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

nbody: add ARMv7a NEON implementation 

Signed-off-by: Steven Noonan <steven@uplinklabs.net>
pull/1/head
Steven Noonan 13 years ago
parent
81d49ebecd
commit
3a655ff1be
5 changed files with 306 additions and 1 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 6
      nbody/GNUmakefile
  2. 116
      nbody/bodybodyInteraction_NEON.h
  3. 90
      nbody/nbody_CPU_NEON.cpp
  4. 93
      nbody/nbody_CPU_NEON_openmp.cpp
  5. 2
      nbody/nbody_CPU_SIMD.h

6
nbody/GNUmakefile
Unescape View File

@ -26,6 +26,10 @@ ifeq ($(uname_M),ppc) @@ -26,6 +26,10 @@ ifeq ($(uname_M),ppc)
CFLAGS += -DHAVE_ALTIVEC
CXXFLAGS += -maltivec
endif
ifeq ($(uname_M),armv7l)
CFLAGS += -DHAVE_NEON
CXXFLAGS += -mfpu=neon
endif
ifeq ($(uname_M),x86_64)
CFLAGS += -DHAVE_SSE
endif
@ -69,12 +73,14 @@ SOURCES := \ @@ -69,12 +73,14 @@ SOURCES := \
nbody_CPU_AOS_tiled.cpp \
nbody_CPU_SOA.cpp \
nbody_CPU_AltiVec.cpp \
nbody_CPU_NEON.cpp \
nbody_CPU_SSE.cpp \
nbody_CPU_SSE_threaded.cpp
ifeq ($(NO_OPENMP),)
SOURCES += \
nbody_CPU_AltiVec_openmp.cpp \
nbody_CPU_NEON_openmp.cpp \
nbody_CPU_SSE_openmp.cpp
endif

116
nbody/bodybodyInteraction_NEON.h
Unescape View File

@ -0,0 +1,116 @@ @@ -0,0 +1,116 @@
/*
*
* bodybodyInteraction_AltiVec.h
*
* SSE implementation of N-body computation.
*
* Copyright (c) 2011-2012, Archaea Software, LLC.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
*/
#ifdef __ARM_NEON__
#include <arm_neon.h>
typedef float vf32x4_t __attribute__ ((vector_size(16),aligned(1)));
static const vf32x4_t vec_zero = {0.0f, 0.0f, 0.0f, 0.0f};
typedef union {
float32x4_t v;
float f[4];
vf32x4_t p;
} v4;
static inline vf32x4_t
_vec_set_ps1(float f)
{
v4 r;
r.v = vdupq_n_f32(f);
return r.p;
}
static inline float
_vec_sum(vf32x4_t const &v)
{
float32x2_t r;
v4 iv;
iv.p = v;
r = vadd_f32(vget_high_f32(iv.v), vget_low_f32(iv.v));
return vget_lane_f32(vpadd_f32(r, r), 0);
}
static inline vf32x4_t
rcp_sqrt_nr_ps(const vf32x4_t& _v) {
v4 vec, result;
vec.p = _v;
result.v = vrsqrteq_f32(vec.v);
result.v = vmulq_f32(vrsqrtsq_f32(vmulq_f32(result.v, result.v), vec.v), result.v);
return result.p;
}
inline void
bodyBodyInteraction(
vf32x4_t& fx,
vf32x4_t& fy,
vf32x4_t& fz,
const vf32x4_t& x0,
const vf32x4_t& y0,
const vf32x4_t& z0,
const vf32x4_t& x1,
const vf32x4_t& y1,
const vf32x4_t& z1,
const vf32x4_t& mass1,
const vf32x4_t& softeningSquared )
{
// r_01 [3 FLOPS]
vf32x4_t dx = x1 - x0;
vf32x4_t dy = y1 - y0;
vf32x4_t dz = z1 - z0;
// d^2 + e^2 [6 FLOPS]
vf32x4_t distSq = ( dx * dx ) + ( dy * dy ) + ( dz * dz );
distSq = distSq + softeningSquared;
// invDistCube =1/distSqr^(3/2) [4 FLOPS (2 mul, 1 sqrt, 1 inv)]
vf32x4_t invDist = rcp_sqrt_nr_ps ( distSq );
vf32x4_t invDistCube = invDist * invDist * invDist;
// s = m_j * invDistCube [1 FLOP]
vf32x4_t s = mass1 * invDistCube;
// (m_1 * r_01) / (d^2 + e^2)^(3/2) [6 FLOPS]
fx = fx + (dx * s);
fy = fy + (dx * s);
fz = fz + (dz * s);
}
#endif

