8 Commits
v9.2 ... 10.1.1

Author SHA1 Message Date
Mahesh Doijade
337815dbee Add and update samples with CUDA 10.1 Update 1 support 2019-04-10 20:12:09 +05:30
Mahesh Doijade
1abc294982 --Update README.md for cuda 10.1 release
-- fix multidevice CG sample to check for concurrentManagedAccess as it needs systemwide atomics
2019-01-23 16:56:20 +05:30
Mahesh Doijade
b458dafcd6 Add and update samples with CUDA 10.1 support 2019-01-23 01:34:43 +05:30
Mahesh Doijade
32f0fc6111 fix the description of CUDA IPC as it is supported on Windows as well since CUDA 9.2 2018-10-24 13:35:39 +05:30
Mahesh Doijade
6bbaf9ac80 -- use DX_SDK env variable only for non-DX12 samples VS files
as DX12 does not need legacy DX SDK
2018-10-09 22:42:25 +05:30
Mahesh Doijade
fcb23487a8 -- Add simpleD3D12 sample which demonstrates DX12-CUDA interop
rendering sinewave
-- Update the documentation with description about this sample
2018-10-04 20:05:23 +05:30
Mahesh Doijade
91dc60dd96 simpleVulkan: windows.h should be included before VersionHelpers.h 2018-09-10 20:41:05 +05:30
Mahesh Doijade
21c36d3568 Add and Update samples for CUDA 10.0 2018-08-24 22:35:15 +05:30
387 changed files with 70041 additions and 1181 deletions

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
@@ -51,457 +51,17 @@
// CUDA Runtime error messages
#ifdef __DRIVER_TYPES_H__
static const char *_cudaGetErrorEnum(cudaError_t error) {
switch (error) {
case cudaSuccess:
return "cudaSuccess";
case cudaErrorMissingConfiguration:
return "cudaErrorMissingConfiguration";
case cudaErrorMemoryAllocation:
return "cudaErrorMemoryAllocation";
case cudaErrorInitializationError:
return "cudaErrorInitializationError";
case cudaErrorLaunchFailure:
return "cudaErrorLaunchFailure";
case cudaErrorPriorLaunchFailure:
return "cudaErrorPriorLaunchFailure";
case cudaErrorLaunchTimeout:
return "cudaErrorLaunchTimeout";
case cudaErrorLaunchOutOfResources:
return "cudaErrorLaunchOutOfResources";
case cudaErrorInvalidDeviceFunction:
return "cudaErrorInvalidDeviceFunction";
case cudaErrorInvalidConfiguration:
return "cudaErrorInvalidConfiguration";
case cudaErrorInvalidDevice:
return "cudaErrorInvalidDevice";
case cudaErrorInvalidValue:
return "cudaErrorInvalidValue";
case cudaErrorInvalidPitchValue:
return "cudaErrorInvalidPitchValue";
case cudaErrorInvalidSymbol:
return "cudaErrorInvalidSymbol";
case cudaErrorMapBufferObjectFailed:
return "cudaErrorMapBufferObjectFailed";
case cudaErrorUnmapBufferObjectFailed:
return "cudaErrorUnmapBufferObjectFailed";
case cudaErrorInvalidHostPointer:
return "cudaErrorInvalidHostPointer";
case cudaErrorInvalidDevicePointer:
return "cudaErrorInvalidDevicePointer";
case cudaErrorInvalidTexture:
return "cudaErrorInvalidTexture";
case cudaErrorInvalidTextureBinding:
return "cudaErrorInvalidTextureBinding";
case cudaErrorInvalidChannelDescriptor:
return "cudaErrorInvalidChannelDescriptor";
case cudaErrorInvalidMemcpyDirection:
return "cudaErrorInvalidMemcpyDirection";
case cudaErrorAddressOfConstant:
return "cudaErrorAddressOfConstant";
case cudaErrorTextureFetchFailed:
return "cudaErrorTextureFetchFailed";
case cudaErrorTextureNotBound:
return "cudaErrorTextureNotBound";
case cudaErrorSynchronizationError:
return "cudaErrorSynchronizationError";
case cudaErrorInvalidFilterSetting:
return "cudaErrorInvalidFilterSetting";
case cudaErrorInvalidNormSetting:
return "cudaErrorInvalidNormSetting";
case cudaErrorMixedDeviceExecution:
return "cudaErrorMixedDeviceExecution";
case cudaErrorCudartUnloading:
return "cudaErrorCudartUnloading";
case cudaErrorUnknown:
return "cudaErrorUnknown";
case cudaErrorNotYetImplemented:
return "cudaErrorNotYetImplemented";
case cudaErrorMemoryValueTooLarge:
return "cudaErrorMemoryValueTooLarge";
case cudaErrorInvalidResourceHandle:
return "cudaErrorInvalidResourceHandle";
case cudaErrorNotReady:
return "cudaErrorNotReady";
case cudaErrorInsufficientDriver:
return "cudaErrorInsufficientDriver";
case cudaErrorSetOnActiveProcess:
return "cudaErrorSetOnActiveProcess";
case cudaErrorInvalidSurface:
return "cudaErrorInvalidSurface";
case cudaErrorNoDevice:
return "cudaErrorNoDevice";
case cudaErrorECCUncorrectable:
return "cudaErrorECCUncorrectable";
case cudaErrorSharedObjectSymbolNotFound:
return "cudaErrorSharedObjectSymbolNotFound";
case cudaErrorSharedObjectInitFailed:
return "cudaErrorSharedObjectInitFailed";
case cudaErrorUnsupportedLimit:
return "cudaErrorUnsupportedLimit";
case cudaErrorDuplicateVariableName:
return "cudaErrorDuplicateVariableName";
case cudaErrorDuplicateTextureName:
return "cudaErrorDuplicateTextureName";
case cudaErrorDuplicateSurfaceName:
return "cudaErrorDuplicateSurfaceName";
case cudaErrorDevicesUnavailable:
return "cudaErrorDevicesUnavailable";
case cudaErrorInvalidKernelImage:
return "cudaErrorInvalidKernelImage";
case cudaErrorNoKernelImageForDevice:
return "cudaErrorNoKernelImageForDevice";
case cudaErrorIncompatibleDriverContext:
return "cudaErrorIncompatibleDriverContext";
case cudaErrorPeerAccessAlreadyEnabled:
return "cudaErrorPeerAccessAlreadyEnabled";
case cudaErrorPeerAccessNotEnabled:
return "cudaErrorPeerAccessNotEnabled";
case cudaErrorDeviceAlreadyInUse:
return "cudaErrorDeviceAlreadyInUse";
case cudaErrorProfilerDisabled:
return "cudaErrorProfilerDisabled";
case cudaErrorProfilerNotInitialized:
return "cudaErrorProfilerNotInitialized";
case cudaErrorProfilerAlreadyStarted:
return "cudaErrorProfilerAlreadyStarted";
case cudaErrorProfilerAlreadyStopped:
return "cudaErrorProfilerAlreadyStopped";
/* Since CUDA 4.0*/
case cudaErrorAssert:
return "cudaErrorAssert";
case cudaErrorTooManyPeers:
return "cudaErrorTooManyPeers";
case cudaErrorHostMemoryAlreadyRegistered:
return "cudaErrorHostMemoryAlreadyRegistered";
case cudaErrorHostMemoryNotRegistered:
return "cudaErrorHostMemoryNotRegistered";
/* Since CUDA 5.0 */
case cudaErrorOperatingSystem:
return "cudaErrorOperatingSystem";
case cudaErrorPeerAccessUnsupported:
return "cudaErrorPeerAccessUnsupported";
case cudaErrorLaunchMaxDepthExceeded:
return "cudaErrorLaunchMaxDepthExceeded";
case cudaErrorLaunchFileScopedTex:
return "cudaErrorLaunchFileScopedTex";
case cudaErrorLaunchFileScopedSurf:
return "cudaErrorLaunchFileScopedSurf";
case cudaErrorSyncDepthExceeded:
return "cudaErrorSyncDepthExceeded";
case cudaErrorLaunchPendingCountExceeded:
return "cudaErrorLaunchPendingCountExceeded";
case cudaErrorNotPermitted:
return "cudaErrorNotPermitted";
case cudaErrorNotSupported:
return "cudaErrorNotSupported";
/* Since CUDA 6.0 */
case cudaErrorHardwareStackError:
return "cudaErrorHardwareStackError";
case cudaErrorIllegalInstruction:
return "cudaErrorIllegalInstruction";
case cudaErrorMisalignedAddress:
return "cudaErrorMisalignedAddress";
case cudaErrorInvalidAddressSpace:
return "cudaErrorInvalidAddressSpace";
case cudaErrorInvalidPc:
return "cudaErrorInvalidPc";
case cudaErrorIllegalAddress:
return "cudaErrorIllegalAddress";
/* Since CUDA 6.5*/
case cudaErrorInvalidPtx:
return "cudaErrorInvalidPtx";
case cudaErrorInvalidGraphicsContext:
return "cudaErrorInvalidGraphicsContext";
case cudaErrorStartupFailure:
return "cudaErrorStartupFailure";
case cudaErrorApiFailureBase:
return "cudaErrorApiFailureBase";
/* Since CUDA 8.0*/
case cudaErrorNvlinkUncorrectable:
return "cudaErrorNvlinkUncorrectable";
/* Since CUDA 8.5*/
case cudaErrorJitCompilerNotFound:
return "cudaErrorJitCompilerNotFound";
/* Since CUDA 9.0*/
case cudaErrorCooperativeLaunchTooLarge:
return "cudaErrorCooperativeLaunchTooLarge";
}
return "<unknown>";
return cudaGetErrorName(error);
}
#endif
#ifdef __cuda_cuda_h__
#ifdef CUDA_DRIVER_API
// CUDA Driver API errors
static const char *_cudaGetErrorEnum(CUresult error) {
switch (error) {
case CUDA_SUCCESS:
return "CUDA_SUCCESS";
case CUDA_ERROR_INVALID_VALUE:
return "CUDA_ERROR_INVALID_VALUE";
case CUDA_ERROR_OUT_OF_MEMORY:
return "CUDA_ERROR_OUT_OF_MEMORY";
case CUDA_ERROR_NOT_INITIALIZED:
return "CUDA_ERROR_NOT_INITIALIZED";
case CUDA_ERROR_DEINITIALIZED:
return "CUDA_ERROR_DEINITIALIZED";
case CUDA_ERROR_PROFILER_DISABLED:
return "CUDA_ERROR_PROFILER_DISABLED";
case CUDA_ERROR_PROFILER_NOT_INITIALIZED:
return "CUDA_ERROR_PROFILER_NOT_INITIALIZED";
case CUDA_ERROR_PROFILER_ALREADY_STARTED:
return "CUDA_ERROR_PROFILER_ALREADY_STARTED";
case CUDA_ERROR_PROFILER_ALREADY_STOPPED:
return "CUDA_ERROR_PROFILER_ALREADY_STOPPED";
case CUDA_ERROR_NO_DEVICE:
return "CUDA_ERROR_NO_DEVICE";
case CUDA_ERROR_INVALID_DEVICE:
return "CUDA_ERROR_INVALID_DEVICE";
case CUDA_ERROR_INVALID_IMAGE:
return "CUDA_ERROR_INVALID_IMAGE";
case CUDA_ERROR_INVALID_CONTEXT:
return "CUDA_ERROR_INVALID_CONTEXT";
case CUDA_ERROR_CONTEXT_ALREADY_CURRENT:
return "CUDA_ERROR_CONTEXT_ALREADY_CURRENT";
case CUDA_ERROR_MAP_FAILED:
return "CUDA_ERROR_MAP_FAILED";
case CUDA_ERROR_UNMAP_FAILED:
return "CUDA_ERROR_UNMAP_FAILED";
case CUDA_ERROR_ARRAY_IS_MAPPED:
return "CUDA_ERROR_ARRAY_IS_MAPPED";
case CUDA_ERROR_ALREADY_MAPPED:
return "CUDA_ERROR_ALREADY_MAPPED";
case CUDA_ERROR_NO_BINARY_FOR_GPU:
return "CUDA_ERROR_NO_BINARY_FOR_GPU";
case CUDA_ERROR_ALREADY_ACQUIRED:
return "CUDA_ERROR_ALREADY_ACQUIRED";
case CUDA_ERROR_NOT_MAPPED:
return "CUDA_ERROR_NOT_MAPPED";
case CUDA_ERROR_NOT_MAPPED_AS_ARRAY:
return "CUDA_ERROR_NOT_MAPPED_AS_ARRAY";
case CUDA_ERROR_NOT_MAPPED_AS_POINTER:
return "CUDA_ERROR_NOT_MAPPED_AS_POINTER";
case CUDA_ERROR_ECC_UNCORRECTABLE:
return "CUDA_ERROR_ECC_UNCORRECTABLE";
case CUDA_ERROR_UNSUPPORTED_LIMIT:
return "CUDA_ERROR_UNSUPPORTED_LIMIT";
case CUDA_ERROR_CONTEXT_ALREADY_IN_USE:
return "CUDA_ERROR_CONTEXT_ALREADY_IN_USE";
case CUDA_ERROR_PEER_ACCESS_UNSUPPORTED:
return "CUDA_ERROR_PEER_ACCESS_UNSUPPORTED";
case CUDA_ERROR_INVALID_PTX:
return "CUDA_ERROR_INVALID_PTX";
case CUDA_ERROR_INVALID_GRAPHICS_CONTEXT:
return "CUDA_ERROR_INVALID_GRAPHICS_CONTEXT";
case CUDA_ERROR_NVLINK_UNCORRECTABLE:
return "CUDA_ERROR_NVLINK_UNCORRECTABLE";
case CUDA_ERROR_JIT_COMPILER_NOT_FOUND:
return "CUDA_ERROR_JIT_COMPILER_NOT_FOUND";
case CUDA_ERROR_INVALID_SOURCE:
return "CUDA_ERROR_INVALID_SOURCE";
case CUDA_ERROR_FILE_NOT_FOUND:
return "CUDA_ERROR_FILE_NOT_FOUND";
case CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND:
return "CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND";
case CUDA_ERROR_SHARED_OBJECT_INIT_FAILED:
return "CUDA_ERROR_SHARED_OBJECT_INIT_FAILED";
case CUDA_ERROR_OPERATING_SYSTEM:
return "CUDA_ERROR_OPERATING_SYSTEM";
case CUDA_ERROR_INVALID_HANDLE:
return "CUDA_ERROR_INVALID_HANDLE";
case CUDA_ERROR_NOT_FOUND:
return "CUDA_ERROR_NOT_FOUND";
case CUDA_ERROR_NOT_READY:
return "CUDA_ERROR_NOT_READY";
case CUDA_ERROR_ILLEGAL_ADDRESS:
return "CUDA_ERROR_ILLEGAL_ADDRESS";
case CUDA_ERROR_LAUNCH_FAILED:
return "CUDA_ERROR_LAUNCH_FAILED";
case CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES:
return "CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES";
case CUDA_ERROR_LAUNCH_TIMEOUT:
return "CUDA_ERROR_LAUNCH_TIMEOUT";
case CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING:
return "CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING";
case CUDA_ERROR_PEER_ACCESS_ALREADY_ENABLED:
return "CUDA_ERROR_PEER_ACCESS_ALREADY_ENABLED";
case CUDA_ERROR_PEER_ACCESS_NOT_ENABLED:
return "CUDA_ERROR_PEER_ACCESS_NOT_ENABLED";
case CUDA_ERROR_PRIMARY_CONTEXT_ACTIVE:
return "CUDA_ERROR_PRIMARY_CONTEXT_ACTIVE";
case CUDA_ERROR_CONTEXT_IS_DESTROYED:
return "CUDA_ERROR_CONTEXT_IS_DESTROYED";
case CUDA_ERROR_ASSERT:
return "CUDA_ERROR_ASSERT";
case CUDA_ERROR_TOO_MANY_PEERS:
return "CUDA_ERROR_TOO_MANY_PEERS";
case CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED:
return "CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED";
case CUDA_ERROR_HOST_MEMORY_NOT_REGISTERED:
return "CUDA_ERROR_HOST_MEMORY_NOT_REGISTERED";
case CUDA_ERROR_HARDWARE_STACK_ERROR:
return "CUDA_ERROR_HARDWARE_STACK_ERROR";
case CUDA_ERROR_ILLEGAL_INSTRUCTION:
return "CUDA_ERROR_ILLEGAL_INSTRUCTION";
case CUDA_ERROR_MISALIGNED_ADDRESS:
return "CUDA_ERROR_MISALIGNED_ADDRESS";
case CUDA_ERROR_INVALID_ADDRESS_SPACE:
return "CUDA_ERROR_INVALID_ADDRESS_SPACE";
case CUDA_ERROR_INVALID_PC:
return "CUDA_ERROR_INVALID_PC";
case CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE:
return "CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE";
case CUDA_ERROR_NOT_PERMITTED:
return "CUDA_ERROR_NOT_PERMITTED";
case CUDA_ERROR_NOT_SUPPORTED:
return "CUDA_ERROR_NOT_SUPPORTED";
case CUDA_ERROR_UNKNOWN:
return "CUDA_ERROR_UNKNOWN";
}
return "<unknown>";
static char unknown[] = "<unknown>";
const char *ret = NULL;
cuGetErrorName(error, &ret);
return ret ? ret : unknown;
}
#endif
@@ -722,6 +282,42 @@ static const char *_cudaGetErrorEnum(curandStatus_t error) {
}
#endif
#ifdef NVJPEGAPI
// nvJPEG API errors
static const char *_cudaGetErrorEnum(nvjpegStatus_t error) {
switch (error) {
case NVJPEG_STATUS_SUCCESS:
return "NVJPEG_STATUS_SUCCESS";
case NVJPEG_STATUS_NOT_INITIALIZED:
return "NVJPEG_STATUS_NOT_INITIALIZED";
case NVJPEG_STATUS_INVALID_PARAMETER:
return "NVJPEG_STATUS_INVALID_PARAMETER";
case NVJPEG_STATUS_BAD_JPEG:
return "NVJPEG_STATUS_BAD_JPEG";
case NVJPEG_STATUS_JPEG_NOT_SUPPORTED:
return "NVJPEG_STATUS_JPEG_NOT_SUPPORTED";
case NVJPEG_STATUS_ALLOCATOR_FAILURE:
return "NVJPEG_STATUS_ALLOCATOR_FAILURE";
case NVJPEG_STATUS_EXECUTION_FAILED:
return "NVJPEG_STATUS_EXECUTION_FAILED";
case NVJPEG_STATUS_ARCH_MISMATCH:
return "NVJPEG_STATUS_ARCH_MISMATCH";
case NVJPEG_STATUS_INTERNAL_ERROR:
return "NVJPEG_STATUS_INTERNAL_ERROR";
}
return "<unknown>";
}
#endif
#ifdef NV_NPPIDEFS_H
// NPP API errors
static const char *_cudaGetErrorEnum(NppStatus error) {
@@ -1067,18 +663,19 @@ inline int _ConvertSMVer2Cores(int major, int minor) {
} sSMtoCores;
sSMtoCores nGpuArchCoresPerSM[] = {
{0x30, 192}, // Kepler Generation (SM 3.0) GK10x class
{0x32, 192}, // Kepler Generation (SM 3.2) GK10x class
{0x35, 192}, // Kepler Generation (SM 3.5) GK11x class
{0x37, 192}, // Kepler Generation (SM 3.7) GK21x class
{0x50, 128}, // Maxwell Generation (SM 5.0) GM10x class
{0x52, 128}, // Maxwell Generation (SM 5.2) GM20x class
{0x53, 128}, // Maxwell Generation (SM 5.3) GM20x class
{0x60, 64}, // Pascal Generation (SM 6.0) GP100 class
{0x61, 128}, // Pascal Generation (SM 6.1) GP10x class
{0x62, 128}, // Pascal Generation (SM 6.2) GP10x class
{0x70, 64}, // Volta Generation (SM 7.0) GV100 class
{0x72, 64}, // Volta Generation (SM 7.2) GV11b class
{0x30, 192},
{0x32, 192},
{0x35, 192},
{0x37, 192},
{0x50, 128},
{0x52, 128},
{0x53, 128},
{0x60, 64},
{0x61, 128},
{0x62, 128},
{0x70, 64},
{0x72, 64},
{0x75, 64},
{-1, -1}};
int index = 0;
@@ -1155,7 +752,7 @@ inline int gpuDeviceInit(int devID) {
inline int gpuGetMaxGflopsDeviceId() {
int current_device = 0, sm_per_multiproc = 0;
int max_perf_device = 0;
int device_count = 0, best_SM_arch = 0;
int device_count = 0;
int devices_prohibited = 0;
uint64_t max_compute_perf = 0;
@@ -1169,30 +766,6 @@ inline int gpuGetMaxGflopsDeviceId() {
exit(EXIT_FAILURE);
}
// Find the best major SM Architecture GPU device
while (current_device < device_count) {
cudaGetDeviceProperties(&deviceProp, current_device);
// If this GPU is not running on Compute Mode prohibited,
// then we can add it to the list
if (deviceProp.computeMode != cudaComputeModeProhibited) {
if (deviceProp.major > 0 && deviceProp.major < 9999) {
best_SM_arch = MAX(best_SM_arch, deviceProp.major);
}
} else {
devices_prohibited++;
}
current_device++;
}
if (devices_prohibited == device_count) {
fprintf(stderr,
"gpuGetMaxGflopsDeviceId() CUDA error:"
" all devices have compute mode prohibited.\n");
exit(EXIT_FAILURE);
}
// Find the best CUDA capable GPU device
current_device = 0;
@@ -1213,23 +786,23 @@ inline int gpuGetMaxGflopsDeviceId() {
sm_per_multiproc * deviceProp.clockRate;
if (compute_perf > max_compute_perf) {
// If we find GPU with SM major > 2, search only these
if (best_SM_arch > 2) {
// If our device==dest_SM_arch, choose this, or else pass
if (deviceProp.major == best_SM_arch) {
max_compute_perf = compute_perf;
max_perf_device = current_device;
}
} else {
max_compute_perf = compute_perf;
max_perf_device = current_device;
}
max_compute_perf = compute_perf;
max_perf_device = current_device;
}
} else {
devices_prohibited++;
}
++current_device;
}
if (devices_prohibited == device_count) {
fprintf(stderr,
"gpuGetMaxGflopsDeviceId() CUDA error:"
" all devices have compute mode prohibited.\n");
exit(EXIT_FAILURE);
}
return max_perf_device;
}

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
@@ -122,18 +122,19 @@ inline int _ConvertSMVer2CoresDRV(int major, int minor) {
} sSMtoCores;
sSMtoCores nGpuArchCoresPerSM[] = {
{0x30, 192}, // Kepler Generation (SM 3.0) GK10x class
{0x32, 192}, // Kepler Generation (SM 3.2) GK10x class
{0x35, 192}, // Kepler Generation (SM 3.5) GK11x class
{0x37, 192}, // Kepler Generation (SM 3.7) GK21x class
{0x50, 128}, // Maxwell Generation (SM 5.0) GM10x class
{0x52, 128}, // Maxwell Generation (SM 5.2) GM20x class
{0x53, 128}, // Maxwell Generation (SM 5.3) GM20x class
{0x60, 64}, // Pascal Generation (SM 6.0) GP100 class
{0x61, 128}, // Pascal Generation (SM 6.1) GP10x class
{0x62, 128}, // Pascal Generation (SM 6.2) GP10x class
{0x70, 64}, // Volta Generation (SM 7.0) GV100 class
{0x72, 64}, // Volta Generation (SM 7.2) GV11b class
{0x30, 192},
{0x32, 192},
{0x35, 192},
{0x37, 192},
{0x50, 128},
{0x52, 128},
{0x53, 128},
{0x60, 64},
{0x61, 128},
{0x62, 128},
{0x70, 64},
{0x72, 64},
{0x75, 64},
{-1, -1}};
int index = 0;

166
Common/helper_cusolver.h Normal file
View File

@@ -0,0 +1,166 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#ifndef HELPER_CUSOLVER
#define HELPER_CUSOLVER
#include <ctype.h>
#include <cuda_runtime.h>
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "cusparse.h"
#define SWITCH_CHAR '-'
struct testOpts {
char *sparse_mat_filename; // by switch -F<filename>
const char *testFunc; // by switch -R<name>
const char *reorder; // by switch -P<name>
int lda; // by switch -lda<int>
};
double vec_norminf(int n, const double *x) {
double norminf = 0;
for (int j = 0; j < n; j++) {
double x_abs = fabs(x[j]);
norminf = (norminf > x_abs) ? norminf : x_abs;
}
return norminf;
}
/*
* |A| = max { |A|*ones(m,1) }
*/
double mat_norminf(int m, int n, const double *A, int lda) {
double norminf = 0;
for (int i = 0; i < m; i++) {
double sum = 0.0;
for (int j = 0; j < n; j++) {
double A_abs = fabs(A[i + j * lda]);
sum += A_abs;
}
norminf = (norminf > sum) ? norminf : sum;
}
return norminf;
}
/*
* |A| = max { |A|*ones(m,1) }
*/
double csr_mat_norminf(int m, int n, int nnzA, const cusparseMatDescr_t descrA,
const double *csrValA, const int *csrRowPtrA,
const int *csrColIndA) {
const int baseA =
(CUSPARSE_INDEX_BASE_ONE == cusparseGetMatIndexBase(descrA)) ? 1 : 0;
double norminf = 0;
for (int i = 0; i < m; i++) {
double sum = 0.0;
const int start = csrRowPtrA[i] - baseA;
const int end = csrRowPtrA[i + 1] - baseA;
for (int colidx = start; colidx < end; colidx++) {
// const int j = csrColIndA[colidx] - baseA;
double A_abs = fabs(csrValA[colidx]);
sum += A_abs;
}
norminf = (norminf > sum) ? norminf : sum;
}
return norminf;
}
void display_matrix(int m, int n, int nnzA, const cusparseMatDescr_t descrA,
const double *csrValA, const int *csrRowPtrA,
const int *csrColIndA) {
const int baseA =
(CUSPARSE_INDEX_BASE_ONE == cusparseGetMatIndexBase(descrA)) ? 1 : 0;
printf("m = %d, n = %d, nnz = %d, matlab base-1\n", m, n, nnzA);
for (int row = 0; row < m; row++) {
const int start = csrRowPtrA[row] - baseA;
const int end = csrRowPtrA[row + 1] - baseA;
for (int colidx = start; colidx < end; colidx++) {
const int col = csrColIndA[colidx] - baseA;
double Areg = csrValA[colidx];
printf("A(%d, %d) = %20.16E\n", row + 1, col + 1, Areg);
}
}
}
#if defined(_WIN32)
#if !defined(WIN32_LEAN_AND_MEAN)
#define WIN32_LEAN_AND_MEAN
#endif
#include <windows.h>
double second(void) {
LARGE_INTEGER t;
static double oofreq;
static int checkedForHighResTimer;
static BOOL hasHighResTimer;
if (!checkedForHighResTimer) {
hasHighResTimer = QueryPerformanceFrequency(&t);
oofreq = 1.0 / (double)t.QuadPart;
checkedForHighResTimer = 1;
}
if (hasHighResTimer) {
QueryPerformanceCounter(&t);
return (double)t.QuadPart * oofreq;
} else {
return (double)GetTickCount() / 1000.0;
}
}
#elif defined(__linux__) || defined(__QNX__)
#include <stddef.h>
#include <sys/resource.h>
#include <sys/time.h>
double second(void) {
struct timeval tv;
gettimeofday(&tv, NULL);
return (double)tv.tv_sec + (double)tv.tv_usec / 1000000.0;
}
#elif defined(__APPLE__)
#include <stddef.h>
#include <sys/resource.h>
#include <sys/sysctl.h>
#include <sys/time.h>
#include <sys/types.h>
double second(void) {
struct timeval tv;
gettimeofday(&tv, NULL);
return (double)tv.tv_sec + (double)tv.tv_usec / 1000000.0;
}
#else
#error unsupported platform
#endif
#endif

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions

View File

@@ -0,0 +1,178 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#include "helper_multiprocess.h"
#include <cstdlib>
#include <string>
int sharedMemoryCreate(const char *name, size_t sz, sharedMemoryInfo *info)
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
info->size = sz;
info->shmHandle = CreateFileMapping(INVALID_HANDLE_VALUE,
NULL,
PAGE_READWRITE,
0,
(DWORD)sz,
name);
if (info->shmHandle == 0) {
return GetLastError();
}
info->addr = MapViewOfFile(info->shmHandle, FILE_MAP_ALL_ACCESS, 0, 0, sz);
if (info->addr == NULL) {
return GetLastError();
}
return 0;
#else
int status = 0;
info->size = sz;
info->shmFd = shm_open(name, O_RDWR | O_CREAT, 0777);
if (info->shmFd < 0) {
return errno;
}
status = ftruncate(info->shmFd, sz);
if (status != 0) {
return status;
}
info->addr = mmap(0, sz, PROT_READ | PROT_WRITE, MAP_SHARED, info->shmFd, 0);
if (info->addr == NULL) {
return errno;
}
return 0;
#endif
}
int sharedMemoryOpen(const char *name, size_t sz, sharedMemoryInfo *info)
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
info->size = sz;
info->shmHandle = OpenFileMapping(FILE_MAP_ALL_ACCESS, FALSE, name);
if (info->shmHandle == 0) {
return GetLastError();
}
info->addr = MapViewOfFile(info->shmHandle, FILE_MAP_ALL_ACCESS, 0, 0, sz);
if (info->addr == NULL) {
return GetLastError();
}
return 0;
#else
info->size = sz;
info->shmFd = shm_open(name, O_RDWR, 0777);
if (info->shmFd < 0) {
return errno;
}
info->addr = mmap(0, sz, PROT_READ | PROT_WRITE, MAP_SHARED, info->shmFd, 0);
if (info->addr == NULL) {
return errno;
}
return 0;
#endif
}
void sharedMemoryClose(sharedMemoryInfo *info)
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
if (info->addr) {
UnmapViewOfFile(info->addr);
}
if (info->shmHandle) {
CloseHandle(info->shmHandle);
}
#else
if (info->addr) {
munmap(info->addr, info->size);
}
if (info->shmFd) {
close(info->shmFd);
}
#endif
}
int spawnProcess(Process *process, const char *app, char * const *args)
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
STARTUPINFO si = {0};
BOOL status;
size_t arglen = 0;
size_t argIdx = 0;
std::string arg_string;
memset(process, 0, sizeof(*process));
while (*args) {
arg_string.append(*args).append(1, ' ');
args++;
}
status = CreateProcess(app, LPSTR(arg_string.c_str()), NULL, NULL, FALSE, 0, NULL, NULL, &si, process);
return status ? 0 : GetLastError();
#else
*process = fork();
if (*process == 0) {
if (0 > execvp(app, args)) {
return errno;
}
}
else if (*process < 0) {
return errno;
}
return 0;
#endif
}
int waitProcess(Process *process)
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
DWORD exitCode;
WaitForSingleObject(process->hProcess, INFINITE);
GetExitCodeProcess(process->hProcess, &exitCode);
CloseHandle(process->hProcess);
CloseHandle(process->hThread);
return (int)exitCode;
#else
int status = 0;
do {
if (0 > waitpid(*process, &status, 0)) {
return errno;
}
} while (!WIFEXITED(status));
return WEXITSTATUS(status);
#endif
}

View File

@@ -0,0 +1,71 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#ifndef HELPER_MULTIPROCESS_H
#define HELPER_MULTIPROCESS_H
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
#ifndef WIN32_LEAN_AND_MEAN
#define WIN32_LEAN_AND_MEAN
#endif
#include <windows.h>
#else
#include <fcntl.h>
#include <sys/mman.h>
#include <unistd.h>
#include <errno.h>
#include <sys/wait.h>
#endif
typedef struct sharedMemoryInfo_st {
void *addr;
size_t size;
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
HANDLE shmHandle;
#else
int shmFd;
#endif
} sharedMemoryInfo;
int sharedMemoryCreate(const char *name, size_t sz, sharedMemoryInfo *info);
int sharedMemoryOpen(const char *name, size_t sz, sharedMemoryInfo *info);
void sharedMemoryClose(sharedMemoryInfo *info);
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
typedef PROCESS_INFORMATION Process;
#else
typedef pid_t Process;
#endif
int spawnProcess(Process *process, const char *app, char * const *args);
int waitProcess(Process *process);
#endif // HELPER_MULTIPROCESS_H

