zara
/
cudaSamples
mirror of https://github.com/ArchaeaSoftware/cudahandbook.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
Browse Source

Fix bit rot, update to use modern idioms

nwilt-techdebt
Nicholas Wilt 4 years ago
parent
d71cd98c31
commit
f9a5490496
56 changed files with 175 additions and 126 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 85
      SMs/divergence.cu
  2. 2
      SMs/testShuffle.cu
  3. 34
      chLib/chError.h
  4. 2
      concurrency/breakevenDtoHMemcpy.cu
  5. 2
      concurrency/breakevenHtoDMemcpy.cu
  6. 4
      concurrency/breakevenKernelAsync.cu
  7. 2
      concurrency/eventRecord.cu
  8. 2
      concurrency/managedOverhead.cu
  9. 2
      concurrency/nullDtoHMemcpyAsync.cu
  10. 2
      concurrency/nullDtoHMemcpySync.cu
  11. 2
      concurrency/nullHtoDMemcpyAsync.cu
  12. 2
      concurrency/nullKernelAsync.cu
  13. 2
      concurrency/nullKernelSync.cu
  14. 2
      concurrency/pageableMemcpyHtoD.cu
  15. 2
      concurrency/pageableMemcpyHtoD16.cu
  16. 2
      concurrency/pageableMemcpyHtoD16Blocking.cu
  17. 2
      concurrency/pageableMemcpyHtoD16Broken.cu
  18. 2
      concurrency/pageableMemcpyHtoD16Synchronous.cu
  19. 2
      concurrency/peer2peerMemcpy.cu
  20. 3
      corr/normalizedCrossCorrelation.cu
  21. 3
      histogram/histogram.cu
  22. 2
      memory/globalCopy.cu
  23. 2
      memory/globalCopy2.cu
  24. 2
      memory/globalRead.cu
  25. 2
      memory/globalReadTex.cu
  26. 2
      memory/globalWrite.cu
  27. 2
      memory/spinlockReduction.cu
  28. 2
      microbench/globalRead.cu
  29. 2
      microbench/globalWrite.cu
  30. 4
      microbench/reportClocks.cu
  31. 6
      nbody/nbody.cu
  32. 18
      nbody/nbody_GPU_AOS_tiled.cuh
  33. 26
      nbody/nbody_GPU_AOS_tiled_const.cuh
  34. 8
      nbody/nbody_GPU_SOA_tiled.cuh
  35. 8
      nbody/nbody_GPU_Shuffle.cuh
  36. 4
      reduction/reduction.cu
  37. 4
      reduction/reductionTemplated.cu
  38. 10
      reduction/reductionWarpShuffle.cuh
  39. 2
      scan/int/timeScan.cu
  40. 2
      scan/streamCompact/timeStreamCompact_odd.cu
  41. 2
      scan/warp/scanWarpShuffle.cuh
  42. 2
      scan/warp/testScanWarp.cu
  43. 2
      texturing/surf2Dmemset.cu
  44. 2
      texturing/tex1d_9bit.cu
  45. 2
      texturing/tex1d_addressing.cu
  46. 2
      texturing/tex1d_unnormalized.cu
  47. 2
      texturing/tex1dfetch.cu
  48. 2
      texturing/tex1dfetch_big.cu
  49. 2
      texturing/tex1dfetch_host.cu
  50. 2
      texturing/tex1dfetch_htod.cu
  51. 2
      texturing/tex1dfetch_int2float.cu
  52. 2
      texturing/tex1dfetch_offset.cu
  53. 2
      texturing/tex2d_addressing.cu
  54. 2
      texturing/tex2d_addressing_device.cu
  55. 2
      texturing/tex2d_memset.cu
  56. 2
      texturing/tex2d_opengl.cu

85
SMs/divergence.cu
Unescape View File

@ -4,7 +4,7 @@
* *
* Microdemo to measure performance implications of conditional code. * Microdemo to measure performance implications of conditional code.
* *
* Build with: nvcc [--gpu-architecture sm_xx] [-D USE_FLOAT] [-D USE_IF_STATEMENT] divergence.cu * Build with: nvcc [--gpu-architecture sm_xx] divergence.cu
* Requires: No minimum SM requirement. * Requires: No minimum SM requirement.
* *
* Copyright (c) 2021, Archaea Software, LLC. * Copyright (c) 2021, Archaea Software, LLC.
@ -46,66 +46,79 @@
// apply blockDim and/or gridDim to n before passing in. // apply blockDim and/or gridDim to n before passing in.
// //
template<int base> template<int base>
void __device__ sumFloats( float *p, size_t N, size_t n ) void __device__ sumInts( uint32_t *p, size_t N, size_t n )
{ {
float f = base;
for ( size_t i = 0; i < N; i++ ) { for ( size_t i = 0; i < N; i++ ) {
*p += f; *p += base;
p += n; p += n;
} }
} }
typedef void(*psumFloats)(float *, size_t, size_t); typedef void(*psumInts)(uint32_t *, size_t, size_t);
__device__ psumFloats rgSumFloats[] = { __device__ psumInts rgSumInts[] = {
sumFloats< 0>, sumFloats< 1>, sumFloats< 2>, sumFloats< 3>, sumInts< 0>, sumInts< 1>, sumInts< 2>, sumInts< 3>,
sumFloats< 4>, sumFloats< 5>, sumFloats< 6>, sumFloats< 7>, sumInts< 4>, sumInts< 5>, sumInts< 6>, sumInts< 7>,
sumFloats< 8>, sumFloats< 9>, sumFloats<10>, sumFloats<11>, sumInts< 8>, sumInts< 9>, sumInts<10>, sumInts<11>,
sumFloats<12>, sumFloats<13>, sumFloats<14>, sumFloats<15>, sumInts<12>, sumInts<13>, sumInts<14>, sumInts<15>,
sumFloats<16>, sumFloats<17>, sumFloats<18>, sumFloats<19>, sumInts<16>, sumInts<17>, sumInts<18>, sumInts<19>,
sumFloats<20>, sumFloats<21>, sumFloats<22>, sumFloats<23>, sumInts<20>, sumInts<21>, sumInts<22>, sumInts<23>,
sumFloats<24>, sumFloats<25>, sumFloats<26>, sumFloats<27>, sumInts<24>, sumInts<25>, sumInts<26>, sumInts<27>,
sumFloats<28>, sumFloats<29>, sumFloats<30>, sumFloats<31> }; sumInts<28>, sumInts<29>, sumInts<30>, sumInts<31> };
template<uint32_t sh>
__global__ void __global__ void
sumFloats_bywarp( float *p, size_t N ) sumInts_bythread( uint32_t *p, size_t N )
{ {
uint32_t warpid = threadIdx.x>>5; uint32_t warpish_id = threadIdx.x>>sh;
N /= blockDim.x*gridDim.x; N /= blockDim.x*gridDim.x;
rgSumFloats[warpid]( p+threadIdx.x+blockIdx.x*blockDim.x, N, blockDim.x*gridDim.x ); rgSumInts[warpish_id&31]( p+threadIdx.x+blockIdx.x*blockDim.x, N, blockDim.x*gridDim.x );
} }
__global__ void template<uint32_t sh>
sumFloats_bythread( float *p, size_t N ) static double
{ timeByThreads( uint32_t *p, size_t N )
}
int
main()
{ {
cudaError_t status; cudaError_t status;
size_t N = 1024*1024*1024UL; float elapsed_time;
float *p = 0; double ret = 0.0;
float et;
cudaEvent_t start = 0, stop = 0; cudaEvent_t start = 0, stop = 0;
cuda(Malloc( (void **) &p, N*sizeof(float)) );
cuda(Memset( p, 0, N*sizeof(float)) );
cuda(EventCreate( &start )); cuda(EventCreate( &start ));
cuda(EventCreate( &stop )); cuda(EventCreate( &stop ));
cuda(EventRecord( start )); cuda(EventRecord( start ));
sumFloats_bywarp<<<3072,256>>>( p, N ); sumInts_bythread<sh><<<3072,1024>>>( p, N );
cuda(EventRecord( stop )); cuda(EventRecord( stop ));
cuda(DeviceSynchronize()); cuda(DeviceSynchronize());
cuda(EventElapsedTime( &et, start, stop )); cuda(EventElapsedTime( &elapsed_time, start, stop ));
ret = N*1000.0/elapsed_time/1e9;
printf( "%.2f ms = %.2f Gops/s\n", et, (double) N*1000.0/et/1e9 ); printf( "%2d threads: %f Gops/s\n", 1<<sh, ret );
Error:
cudaFree( p );
cudaEventDestroy( stop ); cudaEventDestroy( stop );
cudaEventDestroy( start ); cudaEventDestroy( start );
return ret;
}
int
main()
{
cudaError_t status;
size_t N = 1024*1024*1024UL;
uint32_t *p = 0;
cuda(Malloc( (void **) &p, N*sizeof(uint32_t)) );
cuda(Memset( p, 0, N*sizeof(uint32_t)) );
timeByThreads<6>( p, N );
timeByThreads<5>( p, N );
timeByThreads<4>( p, N );
timeByThreads<3>( p, N );
timeByThreads<2>( p, N );
timeByThreads<1>( p, N );
timeByThreads<0>( p, N );
cudaFree( p );
return 0; return 0;
Error: Error:
return 1; return 1;

