mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2024-11-24 12:29:16 +08:00
Fix #194 and add Large Kernel Parameters Sample
This commit is contained in:
Rob Nertney
parent
e612904184
commit
8004ad59ab
|
|
@ -7,6 +7,8 @@ Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This ve
|
|||
This section describes the release notes for the CUDA Samples on GitHub only.
|
||||
|
||||
### CUDA 12.1
|
||||
* Added JIT LTO Sample
|
||||
* Adding Large Kernel Sample
|
||||
|
||||
### [older versions...](./CHANGELOG.md)
|
||||
|
||||
|
|
|
|||
|
|
@ -158,12 +158,6 @@ static void cudaImportNvSciImage(cudaExternalResInterop &cudaExtResObj,
|
|||
pairArrayOut[numAttrs++].key = NvSciBufImageAttrKey_Layout;
|
||||
pairArrayOut[numAttrs++].key = NvSciBufImageAttrKey_PlaneBitsPerPixel;
|
||||
pairArrayOut[numAttrs++].key = NvSciBufImageAttrKey_PlaneOffset;
|
||||
pairArrayOut[numAttrs++].key = NvSciBufImageAttrKey_PlanePitch;
|
||||
pairArrayOut[numAttrs++].key = NvSciBufImageAttrKey_PlaneAlignedHeight;
|
||||
|
||||
uint32_t planePitchs[10];
|
||||
uint32_t planePixel[10];
|
||||
uint32_t planeAlignedHeight[10];
|
||||
|
||||
checkNvSciErrors(NvSciBufAttrListGetAttrs(attrlist, pairArrayOut, numAttrs));
|
||||
|
||||
|
|
@ -183,13 +177,6 @@ static void cudaImportNvSciImage(cudaExternalResInterop &cudaExtResObj,
|
|||
cudaExtResObj.planeCount * sizeof(int32_t));
|
||||
memcpy(cudaExtResObj.planeOffset, (uint64_t *)pairArrayOut[7].value,
|
||||
cudaExtResObj.planeCount * sizeof(uint64_t));
|
||||
memcpy(planePixel, (uint32_t *)pairArrayOut[6].value,
|
||||
cudaExtResObj.planeCount * sizeof(uint32_t));
|
||||
memcpy(planePitchs, (uint32_t *)pairArrayOut[8].value,
|
||||
cudaExtResObj.planeCount * sizeof(uint32_t));
|
||||
memcpy(planeAlignedHeight, (uint32_t *)pairArrayOut[9].value,
|
||||
cudaExtResObj.planeCount * sizeof(uint32_t));
|
||||
|
||||
|
||||
NvSciBufAttrValImageLayoutType layout =
|
||||
*(NvSciBufAttrValImageLayoutType *)pairArrayOut[5].value;
|
||||
|
|
@ -214,8 +201,8 @@ static void cudaImportNvSciImage(cudaExternalResInterop &cudaExtResObj,
|
|||
for (int i = 0; i < cudaExtResObj.planeCount; i++) {
|
||||
cudaExtent extent = {};
|
||||
memset(&extent, 0, sizeof(extent));
|
||||
extent.width = planePitchs[i] / (planePixel[i] / 8);
|
||||
extent.height = planeAlignedHeight[i];
|
||||
extent.width = cudaExtResObj.imageWidth[i];
|
||||
extent.height = cudaExtResObj.imageHeight[i];
|
||||
extent.depth = 0;
|
||||
cudaChannelFormatDesc desc;
|
||||
switch (channelCount) {
|
||||
|
|
|
|||
|
|
@ -384,10 +384,10 @@ else
|
|||
@echo "Sample is ready - all dependencies have been met"
|
||||
endif
|
||||
|
||||
jitlto.o:jitlto.cpp
|
||||
jitLto.o:jitLto.cpp
|
||||
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
|
||||
|
||||
jitLto: jitlto.o
|
||||
jitLto: jitLto.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)
|
||||
|
|
@ -398,7 +398,7 @@ run: build
|
|||
testrun: build
|
||||
|
||||
clean:
|
||||
rm -f jitLto jitlto.o
|
||||
rm -f jitLto jitLto.o
|
||||
rm -rf ../../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/jitLto
|
||||
|
||||
clobber: clean
|
||||
|
|
|
|||
251
Samples/4_CUDA_Libraries/jitLto/jitLto.cpp
Normal file
251
Samples/4_CUDA_Libraries/jitLto/jitLto.cpp
Normal file
|
|
@ -0,0 +1,251 @@
|
|||
/* Copyright (c) 2022, 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 <cuda.h>
