////////
 // print results in a database format
 ///////////////////////////////////////////////////////////////////////////
-void printResultsCSV(unsigned int *memSizes, double *bandwidths,
+void printResultsCSV(unsigned int *memSizes,
-                     unsigned int count, memcpyKind kind, memoryMode memMode,
+                     double       *bandwidths,
-                     int iNumDevs, bool wc) {
+                     unsigned int  count,
+                     memcpyKind    kind,
+                     memoryMode    memMode,
+                     int           iNumDevs,
+                     bool          wc)
+{
     std::string sConfig;
 
     // log config information
     if (kind == DEVICE_TO_DEVICE) {
         sConfig += "D2D";
-  } else {
+    }
+    else {
         if (kind == DEVICE_TO_HOST) {
             sConfig += "D2H";
-    } else if (kind == HOST_TO_DEVICE) {
+        }
+        else if (kind == HOST_TO_DEVICE) {
             sConfig += "H2D";
         }
 
         if (memMode == PAGEABLE) {
             sConfig += "-Paged";
-    } else if (memMode == PINNED) {
+        }
+        else if (memMode == PINNED) {
             sConfig += "-Pinned";
 
             if (wc) {
@@ -916,27 +995,28 @@ void printResultsCSV(unsigned int *memSizes, double *bandwidths,
 
     for (i = 0; i < count; i++) {
         dSeconds = (double)memSizes[i] / (bandwidths[i] * (double)(1e9));
-    printf(
+        printf("bandwidthTest-%s, Bandwidth = %.1f GB/s, Time = %.5f s, Size = %u "
-        "bandwidthTest-%s, Bandwidth = %.1f GB/s, Time = %.5f s, Size = %u "
                "bytes, NumDevsUsed = %d\n",
-        sConfig.c_str(), bandwidths[i], dSeconds, memSizes[i], iNumDevs);
+               sConfig.c_str(),
+               bandwidths[i],
+               dSeconds,
+               memSizes[i],
+               iNumDevs);
     }
 }
 
 ///////////////////////////////////////////////////////////////////////////
 // Print help screen
 ///////////////////////////////////////////////////////////////////////////
-void printHelp(void) {
+void printHelp(void)
+{
     printf("Usage:  bandwidthTest [OPTION]...\n");
-  printf(
+    printf("Test the bandwidth for device to host, host to device, and device to "
-      "Test the bandwidth for device to host, host to device, and device to "
            "device transfers\n");
     printf("\n");
-  printf(
+    printf("Example:  measure the bandwidth of device to host pinned memory copies "
-      "Example:  measure the bandwidth of device to host pinned memory copies "
            "in the range 1024 Bytes to 102400 Bytes in 1024 Byte increments\n");
-  printf(
+    printf("./bandwidthTest --memory=pinned --mode=range --start=1024 --end=102400 "
-      "./bandwidthTest --memory=pinned --mode=range --start=1024 --end=102400 "
            "--increment=1024 --dtoh\n");
 
     printf("\n");

Samples/1_Utilities/deviceQuery/deviceQuery.cpp

@@ -32,7 +32,6 @@
 
 #include <cuda_runtime.h>
 #include <helper_cuda.h>
-
 #include <iostream>
 #include <memory>
 #include <string>
@@ -46,16 +45,13 @@ char **pArgv = NULL;
 #include <cuda.h>
 
 // This function wraps the CUDA Driver API into a template function
-template <class T>
+template <class T> inline void getCudaAttribute(T *attribute, CUdevice_attribute device_attribute, int device)
-inline void getCudaAttribute(T *attribute, CUdevice_attribute device_attribute,
+{
-                             int device) {
     CUresult error = cuDeviceGetAttribute(attribute, device_attribute, device);
 
     if (CUDA_SUCCESS != error) {
         fprintf(
-        stderr,
+            stderr, "cuSafeCallNoSync() Driver API error = %04d from file <%s>, line %i.\n", error, __FILE__, __LINE__);
-        "cuSafeCallNoSync() Driver API error = %04d from file <%s>, line %i.\n",
-        error, __FILE__, __LINE__);
 
         exit(EXIT_FAILURE);
     }
@@ -66,20 +62,19 @@ inline void getCudaAttribute(T *attribute, CUdevice_attribute device_attribute,
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     pArgc = &argc;
     pArgv = argv;
 
     printf("%s Starting...\n\n", argv[0]);
-  printf(
+    printf(" CUDA Device Query (Runtime API) version (CUDART static linking)\n\n");
-      " CUDA Device Query (Runtime API) version (CUDART static linking)\n\n");
 
     int         deviceCount = 0;
     cudaError_t error_id    = cudaGetDeviceCount(&deviceCount);
 
     if (error_id != cudaSuccess) {
-    printf("cudaGetDeviceCount returned %d\n-> %s\n",
+        printf("cudaGetDeviceCount returned %d\n-> %s\n", static_cast<int>(error_id), cudaGetErrorString(error_id));
-           static_cast<int>(error_id), cudaGetErrorString(error_id));
         printf("Result = FAIL\n");
         exit(EXIT_FAILURE);
     }
@@ -87,7 +82,8 @@ int main(int argc, char **argv) {
     // This function call returns 0 if there are no CUDA capable devices.
     if (deviceCount == 0) {
         printf("There are no available device(s) that support CUDA\n");
-  } else {
+    }
+    else {
         printf("Detected %d CUDA Capable device(s)\n", deviceCount);
     }
 
@@ -104,20 +100,23 @@ int main(int argc, char **argv) {
         cudaDriverGetVersion(&driverVersion);
         cudaRuntimeGetVersion(&runtimeVersion);
         printf("  CUDA Driver Version / Runtime Version          %d.%d / %d.%d\n",
-           driverVersion / 1000, (driverVersion % 100) / 10,
+               driverVersion / 1000,
-           runtimeVersion / 1000, (runtimeVersion % 100) / 10);
+               (driverVersion % 100) / 10,
-    printf("  CUDA Capability Major/Minor version number:    %d.%d\n",
+               runtimeVersion / 1000,
-           deviceProp.major, deviceProp.minor);
+               (runtimeVersion % 100) / 10);
+        printf("  CUDA Capability Major/Minor version number:    %d.%d\n", deviceProp.major, deviceProp.minor);
 
         char msg[256];
 #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
-    sprintf_s(msg, sizeof(msg),
+        sprintf_s(msg,
+                  sizeof(msg),
                   "  Total amount of global memory:                 %.0f MBytes "
                   "(%llu bytes)\n",
                   static_cast<float>(deviceProp.totalGlobalMem / 1048576.0f),
                   (unsigned long long)deviceProp.totalGlobalMem);
 #else
-    snprintf(msg, sizeof(msg),
+        snprintf(msg,
+                 sizeof(msg),
                  "  Total amount of global memory:                 %.0f MBytes "
                  "(%llu bytes)\n",
                  static_cast<float>(deviceProp.totalGlobalMem / 1048576.0f),
@@ -128,121 +127,100 @@ int main(int argc, char **argv) {
         printf("  (%03d) Multiprocessors, (%03d) CUDA Cores/MP:    %d CUDA Cores\n",
                deviceProp.multiProcessorCount,
                _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor),
-           _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) *
+               _ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) * deviceProp.multiProcessorCount);
-               deviceProp.multiProcessorCount);
+        printf("  GPU Max Clock rate:                            %.0f MHz (%0.2f "
-    printf(
-        "  GPU Max Clock rate:                            %.0f MHz (%0.2f "
                "GHz)\n",
-        deviceProp.clockRate * 1e-3f, deviceProp.clockRate * 1e-6f);
+               deviceProp.clockRate * 1e-3f,
+               deviceProp.clockRate * 1e-6f);
 