2
SMs/testShuffle.cu
Unescape View File

@ -48,7 +48,7 @@ TestShuffle( int *out, const int *in, size_t N )
size_t i = blockIdx.x*blockDim.x+threadIdx.x; size_t i = blockIdx.x*blockDim.x+threadIdx.x;
int value = (int) i;//in[i]; int value = (int) i;//in[i];
out[i] = __shfl_up( value, 1 ); out[i] = __shfl_up_sync( 0xffffffff, value, 1 );
} }
cudaError_t cudaError_t

34
chLib/chError.h
Unescape View File

@ -112,6 +112,40 @@ chGetErrorString( CUresult status )
ErrorValue(CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED) ErrorValue(CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED)
ErrorValue(CUDA_ERROR_HOST_MEMORY_NOT_REGISTERED) ErrorValue(CUDA_ERROR_HOST_MEMORY_NOT_REGISTERED)
#endif #endif
ErrorValue(CUDA_ERROR_STUB_LIBRARY)
ErrorValue(CUDA_ERROR_PEER_ACCESS_UNSUPPORTED)
ErrorValue(CUDA_ERROR_DEVICE_NOT_LICENSED)
ErrorValue(CUDA_ERROR_INVALID_PTX)
ErrorValue(CUDA_ERROR_INVALID_GRAPHICS_CONTEXT)
ErrorValue(CUDA_ERROR_NVLINK_UNCORRECTABLE)
ErrorValue(CUDA_ERROR_JIT_COMPILER_NOT_FOUND)
ErrorValue(CUDA_ERROR_JIT_COMPILATION_DISABLED)
ErrorValue(CUDA_ERROR_UNSUPPORTED_PTX_VERSION)
ErrorValue(CUDA_ERROR_ILLEGAL_STATE)
ErrorValue(CUDA_ERROR_ILLEGAL_ADDRESS)
ErrorValue(CUDA_ERROR_HARDWARE_STACK_ERROR)
ErrorValue(CUDA_ERROR_ILLEGAL_INSTRUCTION)
ErrorValue(CUDA_ERROR_MISALIGNED_ADDRESS)
ErrorValue(CUDA_ERROR_INVALID_ADDRESS_SPACE)
ErrorValue(CUDA_ERROR_INVALID_PC)
ErrorValue(CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE)
ErrorValue(CUDA_ERROR_NOT_PERMITTED)
ErrorValue(CUDA_ERROR_NOT_SUPPORTED)
ErrorValue(CUDA_ERROR_SYSTEM_NOT_READY)
ErrorValue(CUDA_ERROR_SYSTEM_DRIVER_MISMATCH)
ErrorValue(CUDA_ERROR_COMPAT_NOT_SUPPORTED_ON_DEVICE)
ErrorValue(CUDA_ERROR_STREAM_CAPTURE_UNSUPPORTED)
ErrorValue(CUDA_ERROR_STREAM_CAPTURE_INVALIDATED)
ErrorValue(CUDA_ERROR_STREAM_CAPTURE_MERGE)
ErrorValue(CUDA_ERROR_STREAM_CAPTURE_UNMATCHED)
ErrorValue(CUDA_ERROR_STREAM_CAPTURE_UNJOINED)
ErrorValue(CUDA_ERROR_STREAM_CAPTURE_ISOLATION)
ErrorValue(CUDA_ERROR_STREAM_CAPTURE_IMPLICIT)
ErrorValue(CUDA_ERROR_STREAM_CAPTURE_WRONG_THREAD)
ErrorValue(CUDA_ERROR_TIMEOUT)
ErrorValue(CUDA_ERROR_GRAPH_EXEC_UPDATE_FAILURE)
ErrorValue(CUDA_ERROR_CAPTURED_EVENT)
ErrorValue(CUDA_ERROR_UNKNOWN) ErrorValue(CUDA_ERROR_UNKNOWN)
} }
return "chGetErrorString - unknown error value"; return "chGetErrorString - unknown error value";

