0

異種並列プログラミングでやるべき宿題があります。コードは教職員によって書かれたもので、私たちの義務は でマークされた領域を埋めること//@@です。コードは、CUDA C を使用して 2 つのベクトルを追加することになっています。以下の解決策を試しました。プログラムはエラーなしで実行されますが、コードの出力が期待される結果と一致しないというフィードバックが返されます。必要だと思われるものを追加した後のコードは次のとおりです。 

// MP 1
#include    <wb.h>

__global__ void vecAdd(float* in1, float* in2, float* out, int len) {
//@@ Insert code to implement vector addition here
int i = threadIdx.x + blockDim.x * blockIdx.x;
if (i < len ) out[i] = in1[i] + in2[i]; 
}



int main(int argc, char ** argv) {
wbArg_t args;
int inputLength;
float * hostInput1;
float * hostInput2;
float * hostOutput;
float * deviceInput1;
float * deviceInput2;
float * deviceOutput;
//@@ i added ######
int size = inputLength*sizeof(float);
//@@ ########
args = wbArg_read(argc, argv);

wbTime_start(Generic, "Importing data and creating memory on host");
hostInput1 = (float *) wbImport(wbArg_getInputFile(args, 0), &inputLength);
hostInput2 = (float *) wbImport(wbArg_getInputFile(args, 1), &inputLength);
hostOutput = (float *) malloc(inputLength * sizeof(float));

wbTime_stop(Generic, "Importing data and creating memory on host");

wbLog(TRACE, "The input length is ", inputLength);

wbTime_start(GPU, "Allocating GPU memory.");
//@@ Allocate GPU memory here

cudaMalloc((void**)&deviceInput1 , size);
cudaMalloc((void**)&deviceInput2 , size);
cudaMalloc((void**)&deviceOutput , size);
wbTime_stop(GPU, "Allocating GPU memory.");

wbTime_start(GPU, "Copying input memory to the GPU.");
//@@ Copy memory to the GPU here
cudaMemcpy(deviceInput1, hostInput1, size, cudaMemcpyHostToDevice);
cudaMemcpy(deviceInput2, hostInput2, size, cudaMemcpyHostToDevice);
wbTime_stop(GPU, "Copying input memory to the GPU.");

//@@ Initialize the grid and block dimensions here  
dim3 DimGrid((inputLength -1)/256 +1 , 1 , 1);
dim3 DimBlock(256 , 1, 1); 

wbTime_start(Compute, "Performing CUDA computation");
//@@ Launch the GPU Kernel here     
vecAdd<<<DimGrid , DimBlock>>>(deviceInput1 , deviceInput2 , deviceOutput , inputLength); 
cudaThreadSynchronize();
wbTime_stop(Compute, "Performing CUDA computation");

wbTime_start(Copy, "Copying output memory to the CPU");
//@@ Copy the GPU memory back to the CPU here
cudaMemcpy(hostOutput, deviceOutput, size , cudaMemcpyDeviceToHost);
wbTime_stop(Copy, "Copying output memory to the CPU");

wbTime_start(GPU, "Freeing GPU Memory");
//@@ Free the GPU memory here
free(deviceInput1);
free(deviceInput2);
free(deviceOutput);

wbTime_stop(GPU, "Freeing GPU Memory");

wbSolution(args, hostOutput, inputLength);

free(hostInput1);
free(hostInput2);
free(hostOutput);

return 0;
}
4

2 に答える 2

2

inputLength変数が適切な値を取得した場所にコードを移動します。これを変える：

//@@ i added ######
int size = inputLength*sizeof(float);
//@@ ########
args = wbArg_read(argc, argv);

wbTime_start(Generic, "Importing data and creating memory on host");
hostInput1 = (float *) wbImport(wbArg_getInputFile(args, 0), &inputLength);
hostInput2 = (float *) wbImport(wbArg_getInputFile(args, 1), &inputLength);
hostOutput = (float *) malloc(inputLength * sizeof(float));

これに：

args = wbArg_read(argc, argv);

wbTime_start(Generic, "Importing data and creating memory on host");
hostInput1 = (float *) wbImport(wbArg_getInputFile(args, 0), &inputLength);
hostInput2 = (float *) wbImport(wbArg_getInputFile(args, 1), &inputLength);
hostOutput = (float *) malloc(inputLength * sizeof(float));

//@@ i added ######
int size = inputLength*sizeof(float);
//@@ ########

さらに、コメントでタロンミーによって提案されたことを行います。

于 2012-12-09T17:22:55.593 に答える
1

talonmies と ahmad の両方に感謝します。どちらも私にとって有効な正しい答えを得るのに役立ちました。完全な答え (誰が興味深いか) は次のとおりです。 

// MP 1
#include    <wb.h>

__global__ void vecAdd(float* in1, float* in2, float* out, int len) {
int i = threadIdx.x + blockDim.x * blockIdx.x;
if (i < len ) out[i] = in1[i] + in2[i];
 }



int main(int argc, char ** argv) {
wbArg_t args;
int inputLength;
float * hostInput1;
float * hostInput2;
float * hostOutput;
float * deviceInput1;
float * deviceInput2;
float * deviceOutput;


args = wbArg_read(argc, argv);

wbTime_start(Generic, "Importing data and creating memory on host");
hostInput1 = (float *) wbImport(wbArg_getInputFile(args, 0), &inputLength);
hostInput2 = (float *) wbImport(wbArg_getInputFile(args, 1), &inputLength);
hostOutput = (float *) malloc(inputLength * sizeof(float));
int size = inputLength*sizeof(float);

wbTime_stop(Generic, "Importing data and creating memory on host");

wbLog(TRACE, "The input length is ", inputLength);

wbTime_start(GPU, "Allocating GPU memory.");


cudaMalloc((void**)&deviceInput1 , size);
cudaMalloc((void**)&deviceInput2 , size);
cudaMalloc((void**)&deviceOutput , size);
wbTime_stop(GPU, "Allocating GPU memory.");

wbTime_start(GPU, "Copying input memory to the GPU.");

cudaMemcpy(deviceInput1, hostInput1, size, cudaMemcpyHostToDevice);
cudaMemcpy(deviceInput2, hostInput2, size, cudaMemcpyHostToDevice);
wbTime_stop(GPU, "Copying input memory to the GPU.");


dim3 DimGrid((inputLength -1)/256 +1 , 1 , 1);
dim3 DimBlock(256 , 1, 1); 

wbTime_start(Compute, "Performing CUDA computation");
//@@ Launch the GPU Kernel 

vecAdd<<<DimGrid , DimBlock>>>(deviceInput1 , deviceInput2 , deviceOutput ,  inputLength);

cudaThreadSynchronize();
wbTime_stop(Compute, "Performing CUDA computation");

wbTime_start(Copy, "Copying output memory to the CPU");
//@@ Copy the GPU memory back to the CPU 
cudaMemcpy(hostOutput, deviceOutput, size , cudaMemcpyDeviceToHost);
wbTime_stop(Copy, "Copying output memory to the CPU");

wbTime_start(GPU, "Freeing GPU Memory");
//@@ Free the GPU memory 
cudaFree(deviceInput1);
cudaFree(deviceInput2);
cudaFree(deviceOutput);

wbTime_stop(GPU, "Freeing GPU Memory");

wbSolution(args, hostOutput, inputLength);

free(hostInput1);
free(hostInput2);
free(hostOutput);

return 0;
}
于 2012-12-12T18:38:30.487 に答える