 #if CUDART_VERSION >= 5000
         // This is supported in CUDA 5.0 (runtime API device properties)
-    printf("  Memory Clock rate:                             %.0f Mhz\n",
+        printf("  Memory Clock rate:                             %.0f Mhz\n", deviceProp.memoryClockRate * 1e-3f);
-           deviceProp.memoryClockRate * 1e-3f);
+        printf("  Memory Bus Width:                              %d-bit\n", deviceProp.memoryBusWidth);
-    printf("  Memory Bus Width:                              %d-bit\n",
-           deviceProp.memoryBusWidth);
 
         if (deviceProp.l2CacheSize) {
-      printf("  L2 Cache Size:                                 %d bytes\n",
+            printf("  L2 Cache Size:                                 %d bytes\n", deviceProp.l2CacheSize);
-             deviceProp.l2CacheSize);
         }
 
 #else
         // This only available in CUDA 4.0-4.2 (but these were only exposed in the
         // CUDA Driver API)
         int memoryClock;
-    getCudaAttribute<int>(&memoryClock, CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE,
+        getCudaAttribute<int>(&memoryClock, CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE, dev);
-                          dev);
+        printf("  Memory Clock rate:                             %.0f Mhz\n", memoryClock * 1e-3f);
-    printf("  Memory Clock rate:                             %.0f Mhz\n",
-           memoryClock * 1e-3f);
         int memBusWidth;
-    getCudaAttribute<int>(&memBusWidth,
+        getCudaAttribute<int>(&memBusWidth, CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH, dev);
-                          CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH, dev);
+        printf("  Memory Bus Width:                              %d-bit\n", memBusWidth);
-    printf("  Memory Bus Width:                              %d-bit\n",
-           memBusWidth);
         int L2CacheSize;
         getCudaAttribute<int>(&L2CacheSize, CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE, dev);
 
         if (L2CacheSize) {
-      printf("  L2 Cache Size:                                 %d bytes\n",
+            printf("  L2 Cache Size:                                 %d bytes\n", L2CacheSize);
-             L2CacheSize);
         }
 
 #endif
 
-    printf(
+        printf("  Maximum Texture Dimension Size (x,y,z)         1D=(%d), 2D=(%d, "
-        "  Maximum Texture Dimension Size (x,y,z)         1D=(%d), 2D=(%d, "
                "%d), 3D=(%d, %d, %d)\n",
-        deviceProp.maxTexture1D, deviceProp.maxTexture2D[0],
+               deviceProp.maxTexture1D,
-        deviceProp.maxTexture2D[1], deviceProp.maxTexture3D[0],
+               deviceProp.maxTexture2D[0],
-        deviceProp.maxTexture3D[1], deviceProp.maxTexture3D[2]);
+               deviceProp.maxTexture2D[1],
-    printf(
+               deviceProp.maxTexture3D[0],
-        "  Maximum Layered 1D Texture Size, (num) layers  1D=(%d), %d layers\n",
+               deviceProp.maxTexture3D[1],
-        deviceProp.maxTexture1DLayered[0], deviceProp.maxTexture1DLayered[1]);
+               deviceProp.maxTexture3D[2]);
-    printf(
+        printf("  Maximum Layered 1D Texture Size, (num) layers  1D=(%d), %d layers\n",
-        "  Maximum Layered 2D Texture Size, (num) layers  2D=(%d, %d), %d "
+               deviceProp.maxTexture1DLayered[0],
+               deviceProp.maxTexture1DLayered[1]);
+        printf("  Maximum Layered 2D Texture Size, (num) layers  2D=(%d, %d), %d "
                "layers\n",
-        deviceProp.maxTexture2DLayered[0], deviceProp.maxTexture2DLayered[1],
+               deviceProp.maxTexture2DLayered[0],
+               deviceProp.maxTexture2DLayered[1],
                deviceProp.maxTexture2DLayered[2]);
 