|
||||
#include <nvJitLink.h>
|
||||
#include <nvrtc.h>
|
||||
#include <iostream>
|
||||
#include <cstring>
|
||||
|
||||
#define NUM_THREADS 128
|
||||
#define NUM_BLOCKS 32
|
||||
|
||||
#define NVRTC_SAFE_CALL(x) \
|
||||
do { \
|
||||
nvrtcResult result = x; \
|
||||
if (result != NVRTC_SUCCESS) { \
|
||||
std::cerr << "\nerror: " #x " failed with error " \
|
||||
<< nvrtcGetErrorString(result) << '\n'; \
|
||||
exit(1); \
|
||||
} \
|
||||
} while(0)
|
||||
#define CUDA_SAFE_CALL(x) \
|
||||
do { \
|
||||
CUresult result = x; \
|
||||
if (result != CUDA_SUCCESS) { \
|
||||
const char *msg; \
|
||||
cuGetErrorName(result, &msg); \
|
||||
std::cerr << "\nerror: " #x " failed with error " \
|
||||
<< msg << '\n'; \
|
||||
exit(1); \
|
||||
} \
|
||||
} while(0)
|
||||
#define NVJITLINK_SAFE_CALL(h,x) \
|
||||
do { \
|
||||
nvJitLinkResult result = x; \
|
||||
if (result != NVJITLINK_SUCCESS) { \
|
||||
std::cerr << "\nerror: " #x " failed with error " \
|
||||
<< result << '\n'; \
|
||||
size_t lsize; \
|
||||
result = nvJitLinkGetErrorLogSize(h, &lsize); \
|
||||
if (result == NVJITLINK_SUCCESS && lsize > 0) { \
|
||||
char *log = (char*)malloc(lsize); \
|
||||
result = nvJitLinkGetErrorLog(h, log); \
|
||||
if (result == NVJITLINK_SUCCESS) { \
|
||||
std::cerr << "error log: " << log << '\n'; \
|
||||
free(log); \
|
||||
} \
|
||||
} \
|
||||
exit(1); \
|
||||
} \
|
||||
} while(0)
|
||||
|
||||
const char *lto_saxpy = " \n\
|
||||
extern __device__ float compute(float a, float x, float y); \n\
|
||||
\n\
|
||||
extern \"C\" __global__ \n\
|
||||
void saxpy(float a, float *x, float *y, float *out, size_t n) \n\
|
||||
{ \n\
|
||||
size_t tid = blockIdx.x * blockDim.x + threadIdx.x; \n\
|
||||
if (tid < n) { \n\
|
||||
out[tid] = compute(a, x[tid], y[tid]); \n\
|
||||
} \n\
|
||||
} \n";
|
||||
|
||||
const char *lto_compute = " \n\
|
||||
__device__ float compute(float a, float x, float y) { \n\
|
||||
return a * x + y; \n\
|
||||
} \n";
|
||||
|
||||
// compile code into LTOIR, returning the IR and its size
|
||||
static void getLTOIR (const char *code, const char *name,
|
||||
char **ltoIR, size_t *ltoIRSize)
|
||||
{
|
||||
// Create an instance of nvrtcProgram with the code string.
|
||||
nvrtcProgram prog;
|
||||
NVRTC_SAFE_CALL(
|
||||
nvrtcCreateProgram(&prog, // prog
|
||||
code, // buffer
|
||||
name, // name
|
||||
0, // numHeaders
|
||||
NULL, // headers
|
||||
NULL)); // includeNames
|
||||
|
||||
// specify that LTO IR should be generated for LTO operation
|
||||
const char *opts[] = {"-dlto",
|
||||
"--relocatable-device-code=true"};
|
||||
nvrtcResult compileResult = nvrtcCompileProgram(prog, // prog
|
||||
2, // numOptions
|
||||
opts); // options
|
||||
// Obtain compilation log from the program.
|
||||
size_t logSize;
|
||||
NVRTC_SAFE_CALL(nvrtcGetProgramLogSize(prog, &logSize));
|
||||
char *log = new char[logSize];
|
||||
NVRTC_SAFE_CALL(nvrtcGetProgramLog(prog, log));
|
||||
std::cout << log << '\n';
|
||||
delete[] log;
|
||||
if (compileResult != NVRTC_SUCCESS) {
|
||||
exit(1);
|
||||
}
|
||||
// Obtain generated LTO IR from the program.
|
||||
NVRTC_SAFE_CALL(nvrtcGetLTOIRSize(prog, ltoIRSize));
|
||||
*ltoIR = new char[*ltoIRSize];
|
||||
NVRTC_SAFE_CALL(nvrtcGetLTOIR(prog, *ltoIR));
|
||||
// Destroy the program.