2
concurrency/breakevenDtoHMemcpy.cu
Unescape View File

@ -70,7 +70,7 @@ main( int argc, char *argv[] )
cuda(MemcpyAsync( hostInt, deviceInt, byteCount, cuda(MemcpyAsync( hostInt, deviceInt, byteCount,
cudaMemcpyDeviceToHost, NULL ) ); cudaMemcpyDeviceToHost, NULL ) );
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

2
concurrency/breakevenHtoDMemcpy.cu
Unescape View File

@ -70,7 +70,7 @@ main( int argc, char *argv[] )
cuda(MemcpyAsync( deviceInt, hostInt, byteCount, cuda(MemcpyAsync( deviceInt, hostInt, byteCount,
cudaMemcpyHostToDevice, NULL ) ); cudaMemcpyHostToDevice, NULL ) );
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

4
concurrency/breakevenKernelAsync.cu
Unescape View File

@ -67,7 +67,7 @@ main( int argc, char *argv[] )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
WaitKernel<<<1,1>>>( 0, false ); WaitKernel<<<1,1>>>( 0, false );
} }
cudaThreadSynchronize(); cudaDeviceSynchronize();
printf("Cycles\tus\n" ); printf("Cycles\tus\n" );
for ( int cycles = 0; cycles < 2500; cycles += 100 ) { for ( int cycles = 0; cycles < 2500; cycles += 100 ) {
@ -76,7 +76,7 @@ main( int argc, char *argv[] )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
WaitKernel<<<1,1>>>( cycles, false ); WaitKernel<<<1,1>>>( cycles, false );
} }
cudaThreadSynchronize(); cudaDeviceSynchronize();
chTimerGetTime( &stop ); chTimerGetTime( &stop );
double microseconds = 1e6*chTimerElapsedTime( &start, &stop ); double microseconds = 1e6*chTimerElapsedTime( &start, &stop );
double usPerLaunch = microseconds / (float) cIterations; double usPerLaunch = microseconds / (float) cIterations;

2
concurrency/eventRecord.cu
Unescape View File

@ -72,7 +72,7 @@ usPerLaunch( int cIterations, int cEvents )
cuda(EventRecord( events[j], NULL ) ); cuda(EventRecord( events[j], NULL ) );
} }
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
microseconds = 1e6*chTimerElapsedTime( &start, &stop ); microseconds = 1e6*chTimerElapsedTime( &start, &stop );

2
concurrency/managedOverhead.cu
Unescape View File

@ -73,7 +73,7 @@ usPerLaunch( int cIterations, size_t cPages=0 )
chTimerGetTime( &start ); chTimerGetTime( &start );
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
NullKernel<<<1,1>>>(); NullKernel<<<1,1>>>();
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
if ( bTouch && 0 != p ) { if ( bTouch && 0 != p ) {
for ( int iPage = 0; iPage < cPages; iPage++ ) { for ( int iPage = 0; iPage < cPages; iPage++ ) {
((volatile unsigned char *) p)[iPage*pageSize] |= 1; ((volatile unsigned char *) p)[iPage*pageSize] |= 1;

2
concurrency/nullDtoHMemcpyAsync.cu
Unescape View File

@ -60,7 +60,7 @@ main( int argc, char *argv[] )
cuda(MemcpyAsync( hostInt, deviceInt, sizeof(int), cuda(MemcpyAsync( hostInt, deviceInt, sizeof(int),
cudaMemcpyDeviceToHost, NULL ) ); cudaMemcpyDeviceToHost, NULL ) );
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

2
concurrency/nullDtoHMemcpySync.cu
Unescape View File

@ -61,7 +61,7 @@ main( int argc, char *argv[] )
cuda(Memcpy( hostInt, deviceInt, sizeof(int), cuda(Memcpy( hostInt, deviceInt, sizeof(int),
cudaMemcpyDeviceToHost ) ); cudaMemcpyDeviceToHost ) );
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
cIterations *= 2; cIterations *= 2;
} while ( chTimerElapsedTime( &start, &stop ) < 0.5f ) ; } while ( chTimerElapsedTime( &start, &stop ) < 0.5f ) ;

2
concurrency/nullHtoDMemcpyAsync.cu
Unescape View File

@ -60,7 +60,7 @@ main( int argc, char *argv[] )
cuda(MemcpyAsync( deviceInt, hostInt, sizeof(int), cuda(MemcpyAsync( deviceInt, hostInt, sizeof(int),
cudaMemcpyHostToDevice, NULL ) ); cudaMemcpyHostToDevice, NULL ) );
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

2
concurrency/nullKernelAsync.cu
Unescape View File

@ -60,7 +60,7 @@ usPerLaunch( int cIterations )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
NullKernel<<<1,1>>>(); NullKernel<<<1,1>>>();
} }
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
chTimerGetTime( &stop ); chTimerGetTime( &stop );
microseconds = 1e6*chTimerElapsedTime( &start, &stop ); microseconds = 1e6*chTimerElapsedTime( &start, &stop );

2
concurrency/nullKernelSync.cu
Unescape View File

@ -59,7 +59,7 @@ usPerLaunch( int cIterations )
chTimerGetTime( &start ); chTimerGetTime( &start );
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
NullKernel<<<1,1>>>(); NullKernel<<<1,1>>>();
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
} }
chTimerGetTime( &stop ); chTimerGetTime( &stop );

2
concurrency/pageableMemcpyHtoD.cu
Unescape View File

@ -145,7 +145,7 @@ main( int argc, char *argv[] )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ; chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ;
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

2
concurrency/pageableMemcpyHtoD16.cu
Unescape View File

@ -157,7 +157,7 @@ main( int argc, char *argv[] )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ; chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ;
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

2
concurrency/pageableMemcpyHtoD16Blocking.cu
Unescape View File

@ -156,7 +156,7 @@ main( int argc, char *argv[] )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ; chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ;
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

2
concurrency/pageableMemcpyHtoD16Broken.cu
Unescape View File

@ -157,7 +157,7 @@ main( int argc, char *argv[] )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ; chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ;
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

2
concurrency/pageableMemcpyHtoD16Synchronous.cu
Unescape View File

@ -157,7 +157,7 @@ main( int argc, char *argv[] )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ; chMemcpyHtoD( deviceInt, testVector, numInts*sizeof(int) ) ;
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

2
concurrency/peer2peerMemcpy.cu
Unescape View File

@ -195,7 +195,7 @@ main( int argc, char *argv[] )
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
chMemcpyPeerToPeer( deviceInt[0], 0, deviceInt[1], 1, numInts*sizeof(int) ) ; chMemcpyPeerToPeer( deviceInt[0], 0, deviceInt[1], 1, numInts*sizeof(int) ) ;
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
{ {