-    printf("  Total amount of constant memory:               %zu bytes\n",
+        printf("  Total amount of constant memory:               %zu bytes\n", deviceProp.totalConstMem);
-           deviceProp.totalConstMem);
+        printf("  Total amount of shared memory per block:       %zu bytes\n", deviceProp.sharedMemPerBlock);
-    printf("  Total amount of shared memory per block:       %zu bytes\n",
+        printf("  Total shared memory per multiprocessor:        %zu bytes\n", deviceProp.sharedMemPerMultiprocessor);
-           deviceProp.sharedMemPerBlock);
+        printf("  Total number of registers available per block: %d\n", deviceProp.regsPerBlock);
-    printf("  Total shared memory per multiprocessor:        %zu bytes\n",
+        printf("  Warp size:                                     %d\n", deviceProp.warpSize);
-           deviceProp.sharedMemPerMultiprocessor);
+        printf("  Maximum number of threads per multiprocessor:  %d\n", deviceProp.maxThreadsPerMultiProcessor);
-    printf("  Total number of registers available per block: %d\n",
+        printf("  Maximum number of threads per block:           %d\n", deviceProp.maxThreadsPerBlock);
-           deviceProp.regsPerBlock);
-    printf("  Warp size:                                     %d\n",
-           deviceProp.warpSize);
-    printf("  Maximum number of threads per multiprocessor:  %d\n",
-           deviceProp.maxThreadsPerMultiProcessor);
-    printf("  Maximum number of threads per block:           %d\n",
-           deviceProp.maxThreadsPerBlock);
         printf("  Max dimension size of a thread block (x,y,z): (%d, %d, %d)\n",
-           deviceProp.maxThreadsDim[0], deviceProp.maxThreadsDim[1],
+               deviceProp.maxThreadsDim[0],
+               deviceProp.maxThreadsDim[1],
                deviceProp.maxThreadsDim[2]);
         printf("  Max dimension size of a grid size    (x,y,z): (%d, %d, %d)\n",
-           deviceProp.maxGridSize[0], deviceProp.maxGridSize[1],
+               deviceProp.maxGridSize[0],
+               deviceProp.maxGridSize[1],
                deviceProp.maxGridSize[2]);
-    printf("  Maximum memory pitch:                          %zu bytes\n",
+        printf("  Maximum memory pitch:                          %zu bytes\n", deviceProp.memPitch);
-           deviceProp.memPitch);
+        printf("  Texture alignment:                             %zu bytes\n", deviceProp.textureAlignment);
-    printf("  Texture alignment:                             %zu bytes\n",
+        printf("  Concurrent copy and kernel execution:          %s with %d copy "
-           deviceProp.textureAlignment);
-    printf(
-        "  Concurrent copy and kernel execution:          %s with %d copy "
                "engine(s)\n",
-        (deviceProp.deviceOverlap ? "Yes" : "No"), deviceProp.asyncEngineCount);
+               (deviceProp.deviceOverlap ? "Yes" : "No"),
+               deviceProp.asyncEngineCount);
         printf("  Run time limit on kernels:                     %s\n",
                deviceProp.kernelExecTimeoutEnabled ? "Yes" : "No");
-    printf("  Integrated GPU sharing Host Memory:            %s\n",
+        printf("  Integrated GPU sharing Host Memory:            %s\n", deviceProp.integrated ? "Yes" : "No");
-           deviceProp.integrated ? "Yes" : "No");
+        printf("  Support host page-locked memory mapping:       %s\n", deviceProp.canMapHostMemory ? "Yes" : "No");
-    printf("  Support host page-locked memory mapping:       %s\n",
+        printf("  Alignment requirement for Surfaces:            %s\n", deviceProp.surfaceAlignment ? "Yes" : "No");
-           deviceProp.canMapHostMemory ? "Yes" : "No");
+        printf("  Device has ECC support:                        %s\n", deviceProp.ECCEnabled ? "Enabled" : "Disabled");
-    printf("  Alignment requirement for Surfaces:            %s\n",
-           deviceProp.surfaceAlignment ? "Yes" : "No");
-    printf("  Device has ECC support:                        %s\n",
-           deviceProp.ECCEnabled ? "Enabled" : "Disabled");
 #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
         printf("  CUDA Device Driver Mode (TCC or WDDM):         %s\n",
-           deviceProp.tccDriver ? "TCC (Tesla Compute Cluster Driver)"
+               deviceProp.tccDriver ? "TCC (Tesla Compute Cluster Driver)" : "WDDM (Windows Display Driver Model)");
-                                : "WDDM (Windows Display Driver Model)");
 #endif
-    printf("  Device supports Unified Addressing (UVA):      %s\n",
+        printf("  Device supports Unified Addressing (UVA):      %s\n", deviceProp.unifiedAddressing ? "Yes" : "No");
-           deviceProp.unifiedAddressing ? "Yes" : "No");
+        printf("  Device supports Managed Memory:                %s\n", deviceProp.managedMemory ? "Yes" : "No");
-    printf("  Device supports Managed Memory:                %s\n",
-           deviceProp.managedMemory ? "Yes" : "No");
         printf("  Device supports Compute Preemption:            %s\n",
                deviceProp.computePreemptionSupported ? "Yes" : "No");
-    printf("  Supports Cooperative Kernel Launch:            %s\n",
+        printf("  Supports Cooperative Kernel Launch:            %s\n", deviceProp.cooperativeLaunch ? "Yes" : "No");
-           deviceProp.cooperativeLaunch ? "Yes" : "No");
         printf("  Supports MultiDevice Co-op Kernel Launch:      %s\n",
                deviceProp.cooperativeMultiDeviceLaunch ? "Yes" : "No");
         printf("  Device PCI Domain ID / Bus ID / location ID:   %d / %d / %d\n",
-           deviceProp.pciDomainID, deviceProp.pciBusID, deviceProp.pciDeviceID);
+               deviceProp.pciDomainID,
+               deviceProp.pciBusID,
+               deviceProp.pciDeviceID);
 
-    const char *sComputeMode[] = {
+        const char *sComputeMode[] = {"Default (multiple host threads can use ::cudaSetDevice() with device "
-        "Default (multiple host threads can use ::cudaSetDevice() with device "
                                       "simultaneously)",
                                       "Exclusive (only one host thread in one process is able to use "
                                       "::cudaSetDevice() with this device)",
@@ -250,7 +228,8 @@ int main(int argc, char **argv) {
                                       "device)",
                                       "Exclusive Process (many threads in one process is able to use "
                                       "::cudaSetDevice() with this device)",
-        "Unknown", NULL};
+                                      "Unknown",
+                                      NULL};
         printf("  Compute Mode:\n");
         printf("     < %s >\n", sComputeMode[deviceProp.computeMode]);
     }
@@ -286,10 +265,12 @@ int main(int argc, char **argv) {
                     if (gpuid[i] == gpuid[j]) {
                         continue;
                     }
-          checkCudaErrors(
+                    checkCudaErrors(cudaDeviceCanAccessPeer(&can_access_peer, gpuid[i], gpuid[j]));
-              cudaDeviceCanAccessPeer(&can_access_peer, gpuid[i], gpuid[j]));
                     printf("> Peer access from %s (GPU%d) -> %s (GPU%d) : %s\n",
-                 prop[gpuid[i]].name, gpuid[i], prop[gpuid[j]].name, gpuid[j],
+                           prop[gpuid[i]].name,
+                           gpuid[i],
+                           prop[gpuid[j]].name,
+                           gpuid[j],
                            can_access_peer ? "Yes" : "No");
                 }
             }
@@ -306,22 +287,18 @@ int main(int argc, char **argv) {
     // driver version
     sProfileString += ", CUDA Driver Version = ";
 #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
-  sprintf_s(cTemp, 10, "%d.%d", driverVersion / 1000,
+    sprintf_s(cTemp, 10, "%d.%d", driverVersion / 1000, (driverVersion % 100) / 10);
-            (driverVersion % 100) / 10);
 #else
-  snprintf(cTemp, sizeof(cTemp), "%d.%d", driverVersion / 1000,
+    snprintf(cTemp, sizeof(cTemp), "%d.%d", driverVersion / 1000, (driverVersion % 100) / 10);
-           (driverVersion % 100) / 10);
 #endif
     sProfileString += cTemp;
 
     // Runtime version
     sProfileString += ", CUDA Runtime Version = ";
 #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
-  sprintf_s(cTemp, 10, "%d.%d", runtimeVersion / 1000,
+    sprintf_s(cTemp, 10, "%d.%d", runtimeVersion / 1000, (runtimeVersion % 100) / 10);
-            (runtimeVersion % 100) / 10);
 #else
-  snprintf(cTemp, sizeof(cTemp), "%d.%d", runtimeVersion / 1000,
+    snprintf(cTemp, sizeof(cTemp), "%d.%d", runtimeVersion / 1000, (runtimeVersion % 100) / 10);
-           (runtimeVersion % 100) / 10);
 #endif
     sProfileString += cTemp;
 

Samples/1_Utilities/deviceQueryDrv/deviceQueryDrv.cpp

@@ -30,17 +30,17 @@
  */
 
 // includes, system
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-
 #include <cuda.h>
 #include <helper_cuda_drvapi.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     CUdevice dev;
     int      major = 0, minor = 0;
     int      deviceCount = 0;
@@ -58,15 +58,14 @@ int main(int argc, char **argv) {
     // This function call returns 0 if there are no CUDA capable devices.
     if (deviceCount == 0) {
         printf("There are no available device(s) that support CUDA\n");
-  } else {
+    }
+    else {
         printf("Detected %d CUDA Capable device(s)\n", deviceCount);
     }
 
     for (dev = 0; dev < deviceCount; ++dev) {
-    checkCudaErrors(cuDeviceGetAttribute(
+        checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, dev));
-        &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, dev));
+        checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, dev));
-    checkCudaErrors(cuDeviceGetAttribute(
-        &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, dev));
 
         checkCudaErrors(cuDeviceGetName(deviceName, 256, dev));
 