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions

View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions

View File

@@ -1,4 +1,4 @@
Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
@@ -23,3 +23,6 @@ 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.
For additional information on the license terms, see the CUDA EULA at
https://docs.nvidia.com/cuda/eula/index.html

View File

@@ -1,6 +1,6 @@
###############################################################################
#
# Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions

View File

@@ -1,11 +1,37 @@
# CUDA Samples
Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This version supports [CUDA Toolkit 9.2](https://developer.nvidia.com/cuda-downloads).
Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This version supports [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads).
## Release Notes
This section describes the release notes for the CUDA Samples on GitHub only.
### CUDA 10.1 Update 1
* Added `NV12toBGRandResize`. Demonstrates how to convert and resize NV12 frames to BGR planars frames using CUDA in batch.
* Added `EGLStream_CUDA_Interop`. Demonstrates data exchange between CUDA and EGL Streams.
* Added `cuSolverDn_LinearSolver`. Demonstrates cuSolverDN's LU, QR and Cholesky factorization.
* Added support of Visual Studio 2019 to all samples supported on [Windows](#windows-1).
### CUDA 10.1
* Added `immaTensorCoreGemm`. Demonstrates integer GEMM computation using the Warp Matrix Multiply and Accumulate (WMMA) API for integers employing the Tensor Cores.
* Added `simpleIPC`. Demonstrates Inter Process Communication with one process per GPU for computation.
* Added `nvJPEG`. Demonstrates single and batched decoding of jpeg images using NVJPEG Library.
* Added `bandwidthTest`. It measures the memcopy bandwidth of the GPU and memcpy bandwidth across PCI-e.
* Added `reduction`. Demonstrates several important optimization strategies for Data-Parallel Algorithms like reduction.
* Update all the samples to support CUDA 10.1.
### CUDA 10.0
* Added `simpleCudaGraphs`. Demonstrates CUDA Graphs creation, instantiation and launch using Graphs APIs and Stream Capture APIs.
* Added `conjugateGradientCudaGraphs`. Demonstrates conjugate gradient solver on GPU using CUBLAS and CUSPARSE library calls captured and called using CUDA Graph APIs.
* Added `simpleVulkan`. Demonstrates Vulkan - CUDA Interop.
* Added `simpleD3D12`. Demonstrates DX12 - CUDA Interop.
* Added `UnifiedMemoryPerf`. Demonstrates performance comparision of various memory types involved in system.
* Added `p2pBandwidthLatencyTest`. Demonstrates Peer-To-Peer (P2P) data transfers between pairs of GPUs and computes latency and bandwidth.
* Added `systemWideAtomics`. Demonstrates system wide atomic instructions.
* Added `simpleCUBLASXT`. Demonstrates CUBLAS-XT library which performs GEMM operations over multiple GPUs.
* Added Windows OS support to `conjugateGradientMultiDeviceCG` sample.
* Removed support of Visual Studio 2010 from all samples.
### CUDA 9.2
This is the first release of CUDA Samples on GitHub:
@@ -26,7 +52,7 @@ This is the first release of CUDA Samples on GitHub:
### Prerequisites
Download and install the [CUDA Toolkit 9.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
For system requirements and installation instructions of cuda toolkit, please refer to the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/), the [Windows Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html), and the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html).
### Getting the CUDA Samples
@@ -108,22 +134,32 @@ The samples makefiles can take advantage of certain options:
### Samples by OS
#### Linux
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[shfl_scan](./Samples/shfl_scan)** | **[deviceQuery](./Samples/deviceQuery)** |
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[EGLStream_CUDA_Interop](./Samples/EGLStream_CUDA_Interop)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[simpleIPC](./Samples/simpleIPC)** |
---|---|---|---|
**[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** |
**[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[matrixMul](./Samples/matrixMul)** |
**[shfl_scan](./Samples/shfl_scan)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[nvJPEG](./Samples/nvJPEG)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** |
**[deviceQuery](./Samples/deviceQuery)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** |
**[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** |
**[reduction](./Samples/reduction)** | **[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[bandwidthTest](./Samples/bandwidthTest)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** |
**[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[simpleVulkan](./Samples/simpleVulkan)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** |
**[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[matrixMul](./Samples/matrixMul)** | **[systemWideAtomics](./Samples/systemWideAtomics)** |
#### Windows
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[shfl_scan](./Samples/shfl_scan)** | **[deviceQuery](./Samples/deviceQuery)** |
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[simpleIPC](./Samples/simpleIPC)** | **[shfl_scan](./Samples/shfl_scan)** |
---|---|---|---|
**[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** |
**[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[matrixMul](./Samples/matrixMul)** |
**[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[simpleD3D12](./Samples/simpleD3D12)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[deviceQuery](./Samples/deviceQuery)** |
**[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** |
**[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** | **[reduction](./Samples/reduction)** |
**[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[bandwidthTest](./Samples/bandwidthTest)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[NV12toBGRandResize](./Samples/NV12toBGRandResize)** |
**[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[simpleVulkan](./Samples/simpleVulkan)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** |
**[matrixMul](./Samples/matrixMul)** |
#### Mac OSX
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[shfl_scan](./Samples/shfl_scan)** | **[deviceQuery](./Samples/deviceQuery)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** |
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[shfl_scan](./Samples/shfl_scan)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** |
---|---|---|---|
**[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** |
**[matrixMul](./Samples/matrixMul)** |
**[deviceQuery](./Samples/deviceQuery)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** |
**[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** | **[reduction](./Samples/reduction)** |
**[bandwidthTest](./Samples/bandwidthTest)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** |
**[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[matrixMul](./Samples/matrixMul)** |
## Dependencies
@@ -153,6 +189,10 @@ Some samples can only be run on a 64-bit operating system.
DirectX is a collection of APIs designed to allow development of multimedia applications on Microsoft platforms. For Microsoft platforms, NVIDIA's CUDA Driver supports DirectX. Several CUDA Samples for Windows demonstrates CUDA-DirectX Interoperability, for building such samples one needs to install [Direct X SDK (June 2010 or newer)](http://www.microsoft.com/en-us/download/details.aspx?id=6812) , this is required to be installed on Windows 7, Windows 10 and Windows Server 2008, Other Windows OSes do not need to explicitly install the DirectX SDK.
#### DirectX12
DirectX 12 is a collection of advanced low-level programming APIs which can reduce driver overhead, designed to allow development of multimedia applications on Microsoft platforms starting with Windows 10 OS onwards. For Microsoft platforms, NVIDIA's CUDA Driver supports DirectX. Few CUDA Samples for Windows demonstrates CUDA-DirectX12 Interoperability, for building such samples one needs to install [Windows 10 SDK or higher](https://developer.microsoft.com/en-us/windows/downloads/windows-10-sdk), with VS 2015 or VS 2017.
#### OpenGL
OpenGL is a graphics library used for 2D and 3D rendering. On systems which support OpenGL, NVIDIA's OpenGL implementation is provided with the CUDA Driver.
@@ -161,6 +201,10 @@ OpenGL is a graphics library used for 2D and 3D rendering. On systems which supp
OpenGL ES is an embedded systems graphics library used for 2D and 3D rendering. On systems which support OpenGL ES, NVIDIA's OpenGL ES implementation is provided with the CUDA Driver.
#### Vulkan
Vulkan is a low-overhead, cross-platform 3D graphics and compute API. Vulkan targets high-performance realtime 3D graphics applications such as video games and interactive media across all platforms. On systems which support Vulkan, NVIDIA's Vulkan implementation is provided with the CUDA Driver. For building and running Vulkan applications one needs to install the [Vulkan SDK](https://www.lunarg.com/vulkan-sdk/).
#### OpenMP
OpenMP is an API for multiprocessing programming. OpenMP can be installed using your Linux distribution's package manager system. It usually comes preinstalled with GCC. It can also be found at the [OpenMP website](http://openmp.org/).
@@ -211,7 +255,7 @@ CUBLAS (CUDA Basic Linear Algebra Subroutines) is a GPU-accelerated version of t
#### CUDA Interprocess Communication
IPC (Interprocess Communication) allows processes to share device pointers. IPC is only available on Linux x86_64 and ppc64le systems.
IPC (Interprocess Communication) allows processes to share device pointers.
#### CUFFT
@@ -237,6 +281,10 @@ NPP (NVIDIA Performance Primitives) provides GPU-accelerated image, video, and s
NVGRAPH is a GPU-accelerated graph analytics library.
#### NVJPEG
NVJPEG library provides high-performance, GPU accelerated JPEG decoding functionality for image formats commonly used in deep learning and hyperscale multimedia applications.
#### NVRTC
NVRTC (CUDA RunTime Compilation) is a runtime compilation library for CUDA C++.

View File

@@ -0,0 +1,364 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * 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.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
#
################################################################################
#
# Makefile project only supported on Mac OS X and Linux Platforms)
#
################################################################################
# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda
##############################
# start deprecated interface #
##############################
ifeq ($(x86_64),1)
$(info WARNING - x86_64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
TARGET_ARCH ?= x86_64
endif
ifeq ($(ARMv7),1)
$(info WARNING - ARMv7 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=armv7l instead)
TARGET_ARCH ?= armv7l
endif
ifeq ($(aarch64),1)
$(info WARNING - aarch64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
TARGET_ARCH ?= aarch64
endif
ifeq ($(ppc64le),1)
$(info WARNING - ppc64le variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
TARGET_ARCH ?= ppc64le
endif
ifneq ($(GCC),)
$(info WARNING - GCC variable has been deprecated)
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
HOST_COMPILER ?= $(GCC)
endif
ifneq ($(abi),)
$(error ERROR - abi variable has been removed)
endif
############################
# end deprecated interface #
############################
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
endif
else
TARGET_SIZE := $(shell getconf LONG_BIT)
endif
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
TARGET_ARCH = armv7l
endif
# operating system
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif
# host compiler
ifeq ($(TARGET_OS),darwin)
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
HOST_COMPILER ?= clang++
endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
ifeq ($(TARGET_OS),linux)
HOST_COMPILER ?= arm-linux-gnueabihf-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
else ifeq ($(TARGET_OS),android)
HOST_COMPILER ?= arm-linux-androideabi-g++
endif
else ifeq ($(TARGET_ARCH),aarch64)
ifeq ($(TARGET_OS), linux)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
endif
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
# internal flags
NVCCFLAGS := -m${TARGET_SIZE}
CCFLAGS :=
LDFLAGS :=
# build flags
ifeq ($(TARGET_OS),darwin)
LDFLAGS += -rpath $(CUDA_PATH)/lib
CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
LDFLAGS += -pie
CCFLAGS += -fpie -fpic -fexceptions
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif
# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif
# Debug build flags
ifeq ($(dbg),1)
NVCCFLAGS += -g -G
BUILD_TYPE := debug
else
BUILD_TYPE := release
endif
ALL_CCFLAGS :=
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
UBUNTU = $(shell lsb_release -i -s 2>/dev/null | grep -i ubuntu)
SAMPLE_ENABLED := 1
# This sample is not supported on Mac OSX
ifeq ($(TARGET_OS),darwin)
$(info >>> WARNING - EGLStream_CUDA_Interop is not supported on Mac OSX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
# This sample is not supported on ARMv7
ifeq ($(TARGET_ARCH),armv7l)
$(info >>> WARNING - EGLStream_CUDA_Interop is not supported on ARMv7 - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
# This sample is not supported on android
ifeq ($(TARGET_OS),android)
$(info >>> WARNING - EGLStream_CUDA_Interop is not supported on android - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
# Common includes and paths for CUDA
INCLUDES := -I../../Common
LIBRARIES :=
################################################################################
# Makefile include to help find EGL Libraries
include ./findegl.mk
# EGL specific libraries
ifneq ($(TARGET_OS),darwin)
LIBRARIES += -lEGL
endif
ifeq ($(TARGET_OS),darwin)
ALL_LDFLAGS += -Xcompiler -F/Library/Frameworks -Xlinker -framework -Xlinker CUDA
else
ifeq ($(TARGET_ARCH),x86_64)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/lib64/stubs
CUDA_SEARCH_PATH += $(CUDA_PATH)/targets/x86_64-linux/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/armv7-linux-gnueabihf/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-linux/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/armv7-linux-androideabi/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-linux-androideabi/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/ARMv7-linux-QNX/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-qnx/lib/stubs
endif
ifeq ($(TARGET_ARCH),ppc64le)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/ppc64le-linux/lib/stubs
endif
CUDALIB ?= $(shell find -L $(CUDA_SEARCH_PATH) -maxdepth 1 -name libcuda.so 2> /dev/null)
ifeq ("$(CUDALIB)","")
$(info >>> WARNING - libcuda.so not found, CUDA Driver is not installed. Please re-install the driver. <<<)
SAMPLE_ENABLED := 0
else
CUDALIB := $(shell echo $(CUDALIB) | sed "s/ .*//" | sed "s/\/libcuda.so//" )
LIBRARIES += -L$(CUDALIB) -lcuda
endif
endif
ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif
################################################################################
# Target rules
all: build
build: EGLStream_CUDA_Interop
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@echo "Sample will be waived due to the above missing dependencies"
else
@echo "Sample is ready - all dependencies have been met"
endif
cuda_consumer.o:cuda_consumer.cpp
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
cuda_producer.o:cuda_producer.cpp
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
eglstrm_common.o:eglstrm_common.cpp
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
main.o:main.cpp
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
EGLStream_CUDA_Interop: cuda_consumer.o cuda_producer.o eglstrm_common.o main.o
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
$(EXEC) ./EGLStream_CUDA_Interop
clean:
rm -f EGLStream_CUDA_Interop cuda_consumer.o cuda_producer.o eglstrm_common.o main.o
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/EGLStream_CUDA_Interop
clobber: clean

View File

@@ -0,0 +1,76 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
<entry>
<name>EGLStream_CUDA_Interop</name>
<cuda_api_list>
<driver>cuDeviceGet</driver>
<driver>cuDeviceGetAttribute</driver>
<driver>cuDeviceComputeCapability</driver>
<driver>cuDeviceGetCount</driver>
<driver>cuDeviceGetName</driver>
<driver>cuGraphicsResourceGetMappedEglFrame</driver>
<driver>cuEGLStreamConsumerAcquireFrame</driver>
<driver>cuEGLStreamConsumerReleaseFrame</driver>
<driver>cuEGLStreamProducerPresentFrame</driver>
<driver>cuCtxCreate</driver>
<driver>cuMemAlloc</driver>
<driver>cuMemFree</driver>
<driver>cuMemcpy3D</driver>
<driver>cuStreamCreate</driver>
<driver>cuCtxPushCurrent</driver>
<driver>cuCtxPopCurrent</driver>
<driver>cuCtxDestroy</driver>
</cuda_api_list>
<description><![CDATA[Demonstrates data exchange between CUDA and EGL Streams.]]></description>
<devicecompilation>whole</devicecompilation>
<includepaths>
<path>./</path>
<path>../</path>
<path>../../common/inc</path>
</includepaths>
<keyconcepts>
<concept level="basic">EGLStreams Interop</concept>
</keyconcepts>
<keywords>
<keyword>EGL Streams</keyword>
</keywords>
<libraries>
<library os="linux">cuda</library>
<library framework="true" os="macosx">CUDA</library>
</libraries>
<librarypaths>
</librarypaths>
<nsight_eclipse>true</nsight_eclipse>
<primary_file>main.cpp</primary_file>
<required_dependencies>
<dependency>EGL</dependency>
</required_dependencies>
<scopes>
<scope>1:CUDA Basic Topics</scope>
<scope>2:Graphics Interop</scope>
</scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch>
<sm-arch>sm52</sm-arch>
<sm-arch>sm60</sm-arch>
<sm-arch>sm61</sm-arch>
<sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch>
<supported_envs>
<env>
<arch>x86_64</arch>
<platform>linux</platform>
</env>
<env>
<arch>aarch64</arch>
</env>
</supported_envs>
<supported_sm_architectures>
<include>all</include>
</supported_sm_architectures>
<title>EGLStream CUDA Interop</title>
<type>exe</type>
</entry>

