mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2026-10-13 00:08:26 +08:00
CUDA 13.4 samples update - v13.4-public
Release 13.4 of the CUDA Samples supported by CUDA Toolkit 13.4. See Changelog for more information.
This commit is contained in:
Dheemanth
GitHub
@@ -1,18 +0,0 @@
|
||||
{
|
||||
"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
|
||||
}
|
||||
@@ -1,7 +0,0 @@
|
||||
{
|
||||
"recommendations": [
|
||||
"nvidia.nsight-vscode-edition",
|
||||
"ms-vscode.cpptools",
|
||||
"ms-vscode.makefile-tools"
|
||||
]
|
||||
}
|
||||
@@ -1,34 +1,25 @@
|
||||
cmake_minimum_required(VERSION 3.20)
|
||||
|
||||
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Modules")
|
||||
|
||||
include("${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/DetectCudaArch.cmake")
|
||||
|
||||
if(SAMPLE_SKIP_BUILD)
|
||||
return()
|
||||
endif()
|
||||
|
||||
project(vectorAdd LANGUAGES C CXX CUDA)
|
||||
|
||||
include("${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/CudaSampleCommon.cmake")
|
||||
|
||||
find_package(CUDAToolkit REQUIRED)
|
||||
|
||||
set(CMAKE_POSITION_INDEPENDENT_CODE ON)
|
||||
|
||||
set(CMAKE_CUDA_ARCHITECTURES 75 80 86 87 89 90 100 110 120)
|
||||
set(CMAKE_CUDA_FLAGS "${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)
|
||||
|
||||
# Source file
|
||||
# Add target for vectorAdd
|
||||
add_executable(vectorAdd vectorAdd.cu)
|
||||
|
||||
target_compile_options(vectorAdd PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>)
|
||||
|
||||
target_compile_features(vectorAdd PRIVATE cxx_std_17 cuda_std_17)
|
||||
|
||||
set_target_properties(vectorAdd PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
|
||||
|
||||
# Include installation configuration
|
||||
include(${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/InstallSamples.cmake)
|
||||
include(InstallSamples)
|
||||
setup_samples_install()
|
||||
|
||||
@@ -1,32 +1,82 @@
|
||||
# vectorAdd - Vector Addition
|
||||
# Sample: Vector Addition (Unified Memory)
|
||||
|
||||
## Description
|
||||
|
||||
This CUDA Runtime API sample is a very basic sample that implements element by element vector addition. It is the same as the sample illustrating Chapter 3 of the programming guide with some additions like error checking.
|
||||
A first-look CUDA sample: add two float vectors element-wise on the GPU. This version uses **Unified Memory** (`cudaMallocManaged`) so the same pointer is accessible from both host and device, eliminating the need for explicit `cudaMemcpy` calls. The GPU result is verified against a CPU reference computed on the same buffers.
|
||||
|
||||
The sample mirrors the corresponding Unified Memory example in the CUDA Programming Guide and uses the `//unified-memory-begin` / `//unified-memory-end` markers so the guide can extract the snippet directly.