@@ -75,9 +74,9 @@ int main(int argc, char **argv) {
         int driverVersion = 0;
         checkCudaErrors(cuDriverGetVersion(&driverVersion));
         printf("  CUDA Driver Version:                           %d.%d\n",
-           driverVersion / 1000, (driverVersion % 100) / 10);
+               driverVersion / 1000,
-    printf("  CUDA Capability Major/Minor version number:    %d.%d\n", major,
+               (driverVersion % 100) / 10);
-           minor);
+        printf("  CUDA Capability Major/Minor version number:    %d.%d\n", major, minor);
 
         size_t totalGlobalMem;
         checkCudaErrors(cuDeviceTotalMem(&totalGlobalMem, dev));
@@ -91,231 +90,169 @@ int main(int argc, char **argv) {
         printf("%s", msg);
 
         int multiProcessorCount;
-    getCudaAttribute<int>(&multiProcessorCount,
+        getCudaAttribute<int>(&multiProcessorCount, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, dev);
-                          CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, dev);
 
         printf("  (%2d) Multiprocessors, (%3d) CUDA Cores/MP:     %d CUDA Cores\n",
-           multiProcessorCount, _ConvertSMVer2CoresDRV(major, minor),
+               multiProcessorCount,
+               _ConvertSMVer2CoresDRV(major, minor),
                _ConvertSMVer2CoresDRV(major, minor) * multiProcessorCount);
 
         int clockRate;
         getCudaAttribute<int>(&clockRate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, dev);
-    printf(
+        printf("  GPU Max Clock rate:                            %.0f MHz (%0.2f "
-        "  GPU Max Clock rate:                            %.0f MHz (%0.2f "
                "GHz)\n",
-        clockRate * 1e-3f, clockRate * 1e-6f);
+               clockRate * 1e-3f,
+               clockRate * 1e-6f);
         int memoryClock;
-    getCudaAttribute<int>(&memoryClock, CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE,
+        getCudaAttribute<int>(&memoryClock, CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE, dev);
-                          dev);
+        printf("  Memory Clock rate:                             %.0f Mhz\n", memoryClock * 1e-3f);
-    printf("  Memory Clock rate:                             %.0f Mhz\n",
-           memoryClock * 1e-3f);
         int memBusWidth;
-    getCudaAttribute<int>(&memBusWidth,
+        getCudaAttribute<int>(&memBusWidth, CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH, dev);
-                          CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH, dev);
+        printf("  Memory Bus Width:                              %d-bit\n", memBusWidth);
-    printf("  Memory Bus Width:                              %d-bit\n",
-           memBusWidth);
         int L2CacheSize;
         getCudaAttribute<int>(&L2CacheSize, CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE, dev);
 
         if (L2CacheSize) {
-      printf("  L2 Cache Size:                                 %d bytes\n",
+            printf("  L2 Cache Size:                                 %d bytes\n", L2CacheSize);
-             L2CacheSize);
         }
 
         int maxTex1D, maxTex2D[2], maxTex3D[3];
-    getCudaAttribute<int>(&maxTex1D,
+        getCudaAttribute<int>(&maxTex1D, CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH, dev);
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH, dev);
+        getCudaAttribute<int>(&maxTex2D[0], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH, dev);
-    getCudaAttribute<int>(&maxTex2D[0],
+        getCudaAttribute<int>(&maxTex2D[1], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_HEIGHT, dev);
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH, dev);
+        getCudaAttribute<int>(&maxTex3D[0], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH, dev);
-    getCudaAttribute<int>(&maxTex2D[1],
+        getCudaAttribute<int>(&maxTex3D[1], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT, dev);
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_HEIGHT, dev);
+        getCudaAttribute<int>(&maxTex3D[2], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH, dev);
-    getCudaAttribute<int>(&maxTex3D[0],
+        printf("  Max Texture Dimension Sizes                    1D=(%d) 2D=(%d, %d) "
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH, dev);
-    getCudaAttribute<int>(&maxTex3D[1],
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT, dev);
-    getCudaAttribute<int>(&maxTex3D[2],
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH, dev);
-    printf(
-        "  Max Texture Dimension Sizes                    1D=(%d) 2D=(%d, %d) "
                "3D=(%d, %d, %d)\n",
-        maxTex1D, maxTex2D[0], maxTex2D[1], maxTex3D[0], maxTex3D[1],
+               maxTex1D,
+               maxTex2D[0],
+               maxTex2D[1],
+               maxTex3D[0],
+               maxTex3D[1],
                maxTex3D[2]);
 
         int maxTex1DLayered[2];
-    getCudaAttribute<int>(&maxTex1DLayered[0],
+        getCudaAttribute<int>(&maxTex1DLayered[0], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH, dev);
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH,
+        getCudaAttribute<int>(&maxTex1DLayered[1], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS, dev);
-                          dev);
+        printf("  Maximum Layered 1D Texture Size, (num) layers  1D=(%d), %d layers\n",
-    getCudaAttribute<int>(&maxTex1DLayered[1],
+               maxTex1DLayered[0],
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS,
+               maxTex1DLayered[1]);
-                          dev);
-    printf(
-        "  Maximum Layered 1D Texture Size, (num) layers  1D=(%d), %d layers\n",
-        maxTex1DLayered[0], maxTex1DLayered[1]);
 
         int maxTex2DLayered[3];
-    getCudaAttribute<int>(&maxTex2DLayered[0],
+        getCudaAttribute<int>(&maxTex2DLayered[0], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH, dev);
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH,
+        getCudaAttribute<int>(&maxTex2DLayered[1], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT, dev);
-                          dev);
+        getCudaAttribute<int>(&maxTex2DLayered[2], CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS, dev);
-    getCudaAttribute<int>(&maxTex2DLayered[1],
+        printf("  Maximum Layered 2D Texture Size, (num) layers  2D=(%d, %d), %d "
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT,
-                          dev);
-    getCudaAttribute<int>(&maxTex2DLayered[2],
-                          CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS,
-                          dev);
-    printf(
-        "  Maximum Layered 2D Texture Size, (num) layers  2D=(%d, %d), %d "
                "layers\n",
-        maxTex2DLayered[0], maxTex2DLayered[1], maxTex2DLayered[2]);
+               maxTex2DLayered[0],
+               maxTex2DLayered[1],
+               maxTex2DLayered[2]);
 