View File

@@ -0,0 +1,64 @@
# EGLStream_CUDA_Interop - EGLStream CUDA Interop
## Description
Demonstrates data exchange between CUDA and EGL Streams.
## Key Concepts
EGLStreams Interop
## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
Linux
## Supported CPU Architecture
x86_64, aarch64
## CUDA APIs involved
### [CUDA Driver API](http://docs.nvidia.com/cuda/cuda-driver-api/index.html)
cuDeviceGet, cuDeviceGetAttribute, cuDeviceComputeCapability, cuDeviceGetCount, cuDeviceGetName, cuGraphicsResourceGetMappedEglFrame, cuEGLStreamConsumerAcquireFrame, cuEGLStreamConsumerReleaseFrame, cuEGLStreamProducerPresentFrame, cuCtxCreate, cuMemAlloc, cuMemFree, cuMemcpy3D, cuStreamCreate, cuCtxPushCurrent, cuCtxPopCurrent, cuCtxDestroy
## Dependencies needed to build/run
[EGL](../../README.md#egl)
## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run
### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, aarch64.
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=aarch64` <br/>
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
```
$ make SMS="50 60"
```
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=g++
```
## References (for more details)

View File

@@ -0,0 +1,318 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
//
// DESCRIPTION: Simple CUDA consumer rendering sample app
//
#include "cuda_consumer.h"
#include <helper_cuda_drvapi.h>
#include "eglstrm_common.h"
#if defined(EXTENSION_LIST)
EXTENSION_LIST(EXTLST_EXTERN)
#endif
int checkbuf(FILE *fp1, FILE *fp2);
CUresult cudaConsumerTest(test_cuda_consumer_s *data, char *fileName) {
CUresult cuStatus = CUDA_SUCCESS;
CUarray cudaArr = NULL;
CUeglFrame cudaEgl;
CUgraphicsResource cudaResource;
unsigned int i;
int check_result;
FILE *pInFile1 = NULL, *pInFile2 = NULL, *file_p = NULL;
EGLint streamState = 0;
if (!data) {
printf("%s: Bad parameter\n", __func__);
goto done;
}
if (!eglQueryStreamKHR(g_display, eglStream, EGL_STREAM_STATE_KHR,
&streamState)) {
printf("Cuda consumer, eglQueryStreamKHR EGL_STREAM_STATE_KHR failed\n");
}
if (streamState == EGL_STREAM_STATE_DISCONNECTED_KHR) {
printf("CUDA Consumer: - EGL_STREAM_STATE_DISCONNECTED_KHR received\n");
}
if (streamState == EGL_STREAM_STATE_NEW_FRAME_AVAILABLE_KHR) {
cuStatus = cuEGLStreamConsumerAcquireFrame(&(data->cudaConn), &cudaResource,
NULL, 16000);
if (cuStatus == CUDA_SUCCESS) {
CUdeviceptr pDevPtr = 0;
int bufferSize;
unsigned char *pCudaCopyMem = NULL;
unsigned int copyWidthInBytes = 0, copyHeight = 0;
file_p = fopen(fileName, "wb+");
if (!file_p) {
printf("WriteFrame: file open failed %s\n", fileName);
cuStatus = CUDA_ERROR_UNKNOWN;
goto done;
}
cuStatus =
cuGraphicsResourceGetMappedEglFrame(&cudaEgl, cudaResource, 0, 0);
if (cuStatus != CUDA_SUCCESS) {
printf("Cuda get resource failed with %d\n", cuStatus);
goto done;
}
cuStatus = cuCtxSynchronize();
if (cuStatus != CUDA_SUCCESS) {
printf("cuCtxSynchronize failed \n");
goto done;
}
if (!(cudaEgl.planeCount >= 1 && cudaEgl.planeCount <= 3)) {
printf("Plane count is invalid\nExiting\n");
goto done;
}
for (i = 0; i < cudaEgl.planeCount; i++) {
if (cudaEgl.frameType == CU_EGL_FRAME_TYPE_PITCH) {
pDevPtr = (CUdeviceptr)cudaEgl.frame.pPitch[i];
if (cudaEgl.planeCount == 1) {
bufferSize = cudaEgl.pitch * cudaEgl.height;
copyWidthInBytes = cudaEgl.pitch;
copyHeight = data->height;
} else if (i == 1 && cudaEgl.planeCount == 2) { // YUV 420
// semi-planar
bufferSize = cudaEgl.pitch * cudaEgl.height / 2;
copyWidthInBytes = cudaEgl.pitch;
copyHeight = data->height / 2;
} else {
bufferSize = data->width * data->height;
copyWidthInBytes = data->width;
copyHeight = data->height;
if (i > 0) {
bufferSize >>= 2;
copyWidthInBytes >>= 1;
copyHeight >>= 1;
}
}
} else {
cudaArr = cudaEgl.frame.pArray[i];
if (cudaEgl.planeCount == 1) {
bufferSize = data->width * data->height * 4;
copyWidthInBytes = data->width * 4;
copyHeight = data->height;
} else if (i == 1 && cudaEgl.planeCount == 2) { // YUV 420
// semi-planar
bufferSize = data->width * data->height / 2;
copyWidthInBytes = data->width;
copyHeight = data->height / 2;
} else {
bufferSize = data->width * data->height;
copyWidthInBytes = data->width;
copyHeight = data->height;
if (i > 0) {
bufferSize >>= 2;
copyWidthInBytes >>= 1;
copyHeight >>= 1;
}
}
}
if (i == 0) {
pCudaCopyMem = (unsigned char *)malloc(bufferSize);
if (pCudaCopyMem == NULL) {
printf("pCudaCopyMem malloc failed\n");
goto done;
}
}
memset(pCudaCopyMem, 0, bufferSize);
if (data->pitchLinearOutput) {
cuStatus = cuMemcpyDtoH(pCudaCopyMem, pDevPtr, bufferSize);
if (cuStatus != CUDA_SUCCESS) {
printf(
"cuda_consumer: pitch linear Memcpy failed, bufferSize =%d\n",
bufferSize);
goto done;
}
cuStatus = cuCtxSynchronize();
if (cuStatus != CUDA_SUCCESS) {
printf("cuda_consumer: cuCtxSynchronize failed after memcpy \n");
goto done;
}
} else {
CUDA_MEMCPY3D cpdesc;
memset(&cpdesc, 0, sizeof(cpdesc));
cpdesc.srcXInBytes = cpdesc.srcY = cpdesc.srcZ = cpdesc.srcLOD = 0;
cpdesc.srcMemoryType = CU_MEMORYTYPE_ARRAY;
cpdesc.srcArray = cudaArr;
cpdesc.dstXInBytes = cpdesc.dstY = cpdesc.dstZ = cpdesc.dstLOD = 0;
cpdesc.dstMemoryType = CU_MEMORYTYPE_HOST;
cpdesc.dstHost = (void *)pCudaCopyMem;
cpdesc.WidthInBytes = copyWidthInBytes; // data->width * 4;
cpdesc.Height = copyHeight; // data->height;
cpdesc.Depth = 1;
cuStatus = cuMemcpy3D(&cpdesc);
if (cuStatus != CUDA_SUCCESS) {
printf(
"Cuda consumer: cuMemCpy3D failed, copyWidthInBytes=%d, "
"copyHight=%d\n",
copyWidthInBytes, copyHeight);
}
cuStatus = cuCtxSynchronize();
if (cuStatus != CUDA_SUCCESS) {
printf("cuCtxSynchronize failed after memcpy \n");
}
}
if (cuStatus == CUDA_SUCCESS) {
if (fwrite(pCudaCopyMem, bufferSize, 1, file_p) != 1) {
printf("Cuda consumer: output file write failed\n");
cuStatus = CUDA_ERROR_UNKNOWN;
goto done;
}
}
}
pInFile1 = fopen(data->fileName1, "rb");
if (!pInFile1) {
printf("Failed to open file :%s\n", data->fileName1);
goto done;
}
pInFile2 = fopen(data->fileName2, "rb");
if (!pInFile2) {
printf("Failed to open file :%s\n", data->fileName2);
goto done;
}
rewind(file_p);
check_result = checkbuf(file_p, pInFile1);
if (check_result == -1) {
rewind(file_p);
check_result = checkbuf(file_p, pInFile2);
if (check_result == -1) {
printf("Frame received does not match any valid image: FAILED\n");
} else {
printf("Frame check Passed\n");
}
} else {
printf("Frame check Passed\n");
}
if (pCudaCopyMem) {
free(pCudaCopyMem);
pCudaCopyMem = NULL;
}
cuStatus =
cuEGLStreamConsumerReleaseFrame(&data->cudaConn, cudaResource, NULL);
if (cuStatus != CUDA_SUCCESS) {
printf("cuEGLStreamConsumerReleaseFrame failed with cuStatus = %d\n",
cuStatus);
goto done;
}
} else {
printf("cuda AcquireFrame FAILED with cuStatus=%d\n", cuStatus);
goto done;
}
}
done:
if (file_p) {
fclose(file_p);
file_p = NULL;
}
if (pInFile1) {
fclose(pInFile1);
pInFile1 = NULL;
}
if (pInFile1) {
fclose(pInFile2);
pInFile2 = NULL;
}
return cuStatus;
}
int checkbuf(FILE *fp1, FILE *fp2) {
int match = 0;
int ch1, ch2;
if (fp1 == NULL) {
printf("Invalid file pointer for first file\n");
return -1;
} else if (fp2 == NULL) {
printf("Invalid file pointer for second file\n");
return -1;
} else {
ch1 = getc(fp1);
ch2 = getc(fp2);
while ((ch1 != EOF) && (ch2 != EOF) && (ch1 == ch2)) {
ch1 = getc(fp1);
ch2 = getc(fp2);
}
if (ch1 == ch2) {
match = 1;
} else if (ch1 != ch2) {
match = -1;
}
}
return match;
}
CUresult cudaDeviceCreateConsumer(test_cuda_consumer_s *cudaConsumer,
CUdevice device) {
CUresult status = CUDA_SUCCESS;
if (CUDA_SUCCESS != (status = cuInit(0))) {
printf("Failed to initialize CUDA\n");
return status;
}
int major = 0, minor = 0;
char deviceName[256];
checkCudaErrors(cuDeviceGetAttribute(
&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, device));
checkCudaErrors(cuDeviceGetAttribute(
&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, device));
checkCudaErrors(cuDeviceGetName(deviceName, 256, device));
printf(
"CUDA Consumer on GPU Device %d: \"%s\" with compute capability "
"%d.%d\n\n",
device, deviceName, major, minor);
if (CUDA_SUCCESS !=
(status = cuCtxCreate(&cudaConsumer->context, 0, device))) {
printf("failed to create CUDA context\n");
return status;
}
checkCudaErrors(cuCtxPopCurrent(&cudaConsumer->context));
return status;
}
void cuda_consumer_init(test_cuda_consumer_s *cudaConsumer, TestArgs *args) {
cudaConsumer->pitchLinearOutput = args->pitchLinearOutput;
cudaConsumer->width = args->inputWidth;
cudaConsumer->height = args->inputHeight;
cudaConsumer->fileName1 = args->infile1;
cudaConsumer->fileName2 = args->infile2;
cudaConsumer->outFile1 = "cuda_out1.yuv";
cudaConsumer->outFile2 = "cuda_out2.yuv";
}
CUresult cuda_consumer_deinit(test_cuda_consumer_s *cudaConsumer) {
return cuEGLStreamConsumerDisconnect(&cudaConsumer->cudaConn);
}

View File

@@ -0,0 +1,62 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
//
// DESCRIPTION: CUDA consumer header file
//
#ifndef _CUDA_CONSUMER_H_
#define _CUDA_CONSUMER_H_
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "cudaEGL.h"
#include "eglstrm_common.h"
extern EGLStreamKHR eglStream;
extern EGLDisplay g_display;
typedef struct _test_cuda_consumer_s {
CUcontext context;
CUeglStreamConnection cudaConn;
bool pitchLinearOutput;
unsigned int width;
unsigned int height;
char *fileName1;
char *fileName2;
char *outFile1;
char *outFile2;
unsigned int frameCount;
} test_cuda_consumer_s;
void cuda_consumer_init(test_cuda_consumer_s *cudaConsumer, TestArgs *args);
CUresult cuda_consumer_deinit(test_cuda_consumer_s *cudaConsumer);
CUresult cudaConsumerTest(test_cuda_consumer_s *data, char *outFile);
CUresult cudaDeviceCreateConsumer(test_cuda_consumer_s *cudaConsumer,
CUdevice device);
#endif

Binary file not shown.

Binary file not shown.

View File

@@ -0,0 +1,381 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
//
// DESCRIPTION: Simple cuda EGL stream producer app
//
#include "cuda_producer.h"
#include <helper_cuda_drvapi.h>
#include "cudaEGL.h"
#include "eglstrm_common.h"
#if defined(EXTENSION_LIST)
EXTENSION_LIST(EXTLST_EXTERN)
#endif
static CUresult cudaProducerReadYUVFrame(FILE *file, unsigned int frameNum,
unsigned int width,
unsigned int height,
unsigned char *pBuff) {
int bOrderUV = 0;
unsigned char *pYBuff, *pUBuff, *pVBuff, *pChroma;
unsigned int frameSize = (width * height * 3) / 2;
CUresult ret = CUDA_SUCCESS;
unsigned int i;
if (!pBuff || !file) return CUDA_ERROR_FILE_NOT_FOUND;
pYBuff = pBuff;
// YVU order in the buffer
pVBuff = pYBuff + width * height;
pUBuff = pVBuff + width * height / 4;
if (fseek(file, frameNum * frameSize, SEEK_SET)) {
printf("ReadYUVFrame: Error seeking file: %p\n", file);
ret = CUDA_ERROR_NOT_PERMITTED;
goto done;
}
// read Y U V separately
for (i = 0; i < height; i++) {
if (fread(pYBuff, width, 1, file) != 1) {
printf("ReadYUVFrame: Error reading file: %p\n", file);
ret = CUDA_ERROR_NOT_PERMITTED;
goto done;
}
pYBuff += width;
}
pChroma = bOrderUV ? pUBuff : pVBuff;
for (i = 0; i < height / 2; i++) {
if (fread(pChroma, width / 2, 1, file) != 1) {
printf("ReadYUVFrame: Error reading file: %p\n", file);
ret = CUDA_ERROR_NOT_PERMITTED;
goto done;
}
pChroma += width / 2;
}
pChroma = bOrderUV ? pVBuff : pUBuff;
for (i = 0; i < height / 2; i++) {
if (fread(pChroma, width / 2, 1, file) != 1) {
printf("ReadYUVFrame: Error reading file: %p\n", file);
ret = CUDA_ERROR_NOT_PERMITTED;
goto done;
}
pChroma += width / 2;
}
done:
return ret;
}
static CUresult cudaProducerReadARGBFrame(FILE *file, unsigned int frameNum,
unsigned int width,
unsigned int height,
unsigned char *pBuff) {
unsigned int frameSize = width * height * 4;
CUresult ret = CUDA_SUCCESS;
if (!pBuff || !file) return CUDA_ERROR_FILE_NOT_FOUND;
if (fseek(file, frameNum * frameSize, SEEK_SET)) {
printf("ReadYUVFrame: Error seeking file: %p\n", file);
ret = CUDA_ERROR_NOT_PERMITTED;
goto done;
}
// read ARGB data
if (fread(pBuff, frameSize, 1, file) != 1) {
if (feof(file))
printf("ReadARGBFrame: file read to the end\n");
else
printf("ReadARGBFrame: Error reading file: %p\n", file);
ret = CUDA_ERROR_NOT_PERMITTED;
goto done;
}
done:
return ret;
}
CUresult cudaProducerTest(test_cuda_producer_s *cudaProducer, char *file) {
int framenum = 0;
CUarray cudaArr[3] = {0};
CUdeviceptr cudaPtr[3] = {0, 0, 0};
unsigned int bufferSize;
CUresult cuStatus = CUDA_SUCCESS;
unsigned int i, surfNum, uvOffset[3] = {0};
unsigned int copyWidthInBytes[3] = {0, 0, 0}, copyHeight[3] = {0, 0, 0};
CUeglColorFormat eglColorFormat;
FILE *file_p;
CUeglFrame cudaEgl;
CUcontext oldContext;
file_p = fopen(file, "rb");
if (!file_p) {
printf("CudaProducer: Error opening file: %s\n", file);
goto done;
}
if (cudaProducer->pitchLinearOutput) {
if (cudaProducer->isARGB) {
cudaPtr[0] = cudaProducer->cudaPtrARGB[0];
} else { // YUV case
for (i = 0; i < 3; i++) {
if (i == 0) {
bufferSize = cudaProducer->width * cudaProducer->height;
} else {
bufferSize = cudaProducer->width * cudaProducer->height / 4;
}
cudaPtr[i] = cudaProducer->cudaPtrYUV[i];
}
}
} else {
if (cudaProducer->isARGB) {
cudaArr[0] = cudaProducer->cudaArrARGB[0];
} else {
for (i = 0; i < 3; i++) {
cudaArr[i] = cudaProducer->cudaArrYUV[i];
}
}
}
uvOffset[0] = 0;
if (cudaProducer->isARGB) {
if (CUDA_SUCCESS !=
cudaProducerReadARGBFrame(file_p, framenum, cudaProducer->width,
cudaProducer->height, cudaProducer->pBuff)) {
printf("cuda producer, read ARGB frame failed\n");
goto done;
}
copyWidthInBytes[0] = cudaProducer->width * 4;
copyHeight[0] = cudaProducer->height;
surfNum = 1;
eglColorFormat = CU_EGL_COLOR_FORMAT_ARGB;
} else {
if (CUDA_SUCCESS !=
cudaProducerReadYUVFrame(file_p, framenum, cudaProducer->width,
cudaProducer->height, cudaProducer->pBuff)) {
printf("cuda producer, reading YUV frame failed\n");
goto done;
}
surfNum = 3;
eglColorFormat = CU_EGL_COLOR_FORMAT_YUV420_PLANAR;
copyWidthInBytes[0] = cudaProducer->width;
copyHeight[0] = cudaProducer->height;
copyWidthInBytes[1] = cudaProducer->width / 2;
copyHeight[1] = cudaProducer->height / 2;
copyWidthInBytes[2] = cudaProducer->width / 2;
copyHeight[2] = cudaProducer->height / 2;
uvOffset[1] = cudaProducer->width * cudaProducer->height;
uvOffset[2] =
uvOffset[1] + cudaProducer->width / 2 * cudaProducer->height / 2;
}
if (cudaProducer->pitchLinearOutput) {
for (i = 0; i < surfNum; i++) {
cuStatus =
cuMemcpy(cudaPtr[i], (CUdeviceptr)(cudaProducer->pBuff + uvOffset[i]),
copyWidthInBytes[i] * copyHeight[i]);
if (cuStatus != CUDA_SUCCESS) {
printf("Cuda producer: cuMemCpy pitchlinear failed, cuStatus =%d\n",
cuStatus);
goto done;
}
}
} else {
// copy cudaProducer->pBuff to cudaArray
CUDA_MEMCPY3D cpdesc;
for (i = 0; i < surfNum; i++) {
memset(&cpdesc, 0, sizeof(cpdesc));
cpdesc.srcXInBytes = cpdesc.srcY = cpdesc.srcZ = cpdesc.srcLOD = 0;
cpdesc.srcMemoryType = CU_MEMORYTYPE_HOST;
cpdesc.srcHost = (void *)(cudaProducer->pBuff + uvOffset[i]);
cpdesc.dstXInBytes = cpdesc.dstY = cpdesc.dstZ = cpdesc.dstLOD = 0;
cpdesc.dstMemoryType = CU_MEMORYTYPE_ARRAY;
cpdesc.dstArray = cudaArr[i];
cpdesc.WidthInBytes = copyWidthInBytes[i];
cpdesc.Height = copyHeight[i];
cpdesc.Depth = 1;
cuStatus = cuMemcpy3D(&cpdesc);
if (cuStatus != CUDA_SUCCESS) {
printf("Cuda producer: cuMemCpy failed, cuStatus =%d\n", cuStatus);
goto done;
}
}
}
for (i = 0; i < surfNum; i++) {
if (cudaProducer->pitchLinearOutput)
cudaEgl.frame.pPitch[i] = (void *)cudaPtr[i];
else
cudaEgl.frame.pArray[i] = cudaArr[i];
}
cudaEgl.width = copyWidthInBytes[0];
cudaEgl.depth = 1;
cudaEgl.height = copyHeight[0];
cudaEgl.pitch = cudaProducer->pitchLinearOutput ? cudaEgl.width : 0;
cudaEgl.frameType = cudaProducer->pitchLinearOutput ? CU_EGL_FRAME_TYPE_PITCH
: CU_EGL_FRAME_TYPE_ARRAY;
cudaEgl.planeCount = surfNum;
cudaEgl.numChannels = (eglColorFormat == CU_EGL_COLOR_FORMAT_ARGB) ? 4 : 1;
cudaEgl.eglColorFormat = eglColorFormat;
cudaEgl.cuFormat = CU_AD_FORMAT_UNSIGNED_INT8;
cuStatus =
cuEGLStreamProducerPresentFrame(&cudaProducer->cudaConn, cudaEgl, NULL);
if (cuStatus != CUDA_SUCCESS) {
printf("cuda Producer present frame FAILED with custatus= %d\n", cuStatus);
goto done;
}
done:
if (file_p) {
fclose(file_p);
file_p = NULL;
}
return cuStatus;
}
CUresult cudaDeviceCreateProducer(test_cuda_producer_s *cudaProducer,
CUdevice device) {
CUresult status = CUDA_SUCCESS;
if (CUDA_SUCCESS != (status = cuInit(0))) {
printf("Failed to initialize CUDA\n");
return status;
}
int major = 0, minor = 0;
char deviceName[256];
checkCudaErrors(cuDeviceGetAttribute(
&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, device));
checkCudaErrors(cuDeviceGetAttribute(
&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, device));
checkCudaErrors(cuDeviceGetName(deviceName, 256, device));
printf(
"CUDA Producer on GPU Device %d: \"%s\" with compute capability "
"%d.%d\n\n",
device, deviceName, major, minor);
if (CUDA_SUCCESS !=
(status = cuCtxCreate(&cudaProducer->context, 0, device))) {
printf("failed to create CUDA context\n");
return status;
}
status = cuMemAlloc(&cudaProducer->cudaPtrARGB[0], (WIDTH * HEIGHT * 4));
if (status != CUDA_SUCCESS) {
printf("Create CUDA pointer failed, cuStatus=%d\n", status);
return status;
}
status = cuMemAlloc(&cudaProducer->cudaPtrYUV[0], (WIDTH * HEIGHT));
if (status != CUDA_SUCCESS) {
printf("Create CUDA pointer failed, cuStatus=%d\n", status);
return status;
}
status = cuMemAlloc(&cudaProducer->cudaPtrYUV[1], (WIDTH * HEIGHT) / 4);
if (status != CUDA_SUCCESS) {
printf("Create CUDA pointer failed, cuStatus=%d\n", status);
return status;
}
status = cuMemAlloc(&cudaProducer->cudaPtrYUV[2], (WIDTH * HEIGHT) / 4);
if (status != CUDA_SUCCESS) {
printf("Create CUDA pointer failed, cuStatus=%d\n", status);
return status;
}
CUDA_ARRAY3D_DESCRIPTOR desc = {0};
desc.Format = CU_AD_FORMAT_UNSIGNED_INT8;
desc.Depth = 1;
desc.Flags = CUDA_ARRAY3D_SURFACE_LDST;
desc.NumChannels = 4;
desc.Width = WIDTH * 4;
desc.Height = HEIGHT;
status = cuArray3DCreate(&cudaProducer->cudaArrARGB[0], &desc);
if (status != CUDA_SUCCESS) {
printf("Create CUDA array failed, cuStatus=%d\n", status);
return status;
}
for (int i = 0; i < 3; i++) {
if (i == 0) {
desc.NumChannels = 1;
desc.Width = WIDTH;
desc.Height = HEIGHT;
} else { // U/V surface as planar
desc.NumChannels = 1;
desc.Width = WIDTH / 2;
desc.Height = HEIGHT / 2;
}
status = cuArray3DCreate(&cudaProducer->cudaArrYUV[i], &desc);
if (status != CUDA_SUCCESS) {
printf("Create CUDA array failed, cuStatus=%d\n", status);
return status;
}
}
cudaProducer->pBuff = (unsigned char *)malloc((WIDTH * HEIGHT * 4));
if (!cudaProducer->pBuff) {
printf("CudaProducer: Failed to allocate image buffer\n");
}
checkCudaErrors(cuCtxPopCurrent(&cudaProducer->context));
return status;
}
void cudaProducerInit(test_cuda_producer_s *cudaProducer, EGLDisplay eglDisplay,
EGLStreamKHR eglStream, TestArgs *args) {
cudaProducer->fileName1 = args->infile1;
cudaProducer->fileName2 = args->infile2;
cudaProducer->frameCount = 2;
cudaProducer->width = args->inputWidth;
cudaProducer->height = args->inputHeight;
cudaProducer->isARGB = args->isARGB;
cudaProducer->pitchLinearOutput = args->pitchLinearOutput;
// Set cudaProducer default parameters
cudaProducer->eglDisplay = eglDisplay;
cudaProducer->eglStream = eglStream;
}
CUresult cudaProducerDeinit(test_cuda_producer_s *cudaProducer) {
if (cudaProducer->pBuff) free(cudaProducer->pBuff);
checkCudaErrors(cuMemFree(cudaProducer->cudaPtrARGB[0]));
checkCudaErrors(cuMemFree(cudaProducer->cudaPtrYUV[0]));
checkCudaErrors(cuMemFree(cudaProducer->cudaPtrYUV[1]));
checkCudaErrors(cuMemFree(cudaProducer->cudaPtrYUV[2]));
checkCudaErrors(cuArrayDestroy(cudaProducer->cudaArrARGB[0]));
checkCudaErrors(cuArrayDestroy(cudaProducer->cudaArrYUV[0]));
checkCudaErrors(cuArrayDestroy(cudaProducer->cudaArrYUV[1]));
checkCudaErrors(cuArrayDestroy(cudaProducer->cudaArrYUV[2]));
return cuEGLStreamProducerDisconnect(&cudaProducer->cudaConn);
}

View File

@@ -0,0 +1,68 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
//
// DESCRIPTION: Simple cuda producer header file
//
#ifndef _CUDA_PRODUCER_H_
#define _CUDA_PRODUCER_H_
#include <EGL/egl.h>
#include <EGL/eglext.h>
#include "cudaEGL.h"
#include "eglstrm_common.h"
extern EGLStreamKHR eglStream;
extern EGLDisplay g_display;
typedef struct _test_cuda_producer_s {
// Stream params
char *fileName1;
char *fileName2;
unsigned char *pBuff;
int frameCount;
bool isARGB;
bool pitchLinearOutput;
unsigned int width;
unsigned int height;
CUcontext context;
CUeglStreamConnection cudaConn;
CUdeviceptr cudaPtrARGB[1];
CUdeviceptr cudaPtrYUV[3];
CUarray cudaArrARGB[1];
CUarray cudaArrYUV[3];
EGLStreamKHR eglStream;
EGLDisplay eglDisplay;
} test_cuda_producer_s;
void cudaProducerInit(test_cuda_producer_s *cudaProducer, EGLDisplay eglDisplay,
EGLStreamKHR eglStream, TestArgs *args);
CUresult cudaProducerTest(test_cuda_producer_s *parserArg, char *file);
CUresult cudaProducerDeinit(test_cuda_producer_s *cudaProducer);
CUresult cudaDeviceCreateProducer(test_cuda_producer_s *cudaProducer,
CUdevice device);
#endif

Binary file not shown.

Binary file not shown.

View File

@@ -0,0 +1,139 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
//
// DESCRIPTION: Common egl stream functions
//
#include "eglstrm_common.h"
EGLStreamKHR eglStream;
EGLDisplay g_display;
EGLAttrib cudaIndex;
#if defined(EXTENSION_LIST)
EXTENSION_LIST(EXTLST_DECL)
typedef void (*extlst_fnptr_t)(void);
static struct {
extlst_fnptr_t *fnptr;
char const *name;
} extensionList[] = {EXTENSION_LIST(EXTLST_ENTRY)};
int eglSetupExtensions(void) {
unsigned int i;
for (i = 0; i < (sizeof(extensionList) / sizeof(*extensionList)); i++) {
*extensionList[i].fnptr = eglGetProcAddress(extensionList[i].name);
if (*extensionList[i].fnptr == NULL) {
printf("Couldn't get address of %s()\n", extensionList[i].name);
return 0;
}
}
return 1;
}
int EGLStreamInit(int *cuda_device) {
static const EGLint streamAttrMailboxMode[] = {EGL_SUPPORT_REUSE_NV,
EGL_FALSE, EGL_NONE};
EGLBoolean eglStatus;
#define MAX_EGL_DEVICES 4
EGLint numDevices = 0;
EGLDeviceEXT devices[MAX_EGL_DEVICES];
eglStatus = eglQueryDevicesEXT(MAX_EGL_DEVICES, devices, &numDevices);
if (eglStatus != EGL_TRUE) {
printf("Error querying EGL devices\n");
exit(EXIT_FAILURE);
}
if (numDevices == 0) {
printf("No EGL devices found.. Waiving\n");
eglStatus = EGL_FALSE;
exit(EXIT_WAIVED);
}
int egl_device_id = 0;
for (egl_device_id = 0; egl_device_id < numDevices; egl_device_id++) {
eglStatus = eglQueryDeviceAttribEXT(devices[egl_device_id],
EGL_CUDA_DEVICE_NV, &cudaIndex);
if (eglStatus == EGL_TRUE) {
*cuda_device = cudaIndex; // We select first EGL-CUDA Capable device.
printf("Found EGL-CUDA Capable device with CUDA Device id = %d\n",
(int)cudaIndex);
break;
}
}
if (egl_device_id >= numDevices) {
printf("No CUDA Capable EGL Device found.. Waiving execution\n");
exit(EXIT_WAIVED);
}
g_display = eglGetPlatformDisplayEXT(EGL_PLATFORM_DEVICE_EXT,
(void *)devices[egl_device_id], NULL);
if (g_display == EGL_NO_DISPLAY) {
printf("Could not get EGL display from device. \n");
eglStatus = EGL_FALSE;
exit(EXIT_FAILURE);
}
eglStatus = eglInitialize(g_display, 0, 0);
if (!eglStatus) {
printf("EGL failed to initialize. \n");
eglStatus = EGL_FALSE;
exit(EXIT_FAILURE);
}
eglStream = eglCreateStreamKHR(g_display, streamAttrMailboxMode);
if (eglStream == EGL_NO_STREAM_KHR) {
printf("Could not create EGL stream.\n");
eglStatus = EGL_FALSE;
exit(EXIT_FAILURE);
}
printf("Created EGLStream %p\n", eglStream);
// Set stream attribute
if (!eglStreamAttribKHR(g_display, eglStream, EGL_CONSUMER_LATENCY_USEC_KHR,
16000)) {
printf(
"Consumer: eglStreamAttribKHR EGL_CONSUMER_LATENCY_USEC_KHR failed\n");
return 0;
}
if (!eglStreamAttribKHR(g_display, eglStream,
EGL_CONSUMER_ACQUIRE_TIMEOUT_USEC_KHR, 16000)) {
printf(
"Consumer: eglStreamAttribKHR EGL_CONSUMER_ACQUIRE_TIMEOUT_USEC_KHR "
"failed\n");
return 0;
}
printf("EGLStream initialized\n");
return 1;
}
void EGLStreamFini(void) { eglDestroyStreamKHR(g_display, eglStream); }
#endif

View File

@@ -0,0 +1,103 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
//
// DESCRIPTION: Common EGL stream functions header file
//
#ifndef _EGLSTRM_COMMON_H_
#define _EGLSTRM_COMMON_H_
#include <signal.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <unistd.h>
#include "cuda.h"
#include "cudaEGL.h"
#include "helper_cuda_drvapi.h"
#define EXTENSION_LIST(T) \
T(PFNEGLCREATESTREAMKHRPROC, eglCreateStreamKHR) \
T(PFNEGLDESTROYSTREAMKHRPROC, eglDestroyStreamKHR) \
T(PFNEGLQUERYSTREAMKHRPROC, eglQueryStreamKHR) \
T(PFNEGLQUERYSTREAMU64KHRPROC, eglQueryStreamu64KHR) \
T(PFNEGLQUERYSTREAMTIMEKHRPROC, eglQueryStreamTimeKHR) \
T(PFNEGLSTREAMATTRIBKHRPROC, eglStreamAttribKHR) \
T(PFNEGLSTREAMCONSUMERACQUIREKHRPROC, eglStreamConsumerAcquireKHR) \
T(PFNEGLSTREAMCONSUMERRELEASEKHRPROC, eglStreamConsumerReleaseKHR) \
T(PFNEGLSTREAMCONSUMERGLTEXTUREEXTERNALKHRPROC, \
eglStreamConsumerGLTextureExternalKHR) \
T(PFNEGLGETSTREAMFILEDESCRIPTORKHRPROC, eglGetStreamFileDescriptorKHR) \
T(PFNEGLQUERYDEVICESEXTPROC, eglQueryDevicesEXT) \
T(PFNEGLGETPLATFORMDISPLAYEXTPROC, eglGetPlatformDisplayEXT) \
T(PFNEGLQUERYDEVICEATTRIBEXTPROC, eglQueryDeviceAttribEXT) \
T(PFNEGLCREATESTREAMFROMFILEDESCRIPTORKHRPROC, \
eglCreateStreamFromFileDescriptorKHR)
#define eglCreateStreamKHR my_eglCreateStreamKHR
#define eglDestroyStreamKHR my_eglDestroyStreamKHR
#define eglQueryStreamKHR my_eglQueryStreamKHR
#define eglQueryStreamu64KHR my_eglQueryStreamu64KHR
#define eglQueryStreamTimeKHR my_eglQueryStreamTimeKHR
#define eglStreamAttribKHR my_eglStreamAttribKHR
#define eglStreamConsumerAcquireKHR my_eglStreamConsumerAcquireKHR
#define eglStreamConsumerReleaseKHR my_eglStreamConsumerReleaseKHR
#define eglStreamConsumerGLTextureExternalKHR \
my_eglStreamConsumerGLTextureExternalKHR
#define eglGetStreamFileDescriptorKHR my_eglGetStreamFileDescriptorKHR
#define eglCreateStreamFromFileDescriptorKHR \
my_eglCreateStreamFromFileDescriptorKHR
#define eglQueryDevicesEXT my_eglQueryDevicesEXT
#define eglGetPlatformDisplayEXT my_eglGetPlatformDisplayEXT
#define eglQueryDeviceAttribEXT my_eglQueryDeviceAttribEXT
#define EXTLST_DECL(tx, x) tx my_##x = NULL;
#define EXTLST_EXTERN(tx, x) extern tx my_##x;
#define EXTLST_ENTRY(tx, x) {(extlst_fnptr_t *)&my_##x, #x},
#define MAX_STRING_SIZE 256
#define WIDTH 720
#define HEIGHT 480
typedef struct _TestArgs {
char *infile1;
char *infile2;
bool isARGB;
unsigned int inputWidth;
unsigned int inputHeight;
bool pitchLinearOutput;
} TestArgs;
int eglSetupExtensions(void);
int EGLStreamInit(int *dev);
void EGLStreamFini(void);
#endif

View File

@@ -0,0 +1,156 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * 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.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
#
################################################################################
#
# findegl.mk is used to find the necessary EGL Libraries for specific distributions
# this is supported on Linux
#
################################################################################
# Determine OS platform and unix distribution
ifeq ("$(TARGET_OS)","linux")
# first search lsb_release
DISTRO = $(shell lsb_release -i -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
ifeq ("$(DISTRO)","")
# second search and parse /etc/issue
DISTRO = $(shell more /etc/issue | awk '{print $$1}' | sed '1!d' | sed -e "/^$$/d" 2>/dev/null | tr "[:upper:]" "[:lower:]")
# ensure data from /etc/issue is valid
ifeq (,$(filter $(DISTRO),ubuntu fedora red rhel centos suse))
DISTRO =
endif
ifeq ("$(DISTRO)","")
# third, we can search in /etc/os-release or /etc/{distro}-release
DISTRO = $(shell awk '/ID/' /etc/*-release | sed 's/ID=//' | grep -v "VERSION" | grep -v "ID" | grep -v "DISTRIB")
endif
endif
endif
ifeq ("$(TARGET_OS)","linux")
# $(info) >> findegl.mk -> LINUX path <<<)
# Each set of Linux Distros have different paths for where to find their OpenGL libraries reside
UBUNTU = $(shell echo $(DISTRO) | grep -i ubuntu >/dev/null 2>&1; echo $$?)
FEDORA = $(shell echo $(DISTRO) | grep -i fedora >/dev/null 2>&1; echo $$?)
RHEL = $(shell echo $(DISTRO) | grep -i 'red\|rhel' >/dev/null 2>&1; echo $$?)
CENTOS = $(shell echo $(DISTRO) | grep -i centos >/dev/null 2>&1; echo $$?)
SUSE = $(shell echo $(DISTRO) | grep -i 'suse\|sles' >/dev/null 2>&1; echo $$?)
ifeq ("$(UBUNTU)","0")
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
GLPATH := /usr/arm-linux-gnueabihf/lib
GLLINK := -L/usr/arm-linux-gnueabihf/lib
ifneq ($(TARGET_FS),)
GLPATH += $(TARGET_FS)/usr/lib/arm-linux-gnueabihf
GLLINK += -L$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-aarch64)
GLPATH := /usr/aarch64-linux-gnu/lib
GLLINK := -L/usr/aarch64-linux-gnu/lib
ifneq ($(TARGET_FS),)
GLPATH += $(TARGET_FS)/usr/lib
GLPATH += $(TARGET_FS)/usr/lib/aarch64-linux-gnu
GLLINK += -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
endif
else
UBUNTU_PKG_NAME = $(shell which dpkg >/dev/null 2>&1 && dpkg -l 'nvidia-*' | grep '^ii' | awk '{print $$2}' | head -1)
ifneq ("$(UBUNTU_PKG_NAME)","")
GLPATH ?= /usr/lib/$(UBUNTU_PKG_NAME)
GLLINK ?= -L/usr/lib/$(UBUNTU_PKG_NAME)
endif
DFLT_PATH ?= /usr/lib
endif
endif
ifeq ("$(SUSE)","0")
GLPATH ?= /usr/X11R6/lib64
GLLINK ?= -L/usr/X11R6/lib64
DFLT_PATH ?= /usr/lib64
endif
ifeq ("$(FEDORA)","0")
GLPATH ?= /usr/lib64/nvidia
GLLINK ?= -L/usr/lib64/nvidia
DFLT_PATH ?= /usr/lib64
endif
ifeq ("$(RHEL)","0")
GLPATH ?= /usr/lib64/nvidia
GLLINK ?= -L/usr/lib64/nvidia
DFLT_PATH ?= /usr/lib64
endif
ifeq ("$(CENTOS)","0")
GLPATH ?= /usr/lib64/nvidia
GLLINK ?= -L/usr/lib64/nvidia
DFLT_PATH ?= /usr/lib64
endif
EGLLIB := $(shell find -L $(GLPATH) $(DFLT_PATH) -name libEGL.so -print 2>/dev/null)
ifeq ("$(EGLLIB)","")
$(info >>> WARNING - libEGL.so not found, please install libEGL.so <<<)
SAMPLE_ENABLED := 0
endif
HEADER_SEARCH_PATH ?= $(TARGET_FS)/usr/include
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
HEADER_SEARCH_PATH += /usr/arm-linux-gnueabihf/include
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-aarch64-linux)
HEADER_SEARCH_PATH += /usr/aarch64-linux-gnu/include
endif
EGLHEADER := $(shell find -L $(HEADER_SEARCH_PATH) -name egl.h -print 2>/dev/null)
EGLEXTHEADER := $(shell find -L $(HEADER_SEARCH_PATH) -name eglext.h -print 2>/dev/null)
ifeq ("$(EGLHEADER)","")
$(info >>> WARNING - egl.h not found, please install egl.h <<<)
SAMPLE_ENABLED := 0
endif
ifeq ("$(EGLEXTHEADER)","")
$(info >>> WARNING - eglext.h not found, please install eglext.h <<<)
SAMPLE_ENABLED := 0
endif
else
endif
# Attempt to compile a minimal EGL application and run to check if EGL_SUPPORT_REUSE_NV is supported in the EGL headers available.
ifneq ($(SAMPLE_ENABLED), 0)
$(shell printf "#include <EGL/egl.h>\n#include <EGL/eglext.h>\nint main() {\n#ifdef EGL_SUPPORT_REUSE_NV \n #error \"Compatible EGL header found\" \n return 0;\n#endif \n return 1;\n}" > test.c; )
