NASP/cuda-samples
84
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2026-10-13 00:08:26 +08:00
Code Issues Packages Projects Releases Wiki Activity

CUDA 13.2 samples update (#432)

Browse Source 
- Added Python samples for CUDA Python 1.0 release
- Renamed top-level `Samples` directory to `cpp` to accommodate Python samples.
This commit is contained in:
Dheemanth
2026-05-13 15:13:18 -07:00
committed by GitHub
parent a4526d5229
commit aeab82ff30
2072 changed files with 15077 additions and 1970 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
cpp/3_CUDA_Features/StreamPriorities/.vscode/c_cpp_properties.json vendored Normal file
View 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
cpp/3_CUDA_Features/StreamPriorities/.vscode/extensions.json vendored Normal file
View File

@@ -0,0 +1,7 @@
{
"recommendations": [
"nvidia.nsight-vscode-edition",
"ms-vscode.cpptools",
"ms-vscode.makefile-tools"
]
}

42
cpp/3_CUDA_Features/StreamPriorities/CMakeLists.txt Normal file
View File

@@ -0,0 +1,42 @@
cmake_minimum_required(VERSION 3.20)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Modules")
project(StreamPriorities LANGUAGES C CXX CUDA)
find_package(CUDAToolkit REQUIRED)
set(CMAKE_POSITION_INDEPENDENT_CODE ON)
set(CMAKE_CUDA_ARCHITECTURES 75 80 86 87 89 90 100 110 120)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
if(ENABLE_CUDA_DEBUG)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G") # enable cuda-gdb (may significantly affect performance on some targets)
else()
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
endif()
# Include directories and libraries
include_directories(../../../Common)
if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
# Source file
# Add target for StreamPriorities
if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")
message(STATUS "Will not build sample streamPriorities - not supported on aarch64")
else()
add_executable(StreamPriorities StreamPriorities.cu)
target_compile_options(StreamPriorities PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>)
target_compile_features(StreamPriorities PRIVATE cxx_std_17 cuda_std_17)
set_target_properties(StreamPriorities PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
endif()
else()
message(STATUS "Will not build sample StreamPriorities - requires Linux OS")
endif()
# Include installation configuration
include(${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/InstallSamples.cmake)
setup_samples_install()

34
cpp/3_CUDA_Features/StreamPriorities/README.md Normal file
View File

@@ -0,0 +1,34 @@
# StreamPriorities - Stream Priorities
## Description
This sample demonstrates basic use of stream priorities.
## Key Concepts
CUDA Streams and Events
## Supported SM Architectures
## Supported OSes
Linux
## Supported CPU Architecture
x86_64, ppc64le
## CUDA APIs involved
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
cudaMemcpy, cudaStreamCreateWithPriority, cudaDeviceGetStreamPriorityRange, cudaEventSynchronize, cudaEventRecord, cudaMalloc, cudaEventElapsedTime, cudaGetDeviceProperties, cudaEventCreate
## Dependencies needed to build/run
[Stream-Priorities](../../../README.md#stream-priorities)
## Prerequisites
Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## References (for more details)

193
cpp/3_CUDA_Features/StreamPriorities/StreamPriorities.cu Normal file
View File