90
nbody/nbody_CPU_NEON.cpp
Unescape View File

@ -0,0 +1,90 @@ @@ -0,0 +1,90 @@
/*
*
* nbody_CPU_NEON.cpp
*
* Multithreaded NEON CPU implementation of the O(N^2) N-body calculation.
* Uses SOA (structure of arrays) representation because it is a much
* better fit for NEON.
*
* Copyright (c) 2011-2012, Archaea Software, LLC.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
*/
#ifdef __ARM_NEON__
#include <chTimer.h>
#include "nbody.h"
#include "bodybodyInteraction_NEON.h"
#include "nbody_CPU_SIMD.h"
float
ComputeGravitation_SIMD(
float *force[3],
float *pos[4],
float *mass,
float softeningSquared,
size_t N
)
{
chTimerTimestamp start, end;
chTimerGetTime( &start );
for (size_t i = 0; i < N; i++)
{
vf32x4_t ax = vec_zero;
vf32x4_t ay = vec_zero;
vf32x4_t az = vec_zero;
vf32x4_t *px = (vf32x4_t *) pos[0];
vf32x4_t *py = (vf32x4_t *) pos[1];
vf32x4_t *pz = (vf32x4_t *) pos[2];
vf32x4_t *pmass = (vf32x4_t *) mass;
vf32x4_t x0 = _vec_set_ps1( pos[0][i] );
vf32x4_t y0 = _vec_set_ps1( pos[1][i] );
vf32x4_t z0 = _vec_set_ps1( pos[2][i] );
for ( size_t j = 0; j < N/4; j++ ) {
bodyBodyInteraction(
ax, ay, az,
x0, y0, z0,
px[j], py[j], pz[j], pmass[j],
_vec_set_ps1( softeningSquared ) );
}
// Accumulate sum of four floats in the NEON register
force[0][i] = _vec_sum( ax );
force[1][i] = _vec_sum( ay );
force[2][i] = _vec_sum( az );
}
chTimerGetTime( &end );
return (float) chTimerElapsedTime( &start, &end ) * 1000.0f;
}
#endif

93
nbody/nbody_CPU_NEON_openmp.cpp
Unescape View File

@ -0,0 +1,93 @@ @@ -0,0 +1,93 @@
/*
*
* nbody_CPU_NEON.cpp
*
* Multithreaded NEON CPU implementation of the O(N^2) N-body calculation.
* Uses SOA (structure of arrays) representation because it is a much
* better fit for NEON.
*
* Copyright (c) 2011-2012, Archaea Software, LLC.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
*/
#ifdef __ARM_NEON__
#ifdef _OPENMP
#include <chTimer.h>
#include "nbody.h"
#include "bodybodyInteraction_NEON.h"
#include "nbody_CPU_SIMD.h"
float
ComputeGravitation_SIMD_openmp(
float *force[3],
float *pos[4],
float *mass,
float softeningSquared,
size_t N
)
{
chTimerTimestamp start, end;
chTimerGetTime( &start );
#pragma omp parallel for
for (size_t i = 0; i < N; i++)
{
vf32x4_t ax = vec_zero;
vf32x4_t ay = vec_zero;
vf32x4_t az = vec_zero;
vf32x4_t *px = (vf32x4_t *) pos[0];
vf32x4_t *py = (vf32x4_t *) pos[1];
vf32x4_t *pz = (vf32x4_t *) pos[2];
vf32x4_t *pmass = (vf32x4_t *) mass;
vf32x4_t x0 = _vec_set_ps1( pos[0][i] );
vf32x4_t y0 = _vec_set_ps1( pos[1][i] );
vf32x4_t z0 = _vec_set_ps1( pos[2][i] );
for ( size_t j = 0; j < N/4; j++ ) {
bodyBodyInteraction(
ax, ay, az,
x0, y0, z0,
px[j], py[j], pz[j], pmass[j],
_vec_set_ps1( softeningSquared ) );
}
// Accumulate sum of four floats in the NEON register
force[0][i] = _vec_sum( ax );
force[1][i] = _vec_sum( ay );
force[2][i] = _vec_sum( az );
}
chTimerGetTime( &end );
return (float) chTimerElapsedTime( &start, &end ) * 1000.0f;
}
#endif
#endif

2
nbody/nbody_CPU_SIMD.h
Unescape View File

@ -43,7 +43,7 @@ @@ -43,7 +43,7 @@
#define HAVE_SIMD_OPENMP
#endif
#elif defined(HAVE_ALTIVEC)
#elif defined(HAVE_ALTIVEC) || defined(HAVE_NEON)
#define HAVE_SIMD
#ifdef USE_OPENMP

Write Preview
Loading…
Cancel
Save