         int totalConstantMemory;
-    getCudaAttribute<int>(&totalConstantMemory,
+        getCudaAttribute<int>(&totalConstantMemory, CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY, dev);
-                          CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY, dev);
+        printf("  Total amount of constant memory:               %u bytes\n", totalConstantMemory);
-    printf("  Total amount of constant memory:               %u bytes\n",
-           totalConstantMemory);
         int sharedMemPerBlock;
-    getCudaAttribute<int>(&sharedMemPerBlock,
+        getCudaAttribute<int>(&sharedMemPerBlock, CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK, dev);
-                          CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK, dev);
+        printf("  Total amount of shared memory per block:       %u bytes\n", sharedMemPerBlock);
-    printf("  Total amount of shared memory per block:       %u bytes\n",
-           sharedMemPerBlock);
         int regsPerBlock;
-    getCudaAttribute<int>(&regsPerBlock,
+        getCudaAttribute<int>(&regsPerBlock, CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK, dev);
-                          CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK, dev);
+        printf("  Total number of registers available per block: %d\n", regsPerBlock);
-    printf("  Total number of registers available per block: %d\n",
-           regsPerBlock);
         int warpSize;
         getCudaAttribute<int>(&warpSize, CU_DEVICE_ATTRIBUTE_WARP_SIZE, dev);
         printf("  Warp size:                                     %d\n", warpSize);
         int maxThreadsPerMultiProcessor;
-    getCudaAttribute<int>(&maxThreadsPerMultiProcessor,
+        getCudaAttribute<int>(&maxThreadsPerMultiProcessor, CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR, dev);
-                          CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR,
+        printf("  Maximum number of threads per multiprocessor:  %d\n", maxThreadsPerMultiProcessor);
-                          dev);
-    printf("  Maximum number of threads per multiprocessor:  %d\n",
-           maxThreadsPerMultiProcessor);
         int maxThreadsPerBlock;
-    getCudaAttribute<int>(&maxThreadsPerBlock,
+        getCudaAttribute<int>(&maxThreadsPerBlock, CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK, dev);
-                          CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK, dev);
+        printf("  Maximum number of threads per block:           %d\n", maxThreadsPerBlock);
-    printf("  Maximum number of threads per block:           %d\n",
-           maxThreadsPerBlock);
 
         int blockDim[3];
-    getCudaAttribute<int>(&blockDim[0], CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X,
+        getCudaAttribute<int>(&blockDim[0], CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X, dev);
-                          dev);
+        getCudaAttribute<int>(&blockDim[1], CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y, dev);
-    getCudaAttribute<int>(&blockDim[1], CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y,
+        getCudaAttribute<int>(&blockDim[2], CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Z, dev);
-                          dev);
+        printf("  Max dimension size of a thread block (x,y,z): (%d, %d, %d)\n", blockDim[0], blockDim[1], blockDim[2]);
-    getCudaAttribute<int>(&blockDim[2], CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Z,
-                          dev);
-    printf("  Max dimension size of a thread block (x,y,z): (%d, %d, %d)\n",
-           blockDim[0], blockDim[1], blockDim[2]);
         int gridDim[3];
         getCudaAttribute<int>(&gridDim[0], CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X, dev);
         getCudaAttribute<int>(&gridDim[1], CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y, dev);
         getCudaAttribute<int>(&gridDim[2], CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z, dev);
-    printf("  Max dimension size of a grid size (x,y,z):    (%d, %d, %d)\n",
+        printf("  Max dimension size of a grid size (x,y,z):    (%d, %d, %d)\n", gridDim[0], gridDim[1], gridDim[2]);
-           gridDim[0], gridDim[1], gridDim[2]);
 
         int textureAlign;
-    getCudaAttribute<int>(&textureAlign, CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT,
+        getCudaAttribute<int>(&textureAlign, CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT, dev);
-                          dev);
+        printf("  Texture alignment:                             %u bytes\n", textureAlign);
-    printf("  Texture alignment:                             %u bytes\n",
-           textureAlign);
 
         int memPitch;
         getCudaAttribute<int>(&memPitch, CU_DEVICE_ATTRIBUTE_MAX_PITCH, dev);
-    printf("  Maximum memory pitch:                          %u bytes\n",
+        printf("  Maximum memory pitch:                          %u bytes\n", memPitch);
-           memPitch);
 
         int gpuOverlap;
         getCudaAttribute<int>(&gpuOverlap, CU_DEVICE_ATTRIBUTE_GPU_OVERLAP, dev);
 
         int asyncEngineCount;
-    getCudaAttribute<int>(&asyncEngineCount,
+        getCudaAttribute<int>(&asyncEngineCount, CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT, dev);
-                          CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT, dev);
+        printf("  Concurrent copy and kernel execution:          %s with %d copy "
-    printf(
-        "  Concurrent copy and kernel execution:          %s with %d copy "
                "engine(s)\n",
-        (gpuOverlap ? "Yes" : "No"), asyncEngineCount);
+               (gpuOverlap ? "Yes" : "No"),
+               asyncEngineCount);
 
         int kernelExecTimeoutEnabled;
-    getCudaAttribute<int>(&kernelExecTimeoutEnabled,
+        getCudaAttribute<int>(&kernelExecTimeoutEnabled, CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT, dev);
-                          CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT, dev);
+        printf("  Run time limit on kernels:                     %s\n", kernelExecTimeoutEnabled ? "Yes" : "No");
-    printf("  Run time limit on kernels:                     %s\n",
-           kernelExecTimeoutEnabled ? "Yes" : "No");
         int integrated;
         getCudaAttribute<int>(&integrated, CU_DEVICE_ATTRIBUTE_INTEGRATED, dev);
-    printf("  Integrated GPU sharing Host Memory:            %s\n",
+        printf("  Integrated GPU sharing Host Memory:            %s\n", integrated ? "Yes" : "No");
-           integrated ? "Yes" : "No");
         int canMapHostMemory;
-    getCudaAttribute<int>(&canMapHostMemory,
+        getCudaAttribute<int>(&canMapHostMemory, CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY, dev);
-                          CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY, dev);
+        printf("  Support host page-locked memory mapping:       %s\n", canMapHostMemory ? "Yes" : "No");
-    printf("  Support host page-locked memory mapping:       %s\n",
-           canMapHostMemory ? "Yes" : "No");
 
         int concurrentKernels;
-    getCudaAttribute<int>(&concurrentKernels,
+        getCudaAttribute<int>(&concurrentKernels, CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS, dev);
-                          CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS, dev);
+        printf("  Concurrent kernel execution:                   %s\n", concurrentKernels ? "Yes" : "No");
-    printf("  Concurrent kernel execution:                   %s\n",
-           concurrentKernels ? "Yes" : "No");
 
         int surfaceAlignment;
-    getCudaAttribute<int>(&surfaceAlignment,
+        getCudaAttribute<int>(&surfaceAlignment, CU_DEVICE_ATTRIBUTE_SURFACE_ALIGNMENT, dev);
-                          CU_DEVICE_ATTRIBUTE_SURFACE_ALIGNMENT, dev);
+        printf("  Alignment requirement for Surfaces:            %s\n", surfaceAlignment ? "Yes" : "No");
-    printf("  Alignment requirement for Surfaces:            %s\n",
-           surfaceAlignment ? "Yes" : "No");
 
         int eccEnabled;
         getCudaAttribute<int>(&eccEnabled, CU_DEVICE_ATTRIBUTE_ECC_ENABLED, dev);
-    printf("  Device has ECC support:                        %s\n",
+        printf("  Device has ECC support:                        %s\n", eccEnabled ? "Enabled" : "Disabled");
-           eccEnabled ? "Enabled" : "Disabled");
 