3
corr/normalizedCrossCorrelation.cu
Unescape View File

@ -476,7 +476,8 @@ main(int argc, char *argv[])
int sharedPitch; int sharedPitch;
int sharedMem; int sharedMem;
char *inputFilename = "coins.pgm"; char defaultInputFilename[] = "coins.pgm";
char *inputFilename = defaultInputFilename;
char *outputFilename = NULL; char *outputFilename = NULL;
cudaArray *pArrayImage = NULL; cudaArray *pArrayImage = NULL;

3
histogram/histogram.cu
Unescape View File

@ -261,7 +261,8 @@ main(int argc, char *argv[])
dim3 threads; dim3 threads;
char *inputFilename = "coins.pgm"; char defaultInputFilename[] = "coins.pgm";
char *inputFilename = defaultInputFilename;
char *outputFilename = NULL; char *outputFilename = NULL;
cudaArray *pArrayImage = NULL; cudaArray *pArrayImage = NULL;

2
memory/globalCopy.cu
Unescape View File

@ -114,7 +114,7 @@ BandwidthCopy( T *deviceOut, T *deviceIn,
GlobalCopy<T,n><<<cBlocks,cThreads>>>( deviceOut+bOffsetDst, deviceIn+bOffsetSrc, N-bOffsetDst-bOffsetSrc ); GlobalCopy<T,n><<<cBlocks,cThreads>>>( deviceOut+bOffsetDst, deviceIn+bOffsetSrc, N-bOffsetDst-bOffsetSrc );
} }
cudaEventRecord( evStop ); cudaEventRecord( evStop );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
// make configurations that cannot launch error-out with 0 bandwidth // make configurations that cannot launch error-out with 0 bandwidth
cuda(GetLastError() ); cuda(GetLastError() );
cuda(EventElapsedTime( &ms, evStart, evStop ) ); cuda(EventElapsedTime( &ms, evStart, evStop ) );

2
memory/globalCopy2.cu
Unescape View File

@ -160,7 +160,7 @@ BandwidthCopy( T *deviceOut, T *deviceIn0, T *deviceIn1,
GlobalCopy<T,n><<<cBlocks,cThreads>>>( deviceOut+bOffsetDst, deviceIn0+bOffsetSrc, deviceIn1+bOffsetSrc, N-bOffsetDst-bOffsetSrc ); GlobalCopy<T,n><<<cBlocks,cThreads>>>( deviceOut+bOffsetDst, deviceIn0+bOffsetSrc, deviceIn1+bOffsetSrc, N-bOffsetDst-bOffsetSrc );
} }
cudaEventRecord( evStop ); cudaEventRecord( evStop );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
// make configurations that cannot launch error-out with 0 bandwidth // make configurations that cannot launch error-out with 0 bandwidth
cuda(GetLastError() ); cuda(GetLastError() );
cuda(EventElapsedTime( &ms, evStart, evStop ) ); cuda(EventElapsedTime( &ms, evStart, evStop ) );

2
memory/globalRead.cu
Unescape View File

@ -179,7 +179,7 @@ BandwidthReads( size_t N, int cBlocks, int cThreads )
GlobalReads<T,n><<<cBlocks,cThreads>>>( out, in+bOffset, N-bOffset, false ); GlobalReads<T,n><<<cBlocks,cThreads>>>( out, in+bOffset, N-bOffset, false );
} }
cudaEventRecord( evStop ); cudaEventRecord( evStop );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
// make configurations that cannot launch error-out with 0 bandwidth // make configurations that cannot launch error-out with 0 bandwidth
cuda(GetLastError() ); cuda(GetLastError() );
cuda(EventElapsedTime( &ms, evStart, evStop ) ); cuda(EventElapsedTime( &ms, evStart, evStop ) );

2
memory/globalReadTex.cu
Unescape View File

@ -236,7 +236,7 @@ BandwidthReads( size_t N, int cBlocks, int cThreads )
GlobalReads<T,n><<<cBlocks,cThreads>>>( out, bOffset, N-bOffset, false ); GlobalReads<T,n><<<cBlocks,cThreads>>>( out, bOffset, N-bOffset, false );
} }
cudaEventRecord( evStop ); cudaEventRecord( evStop );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
// make configurations that cannot launch error-out with 0 bandwidth // make configurations that cannot launch error-out with 0 bandwidth
cuda(GetLastError() ); cuda(GetLastError() );
cuda(EventElapsedTime( &ms, evStart, evStop ) ); cuda(EventElapsedTime( &ms, evStart, evStop ) );

2
memory/globalWrite.cu
Unescape View File

@ -131,7 +131,7 @@ BandwidthWrites( size_t N, int cBlocks, int cThreads )
GlobalWrites<T,n><<<cBlocks,cThreads>>>( out+bOffset, (T) 0xcc, N-bOffset ); GlobalWrites<T,n><<<cBlocks,cThreads>>>( out+bOffset, (T) 0xcc, N-bOffset );
} }
cudaEventRecord( evStop ); cudaEventRecord( evStop );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
// make configurations that cannot launch error-out with 0 bandwidth // make configurations that cannot launch error-out with 0 bandwidth
cuda(GetLastError() ); cuda(GetLastError() );
cuda(EventElapsedTime( &ms, evStart, evStop ) ); cuda(EventElapsedTime( &ms, evStart, evStop ) );

2
memory/spinlockReduction.cu
Unescape View File

@ -208,7 +208,7 @@ AtomicsPerSecond( size_t N, int cBlocks, int cThreads )
} }
cudaEventRecord( evStop ); cudaEventRecord( evStop );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
// make configurations that cannot launch error-out with 0 bandwidth // make configurations that cannot launch error-out with 0 bandwidth
cuda(GetLastError() ); cuda(GetLastError() );

2
microbench/globalRead.cu
Unescape View File

@ -179,7 +179,7 @@ BandwidthReads( size_t N, int cBlocks, int cThreads )
GlobalReads<T,n><<<cBlocks,cThreads>>>( out, in+bOffset, N-bOffset, false ); GlobalReads<T,n><<<cBlocks,cThreads>>>( out, in+bOffset, N-bOffset, false );
} }
cudaEventRecord( evStop ); cudaEventRecord( evStop );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
// make configurations that cannot launch error-out with 0 bandwidth // make configurations that cannot launch error-out with 0 bandwidth
cuda(GetLastError() ); cuda(GetLastError() );
cuda(EventElapsedTime( &ms, evStart, evStop ) ); cuda(EventElapsedTime( &ms, evStart, evStop ) );

2
microbench/globalWrite.cu
Unescape View File

@ -131,7 +131,7 @@ BandwidthWrites( size_t N, int cBlocks, int cThreads )
GlobalWrites<T,n><<<cBlocks,cThreads>>>( out+bOffset, (T) 0xcc, N-bOffset ); GlobalWrites<T,n><<<cBlocks,cThreads>>>( out+bOffset, (T) 0xcc, N-bOffset );
} }
cudaEventRecord( evStop ); cudaEventRecord( evStop );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
// make configurations that cannot launch error-out with 0 bandwidth // make configurations that cannot launch error-out with 0 bandwidth
cuda(GetLastError() ); cuda(GetLastError() );
cuda(EventElapsedTime( &ms, evStart, evStop ) ); cuda(EventElapsedTime( &ms, evStart, evStop ) );

