Ptx程序集由于错误而中止

Ptx assembly aborted due to errors

本文关键字：错误 程序集 于错误 Ptx      更新时间：2023-10-16

我曾尝试在MS Visual Studio 2012中的C/CUDA 5.5上编写CSR矩阵和向量相乘的程序，但遇到了ptx错误。我的代码清单：

__global__ void multKernelSymm(double* s, double* u, double* val, int* rowPtr, int* colInd)
    {
        int l = 0, jl, i;
        int idx = blockDim.x*blockIdx.x;
        l = rowPtr[idx] - 1;
        for ( i = 0; i < (rowPtr[idx + 1] - rowPtr[idx]); i++){
            jl = colInd[l] - 1;
            s[idx] = s[idx] + val[l] * u[jl];
            l+=1;
        }
        l = 0;
        l = rowPtr[idx] - 1;
        for (int i = 0; i < (rowPtr[idx + 1] - rowPtr[idx]); i++){
            jl = colInd[l] - 1;
            if(jl > idx)
                atomicAdd(&s[jl], val[l] * u[idx]);
                //s[jl] = s[jl] + val[l] * u[idx];
            l+=1;
        }

    }
    __device__ double atomicAdd(double* address, double val)
    {
         unsigned long long int* address_as_ull =
         (unsigned long long int*)address;
         unsigned long long int old = *address_as_ull, assumed;
         do {
            assumed = old;
            old = atomicCAS(address_as_ull, assumed, __double_as_longlong(val + __longlong_as_double(assumed)));
         } while (assumed != old);
         return __longlong_as_double(old);
    }

我在VS:中得到的

 C:UsersДмитрийDocumentsVisual Studio 2012Projectsmatrix mult simplematrix mult simple>"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv6.0binnvcc.exe" -gencode=arch=compute_20,code="sm_20,compute_20" --use-local-env --cl-version 2012 -ccbin "C:Program Files (x86)Microsoft Visual Studio 11.0VCbinx86_amd64"  -I"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv6.0include" -I"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv6.0include"  -G   --keep-dir x64Debug -maxrregcount=0  --machine 64 --compile -cudart static  -g   -DWIN64 -D_DEBUG -D_CONSOLE -D_MBCS -Xcompiler "/EHsc /W3 /nologo /Od /Zi /RTC1 /MDd  " -o x64Debugkernel.cu.obj "C:UsersДмитрийDocumentsVisual Studio 2012Projectsmatrix mult simplematrix mult simplekernel.cu" 
1>  ptxas C:/Users/AACE~1/AppData/Local/Temp/tmpxft_00001b88_00000000-4_kernel.ptx, line 479; : fatal error : Parsing error near '-': syntax error
1>ptxas : fatal error : Ptx assembly aborted due to errors
1>  kernel.cu
1>C:Program Files (x86)MSBuildMicrosoft.Cppv4.0V110BuildCustomizationsCUDA 6.0.targets(597,9): error MSB3721: The command ""C:Program FilesNVIDIA GPU Computing ToolkitCUDAv6.0binnvcc.exe" -gencode=arch=compute_20,code="sm_20,compute_20" --use-local-env --cl-version 2012 -ccbin "C:Program Files (x86)Microsoft Visual Studio 11.0VCbinx86_amd64"  -I"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv6.0include" -I"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv6.0include"  -G   --keep-dir x64Debug -maxrregcount=0  --machine 64 --compile -cudart static  -g   -DWIN64 -D_DEBUG -D_CONSOLE -D_MBCS -Xcompiler "/EHsc /W3 /nologo /Od /Zi /RTC1 /MDd  " -o x64Debugkernel.cu.obj "C:UsersДмитрийDocumentsVisual Studio 2012Projectsmatrix mult simplematrix mult simplekernel.cu"" exited with code -1.
1>
 >     ========== Build: 0 succeeded, 1 failed, 0 up-to-date, 0 skipped ==========

我能修吗？

我通过做3个更改来编译您的代码：

  1. 如@PaulR所示，在调用自定义atomicAdd函数时，请删除double关键字

  2. atomicAdd的第一个参数应该是指向要更新的地址（要将值添加到的地址）的指针。所以你的函数调用应该是这样的：

    atomicAdd(&(s[jl]), val[l] * u[idx]);

  3. 最后，将自定义atomicAdd函数的声明和定义移动到内核定义之前的。

通过这些更改，我可以编译您的代码。

相关文章：