EGL_DEFINES := $(shell $(HOST_COMPILER) $(CCFLAGS) $(EXTRA_CCFLAGS) -lEGL test.c -c 2>&1 | grep -ic "Compatible EGL header found";)
SHOULD_WAIVE := 0
ifeq ($(EGL_DEFINES),0)
SHOULD_WAIVE := 1
endif
ifeq ($(SHOULD_WAIVE),1)
$(info -----------------------------------------------------------------------------------------------)
$(info WARNING - NVIDIA EGL EXTENSIONS are not available in the present EGL headers)
$(info -----------------------------------------------------------------------------------------------)
$(info This CUDA Sample cannot be built if the EGL NVIDIA EXTENSIONS like EGL_SUPPORT_REUSE_NV are not supported in EGL headers.)
$(info This will be a dry-run of the Makefile.)
$(info Please install the latest khronos EGL headers and libs to build this sample)
$(info -----------------------------------------------------------------------------------------------)
SAMPLE_ENABLED := 0
endif
$(shell rm test.o test.c 2>/dev/null)
endif

View File

@@ -0,0 +1,231 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
//
// DESCRIPTION: Simple EGL stream sample app
//
//
//#define EGL_EGLEXT_PROTOTYPES
#include "cudaEGL.h"
#include "cuda_consumer.h"
#include "cuda_producer.h"
#include "eglstrm_common.h"
/* ------ globals ---------*/
#if defined(EXTENSION_LIST)
EXTENSION_LIST(EXTLST_EXTERN)
#endif
#define NUM_TRAILS 4
bool signal_stop = 0;
static void sig_handler(int sig) {
signal_stop = 1;
printf("Signal: %d\n", sig);
}
int main(int argc, char **argv) {
TestArgs args;
CUresult curesult = CUDA_SUCCESS;
unsigned int i, j;
EGLint streamState = 0;
test_cuda_consumer_s cudaConsumer;
test_cuda_producer_s cudaProducer;
memset(&cudaProducer, 0, sizeof(test_cuda_producer_s));
memset(&cudaConsumer, 0, sizeof(test_cuda_consumer_s));
// Hook up Ctrl-C handler
signal(SIGINT, sig_handler);
if (!eglSetupExtensions()) {
printf("SetupExtentions failed \n");
curesult = CUDA_ERROR_UNKNOWN;
goto done;
}
checkCudaErrors(cuInit(0));
int count;
checkCudaErrors(cuDeviceGetCount(&count));
printf("Found %d cuda devices\n", count);
CUdevice devId;
if (!EGLStreamInit(&devId)) {
printf("EGLStream Init failed.\n");
curesult = CUDA_ERROR_UNKNOWN;
goto done;
}
curesult = cudaDeviceCreateProducer(&cudaProducer, devId);
if (curesult != CUDA_SUCCESS) {
goto done;
}
curesult = cudaDeviceCreateConsumer(&cudaConsumer, devId);
if (curesult != CUDA_SUCCESS) {
goto done;
}
checkCudaErrors(cuCtxPushCurrent(cudaConsumer.context));
if (CUDA_SUCCESS != (curesult = cuEGLStreamConsumerConnect(
&(cudaConsumer.cudaConn), eglStream))) {
printf("FAILED Connect CUDA consumer with error %d\n", curesult);
goto done;
} else {
printf("Connected CUDA consumer, CudaConsumer %p\n", cudaConsumer.cudaConn);
}
checkCudaErrors(cuCtxPopCurrent(&cudaConsumer.context));
checkCudaErrors(cuCtxPushCurrent(cudaProducer.context));
if (CUDA_SUCCESS ==
(curesult = cuEGLStreamProducerConnect(&(cudaProducer.cudaConn),
eglStream, WIDTH, HEIGHT))) {
printf("Connect CUDA producer Done, CudaProducer %p\n",
cudaProducer.cudaConn);
} else {
printf("Connect CUDA producer FAILED with error %d\n", curesult);
goto done;
}
checkCudaErrors(cuCtxPopCurrent(&cudaProducer.context));
// Initialize producer
for (i = 0; i < NUM_TRAILS; i++) {
if (streamState != EGL_STREAM_STATE_CONNECTING_KHR) {
if (!eglQueryStreamKHR(g_display, eglStream, EGL_STREAM_STATE_KHR,
&streamState)) {
printf("main: eglQueryStreamKHR EGL_STREAM_STATE_KHR failed\n");
curesult = CUDA_ERROR_UNKNOWN;
goto done;
}
}
args.inputWidth = WIDTH;
args.inputHeight = HEIGHT;
if (i % 2 != 0) {
args.isARGB = 1;
args.infile1 = sdkFindFilePath("cuda_f_1.yuv", argv[0]);
args.infile2 = sdkFindFilePath("cuda_f_2.yuv", argv[0]);
} else {
args.isARGB = 0;
args.infile1 = sdkFindFilePath("cuda_yuv_f_1.yuv", argv[0]);
args.infile2 = sdkFindFilePath("cuda_yuv_f_2.yuv", argv[0]);
}
if ((i % 4) < 2) {
args.pitchLinearOutput = 1;
} else {
args.pitchLinearOutput = 0;
}
checkCudaErrors(cuCtxPushCurrent(cudaProducer.context));
cudaProducerInit(&cudaProducer, g_display, eglStream, &args);
checkCudaErrors(cuCtxPopCurrent(&cudaProducer.context));
checkCudaErrors(cuCtxPushCurrent(cudaConsumer.context));
cuda_consumer_init(&cudaConsumer, &args);
checkCudaErrors(cuCtxPopCurrent(&cudaConsumer.context));
printf("main - Cuda Producer and Consumer Initialized.\n");
for (j = 0; j < 2; j++) {
printf("Running for %s frame and %s input\n",
args.isARGB ? "ARGB" : "YUV",
args.pitchLinearOutput ? "Pitchlinear" : "BlockLinear");
if (j == 0) {
checkCudaErrors(cuCtxPushCurrent(cudaProducer.context));
curesult = cudaProducerTest(&cudaProducer, cudaProducer.fileName1);
if (curesult != CUDA_SUCCESS) {
printf("Cuda Producer Test failed for frame = %d\n", j + 1);
goto done;
}
checkCudaErrors(cuCtxPopCurrent(&cudaProducer.context));
checkCudaErrors(cuCtxPushCurrent(cudaConsumer.context));
curesult = cudaConsumerTest(&cudaConsumer, cudaConsumer.outFile1);
if (curesult != CUDA_SUCCESS) {
printf("Cuda Consumer Test failed for frame = %d\n", j + 1);
goto done;
}
checkCudaErrors(cuCtxPopCurrent(&cudaConsumer.context));
} else {
checkCudaErrors(cuCtxPushCurrent(cudaProducer.context));
curesult = cudaProducerTest(&cudaProducer, cudaProducer.fileName2);
if (curesult != CUDA_SUCCESS) {
printf("Cuda Producer Test failed for frame = %d\n", j + 1);
goto done;
}
checkCudaErrors(cuCtxPopCurrent(&cudaProducer.context));
checkCudaErrors(cuCtxPushCurrent(cudaConsumer.context));
curesult = cudaConsumerTest(&cudaConsumer, cudaConsumer.outFile2);
if (curesult != CUDA_SUCCESS) {
printf("Cuda Consumer Test failed for frame = %d\n", j + 1);
goto done;
}
checkCudaErrors(cuCtxPopCurrent(&cudaConsumer.context));
}
}
}
checkCudaErrors(cuCtxPushCurrent(cudaProducer.context));
if (CUDA_SUCCESS != (curesult = cudaProducerDeinit(&cudaProducer))) {
printf("Producer Disconnect FAILED. \n");
goto done;
}
checkCudaErrors(cuCtxPopCurrent(&cudaProducer.context));
if (!eglQueryStreamKHR(g_display, eglStream, EGL_STREAM_STATE_KHR,
&streamState)) {
printf("Cuda consumer, eglQueryStreamKHR EGL_STREAM_STATE_KHR failed\n");
curesult = CUDA_ERROR_UNKNOWN;
goto done;
}
if (streamState != EGL_STREAM_STATE_DISCONNECTED_KHR) {
if (CUDA_SUCCESS != (curesult = cuda_consumer_deinit(&cudaConsumer))) {
printf("Consumer Disconnect FAILED.\n");
goto done;
}
}
printf("Producer and Consumer Disconnected \n");
done:
if (!eglQueryStreamKHR(g_display, eglStream, EGL_STREAM_STATE_KHR,
&streamState)) {
printf("Cuda consumer, eglQueryStreamKHR EGL_STREAM_STATE_KHR failed\n");
curesult = CUDA_ERROR_UNKNOWN;
}
if (streamState != EGL_STREAM_STATE_DISCONNECTED_KHR) {
EGLStreamFini();
}
if (curesult == CUDA_SUCCESS) {
printf("&&&& EGLStream interop test PASSED\n");
} else {
printf("&&&& EGLStream interop test FAILED\n");
}
return 0;
}

View File

@@ -0,0 +1,322 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * 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.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
#
################################################################################
#
# Makefile project only supported on Mac OS X and Linux Platforms)
#
################################################################################
# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda
##############################
# start deprecated interface #
##############################
ifeq ($(x86_64),1)
$(info WARNING - x86_64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
TARGET_ARCH ?= x86_64
endif
ifeq ($(ARMv7),1)
$(info WARNING - ARMv7 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=armv7l instead)
TARGET_ARCH ?= armv7l
endif
ifeq ($(aarch64),1)
$(info WARNING - aarch64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
TARGET_ARCH ?= aarch64
endif
ifeq ($(ppc64le),1)
$(info WARNING - ppc64le variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
TARGET_ARCH ?= ppc64le
endif
ifneq ($(GCC),)
$(info WARNING - GCC variable has been deprecated)
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
HOST_COMPILER ?= $(GCC)
endif
ifneq ($(abi),)
$(error ERROR - abi variable has been removed)
endif
############################
# end deprecated interface #
############################
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
endif
else
TARGET_SIZE := $(shell getconf LONG_BIT)
endif
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
TARGET_ARCH = armv7l
endif
# operating system
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif
# host compiler
ifeq ($(TARGET_OS),darwin)
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
HOST_COMPILER ?= clang++
endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
ifeq ($(TARGET_OS),linux)
HOST_COMPILER ?= arm-linux-gnueabihf-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
else ifeq ($(TARGET_OS),android)
HOST_COMPILER ?= arm-linux-androideabi-g++
endif
else ifeq ($(TARGET_ARCH),aarch64)
ifeq ($(TARGET_OS), linux)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
endif
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
# internal flags
NVCCFLAGS := -m${TARGET_SIZE}
CCFLAGS :=
LDFLAGS :=
# build flags
ifeq ($(TARGET_OS),darwin)
LDFLAGS += -rpath $(CUDA_PATH)/lib
CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
LDFLAGS += -pie
CCFLAGS += -fpie -fpic -fexceptions
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif
# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif
# Debug build flags
ifeq ($(dbg),1)
NVCCFLAGS += -g -G
BUILD_TYPE := debug
else
BUILD_TYPE := release
endif
ALL_CCFLAGS :=
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
# This sample is not supported on ARMv7
ifeq ($(TARGET_ARCH),armv7l)
$(info >>> WARNING - NV12toBGRandResize is not supported on ARMv7 - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
# Common includes and paths for CUDA
INCLUDES := -I../../Common
LIBRARIES :=
################################################################################
# Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 30 35 37 50 52 60 61 70 72 75
else
SMS ?= 30 35 37 50 52 60 61 70 75
endif
ifeq ($(SMS),)
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
HIGHEST_SM := $(lastword $(sort $(SMS)))
ifneq ($(HIGHEST_SM),)
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
endif
endif
ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif
################################################################################
# Target rules
all: build
build: NV12toBGRandResize
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@echo "Sample will be waived due to the above missing dependencies"
else
@echo "Sample is ready - all dependencies have been met"
endif
bgr_resize.o:bgr_resize.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
nv12_resize.o:nv12_resize.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
nv12_to_bgr_planar.o:nv12_to_bgr_planar.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
resize_convert_main.o:resize_convert_main.cpp
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
utils.o:utils.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
NV12toBGRandResize: bgr_resize.o nv12_resize.o nv12_to_bgr_planar.o resize_convert_main.o utils.o
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
$(EXEC) ./NV12toBGRandResize
clean:
rm -f NV12toBGRandResize bgr_resize.o nv12_resize.o nv12_to_bgr_planar.o resize_convert_main.o utils.o
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/NV12toBGRandResize
clobber: clean

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2012
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "NV12toBGRandResize", "NV12toBGRandResize_vs2012.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,112 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>NV12toBGRandResize_vs2012</RootNamespace>
<ProjectName>NV12toBGRandResize</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v110</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bgr_resize.cu" />
<CudaCompile Include="nv12_resize.cu" />
<CudaCompile Include="nv12_to_bgr_planar.cu" />
<ClCompile Include="resize_convert_main.cpp" />
<CudaCompile Include="utils.cu" />
<ClInclude Include="resize_convert.h" />
<ClInclude Include="utils.h" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 13.00
# Visual Studio 2013
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "NV12toBGRandResize", "NV12toBGRandResize_vs2013.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,112 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>NV12toBGRandResize_vs2013</RootNamespace>
<ProjectName>NV12toBGRandResize</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v120</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bgr_resize.cu" />
<CudaCompile Include="nv12_resize.cu" />
<CudaCompile Include="nv12_to_bgr_planar.cu" />
<ClCompile Include="resize_convert_main.cpp" />
<CudaCompile Include="utils.cu" />
<ClInclude Include="resize_convert.h" />
<ClInclude Include="utils.h" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 14.00
# Visual Studio 2015
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "NV12toBGRandResize", "NV12toBGRandResize_vs2015.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,112 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>NV12toBGRandResize_vs2015</RootNamespace>
<ProjectName>NV12toBGRandResize</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v140</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bgr_resize.cu" />
<CudaCompile Include="nv12_resize.cu" />
<CudaCompile Include="nv12_to_bgr_planar.cu" />
<ClCompile Include="resize_convert_main.cpp" />
<CudaCompile Include="utils.cu" />
<ClInclude Include="resize_convert.h" />
<ClInclude Include="utils.h" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2017
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "NV12toBGRandResize", "NV12toBGRandResize_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,117 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>NV12toBGRandResize_vs2017</RootNamespace>
<ProjectName>NV12toBGRandResize</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
<LatestTargetPlatformVersion>$([Microsoft.Build.Utilities.ToolLocationHelper]::GetLatestSDKTargetPlatformVersion('Windows', '10.0'))</LatestTargetPlatformVersion>
<WindowsTargetPlatformVersion Condition="'$(WindowsTargetPlatformVersion)' == ''">$(LatestTargetPlatformVersion)</WindowsTargetPlatformVersion>
<TargetPlatformVersion>$(WindowsTargetPlatformVersion)</TargetPlatformVersion>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v141</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bgr_resize.cu" />
<CudaCompile Include="nv12_resize.cu" />
<CudaCompile Include="nv12_to_bgr_planar.cu" />
<ClCompile Include="resize_convert_main.cpp" />
<CudaCompile Include="utils.cu" />
<ClInclude Include="resize_convert.h" />
<ClInclude Include="utils.h" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2019
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "NV12toBGRandResize", "NV12toBGRandResize_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,113 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>NV12toBGRandResize_vs2019</RootNamespace>
<ProjectName>NV12toBGRandResize</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v142</PlatformToolset>
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bgr_resize.cu" />
<CudaCompile Include="nv12_resize.cu" />
<CudaCompile Include="nv12_to_bgr_planar.cu" />
<ClCompile Include="resize_convert_main.cpp" />
<CudaCompile Include="utils.cu" />
<ClInclude Include="resize_convert.h" />
<ClInclude Include="utils.h" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,70 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
<entry>
<name>NV12toBGRandResize</name>
<cuda_api_list>
<toolkit>cudaMemcpy2D</toolkit>
<toolkit>cudaMallocManaged</toolkit>
</cuda_api_list>
<description><![CDATA[This code shows two ways to convert and resize NV12 frames to BGR 3 planars frames using CUDA in batch. Way-1, Convert NV12 Input to BGR @ Input Resolution-1, then Resize to Resolution#2. Way-2, resize NV12 Input to Resolution#2 then convert it to BGR Output. NVIDIA HW Decoder, both dGPU and Tegra, normally outputs NV12 pitch format frames. For the inference using TensorRT, the input frame needs to be BGR planar format with possibly different size. So, conversion and resizing from NV12 to BGR planar is usually required for the inference following decoding. This CUDA code provides a reference implementation for conversion and resizing.]]></description>
<devicecompilation>whole</devicecompilation>
<includepaths>
<path>./</path>
<path>../</path>
<path>../../common/inc</path>
</includepaths>
<keyconcepts>
<concept level="basic">Graphics Interop</concept>
<concept level="basic">Image Processing</concept>
<concept level="basic">Video Processing</concept>
</keyconcepts>
<keywords>
<keyword>GPGPU</keyword>
</keywords>
<libraries>
</libraries>
<librarypaths>
</librarypaths>
<nsight_eclipse>true</nsight_eclipse>
<primary_file>resize_convert_main.cpp</primary_file>
<scopes>
<scope>1:CUDA Basic Topics</scope>
<scope>2:Image Processing</scope>
<scope>2:Computer Vision</scope>
</scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch>
<sm-arch>sm52</sm-arch>
<sm-arch>sm60</sm-arch>
<sm-arch>sm61</sm-arch>
<sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch>
<supported_envs>
<env>
<arch>x86_64</arch>
<platform>linux</platform>
</env>
<env>
<platform>windows7</platform>
</env>
<env>
<arch>x86_64</arch>
<platform>macosx</platform>
</env>
<env>
<arch>aarch64</arch>
</env>
<env>
<arch>ppc64le</arch>
<platform>linux</platform>
</env>
</supported_envs>
<supported_sm_architectures>
<include>all</include>
</supported_sm_architectures>
<title>NV12toBGRandResize</title>
<type>exe</type>
</entry>

View File

@@ -0,0 +1,94 @@
# NV12toBGRandResize - NV12toBGRandResize
## Description
This code shows two ways to convert and resize NV12 frames to BGR 3 planars frames using CUDA in batch. Way-1, Convert NV12 Input to BGR @ Input Resolution-1, then Resize to Resolution#2. Way-2, resize NV12 Input to Resolution#2 then convert it to BGR Output. NVIDIA HW Decoder, both dGPU and Tegra, normally outputs NV12 pitch format frames. For the inference using TensorRT, the input frame needs to be BGR planar format with possibly different size. So, conversion and resizing from NV12 to BGR planar is usually required for the inference following decoding. This CUDA code provides a reference implementation for conversion and resizing.
## Key Concepts
Graphics Interop, Image Processing, Video Processing
## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
Linux, Windows, MacOSX
## Supported CPU Architecture
x86_64, ppc64le, aarch64
## CUDA APIs involved
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
cudaMemcpy2D, cudaMallocManaged
## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run
### Windows
The Windows samples are built using the Visual Studio IDE. Solution files (.sln) are provided for each supported version of Visual Studio, using the format:
```
*_vs<version>.sln - for Visual Studio <version>
```
Each individual sample has its own set of solution files in its directory:
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check DirectX Dependencies section for details."
### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, ppc64le, aarch64.
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=ppc64le` <br/> `$ make TARGET_ARCH=aarch64` <br/>
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
```
$ make SMS="50 60"
```
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=g++
```
### Mac
The Mac samples are built using makefiles. To use the makefiles, change directory into the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where "A B ..." is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use SMS="50 60".
```
$ make SMS="A B ..."
```
* **HOST_COMPILER=<host_compiler>** - override the default clang host compiler. See the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=clang
```
## References (for more details)

View File

@@ -0,0 +1,134 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
// Implements BGR 3 progressive planars frames batch resize
#include <cuda.h>
#include <cuda_runtime.h>
#include "resize_convert.h"
__global__ void resizeBGRplanarBatchKernel(cudaTextureObject_t texSrc,
float *pDst, int nDstPitch, int nDstHeight, int nSrcHeight,
int batch, float scaleX, float scaleY,
int cropX, int cropY, int cropW, int cropH) {
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
if (x >= (int)(cropW/scaleX) || y >= (int)(cropH/scaleY))
return;
int frameSize = nDstPitch*nDstHeight;
float *p = NULL;
for (int i = blockIdx.z; i < batch; i += gridDim.z) {
#pragma unroll
for (int channel=0; channel < 3; channel++){
p = pDst + i * 3 * frameSize + y * nDstPitch + x + channel * frameSize;
*p = tex2D<float>(texSrc, x * scaleX + cropX,
((3 * i + channel) * nSrcHeight + y * scaleY + cropY));
}
}
}
static void resizeBGRplanarBatchCore(
float *dpSrc, int nSrcPitch, int nSrcWidth, int nSrcHeight,
float *dpDst, int nDstPitch, int nDstWidth, int nDstHeight,
int nBatchSize, cudaStream_t stream, bool whSameResizeRatio,
int cropX, int cropY, int cropW, int cropH) {
cudaTextureObject_t texSrc[2];
int nTiles = 1, h, iTile;
h = nSrcHeight * 3 * nBatchSize;
while ((h + nTiles - 1) / nTiles > 65536)
nTiles++;
if (nTiles > 2)
return;
int batchTile = nBatchSize / nTiles;
int batchTileLast = nBatchSize - batchTile * (nTiles-1);
for (iTile = 0; iTile < nTiles; ++iTile) {
int bs = (iTile == nTiles - 1) ? batchTileLast : batchTile;
float *dpSrcNew = dpSrc +
iTile * (batchTile * 3 * nSrcHeight * nSrcPitch);
cudaResourceDesc resDesc = {};
resDesc.resType = cudaResourceTypePitch2D;
resDesc.res.pitch2D.devPtr = dpSrcNew;
resDesc.res.pitch2D.desc = cudaCreateChannelDesc<float>();
resDesc.res.pitch2D.width = nSrcWidth;
resDesc.res.pitch2D.height = bs * 3 * nSrcHeight;
resDesc.res.pitch2D.pitchInBytes = nSrcPitch * sizeof(float);
cudaTextureDesc texDesc = {};
texDesc.filterMode = cudaFilterModeLinear;
texDesc.readMode = cudaReadModeElementType;
checkCudaErrors(cudaCreateTextureObject(&texSrc[iTile], &resDesc, &texDesc, NULL));
float *dpDstNew = dpDst +
iTile * (batchTile * 3 * nDstHeight * nDstPitch);
if(cropW == 0 || cropH == 0) {
cropX = 0;
cropY = 0;
cropW = nSrcWidth;
cropH = nSrcHeight;
}
float scaleX = (cropW*1.0f / nDstWidth);
float scaleY = (cropH*1.0f / nDstHeight);
if(whSameResizeRatio == true)
scaleX = scaleY = scaleX > scaleY ? scaleX : scaleY;
dim3 block(32, 32, 1);
size_t blockDimZ = bs;
// Restricting blocks in Z-dim till 32 to not launch too many blocks
blockDimZ = (blockDimZ > 32) ? 32 : blockDimZ;
dim3 grid((cropW*1.0f/scaleX + block.x - 1) / block.x,
(cropH*1.0f/scaleY + block.y - 1) / block.y, blockDimZ);
resizeBGRplanarBatchKernel<<<grid, block, 0, stream>>>
(texSrc[iTile], dpDstNew, nDstPitch, nDstHeight, nSrcHeight,
bs, scaleX, scaleY, cropX, cropY, cropW, cropH);
}
for (iTile = 0; iTile < nTiles; ++iTile)
checkCudaErrors(cudaDestroyTextureObject(texSrc[iTile]));
}
void resizeBGRplanarBatch(
float *dpSrc, int nSrcPitch, int nSrcWidth, int nSrcHeight,
float *dpDst, int nDstPitch, int nDstWidth, int nDstHeight,
int nBatchSize, cudaStream_t stream,
int cropX, int cropY, int cropW, int cropH, bool whSameResizeRatio) {
resizeBGRplanarBatchCore(dpSrc, nSrcPitch, nSrcWidth, nSrcHeight,
dpDst, nDstPitch, nDstWidth, nDstHeight, nBatchSize, stream,
whSameResizeRatio, cropX, cropY, cropW, cropH);
}

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

View File

@@ -0,0 +1,112 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
// Implements interlace NV12 frames batch resize
#include <cuda.h>
#include <cuda_runtime.h>
#include "resize_convert.h"
__global__ static void resizeNV12BatchKernel(cudaTextureObject_t texSrcLuma,
cudaTextureObject_t texSrcChroma,
uint8_t *pDstNv12, int nSrcWidth,
int nSrcHeight, int nDstPitch,
int nDstWidth, int nDstHeight,
int nBatchSize) {
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
int px = x * 2, py = y * 2;
if ((px + 1) >= nDstWidth || (py + 1) >= nDstHeight) return;
float fxScale = 1.0f * nSrcWidth / nDstWidth;
float fyScale = 1.0f * nSrcHeight / nDstHeight;
uint8_t *p = pDstNv12 + px + py * nDstPitch;
int hh = nDstHeight * 3 / 2;
int nByte = nDstPitch * hh;
int px_fxScale = px * fxScale;
int px_fxScale_1 = (px + 1) * fxScale;
int py_fyScale = py * fyScale;
int py_fyScale_1 = (py + 1) * fyScale;
for (int i = blockIdx.z; i < nBatchSize; i+=gridDim.z) {
*(uchar2 *)p = make_uchar2(tex2D<uint8_t>(texSrcLuma, px_fxScale, py_fyScale),
tex2D<uint8_t>(texSrcLuma, px_fxScale_1, py_fyScale));
*(uchar2 *)(p + nDstPitch) =
make_uchar2(tex2D<uint8_t>(texSrcLuma, px_fxScale, py_fyScale_1),
tex2D<uint8_t>(texSrcLuma, px_fxScale_1, py_fyScale_1));
*(uchar2 *)(p + (nDstHeight - y) * nDstPitch) = tex2D<uchar2>(
texSrcChroma, x * fxScale, (hh * i + nDstHeight + y) * fyScale);
p += nByte;
py += hh;
}
}
void resizeNV12Batch(uint8_t *dpSrc, int nSrcPitch, int nSrcWidth,
int nSrcHeight, uint8_t *dpDst, int nDstPitch,
int nDstWidth, int nDstHeight, int nBatchSize,
cudaStream_t stream) {
int hhSrc = ceilf(nSrcHeight * 3.0f / 2.0f);
cudaResourceDesc resDesc = {};
resDesc.resType = cudaResourceTypePitch2D;
resDesc.res.pitch2D.devPtr = dpSrc;
resDesc.res.pitch2D.desc = cudaCreateChannelDesc<uint8_t>();
resDesc.res.pitch2D.width = nSrcWidth;
resDesc.res.pitch2D.height = hhSrc * nBatchSize;
resDesc.res.pitch2D.pitchInBytes = nSrcPitch;
cudaTextureDesc texDesc = {};
texDesc.filterMode = cudaFilterModePoint;
texDesc.readMode = cudaReadModeElementType;
cudaTextureObject_t texLuma = 0;
checkCudaErrors(cudaCreateTextureObject(&texLuma, &resDesc, &texDesc, NULL));
resDesc.res.pitch2D.desc = cudaCreateChannelDesc<uchar2>();
resDesc.res.pitch2D.width /= 2;
cudaTextureObject_t texChroma = 0;
checkCudaErrors(cudaCreateTextureObject(&texChroma, &resDesc, &texDesc, NULL));
dim3 block(32, 32, 1);
size_t blockDimZ = nBatchSize;
// Restricting blocks in Z-dim till 32 to not launch too many blocks
blockDimZ = (blockDimZ > 32) ? 32 : blockDimZ;
dim3 grid((nDstWidth / 2 + block.x) / block.x,
(nDstHeight / 2 + block.y) / block.y, blockDimZ);
resizeNV12BatchKernel<<<grid, block, 0, stream>>>(
texLuma, texChroma, dpDst, nSrcWidth, nSrcHeight, nDstPitch, nDstWidth,
nDstHeight, nBatchSize);
checkCudaErrors(cudaDestroyTextureObject(texLuma));
checkCudaErrors(cudaDestroyTextureObject(texChroma));
}

View File

@@ -0,0 +1,154 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
// Implements NV12 to BGR batch conversion
#include <cuda.h>
#include <cuda_runtime.h>
#include "resize_convert.h"
#define CONV_THREADS_X 64
#define CONV_THREADS_Y 10
__forceinline__ __device__ static float clampF(float x, float lower,
float upper) {
return x < lower ? lower : (x > upper ? upper : x);
}
__global__ static void nv12ToBGRplanarBatchKernel(const uint8_t *pNv12,
int nNv12Pitch, float *pBgr,
int nRgbPitch, int nWidth,
int nHeight, int nBatchSize) {
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
if ((x << 2) + 1 > nWidth || (y << 1) + 1 > nHeight) return;
const uint8_t *__restrict__ pSrc = pNv12;
for (int i = blockIdx.z; i < nBatchSize; i += gridDim.z) {
pSrc = pNv12 + i * ((nHeight * nNv12Pitch * 3) >> 1) + (x << 2) +
(y << 1) * nNv12Pitch;
uchar4 luma2x01, luma2x23, uv2;
*(uint32_t *)&luma2x01 = *(uint32_t *)pSrc;
*(uint32_t *)&luma2x23 = *(uint32_t *)(pSrc + nNv12Pitch);
*(uint32_t *)&uv2 = *(uint32_t *)(pSrc + (nHeight - y) * nNv12Pitch);
float *pDstBlock = (pBgr + i * ((nHeight * nRgbPitch * 3) >> 2) +
((blockIdx.x * blockDim.x) << 2) +
((blockIdx.y * blockDim.y) << 1) * (nRgbPitch >> 2));
float2 add1;
float2 add2;
float2 add3;
float2 add00, add01, add02, add03;
float2 d, e;
add00.x = 1.1644f * luma2x01.x;
add01.x = 1.1644f * luma2x01.y;
add00.y = 1.1644f * luma2x01.z;
add01.y = 1.1644f * luma2x01.w;
add02.x = 1.1644f * luma2x23.x;
add03.x = 1.1644f * luma2x23.y;
add02.y = 1.1644f * luma2x23.z;
add03.y = 1.1644f * luma2x23.w;
d.x = uv2.x - 128.0f;
e.x = uv2.y - 128.0f;
d.y = uv2.z - 128.0f;
e.y = uv2.w - 128.0f;
add1.x = 2.0172f * d.x;
add1.y = 2.0172f * d.y;
add2.x = (-0.3918f) * d.x + (-0.8130f) * e.x;
add2.y = (-0.3918f) * d.y + (-0.8130f) * e.y;
add3.x = 1.5960f * e.x;
add3.y = 1.5960f * e.y;
int rowStride = (threadIdx.y << 1) * (nRgbPitch >> 2);
int nextRowStride = ((threadIdx.y << 1) + 1) * (nRgbPitch >> 2);
// B
*((float4 *)&pDstBlock[rowStride + (threadIdx.x << 2)]) =
make_float4(clampF(add00.x + add1.x, 0.0f, 255.0f),
clampF(add01.x + add1.x, 0.0f, 255.0f),
clampF(add00.y + add1.y, 0.0f, 255.0f),
clampF(add01.y + add1.y, 0.0f, 255.0f));
*((float4 *)&pDstBlock[nextRowStride + (threadIdx.x << 2)]) =
make_float4(clampF(add02.x + add1.x, 0.0f, 255.0f),
clampF(add03.x + add1.x, 0.0f, 255.0f),
clampF(add02.y + add1.y, 0.0f, 255.0f),
clampF(add03.y + add1.y, 0.0f, 255.0f));
int planeStride = nHeight * nRgbPitch >> 2;
// G
*((float4 *)&pDstBlock[planeStride + rowStride + (threadIdx.x << 2)]) =
make_float4(clampF(add00.x + add2.x, 0.0f, 255.0f),
clampF(add01.x + add2.x, 0.0f, 255.0f),
clampF(add00.y + add2.y, 0.0f, 255.0f),
clampF(add01.y + add2.y, 0.0f, 255.0f));
*((float4 *)&pDstBlock[planeStride + nextRowStride + (threadIdx.x << 2)]) =
make_float4(clampF(add02.x + add2.x, 0.0f, 255.0f),
clampF(add03.x + add2.x, 0.0f, 255.0f),
clampF(add02.y + add2.y, 0.0f, 255.0f),
clampF(add03.y + add2.y, 0.0f, 255.0f));
// R
*((float4
*)&pDstBlock[(planeStride << 1) + rowStride + (threadIdx.x << 2)]) =
make_float4(clampF(add00.x + add3.x, 0.0f, 255.0f),
clampF(add01.x + add3.x, 0.0f, 255.0f),
clampF(add00.y + add3.y, 0.0f, 255.0f),
clampF(add01.y + add3.y, 0.0f, 255.0f));
*((float4 *)&pDstBlock[(planeStride << 1) + nextRowStride +
(threadIdx.x << 2)]) =
make_float4(clampF(add02.x + add3.x, 0.0f, 255.0f),
clampF(add03.x + add3.x, 0.0f, 255.0f),
clampF(add02.y + add3.y, 0.0f, 255.0f),
clampF(add03.y + add3.y, 0.0f, 255.0f));
}
}
void nv12ToBGRplanarBatch(uint8_t *pNv12, int nNv12Pitch, float *pBgr,
int nRgbPitch, int nWidth, int nHeight,
int nBatchSize, cudaStream_t stream) {
dim3 threads(CONV_THREADS_X, CONV_THREADS_Y);
size_t blockDimZ = nBatchSize;
// Restricting blocks in Z-dim till 32 to not launch too many blocks
blockDimZ = (blockDimZ > 32) ? 32 : blockDimZ;
dim3 blocks((nWidth / 4 - 1) / threads.x + 1,
(nHeight / 2 - 1) / threads.y + 1, blockDimZ);
nv12ToBGRplanarBatchKernel<<<blocks, threads, 0, stream>>>(
pNv12, nNv12Pitch, pBgr, nRgbPitch, nWidth, nHeight, nBatchSize);
}

View File

@@ -0,0 +1,56 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#ifndef __H_RESIZE_CONVERT__
#define __H_RESIZE_CONVERT__
#include <iostream>
#include <helper_cuda.h>
// nv12 resize
extern "C"
void resizeNV12Batch(
uint8_t *dpSrc, int nSrcPitch, int nSrcWidth, int nSrcHeight,