|
||||
|
||||
## What You'll Learn
|
||||
|
||||
- Allocating memory accessible to both host and device with `cudaMallocManaged`
|
||||
- Launching a 1D CUDA kernel that operates on a Unified Memory buffer with no explicit transfers
|
||||
- Computing grid dimensions with `cuda::ceil_div` (from `<cuda/cmath>`)
|
||||
- Synchronizing host execution with device work via `cudaDeviceSynchronize`
|
||||
- Validating GPU output against a serial CPU reference
|
||||
|
||||
## Key Concepts
|
||||
|
||||
CUDA Runtime API, Vector Addition
|
||||
- **Unified Memory** — one pointer for both CPU and GPU; the CUDA runtime migrates pages on demand
|
||||
- **1D thread indexing** — `threadIdx.x + blockIdx.x * blockDim.x`
|
||||
- **Bounds checking** in the kernel so the vector length need not be a multiple of the block size
|
||||
|
||||
## Supported SM Architectures
|
||||
## Key APIs
|
||||
|
||||
[SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 5.3 ](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) [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)
|
||||
### CUDA Runtime
|
||||
- `cudaMallocManaged` — allocate Unified Memory accessible to both host and device
|
||||
- `cudaDeviceSynchronize` — block the host until all submitted device work completes
|
||||
- `cudaFree` — release a Unified Memory allocation
|
||||
|
||||
## Supported OSes
|
||||
### libcudacxx
|
||||
- `cuda::ceil_div` — `ceil(a / b)` used to compute the number of blocks; from `<cuda/cmath>`
|
||||
|
||||
Linux, Windows
|
||||
## Requirements
|
||||
|
||||
## Supported CPU Architecture
|
||||
### Hardware
|
||||
- NVIDIA GPU with Compute Capability 7.5 or higher
|
||||
|
||||
x86_64, armv7l
|
||||
### Software
|
||||
- CMake 3.20 or newer
|
||||
- A C++17-capable host compiler
|
||||
|
||||
## CUDA APIs involved
|
||||
## How to Build
|
||||
|
||||
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
|
||||
cudaMemcpy, cudaGetErrorString, cudaFree, cudaGetLastError, cudaMalloc
|
||||
See the [top-level README](../../../README.md#building-cuda-samples) for full build instructions, including how to build all samples or a single sample standalone.
|
||||
|
||||
## Prerequisites
|
||||
## How to Run
|
||||
|
||||
Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
### Basic usage (default vector length = 1024)
|
||||
```bash
|
||||
./vectorAdd
|
||||
```
|
||||
|
||||
### Custom vector length
|
||||
```bash
|
||||
./vectorAdd 1000000
|
||||
```
|
||||
|
||||
## Expected Output
|
||||
|
||||
On success:
|
||||
|
||||
```text
|
||||
Unified Memory: CPU and GPU answers match
|
||||
```
|
||||
|
||||
On failure, the first mismatching index and the two diverging values are printed, followed by an error line:
|
||||
|
||||
```text
|
||||
Index 0 mismatch: 0.123456 != 0.123400
|
||||
Unified Memory: Error - CPU and GPU answers do not match
|
||||
```
|
||||
|
||||
## Files
|
||||
|
||||
- `vectorAdd.cu` — Unified Memory vector addition implementation (kernel + host driver)
|
||||
- `README.md` — this file
|
||||
- `CMakeLists.txt` — build configuration
|
||||
|
||||
## See Also
|
||||
|
||||
- [CUDA C++ Programming Guide — vectorAdd example with Unified Memory](https://docs.nvidia.com/cuda/cuda-programming-guide/02-basics/intro-to-cuda-cpp.html)
|
||||
|
||||
## References (for more details)
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
/* Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
|
||||
/* Copyright (c) 2026, NVIDIA CORPORATION. All rights reserved.
|
||||
*
|
||||
* Redistribution and use in source and binary forms, with or without
|
||||
* modification, are permitted provided that the following conditions
|
||||
@@ -33,164 +33,112 @@
|
||||
* of the programming guide with some additions like error checking.
|
||||
*/
|
||||
|
||||
#include <cuda_runtime_api.h>
|
||||
#include <memory.h>
|
||||
#include <cstdlib>
|
||||
#include <ctime>
|
||||
#include <stdio.h>
|
||||
|
||||
// For the CUDA runtime routines (prefixed with "cuda_")
|
||||
#include <cuda_runtime.h>
|
||||
#include <helper_cuda.h>
|
||||
#include <cuda/cmath>
|
||||
/**
|
||||
* CUDA Kernel Device code
|
||||
*
|
||||
* Computes the vector addition of A and B into C. The 3 vectors have the same
|
||||
* number of elements numElements.
|
||||
* number of elements. This exmple shows Vector addition using Unified memory.