@@ -0,0 +1,193 @@
/* 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.
*/
// std::system includes
#include <cstdio>
// CUDA-C includes
#include <cuda_runtime.h>
#include <helper_cuda.h>
#define TOTAL_SIZE 256 * 1024 * 1024
#define EACH_SIZE 128 * 1024 * 1024
// # threadblocks
#define TBLOCKS 1024
#define THREADS 512
// throw error on equality
#define ERR_EQ(X, Y) \
do { \
if ((X) == (Y)) { \
fprintf(stderr, "Error in %s at %s:%d\n", __func__, __FILE__, __LINE__); \
exit(-1); \
} \
} while (0)
// throw error on difference
#define ERR_NE(X, Y) \
do { \
if ((X) != (Y)) { \
fprintf(stderr, "Error in %s at %s:%d\n", __func__, __FILE__, __LINE__); \
exit(-1); \
} \
} while (0)
// copy from source -> destination arrays
__global__ void memcpy_kernel(int *dst, int *src, size_t n)
{
int num = gridDim.x * blockDim.x;
int id = blockDim.x * blockIdx.x + threadIdx.x;
for (int i = id; i < n / sizeof(int); i += num) {
dst[i] = src[i];
}
}
// initialise memory
void mem_init(int *buf, size_t n)
{
for (int i = 0; i < n / sizeof(int); i++) {
buf[i] = i;
}
}
int main(int argc, char **argv)
{
cudaDeviceProp device_prop;
int dev_id;
printf("Starting [%s]...\n", argv[0]);
// set device
dev_id = findCudaDevice(argc, (const char **)argv);
checkCudaErrors(cudaGetDeviceProperties(&device_prop, dev_id));
if ((device_prop.major << 4) + device_prop.minor < 0x35) {
fprintf(stderr,
"%s requires Compute Capability of SM 3.5 or higher to "
"run.\nexiting...\n",
argv[0]);
exit(EXIT_WAIVED);
}
// get the range of priorities available
// [ greatest_priority, lowest_priority ]
int priority_low;
int priority_hi;
checkCudaErrors(cudaDeviceGetStreamPriorityRange(&priority_low, &priority_hi));
printf("CUDA stream priority range: LOW: %d to HIGH: %d\n", priority_low, priority_hi);
// create streams with highest and lowest available priorities
cudaStream_t st_low;
cudaStream_t st_hi;
checkCudaErrors(cudaStreamCreateWithPriority(&st_low, cudaStreamNonBlocking, priority_low));
checkCudaErrors(cudaStreamCreateWithPriority(&st_hi, cudaStreamNonBlocking, priority_hi));
size_t size;
size = TOTAL_SIZE;
// initialise host data
int *h_src_low;
int *h_src_hi;
ERR_EQ(h_src_low = (int *)malloc(size), NULL);
ERR_EQ(h_src_hi = (int *)malloc(size), NULL);
mem_init(h_src_low, size);
mem_init(h_src_hi, size);
// initialise device data
int *h_dst_low;
int *h_dst_hi;
ERR_EQ(h_dst_low = (int *)malloc(size), NULL);
ERR_EQ(h_dst_hi = (int *)malloc(size), NULL);
memset(h_dst_low, 0, size);
memset(h_dst_hi, 0, size);
// copy source data -> device
int *d_src_low;
int *d_src_hi;
checkCudaErrors(cudaMalloc(&d_src_low, size));
checkCudaErrors(cudaMalloc(&d_src_hi, size));
checkCudaErrors(cudaMemcpy(d_src_low, h_src_low, size, cudaMemcpyHostToDevice));
checkCudaErrors(cudaMemcpy(d_src_hi, h_src_hi, size, cudaMemcpyHostToDevice));
// allocate memory for memcopy destination
int *d_dst_low;
int *d_dst_hi;
checkCudaErrors(cudaMalloc(&d_dst_low, size));
checkCudaErrors(cudaMalloc(&d_dst_hi, size));
// create some events
cudaEvent_t ev_start_low;
cudaEvent_t ev_start_hi;
cudaEvent_t ev_end_low;
cudaEvent_t ev_end_hi;
checkCudaErrors(cudaEventCreate(&ev_start_low));
checkCudaErrors(cudaEventCreate(&ev_start_hi));
checkCudaErrors(cudaEventCreate(&ev_end_low));
checkCudaErrors(cudaEventCreate(&ev_end_hi));
/* */
// call pair of kernels repeatedly (with different priority streams)
checkCudaErrors(cudaEventRecord(ev_start_low, st_low));
checkCudaErrors(cudaEventRecord(ev_start_hi, st_hi));
for (int i = 0; i < TOTAL_SIZE; i += EACH_SIZE) {
int j = i / sizeof(int);
memcpy_kernel<<<TBLOCKS, THREADS, 0, st_low>>>(d_dst_low + j, d_src_low + j, EACH_SIZE);
memcpy_kernel<<<TBLOCKS, THREADS, 0, st_hi>>>(d_dst_hi + j, d_src_hi + j, EACH_SIZE);
}
checkCudaErrors(cudaEventRecord(ev_end_low, st_low));
checkCudaErrors(cudaEventRecord(ev_end_hi, st_hi));
checkCudaErrors(cudaEventSynchronize(ev_end_low));
checkCudaErrors(cudaEventSynchronize(ev_end_hi));
/* */
size = TOTAL_SIZE;
checkCudaErrors(cudaMemcpy(h_dst_low, d_dst_low, size, cudaMemcpyDeviceToHost));
checkCudaErrors(cudaMemcpy(h_dst_hi, d_dst_hi, size, cudaMemcpyDeviceToHost));
// check results of kernels
ERR_NE(memcmp(h_dst_low, h_src_low, size), 0);
ERR_NE(memcmp(h_dst_hi, h_src_hi, size), 0);
// check timings
float ms_low;
float ms_hi;
checkCudaErrors(cudaEventElapsedTime(&ms_low, ev_start_low, ev_end_low));
checkCudaErrors(cudaEventElapsedTime(&ms_hi, ev_start_hi, ev_end_hi));
printf("elapsed time of kernels launched to LOW priority stream: %.3lf ms\n", ms_low);
printf("elapsed time of kernels launched to HI priority stream: %.3lf ms\n", ms_hi);
exit(EXIT_SUCCESS);
}