uint8_t *dpDst, int nDstPitch, int nDstWidth, int nDstHeight,
int nBatchSize, cudaStream_t stream = 0);
// bgr resize
extern "C"
void resizeBGRplanarBatch(
float *dpSrc, int nSrcPitch, int nSrcWidth, int nSrcHeight,
float *dpDst, int nDstPitch, int nDstWidth, int nDstHeight,
int nBatchSize, cudaStream_t stream = 0,
int cropX = 0, int cropY = 0, int cropW = 0, int cropH = 0,
bool whSameResizeRatio = false);
//NV12 to bgr planar
extern "C"
void nv12ToBGRplanarBatch(uint8_t *pNv12, int nNv12Pitch,
float *pRgb, int nRgbPitch, int nWidth, int nHeight,
int nBatchSize, cudaStream_t stream=0);
#endif

View File

@@ -0,0 +1,448 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
/*
NVIDIA HW Decoder, both dGPU and Tegra, normally outputs NV12 pitch format
frames. For the inference using TensorRT, the input frame needs to be BGR planar
format with possibly different size. So, conversion and resizing from NV12 to
BGR planar is usually required for the inference following decoding.
This CUDA code is to provide a reference implementation for conversion and
resizing.
Limitaion
=========
NV12resize needs the height to be a even value.
Note
====
Resize function needs the pitch of image buffer to be 32 alignment.
Run
====
./NV12toBGRandResize
OR
./NV12toBGRandResize -input=data/test1920x1080.nv12 -width=1920 -height=1080 \
-dst_width=640 -dst_height=480 -batch=40 -device=0
*/
#include <cuda.h>
#include <cuda_runtime.h>
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <cassert>
#include <fstream>
#include <iostream>
#include <memory>
#include "resize_convert.h"
#include "utils.h"
#define TEST_LOOP 20
typedef struct _nv12_to_bgr24_context_t {
int width;
int height;
int pitch;
int dst_width;
int dst_height;
int dst_pitch;
int batch;
int device; // cuda device ID
char *input_nv12_file;
int ctx_pitch; // the value will be suitable for Texture memroy.
int ctx_heights; // the value will be even.
} nv12_to_bgr24_context;
nv12_to_bgr24_context g_ctx;
static void printHelp(const char *app_name) {
std::cout << "Usage:" << app_name << " [options]\n\n";
std::cout << "OPTIONS:\n";
std::cout << "\t-h,--help\n\n";
std::cout << "\t-input=nv12file nv12 input file\n";
std::cout
<< "\t-width=width input nv12 image width, <1 -- 4096>\n";
std::cout
<< "\t-height=height input nv12 image height, <1 -- 4096>\n";
std::cout
<< "\t-pitch=pitch(optional) input nv12 image pitch, <0 -- 4096>\n";
std::cout
<< "\t-dst_width=width output BGR image width, <1 -- 4096>\n";
std::cout
<< "\t-dst_height=height output BGR image height, <1 -- 4096>\n";
std::cout
<< "\t-dst_pitch=pitch(optional) output BGR image pitch, <0 -- 4096>\n";
std::cout
<< "\t-batch=batch process frames count, <1 -- 4096>\n\n";
std::cout
<< "\t-device=device_num(optional) cuda device number, <0 -- 4096>\n\n";
return;
}
int parseCmdLine(int argc, char *argv[]) {
char **argp = (char **)argv;
char *arg = (char *)argv[0];
memset(&g_ctx, 0, sizeof(g_ctx));
if ((arg && (!strcmp(arg, "-h") || !strcmp(arg, "--help")))) {
printHelp(argv[0]);
return -1;
}
if (argc == 1) {
// Run using default arguments
g_ctx.input_nv12_file = sdkFindFilePath("test1920x1080.nv12", argv[0]);
if (g_ctx.input_nv12_file == NULL) {
printf("Cannot find input file test1920x1080.nv12\n Exiting\n");
return EXIT_FAILURE;
}
g_ctx.width = 1920;
g_ctx.height = 1080;
g_ctx.dst_width = 640;
g_ctx.dst_height = 480;
g_ctx.batch = 24;
} else if (argc > 1) {
if (checkCmdLineFlag(argc, (const char **)argv, "width")) {
g_ctx.width = getCmdLineArgumentInt(argc, (const char **)argv, "width");
}
if (checkCmdLineFlag(argc, (const char **)argv, "height")) {
g_ctx.height = getCmdLineArgumentInt(argc, (const char **)argv, "height");
}
if (checkCmdLineFlag(argc, (const char **)argv, "pitch")) {
g_ctx.pitch = getCmdLineArgumentInt(argc, (const char **)argv, "pitch");
}
if (checkCmdLineFlag(argc, (const char **)argv, "input")) {
getCmdLineArgumentString(argc, (const char **)argv, "input",
(char **)&g_ctx.input_nv12_file);
}
if (checkCmdLineFlag(argc, (const char **)argv, "dst_width")) {
g_ctx.dst_width =
getCmdLineArgumentInt(argc, (const char **)argv, "dst_width");
}
if (checkCmdLineFlag(argc, (const char **)argv, "dst_height")) {
g_ctx.dst_height =
getCmdLineArgumentInt(argc, (const char **)argv, "dst_height");
}
if (checkCmdLineFlag(argc, (const char **)argv, "dst_pitch")) {
g_ctx.dst_pitch =
getCmdLineArgumentInt(argc, (const char **)argv, "dst_pitch");
}
if (checkCmdLineFlag(argc, (const char **)argv, "batch")) {
g_ctx.batch = getCmdLineArgumentInt(argc, (const char **)argv, "batch");
}
}
g_ctx.device = findCudaDevice(argc, (const char **)argv);
if ((g_ctx.width == 0) || (g_ctx.height == 0) || (g_ctx.dst_width == 0) ||
(g_ctx.dst_height == 0) || !g_ctx.input_nv12_file) {
printHelp(argv[0]);
return -1;
}
if (g_ctx.pitch == 0) g_ctx.pitch = g_ctx.width;
if (g_ctx.dst_pitch == 0) g_ctx.dst_pitch = g_ctx.dst_width;
return 0;
}
/*
load nv12 yuvfile data into GPU device memory with batch of copy
*/
static int loadNV12Frame(unsigned char *d_inputNV12) {
unsigned char *pNV12FrameData;
unsigned char *d_nv12;
int frameSize;
std::ifstream nv12File(g_ctx.input_nv12_file, std::ifstream::in | std::ios::binary);
if (!nv12File.is_open()) {
std::cerr << "Can't open files\n";
return -1;
}
frameSize = g_ctx.pitch * g_ctx.ctx_heights;
#if USE_UVM_MEM
pNV12FrameData = d_inputNV12;
#else
pNV12FrameData = (unsigned char *)malloc(frameSize);
if (pNV12FrameData == NULL) {
std::cerr << "Failed to malloc pNV12FrameData\n";
return -1;
}
#endif
nv12File.read((char *)pNV12FrameData, frameSize);
if (nv12File.gcount() < frameSize) {
std::cerr << "can't get one frame!\n";
return -1;
}
#if USE_UVM_MEM
// Prefetch to GPU for following GPU operation
cudaStreamAttachMemAsync(NULL, pNV12FrameData, 0, cudaMemAttachGlobal);
#endif
// expand one frame to multi frames for batch processing
d_nv12 = d_inputNV12;
for (int i = 0; i < g_ctx.batch; i++) {
checkCudaErrors(cudaMemcpy2D((void *)d_nv12, g_ctx.ctx_pitch,
pNV12FrameData, g_ctx.width, g_ctx.width,
g_ctx.ctx_heights, cudaMemcpyHostToDevice));
d_nv12 += g_ctx.ctx_pitch * g_ctx.ctx_heights;
}
#if (USE_UVM_MEM == 0)
free(pNV12FrameData);
#endif
nv12File.close();
return 0;
}
/*
1. resize interlace nv12 to target size
2. convert nv12 to bgr 3 progressive planars
*/
void nv12ResizeAndNV12ToBGR(unsigned char *d_inputNV12) {
unsigned char *d_resizedNV12;
float *d_outputBGR;
int size;
char filename[40];
/* allocate device memory for resized nv12 output */
size = g_ctx.dst_width * ceil(g_ctx.dst_height * 3.0f / 2.0f) * g_ctx.batch *
sizeof(unsigned char);
checkCudaErrors(cudaMalloc((void **)&d_resizedNV12, size));
/* allocate device memory for bgr output */
size = g_ctx.dst_pitch * g_ctx.dst_height * 3 * g_ctx.batch * sizeof(float);
checkCudaErrors(cudaMalloc((void **)&d_outputBGR, size));
cudaStream_t stream;
checkCudaErrors(cudaStreamCreate(&stream));
/* create cuda event handles */
cudaEvent_t start, stop;
checkCudaErrors(cudaEventCreate(&start));
checkCudaErrors(cudaEventCreate(&stop));
float elapsedTime = 0.0f;
/* resize interlace nv12 */
cudaEventRecord(start, 0);
for (int i = 0; i < TEST_LOOP; i++) {
resizeNV12Batch(d_inputNV12, g_ctx.ctx_pitch, g_ctx.width, g_ctx.height,
d_resizedNV12, g_ctx.dst_width, g_ctx.dst_width,
g_ctx.dst_height, g_ctx.batch);
}
cudaEventRecord(stop, 0);
cudaEventSynchronize(stop);
cudaEventElapsedTime(&elapsedTime, start, stop);
printf(
" CUDA resize nv12(%dx%d --> %dx%d), batch: %d,"
" average time: %.3f ms ==> %.3f ms/frame\n",
g_ctx.width, g_ctx.height, g_ctx.dst_width, g_ctx.dst_height, g_ctx.batch,
(elapsedTime / (TEST_LOOP * 1.0f)),
(elapsedTime / (TEST_LOOP * 1.0f)) / g_ctx.batch);
sprintf(filename, "resized_nv12_%dx%d", g_ctx.dst_width, g_ctx.dst_height);
/* convert nv12 to bgr 3 progressive planars */
cudaEventRecord(start, 0);
for (int i = 0; i < TEST_LOOP; i++) {
nv12ToBGRplanarBatch(d_resizedNV12, g_ctx.dst_pitch, // intput
d_outputBGR,
g_ctx.dst_pitch * sizeof(float), // output
g_ctx.dst_width, g_ctx.dst_height, // output
g_ctx.batch, 0);
}
cudaEventRecord(stop, 0);
cudaEventSynchronize(stop);
cudaEventElapsedTime(&elapsedTime, start, stop);
printf(
" CUDA convert nv12(%dx%d) to bgr(%dx%d), batch: %d,"
" average time: %.3f ms ==> %.3f ms/frame\n",
g_ctx.dst_width, g_ctx.dst_height, g_ctx.dst_width, g_ctx.dst_height,
g_ctx.batch, (elapsedTime / (TEST_LOOP * 1.0f)),
(elapsedTime / (TEST_LOOP * 1.0f)) / g_ctx.batch);
sprintf(filename, "converted_bgr_%dx%d", g_ctx.dst_width, g_ctx.dst_height);
dumpBGR(d_outputBGR, g_ctx.dst_pitch, g_ctx.dst_width, g_ctx.dst_height,
g_ctx.batch, (char *)"t1", filename);
/* release resources */
checkCudaErrors(cudaEventDestroy(start));
checkCudaErrors(cudaEventDestroy(stop));
checkCudaErrors(cudaStreamDestroy(stream));
checkCudaErrors(cudaFree(d_resizedNV12));
checkCudaErrors(cudaFree(d_outputBGR));
}
/*
1. convert nv12 to bgr 3 progressive planars
2. resize bgr 3 planars to target size
*/
void nv12ToBGRandBGRresize(unsigned char *d_inputNV12) {
float *d_bgr;
float *d_resizedBGR;
int size;
char filename[40];
/* allocate device memory for bgr output */
size = g_ctx.ctx_pitch * g_ctx.height * 3 * g_ctx.batch * sizeof(float);
checkCudaErrors(cudaMalloc((void **)&d_bgr, size));
/* allocate device memory for resized bgr output */
size = g_ctx.dst_width * g_ctx.dst_height * 3 * g_ctx.batch * sizeof(float);
checkCudaErrors(cudaMalloc((void **)&d_resizedBGR, size));
cudaStream_t stream;
checkCudaErrors(cudaStreamCreate(&stream));
/* create cuda event handles */
cudaEvent_t start, stop;
checkCudaErrors(cudaEventCreate(&start));
checkCudaErrors(cudaEventCreate(&stop));
float elapsedTime = 0.0f;
/* convert interlace nv12 to bgr 3 progressive planars */
cudaEventRecord(start, 0);
cudaDeviceSynchronize();
for (int i = 0; i < TEST_LOOP; i++) {
nv12ToBGRplanarBatch(d_inputNV12, g_ctx.ctx_pitch, d_bgr,
g_ctx.ctx_pitch * sizeof(float), g_ctx.width,
g_ctx.height, g_ctx.batch, 0);
}
cudaEventRecord(stop, 0);
cudaEventSynchronize(stop);
cudaEventElapsedTime(&elapsedTime, start, stop);
printf(
" CUDA convert nv12(%dx%d) to bgr(%dx%d), batch: %d,"
" average time: %.3f ms ==> %.3f ms/frame\n",
g_ctx.width, g_ctx.height, g_ctx.width, g_ctx.height, g_ctx.batch,
(elapsedTime / (TEST_LOOP * 1.0f)),
(elapsedTime / (TEST_LOOP * 1.0f)) / g_ctx.batch);
sprintf(filename, "converted_bgr_%dx%d", g_ctx.width, g_ctx.height);
/* resize bgr 3 progressive planars */
cudaEventRecord(start, 0);
for (int i = 0; i < TEST_LOOP; i++) {
resizeBGRplanarBatch(d_bgr, g_ctx.ctx_pitch, g_ctx.width, g_ctx.height,
d_resizedBGR, g_ctx.dst_width, g_ctx.dst_width,
g_ctx.dst_height, g_ctx.batch);
}
cudaEventRecord(stop, 0);
cudaEventSynchronize(stop);
cudaEventElapsedTime(&elapsedTime, start, stop);
printf(
" CUDA resize bgr(%dx%d --> %dx%d), batch: %d,"
" average time: %.3f ms ==> %.3f ms/frame\n",
g_ctx.width, g_ctx.height, g_ctx.dst_width, g_ctx.dst_height, g_ctx.batch,
(elapsedTime / (TEST_LOOP * 1.0f)),
(elapsedTime / (TEST_LOOP * 1.0f)) / g_ctx.batch);
memset(filename, 0, sizeof(filename));
sprintf(filename, "resized_bgr_%dx%d", g_ctx.dst_width, g_ctx.dst_height);
dumpBGR(d_resizedBGR, g_ctx.dst_pitch, g_ctx.dst_width, g_ctx.dst_height,
g_ctx.batch, (char *)"t2", filename);
/* release resources */
checkCudaErrors(cudaEventDestroy(start));
checkCudaErrors(cudaEventDestroy(stop));
checkCudaErrors(cudaStreamDestroy(stream));
checkCudaErrors(cudaFree(d_bgr));
checkCudaErrors(cudaFree(d_resizedBGR));
}
int main(int argc, char *argv[]) {
unsigned char *d_inputNV12;
if (parseCmdLine(argc, argv) < 0) return EXIT_FAILURE;
g_ctx.ctx_pitch = g_ctx.width;
int ctx_alignment = 32;
g_ctx.ctx_pitch += (g_ctx.ctx_pitch % ctx_alignment != 0)
? (ctx_alignment - g_ctx.ctx_pitch % ctx_alignment)
: 0;
g_ctx.ctx_heights = ceil(g_ctx.height * 3.0f / 2.0f);
/* load nv12 yuv data into d_inputNV12 with batch of copies */
#if USE_UVM_MEM
checkCudaErrors(cudaMallocManaged(
(void **)&d_inputNV12,
(g_ctx.ctx_pitch * g_ctx.ctx_heights * g_ctx.batch), cudaMemAttachHost));
printf("\nUSE_UVM_MEM\n");
#else
checkCudaErrors(
cudaMalloc((void **)&d_inputNV12,
(g_ctx.ctx_pitch * g_ctx.ctx_heights * g_ctx.batch)));
#endif
if (loadNV12Frame(d_inputNV12)) {
std::cerr << "failed to load batch data!\n";
return EXIT_FAILURE;
}
/* firstly resize nv12, then convert nv12 to bgr */
printf("\nTEST#1:\n");
nv12ResizeAndNV12ToBGR(d_inputNV12);
/* first convert nv12 to bgr, then resize bgr */
printf("\nTEST#2:\n");
nv12ToBGRandBGRresize(d_inputNV12);
checkCudaErrors(cudaFree(d_inputNV12));
return EXIT_SUCCESS;
}

View File

@@ -0,0 +1,152 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#include <stdlib.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <fstream>
#include <iostream>
#include <cuda.h>
#include <cuda_runtime.h>
#include "resize_convert.h"
#include "utils.h"
__global__ void floatToChar(float *src, unsigned char *dst, int height,
int width, int batchSize) {
int x = threadIdx.x + blockIdx.x * blockDim.x;
if (x >= height * width) return;
int offset = height * width * 3;
for (int j = 0; j < batchSize; j++) {
// b
*(dst + j * offset + x * 3 + 0) =
(unsigned char)*(src + j * offset + height * width * 0 + x);
// g
*(dst + j * offset + x * 3 + 1) =
(unsigned char)*(src + j * offset + height * width * 1 + x);
// r
*(dst + j * offset + x * 3 + 2) =
(unsigned char)*(src + j * offset + height * width * 2 + x);
}
}
void floatPlanarToChar(float *src, unsigned char *dst, int height, int width,
int batchSize) {
floatToChar<<<(height * width - 1) / 1024 + 1, 1024, 0, NULL>>>(
src, dst, height, width, batchSize);
}
void dumpRawBGR(float *d_srcBGR, int pitch, int width, int height,
int batchSize, char *folder, char *tag) {
float *bgr, *d_bgr;
int frameSize;
char directory[120];
char mkdir_cmd[256];
#if !defined(_WIN32)
sprintf(directory, "output/%s", folder);
sprintf(mkdir_cmd, "mkdir -p %s 2> /dev/null", directory);
#else
sprintf(directory, "output\\%s", folder);
sprintf(mkdir_cmd, "mkdir %s 2> nul", directory);
#endif
int ret = system(mkdir_cmd);
frameSize = width * height * 3 * sizeof(float);
bgr = (float *)malloc(frameSize);
if (bgr == NULL) {
std::cerr << "Failed malloc for bgr\n";
return;
}
d_bgr = d_srcBGR;
for (int i = 0; i < batchSize; i++) {
char filename[120];
std::ofstream *outputFile;
checkCudaErrors(cudaMemcpy((void *)bgr, (void *)d_bgr, frameSize,
cudaMemcpyDeviceToHost));
sprintf(filename, "%s/%s_%d.raw", directory, tag, (i + 1));
outputFile = new std::ofstream(filename);
if (outputFile) {
outputFile->write((char *)bgr, frameSize);
delete outputFile;
}
d_bgr += pitch * height * 3;
}
free(bgr);
}
void dumpBGR(float *d_srcBGR, int pitch, int width, int height, int batchSize,
char *folder, char *tag) {
dumpRawBGR(d_srcBGR, pitch, width, height, batchSize, folder, tag);
}
void dumpYUV(unsigned char *d_nv12, int size, char *folder, char *tag) {
unsigned char *nv12Data;
std::ofstream *nv12File;
char filename[120];
char directory[60];
char mkdir_cmd[256];
#if !defined(_WIN32)
sprintf(directory, "output/%s", folder);
sprintf(mkdir_cmd, "mkdir -p %s 2> /dev/null", directory);
#else
sprintf(directory, "output\\%s", folder);
sprintf(mkdir_cmd, "mkdir %s 2> nul", directory);
#endif
int ret = system(mkdir_cmd);
sprintf(filename, "%s/%s.nv12", directory, tag);
nv12File = new std::ofstream(filename);
if (nv12File == NULL) {
std::cerr << "Failed to new " << filename;
return;
}
nv12Data = (unsigned char *)malloc(size * (sizeof(char)));
if (nv12Data == NULL) {
std::cerr << "Failed to allcoate memory\n";
return;
}
cudaMemcpy((void *)nv12Data, (void *)d_nv12, size, cudaMemcpyDeviceToHost);
nv12File->write((const char *)nv12Data, size);
free(nv12Data);
delete nv12File;
}

View File

@@ -0,0 +1,37 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#ifndef __H_UTIL_
#define __H_UTIL_
extern "C"
void dumpBGR(float *d_srcBGR, int pitch, int width, int height,
int batchSize, char *folder, char *tag);
extern "C"
void dumpYUV(unsigned char *d_nv12, int size, char *folder, char *tag);
#endif

View File

@@ -0,0 +1,316 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * 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.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
#
################################################################################
#
# Makefile project only supported on Mac OS X and Linux Platforms)
#
################################################################################
# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda
##############################
# start deprecated interface #
##############################
ifeq ($(x86_64),1)
$(info WARNING - x86_64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
TARGET_ARCH ?= x86_64
endif
ifeq ($(ARMv7),1)
$(info WARNING - ARMv7 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=armv7l instead)
TARGET_ARCH ?= armv7l
endif
ifeq ($(aarch64),1)
$(info WARNING - aarch64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
TARGET_ARCH ?= aarch64
endif
ifeq ($(ppc64le),1)
$(info WARNING - ppc64le variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
TARGET_ARCH ?= ppc64le
endif
ifneq ($(GCC),)
$(info WARNING - GCC variable has been deprecated)
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
HOST_COMPILER ?= $(GCC)
endif
ifneq ($(abi),)
$(error ERROR - abi variable has been removed)
endif
############################
# end deprecated interface #
############################
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
endif
else
TARGET_SIZE := $(shell getconf LONG_BIT)
endif
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
TARGET_ARCH = armv7l
endif
# operating system
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif
# host compiler
ifeq ($(TARGET_OS),darwin)
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
HOST_COMPILER ?= clang++
endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
ifeq ($(TARGET_OS),linux)
HOST_COMPILER ?= arm-linux-gnueabihf-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
else ifeq ($(TARGET_OS),android)
HOST_COMPILER ?= arm-linux-androideabi-g++
endif
else ifeq ($(TARGET_ARCH),aarch64)
ifeq ($(TARGET_OS), linux)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
endif
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
# internal flags
NVCCFLAGS := -m${TARGET_SIZE}
CCFLAGS :=
LDFLAGS :=
# build flags
ifeq ($(TARGET_OS),darwin)
LDFLAGS += -rpath $(CUDA_PATH)/lib
CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
LDFLAGS += -pie
CCFLAGS += -fpie -fpic -fexceptions
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif
# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif
# Debug build flags
ifeq ($(dbg),1)
NVCCFLAGS += -g -G
BUILD_TYPE := debug
else
BUILD_TYPE := release
endif
ALL_CCFLAGS :=
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
# This sample is not supported on QNX
ifeq ($(TARGET_OS),qnx)
$(info >>> WARNING - UnifiedMemoryPerf is not supported on QNX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
# Common includes and paths for CUDA
INCLUDES := -I../../Common
LIBRARIES :=
################################################################################
# Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 30 35 37 50 52 60 61 70 72 75
else
SMS ?= 30 35 37 50 52 60 61 70 75
endif
ifeq ($(SMS),)
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
HIGHEST_SM := $(lastword $(sort $(SMS)))
ifneq ($(HIGHEST_SM),)
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
endif
endif
ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif
################################################################################
# Target rules
all: build
build: UnifiedMemoryPerf
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@echo "Sample will be waived due to the above missing dependencies"
else
@echo "Sample is ready - all dependencies have been met"
endif
commonKernels.o:commonKernels.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
helperFunctions.o:helperFunctions.cpp
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
matrixMultiplyPerf.o:matrixMultiplyPerf.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
UnifiedMemoryPerf: commonKernels.o helperFunctions.o matrixMultiplyPerf.o
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
$(EXEC) ./UnifiedMemoryPerf
clean:
rm -f UnifiedMemoryPerf commonKernels.o helperFunctions.o matrixMultiplyPerf.o
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/UnifiedMemoryPerf
clobber: clean

View File

@@ -0,0 +1,85 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
<entry>
<name>UnifiedMemoryPerf</name>
<cuda_api_list>
<toolkit>cudaMallocManaged</toolkit>
<toolkit>cudaStreamAttachMemAsync</toolkit>
<toolkit>cudaMemcpyAsync</toolkit>
<toolkit>cudaMallocHost</toolkit>
<toolkit>cudaMalloc</toolkit>
</cuda_api_list>
<description><![CDATA[This sample demonstrates the performance comparision using matrix multiplication kernel of Unified Memory with/without hints and other types of memory like zero copy buffers, pageable, pagelocked memory performing synchronous and Asynchronous transfers on a single GPU.]]></description>
<devicecompilation>whole</devicecompilation>
<includepaths>
<path>./</path>
<path>../</path>
<path>../../common/inc</path>
</includepaths>
<keyconcepts>
<concept level="basic">CUDA Systems Integration</concept>
<concept level="basic">Unified Memory</concept>
<concept level="basic">CUDA Streams and Events</concept>
<concept level="basic">Pinned System Paged Memory</concept>
</keyconcepts>
<keywords>
<keyword>CUDA</keyword>
<keyword>Unified Memory</keyword>
<keyword>Pinned Memory</keyword>
<keyword>Zero copy buffer</keyword>
<keyword>UVM</keyword>
<keyword>Streams</keyword>
</keywords>
<libraries>
</libraries>
<librarypaths>
</librarypaths>
<nsight_eclipse>true</nsight_eclipse>
<primary_file>matrixMultiplyPerf.cu</primary_file>
<required_dependencies>
<dependency>UVM</dependency>
</required_dependencies>
<scopes>
<scope>1:CUDA Basic Topics</scope>
<scope>1:CUDA Systems Integration</scope>
<scope>1:Unified Memory</scope>
</scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch>
<sm-arch>sm52</sm-arch>
<sm-arch>sm60</sm-arch>
<sm-arch>sm61</sm-arch>
<sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch>
<supported_envs>
<env>
<arch>x86_64</arch>
<platform>linux</platform>
</env>
<env>
<arch>x86_64</arch>
<platform>macosx</platform>
</env>
<env>
<platform>windows7</platform>
</env>
<env>
<arch>arm</arch>
</env>
<env>
<arch>aarch64</arch>
</env>
<env>
<arch>ppc64le</arch>
<platform>linux</platform>
</env>
</supported_envs>
<supported_sm_architectures>
<from>3.0</from>
</supported_sm_architectures>
<title>Unified and other CUDA Memories Performance</title>
<type>exe</type>
</entry>

View File

@@ -0,0 +1,98 @@
# UnifiedMemoryPerf - Unified and other CUDA Memories Performance
## Description
This sample demonstrates the performance comparision using matrix multiplication kernel of Unified Memory with/without hints and other types of memory like zero copy buffers, pageable, pagelocked memory performing synchronous and Asynchronous transfers on a single GPU.
## Key Concepts
CUDA Systems Integration, Unified Memory, CUDA Streams and Events, Pinned System Paged Memory
## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
Linux, Windows, MacOSX
## Supported CPU Architecture
x86_64, ppc64le, armv7l, aarch64
## CUDA APIs involved
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
cudaMallocManaged, cudaStreamAttachMemAsync, cudaMemcpyAsync, cudaMallocHost, cudaMalloc
## Dependencies needed to build/run
[UVM](../../README.md#uvm)
## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run
### Windows
The Windows samples are built using the Visual Studio IDE. Solution files (.sln) are provided for each supported version of Visual Studio, using the format:
```
*_vs<version>.sln - for Visual Studio <version>
```
Each individual sample has its own set of solution files in its directory:
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check DirectX Dependencies section for details."
### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, ppc64le, armv7l, aarch64.
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=ppc64le` <br/> `$ make TARGET_ARCH=armv7l` <br/> `$ make TARGET_ARCH=aarch64` <br/>
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
```
$ make SMS="50 60"
```
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=g++
```
### Mac
The Mac samples are built using makefiles. To use the makefiles, change directory into the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where "A B ..." is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use SMS="50 60".
```
$ make SMS="A B ..."
```
* **HOST_COMPILER=<host_compiler>** - override the default clang host compiler. See the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=clang
```
## References (for more details)

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2012
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "UnifiedMemoryPerf", "UnifiedMemoryPerf_vs2012.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,110 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>UnifiedMemoryPerf_vs2012</RootNamespace>
<ProjectName>UnifiedMemoryPerf</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v110</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="commonKernels.cu" />
<ClCompile Include="helperFunctions.cpp" />
<CudaCompile Include="matrixMultiplyPerf.cu" />
<ClInclude Include="commonDefs.hpp" />
<ClInclude Include="commonKernels.hpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 13.00
# Visual Studio 2013
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "UnifiedMemoryPerf", "UnifiedMemoryPerf_vs2013.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,110 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>UnifiedMemoryPerf_vs2013</RootNamespace>
<ProjectName>UnifiedMemoryPerf</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v120</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="commonKernels.cu" />
<ClCompile Include="helperFunctions.cpp" />
<CudaCompile Include="matrixMultiplyPerf.cu" />
<ClInclude Include="commonDefs.hpp" />
<ClInclude Include="commonKernels.hpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 14.00
# Visual Studio 2015
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "UnifiedMemoryPerf", "UnifiedMemoryPerf_vs2015.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,110 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>UnifiedMemoryPerf_vs2015</RootNamespace>
<ProjectName>UnifiedMemoryPerf</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v140</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="commonKernels.cu" />
<ClCompile Include="helperFunctions.cpp" />
<CudaCompile Include="matrixMultiplyPerf.cu" />
<ClInclude Include="commonDefs.hpp" />
<ClInclude Include="commonKernels.hpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2017
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "UnifiedMemoryPerf", "UnifiedMemoryPerf_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,115 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>UnifiedMemoryPerf_vs2017</RootNamespace>
<ProjectName>UnifiedMemoryPerf</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
<LatestTargetPlatformVersion>$([Microsoft.Build.Utilities.ToolLocationHelper]::GetLatestSDKTargetPlatformVersion('Windows', '10.0'))</LatestTargetPlatformVersion>
<WindowsTargetPlatformVersion Condition="'$(WindowsTargetPlatformVersion)' == ''">$(LatestTargetPlatformVersion)</WindowsTargetPlatformVersion>
<TargetPlatformVersion>$(WindowsTargetPlatformVersion)</TargetPlatformVersion>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v141</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="commonKernels.cu" />
<ClCompile Include="helperFunctions.cpp" />
<CudaCompile Include="matrixMultiplyPerf.cu" />
<ClInclude Include="commonDefs.hpp" />
<ClInclude Include="commonKernels.hpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2019
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "UnifiedMemoryPerf", "UnifiedMemoryPerf_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,111 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>UnifiedMemoryPerf_vs2019</RootNamespace>
<ProjectName>UnifiedMemoryPerf</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v142</PlatformToolset>
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="commonKernels.cu" />
<ClCompile Include="helperFunctions.cpp" />
<CudaCompile Include="matrixMultiplyPerf.cu" />
<ClInclude Include="commonDefs.hpp" />
<ClInclude Include="commonKernels.hpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,88 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#ifndef _COMMON_DEFS_
#define _COMMON_DEFS_
#include <cuda.h>
#define ONE_KB 1024
#define ONE_MB (ONE_KB * ONE_KB)
extern size_t maxSampleSizeInMb;
extern int numKernelRuns;
extern int verboseResults;
extern unsigned int findNumSizesToTest(unsigned int minSize,
unsigned int maxSize,
unsigned int multiplier);
// For Tracking the different memory allocation types
typedef enum memAllocType_enum {
MEMALLOC_TYPE_START,
USE_MANAGED_MEMORY_WITH_HINTS = MEMALLOC_TYPE_START,
USE_MANAGED_MEMORY_WITH_HINTS_ASYNC,
USE_MANAGED_MEMORY,
USE_ZERO_COPY,
USE_HOST_PAGEABLE_AND_DEVICE_MEMORY,
USE_HOST_PAGEABLE_AND_DEVICE_MEMORY_ASYNC,
USE_HOST_PAGELOCKED_AND_DEVICE_MEMORY,
USE_HOST_PAGELOCKED_AND_DEVICE_MEMORY_ASYNC,
MEMALLOC_TYPE_END = USE_HOST_PAGELOCKED_AND_DEVICE_MEMORY_ASYNC,
MEMALLOC_TYPE_INVALID,
MEMALLOC_TYPE_COUNT = MEMALLOC_TYPE_INVALID
} MemAllocType;
typedef enum bandwidthType_enum {
READ_BANDWIDTH,
WRITE_BANDWIDTH
} BandwidthType;
extern const char *memAllocTypeStr[];
extern const char *memAllocTypeShortStr[];
struct resultsData;
struct testResults;
void createAndInitTestResults(struct testResults **results,
const char *testName,
unsigned int numMeasurements,
unsigned int numSizesToTest);
unsigned long *getPtrSizesToTest(struct testResults *results);
void freeTestResultsAndAllResultsData(struct testResults *results);
void createResultDataAndAddToTestResults(struct resultsData **ptrData,
struct testResults *results,
const char *resultsName,
bool printOnlyInVerbose,
bool reportAsBandwidth);
double *getPtrRunTimesInMs(struct resultsData *data, int allocType,
int sizeIndex);
void printResults(struct testResults *results,
bool print_launch_transfer_results, bool print_std_deviation);
#endif

View File

@@ -0,0 +1,33 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#include "commonKernels.hpp"
__global__ void spinWhileLessThanOne(volatile unsigned int *latch) {
while (latch[0] < 1)
;
}

View File

@@ -0,0 +1,28 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
__global__ void spinWhileLessThanOne(volatile unsigned int *latch);

View File