4
microbench/reportClocks.cu
Unescape View File

@ -97,13 +97,13 @@ ReportTimesAndIDs( FILE *clocksFile, FILE *tidsFile, dim3 gridSize, dim3 blockSi
cuda(EventCreate( &stop ) ); cuda(EventCreate( &stop ) );
WriteClockValues<<<gridSize, blockSize>>>( deviceClockValues, deviceThreadIDs ); WriteClockValues<<<gridSize, blockSize>>>( deviceClockValues, deviceThreadIDs );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
cuda(EventRecord( start, 0 ) ); cuda(EventRecord( start, 0 ) );
WriteClockValues<<<gridSize, blockSize>>>( deviceClockValues, deviceThreadIDs ); WriteClockValues<<<gridSize, blockSize>>>( deviceClockValues, deviceThreadIDs );
cuda(EventRecord( stop, 0 ) ); cuda(EventRecord( stop, 0 ) );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
{ {
float ms; float ms;

6
nbody/nbody.cu
Unescape View File

@ -7,8 +7,8 @@
* parallelizable, with lots of FLOPS per unit of external * parallelizable, with lots of FLOPS per unit of external
* memory bandwidth required. * memory bandwidth required.
* *
* Build with: nvcc -I ../chLib <options> nbody.cu nbody_CPU_SSE.cpp nbody_CPU_SSE_threaded.cpp nbody_GPU_shared.cu nbody_multiGPU.cu nbody_multiGPU_threaded.cu * Build with: nvcc -I ../chLib nbody.cu nbody_CPU_SSE.cpp nbody_CPU_AOS.cpp nbody_CPU_AOS_tiled.cpp nbody_CPU_SSE_threaded.cpp nbody_CPU_SOA.cpp nbody_GPU_shared.cu nbody_multiGPU.cu nbody_multiGPU_threaded.cu
* On Linux: nvcc -I ../chLib <options> nbody.cu nbody_CPU_SSE.cpp nbody_CPU_SSE_threaded.cpp nbody_GPU_shared.cu nbody_multiGPU.cu nbody_multiGPU_threaded.cu -lpthread -lrt * On Linux: nvcc -I ../chLib nbody.cu nbody_CPU_SSE.cpp nbody_CPU_AOS.cpp nbody_CPU_AOS_tiled.cpp nbody_CPU_SSE_threaded.cpp nbody_CPU_SOA.cpp nbody_GPU_shared.cu nbody_multiGPU.cu nbody_multiGPU_threaded.cu -lcudart_static -ldl -lrt
* Requires: No minimum SM requirement. If SM 3.x is not available, * Requires: No minimum SM requirement. If SM 3.x is not available,
* this application quietly replaces the shuffle and fast-atomic * this application quietly replaces the shuffle and fast-atomic
* implementations with the shared memory implementation. * implementations with the shared memory implementation.
@ -482,7 +482,7 @@ ComputeGravitation(
sumY += g_hostAOS_Force[i*3+1]; sumY += g_hostAOS_Force[i*3+1];
sumZ += g_hostAOS_Force[i*3+2]; sumZ += g_hostAOS_Force[i*3+2];
} }
*maxRelError = max( fabs(sumX), max(fabs(sumY), fabs(sumZ)) ); *maxRelError = std::max( fabs(sumX), std::max(fabs(sumY), fabs(sumZ)) );
if ( g_ZeroThreshold != 0.0 && if ( g_ZeroThreshold != 0.0 &&
fabs( *maxRelError ) > g_ZeroThreshold ) { fabs( *maxRelError ) > g_ZeroThreshold ) {
printf( "Maximum sum of forces > threshold (%E > %E)\n", printf( "Maximum sum of forces > threshold (%E > %E)\n",

18
nbody/nbody_GPU_AOS_tiled.cuh
Unescape View File

@ -78,11 +78,11 @@ inline float
__device__ __device__
warpReduce( float x ) warpReduce( float x )
{ {
x += __int_as_float( __shfl_xor( __float_as_int(x), 16 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 16 ) );
x += __int_as_float( __shfl_xor( __float_as_int(x), 8 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 8 ) );
x += __int_as_float( __shfl_xor( __float_as_int(x), 4 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 4 ) );
x += __int_as_float( __shfl_xor( __float_as_int(x), 2 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 2 ) );
x += __int_as_float( __shfl_xor( __float_as_int(x), 1 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 1 ) );
return x; return x;
} }
@ -111,10 +111,10 @@ DoNondiagonalTile_GPU(
float fx, fy, fz; float fx, fy, fz;
float4 bodyPosMass; float4 bodyPosMass;
bodyPosMass.x = __shfl( shufSrcPosMass.x, _j ); bodyPosMass.x = __shfl_sync( 0xffffffff, shufSrcPosMass.x, _j );
bodyPosMass.y = __shfl( shufSrcPosMass.y, _j ); bodyPosMass.y = __shfl_sync( 0xffffffff, shufSrcPosMass.y, _j );
bodyPosMass.z = __shfl( shufSrcPosMass.z, _j ); bodyPosMass.z = __shfl_sync( 0xffffffff, shufSrcPosMass.z, _j );
bodyPosMass.w = __shfl( shufSrcPosMass.w, _j ); bodyPosMass.w = __shfl_sync( 0xffffffff, shufSrcPosMass.w, _j );
bodyBodyInteraction<float>( bodyBodyInteraction<float>(
&fx, &fy, &fz, &fx, &fy, &fz,