|
||||
NVRTC_SAFE_CALL(nvrtcDestroyProgram(&prog));
|
||||
}
|
||||
|
||||
int main(int argc, char *argv[])
|
||||
{
|
||||
char *ltoIR1;
|
||||
char *ltoIR2;
|
||||
size_t ltoIR1Size;
|
||||
size_t ltoIR2Size;
|
||||
// getLTOIR uses nvrtc to get the LTOIR.
|
||||
// We could also use nvcc offline with -dlto -fatbin
|
||||
// to generate the IR, but using nvrtc keeps the build simpler.
|
||||
getLTOIR(lto_saxpy, "lto_saxpy.cu", <oIR1, <oIR1Size);
|
||||
getLTOIR(lto_compute, "lto_compute.cu", <oIR2, <oIR2Size);
|
||||
|
||||
CUdevice cuDevice;
|
||||
CUcontext context;
|
||||
CUmodule module;
|
||||
CUfunction kernel;
|
||||
CUDA_SAFE_CALL(cuInit(0));
|
||||
CUDA_SAFE_CALL(cuDeviceGet(&cuDevice, 0));
|
||||
CUDA_SAFE_CALL(cuCtxCreate(&context, 0, cuDevice));
|
||||
|
||||
// Dynamically determine the arch to link for
|
||||
int major = 0;
|
||||
int minor = 0;
|
||||
CUDA_SAFE_CALL(cuDeviceGetAttribute(&major,
|
||||
CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
|
||||
CUDA_SAFE_CALL(cuDeviceGetAttribute(&minor,
|
||||
CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
|
||||
int arch = major*10 + minor;
|
||||
char smbuf[16];
|
||||
memset(smbuf,0,16);
|
||||
sprintf(smbuf, "-arch=sm_%d", arch);
|
||||
|
||||
// Load the generated LTO IR and link them together
|
||||
nvJitLinkHandle handle;
|
||||
const char *lopts[] = {"-lto", smbuf};
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkCreate(&handle, 2, lopts));
|
||||
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkAddData(handle, NVJITLINK_INPUT_LTOIR,
|
||||
(void *)ltoIR1, ltoIR1Size, "lto_saxpy"));
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkAddData(handle, NVJITLINK_INPUT_LTOIR,
|
||||
(void *)ltoIR2, ltoIR2Size, "lto_compute"));
|
||||
|
||||
// The call to nvJitLinkComplete causes linker to link together the two
|
||||
// LTO IR modules, do optimization on the linked LTO IR,
|
||||
// and generate cubin from it.
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkComplete(handle));
|
||||
|
||||
// check error log
|
||||
size_t logSize;
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkGetErrorLogSize(handle, &logSize));
|
||||
if (logSize > 0) {
|
||||
char *log = (char*)malloc(logSize+1);
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkGetErrorLog(handle, log));
|
||||
std::cout << "Error log: " << log << std::endl;
|
||||
free(log);
|
||||
}
|
||||
|
||||
// get linked cubin
|
||||
size_t cubinSize;
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkGetLinkedCubinSize(handle, &cubinSize));
|
||||
void *cubin = malloc(cubinSize);
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkGetLinkedCubin(handle, cubin));
|
||||
|
||||
NVJITLINK_SAFE_CALL(handle, nvJitLinkDestroy(&handle));
|
||||
delete[] ltoIR1;
|
||||
delete[] ltoIR2;
|
||||
|
||||
// cubin is linked, so now load it
|
||||
CUDA_SAFE_CALL(cuModuleLoadData(&module, cubin));
|
||||
CUDA_SAFE_CALL(cuModuleGetFunction(&kernel, module, "saxpy"));
|
||||
|
||||
// Generate input for execution, and create output buffers.
|
||||
size_t n = NUM_THREADS * NUM_BLOCKS;
|
||||
size_t bufferSize = n * sizeof(float);
|
||||
float a = 5.1f;
|
||||
float *hX = new float[n], *hY = new float[n], *hOut = new float[n];
|
||||
for (size_t i = 0; i < n; ++i) {
|
||||
hX[i] = static_cast<float>(i);
|
||||
hY[i] = static_cast<float>(i * 2);
|
||||
}
|
||||
CUdeviceptr dX, dY, dOut;
|
||||
CUDA_SAFE_CALL(cuMemAlloc(&dX, bufferSize));
|
||||
CUDA_SAFE_CALL(cuMemAlloc(&dY, bufferSize));
|
||||
CUDA_SAFE_CALL(cuMemAlloc(&dOut, bufferSize));
|
||||
CUDA_SAFE_CALL(cuMemcpyHtoD(dX, hX, bufferSize));
|
||||
CUDA_SAFE_CALL(cuMemcpyHtoD(dY, hY, bufferSize));
|
||||
// Execute SAXPY.