|
||||
*/
|
||||
__global__ void vectorAdd(const float *A, const float *B, float *C, int numElements)
|
||||
{
|
||||
int i = blockDim.x * blockIdx.x + threadIdx.x;
|
||||
|
||||
if (i < numElements) {
|
||||
C[i] = A[i] + B[i] + 0.0f;
|
||||
|
||||
__global__ void vecAdd(float* A, float* B, float* C, int vectorLength)
|
||||
{
|
||||
int workIndex = threadIdx.x + blockIdx.x*blockDim.x;
|
||||
if(workIndex < vectorLength)
|
||||
{
|
||||
C[workIndex] = A[workIndex] + B[workIndex];
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Host main routine
|
||||
*/
|
||||
int main(void)
|
||||
void initArray(float* A, int length)
|
||||
{
|
||||
// Error code to check return values for CUDA calls
|
||||
cudaError_t err = cudaSuccess;
|
||||
|
||||
// Print the vector length to be used, and compute its size
|
||||
int numElements = 50000;
|
||||
size_t size = numElements * sizeof(float);
|
||||
printf("[Vector addition of %d elements]\n", numElements);
|
||||
|
||||
// Allocate the host input vector A
|
||||
float *h_A = (float *)malloc(size);
|
||||
|
||||
// Allocate the host input vector B
|
||||
float *h_B = (float *)malloc(size);
|
||||
|
||||
// Allocate the host output vector C
|
||||
float *h_C = (float *)malloc(size);
|
||||
|
||||
// Verify that allocations succeeded
|
||||
if (h_A == NULL || h_B == NULL || h_C == NULL) {
|
||||
fprintf(stderr, "Failed to allocate host vectors!\n");
|
||||
exit(EXIT_FAILURE);
|
||||
std::srand(std::time({}));
|
||||
for(int i=0; i<length; i++)
|
||||
{
|
||||
A[i] = rand() / (float)RAND_MAX;
|
||||
}
|
||||
}
|
||||
|
||||
// Initialize the host input vectors
|
||||
for (int i = 0; i < numElements; ++i) {
|
||||
h_A[i] = rand() / (float)RAND_MAX;
|
||||
h_B[i] = rand() / (float)RAND_MAX;
|
||||
void serialVecAdd(float* A, float* B, float* C, int length)
|
||||
{
|
||||
for(int i=0; i<length; i++)
|
||||
{
|
||||
C[i] = A[i] + B[i];
|
||||
}
|
||||
}
|
||||
|
||||
// Allocate the device input vector A
|
||||
float *d_A = NULL;
|
||||
err = cudaMalloc((void **)&d_A, size);
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to allocate device vector A (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// Allocate the device input vector B
|
||||
float *d_B = NULL;
|
||||
err = cudaMalloc((void **)&d_B, size);
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to allocate device vector B (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// Allocate the device output vector C
|
||||
float *d_C = NULL;
|
||||
err = cudaMalloc((void **)&d_C, size);
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to allocate device vector C (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// Copy the host input vectors A and B in host memory to the device input
|
||||
// vectors in
|
||||
// device memory
|
||||
printf("Copy input data from the host memory to the CUDA device\n");
|
||||
err = cudaMemcpy(d_A, h_A, size, cudaMemcpyHostToDevice);
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to copy vector A from host to device (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
err = cudaMemcpy(d_B, h_B, size, cudaMemcpyHostToDevice);
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to copy vector B from host to device (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// Launch the Vector Add CUDA Kernel
|
||||
int threadsPerBlock = 256;
|
||||
int blocksPerGrid = (numElements + threadsPerBlock - 1) / threadsPerBlock;
|
||||
printf("CUDA kernel launch with %d blocks of %d threads\n", blocksPerGrid, threadsPerBlock);
|
||||
vectorAdd<<<blocksPerGrid, threadsPerBlock>>>(d_A, d_B, d_C, numElements);
|
||||
err = cudaGetLastError();
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to launch vectorAdd kernel (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// Copy the device result vector in device memory to the host result vector
|
||||
// in host memory.