26
nbody/nbody_GPU_AOS_tiled_const.cuh
Unescape View File

@ -82,11 +82,11 @@ __device__
warpReduce_const( float x ) warpReduce_const( float x )
{ {
#if __CUDA_ARCH__ && __CUDA_ARCH__ > 300 #if __CUDA_ARCH__ && __CUDA_ARCH__ > 300
x += __int_as_float( __shfl_xor( __float_as_int(x), 16 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 16 ) );
x += __int_as_float( __shfl_xor( __float_as_int(x), 8 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 8 ) );
x += __int_as_float( __shfl_xor( __float_as_int(x), 4 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 4 ) );
x += __int_as_float( __shfl_xor( __float_as_int(x), 2 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 2 ) );
x += __int_as_float( __shfl_xor( __float_as_int(x), 1 ) ); x += __int_as_float( __shfl_xor_sync( 0xffffffff, __float_as_int(x), 1 ) );
#endif #endif
return x; return x;
} }
@ -117,10 +117,10 @@ DoNondiagonalTile_GPU_const(
float fx, fy, fz; float fx, fy, fz;
float4 bodyPosMass; float4 bodyPosMass;
bodyPosMass.x = __shfl( shufSrcPosMass.x, _j ); bodyPosMass.x = __shfl_sync( 0xffffffff, shufSrcPosMass.x, _j );
bodyPosMass.y = __shfl( shufSrcPosMass.y, _j ); bodyPosMass.y = __shfl_sync( 0xffffffff, shufSrcPosMass.y, _j );
bodyPosMass.z = __shfl( shufSrcPosMass.z, _j ); bodyPosMass.z = __shfl_sync( 0xffffffff, shufSrcPosMass.z, _j );
bodyPosMass.w = __shfl( shufSrcPosMass.w, _j ); bodyPosMass.w = __shfl_sync( 0xffffffff, shufSrcPosMass.w, _j );
bodyBodyInteraction<float>( bodyBodyInteraction<float>(
&fx, &fy, &fz, &fx, &fy, &fz,
@ -200,10 +200,10 @@ DoNondiagonalTile_GPU_const(
float fx, fy, fz; float fx, fy, fz;
float4 bodyPosMass; float4 bodyPosMass;
bodyPosMass.x = __shfl( shufSrcPosMass.x, _j ); bodyPosMass.x = __shfl_sync( 0xffffffff, shufSrcPosMass.x, _j );
bodyPosMass.y = __shfl( shufSrcPosMass.y, _j ); bodyPosMass.y = __shfl_sync( 0xffffffff, shufSrcPosMass.y, _j );
bodyPosMass.z = __shfl( shufSrcPosMass.z, _j ); bodyPosMass.z = __shfl_sync( 0xffffffff, shufSrcPosMass.z, _j );
bodyPosMass.w = __shfl( shufSrcPosMass.w, _j ); bodyPosMass.w = __shfl_sync( 0xffffffff, shufSrcPosMass.w, _j );
bodyBodyInteraction<float>( bodyBodyInteraction<float>(
&fx, &fy, &fz, &fx, &fy, &fz,

8
nbody/nbody_GPU_SOA_tiled.cuh
Unescape View File

@ -98,10 +98,10 @@ DoNondiagonalTile_GPU_SOA(
float fx, fy, fz; float fx, fy, fz;
float4 bodyPosMass; float4 bodyPosMass;
bodyPosMass.x = __shfl( shufSrcPosMass.x, _j ); bodyPosMass.x = __shfl_sync( 0xffffffff, shufSrcPosMass.x, _j );
bodyPosMass.y = __shfl( shufSrcPosMass.y, _j ); bodyPosMass.y = __shfl_sync( 0xffffffff, shufSrcPosMass.y, _j );
bodyPosMass.z = __shfl( shufSrcPosMass.z, _j ); bodyPosMass.z = __shfl_sync( 0xffffffff, shufSrcPosMass.z, _j );
bodyPosMass.w = __shfl( shufSrcPosMass.w, _j ); bodyPosMass.w = __shfl_sync( 0xffffffff, shufSrcPosMass.w, _j );
bodyBodyInteraction<float>( bodyBodyInteraction<float>(
&fx, &fy, &fz, &fx, &fy, &fz,

8
nbody/nbody_GPU_Shuffle.cuh
Unescape View File

@ -56,10 +56,10 @@ ComputeNBodyGravitation_Shuffle(
float fx, fy, fz; float fx, fy, fz;
float4 shufDstPosMass; float4 shufDstPosMass;
shufDstPosMass.x = __shfl( shufSrcPosMass.x, k ); shufDstPosMass.x = __shfl_sync( 0xffffffff, shufSrcPosMass.x, k );
shufDstPosMass.y = __shfl( shufSrcPosMass.y, k ); shufDstPosMass.y = __shfl_sync( 0xffffffff, shufSrcPosMass.y, k );
shufDstPosMass.z = __shfl( shufSrcPosMass.z, k ); shufDstPosMass.z = __shfl_sync( 0xffffffff, shufSrcPosMass.z, k );
shufDstPosMass.w = __shfl( shufSrcPosMass.w, k ); shufDstPosMass.w = __shfl_sync( 0xffffffff, shufSrcPosMass.w, k );
bodyBodyInteraction( bodyBodyInteraction(
&fx, &fy, &fz, &fx, &fy, &fz,

4
reduction/reduction.cu
Unescape View File

@ -85,7 +85,7 @@ TimedReduction(
cuda(Malloc( &partialSums, cBlocks*sizeof(int) ) ); cuda(Malloc( &partialSums, cBlocks*sizeof(int) ) );
cuda(EventCreate( &start ) ); cuda(EventCreate( &start ) );
cuda(EventCreate( &stop ) ); cuda(EventCreate( &stop ) );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
cuda(EventRecord( start, 0 ) ); cuda(EventRecord( start, 0 ) );
hostReduction( hostReduction(
@ -165,7 +165,7 @@ usPerInvocation( int cIterations, size_t N,
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
pfnReduction( partialSums, partialSums, smallArray, N, 1, 256 ); pfnReduction( partialSums, partialSums, smallArray, N, 1, 256 );
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
ret = chTimerElapsedTime( &start, &stop ); ret = chTimerElapsedTime( &start, &stop );
ret = (ret / (double) cIterations) * 1e6; ret = (ret / (double) cIterations) * 1e6;

4
reduction/reductionTemplated.cu
Unescape View File

@ -129,7 +129,7 @@ TimedReduction(
cuda(Malloc( &partialSums, cBlocks*sizeof(ReductionType) ) ); cuda(Malloc( &partialSums, cBlocks*sizeof(ReductionType) ) );
cuda(EventCreate( &start ) ); cuda(EventCreate( &start ) );
cuda(EventCreate( &stop ) ); cuda(EventCreate( &stop ) );
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
cuda(EventRecord( start, 0 ) ); cuda(EventRecord( start, 0 ) );
hostReduction( deviceAnswer, partialSums, deviceIn, N, cBlocks, cThreads ); hostReduction( deviceAnswer, partialSums, deviceIn, N, cBlocks, cThreads );
@ -209,7 +209,7 @@ usPerInvocation( int cIterations, size_t N,
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
pfnReduction( partialSums, partialSums, smallArray, N, 1, 256 ); pfnReduction( partialSums, partialSums, smallArray, N, 1, 256 );
} }
cuda(ThreadSynchronize() ); cuda(DeviceSynchronize() );
chTimerGetTime( &stop ); chTimerGetTime( &stop );
ret = chTimerElapsedTime( &start, &stop ); ret = chTimerElapsedTime( &start, &stop );
ret = (ret / (double) cIterations) * 1e6; ret = (ret / (double) cIterations) * 1e6;