 #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
         int tccDriver;
         getCudaAttribute<int>(&tccDriver, CU_DEVICE_ATTRIBUTE_TCC_DRIVER, dev);
         printf("  CUDA Device Driver Mode (TCC or WDDM):         %s\n",
-           tccDriver ? "TCC (Tesla Compute Cluster Driver)"
+               tccDriver ? "TCC (Tesla Compute Cluster Driver)" : "WDDM (Windows Display Driver Model)");
-                     : "WDDM (Windows Display Driver Model)");
 #endif
 
         int unifiedAddressing;
-    getCudaAttribute<int>(&unifiedAddressing,
+        getCudaAttribute<int>(&unifiedAddressing, CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING, dev);
-                          CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING, dev);
+        printf("  Device supports Unified Addressing (UVA):      %s\n", unifiedAddressing ? "Yes" : "No");
-    printf("  Device supports Unified Addressing (UVA):      %s\n",
-           unifiedAddressing ? "Yes" : "No");
 
         int managedMemory;
-    getCudaAttribute<int>(&managedMemory, CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY,
+        getCudaAttribute<int>(&managedMemory, CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY, dev);
-                          dev);
+        printf("  Device supports Managed Memory:                %s\n", managedMemory ? "Yes" : "No");
-    printf("  Device supports Managed Memory:                %s\n",
-           managedMemory ? "Yes" : "No");
 
         int computePreemption;
-    getCudaAttribute<int>(&computePreemption,
+        getCudaAttribute<int>(&computePreemption, CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED, dev);
-                          CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED,
+        printf("  Device supports Compute Preemption:            %s\n", computePreemption ? "Yes" : "No");
-                          dev);
-    printf("  Device supports Compute Preemption:            %s\n",
-           computePreemption ? "Yes" : "No");
 
         int cooperativeLaunch;
-    getCudaAttribute<int>(&cooperativeLaunch,
+        getCudaAttribute<int>(&cooperativeLaunch, CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH, dev);
-                          CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH, dev);
+        printf("  Supports Cooperative Kernel Launch:            %s\n", cooperativeLaunch ? "Yes" : "No");
-    printf("  Supports Cooperative Kernel Launch:            %s\n",
-           cooperativeLaunch ? "Yes" : "No");
 
         int cooperativeMultiDevLaunch;
-    getCudaAttribute<int>(&cooperativeMultiDevLaunch,
+        getCudaAttribute<int>(&cooperativeMultiDevLaunch, CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH, dev);
-                          CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH,
+        printf("  Supports MultiDevice Co-op Kernel Launch:      %s\n", cooperativeMultiDevLaunch ? "Yes" : "No");
-                          dev);
-    printf("  Supports MultiDevice Co-op Kernel Launch:      %s\n",
-           cooperativeMultiDevLaunch ? "Yes" : "No");
 
         int pciDomainID, pciBusID, pciDeviceID;
         getCudaAttribute<int>(&pciDomainID, CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID, dev);
         getCudaAttribute<int>(&pciBusID, CU_DEVICE_ATTRIBUTE_PCI_BUS_ID, dev);
         getCudaAttribute<int>(&pciDeviceID, CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID, dev);
-    printf("  Device PCI Domain ID / Bus ID / location ID:   %d / %d / %d\n",
+        printf("  Device PCI Domain ID / Bus ID / location ID:   %d / %d / %d\n", pciDomainID, pciBusID, pciDeviceID);
-           pciDomainID, pciBusID, pciDeviceID);
 
-    const char *sComputeMode[] = {
+        const char *sComputeMode[] = {"Default (multiple host threads can use ::cudaSetDevice() with device "
-        "Default (multiple host threads can use ::cudaSetDevice() with device "
                                       "simultaneously)",
                                       "Exclusive (only one host thread in one process is able to use "
                                       "::cudaSetDevice() with this device)",
@@ -323,7 +260,8 @@ int main(int argc, char **argv) {
                                       "device)",
                                       "Exclusive Process (many threads in one process is able to use "
                                       "::cudaSetDevice() with this device)",
-        "Unknown", NULL};
+                                      "Unknown",
+                                      NULL};
 
         int computeMode;
         getCudaAttribute<int>(&computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, dev);
@@ -338,10 +276,8 @@ int main(int argc, char **argv) {
         int tccDriver     = 0;
 
         for (int i = 0; i < deviceCount; i++) {
-      checkCudaErrors(cuDeviceGetAttribute(
+            checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, i));
-          &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, i));
+            checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, i));
-      checkCudaErrors(cuDeviceGetAttribute(
-          &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, i));
             getCudaAttribute<int>(&tccDriver, CU_DEVICE_ATTRIBUTE_TCC_DRIVER, i);
 
             // Only boards based on Fermi or later can support P2P
@@ -367,14 +303,15 @@ int main(int argc, char **argv) {
                     if (gpuid[i] == gpuid[j]) {
                         continue;
                     }
-          checkCudaErrors(
+                    checkCudaErrors(cuDeviceCanAccessPeer(&can_access_peer, gpuid[i], gpuid[j]));
-              cuDeviceCanAccessPeer(&can_access_peer, gpuid[i], gpuid[j]));
                     checkCudaErrors(cuDeviceGetName(deviceName0, 256, gpuid[i]));
                     checkCudaErrors(cuDeviceGetName(deviceName1, 256, gpuid[j]));
-          printf(
+                    printf("> Peer-to-Peer (P2P) access from %s (GPU%d) -> %s (GPU%d) : "
-              "> Peer-to-Peer (P2P) access from %s (GPU%d) -> %s (GPU%d) : "
                            "%s\n",
-              deviceName0, gpuid[i], deviceName1, gpuid[j],
+                           deviceName0,
+                           gpuid[i],
+                           deviceName1,
+                           gpuid[j],
                            can_access_peer ? "Yes" : "No");
                 }
             }

Samples/1_Utilities/topologyQuery/README.md

@@ -30,4 +30,3 @@ cudaGetDeviceCount, cudaDeviceGetAttribute
 Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)
-

Samples/1_Utilities/topologyQuery/topologyQuery.cu

@@ -37,32 +37,30 @@
 #include <helper_cuda.h>
 #include <helper_functions.h> // helper for shared that are common to CUDA Samples
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     int deviceCount = 0;
     checkCudaErrors(cudaGetDeviceCount(&deviceCount));
 
     // Enumerates Device <-> Device links
     for (int device1 = 0; device1 < deviceCount; device1++) {
         for (int device2 = 0; device2 < deviceCount; device2++) {
-      if (device1 == device2) continue;
+            if (device1 == device2)
+                continue;
 
             int perfRank        = 0;
             int atomicSupported = 0;
             int accessSupported = 0;
 