@@ -0,0 +1,303 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#include <stdio.h>
#include <string.h>
#include "commonDefs.hpp"
#define CU_INIT_UUID
#include <cmath>
#define UNITS_Time "ms"
#define UNITS_BW "MB/s"
#define KB_str "KB"
#define MB_str "MB"
struct resultsData {
char resultsName[64];
struct testResults *results;
// this has MEMALLOC_TYPE_COUNT * results->numSizesToTest *
// results->numMeasurements elements
double **runTimesInMs[MEMALLOC_TYPE_COUNT];
double *averageRunTimesInMs[MEMALLOC_TYPE_COUNT];
double *stdDevRunTimesInMs[MEMALLOC_TYPE_COUNT];
double *stdDevBandwidthInMBps[MEMALLOC_TYPE_COUNT];
bool printOnlyInVerbose;
bool reportAsBandwidth;
struct resultsData *next;
};
struct testResults {
char testName[64];
unsigned int numMeasurements;
unsigned long *sizesToTest;
unsigned int numSizesToTest;
struct resultsData *resultsDataHead;
struct resultsData *resultsDataTail;
};
unsigned int findNumSizesToTest(unsigned int minSize, unsigned int maxSize,
unsigned int multiplier) {
unsigned int numSizesToTest = 0;
while (minSize <= maxSize) {
numSizesToTest++;
minSize *= multiplier;
}
return numSizesToTest;
}
int compareDoubles(const void *ptr1, const void *ptr2) {
return (*(double *)ptr1 > *(double *)ptr2) ? 1 : -1;
}
static inline double getTimeOrBandwidth(double runTimeInMs, unsigned long size,
bool getBandwidth) {
return (getBandwidth) ? (1000 * (size / runTimeInMs)) / ONE_MB : runTimeInMs;
}
void createAndInitTestResults(struct testResults **ptrResults,
const char *testName,
unsigned int numMeasurements,
unsigned int numSizesToTest) {
unsigned int i;
struct testResults *results;
results = (struct testResults *)malloc(sizeof(struct testResults));
memset(results, 0, sizeof(struct testResults));
strcpy(results->testName, testName);
results->numMeasurements = numMeasurements;
results->numSizesToTest = numSizesToTest;
results->sizesToTest =
(unsigned long *)malloc(numSizesToTest * sizeof(unsigned long));
results->resultsDataHead = NULL;
results->resultsDataTail = NULL;
*ptrResults = results;
}
unsigned long *getPtrSizesToTest(struct testResults *results) {
return results->sizesToTest;
}
void createResultDataAndAddToTestResults(struct resultsData **ptrData,
struct testResults *results,
const char *resultsName,
bool printOnlyInVerbose,
bool reportAsBandwidth) {
unsigned int i, j;
struct resultsData *data;
data = (struct resultsData *)malloc(sizeof(struct resultsData));
memset(data, 0, sizeof(struct resultsData));
strcpy(data->resultsName, resultsName);
data->results = results;
for (i = 0; i < MEMALLOC_TYPE_COUNT; i++) {
data->runTimesInMs[i] =
(double **)malloc(results->numSizesToTest * sizeof(double *));
for (j = 0; j < results->numSizesToTest; j++) {
data->runTimesInMs[i][j] =
(double *)malloc(results->numMeasurements * sizeof(double));
}
data->averageRunTimesInMs[i] =
(double *)malloc(results->numSizesToTest * sizeof(double));
data->stdDevRunTimesInMs[i] =
(double *)malloc(results->numSizesToTest * sizeof(double));
data->stdDevBandwidthInMBps[i] =
(double *)malloc(results->numSizesToTest * sizeof(double));
}
data->printOnlyInVerbose = printOnlyInVerbose;
data->reportAsBandwidth = reportAsBandwidth;
data->next = NULL;
*ptrData = data;
if (results->resultsDataHead == NULL) {
results->resultsDataHead = data;
results->resultsDataTail = data;
} else {
results->resultsDataTail->next = data;
results->resultsDataTail = data;
}
}
double *getPtrRunTimesInMs(struct resultsData *data, int allocType,
int sizeIndex) {
return data->runTimesInMs[allocType][sizeIndex];
}
void freeTestResultsAndAllResultsData(struct testResults *results) {
struct resultsData *data, *dataToFree;
unsigned int i, j;
for (data = results->resultsDataHead; data != NULL;) {
for (i = 0; i < MEMALLOC_TYPE_COUNT; i++) {
for (j = 0; j < results->numSizesToTest; j++) {
free(data->runTimesInMs[i][j]);
}
free(data->runTimesInMs[i]);
free(data->averageRunTimesInMs[i]);
free(data->stdDevRunTimesInMs[i]);
free(data->stdDevBandwidthInMBps[i]);
}
dataToFree = data;
data = data->next;
free(dataToFree);
}
free(results->sizesToTest);
free(results);
}
void calculateAverageAndStdDev(double *pAverage, double *pStdDev,
double *allResults, unsigned int count) {
unsigned int i;
double average = 0.0;
double stdDev = 0.0;
for (i = 0; i < count; i++) {
average += allResults[i];
}
average /= count;
for (i = 0; i < count; i++) {
stdDev += (allResults[i] - average) * (allResults[i] - average);
}
stdDev /= count;
stdDev = sqrt(stdDev);
*pAverage = average;
*pStdDev = (average == 0.0) ? 0.0 : ((100.0 * stdDev) / average);
}
void calculateStdDevBandwidth(double *pStdDev, double *allResults,
unsigned int count, unsigned long size) {
unsigned int i;
double bandwidth;
double average = 0.0;
double stdDev = 0.0;
for (i = 0; i < count; i++) {
bandwidth = (1000 * (size / allResults[i])) / ONE_MB;
average += bandwidth;
}
average /= count;
for (i = 0; i < count; i++) {
bandwidth = (1000 * (size / allResults[i])) / ONE_MB;
stdDev += (bandwidth - average) * (bandwidth - average);
}
stdDev /= count;
stdDev = sqrt(stdDev);
*pStdDev = (average == 0.0) ? 0.0 : ((100.0 * stdDev) / average);
}
void printTimesInTableFormat(struct testResults *results,
struct resultsData *data, bool printAverage,
bool printStdDev) {
unsigned int i, j;
bool printStdDevBandwidth = printStdDev && data->reportAsBandwidth;
printf("Size_KB");
for (i = 0; i < MEMALLOC_TYPE_COUNT; i++) {
printf("\t%7s", memAllocTypeShortStr[i]);
}
printf("\n");
for (j = 0; j < results->numSizesToTest; j++) {
printf("%lu", results->sizesToTest[j] / ONE_KB);
for (i = 0; i < MEMALLOC_TYPE_COUNT; i++) {
printf(data->reportAsBandwidth ? "\t%7.2lf" : "\t%7.3lf",
printStdDevBandwidth
? data->stdDevBandwidthInMBps[i][j]
: getTimeOrBandwidth(
printAverage ? data->averageRunTimesInMs[i][j]
: data->stdDevRunTimesInMs[i][j],
results->sizesToTest[j], data->reportAsBandwidth));
}
printf("\n");
}
}
void printAllResultsInVerboseMode(struct testResults *results,
struct resultsData *data) {
unsigned int i, j, k;
for (i = 0; i < MEMALLOC_TYPE_COUNT; i++) {
printf("Verbose mode, printing all results for %s\n", memAllocTypeStr[i]);
printf("Instance");
for (j = 0; j < results->numSizesToTest; j++) {
printf("\t%lu", results->sizesToTest[j] / ONE_KB);
}
printf("\n");
for (k = 0; k < results->numMeasurements; k++) {
printf("%u", k);
for (j = 0; j < results->numSizesToTest; j++) {
printf(data->reportAsBandwidth ? "\t%7.2lf" : "\t%7.3lf",
getTimeOrBandwidth(data->runTimesInMs[i][j][k],
results->sizesToTest[j],
data->reportAsBandwidth));
}
printf("\n");
}
}
}
void printResults(struct testResults *results,
bool print_launch_transfer_results,
bool print_std_deviation) {
char vulcanPrint[256];
char resultNameNoSpaces[64];
unsigned int i, j, k;
struct resultsData *resultsIter;
bool sizeGreaterThan1MB;
for (resultsIter = results->resultsDataHead; resultsIter != NULL;
resultsIter = resultsIter->next) {
if (!verboseResults && resultsIter->printOnlyInVerbose) {
continue;
}
if (!print_launch_transfer_results) {
if (!(strcmp(resultsIter->resultsName, "Overall Time") == 0)) {
continue;
}
}
// regular print
printf("\n%s For %s ", resultsIter->resultsName, results->testName);
printf("\n");
for (j = 0; j < results->numSizesToTest; j++) {
for (i = 0; i < MEMALLOC_TYPE_COUNT; i++) {
calculateAverageAndStdDev(&resultsIter->averageRunTimesInMs[i][j],
&resultsIter->stdDevRunTimesInMs[i][j],
resultsIter->runTimesInMs[i][j],
results->numMeasurements);
if (resultsIter->reportAsBandwidth) {
calculateStdDevBandwidth(&resultsIter->stdDevBandwidthInMBps[i][j],
resultsIter->runTimesInMs[i][j],
results->numMeasurements,
results->sizesToTest[j]);
}
}
}
printf("\nPrinting Average of %u measurements in (%s)\n",
results->numMeasurements,
resultsIter->reportAsBandwidth ? UNITS_BW : UNITS_Time);
printTimesInTableFormat(results, resultsIter, true, false);
if (print_std_deviation) {
printf(
"\nPrinting Standard Deviation as %% of average of %u measurements\n",
results->numMeasurements);
printTimesInTableFormat(results, resultsIter, false, true);
}
if (verboseResults) {
printAllResultsInVerboseMode(results, resultsIter);
}
}
}

View File

@@ -0,0 +1,697 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
#include <helper_cuda.h>
#include <helper_timer.h>
#include "commonDefs.hpp"
#include "commonKernels.hpp"
#define VERIFY_GPU_CORRECTNESS 0
size_t maxSampleSizeInMb = 64;
int numKernelRuns = 100;
int verboseResults = 0;
const char *memAllocTypeStr[MEMALLOC_TYPE_COUNT] = {
"Managed_Memory_With_Hints",
"Managed_Memory_With_Hints_FullyAsync",
"Managed_Memory_NoHints",
"Zero_Copy",
"Memcpy_HostMalloc_DeviceCudaMalloc",
"MemcpyAsync_HostMalloc_DeviceCudaMalloc",
"Memcpy_HostCudaHostAlloc_DeviceCudaMalloc",
"MemcpyAsync_HostCudaHostAlloc_DeviceCudaMalloc"};
const char *memAllocTypeShortStr[MEMALLOC_TYPE_COUNT] = {
"UMhint", // Managed Memory With Hints
"UMhntAs", // Managed Memory With_Hints Async
"UMeasy", // Managed_Memory with No Hints
"0Copy", // Zero Copy
"MemCopy", // USE HOST PAGEABLE AND DEVICE_MEMORY
"CpAsync", // USE HOST PAGEABLE AND DEVICE_MEMORY ASYNC
"CpHpglk", // USE HOST PAGELOCKED AND DEVICE MEMORY
"CpPglAs" // USE HOST PAGELOCKED AND DEVICE MEMORY ASYNC
};
static float RandFloat(float low, float high) {
float t = (float)rand() / (float)RAND_MAX;
return (1.0f - t) * low + t * high;
}
void fillMatrixWithRandomValues(float *matrix, unsigned int matrixDim) {
unsigned int i, j;
for (i = 0; i < matrixDim; ++i) {
for (j = 0; j < matrixDim; ++j) {
matrix[j + i * matrixDim] = RandFloat(0.0f, 10.0f);
}
}
}
#if VERIFY_GPU_CORRECTNESS
void verifyMatrixMultiplyCorrectness(float *C, float *A, float *B,
unsigned int matrixDim) {
unsigned int i, j, k, numErrors = 0;
for (i = 0; i < matrixDim; ++i) {
for (j = 0; j < matrixDim; ++j) {
float result = 0.0f;
for (k = 0; k < matrixDim; ++k) {
result += A[k + i * matrixDim] * B[j + k * matrixDim];
}
if (fabs(C[j + i * matrixDim] - result) > 0.001 * matrixDim) {
printf("At [%u, %u]: Expected %f, Found %f\n", i, j, result,
C[j + i * matrixDim]);
++numErrors;
}
}
}
if (numErrors != 0) {
printf("%d value mismatches occured\n", numErrors);
fflush(stdout);
exit(EXIT_FAILURE); // exit since value mismatches occured
}
}
#endif
void copyMatrix(float *dstMatrix, float *srcMatrix, unsigned int matrixDim) {
size_t size = matrixDim * matrixDim * sizeof(float);
memcpy(dstMatrix, srcMatrix, size);
}
void verifyMatrixData(float *expectedData, float *observedData,
unsigned int matrixDim) {
unsigned int i, j, numErrors = 0;
for (i = 0; i < matrixDim; ++i) {
for (j = 0; j < matrixDim; ++j) {
if (expectedData[j + i * matrixDim] != observedData[j + i * matrixDim]) {
++numErrors;
if (verboseResults) {
printf("At [%u, %u]: Expected %f, Found %f\n", i, j,
expectedData[j + i * matrixDim],
observedData[j + i * matrixDim]);
}
}
}
}
if (numErrors != 0) {
printf("%d value mismatches occured\n", numErrors);
fflush(stdout);
exit(EXIT_FAILURE); // exit since value mismatches occured
}
}
#define BLOCK_SIZE 32
__global__ void matrixMultiplyKernel(float *C, float *A, float *B,
unsigned int matrixDim) {
// Block index
int bx = blockIdx.x;
int by = blockIdx.y;
// Thread index
int tx = threadIdx.x;
int ty = threadIdx.y;
unsigned int wA = matrixDim;
unsigned int wB = matrixDim;
// Index of the first sub-matrix of A processed by the block
int aBegin = matrixDim * BLOCK_SIZE * by;
// Index of the last sub-matrix of A processed by the block
int aEnd = aBegin + wA - 1;
// Step size used to iterate through the sub-matrices of A
int aStep = BLOCK_SIZE;
// Index of the first sub-matrix of B processed by the block
int bBegin = BLOCK_SIZE * bx;
// Step size used to iterate through the sub-matrices of B
int bStep = BLOCK_SIZE * wB;
// Csub is used to store the element of the block sub-matrix
// that is computed by the thread
float Csub = 0;
// Loop over all the sub-matrices of A and B
// required to compute the block sub-matrix
for (int a = aBegin, b = bBegin; 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];
// Declaration of the shared memory array Bs used to
// store the sub-matrix of B
__shared__ float Bs[BLOCK_SIZE][BLOCK_SIZE];
// Load the matrices from device memory
// to shared memory; each thread loads
// one element of each matrix
As[ty][tx] = A[a + wA * ty + tx];
Bs[ty][tx] = B[b + wB * ty + tx];
// Synchronize to make sure the matrices are loaded
__syncthreads();
// Multiply the two matrices together;
// each thread computes one element
// of the block sub-matrix
#pragma unroll
for (int 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
// sub-matrices of A and B in the next iteration
__syncthreads();
}
// Write the block sub-matrix to device memory;
// each thread writes one element
int c = wB * BLOCK_SIZE * by + BLOCK_SIZE * bx;
C[c + wB * ty + tx] = Csub;
}
void runMatrixMultiplyKernel(unsigned int matrixDim, int allocType,
unsigned int numLoops, double *gpuLaunchCallsTimes,
double *gpuTransferToCallsTimes,
double *gpuTransferFromCallsTimes,
double *gpuLaunchAndTransferCallsTimes,
double *gpuLaunchTransferSyncTimes,
double *cpuAccessTimes, double *overallTimes,
int device_id) {
float *dptrA = NULL, *hptrA = NULL;
float *dptrB = NULL, *hptrB = NULL;
float *dptrC = NULL, *hptrC = NULL;
float *randValuesX = NULL, *randValuesY = NULL;
float *randValuesVerifyXmulY = NULL, *randValuesVerifyYmulX = NULL;
bool copyRequired = false, hintsRequired = false;
bool someTransferOpRequired;
bool isAsync = false;
cudaStream_t streamToRunOn;
unsigned int *latch;
size_t size = matrixDim * matrixDim * sizeof(float);
dim3 threads(32, 32);
dim3 grid(matrixDim / threads.x, matrixDim / threads.y);
StopWatchInterface *gpuLaunchCallsTimer = 0, *gpuTransferCallsTimer = 0;
StopWatchInterface *gpuSyncTimer = 0, *cpuAccessTimer = 0;
sdkCreateTimer(&gpuLaunchCallsTimer);
sdkCreateTimer(&gpuTransferCallsTimer);
sdkCreateTimer(&gpuSyncTimer);
sdkCreateTimer(&cpuAccessTimer);
unsigned int i;
cudaDeviceProp deviceProp;
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, device_id));
checkCudaErrors(cudaStreamCreate(&streamToRunOn));
randValuesX = (float *)malloc(size);
if (!randValuesX) {
exit(EXIT_FAILURE); // exit since memory allocation error
}
randValuesY = (float *)malloc(size);
if (!randValuesY) {
exit(EXIT_FAILURE); // exit since memory allocation error
}
randValuesVerifyXmulY = (float *)malloc(size);
if (!randValuesVerifyXmulY) {
exit(EXIT_FAILURE); // exit since memory allocation error
}
randValuesVerifyYmulX = (float *)malloc(size);
if (!randValuesVerifyYmulX) {
exit(EXIT_FAILURE); // exit since memory allocation error
}
checkCudaErrors(cudaMalloc(&dptrA, size));
checkCudaErrors(cudaMalloc(&dptrB, size));
checkCudaErrors(cudaMalloc(&dptrC, size));
fillMatrixWithRandomValues(randValuesX, matrixDim);
fillMatrixWithRandomValues(randValuesY, matrixDim);
checkCudaErrors(
cudaMemcpyAsync(dptrA, randValuesX, size, cudaMemcpyHostToDevice));
checkCudaErrors(
cudaMemcpyAsync(dptrB, randValuesY, size, cudaMemcpyHostToDevice));
matrixMultiplyKernel<<<grid, threads>>>(dptrC, dptrA, dptrB, matrixDim);
checkCudaErrors(cudaMemcpyAsync(randValuesVerifyXmulY, dptrC, size,
cudaMemcpyDeviceToHost));
checkCudaErrors(cudaStreamSynchronize(NULL));
matrixMultiplyKernel<<<grid, threads>>>(dptrC, dptrB, dptrA, matrixDim);
checkCudaErrors(cudaMemcpyAsync(randValuesVerifyYmulX, dptrC, size,
cudaMemcpyDeviceToHost));
checkCudaErrors(cudaStreamSynchronize(NULL));
#if VERIFY_GPU_CORRECTNESS
verifyMatrixMultiplyCorrectness(randValuesVerifyXmulY, randValuesX,
randValuesY, matrixDim);
verifyMatrixMultiplyCorrectness(randValuesVerifyYmulX, randValuesY,
randValuesX, matrixDim);
#endif
checkCudaErrors(cudaFree(dptrA));
checkCudaErrors(cudaFree(dptrB));
checkCudaErrors(cudaFree(dptrC));
checkCudaErrors(cudaMallocHost(&latch, sizeof(unsigned int)));
switch (allocType) {
case USE_HOST_PAGEABLE_AND_DEVICE_MEMORY:
case USE_HOST_PAGEABLE_AND_DEVICE_MEMORY_ASYNC:
hptrA = (float *)malloc(size);
if (!hptrA) {
exit(EXIT_FAILURE); // exit since memory allocation error
}
hptrB = (float *)malloc(size);
if (!hptrB) {
exit(EXIT_FAILURE); // exit since memory allocation error
}
hptrC = (float *)malloc(size);
if (!hptrC) {
exit(EXIT_FAILURE); // exit since memory allocation error
}
checkCudaErrors(cudaMalloc(&dptrA, size));
checkCudaErrors(cudaMalloc(&dptrB, size));
checkCudaErrors(cudaMalloc(&dptrC, size));
copyRequired = true;
break;
case USE_HOST_PAGELOCKED_AND_DEVICE_MEMORY:
case USE_HOST_PAGELOCKED_AND_DEVICE_MEMORY_ASYNC:
checkCudaErrors(cudaMallocHost(&hptrA, size));
checkCudaErrors(cudaMallocHost(&hptrB, size));
checkCudaErrors(cudaMallocHost(&hptrC, size));
checkCudaErrors(cudaMalloc(&dptrA, size));
checkCudaErrors(cudaMalloc(&dptrB, size));
checkCudaErrors(cudaMalloc(&dptrC, size));
copyRequired = true;
break;
case USE_ZERO_COPY:
checkCudaErrors(cudaMallocHost(&hptrA, size));
checkCudaErrors(cudaMallocHost(&hptrB, size));
checkCudaErrors(cudaMallocHost(&hptrC, size));
checkCudaErrors(cudaHostGetDevicePointer(&dptrA, hptrA, 0));
checkCudaErrors(cudaHostGetDevicePointer(&dptrB, hptrB, 0));
checkCudaErrors(cudaHostGetDevicePointer(&dptrC, hptrC, 0));
break;
case USE_MANAGED_MEMORY:
checkCudaErrors(cudaMallocManaged(&dptrA, size));
checkCudaErrors(cudaMallocManaged(&dptrB, size));
checkCudaErrors(cudaMallocManaged(&dptrC, size));
hptrA = dptrA;
hptrB = dptrB;
hptrC = dptrC;
break;
case USE_MANAGED_MEMORY_WITH_HINTS:
case USE_MANAGED_MEMORY_WITH_HINTS_ASYNC:
if (deviceProp.concurrentManagedAccess) {
checkCudaErrors(cudaMallocManaged(&dptrA, size));
checkCudaErrors(cudaMallocManaged(&dptrB, size));
checkCudaErrors(cudaMallocManaged(&dptrC, size));
checkCudaErrors(cudaMemPrefetchAsync(dptrA, size, cudaCpuDeviceId));
checkCudaErrors(cudaMemPrefetchAsync(dptrB, size, cudaCpuDeviceId));
checkCudaErrors(cudaMemPrefetchAsync(dptrC, size, cudaCpuDeviceId));
} else {
checkCudaErrors(cudaMallocManaged(&dptrA, size, cudaMemAttachHost));
checkCudaErrors(cudaMallocManaged(&dptrB, size, cudaMemAttachHost));
checkCudaErrors(cudaMallocManaged(&dptrC, size, cudaMemAttachHost));
}
hptrA = dptrA;
hptrB = dptrB;
hptrC = dptrC;
hintsRequired = true;
break;
default:
exit(EXIT_FAILURE); // exit with error
}
if (allocType == USE_HOST_PAGEABLE_AND_DEVICE_MEMORY_ASYNC ||
allocType == USE_HOST_PAGELOCKED_AND_DEVICE_MEMORY_ASYNC ||
allocType == USE_MANAGED_MEMORY_WITH_HINTS_ASYNC) {
isAsync = true;
}
someTransferOpRequired = copyRequired || hintsRequired;
// fill buffers with 0 to avoid any first access page-fault overheads.
memset(hptrA, 0, size);
memset(hptrB, 0, size);
memset(hptrC, 0, size);
for (i = 0; i < numLoops; i++) {
cpuAccessTimes[i] = 0.0;
gpuLaunchCallsTimes[i] = 0.0;
gpuTransferToCallsTimes[i] = 0.0;
gpuTransferFromCallsTimes[i] = 0.0;
sdkStartTimer(&cpuAccessTimer);
{
copyMatrix(hptrA, (i & 0x1 == 0) ? randValuesX : randValuesY, matrixDim);
copyMatrix(hptrB, (i & 0x1 == 0) ? randValuesY : randValuesX, matrixDim);
}
sdkStopTimer(&cpuAccessTimer);
cpuAccessTimes[i] += sdkGetAverageTimerValue(&cpuAccessTimer);
sdkResetTimer(&cpuAccessTimer);
if (isAsync && hintsRequired) {
*latch = 0;
// Prevent any work on stream from starting until all work is pushed
spinWhileLessThanOne<<<1, 1, 0, streamToRunOn>>>(latch);
}
if (someTransferOpRequired) {
sdkStartTimer(&gpuTransferCallsTimer);
if (copyRequired) {
if (isAsync) {
checkCudaErrors(cudaMemcpyAsync(
dptrA, hptrA, size, cudaMemcpyHostToDevice, streamToRunOn));
checkCudaErrors(cudaMemcpyAsync(
dptrB, hptrB, size, cudaMemcpyHostToDevice, streamToRunOn));
} else {
checkCudaErrors(
cudaMemcpy(dptrA, hptrA, size, cudaMemcpyHostToDevice));
checkCudaErrors(
cudaMemcpy(dptrB, hptrB, size, cudaMemcpyHostToDevice));
}
}
if (hintsRequired) {
if (deviceProp.concurrentManagedAccess) {
checkCudaErrors(
cudaMemPrefetchAsync(dptrA, size, device_id, streamToRunOn));
checkCudaErrors(
cudaMemPrefetchAsync(dptrB, size, device_id, streamToRunOn));
checkCudaErrors(
cudaMemPrefetchAsync(dptrC, size, device_id, streamToRunOn));
} else {
checkCudaErrors(cudaStreamAttachMemAsync(streamToRunOn, dptrA, 0,
cudaMemAttachGlobal));
checkCudaErrors(cudaStreamAttachMemAsync(streamToRunOn, dptrB, 0,
cudaMemAttachGlobal));
checkCudaErrors(cudaStreamAttachMemAsync(streamToRunOn, dptrC, 0,
cudaMemAttachGlobal));
}
if (!isAsync) {
checkCudaErrors(cudaStreamSynchronize(streamToRunOn));
}
}
sdkStopTimer(&gpuTransferCallsTimer);
gpuTransferToCallsTimes[i] +=
sdkGetAverageTimerValue(&gpuTransferCallsTimer);
sdkResetTimer(&gpuTransferCallsTimer);
}
sdkStartTimer(&gpuLaunchCallsTimer);
{
matrixMultiplyKernel<<<grid, threads, 0, streamToRunOn>>>(
dptrC, dptrA, dptrB, matrixDim);
if (!isAsync) {
checkCudaErrors(cudaStreamSynchronize(streamToRunOn));
}
}
sdkStopTimer(&gpuLaunchCallsTimer);
gpuLaunchCallsTimes[i] += sdkGetAverageTimerValue(&gpuLaunchCallsTimer);
sdkResetTimer(&gpuLaunchCallsTimer);
if (someTransferOpRequired) {
sdkStartTimer(&gpuTransferCallsTimer);
if (hintsRequired) {
if (deviceProp.concurrentManagedAccess) {
checkCudaErrors(cudaMemPrefetchAsync(dptrA, size, cudaCpuDeviceId));
checkCudaErrors(cudaMemPrefetchAsync(dptrB, size, cudaCpuDeviceId));
checkCudaErrors(cudaMemPrefetchAsync(dptrC, size, cudaCpuDeviceId));
} else {
checkCudaErrors(cudaStreamAttachMemAsync(streamToRunOn, dptrA, 0,
cudaMemAttachHost));
checkCudaErrors(cudaStreamAttachMemAsync(streamToRunOn, dptrB, 0,
cudaMemAttachHost));
checkCudaErrors(cudaStreamAttachMemAsync(streamToRunOn, dptrC, 0,
cudaMemAttachHost));
}
if (!isAsync) {
checkCudaErrors(cudaStreamSynchronize(streamToRunOn));
}
}
if (copyRequired) {
if (isAsync) {
checkCudaErrors(cudaMemcpyAsync(
hptrC, dptrC, size, cudaMemcpyDeviceToHost, streamToRunOn));
} else {
checkCudaErrors(
cudaMemcpy(hptrC, dptrC, size, cudaMemcpyDeviceToHost));
}
}
sdkStopTimer(&gpuTransferCallsTimer);
gpuTransferFromCallsTimes[i] +=
sdkGetAverageTimerValue(&gpuTransferCallsTimer);
sdkResetTimer(&gpuTransferCallsTimer);
}
gpuLaunchAndTransferCallsTimes[i] = gpuLaunchCallsTimes[i] +
gpuTransferToCallsTimes[i] +
gpuTransferFromCallsTimes[i];
gpuLaunchTransferSyncTimes[i] = gpuLaunchAndTransferCallsTimes[i];
if (isAsync) {
sdkStartTimer(&gpuSyncTimer);
{
if (hintsRequired) {
*latch = 1;
}
checkCudaErrors(cudaStreamSynchronize(streamToRunOn));
}
sdkStopTimer(&gpuSyncTimer);
gpuLaunchTransferSyncTimes[i] += sdkGetAverageTimerValue(&gpuSyncTimer);
sdkResetTimer(&gpuSyncTimer);
}
sdkStartTimer(&cpuAccessTimer);
{
verifyMatrixData(
(i & 0x1 == 0) ? randValuesVerifyXmulY : randValuesVerifyYmulX, hptrC,
matrixDim);
}
sdkStopTimer(&cpuAccessTimer);
cpuAccessTimes[i] += sdkGetAverageTimerValue(&cpuAccessTimer);
sdkResetTimer(&cpuAccessTimer);
overallTimes[i] = cpuAccessTimes[i] + gpuLaunchTransferSyncTimes[i];
}
switch (allocType) {
case USE_HOST_PAGEABLE_AND_DEVICE_MEMORY:
case USE_HOST_PAGEABLE_AND_DEVICE_MEMORY_ASYNC:
free(hptrA);
free(hptrB);
free(hptrC);
checkCudaErrors(cudaFree(dptrA));
checkCudaErrors(cudaFree(dptrB));
checkCudaErrors(cudaFree(dptrC));
break;
case USE_HOST_PAGELOCKED_AND_DEVICE_MEMORY:
case USE_HOST_PAGELOCKED_AND_DEVICE_MEMORY_ASYNC:
checkCudaErrors(cudaFreeHost(hptrA));
checkCudaErrors(cudaFreeHost(hptrB));
checkCudaErrors(cudaFreeHost(hptrC));
checkCudaErrors(cudaFree(dptrA));
checkCudaErrors(cudaFree(dptrB));
checkCudaErrors(cudaFree(dptrC));
break;
case USE_ZERO_COPY:
checkCudaErrors(cudaFreeHost(hptrA));
checkCudaErrors(cudaFreeHost(hptrB));
checkCudaErrors(cudaFreeHost(hptrC));
break;
case USE_MANAGED_MEMORY:
case USE_MANAGED_MEMORY_WITH_HINTS:
case USE_MANAGED_MEMORY_WITH_HINTS_ASYNC:
checkCudaErrors(cudaFree(dptrA));
checkCudaErrors(cudaFree(dptrB));
checkCudaErrors(cudaFree(dptrC));
break;
default:
exit(EXIT_FAILURE); // exit due to error
}
checkCudaErrors(cudaStreamDestroy(streamToRunOn));
checkCudaErrors(cudaFreeHost(latch));
free(randValuesX);
free(randValuesY);
free(randValuesVerifyXmulY);
free(randValuesVerifyYmulX);
sdkDeleteTimer(&gpuLaunchCallsTimer);
sdkDeleteTimer(&gpuTransferCallsTimer);
sdkDeleteTimer(&gpuSyncTimer);
sdkDeleteTimer(&cpuAccessTimer);
}
void matrixMultiplyPerfRunner(bool reportAsBandwidth,
bool print_launch_transfer_results,
bool print_std_deviation, int device_id) {
int i;
unsigned int minMatrixDim = 32;
unsigned int multiplierDim = 2;
unsigned int matrixDim;
unsigned int minSize = minMatrixDim * minMatrixDim * sizeof(float);
unsigned int maxSize =
(maxSampleSizeInMb * ONE_MB) /
4; // 3 buffers are used, but dividing by 4 (power of 2)
unsigned int multiplier = multiplierDim * multiplierDim;
unsigned int numSizesToTest;
struct testResults *results;
struct resultsData *gpuLaunchCallsTimes;
struct resultsData *gpuTransferToCallsTimes;
struct resultsData *gpuTransferFromCallsTimes;
struct resultsData *gpuLaunchAndTransferCallsTimes;
struct resultsData *gpuLaunchTransferSyncTimes;
struct resultsData *cpuAccessTimes;
struct resultsData *overallTimes;
unsigned long *sizesToTest;
unsigned int j;
numSizesToTest = findNumSizesToTest(minSize, maxSize, multiplier);
createAndInitTestResults(&results, "matrixMultiplyPerf", numKernelRuns,
numSizesToTest);
sizesToTest = getPtrSizesToTest(results);
createResultDataAndAddToTestResults(&gpuLaunchCallsTimes, results,
"GPU Kernel Launch Call Time", false,
reportAsBandwidth);
createResultDataAndAddToTestResults(&gpuTransferToCallsTimes, results,
"CPU to GPU Transfer Calls Time", false,
reportAsBandwidth);
createResultDataAndAddToTestResults(&gpuTransferFromCallsTimes, results,
"GPU to CPU Transfer Calls Time", false,
reportAsBandwidth);
createResultDataAndAddToTestResults(&gpuLaunchAndTransferCallsTimes, results,
"GPU Launch and Transfer Calls Time",
false, reportAsBandwidth);
createResultDataAndAddToTestResults(&gpuLaunchTransferSyncTimes, results,
"GPU Launch Transfer and Sync Time",
false, reportAsBandwidth);
createResultDataAndAddToTestResults(
&cpuAccessTimes, results, "CPU Access Time", false, reportAsBandwidth);
createResultDataAndAddToTestResults(&overallTimes, results, "Overall Time",
false, reportAsBandwidth);
printf("Running ");
for (matrixDim = minMatrixDim, j = 0;
matrixDim * matrixDim <= maxSize / sizeof(float);
matrixDim *= multiplierDim, ++j) {
sizesToTest[j] = matrixDim * matrixDim * sizeof(float);
for (i = MEMALLOC_TYPE_START; i <= MEMALLOC_TYPE_END; i++) {
printf(".");
fflush(stdout);
runMatrixMultiplyKernel(
matrixDim, i, numKernelRuns,
getPtrRunTimesInMs(gpuLaunchCallsTimes, i, j),
getPtrRunTimesInMs(gpuTransferToCallsTimes, i, j),
getPtrRunTimesInMs(gpuTransferFromCallsTimes, i, j),
getPtrRunTimesInMs(gpuLaunchAndTransferCallsTimes, i, j),
getPtrRunTimesInMs(gpuLaunchTransferSyncTimes, i, j),
getPtrRunTimesInMs(cpuAccessTimes, i, j),
getPtrRunTimesInMs(overallTimes, i, j), device_id);
}
}
printf("\n");
printResults(results, print_launch_transfer_results, print_std_deviation);
freeTestResultsAndAllResultsData(results);
}
static void usage() {
printf(
"./cudaMemoryTypesPerf [-device=<device_id>] [-reportAsBandwidth] "
"[-print-launch-transfer-results] [-print-std-deviation] [-verbose]\n");
printf("Options:\n");
printf(
"-reportAsBandwidth: By default time taken is printed, this "
"option allows to instead print bandwidth.\n");
printf(
"-print-launch-transfer-results: By default overall results are printed, "
"this option allows to print data transfers and kernel time as well.\n");
printf(
"-print-std-deviation: Prints std deviation of the results.\n");
printf(
"-kernel-iterations=<num>: Number of times the kernel tests should "
"be run[default is 100 iterations].\n");
printf(
"-device=<device_id>: Allows to pass GPU Device ID on which "
"the tests will be run.\n");
printf("-verbose: Prints highly verbose output.\n");
}
int main(int argc, char **argv) {
bool reportAsBandwidth = false;
bool print_launch_transfer_results = false;
bool print_std_deviation = false;
if (checkCmdLineFlag(argc, (const char **)argv, "help") ||
checkCmdLineFlag(argc, (const char **)argv, "h")) {
usage();
printf("&&&& %s WAIVED\n", argv[0]);
exit(EXIT_WAIVED);
}
if (checkCmdLineFlag(argc, (const char **)argv, "reportAsBandwidth")) {
reportAsBandwidth = true;
}
if (checkCmdLineFlag(argc, (const char **)argv,
"print-launch-transfer-results")) {
print_launch_transfer_results = true;
}
if (checkCmdLineFlag(argc, (const char **)argv, "print-std-deviation")) {
print_std_deviation = true;
}
if (checkCmdLineFlag(argc, (const char **)argv, "kernel-iterations")) {
numKernelRuns =
getCmdLineArgumentInt(argc, (const char **)argv, "kernel-iterations");
}
if (checkCmdLineFlag(argc, (const char **)argv, "verbose")) {
verboseResults = 1;
}
int device_id = findCudaDevice(argc, (const char **)argv);
matrixMultiplyPerfRunner(reportAsBandwidth, print_launch_transfer_results,
print_std_deviation, device_id);
printf(
"\nNOTE: The CUDA Samples are not meant for performance measurements. "
"Results may vary when GPU Boost is enabled.\n");
exit(EXIT_SUCCESS);
}

View File

@@ -0,0 +1,304 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * 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.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
#
################################################################################
#
# Makefile project only supported on Mac OS X and Linux Platforms)
#
################################################################################
# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda
##############################
# start deprecated interface #
##############################
ifeq ($(x86_64),1)
$(info WARNING - x86_64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
TARGET_ARCH ?= x86_64
endif
ifeq ($(ARMv7),1)
$(info WARNING - ARMv7 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=armv7l instead)
TARGET_ARCH ?= armv7l
endif
ifeq ($(aarch64),1)
$(info WARNING - aarch64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
TARGET_ARCH ?= aarch64
endif
ifeq ($(ppc64le),1)
$(info WARNING - ppc64le variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
TARGET_ARCH ?= ppc64le
endif
ifneq ($(GCC),)
$(info WARNING - GCC variable has been deprecated)
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
HOST_COMPILER ?= $(GCC)
endif
ifneq ($(abi),)
$(error ERROR - abi variable has been removed)
endif
############################
# end deprecated interface #
############################
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
endif
else
TARGET_SIZE := $(shell getconf LONG_BIT)
endif
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
TARGET_ARCH = armv7l
endif
# operating system
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif
# host compiler
ifeq ($(TARGET_OS),darwin)
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
HOST_COMPILER ?= clang++
endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
ifeq ($(TARGET_OS),linux)
