IntelliSense 对于“使用命名空间 nvcuda”显示“名称必须是命名空间名称”

Question

大家好，实际上我正在 Cuda 上进行编程，并且正在测试一个简单的张量核心示例，但是我在智能感知方面遇到了问题，实际上它在这个命令上显示了错误（参见图片），我不知道为什么，因为当我编译并运行程序（带任务）其工作正常，一些想法？

错误：

完整代码：

#include <stdio.h>
#include <stdlib.h>
#include <cuda.h>
#include <mma.h>

using namespace nvcuda;

// The only dimensions currently supported by WMMA
const int WMMA_M = 16;
const int WMMA_N = 16;
const int WMMA_K = 16;

__global__ void wmma_example(half *a, half *b, float *c, 
                             int M, int N, int K, 
                             float alpha, float beta) 
{

    // Leading dimensions. Packed with no transpositions.
    int lda = M;
    int ldb = K;
    int ldc = M;
    
    // Tile using a 2D grid
    int warpM = (blockIdx.x * blockDim.x + threadIdx.x) / warpSize;
    int warpN = (blockIdx.y * blockDim.y + threadIdx.y);

    // Declare the fragments
    wmma::fragment<wmma::matrix_a, WMMA_M, WMMA_N, WMMA_K, half, wmma::col_major> a_frag;
    wmma::fragment<wmma::matrix_b, WMMA_M, WMMA_N, WMMA_K, half, wmma::col_major> b_frag;
    wmma::fragment<wmma::accumulator, WMMA_M, WMMA_N, WMMA_K, float> acc_frag;
    wmma::fragment<wmma::accumulator, WMMA_M, WMMA_N, WMMA_K, float> c_frag;

    wmma::fill_fragment(acc_frag, 0.0f);

    // Loop over the K-dimension
    for (int i = 0; i < K; i += WMMA_K) {
        int aRow = warpM * WMMA_M;
        int aCol = i;
        int bRow = i;
        int bCol = warpN * WMMA_N;
        
        // Bounds checking
        if (aRow < M && aCol < K && bRow < K && bCol < N) {
            // Load the inputs
            wmma::load_matrix_sync(a_frag, a + aRow + aCol * lda, lda);
            wmma::load_matrix_sync(b_frag, b + bRow + bCol * ldb, ldb);

            // Perform the matrix multiplication
            wmma::mma_sync(acc_frag, a_frag, b_frag, acc_frag);
        }
    }

    // Load in current value of c, scale by beta, and add to result scaled by alpha
    int cRow = warpM * WMMA_M;
    int cCol = warpN * WMMA_N;
    
    if (cRow < M && cCol < N) {
        wmma::load_matrix_sync(c_frag, c + cRow + cCol * ldc, ldc, wmma::mem_col_major);
        
        for(int i=0; i < c_frag.num_elements; i++) {
            c_frag.x[i] = alpha * acc_frag.x[i] + beta * c_frag.x[i];
        }

                // Store the output
        wmma::store_matrix_sync(c + cRow + cCol * ldc, c_frag, ldc, wmma::mem_col_major);
    }

}

Answer 1

在 c_cpp_properties 中设置 CUDA_ARCH 变量如下：

{
    "configurations": [
        {
            "name": "Linux",
            "includePath": [
                "${workspaceFolder}/**"
            ],
            "defines": [
                "__CUDA_ARCH__=750"
            ],
            "compilerPath": "/usr/bin/gcc",
            "cStandard": "gnu17",
            "cppStandard": "gnu++17",
            "intelliSenseMode": "linux-gcc-x64"
        }
    ],
    "version": 4
}