-      checkCudaErrors(cudaDeviceGetP2PAttribute(
+            checkCudaErrors(
-          &accessSupported, cudaDevP2PAttrAccessSupported, device1, device2));
+                cudaDeviceGetP2PAttribute(&accessSupported, cudaDevP2PAttrAccessSupported, device1, device2));
-      checkCudaErrors(cudaDeviceGetP2PAttribute(
+            checkCudaErrors(cudaDeviceGetP2PAttribute(&perfRank, cudaDevP2PAttrPerformanceRank, device1, device2));
-          &perfRank, cudaDevP2PAttrPerformanceRank, device1, device2));
+            checkCudaErrors(
-      checkCudaErrors(cudaDeviceGetP2PAttribute(
+                cudaDeviceGetP2PAttribute(&atomicSupported, cudaDevP2PAttrNativeAtomicSupported, device1, device2));
-          &atomicSupported, cudaDevP2PAttrNativeAtomicSupported, device1,
-          device2));
 
             if (accessSupported) {
-        std::cout << "GPU" << device1 << " <-> GPU" << device2 << ":"
+                std::cout << "GPU" << device1 << " <-> GPU" << device2 << ":" << std::endl;
-                  << std::endl;
+                std::cout << "  * Atomic Supported: " << (atomicSupported ? "yes" : "no") << std::endl;
-        std::cout << "  * Atomic Supported: "
-                  << (atomicSupported ? "yes" : "no") << std::endl;
                 std::cout << "  * Perf Rank: " << perfRank << std::endl;
             }
         }
@@ -71,11 +69,9 @@ int main(int argc, char **argv) {
     // Enumerates Device <-> Host links
     for (int device = 0; device < deviceCount; device++) {
         int atomicSupported = 0;
-    checkCudaErrors(cudaDeviceGetAttribute(
+        checkCudaErrors(cudaDeviceGetAttribute(&atomicSupported, cudaDevAttrHostNativeAtomicSupported, device));
-        &atomicSupported, cudaDevAttrHostNativeAtomicSupported, device));
         std::cout << "GPU" << device << " <-> CPU:" << std::endl;
-    std::cout << "  * Atomic Supported: " << (atomicSupported ? "yes" : "no")
+        std::cout << "  * Atomic Supported: " << (atomicSupported ? "yes" : "no") << std::endl;
-              << std::endl;
     }
 
     return 0;

Samples/2_Concepts_and_Techniques/EGLStream_CUDA_CrossGPU/README.md

@@ -37,4 +37,3 @@ Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-d
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)
-

Samples/2_Concepts_and_Techniques/EGLStream_CUDA_CrossGPU/cuda_consumer.cpp

@@ -29,12 +29,14 @@
 // DESCRIPTION:   Simple CUDA consumer rendering sample app
 //
 
-#include <cuda_runtime.h>
 #include "cuda_consumer.h"
-#include "eglstrm_common.h"
+
+#include <cuda_runtime.h>
 #include <math.h>
 #include <unistd.h>
 
+#include "eglstrm_common.h"
+
 #if defined(EXTENSION_LIST)
 EXTENSION_LIST(EXTLST_EXTERN)
 #endif
@@ -47,19 +49,22 @@ static int count_rel = 0;
 static double rel_time[25000] = {0}, total_time_rel = 0;
 
 void acquireApiStat(void);
-void acquireApiStat(void) {
+void acquireApiStat(void)
+{
     int    i   = 0;
     double min = 10000000, max = 0;
     double average_launch_time = 0, standard_deviation = 0;
-  if (count_acq == 0) return;
+    if (count_acq == 0)
+        return;
     // lets compute the standard deviation
     min = max           = acquire_time[1];
     average_launch_time = (total_time_acq - acquire_time[0]) / count_acq;
     for (i = 1; i < count_acq; i++) {
-    standard_deviation += (acquire_time[i] - average_launch_time) *
+        standard_deviation += (acquire_time[i] - average_launch_time) * (acquire_time[i] - average_launch_time);
-                          (acquire_time[i] - average_launch_time);
+        if (acquire_time[i] < min)
-    if (acquire_time[i] < min) min = acquire_time[i];
+            min = acquire_time[i];
-    if (acquire_time[i] > max) max = acquire_time[i];
+        if (acquire_time[i] > max)
+            max = acquire_time[i];
     }
     standard_deviation = sqrt(standard_deviation / count_acq);
     printf("acquire Avg: %lf\n", average_launch_time);
@@ -70,10 +75,11 @@ void acquireApiStat(void) {
     min = max           = rel_time[1];
     average_launch_time = (total_time_rel - rel_time[0]) / count_rel;
     for (i = 1; i < count_rel; i++) {
-    standard_deviation += (rel_time[i] - average_launch_time) *
+        standard_deviation += (rel_time[i] - average_launch_time) * (rel_time[i] - average_launch_time);
-                          (rel_time[i] - average_launch_time);
+        if (rel_time[i] < min)
-    if (rel_time[i] < min) min = rel_time[i];
+            min = rel_time[i];
-    if (rel_time[i] > max) max = rel_time[i];
+        if (rel_time[i] > max)
+            max = rel_time[i];
     }
     standard_deviation = sqrt(standard_deviation / count_rel);
     printf("release Avg: %lf\n", average_launch_time);
@@ -81,8 +87,8 @@ void acquireApiStat(void) {
     printf("release min: %lf\n", min);
     printf("release max: %lf\n", max);
 }
-CUresult cudaConsumerAcquireFrame(test_cuda_consumer_s *cudaConsumer,
+CUresult cudaConsumerAcquireFrame(test_cuda_consumer_s *cudaConsumer, int frameNumber)
-                                  int frameNumber) {
+{
     CUresult        cuStatus = CUDA_SUCCESS;
     CUeglFrame      cudaEgl;
     struct timespec start, end;
@@ -95,8 +101,7 @@ CUresult cudaConsumerAcquireFrame(test_cuda_consumer_s *cudaConsumer,
     }
 
     while (1) {
-    if (!eglQueryStreamKHR(cudaConsumer->eglDisplay, cudaConsumer->eglStream,
+        if (!eglQueryStreamKHR(cudaConsumer->eglDisplay, cudaConsumer->eglStream, EGL_STREAM_STATE_KHR, &streamState)) {
-                           EGL_STREAM_STATE_KHR, &streamState)) {
             printf("Cuda Consumer: eglQueryStreamKHR EGL_STREAM_STATE_KHR failed\n");
             cuStatus = CUDA_ERROR_UNKNOWN;
             goto done;
@@ -115,33 +120,35 @@ CUresult cudaConsumerAcquireFrame(test_cuda_consumer_s *cudaConsumer,
         getTime(&start);
     }
     cuStatus =
-      cuEGLStreamConsumerAcquireFrame(&(cudaConsumer->cudaConn), &cudaResource,
+        cuEGLStreamConsumerAcquireFrame(&(cudaConsumer->cudaConn), &cudaResource, &cudaConsumer->consCudaStream, 16000);
-                                      &cudaConsumer->consCudaStream, 16000);
     if (cudaConsumer->profileAPI) {
         getTime(&end);
         curTime                   = TIME_DIFF(end, start);
         acquire_time[count_acq++] = curTime;
-    if (count_acq == 25000) count_acq = 0;
+        if (count_acq == 25000)
+            count_acq = 0;
         total_time_acq += curTime;
     }
     if (cuStatus == CUDA_SUCCESS) {
         CUdeviceptr pDevPtr = 0;
         cudaError_t err;
 