HOST_COMPILER ?= arm-linux-gnueabihf-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
else ifeq ($(TARGET_OS),android)
HOST_COMPILER ?= arm-linux-androideabi-g++
endif
else ifeq ($(TARGET_ARCH),aarch64)
ifeq ($(TARGET_OS), linux)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
endif
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
# internal flags
NVCCFLAGS := -m${TARGET_SIZE}
CCFLAGS :=
LDFLAGS :=
# build flags
ifeq ($(TARGET_OS),darwin)
LDFLAGS += -rpath $(CUDA_PATH)/lib
CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
LDFLAGS += -pie
CCFLAGS += -fpie -fpic -fexceptions
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif
# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif
# Debug build flags
ifeq ($(dbg),1)
NVCCFLAGS += -g -G
BUILD_TYPE := debug
else
BUILD_TYPE := release
endif
ALL_CCFLAGS :=
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
# Common includes and paths for CUDA
INCLUDES := -I../../Common
LIBRARIES :=
################################################################################
# Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 30 35 37 50 52 60 61 70 72 75
else
SMS ?= 30 35 37 50 52 60 61 70 75
endif
ifeq ($(SMS),)
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
HIGHEST_SM := $(lastword $(sort $(SMS)))
ifneq ($(HIGHEST_SM),)
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
endif
endif
ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif
################################################################################
# Target rules
all: build
build: bandwidthTest
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@echo "Sample will be waived due to the above missing dependencies"
else
@echo "Sample is ready - all dependencies have been met"
endif
bandwidthTest.o:bandwidthTest.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
bandwidthTest: bandwidthTest.o
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
$(EXEC) ./bandwidthTest
clean:
rm -f bandwidthTest bandwidthTest.o
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/bandwidthTest
clobber: clean

View File

@@ -0,0 +1,79 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
<entry>
<name>bandwidthTest</name>
<cuda_api_list>
<toolkit>cudaSetDevice</toolkit>
<toolkit>cudaHostAlloc</toolkit>
<toolkit>cudaFree</toolkit>
<toolkit>cudaMallocHost</toolkit>
<toolkit>cudaFreeHost</toolkit>
<toolkit>cudaMemcpy</toolkit>
<toolkit>cudaMemcpyAsync</toolkit>
<toolkit>cudaEventCreate</toolkit>
<toolkit>cudaEventRecord</toolkit>
<toolkit>cudaEventDestroy</toolkit>
<toolkit>cudaDeviceSynchronize</toolkit>
<toolkit>cudaEventElapsedTime</toolkit>
</cuda_api_list>
<description><![CDATA[This is a simple test program to measure the memcopy bandwidth of the GPU and memcpy bandwidth across PCI-e. This test application is capable of measuring device to device copy bandwidth, host to device copy bandwidth for pageable and page-locked memory, and device to host copy bandwidth for pageable and page-locked memory.]]></description>
<devicecompilation>whole</devicecompilation>
<includepaths>
<path>./</path>
<path>../</path>
<path>../../common/inc</path>
</includepaths>
<keyconcepts>
<concept level="basic">CUDA Streams and Events</concept>
<concept level="basic">Performance Strategies</concept>
</keyconcepts>
<keywords>
<keyword>GPGPU</keyword>
<keyword>bandwidth</keyword>
</keywords>
<libraries>
</libraries>
<librarypaths>
</librarypaths>
<nsight_eclipse>true</nsight_eclipse>
<primary_file>bandwidthTest.cu</primary_file>
<scopes>
<scope>1:CUDA Basic Topics</scope>
<scope>1:Performance Strategies</scope>
</scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch>
<sm-arch>sm52</sm-arch>
<sm-arch>sm60</sm-arch>
<sm-arch>sm61</sm-arch>
<sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch>
<supported_envs>
<env>
<arch>x86_64</arch>
<platform>linux</platform>
</env>
<env>
<platform>windows7</platform>
</env>
<env>
<arch>x86_64</arch>
<platform>macosx</platform>
</env>
<env>
<arch>arm</arch>
</env>
<env>
<arch>ppc64le</arch>
<platform>linux</platform>
</env>
</supported_envs>
<supported_sm_architectures>
<include>all</include>
</supported_sm_architectures>
<title>Bandwidth Test</title>
<type>exe</type>
</entry>

View File

@@ -0,0 +1,94 @@
# bandwidthTest - Bandwidth Test
## Description
This is a simple test program to measure the memcopy bandwidth of the GPU and memcpy bandwidth across PCI-e. This test application is capable of measuring device to device copy bandwidth, host to device copy bandwidth for pageable and page-locked memory, and device to host copy bandwidth for pageable and page-locked memory.
## Key Concepts
CUDA Streams and Events, Performance Strategies
## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
Linux, Windows, MacOSX
## Supported CPU Architecture
x86_64, ppc64le, armv7l
## CUDA APIs involved
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
cudaSetDevice, cudaHostAlloc, cudaFree, cudaMallocHost, cudaFreeHost, cudaMemcpy, cudaMemcpyAsync, cudaEventCreate, cudaEventRecord, cudaEventDestroy, cudaDeviceSynchronize, cudaEventElapsedTime
## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run
### Windows
The Windows samples are built using the Visual Studio IDE. Solution files (.sln) are provided for each supported version of Visual Studio, using the format:
```
*_vs<version>.sln - for Visual Studio <version>
```
Each individual sample has its own set of solution files in its directory:
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check DirectX Dependencies section for details."
### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, ppc64le, armv7l.
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=ppc64le` <br/> `$ make TARGET_ARCH=armv7l` <br/>
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
```
$ make SMS="50 60"
```
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=g++
```
### Mac
The Mac samples are built using makefiles. To use the makefiles, change directory into the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where "A B ..." is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use SMS="50 60".
```
$ make SMS="A B ..."
```
* **HOST_COMPILER=<host_compiler>** - override the default clang host compiler. See the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=clang
```
## References (for more details)

View File

@@ -0,0 +1,969 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
/*
* This is a simple test program to measure the memcopy bandwidth of the GPU.
* It can measure device to device copy bandwidth, host to device copy bandwidth
* for pageable and pinned memory, and device to host copy bandwidth for
* pageable and pinned memory.
*
* Usage:
* ./bandwidthTest [option]...
*/
// CUDA runtime
#include <cuda_runtime.h>
// includes
#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>
static const char *sSDKsample = "CUDA Bandwidth Test";
// defines, project
#define MEMCOPY_ITERATIONS 100
#define DEFAULT_SIZE (32 * (1e6)) // 32 M
#define DEFAULT_INCREMENT (4 * (1e6)) // 4 M
#define CACHE_CLEAR_SIZE (16 * (1e6)) // 16 M
// shmoo mode defines
#define SHMOO_MEMSIZE_MAX (64 * (1e6)) // 64 M
#define SHMOO_MEMSIZE_START (1e3) // 1 KB
#define SHMOO_INCREMENT_1KB (1e3) // 1 KB
#define SHMOO_INCREMENT_2KB (2 * 1e3) // 2 KB
#define SHMOO_INCREMENT_10KB (10 * (1e3)) // 10KB
#define SHMOO_INCREMENT_100KB (100 * (1e3)) // 100 KB
#define SHMOO_INCREMENT_1MB (1e6) // 1 MB
#define SHMOO_INCREMENT_2MB (2 * 1e6) // 2 MB
#define SHMOO_INCREMENT_4MB (4 * 1e6) // 4 MB
#define SHMOO_LIMIT_20KB (20 * (1e3)) // 20 KB
#define SHMOO_LIMIT_50KB (50 * (1e3)) // 50 KB
#define SHMOO_LIMIT_100KB (100 * (1e3)) // 100 KB
#define SHMOO_LIMIT_1MB (1e6) // 1 MB
#define SHMOO_LIMIT_16MB (16 * 1e6) // 16 MB
#define SHMOO_LIMIT_32MB (32 * 1e6) // 32 MB
// CPU cache flush
#define FLUSH_SIZE (256 * 1024 * 1024)
char *flush_buf;
// enums, project
enum testMode { QUICK_MODE, RANGE_MODE, SHMOO_MODE };
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",
"Device to Device", NULL};
const char *sMemoryMode[] = {"PINNED", "PAGEABLE", NULL};
// if true, use CPU based timing for everything
static bool bDontUseGPUTiming;
int *pArgc = NULL;
char **pArgv = NULL;
////////////////////////////////////////////////////////////////////////////////
// declaration, forward
int runTest(const int argc, const char **argv);
void testBandwidth(unsigned int start, unsigned int end, unsigned int increment,
testMode mode, memcpyKind kind, printMode printmode,
memoryMode memMode, int startDevice, int endDevice, bool wc);
void testBandwidthQuick(unsigned int size, memcpyKind kind, printMode printmode,
memoryMode memMode, int startDevice, int endDevice,
bool wc);
void testBandwidthRange(unsigned int start, unsigned int end,
unsigned int increment, memcpyKind kind,
printMode printmode, memoryMode memMode,
int startDevice, int endDevice, bool wc);
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,
unsigned int count, memcpyKind kind,
memoryMode memMode, 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) {
pArgc = &argc;
pArgv = argv;
flush_buf = (char *)malloc(FLUSH_SIZE);
// set logfile name and start logs
printf("[%s] - Starting...\n", sSDKsample);
int iRetVal = runTest(argc, (const char **)argv);
if (iRetVal < 0) {
checkCudaErrors(cudaSetDevice(0));
}
// finish
printf("%s\n", (iRetVal == 0) ? "Result = PASS" : "Result = FAIL");
printf(
"\nNOTE: The CUDA Samples are not meant for performance measurements. "
"Results may vary when GPU Boost is enabled.\n");
free(flush_buf);
exit((iRetVal == 0) ? EXIT_SUCCESS : EXIT_FAILURE);
}
///////////////////////////////////////////////////////////////////////////////
// Parse args, run the appropriate tests
///////////////////////////////////////////////////////////////////////////////
int runTest(const int argc, const char **argv) {
int start = DEFAULT_SIZE;
int end = DEFAULT_SIZE;
int startDevice = 0;
int endDevice = 0;
int increment = DEFAULT_INCREMENT;
testMode mode = QUICK_MODE;
bool htod = false;
bool dtoh = false;
bool dtod = false;
bool wc = false;
char *modeStr;
char *device = NULL;
printMode printmode = USER_READABLE;
char *memModeStr = NULL;
memoryMode memMode = PINNED;
// process command line args
if (checkCmdLineFlag(argc, argv, "help")) {
printHelp();
return 0;
}
if (checkCmdLineFlag(argc, argv, "csv")) {
printmode = CSV;
}
if (getCmdLineArgumentString(argc, argv, "memory", &memModeStr)) {
if (strcmp(memModeStr, "pageable") == 0) {
memMode = PAGEABLE;
} else if (strcmp(memModeStr, "pinned") == 0) {
memMode = PINNED;
} else {
printf("Invalid memory mode - valid modes are pageable or pinned\n");
printf("See --help for more information\n");
return -1000;
}
} else {
// default - pinned memory
memMode = PINNED;
}
if (getCmdLineArgumentString(argc, argv, "device", &device)) {
int deviceCount;
cudaError_t error_id = cudaGetDeviceCount(&deviceCount);
if (error_id != cudaSuccess) {
printf("cudaGetDeviceCount returned %d\n-> %s\n", (int)error_id,
cudaGetErrorString(error_id));
exit(EXIT_FAILURE);
}
if (deviceCount == 0) {
printf("!!!!!No devices found!!!!!\n");
return -2000;
}
if (strcmp(device, "all") == 0) {
printf(
"\n!!!!!Cumulative Bandwidth to be computed from all the devices "
"!!!!!!\n\n");
startDevice = 0;
endDevice = deviceCount - 1;
} else {
startDevice = endDevice = atoi(device);
if (startDevice >= deviceCount || startDevice < 0) {
printf(
"\n!!!!!Invalid GPU number %d given hence default gpu %d will be "
"used !!!!!\n",
startDevice, 0);
startDevice = endDevice = 0;
}
}
}
printf("Running on...\n\n");
for (int currentDevice = startDevice; currentDevice <= endDevice;
currentDevice++) {
cudaDeviceProp deviceProp;
cudaError_t error_id = cudaGetDeviceProperties(&deviceProp, currentDevice);
if (error_id == cudaSuccess) {
printf(" Device %d: %s\n", currentDevice, deviceProp.name);
if (deviceProp.computeMode == cudaComputeModeProhibited) {
fprintf(stderr,
"Error: device is running in <Compute Mode Prohibited>, no "
"threads can use ::cudaSetDevice().\n");
checkCudaErrors(cudaSetDevice(currentDevice));
exit(EXIT_FAILURE);
}
} else {
printf("cudaGetDeviceProperties returned %d\n-> %s\n", (int)error_id,
cudaGetErrorString(error_id));
checkCudaErrors(cudaSetDevice(currentDevice));
exit(EXIT_FAILURE);
}
}
if (getCmdLineArgumentString(argc, argv, "mode", &modeStr)) {
// figure out the mode
if (strcmp(modeStr, "quick") == 0) {
printf(" Quick Mode\n\n");
mode = QUICK_MODE;
} else if (strcmp(modeStr, "shmoo") == 0) {
printf(" Shmoo Mode\n\n");
mode = SHMOO_MODE;
} else if (strcmp(modeStr, "range") == 0) {
printf(" Range Mode\n\n");
mode = RANGE_MODE;
} else {
printf("Invalid mode - valid modes are quick, range, or shmoo\n");
printf("See --help for more information\n");
return -3000;
}
} else {
// default mode - quick
printf(" Quick Mode\n\n");
mode = QUICK_MODE;
}
if (checkCmdLineFlag(argc, argv, "htod")) {
htod = true;
}
if (checkCmdLineFlag(argc, argv, "dtoh")) {
dtoh = true;
}
if (checkCmdLineFlag(argc, argv, "dtod")) {
dtod = true;
}
#if CUDART_VERSION >= 2020
if (checkCmdLineFlag(argc, argv, "wc")) {
wc = true;
}
#endif
if (checkCmdLineFlag(argc, argv, "cputiming")) {
bDontUseGPUTiming = true;
}
if (!htod && !dtoh && !dtod) {
// default: All
htod = true;
dtoh = true;
dtod = true;
}
if (RANGE_MODE == mode) {
if (checkCmdLineFlag(argc, (const char **)argv, "start")) {
start = getCmdLineArgumentInt(argc, argv, "start");
if (start <= 0) {
printf("Illegal argument - start must be greater than zero\n");
return -4000;
}
} else {
printf("Must specify a starting size in range mode\n");
printf("See --help for more information\n");
return -5000;
}
if (checkCmdLineFlag(argc, (const char **)argv, "end")) {
end = getCmdLineArgumentInt(argc, argv, "end");
if (end <= 0) {
printf("Illegal argument - end must be greater than zero\n");
return -6000;
}
if (start > end) {
printf("Illegal argument - start is greater than end\n");
return -7000;
}
} else {
printf("Must specify an end size in range mode.\n");
printf("See --help for more information\n");
return -8000;
}
if (checkCmdLineFlag(argc, argv, "increment")) {
increment = getCmdLineArgumentInt(argc, argv, "increment");
if (increment <= 0) {
printf("Illegal argument - increment must be greater than zero\n");
return -9000;
}
} else {
printf("Must specify an increment in user mode\n");
printf("See --help for more information\n");
return -10000;
}
}
if (htod) {
testBandwidth((unsigned int)start, (unsigned int)end,
(unsigned int)increment, mode, HOST_TO_DEVICE, printmode,
memMode, startDevice, endDevice, wc);
}
if (dtoh) {
testBandwidth((unsigned int)start, (unsigned int)end,
(unsigned int)increment, mode, DEVICE_TO_HOST, printmode,
memMode, startDevice, endDevice, wc);
}
if (dtod) {
testBandwidth((unsigned int)start, (unsigned int)end,
(unsigned int)increment, mode, DEVICE_TO_DEVICE, printmode,
memMode, startDevice, endDevice, wc);
}
// Ensure that we reset all CUDA Devices in question
for (int nDevice = startDevice; nDevice <= endDevice; nDevice++) {
cudaSetDevice(nDevice);
}
return 0;
}
///////////////////////////////////////////////////////////////////////////////
// Run a bandwidth test
///////////////////////////////////////////////////////////////////////////////
void testBandwidth(unsigned int start, unsigned int end, unsigned int increment,
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,
endDevice, wc);
break;
case RANGE_MODE:
testBandwidthRange(start, end, increment, kind, printmode, memMode,
startDevice, endDevice, wc);
break;
case SHMOO_MODE:
testBandwidthShmoo(kind, printmode, memMode, startDevice, endDevice, wc);
break;
default:
break;
}
}
//////////////////////////////////////////////////////////////////////
// Run a quick mode bandwidth test
//////////////////////////////////////////////////////////////////////
void testBandwidthQuick(unsigned int size, memcpyKind kind, printMode printmode,
memoryMode memMode, 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,
unsigned int increment, 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);
unsigned int *memSizes = (unsigned int *)malloc(count * sizeof(unsigned int));
double *bandwidths = (double *)malloc(count * sizeof(double));
// Before calculating the cumulative bandwidth, initialize bandwidths array to
// NULL
for (unsigned int i = 0; i < count; i++) {
bandwidths[i] = 0.0;
}
// Use the device asked by the user
for (int currentDevice = startDevice; currentDevice <= endDevice;
currentDevice++) {
cudaSetDevice(currentDevice);
// run each of the copies
for (unsigned int i = 0; i < count; i++) {
memSizes[i] = start + i * increment;
switch (kind) {
case DEVICE_TO_HOST:
bandwidths[i] += testDeviceToHostTransfer(memSizes[i], memMode, wc);
break;
case HOST_TO_DEVICE:
bandwidths[i] += testHostToDeviceTransfer(memSizes[i], memMode, wc);
break;
case DEVICE_TO_DEVICE:
bandwidths[i] += testDeviceToDeviceTransfer(memSizes[i]);
break;
}
}
} // Complete the bandwidth computation on all the devices
// print results
if (printmode == CSV) {
printResultsCSV(memSizes, bandwidths, count, kind, memMode,
(1 + endDevice - startDevice), wc);
} else {
printResultsReadable(memSizes, bandwidths, count, kind, memMode,
(1 + endDevice - startDevice), wc);
}
// clean up
free(memSizes);
free(bandwidths);
}
//////////////////////////////////////////////////////////////////////////////
// Intense shmoo mode - covers a large range of values with varying increments
//////////////////////////////////////////////////////////////////////////////
void testBandwidthShmoo(memcpyKind kind, printMode printmode,
memoryMode memMode, int startDevice, int endDevice,
bool wc) {
// count the number of copies to make
unsigned int count =
1 + (SHMOO_LIMIT_20KB / SHMOO_INCREMENT_1KB) +
((SHMOO_LIMIT_50KB - SHMOO_LIMIT_20KB) / SHMOO_INCREMENT_2KB) +
((SHMOO_LIMIT_100KB - SHMOO_LIMIT_50KB) / SHMOO_INCREMENT_10KB) +
((SHMOO_LIMIT_1MB - SHMOO_LIMIT_100KB) / SHMOO_INCREMENT_100KB) +
((SHMOO_LIMIT_16MB - SHMOO_LIMIT_1MB) / SHMOO_INCREMENT_1MB) +
((SHMOO_LIMIT_32MB - SHMOO_LIMIT_16MB) / SHMOO_INCREMENT_2MB) +
((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));
// Before calculating the cumulative bandwidth, initialize bandwidths array to
// NULL
for (unsigned int i = 0; i < count; i++) {
bandwidths[i] = 0.0;
}
// Use the device asked by the user
for (int currentDevice = startDevice; currentDevice <= endDevice;
currentDevice++) {
cudaSetDevice(currentDevice);
// Run the shmoo
int iteration = 0;
unsigned int memSize = 0;
while (memSize <= SHMOO_MEMSIZE_MAX) {
if (memSize < SHMOO_LIMIT_20KB) {
memSize += SHMOO_INCREMENT_1KB;
} else if (memSize < SHMOO_LIMIT_50KB) {
memSize += SHMOO_INCREMENT_2KB;
} else if (memSize < SHMOO_LIMIT_100KB) {
memSize += SHMOO_INCREMENT_10KB;
} else if (memSize < SHMOO_LIMIT_1MB) {
memSize += SHMOO_INCREMENT_100KB;
} else if (memSize < SHMOO_LIMIT_16MB) {
memSize += SHMOO_INCREMENT_1MB;
} else if (memSize < SHMOO_LIMIT_32MB) {
memSize += SHMOO_INCREMENT_2MB;
} else {
memSize += SHMOO_INCREMENT_4MB;
}
memSizes[iteration] = memSize;
switch (kind) {
case DEVICE_TO_HOST:
bandwidths[iteration] +=
testDeviceToHostTransfer(memSizes[iteration], memMode, wc);
break;
case HOST_TO_DEVICE:
bandwidths[iteration] +=
testHostToDeviceTransfer(memSizes[iteration], memMode, wc);
break;
case DEVICE_TO_DEVICE:
bandwidths[iteration] +=
testDeviceToDeviceTransfer(memSizes[iteration]);
break;
}
iteration++;
printf(".");
fflush(0);
}
} // Complete the bandwidth computation on all the devices
// print results
printf("\n");
if (CSV == printmode) {
printResultsCSV(memSizes, bandwidths, count, kind, memMode,
(1 + endDevice - startDevice), wc);
} else {
printResultsReadable(memSizes, bandwidths, count, kind, memMode,
(1 + endDevice - startDevice), wc);
}
// clean up
free(memSizes);
free(bandwidths);
}
///////////////////////////////////////////////////////////////////////////////
// test the bandwidth of a device to host memcopy of a specific size
///////////////////////////////////////////////////////////////////////////////
float testDeviceToHostTransfer(unsigned int memSize, memoryMode memMode,
bool wc) {
StopWatchInterface *timer = NULL;
float elapsedTimeInMs = 0.0f;
float bandwidthInGBs = 0.0f;
unsigned char *h_idata = NULL;
unsigned char *h_odata = NULL;
cudaEvent_t start, stop;
sdkCreateTimer(&timer);
checkCudaErrors(cudaEventCreate(&start));
checkCudaErrors(cudaEventCreate(&stop));
// allocate host memory
if (PINNED == memMode) {
// pinned memory mode - use special function to get OS-pinned memory
#if CUDART_VERSION >= 2020
checkCudaErrors(cudaHostAlloc((void **)&h_idata, 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 {
// pageable memory mode - use malloc
h_idata = (unsigned char *)malloc(memSize);
h_odata = (unsigned char *)malloc(memSize);
if (h_idata == 0 || h_odata == 0) {
fprintf(stderr, "Not enough memory avaialable on host to run test!\n");
exit(EXIT_FAILURE);
}
}
// initialize the memory
for (unsigned int i = 0; i < memSize / sizeof(unsigned char); i++) {
h_idata[i] = (unsigned char)(i & 0xff);
}
// allocate device memory
unsigned char *d_idata;
checkCudaErrors(cudaMalloc((void **)&d_idata, memSize));
// initialize the device memory
checkCudaErrors(
cudaMemcpy(d_idata, h_idata, memSize, cudaMemcpyHostToDevice));
// copy data from GPU to Host
if (PINNED == memMode) {
if (bDontUseGPUTiming) sdkStartTimer(&timer);
checkCudaErrors(cudaEventRecord(start, 0));
for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
checkCudaErrors(cudaMemcpyAsync(h_odata, d_idata, memSize,
cudaMemcpyDeviceToHost, 0));
}
checkCudaErrors(cudaEventRecord(stop, 0));
checkCudaErrors(cudaDeviceSynchronize());
checkCudaErrors(cudaEventElapsedTime(&elapsedTimeInMs, start, stop));
if (bDontUseGPUTiming) {
sdkStopTimer(&timer);
elapsedTimeInMs = sdkGetTimerValue(&timer);
sdkResetTimer(&timer);
}
} else {
elapsedTimeInMs = 0;
for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
sdkStartTimer(&timer);
checkCudaErrors(
cudaMemcpy(h_odata, d_idata, memSize, cudaMemcpyDeviceToHost));
sdkStopTimer(&timer);
elapsedTimeInMs += sdkGetTimerValue(&timer);
sdkResetTimer(&timer);
memset(flush_buf, i, FLUSH_SIZE);
}
}
// calculate bandwidth in GB/s
double time_s = elapsedTimeInMs / 1e3;
bandwidthInGBs = (memSize * (float)MEMCOPY_ITERATIONS) / (double)1e9;
bandwidthInGBs = bandwidthInGBs / time_s;
// clean up memory
checkCudaErrors(cudaEventDestroy(stop));
checkCudaErrors(cudaEventDestroy(start));
sdkDeleteTimer(&timer);
if (PINNED == memMode) {
checkCudaErrors(cudaFreeHost(h_idata));
checkCudaErrors(cudaFreeHost(h_odata));
} else {
free(h_idata);
free(h_odata);
}
checkCudaErrors(cudaFree(d_idata));
return bandwidthInGBs;
}
///////////////////////////////////////////////////////////////////////////////
//! test the bandwidth of a host to device memcopy of a specific size
///////////////////////////////////////////////////////////////////////////////
float testHostToDeviceTransfer(unsigned int memSize, memoryMode memMode,
bool wc) {
StopWatchInterface *timer = NULL;
float elapsedTimeInMs = 0.0f;
float bandwidthInGBs = 0.0f;
cudaEvent_t start, stop;
sdkCreateTimer(&timer);
checkCudaErrors(cudaEventCreate(&start));
checkCudaErrors(cudaEventCreate(&stop));
// allocate host memory
unsigned char *h_odata = NULL;
if (PINNED == memMode) {
#if CUDART_VERSION >= 2020
// pinned memory mode - use special function to get OS-pinned memory
checkCudaErrors(cudaHostAlloc((void **)&h_odata, memSize,
(wc) ? cudaHostAllocWriteCombined : 0));
#else
// pinned memory mode - use special function to get OS-pinned memory
checkCudaErrors(cudaMallocHost((void **)&h_odata, memSize));
#endif
} else {
// pageable memory mode - use malloc
h_odata = (unsigned char *)malloc(memSize);
if (h_odata == 0) {
fprintf(stderr, "Not enough memory available on host to run test!\n");
exit(EXIT_FAILURE);
}
}
unsigned char *h_cacheClear1 = (unsigned char *)malloc(CACHE_CLEAR_SIZE);
unsigned char *h_cacheClear2 = (unsigned char *)malloc(CACHE_CLEAR_SIZE);
if (h_cacheClear1 == 0 || h_cacheClear2 == 0) {
fprintf(stderr, "Not enough memory available on host to run test!\n");
exit(EXIT_FAILURE);
}
// initialize the memory
for (unsigned int i = 0; i < memSize / sizeof(unsigned char); i++) {
h_odata[i] = (unsigned char)(i & 0xff);
}
for (unsigned int i = 0; i < CACHE_CLEAR_SIZE / sizeof(unsigned char); i++) {
h_cacheClear1[i] = (unsigned char)(i & 0xff);
h_cacheClear2[i] = (unsigned char)(0xff - (i & 0xff));
}
// allocate device memory
unsigned char *d_idata;
checkCudaErrors(cudaMalloc((void **)&d_idata, memSize));
// copy host memory to device memory
if (PINNED == memMode) {
if (bDontUseGPUTiming) sdkStartTimer(&timer);
checkCudaErrors(cudaEventRecord(start, 0));
for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
checkCudaErrors(cudaMemcpyAsync(d_idata, h_odata, memSize,
cudaMemcpyHostToDevice, 0));
}
checkCudaErrors(cudaEventRecord(stop, 0));
checkCudaErrors(cudaDeviceSynchronize());
checkCudaErrors(cudaEventElapsedTime(&elapsedTimeInMs, start, stop));
if (bDontUseGPUTiming) {
sdkStopTimer(&timer);
elapsedTimeInMs = sdkGetTimerValue(&timer);
sdkResetTimer(&timer);
}
} else {
elapsedTimeInMs = 0;
for (unsigned int i = 0; i < MEMCOPY_ITERATIONS; i++) {
sdkStartTimer(&timer);
checkCudaErrors(
cudaMemcpy(d_idata, h_odata, memSize, cudaMemcpyHostToDevice));
sdkStopTimer(&timer);
elapsedTimeInMs += sdkGetTimerValue(&timer);
sdkResetTimer(&timer);
memset(flush_buf, i, FLUSH_SIZE);
}
}
// calculate bandwidth in GB/s
double time_s = elapsedTimeInMs / 1e3;
bandwidthInGBs = (memSize * (float)MEMCOPY_ITERATIONS) / (double)1e9;
bandwidthInGBs = bandwidthInGBs / time_s;
// clean up memory
checkCudaErrors(cudaEventDestroy(stop));
checkCudaErrors(cudaEventDestroy(start));
sdkDeleteTimer(&timer);
if (PINNED == memMode) {
checkCudaErrors(cudaFreeHost(h_odata));
} else {
free(h_odata);
}
free(h_cacheClear1);
free(h_cacheClear2);
checkCudaErrors(cudaFree(d_idata));
return bandwidthInGBs;
}
///////////////////////////////////////////////////////////////////////////////
//! test the bandwidth of a device to device memcopy of a specific size
///////////////////////////////////////////////////////////////////////////////
float testDeviceToDeviceTransfer(unsigned int memSize) {
StopWatchInterface *timer = NULL;
float elapsedTimeInMs = 0.0f;
float bandwidthInGBs = 0.0f;
cudaEvent_t start, stop;
sdkCreateTimer(&timer);
checkCudaErrors(cudaEventCreate(&start));
checkCudaErrors(cudaEventCreate(&stop));
// allocate host memory
unsigned char *h_idata = (unsigned char *)malloc(memSize);
if (h_idata == 0) {
fprintf(stderr, "Not enough memory avaialable on host to run test!\n");
exit(EXIT_FAILURE);
}
// initialize the host memory
for (unsigned int i = 0; i < memSize / sizeof(unsigned char); i++) {
h_idata[i] = (unsigned char)(i & 0xff);
}
// allocate device memory
unsigned char *d_idata;
checkCudaErrors(cudaMalloc((void **)&d_idata, memSize));
unsigned char *d_odata;
checkCudaErrors(cudaMalloc((void **)&d_odata, memSize));
// initialize memory
checkCudaErrors(
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(
cudaMemcpy(d_odata, d_idata, memSize, cudaMemcpyDeviceToDevice));
}
checkCudaErrors(cudaEventRecord(stop, 0));
// Since device to device memory copies are non-blocking,
// cudaDeviceSynchronize() is required in order to get
// proper timing.
checkCudaErrors(cudaDeviceSynchronize());
// get the total elapsed time in ms
sdkStopTimer(&timer);
checkCudaErrors(cudaEventElapsedTime(&elapsedTimeInMs, start, stop));
if (bDontUseGPUTiming) {
elapsedTimeInMs = sdkGetTimerValue(&timer);
}
// calculate bandwidth in GB/s
double time_s = elapsedTimeInMs / 1e3;
bandwidthInGBs = (2.0f * memSize * (float)MEMCOPY_ITERATIONS) / (double)1e9;
bandwidthInGBs = bandwidthInGBs / time_s;
// clean up memory
sdkDeleteTimer(&timer);
free(h_idata);
checkCudaErrors(cudaEventDestroy(stop));
checkCudaErrors(cudaEventDestroy(start));
checkCudaErrors(cudaFree(d_idata));
checkCudaErrors(cudaFree(d_odata));
return bandwidthInGBs;
}
/////////////////////////////////////////////////////////
// print results in an easily read format
////////////////////////////////////////////////////////
void printResultsReadable(unsigned int *memSizes, double *bandwidths,
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]);
if (wc) {
printf(" Write-Combined Memory Writes are Enabled");
}
printf(" Transfer Size (Bytes)\tBandwidth(GB/s)\n");
unsigned int i;
for (i = 0; i < (count - 1); i++) {
printf(" %u\t\t\t%s%.1f\n", memSizes[i],
(memSizes[i] < 10000) ? "\t" : "", bandwidths[i]);
}
printf(" %u\t\t\t%s%.1f\n\n", memSizes[i],
(memSizes[i] < 10000) ? "\t" : "", bandwidths[i]);
}
///////////////////////////////////////////////////////////////////////////
// print results in a database format
///////////////////////////////////////////////////////////////////////////
void printResultsCSV(unsigned int *memSizes, double *bandwidths,
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 {
if (kind == DEVICE_TO_HOST) {
sConfig += "D2H";
} else if (kind == HOST_TO_DEVICE) {
sConfig += "H2D";
}
if (memMode == PAGEABLE) {
sConfig += "-Paged";
} else if (memMode == PINNED) {
sConfig += "-Pinned";
if (wc) {
sConfig += "-WriteCombined";
}
}
}
unsigned int i;
double dSeconds = 0.0;
for (i = 0; i < count; i++) {
dSeconds = (double)memSizes[i] / (bandwidths[i] * (double)(1 << 20));
printf(
"bandwidthTest-%s, Bandwidth = %.1f GB/s, Time = %.5f s, Size = %u "
"bytes, NumDevsUsed = %d\n",
sConfig.c_str(), bandwidths[i], dSeconds, memSizes[i], iNumDevs);
}
}
///////////////////////////////////////////////////////////////////////////
// Print help screen
///////////////////////////////////////////////////////////////////////////
void printHelp(void) {
printf("Usage: bandwidthTest [OPTION]...\n");
printf(
"Test the bandwidth for device to host, host to device, and device to "
"device transfers\n");
printf("\n");
printf(
"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(
"./bandwidthTest --memory=pinned --mode=range --start=1024 --end=102400 "
"--increment=1024 --dtoh\n");
printf("\n");
printf("Options:\n");
printf("--help\tDisplay this help menu\n");
printf("--csv\tPrint results as a CSV\n");
printf("--device=[deviceno]\tSpecify the device device to be used\n");
printf(" all - compute cumulative bandwidth on all the devices\n");
printf(" 0,1,2,...,n - Specify any particular device to be used\n");
printf("--memory=[MEMMODE]\tSpecify which memory mode to use\n");
printf(" pageable - pageable memory\n");
printf(" pinned - non-pageable system memory\n");
printf("--mode=[MODE]\tSpecify the mode to use\n");
printf(" quick - performs a quick measurement\n");
printf(" range - measures a user-specified range of values\n");
printf(" shmoo - performs an intense shmoo of a large range of values\n");