|
||||
void *args[] = { &a, &dX, &dY, &dOut, &n };
|
||||
CUDA_SAFE_CALL(
|
||||
cuLaunchKernel(kernel,
|
||||
NUM_BLOCKS, 1, 1, // grid dim
|
||||
NUM_THREADS, 1, 1, // block dim
|
||||
0, NULL, // shared mem and stream
|
||||
args, 0)); // arguments
|
||||
CUDA_SAFE_CALL(cuCtxSynchronize());
|
||||
// Retrieve and print output.
|
||||
CUDA_SAFE_CALL(cuMemcpyDtoH(hOut, dOut, bufferSize));
|
||||
|
||||
for (size_t i = 0; i < n; ++i) {
|
||||
std::cout << a << " * " << hX[i] << " + " << hY[i]
|
||||
<< " = " << hOut[i] << '\n';
|
||||
}
|
||||
// check last value to verify
|
||||
if (hOut[n-1] == 29074.5) {
|
||||
std::cout << "PASSED!\n";
|
||||
} else {
|
||||
std::cout << "values not expected?\n";
|
||||
}
|
||||
// Release resources.
|
||||
CUDA_SAFE_CALL(cuMemFree(dX));
|
||||
CUDA_SAFE_CALL(cuMemFree(dY));
|
||||
CUDA_SAFE_CALL(cuMemFree(dOut));
|
||||
CUDA_SAFE_CALL(cuModuleUnload(module));
|
||||
CUDA_SAFE_CALL(cuCtxDestroy(context));
|
||||
free(cubin);
|
||||
delete[] hX;
|
||||
delete[] hY;
|
||||
delete[] hOut;
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -102,7 +102,7 @@
|
|||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="jitlto.cpp" />
|
||||
<ClCompile Include="jitLto.cpp" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
|
|
|
|||
|
|
@ -98,7 +98,7 @@
|
|||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="jitlto.cpp" />
|
||||
<ClCompile Include="jitLto.cpp" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
|
|
|
|||
|
|
@ -98,7 +98,7 @@
|
|||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="jitlto.cpp" />
|
||||
<ClCompile Include="jitLto.cpp" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
|
|
|
|||
18
Samples/6_Performance/LargeKernelParameter/.vscode/c_cpp_properties.json
vendored
Normal file
18
Samples/6_Performance/LargeKernelParameter/.vscode/c_cpp_properties.json
vendored
Normal file
|
|
@ -0,0 +1,18 @@
|
|||
{
|
||||
"configurations": [
|
||||
{
|
||||
"name": "Linux",
|
||||
"includePath": [
|
||||
"${workspaceFolder}/**",
|
||||
"${workspaceFolder}/../../../Common"
|
||||
],
|
||||
"defines": [],
|
||||
"compilerPath": "/usr/local/cuda/bin/nvcc",
|
||||
"cStandard": "gnu17",
|
||||
"cppStandard": "gnu++14",
|
||||
"intelliSenseMode": "linux-gcc-x64",
|
||||
"configurationProvider": "ms-vscode.makefile-tools"
|
||||
}
|
||||
],
|
||||
"version": 4
|
||||
}
|
||||
7
Samples/6_Performance/LargeKernelParameter/.vscode/extensions.json
vendored
Normal file
7
Samples/6_Performance/LargeKernelParameter/.vscode/extensions.json
vendored
Normal file
|
|
@ -0,0 +1,7 @@
|
|||
{
|
||||
"recommendations": [
|
||||
"nvidia.nsight-vscode-edition",
|
||||
"ms-vscode.cpptools",
|
||||
"ms-vscode.makefile-tools"
|
||||
]
|
||||
}
|
||||
10
Samples/6_Performance/LargeKernelParameter/.vscode/launch.json
vendored
Normal file
10
Samples/6_Performance/LargeKernelParameter/.vscode/launch.json
vendored
Normal file
|
|
@ -0,0 +1,10 @@
|
|||
{
|
||||
"configurations": [
|
||||
{
|
||||
"name": "CUDA C++: Launch",
|
||||
"type": "cuda-gdb",
|
||||
"request": "launch",
|
||||
"program": "${workspaceFolder}/LargeKernelParameter"
|
||||
}
|
||||
]
|
||||
}
|
||||
15
Samples/6_Performance/LargeKernelParameter/.vscode/tasks.json
vendored
Normal file
15
Samples/6_Performance/LargeKernelParameter/.vscode/tasks.json
vendored
Normal file
|
|
@ -0,0 +1,15 @@
|
|||
{
|
||||
"version": "2.0.0",
|
||||
"tasks": [
|
||||
{
|
||||
"label": "sample",
|
||||
"type": "shell",
|
||||
"command": "make dbg=1",
|
||||
"problemMatcher": ["$nvcc"],
|
||||
"group": {
|
||||
"kind": "build",
|
||||
"isDefault": true
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
|
|
@ -0,0 +1,173 @@
|
|||
/* Copyright (c) 2023, 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 showing performance and usability
|
||||
* improvements with large kernel parameters introduced in CUDA 12.1
|
||||
*/
|
||||
#include <chrono>
|
||||
#include <iostream>
|
||||
#include <cassert>
|
||||
|
||||
// Utility includes
|
||||
#include <helper_cuda.h>
|
||||
|
||||
using namespace std;
|
||||
using namespace std::chrono;
|
||||
|
||||
#define TEST_ITERATIONS (1000)
|
||||
#define TOTAL_PARAMS (8000) // ints
|
||||
#define KERNEL_PARAM_LIMIT (1024) // ints