-    cuStatus =
+        cuStatus = cuGraphicsResourceGetMappedEglFrame(&cudaEgl, cudaResource, 0, 0);
-        cuGraphicsResourceGetMappedEglFrame(&cudaEgl, cudaResource, 0, 0);
         if (cuStatus != CUDA_SUCCESS) {
             printf("Cuda get resource failed with %d\n", cuStatus);
             goto done;
         }
         pDevPtr = (CUdeviceptr)cudaEgl.frame.pPitch[0];
 
-    err = cudaConsumer_filter(cudaConsumer->consCudaStream, (char *)pDevPtr,
+        err = cudaConsumer_filter(cudaConsumer->consCudaStream,
-                              WIDTH * 4, HEIGHT, PROD_DATA + frameNumber,
+                                  (char *)pDevPtr,
-                              CONS_DATA + frameNumber, frameNumber);
+                                  WIDTH * 4,
+                                  HEIGHT,
+                                  PROD_DATA + frameNumber,
+                                  CONS_DATA + frameNumber,
+                                  frameNumber);
         if (err != cudaSuccess) {
-      printf("Cuda Consumer: kernel failed with: %s\n",
+            printf("Cuda Consumer: kernel failed with: %s\n", cudaGetErrorString(err));
-             cudaGetErrorString(err));
             goto done;
         }
     }
@@ -150,8 +157,8 @@ done:
     return cuStatus;
 }
 
-CUresult cudaConsumerReleaseFrame(test_cuda_consumer_s *cudaConsumer,
+CUresult cudaConsumerReleaseFrame(test_cuda_consumer_s *cudaConsumer, int frameNumber)
-                                  int frameNumber) {
+{
     CUresult        cuStatus = CUDA_SUCCESS;
     struct timespec start, end;
     double          curTime;
@@ -163,13 +170,13 @@ CUresult cudaConsumerReleaseFrame(test_cuda_consumer_s *cudaConsumer,
     if (cudaConsumer->profileAPI) {
         getTime(&start);
     }
-  cuStatus = cuEGLStreamConsumerReleaseFrame(
+    cuStatus = cuEGLStreamConsumerReleaseFrame(&cudaConsumer->cudaConn, cudaResource, &cudaConsumer->consCudaStream);
-      &cudaConsumer->cudaConn, cudaResource, &cudaConsumer->consCudaStream);
     if (cudaConsumer->profileAPI) {
         getTime(&end);
         curTime               = TIME_DIFF(end, start);
         rel_time[count_rel++] = curTime;
-    if (count_rel == 25000) count_rel = 0;
+        if (count_rel == 25000)
+            count_rel = 0;
         total_time_rel += curTime;
     }
     if (cuStatus != CUDA_SUCCESS) {
@@ -181,7 +188,8 @@ done:
     return cuStatus;
 }
 
-CUresult cudaDeviceCreateConsumer(test_cuda_consumer_s *cudaConsumer) {
+CUresult cudaDeviceCreateConsumer(test_cuda_consumer_s *cudaConsumer)
+{
     CUdevice device;
     CUresult status = CUDA_SUCCESS;
 
@@ -190,34 +198,31 @@ CUresult cudaDeviceCreateConsumer(test_cuda_consumer_s *cudaConsumer) {
         return status;
     }
 
-  if (CUDA_SUCCESS !=
+    if (CUDA_SUCCESS != (status = cuDeviceGet(&device, cudaConsumer->cudaDevId))) {
-      (status = cuDeviceGet(&device, cudaConsumer->cudaDevId))) {
         printf("failed to get CUDA device\n");
         return status;
     }
 
-  if (CUDA_SUCCESS !=
+    if (CUDA_SUCCESS != (status = cuCtxCreate(&cudaConsumer->context, 0, device))) {
-      (status = cuCtxCreate(&cudaConsumer->context, 0, device))) {
         printf("failed to create CUDA context\n");
         return status;
     }
 
     int  major = 0, minor = 0;
     char deviceName[256];
-  cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR,
+    cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, device);
-                       device);
+    cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, device);
-  cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR,
-                       device);
     cuDeviceGetName(deviceName, 256, device);
-  printf(
+    printf("CUDA Consumer on GPU Device %d: \"%s\" with compute capability "
-      "CUDA Consumer on GPU Device %d: \"%s\" with compute capability "
            "%d.%d\n\n",
-      device, deviceName, major, minor);
+           device,
+           deviceName,
+           major,
+           minor);
 
     cuCtxPopCurrent(&cudaConsumer->context);
     if (major < 6) {
-    printf(
+        printf("EGLStream_CUDA_CrossGPU requires SM 6.0 or higher arch GPU.  "
-        "EGLStream_CUDA_CrossGPU requires SM 6.0 or higher arch GPU.  "
                "Exiting...\n");
         exit(2); // EXIT_WAIVED
     }
@@ -225,8 +230,8 @@ CUresult cudaDeviceCreateConsumer(test_cuda_consumer_s *cudaConsumer) {
     return status;
 }
 
-CUresult cuda_consumer_init(test_cuda_consumer_s *cudaConsumer,
+CUresult cuda_consumer_init(test_cuda_consumer_s *cudaConsumer, TestArgs *args)
-                            TestArgs *args) {
+{
     CUresult status = CUDA_SUCCESS;
     int      bufferSize;
 
@@ -250,7 +255,8 @@ done:
     return status;
 }
 
-CUresult cuda_consumer_Deinit(test_cuda_consumer_s *cudaConsumer) {
+CUresult cuda_consumer_Deinit(test_cuda_consumer_s *cudaConsumer)
+{
     if (cudaConsumer->pCudaCopyMem) {
         free(cudaConsumer->pCudaCopyMem);
     }

Samples/2_Concepts_and_Techniques/EGLStream_CUDA_CrossGPU/cuda_consumer.h

@@ -32,15 +32,17 @@
 #ifndef _CUDA_CONSUMER_H_
 #define _CUDA_CONSUMER_H_
 
+#include <cuda.h>
+#include <cuda_runtime.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+
 #include "cudaEGL.h"
 #include "eglstrm_common.h"
-#include <cuda_runtime.h>
-#include <cuda.h>
 
-typedef struct _test_cuda_consumer_s {
+typedef struct _test_cuda_consumer_s
+{
     CUcontext             context;
     CUeglStreamConnection cudaConn;
     int                   cudaDevId;
@@ -58,8 +60,12 @@ CUresult cuda_consumer_Deinit(test_cuda_consumer_s *cudaConsumer);
 CUresult    cudaConsumerAcquireFrame(test_cuda_consumer_s *data, int frameNumber);
 CUresult    cudaConsumerReleaseFrame(test_cuda_consumer_s *data, int frameNumber);
 CUresult    cudaDeviceCreateConsumer(test_cuda_consumer_s *cudaConsumer);
-cudaError_t cudaConsumer_filter(CUstream cStream, char *pSrc, int width,
+cudaError_t cudaConsumer_filter(CUstream cStream,
-                                int height, char expectedVal, char newVal,
+                                char    *pSrc,
+                                int      width,
+                                int      height,
+                                char     expectedVal,
+                                char     newVal,
                                 int      frameNumber);
 cudaError_t cudaGetValueMismatch(void);