|
||||
printf("Copy output data from the CUDA device to the host memory\n");
|
||||
err = cudaMemcpy(h_C, d_C, size, cudaMemcpyDeviceToHost);
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to copy vector C from device to host (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// Verify that the result vector is correct
|
||||
for (int i = 0; i < numElements; ++i) {
|
||||
if (fabs(h_A[i] + h_B[i] - h_C[i]) > 1e-5) {
|
||||
fprintf(stderr, "Result verification failed at element %d!\n", i);
|
||||
exit(EXIT_FAILURE);
|
||||
bool vectorApproximatelyEqual(float* A, float* B, int length, float epsilon=0.00001)
|
||||
{
|
||||
for(int i=0; i<length; i++)
|
||||
{
|
||||
if(fabs(A[i] -B[i]) > epsilon)
|
||||
{
|
||||
printf("Index %d mismatch: %f != %f", i, A[i], B[i]);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
printf("Test PASSED\n");
|
||||
|
||||
// Free device global memory
|
||||
err = cudaFree(d_A);
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to free device vector A (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
int main(int argc, char** argv)
|
||||
{
|
||||
int vectorLength = 1024;
|
||||
if(argc >=2)
|
||||
{
|
||||
vectorLength = std::atoi(argv[1]);
|
||||
}
|
||||
|
||||
err = cudaFree(d_B);
|
||||
//unified-memory-example-begin
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to free device vector B (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
// Pointers to memory vectors
|
||||
float* A = nullptr;
|
||||
float* B = nullptr;
|
||||
float* C = nullptr;
|
||||
float* comparisonResult = (float*)malloc(vectorLength*sizeof(float));
|
||||
|
||||
// Use unified memory to allocate buffers
|
||||
cudaMallocManaged(&A, vectorLength*sizeof(float));
|
||||
cudaMallocManaged(&B, vectorLength*sizeof(float));
|
||||
cudaMallocManaged(&C, vectorLength*sizeof(float));
|
||||
|
||||
// Initialize vectors on the host
|
||||
initArray(A, vectorLength);
|
||||
initArray(B, vectorLength);
|
||||
|
||||
// Launch the kernel. Unified memory will make sure A, B, and C are
|
||||
// accessible to the GPU
|
||||
int threads = 256;
|
||||
int blocks = cuda::ceil_div(vectorLength, threads);
|
||||
vecAdd<<<blocks, threads>>>(A, B, C, vectorLength);
|
||||
// Wait for the kernel to complete execution
|
||||
cudaDeviceSynchronize();
|
||||
|
||||
// Perform computation serially on CPU for comparison
|
||||
serialVecAdd(A, B, comparisonResult, vectorLength);
|
||||
|
||||
// Confirm that CPU and GPU got the same answer
|
||||
if(vectorApproximatelyEqual(C, comparisonResult, vectorLength))
|
||||
{
|
||||
printf("Unified Memory: CPU and GPU answers match\n");
|
||||
}
|
||||
else
|
||||
{
|
||||
printf("Unified Memory: Error - CPU and GPU answers do not match\n");
|
||||
}
|
||||
|
||||
err = cudaFree(d_C);
|
||||
// Clean Up
|
||||
cudaFree(A);
|
||||
cudaFree(B);
|
||||
cudaFree(C);
|
||||
free(comparisonResult);
|
||||
|
||||
if (err != cudaSuccess) {
|
||||
fprintf(stderr, "Failed to free device vector C (error code %s)!\n", cudaGetErrorString(err));
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
//unified-memory-example-end
|
||||
|
||||
// Free host memory
|
||||
free(h_A);
|
||||
free(h_B);
|
||||
free(h_C);
|
||||
|
||||
printf("Done\n");
|
||||
return 0;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user