|
||||
#define CONST_COPIED_PARAMS (TOTAL_PARAMS - KERNEL_PARAM_LIMIT)
|
||||
|
||||
__constant__ int excess_params[CONST_COPIED_PARAMS];
|
||||
|
||||
typedef struct {
|
||||
int param[KERNEL_PARAM_LIMIT];
|
||||
} param_t;
|
||||
|
||||
typedef struct {
|
||||
int param[TOTAL_PARAMS];
|
||||
} param_large_t;
|
||||
|
||||
// Kernel with 4KB kernel parameter limit
|
||||
__global__ void kernelDefault(__grid_constant__ const param_t p, int *result) {
|
||||
int tmp = 0;
|
||||
|
||||
// accumulate kernel parameters
|
||||
for (int i = 0; i < KERNEL_PARAM_LIMIT; ++i) {
|
||||
tmp += p.param[i];
|
||||
}
|
||||
|
||||
// accumulate excess values passed via const memory
|
||||
for (int i = 0; i < CONST_COPIED_PARAMS; ++i) {
|
||||
tmp += excess_params[i];
|
||||
}
|
||||
|
||||
*result = tmp;
|
||||
}
|
||||
|
||||
// Kernel with 32,764 byte kernel parameter limit
|
||||
__global__ void kernelLargeParam(__grid_constant__ const param_large_t p, int *result) {
|
||||
int tmp = 0;
|
||||
|
||||
// accumulate kernel parameters
|
||||
for (int i = 0; i < TOTAL_PARAMS; ++i) {
|
||||
tmp += p.param[i];
|
||||
}
|
||||
|
||||
*result = tmp;
|
||||
}
|
||||
|
||||
static void report_time(std::chrono::time_point<std::chrono::steady_clock> start,
|
||||
std::chrono::time_point<std::chrono::steady_clock> end,
|
||||
int iters) {
|
||||
auto usecs = duration_cast<duration<float,
|
||||
microseconds::period>>(end - start);
|
||||
cout << usecs.count() / iters << endl;
|
||||
}
|
||||
|
||||
int main() {
|
||||
int rc;
|
||||
cudaFree(0);
|
||||
|
||||
param_t p;
|
||||
param_large_t p_large;
|
||||
|
||||
// pageable host memory that holds excess constants passed via constant memory
|
||||
int *copied_params = (int *)malloc(CONST_COPIED_PARAMS * sizeof(int));
|
||||
assert(copied_params);
|
||||
|
||||
// storage for computed result
|
||||
int *d_result;
|
||||
int h_result;
|
||||
checkCudaErrors(cudaMalloc(&d_result, sizeof(int)));
|
||||
|
||||
int expected_result = 0;
|
||||
|
||||
// fill in data for validation
|
||||
for (int i = 0; i < KERNEL_PARAM_LIMIT; ++i) {
|
||||
p.param[i] = (i & 0xFF);
|
||||
}
|
||||
for (int i = KERNEL_PARAM_LIMIT; i < TOTAL_PARAMS; ++i) {
|
||||
copied_params[i - KERNEL_PARAM_LIMIT] = (i & 0xFF);
|
||||
}
|
||||
for (int i = 0; i < TOTAL_PARAMS; ++i) {
|
||||
p_large.param[i] = (i & 0xFF);
|
||||
expected_result += (i & 0xFF);
|
||||
}
|
||||
|
||||
// warmup, verify correctness
|
||||
checkCudaErrors(cudaMemcpyToSymbol(excess_params, copied_params, CONST_COPIED_PARAMS * sizeof(int), 0, cudaMemcpyHostToDevice));
|
||||
kernelDefault<<<1,1>>>(p, d_result);
|
||||
checkCudaErrors(cudaMemcpy(&h_result, d_result, sizeof(int), cudaMemcpyDeviceToHost));
|
||||
checkCudaErrors(cudaDeviceSynchronize());
|
||||
if(h_result != expected_result) {
|
||||
std::cout << "Test failed" << std::endl;
|
||||
rc=-1;
|
||||
goto Exit;
|
||||
}
|
||||
|
||||
kernelLargeParam<<<1,1>>>(p_large, d_result);
|
||||
checkCudaErrors(cudaMemcpy(&h_result, d_result, sizeof(int), cudaMemcpyDeviceToHost));
|
||||
checkCudaErrors(cudaDeviceSynchronize());
|
||||
if(h_result != expected_result) {
|
||||
std::cout << "Test failed" << std::endl;
|
||||
rc=-1;
|
||||
goto Exit;
|
||||
}
|
||||
|
||||
// benchmark default kernel parameter limit
|
||||
{
|
||||
auto start = steady_clock::now();
|
||||
for (int i = 0; i < TEST_ITERATIONS; ++i) {
|
||||
checkCudaErrors(cudaMemcpyToSymbol(excess_params, copied_params, CONST_COPIED_PARAMS * sizeof(int), 0, cudaMemcpyHostToDevice));
|
||||
kernelDefault<<<1, 1>>>(p, d_result);
|
||||
}
|
||||
checkCudaErrors(cudaDeviceSynchronize());
|
||||
auto end = steady_clock::now();
|
||||
std::cout << "Kernel 4KB parameter limit - time (us):";
|
||||
report_time(start, end, TEST_ITERATIONS);
|
||||
|
||||
// benchmark large kernel parameter limit
|
||||
start = steady_clock::now();
|
||||
for (int i = 0; i < TEST_ITERATIONS; ++i) {
|
||||
kernelLargeParam<<<1, 1>>>(p_large, d_result);
|
||||
}
|
||||
checkCudaErrors(cudaDeviceSynchronize());
|
||||
end = steady_clock::now();
|
||||
std::cout << "Kernel 32,764 byte parameter limit - time (us):";
|
||||
report_time(start, end, TEST_ITERATIONS);
|