printf("--htod\tMeasure host to device transfers\n");
printf("--dtoh\tMeasure device to host transfers\n");
printf("--dtod\tMeasure device to device transfers\n");
#if CUDART_VERSION >= 2020
printf("--wc\tAllocate pinned memory as write-combined\n");
#endif
printf("--cputiming\tForce CPU-based timing always\n");
printf("Range mode options\n");
printf("--start=[SIZE]\tStarting transfer size in bytes\n");
printf("--end=[SIZE]\tEnding transfer size in bytes\n");
printf("--increment=[SIZE]\tIncrement size in bytes\n");
}

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2012
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "bandwidthTest", "bandwidthTest_vs2012.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>bandwidthTest_vs2012</RootNamespace>
<ProjectName>bandwidthTest</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v110</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bandwidthTest.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 13.00
# Visual Studio 2013
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "bandwidthTest", "bandwidthTest_vs2013.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>bandwidthTest_vs2013</RootNamespace>
<ProjectName>bandwidthTest</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v120</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bandwidthTest.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 14.00
# Visual Studio 2015
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "bandwidthTest", "bandwidthTest_vs2015.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>bandwidthTest_vs2015</RootNamespace>
<ProjectName>bandwidthTest</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v140</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bandwidthTest.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2017
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "bandwidthTest", "bandwidthTest_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,112 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>bandwidthTest_vs2017</RootNamespace>
<ProjectName>bandwidthTest</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
<LatestTargetPlatformVersion>$([Microsoft.Build.Utilities.ToolLocationHelper]::GetLatestSDKTargetPlatformVersion('Windows', '10.0'))</LatestTargetPlatformVersion>
<WindowsTargetPlatformVersion Condition="'$(WindowsTargetPlatformVersion)' == ''">$(LatestTargetPlatformVersion)</WindowsTargetPlatformVersion>
<TargetPlatformVersion>$(WindowsTargetPlatformVersion)</TargetPlatformVersion>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v141</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bandwidthTest.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2019
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "bandwidthTest", "bandwidthTest_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,108 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>bandwidthTest_vs2019</RootNamespace>
<ProjectName>bandwidthTest</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v142</PlatformToolset>
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bandwidthTest.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,306 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * 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.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
#
################################################################################
#
# Makefile project only supported on Mac OS X and Linux Platforms)
#
################################################################################
# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda
##############################
# start deprecated interface #
##############################
ifeq ($(x86_64),1)
$(info WARNING - x86_64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
TARGET_ARCH ?= x86_64
endif
ifeq ($(ARMv7),1)
$(info WARNING - ARMv7 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=armv7l instead)
TARGET_ARCH ?= armv7l
endif
ifeq ($(aarch64),1)
$(info WARNING - aarch64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
TARGET_ARCH ?= aarch64
endif
ifeq ($(ppc64le),1)
$(info WARNING - ppc64le variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
TARGET_ARCH ?= ppc64le
endif
ifneq ($(GCC),)
$(info WARNING - GCC variable has been deprecated)
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
HOST_COMPILER ?= $(GCC)
endif
ifneq ($(abi),)
$(error ERROR - abi variable has been removed)
endif
############################
# end deprecated interface #
############################
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
endif
else
TARGET_SIZE := $(shell getconf LONG_BIT)
endif
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
TARGET_ARCH = armv7l
endif
# operating system
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif
# host compiler
ifeq ($(TARGET_OS),darwin)
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
HOST_COMPILER ?= clang++
endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
ifeq ($(TARGET_OS),linux)
HOST_COMPILER ?= arm-linux-gnueabihf-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
else ifeq ($(TARGET_OS),android)
HOST_COMPILER ?= arm-linux-androideabi-g++
endif
else ifeq ($(TARGET_ARCH),aarch64)
ifeq ($(TARGET_OS), linux)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
endif
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
# internal flags
NVCCFLAGS := -m${TARGET_SIZE}
CCFLAGS :=
LDFLAGS :=
# build flags
ifeq ($(TARGET_OS),darwin)
LDFLAGS += -rpath $(CUDA_PATH)/lib
CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
LDFLAGS += -pie
CCFLAGS += -fpie -fpic -fexceptions
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif
# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif
# Debug build flags
ifeq ($(dbg),1)
NVCCFLAGS += -g -G
BUILD_TYPE := debug
else
BUILD_TYPE := release
endif
ALL_CCFLAGS :=
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
# Common includes and paths for CUDA
INCLUDES := -I../../Common
LIBRARIES :=
################################################################################
# Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 30 35 37 50 52 60 61 70 72 75
else
SMS ?= 30 35 37 50 52 60 61 70 75
endif
ifeq ($(SMS),)
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
HIGHEST_SM := $(lastword $(sort $(SMS)))
ifneq ($(HIGHEST_SM),)
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
endif
endif
LIBRARIES += -lcublas_static -lcublasLt_static -lcusparse_static -lculibos
ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif
################################################################################
# Target rules
all: build
build: conjugateGradientCudaGraphs
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@echo "Sample will be waived due to the above missing dependencies"
else
@echo "Sample is ready - all dependencies have been met"
endif
conjugateGradientCudaGraphs.o:conjugateGradientCudaGraphs.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
conjugateGradientCudaGraphs: conjugateGradientCudaGraphs.o
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
$(EXEC) ./conjugateGradientCudaGraphs
clean:
rm -f conjugateGradientCudaGraphs conjugateGradientCudaGraphs.o
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/conjugateGradientCudaGraphs
clobber: clean

View File

@@ -0,0 +1,86 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
<entry>
<name>conjugateGradientCudaGraphs</name>
<cuda_api_list>
<toolkit>cudaStreamBeginCapture</toolkit>
<toolkit>cudaStreamEndCapture</toolkit>
<toolkit>cudaGraphCreate</toolkit>
<toolkit>cudaGraphLaunch</toolkit>
<toolkit>cudaGraphInstantiate</toolkit>
<toolkit>cudaGraphExecDestroy</toolkit>
<toolkit>cudaGraphDestroy</toolkit>
</cuda_api_list>
<description><![CDATA[This sample implements a conjugate gradient solver on GPU using CUBLAS and CUSPARSE library calls captured and called using CUDA Graph APIs.]]></description>
<devicecompilation>whole</devicecompilation>
<includepaths>
<path>./</path>
<path>../</path>
<path>../../common/inc</path>
</includepaths>
<keyconcepts>
<concept level="advanced">Linear Algebra</concept>
<concept level="advanced">CUBLAS Library</concept>
<concept level="advanced">CUSPARSE Library</concept>
</keyconcepts>
<keywords>
<keyword>CUDA</keyword>
<keyword>CUBLAS</keyword>
<keyword>CUSPARSE</keyword>
<keyword>Sparse Matrix</keyword>
</keywords>
<libraries>
<library>cublas_static</library>
<library>cublasLt_static</library>
<library>cusparse_static</library>
<library>culibos</library>
</libraries>
<librarypaths>
</librarypaths>
<nsight_eclipse>true</nsight_eclipse>
<primary_file>conjugateGradientCudaGraphs.cu</primary_file>
<required_dependencies>
<dependency>CUBLAS</dependency>
<dependency>CUSPARSE</dependency>
</required_dependencies>
<scopes>
<scope>1:CUDA Advanced Topics</scope>
<scope>3:Linear Algebra</scope>
<scope>1:CUDA Graphs</scope>
</scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch>
<sm-arch>sm52</sm-arch>
<sm-arch>sm60</sm-arch>
<sm-arch>sm61</sm-arch>
<sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch>
<supported_envs>
<env>
<arch>x86_64</arch>
<platform>linux</platform>
</env>
<env>
<platform>windows7</platform>
</env>
<env>
<arch>x86_64</arch>
<platform>macosx</platform>
</env>
<env>
<arch>arm</arch>
</env>
<env>
<arch>ppc64le</arch>
<platform>linux</platform>
</env>
</supported_envs>
<supported_sm_architectures>
<include>all</include>
</supported_sm_architectures>
<title>Conjugate Gradient using Cuda Graphs</title>
<type>exe</type>
</entry>

View File

@@ -0,0 +1,98 @@
# conjugateGradientCudaGraphs - Conjugate Gradient using Cuda Graphs
## Description
This sample implements a conjugate gradient solver on GPU using CUBLAS and CUSPARSE library calls captured and called using CUDA Graph APIs.
## Key Concepts
Linear Algebra, CUBLAS Library, CUSPARSE Library
## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
Linux, Windows, MacOSX
## Supported CPU Architecture
x86_64, ppc64le, armv7l
## CUDA APIs involved
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
cudaStreamBeginCapture, cudaStreamEndCapture, cudaGraphCreate, cudaGraphLaunch, cudaGraphInstantiate, cudaGraphExecDestroy, cudaGraphDestroy
## Dependencies needed to build/run
[CUBLAS](../../README.md#cublas), [CUSPARSE](../../README.md#cusparse)
## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run
### Windows
The Windows samples are built using the Visual Studio IDE. Solution files (.sln) are provided for each supported version of Visual Studio, using the format:
```
*_vs<version>.sln - for Visual Studio <version>
```
Each individual sample has its own set of solution files in its directory:
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check DirectX Dependencies section for details."
### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, ppc64le, armv7l.
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=ppc64le` <br/> `$ make TARGET_ARCH=armv7l` <br/>
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
```
$ make SMS="50 60"
```
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=g++
```
### Mac
The Mac samples are built using makefiles. To use the makefiles, change directory into the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where "A B ..." is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use SMS="50 60".
```
$ make SMS="A B ..."
```
* **HOST_COMPILER=<host_compiler>** - override the default clang host compiler. See the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=clang
```
## References (for more details)

View File

@@ -0,0 +1,466 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * 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.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
*/
/*
* This sample implements a conjugate gradient solver on GPU
* using CUBLAS and CUSPARSE with CUDA Graphs
*
*/
// includes, system
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
/* Using updated (v2) interfaces to cublas */
#include <cublas_v2.h>
#include <cuda_runtime.h>
#include <cusparse.h>
#include <cooperative_groups.h>
// Utilities and system includes
#include <helper_cuda.h> // helper function CUDA error checking and initialization
#include <helper_functions.h> // helper for shared functions common to CUDA Samples
namespace cg = cooperative_groups;
const char *sSDKname = "conjugateGradientCudaGraphs";
#ifndef WITH_GRAPH
#define WITH_GRAPH 1
#endif
/* genTridiag: generate a random tridiagonal symmetric matrix */
void genTridiag(int *I, int *J, float *val, int N, int nz) {
I[0] = 0, J[0] = 0, J[1] = 1;
val[0] = (float)rand() / RAND_MAX + 10.0f;
val[1] = (float)rand() / RAND_MAX;
int start;
for (int i = 1; i < N; i++) {
if (i > 1) {
I[i] = I[i - 1] + 3;
} else {
I[1] = 2;
}
start = (i - 1) * 3 + 2;
J[start] = i - 1;
J[start + 1] = i;
if (i < N - 1) {
J[start + 2] = i + 1;
}
val[start] = val[start - 1];
val[start + 1] = (float)rand() / RAND_MAX + 10.0f;
if (i < N - 1) {
val[start + 2] = (float)rand() / RAND_MAX;
}
}
I[N] = nz;
}
__global__ void initVectors(float *rhs, float *x, int N) {
size_t gid = blockIdx.x * blockDim.x + threadIdx.x;
for (size_t i = gid; i < N; i += gridDim.x * blockDim.x) {
rhs[i] = 1.0;
x[i] = 0.0;
}
}
__global__ void gpuDotProduct(float *vecA, float *vecB, float *result,
int size) {
cg::thread_block cta = cg::this_thread_block();
int gid = blockIdx.x * blockDim.x + threadIdx.x;
extern __shared__ double tmp[];
double temp_sum = 0.0;
for (int i = gid; i < size; i += gridDim.x * blockDim.x) {
temp_sum += (double)(vecA[i] * vecB[i]);
}
tmp[cta.thread_rank()] = temp_sum;
cg::sync(cta);
cg::thread_block_tile<32> tile32 = cg::tiled_partition<32>(cta);
double beta = temp_sum;
double temp;
for (int i = tile32.size() / 2; i > 0; i >>= 1) {
if (tile32.thread_rank() < i) {
temp = tmp[cta.thread_rank() + i];
beta += temp;
tmp[cta.thread_rank()] = beta;
}
cg::sync(tile32);
}
cg::sync(cta);
if (cta.thread_rank() == 0) {
beta = 0.0;
for (int i = 0; i < cta.size(); i += tile32.size()) {
beta += tmp[i];
}
atomicAdd(result, (float)beta);
}
}
__global__ void gpuSpMV(int *I, int *J, float *val, int nnz, int num_rows,
float alpha, float *inputVecX, float *outputVecY) {
size_t gid = blockIdx.x * blockDim.x + threadIdx.x;
for (size_t i = gid; i < num_rows; i += blockDim.x * gridDim.x) {
int row_elem = I[i];
int next_row_elem = I[i + 1];
int num_elems_this_row = next_row_elem - row_elem;
float output = 0.0;
for (int j = 0; j < num_elems_this_row; j++) {
output += alpha * val[row_elem + j] * inputVecX[J[row_elem + j]];
}
outputVecY[i] = output;
}
}
__global__ void r1_div_x(float *r1, float *r0, float *b) {
int gid = blockIdx.x * blockDim.x + threadIdx.x;
if (gid == 0) {
b[0] = r1[0] / r0[0];
}
}
__global__ void a_minus(float *a, float *na) {
int gid = blockIdx.x * blockDim.x + threadIdx.x;
if (gid == 0) {
na[0] = -(a[0]);
}
}
int main(int argc, char **argv) {
int N = 0, nz = 0, *I = NULL, *J = NULL;
float *val = NULL;
const float tol = 1e-5f;
const int max_iter = 10000;
float *x;
float *rhs;
float r1;
int *d_col, *d_row;
float *d_val, *d_x;
float *d_r, *d_p, *d_Ax;
int k;
float alpha, beta, alpham1;
cudaStream_t stream1, streamForGraph;
// This will pick the best possible CUDA capable device
cudaDeviceProp deviceProp;
int devID = findCudaDevice(argc, (const char **)argv);
if (devID < 0) {
printf("exiting...\n");
exit(EXIT_SUCCESS);
}
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, devID));
// Statistics about the GPU device
printf(
"> GPU device has %d Multi-Processors, SM %d.%d compute capabilities\n\n",
deviceProp.multiProcessorCount, deviceProp.major, deviceProp.minor);
/* Generate a random tridiagonal symmetric matrix in CSR format */
N = 1048576;
nz = (N - 2) * 3 + 4;
I = (int *)malloc(sizeof(int) * (N + 1));
J = (int *)malloc(sizeof(int) * nz);
val = (float *)malloc(sizeof(float) * nz);
genTridiag(I, J, val, N, nz);
x = (float *)malloc(sizeof(float) * N);
rhs = (float *)malloc(sizeof(float) * N);
for (int i = 0; i < N; i++) {
rhs[i] = 1.0;
x[i] = 0.0;
}
/* Get handle to the CUBLAS context */
cublasHandle_t cublasHandle = 0;
cublasStatus_t cublasStatus;
cublasStatus = cublasCreate(&cublasHandle);
checkCudaErrors(cublasStatus);
/* Get handle to the CUSPARSE context */
cusparseHandle_t cusparseHandle = 0;
cusparseStatus_t cusparseStatus;
cusparseStatus = cusparseCreate(&cusparseHandle);
checkCudaErrors(cusparseStatus);
checkCudaErrors(cudaStreamCreate(&stream1));
checkCudaErrors(cudaMalloc((void **)&d_col, nz * sizeof(int)));
checkCudaErrors(cudaMalloc((void **)&d_row, (N + 1) * sizeof(int)));
checkCudaErrors(cudaMalloc((void **)&d_val, nz * sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_x, N * sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_r, N * sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_p, N * sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_Ax, N * sizeof(float)));
float *d_r1, *d_r0, *d_dot, *d_a, *d_na, *d_b;
checkCudaErrors(cudaMalloc((void **)&d_r1, sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_r0, sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_dot, sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_a, sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_na, sizeof(float)));
checkCudaErrors(cudaMalloc((void **)&d_b, sizeof(float)));
cusparseMatDescr_t descr = 0;
checkCudaErrors(cusparseCreateMatDescr(&descr));
checkCudaErrors(cusparseSetMatType(descr, CUSPARSE_MATRIX_TYPE_GENERAL));
checkCudaErrors(cusparseSetMatIndexBase(descr, CUSPARSE_INDEX_BASE_ZERO));
int numBlocks = 0, blockSize = 0, numBlocks2 = 0, blockSize2 = 0;
checkCudaErrors(
cudaOccupancyMaxPotentialBlockSize(&numBlocks, &blockSize, initVectors));
checkCudaErrors(cudaMemcpyAsync(d_col, J, nz * sizeof(int),
cudaMemcpyHostToDevice, stream1));
checkCudaErrors(cudaMemcpyAsync(d_row, I, (N + 1) * sizeof(int),
cudaMemcpyHostToDevice, stream1));
checkCudaErrors(cudaMemcpyAsync(d_val, val, nz * sizeof(float),
cudaMemcpyHostToDevice, stream1));
initVectors<<<numBlocks, blockSize, 0, stream1>>>(d_r, d_x, N);
checkCudaErrors(cudaOccupancyMaxPotentialBlockSize(&numBlocks2, &blockSize2,
gpuSpMV));
checkCudaErrors(cudaOccupancyMaxPotentialBlockSize(&numBlocks, &blockSize,
gpuDotProduct));
alpha = 1.0;
alpham1 = -1.0;
beta = 0.0;
checkCudaErrors(cusparseSetStream(cusparseHandle, stream1));
checkCudaErrors(
cusparseScsrmv(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, N, N, nz,
&alpha, descr, d_val, d_row, d_col, d_x, &beta, d_Ax));
checkCudaErrors(cublasSetStream(cublasHandle, stream1));
checkCudaErrors(cublasSaxpy(cublasHandle, N, &alpham1, d_Ax, 1, d_r, 1));
checkCudaErrors(
cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_DEVICE));
checkCudaErrors(cublasSdot(cublasHandle, N, d_r, 1, d_r, 1, d_r1));
k = 1;
// First Iteration when k=1 starts
checkCudaErrors(cublasScopy(cublasHandle, N, d_r, 1, d_p, 1));
checkCudaErrors(
cusparseScsrmv(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, N, N, nz,
&alpha, descr, d_val, d_row, d_col, d_p, &beta, d_Ax));
checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));
r1_div_x<<<1, 1, 0, stream1>>>(d_r1, d_dot, d_a);
checkCudaErrors(cublasSaxpy(cublasHandle, N, d_a, d_p, 1, d_x, 1));
a_minus<<<1, 1, 0, stream1>>>(d_a, d_na);
checkCudaErrors(cublasSaxpy(cublasHandle, N, d_na, d_Ax, 1, d_r, 1));
checkCudaErrors(cudaMemcpyAsync(d_r0, d_r1, sizeof(float),
cudaMemcpyDeviceToDevice, stream1));
checkCudaErrors(cublasSdot(cublasHandle, N, d_r, 1, d_r, 1, d_r1));
checkCudaErrors(cudaMemcpyAsync(&r1, d_r1, sizeof(float),
cudaMemcpyDeviceToHost, stream1));
checkCudaErrors(cudaStreamSynchronize(stream1));
printf("iteration = %3d, residual = %e\n", k, sqrt(r1));
// First Iteration when k=1 ends
k++;
#if WITH_GRAPH
cudaGraph_t initGraph;
checkCudaErrors(cudaStreamCreate(&streamForGraph));
checkCudaErrors(cublasSetStream(cublasHandle, stream1));
checkCudaErrors(cusparseSetStream(cusparseHandle, stream1));
checkCudaErrors(cudaStreamBeginCapture(stream1, cudaStreamCaptureModeGlobal));
r1_div_x<<<1, 1, 0, stream1>>>(d_r1, d_r0, d_b);
cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_DEVICE);
checkCudaErrors(cublasSscal(cublasHandle, N, d_b, d_p, 1));
cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_HOST);
checkCudaErrors(cublasSaxpy(cublasHandle, N, &alpha, d_r, 1, d_p, 1));
cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_DEVICE);
#if 0 // Use cusparseScsrmv API when it is cuda graph compliant
checkCudaErrors(
cusparseSetPointerMode(cusparseHandle, CUSPARSE_POINTER_MODE_HOST));
checkCudaErrors(
cusparseScsrmv(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, N, N, nz,
&alpha, descr, d_val, d_row, d_col, d_p, &beta, d_Ax));
#else
gpuSpMV<<<numBlocks2, blockSize2, 0, stream1>>>(d_row, d_col, d_val, nz,
N, alpha, d_p, d_Ax);
#endif
checkCudaErrors(cudaMemsetAsync(d_dot, 0, sizeof(float), stream1));
// Use cublasSdot API when it is cuda graph compliant.
// checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));
gpuDotProduct<<<numBlocks, blockSize, blockSize * sizeof(double), stream1>>>(
d_p, d_Ax, d_dot, N);
r1_div_x<<<1, 1, 0, stream1>>>(d_r1, d_dot, d_a);
checkCudaErrors(cublasSaxpy(cublasHandle, N, d_a, d_p, 1, d_x, 1));
a_minus<<<1, 1, 0, stream1>>>(d_a, d_na);
checkCudaErrors(cublasSaxpy(cublasHandle, N, d_na, d_Ax, 1, d_r, 1));
checkCudaErrors(cudaMemcpyAsync(d_r0, d_r1, sizeof(float),
cudaMemcpyDeviceToDevice, stream1));
checkCudaErrors(cudaMemsetAsync(d_r1, 0, sizeof(float), stream1));
// Use cublasSdot API when it is cuda graph compliant.
// checkCudaErrors(cublasSdot(cublasHandle, N, d_r, 1, d_r, 1, d_r1));
gpuDotProduct<<<numBlocks, blockSize, blockSize * sizeof(double), stream1>>>(
d_r, d_r, d_r1, N);
checkCudaErrors(cudaMemcpyAsync((float *)&r1, d_r1, sizeof(float),
cudaMemcpyDeviceToHost, stream1));
checkCudaErrors(cudaStreamEndCapture(stream1, &initGraph));
cudaGraphExec_t graphExec;
checkCudaErrors(cudaGraphInstantiate(&graphExec, initGraph, NULL, NULL, 0));
#endif
checkCudaErrors(cublasSetStream(cublasHandle, stream1));
checkCudaErrors(cusparseSetStream(cusparseHandle, stream1));
while (r1 > tol * tol && k <= max_iter) {
#if WITH_GRAPH
checkCudaErrors(cudaGraphLaunch(graphExec, streamForGraph));
checkCudaErrors(cudaStreamSynchronize(streamForGraph));
#else
r1_div_x<<<1, 1, 0, stream1>>>(d_r1, d_r0, d_b);
cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_DEVICE);
checkCudaErrors(cublasSscal(cublasHandle, N, d_b, d_p, 1));
cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_HOST);
checkCudaErrors(cublasSaxpy(cublasHandle, N, &alpha, d_r, 1, d_p, 1));
checkCudaErrors(cusparseScsrmv(
cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, N, N, nz, &alpha,
descr, d_val, d_row, d_col, d_p, &beta, d_Ax));
cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_DEVICE);
checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));
r1_div_x<<<1, 1, 0, stream1>>>(d_r1, d_dot, d_a);
checkCudaErrors(cublasSaxpy(cublasHandle, N, d_a, d_p, 1, d_x, 1));
a_minus<<<1, 1, 0, stream1>>>(d_a, d_na);
checkCudaErrors(cublasSaxpy(cublasHandle, N, d_na, d_Ax, 1, d_r, 1));
checkCudaErrors(cudaMemcpyAsync(d_r0, d_r1, sizeof(float),
cudaMemcpyDeviceToDevice, stream1));
checkCudaErrors(cublasSdot(cublasHandle, N, d_r, 1, d_r, 1, d_r1));
checkCudaErrors(cudaMemcpyAsync((float *)&r1, d_r1, sizeof(float),
cudaMemcpyDeviceToHost, stream1));
checkCudaErrors(cudaStreamSynchronize(stream1));
#endif
printf("iteration = %3d, residual = %e\n", k, sqrt(r1));
k++;
}
#if WITH_GRAPH
checkCudaErrors(cudaMemcpyAsync(x, d_x, N * sizeof(float),
cudaMemcpyDeviceToHost, streamForGraph));
checkCudaErrors(cudaStreamSynchronize(streamForGraph));
#else
checkCudaErrors(cudaMemcpyAsync(x, d_x, N * sizeof(float),
cudaMemcpyDeviceToHost, stream1));
checkCudaErrors(cudaStreamSynchronize(stream1));
#endif
float rsum, diff, err = 0.0;
for (int i = 0; i < N; i++) {
rsum = 0.0;
for (int j = I[i]; j < I[i + 1]; j++) {
rsum += val[j] * x[J[j]];
}
diff = fabs(rsum - rhs[i]);
if (diff > err) {
err = diff;
}
}
#if WITH_GRAPH
checkCudaErrors(cudaGraphExecDestroy(graphExec));
checkCudaErrors(cudaGraphDestroy(initGraph));
checkCudaErrors(cudaStreamDestroy(streamForGraph));
#endif
checkCudaErrors(cudaStreamDestroy(stream1));
cusparseDestroy(cusparseHandle);
cublasDestroy(cublasHandle);
free(I);
free(J);
free(val);
free(x);
free(rhs);
cudaFree(d_col);
cudaFree(d_row);
cudaFree(d_val);
cudaFree(d_x);
cudaFree(d_r);
cudaFree(d_p);
cudaFree(d_Ax);
printf("Test Summary: Error amount = %f\n", err);
exit((k <= max_iter) ? 0 : 1);
}

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2012
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "conjugateGradientCudaGraphs", "conjugateGradientCudaGraphs_vs2012.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>conjugateGradientCudaGraphs_vs2012</RootNamespace>
<ProjectName>conjugateGradientCudaGraphs</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v110</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cublas.lib;cusparse.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/conjugateGradientCudaGraphs.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="conjugateGradientCudaGraphs.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 13.00
# Visual Studio 2013
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "conjugateGradientCudaGraphs", "conjugateGradientCudaGraphs_vs2013.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>conjugateGradientCudaGraphs_vs2013</RootNamespace>
<ProjectName>conjugateGradientCudaGraphs</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v120</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cublas.lib;cusparse.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/conjugateGradientCudaGraphs.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="conjugateGradientCudaGraphs.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 14.00
# Visual Studio 2015
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "conjugateGradientCudaGraphs", "conjugateGradientCudaGraphs_vs2015.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>conjugateGradientCudaGraphs_vs2015</RootNamespace>
<ProjectName>conjugateGradientCudaGraphs</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v140</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cublas.lib;cusparse.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/conjugateGradientCudaGraphs.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="conjugateGradientCudaGraphs.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2017
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "conjugateGradientCudaGraphs", "conjugateGradientCudaGraphs_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,112 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>conjugateGradientCudaGraphs_vs2017</RootNamespace>
<ProjectName>conjugateGradientCudaGraphs</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
<LatestTargetPlatformVersion>$([Microsoft.Build.Utilities.ToolLocationHelper]::GetLatestSDKTargetPlatformVersion('Windows', '10.0'))</LatestTargetPlatformVersion>
<WindowsTargetPlatformVersion Condition="'$(WindowsTargetPlatformVersion)' == ''">$(LatestTargetPlatformVersion)</WindowsTargetPlatformVersion>
<TargetPlatformVersion>$(WindowsTargetPlatformVersion)</TargetPlatformVersion>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v141</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cublas.lib;cusparse.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/conjugateGradientCudaGraphs.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="conjugateGradientCudaGraphs.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2019
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "conjugateGradientCudaGraphs", "conjugateGradientCudaGraphs_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,108 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>conjugateGradientCudaGraphs_vs2019</RootNamespace>
<ProjectName>conjugateGradientCudaGraphs</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v142</PlatformToolset>
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cublas.lib;cusparse.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/conjugateGradientCudaGraphs.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="conjugateGradientCudaGraphs.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" />
</ImportGroup>
</Project>

View File

@@ -1,31 +1,29 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Copyright 1993-2015 NVIDIA Corporation. All rights reserved.
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * 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.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# NOTICE TO USER:
#
# This source code is subject to NVIDIA ownership rights under U.S. and
# international Copyright laws.
#
# NVIDIA MAKES NO REPRESENTATION ABOUT THE SUITABILITY OF THIS SOURCE
# CODE FOR ANY PURPOSE. IT IS PROVIDED "AS IS" WITHOUT EXPRESS OR
# IMPLIED WARRANTY OF ANY KIND. NVIDIA DISCLAIMS ALL WARRANTIES WITH
# REGARD TO THIS SOURCE CODE, INCLUDING ALL IMPLIED WARRANTIES OF
# MERCHANTABILITY, NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR PURPOSE.
# IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY SPECIAL, INDIRECT, INCIDENTAL,
# OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS
# OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
# OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE
# OR PERFORMANCE OF THIS SOURCE CODE.
#
# U.S. Government End Users. This source code is a "commercial item" as
# that term is defined at 48 C.F.R. 2.101 (OCT 1995), consisting of
# "commercial computer software" and "commercial computer software
# documentation" as such terms are used in 48 C.F.R. 12.212 (SEPT 1995)
# and is provided to the U.S. Government only as a commercial end item.
# Consistent with 48 C.F.R.12.212 and 48 C.F.R. 227.7202-1 through
# 227.7202-4 (JUNE 1995), all U.S. Government End Users acquire the
# source code with only those rights set forth herein.
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``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.
#
################################################################################
#
@@ -141,7 +139,7 @@ else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-g++
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
@@ -266,7 +264,11 @@ LIBRARIES :=
################################################################################
# Gencode arguments
SMS ?= 60 61 70
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 60 61 70 72 75
else
SMS ?= 60 61 70 75
endif
ifeq ($(SMS),)
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)

Some files were not shown because too many files have changed in this diff Show More