10
reduction/reductionWarpShuffle.cuh
Unescape View File

@ -61,11 +61,11 @@ Reduction2_kernel( int *out, const int *in, size_t N )
volatile int *wsSum = sPartials; volatile int *wsSum = sPartials;
if ( blockDim.x > 32 ) wsSum[tid] += wsSum[tid + 32]; if ( blockDim.x > 32 ) wsSum[tid] += wsSum[tid + 32];
int mySum = wsSum[tid]; int mySum = wsSum[tid];
mySum += __shfl_xor( mySum, 16 ); mySum += __shfl_xor_sync( 0xffffffff, mySum, 16 );
mySum += __shfl_xor( mySum, 8 ); mySum += __shfl_xor_sync( 0xffffffff, mySum, 8 );
mySum += __shfl_xor( mySum, 4 ); mySum += __shfl_xor_sync( 0xffffffff, mySum, 4 );
mySum += __shfl_xor( mySum, 2 ); mySum += __shfl_xor_sync( 0xffffffff, mySum, 2 );
mySum += __shfl_xor( mySum, 1 ); mySum += __shfl_xor_sync( 0xffffffff, mySum, 1 );
/* wsSum[tid] += wsSum[tid + 16]; /* wsSum[tid] += wsSum[tid + 16];
wsSum[tid] += wsSum[tid + 8]; wsSum[tid] += wsSum[tid + 8];
wsSum[tid] += wsSum[tid + 4]; wsSum[tid] += wsSum[tid + 4];

2
scan/int/timeScan.cu
Unescape View File

@ -96,7 +96,7 @@ TimeScan( void (*pfnScanGPU)(T *, const T *, size_t, int),
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
pfnScanGPU( outGPU, inGPU, N, numThreads ); pfnScanGPU( outGPU, inGPU, N, numThreads );
} }
if ( cudaSuccess != cudaThreadSynchronize() ) if ( cudaSuccess != cudaDeviceSynchronize() )
goto Error; goto Error;
chTimerGetTime( &stop ); chTimerGetTime( &stop );

2
scan/streamCompact/timeStreamCompact_odd.cu
Unescape View File

@ -122,7 +122,7 @@ TimeStreamCompact(
for ( int i = 0; i < cIterations; i++ ) { for ( int i = 0; i < cIterations; i++ ) {
pfnScanGPU( outGPU, deviceTotal, inGPU, N, numThreads ); pfnScanGPU( outGPU, deviceTotal, inGPU, N, numThreads );
} }
if ( cudaSuccess != cudaThreadSynchronize() ) if ( cudaSuccess != cudaDeviceSynchronize() )
goto Error; goto Error;
chTimerGetTime( &stop ); chTimerGetTime( &stop );

2
scan/warp/scanWarpShuffle.cuh
Unescape View File

@ -69,7 +69,7 @@ exclusive_scan_warp_shfl(int mysum)
{ {
const unsigned int lane = threadIdx.x & 31; const unsigned int lane = threadIdx.x & 31;
for(int i = 0; i < levels; ++i) for(int i = 0; i < levels; ++i)
mysum = shfl_scan_add_step(mysum, 1 << i); mysum = scanWarpShuffle_step( mysum, 1 << i);
mysum = __shfl_up(mysum, 1); mysum = __shfl_up(mysum, 1);
return (lane) ? mysum : 0; return (lane) ? mysum : 0;
} }

2
scan/warp/testScanWarp.cu
Unescape View File

@ -63,7 +63,7 @@ enum ScanType {
#include "scanReduceThenScan.cuh" #include "scanReduceThenScan.cuh"
#include "scanReduceThenScan_0.cuh" #include "scanReduceThenScan_0.cuh"
#include "scan2Level.cuh" #include "scan2Level.cuh"
#include "ScanThrust.cuh" #include "scanThrust.cuh"
template<int period> template<int period>
void void

2
texturing/surf2Dmemset.cu
Unescape View File

@ -184,7 +184,7 @@ CreateAndPrintTex(
blocks.y = 1; blocks.y = 1;
threads.x = 64; threads.y = 4; threads.x = 64; threads.y = 4;
TexReadout<<<blocks,threads>>>( outDevice, outWidth, outPitch, outHeight, base, increment ); TexReadout<<<blocks,threads>>>( outDevice, outWidth, outPitch, outHeight, base, increment );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int row = 0; row < outHeight; row++ ) { for ( int row = 0; row < outHeight; row++ ) {
float4 *outrow = (float4 *) ((char *) outHost + row*outPitch); float4 *outrow = (float4 *) ((char *) outHost + row*outPitch);

2
texturing/tex1d_9bit.cu
Unescape View File

@ -98,7 +98,7 @@ CreateAndPrintTex( T *initTex, size_t texN, size_t outN,
tex.addressMode[0] = addressMode; tex.addressMode[0] = addressMode;
cuda(HostGetDevicePointer(&outDevice, outHost, 0)); cuda(HostGetDevicePointer(&outDevice, outHost, 0));
TexReadout<<<2,384>>>( outDevice, outN, base, increment ); TexReadout<<<2,384>>>( outDevice, outN, base, increment );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
printf( "X\tY\tActual Value\tExpected Value\tDiff\n" ); printf( "X\tY\tActual Value\tExpected Value\tDiff\n" );
for ( int i = 0; i < outN; i++ ) { for ( int i = 0; i < outN; i++ ) {

2
texturing/tex1d_addressing.cu
Unescape View File

@ -169,7 +169,7 @@ CreateAndPrintTex( T *initTex, size_t texN, size_t outN,
tex.filterMode = filterMode; tex.filterMode = filterMode;
tex.addressMode[0] = addressMode; tex.addressMode[0] = addressMode;
TexReadout<<<2,384>>>( outDevice, outN, base, increment ); TexReadout<<<2,384>>>( outDevice, outN, base, increment );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int i = 0; i < outN; i++ ) { for ( int i = 0; i < outN; i++ ) {
float x = base+(float)i*increment; float x = base+(float)i*increment;

2
texturing/tex1d_unnormalized.cu
Unescape View File

@ -112,7 +112,7 @@ CreateAndPrintTex( T *initTex, size_t texN, size_t outN,
tex.addressMode[0] = addressMode; tex.addressMode[0] = addressMode;
cuda(HostGetDevicePointer(&outDevice, outHost, 0)); cuda(HostGetDevicePointer(&outDevice, outHost, 0));
TexReadout<<<2,384>>>( outDevice, outN, base, increment ); TexReadout<<<2,384>>>( outDevice, outN, base, increment );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int i = 0; i < outN; i++ ) { for ( int i = 0; i < outN; i++ ) {
printf( "(%.2f, %.2f)\n", outHost[i].x, outHost[i].y ); printf( "(%.2f, %.2f)\n", outHost[i].x, outHost[i].y );