||||
}
|
||||
std::cout << "Test passed!" << std::endl;
|
||||
rc=0;
|
||||
Exit:
|
||||
// cleanup
|
||||
cudaFree(d_result);
|
||||
free(copied_params);
|
||||
return rc;
|
||||
}
|
||||
|
|
@ -1,7 +1,7 @@
|
|||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2017
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "jitlto", "jitlto_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "LargeKernelParameter", "LargeKernelParameter_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
|
|
@ -15,8 +15,8 @@
|
|||
</ItemGroup>
|
||||
<PropertyGroup Label="Globals">
|
||||
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
|
||||
<RootNamespace>jitlto_vs2017</RootNamespace>
|
||||
<ProjectName>jitlto</ProjectName>
|
||||
<RootNamespace>LargeKernelParameter_vs2017</RootNamespace>
|
||||
<ProjectName>LargeKernelParameter</ProjectName>
|
||||
<CudaToolkitCustomDir />
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
|
||||
|
|
@ -38,7 +38,7 @@
|
|||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.0.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.1.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
|
|
@ -58,16 +58,16 @@
|
|||
<ClCompile>
|
||||
<WarningLevel>Level3</WarningLevel>
|
||||
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;$(CudaToolkitIncludeDir);$(CUDA_PATH)/include;</AdditionalIncludeDirectories>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;</AdditionalIncludeDirectories>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<SubSystem>Console</SubSystem>
|
||||
<AdditionalDependencies>cuda.lib;nvrtc.lib;nvJitLink.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>
|
||||
<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)/jitlto.exe</OutputFile>
|
||||
<OutputFile>$(OutDir)/LargeKernelParameter.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration></CodeGeneration>
|
||||
<CodeGeneration>compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;compute_89,sm_89;compute_90,sm_90;</CodeGeneration>
|
||||
<AdditionalOptions>-Xcompiler "/wd 4819" --threads 0 </AdditionalOptions>
|
||||
<Include>./;../../../Common</Include>
|
||||
<Defines>WIN32</Defines>
|
||||
|
|
@ -102,11 +102,11 @@
|
|||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="jitlto.cpp" />
|
||||
<CudaCompile Include="LargeKernelParameter.cu" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.0.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.1.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
|
@ -1,7 +1,7 @@
|
|||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2019
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "jitlto", "jitlto_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "LargeKernelParameter", "LargeKernelParameter_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
|
|
@ -15,8 +15,8 @@
|
|||
</ItemGroup>
|
||||
<PropertyGroup Label="Globals">
|
||||
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
|
||||
<RootNamespace>jitlto_vs2019</RootNamespace>
|
||||
<ProjectName>jitlto</ProjectName>
|
||||
<RootNamespace>LargeKernelParameter_vs2019</RootNamespace>
|
||||
<ProjectName>LargeKernelParameter</ProjectName>
|
||||
<CudaToolkitCustomDir />
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
|
||||
|
|
@ -34,7 +34,7 @@
|
|||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.0.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.1.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
|
|
@ -54,16 +54,16 @@
|
|||
<ClCompile>
|
||||
<WarningLevel>Level3</WarningLevel>
|
||||
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;$(CudaToolkitIncludeDir);$(CUDA_PATH)/include;</AdditionalIncludeDirectories>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;</AdditionalIncludeDirectories>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<SubSystem>Console</SubSystem>
|
||||
<AdditionalDependencies>cuda.lib;nvrtc.lib;nvJitLink.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>
|
||||
<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)/jitlto.exe</OutputFile>
|
||||
<OutputFile>$(OutDir)/LargeKernelParameter.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration></CodeGeneration>
|
||||