2
texturing/tex1dfetch.cu
Unescape View File

@ -65,7 +65,7 @@ PrintTex( float *host, size_t N )
cuda(HostGetDevicePointer( (void **) &device, host, 0 )); cuda(HostGetDevicePointer( (void **) &device, host, 0 ));
TexReadout<<<2,384>>>( device, N ); TexReadout<<<2,384>>>( device, N );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int i = 0; i < N; i++ ) { for ( int i = 0; i < N; i++ ) {
printf( "%.2f ", host[i] ); printf( "%.2f ", host[i] );
} }

2
texturing/tex1dfetch_big.cu
Unescape View File

@ -177,7 +177,7 @@ TexChecksum( int *out, int c, size_t N )
default: default:
goto Error; goto Error;
} }
if ( cudaSuccess != cudaThreadSynchronize() ) if ( cudaSuccess != cudaDeviceSynchronize() )
goto Error; goto Error;
*out = checksumGPU(); *out = checksumGPU();
ret = true; ret = true;

2
texturing/tex1dfetch_host.cu
Unescape View File

@ -65,7 +65,7 @@ PrintTex( float *host, size_t N )
cuda(HostGetDevicePointer( (void **) &device, host, 0 )); cuda(HostGetDevicePointer( (void **) &device, host, 0 ));
TexReadout<<<2,384>>>( device, N ); TexReadout<<<2,384>>>( device, N );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int i = 0; i < N; i++ ) { for ( int i = 0; i < N; i++ ) {
printf( "%.2f ", host[i] ); printf( "%.2f ", host[i] );
} }

2
texturing/tex1dfetch_htod.cu
Unescape View File

@ -68,7 +68,7 @@ MeasureBandwidth( void *out, size_t N, int blocks, int threads )
chTimerGetTime( &start ); chTimerGetTime( &start );
TexReadout<<<2,384>>>( (float *) out, N ); TexReadout<<<2,384>>>( (float *) out, N );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
chTimerGetTime( &stop ); chTimerGetTime( &stop );

2
texturing/tex1dfetch_int2float.cu
Unescape View File

@ -119,7 +119,7 @@ CheckTexPromoteToFloat( size_t N )
cudaCreateChannelDesc<T>(), cudaCreateChannelDesc<T>(),
N*sizeof(T))); N*sizeof(T)));
TexReadout<<<2,384>>>( foutDevice, N ); TexReadout<<<2,384>>>( foutDevice, N );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int i = 0; i < N; i++ ) { for ( int i = 0; i < N; i++ ) {
printf( "%.2f ", foutHost[i] ); printf( "%.2f ", foutHost[i] );

2
texturing/tex1dfetch_offset.cu
Unescape View File

@ -66,7 +66,7 @@ CheckTex( float *hostOut, const float *in, size_t offset, size_t N )
cuda(HostGetDevicePointer( (void **) &deviceOut, hostOut, 0 )); cuda(HostGetDevicePointer( (void **) &deviceOut, hostOut, 0 ));
TexReadout<<<2,384>>>( deviceOut, offset>>2, N ); TexReadout<<<2,384>>>( deviceOut, offset>>2, N );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int i = 0; i < N; i++ ) { for ( int i = 0; i < N; i++ ) {
if ( in[i] != hostOut[i] ) { if ( in[i] != hostOut[i] ) {
printf( "Mismatch at index %d\n", i ); printf( "Mismatch at index %d\n", i );

2
texturing/tex2d_addressing.cu
Unescape View File

@ -139,7 +139,7 @@ CreateAndPrintTex(
blocks.y = 1; blocks.y = 1;
threads.x = 64; threads.y = 4; threads.x = 64; threads.y = 4;
TexReadout<<<blocks,threads>>>( outDevice, outWidth, outPitch, outHeight, base, increment ); TexReadout<<<blocks,threads>>>( outDevice, outWidth, outPitch, outHeight, base, increment );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int row = 0; row < outHeight; row++ ) { for ( int row = 0; row < outHeight; row++ ) {
float4 *outrow = (float4 *) ((char *) outHost + row*outPitch); float4 *outrow = (float4 *) ((char *) outHost + row*outPitch);

2
texturing/tex2d_addressing_device.cu
Unescape View File

@ -141,7 +141,7 @@ CreateAndPrintTex(
blocks.y = 1; blocks.y = 1;
threads.x = 64; threads.y = 4; threads.x = 64; threads.y = 4;
TexReadout<<<blocks,threads>>>( outDevice, outWidth, outPitch, outHeight, base, increment ); TexReadout<<<blocks,threads>>>( outDevice, outWidth, outPitch, outHeight, base, increment );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int row = 0; row < outHeight; row++ ) { for ( int row = 0; row < outHeight; row++ ) {
float4 *outrow = (float4 *) ((char *) outHost + row*outPitch); float4 *outrow = (float4 *) ((char *) outHost + row*outPitch);

2
texturing/tex2d_memset.cu
Unescape View File

@ -157,7 +157,7 @@ CreateAndPrintTex(
blocks.y = 1; blocks.y = 1;
threads.x = 64; threads.y = 4; threads.x = 64; threads.y = 4;
TexReadout<<<blocks,threads>>>( outDevice, outWidth, outPitch, outHeight, base, increment ); TexReadout<<<blocks,threads>>>( outDevice, outWidth, outPitch, outHeight, base, increment );
cuda(ThreadSynchronize()); cuda(DeviceSynchronize());
for ( int row = 0; row < outHeight; row++ ) { for ( int row = 0; row < outHeight; row++ ) {
float4 *outrow = (float4 *) ((char *) outHost + row*outPitch); float4 *outrow = (float4 *) ((char *) outHost + row*outPitch);

2
texturing/tex2d_opengl.cu
Unescape View File

@ -189,7 +189,7 @@ void displayCB(void) /* function called whenever redisplay needed */
else { else {
RenderTextureUnnormalized<<<g_height, 384>>>( g_deviceFrameBuffer, g_width, g_height ); RenderTextureUnnormalized<<<g_height, 384>>>( g_deviceFrameBuffer, g_width, g_height );
} }
if ( cudaSuccess != cudaThreadSynchronize() ) if ( cudaSuccess != cudaDeviceSynchronize() )
return; return;
glRasterPos2f( 0.0f, 0.0f ); glRasterPos2f( 0.0f, 0.0f );
glDrawPixels( g_width, g_height, GL_RGBA, GL_UNSIGNED_BYTE, g_hostFrameBuffer ); glDrawPixels( g_width, g_height, GL_RGBA, GL_UNSIGNED_BYTE, g_hostFrameBuffer );

Write Preview
Loading…
Cancel
Save