<CodeGeneration>compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;compute_89,sm_89;compute_90,sm_90;</CodeGeneration>
|
||||
<AdditionalOptions>-Xcompiler "/wd 4819" --threads 0 </AdditionalOptions>
|
||||
<Include>./;../../../Common</Include>
|
||||
<Defines>WIN32</Defines>
|
||||
|
|
@ -98,11 +98,11 @@
|
|||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="jitlto.cpp" />
|
||||
<CudaCompile Include="LargeKernelParameter.cu" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.0.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.1.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
|
@ -1,7 +1,7 @@
|
|||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2022
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "jitlto", "jitlto_vs2022.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "LargeKernelParameter", "LargeKernelParameter_vs2022.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
|
|
@ -15,8 +15,8 @@
|
|||
</ItemGroup>
|
||||
<PropertyGroup Label="Globals">
|
||||
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
|
||||
<RootNamespace>jitlto_vs2022</RootNamespace>
|
||||
<ProjectName>jitlto</ProjectName>
|
||||
<RootNamespace>LargeKernelParameter_vs2022</RootNamespace>
|
||||
<ProjectName>LargeKernelParameter</ProjectName>
|
||||
<CudaToolkitCustomDir />
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
|
||||
|
|
@ -34,7 +34,7 @@
|
|||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.0.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.1.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
|
|
@ -54,16 +54,16 @@
|
|||
<ClCompile>
|
||||
<WarningLevel>Level3</WarningLevel>
|
||||
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;$(CudaToolkitIncludeDir);$(CUDA_PATH)/include;</AdditionalIncludeDirectories>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;</AdditionalIncludeDirectories>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<SubSystem>Console</SubSystem>
|
||||
<AdditionalDependencies>cuda.lib;nvrtc.lib;nvJitLink.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>
|
||||
<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)/jitlto.exe</OutputFile>
|
||||
<OutputFile>$(OutDir)/LargeKernelParameter.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration></CodeGeneration>
|
||||
<CodeGeneration>compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;compute_89,sm_89;compute_90,sm_90;</CodeGeneration>
|
||||
<AdditionalOptions>-Xcompiler "/wd 4819" --threads 0 </AdditionalOptions>
|
||||
<Include>./;../../../Common</Include>
|
||||
<Defines>WIN32</Defines>
|
||||
|
|
@ -98,11 +98,11 @@
|
|||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<ClCompile Include="jitlto.cpp" />
|
||||
<CudaCompile Include="LargeKernelParameter.cu" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.0.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.1.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
340
Samples/6_Performance/LargeKernelParameter/Makefile
Normal file
340
Samples/6_Performance/LargeKernelParameter/Makefile
Normal file
|
|
@ -0,0 +1,340 @@
|
|||
################################################################################
|
||||
# Copyright (c) 2022, 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 $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
|
||||
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
|
||||
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
|
||||
TARGET_SIZE := 64
|
||||
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
|
||||
TARGET_SIZE := 32
|
||||
endif
|
||||
else
|
||||
TARGET_SIZE := $(shell getconf LONG_BIT)
|
||||
endif
|
||||
else
|
||||
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
|
||||
endif
|
||||
|
||||
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
|
||||
ifeq ($(HOST_ARCH),aarch64)
|
||||
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux 2>/dev/null))
|
||||
HOST_ARCH := sbsa
|
||||
TARGET_ARCH := sbsa
|
||||
endif
|
||||
endif
|
||||
|
||||
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
|
||||
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa 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 ?= $(QNX_HOST)/usr/bin/q++
|
||||
else ifeq ($(TARGET_OS), android)
|
||||
HOST_COMPILER ?= aarch64-linux-android-clang++
|
||||
endif
|
||||
else ifeq ($(TARGET_ARCH),sbsa)
|
||||
HOST_COMPILER ?= aarch64-linux-gnu-g++
|
||||
else ifeq ($(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)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
|
||||
LDFLAGS += -rpath-link=$(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 -I$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include/libdrm
|
||||
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
|
||||
endif
|
||||
endif
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
|
||||
NVCCFLAGS += -D_QNX_SOURCE
|
||||
NVCCFLAGS += --qpp-config 8.3.0,gcc_ntoaarch64le
|
||||
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
|
||||
LDFLAGS += -lsocket
|
||||
LDFLAGS += -L/usr/lib/aarch64-qnx-gnu
|
||||
CCFLAGS += "-Wl\,-rpath-link\,/usr/lib/aarch64-qnx-gnu"
|
||||
ifdef TARGET_OVERRIDE
|
||||
LDFLAGS += -lslog2
|
||||
endif
|
||||
|
||||
ifneq ($(TARGET_FS),)
|
||||
LDFLAGS += -L$(TARGET_FS)/usr/lib
|
||||
CCFLAGS += "-Wl\,-rpath-link\,$(TARGET_FS)/usr/lib"
|
||||
LDFLAGS += -L$(TARGET_FS)/usr/libnvidia
|
||||
CCFLAGS += "-Wl\,-rpath-link\,$(TARGET_FS)/usr/libnvidia"
|
||||
CCFLAGS += -I$(TARGET_FS)/../include
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
|
||||
ifdef TARGET_OVERRIDE # cuda toolkit targets override
|
||||
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
|
||||
endif
|
||||
|
||||
# Install directory of different arch
|
||||
CUDA_INSTALL_TARGET_DIR :=
|
||||
ifeq ($(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)-$(TARGET_OS),sbsa-linux)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
|
||||
else ifeq ($(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 $(TARGET_ARCH),armv7l aarch64 sbsa))
|
||||
SMS ?= 70 72 75 80 86 87 90
|
||||
else
|
||||
SMS ?= 70 75 80 86 89 90
|
||||
endif
|
||||
|
||||
ifeq ($(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
|
||||
|
||||
ALL_CCFLAGS += --std=c++11 --threads 0
|
||||
|
||||
ifeq ($(SAMPLE_ENABLED),0)
|
||||
EXEC ?= @echo "[@]"
|
||||
endif
|
||||
|
||||
################################################################################
|
||||
|
||||
# Target rules
|
||||
all: build
|
||||
|
||||
build: LargeKernelParameter
|
||||
|
||||
check.deps:
|
||||
ifeq ($(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
|
||||
|
||||
LargeKernelParameter.o:LargeKernelParameter.cu
|
||||
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
|
||||
|
||||
LargeKernelParameter: LargeKernelParameter.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
|
||||
$(EXEC) ./LargeKernelParameter
|
||||
|
||||
testrun: build
|
||||
|
||||
clean:
|
||||
rm -f LargeKernelParameter LargeKernelParameter.o
|
||||
rm -rf ../../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/LargeKernelParameter
|
||||
|
||||
clobber: clean
|
||||
70
Samples/6_Performance/LargeKernelParameter/README.md
Normal file
70
Samples/6_Performance/LargeKernelParameter/README.md
Normal file
|
|
@ -0,0 +1,70 @@
|
|||
# LargeKernelParameter - Large Kernel Parameters
|
||||
|
||||
## Description
|
||||
|
||||
Simple test that demonstrates performance and usability improvements with large kernel parameters introduced in CUDA 12.1
|
||||
|
||||
## Key Concepts
|
||||
|
||||
Performance Strategies
|
||||
|
||||
## Supported SM Architectures
|
||||
|
||||
[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) [SM 8.0 ](https://developer.nvidia.com/cuda-gpus) [SM 8.6 ](https://developer.nvidia.com/cuda-gpus) [SM 8.7 ](https://developer.nvidia.com/cuda-gpus) [SM 8.9 ](https://developer.nvidia.com/cuda-gpus) [SM 9.0 ](https://developer.nvidia.com/cuda-gpus)
|
||||
|
||||
## Supported OSes
|
||||
|
||||
Linux, Windows
|
||||
|
||||
## Supported CPU Architecture
|
||||
|
||||
x86_64, ppc64le, armv7l
|
||||
|
||||
## CUDA APIs involved
|
||||
|
||||
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
|
||||
cudaMemcpyToSymbol, cudaMemcpy, cudaFree, cudaDeviceSynchronize, cudaMalloc
|
||||
|
||||
## Prerequisites
|
||||
|
||||
Download and install the [CUDA Toolkit 12.1](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++
|
||||
```
|
||||
|
||||
## References (for more details)
|
||||
|
||||
Loading…
Reference in New Issue
Block a user