我正在尝试复制matlab fft功能，它在矩阵中逐行（或逐列）地执行。每一行都是袖口计划中的一个批次。
我可以使用cufftExecC2C（下面的代码中注释掉的部分可以使用），但不能使用cufftExecR2C。我的代码使用的是cufftPlan1d，但理想情况下我想用cufftPlanMany实现它。
我想知道我做错了什么，如果有更好的方法可以做到这一点。谢谢。
// linker -> input -> additional dependencies -> add 'cufft.lib'
// VC++ Directories -> include directories - > add 'C:\ProgramData\NVIDIA Corporation\CUDA Samples\v6.0\common\inc'
#include <stdio.h>
#include <stdlib.h>
#include <cufft.h>
#include <cuda_runtime.h>
#include <iostream>
#define NX 6
#define NY 5
void printArray(float *my_array);
void printComplexArray(float2 *my_array);
int main(){
/************************************************************ C2C ************************************************************/
/*
float2 *initial_array = (float2 *)malloc(sizeof(float2) * NX * NY);
for (int h = 0; h < NX; h++){
for (int w = 0; w < NY; w++){
initial_array[NY * h + w].x = 0;
initial_array[NY * h + w].y = 0;
}
}
initial_array[NY*3 + 0].x = 1;
initial_array[NY*5 + 0].x = 1;
printComplexArray(initial_array);
float2 *transformed_array= (float2 *)malloc(sizeof(float2) * NX * NY);
cufftComplex *gpu_initial_array;
cufftComplex *gpu_transformed_array;
cudaMalloc((void **)&gpu_initial_array, NX*NY*sizeof(cufftComplex));
cudaMalloc((void **)&gpu_transformed_array, NX*NY*sizeof(cufftComplex));
cudaMemcpy(gpu_initial_array, initial_array, NX*NY*sizeof(float2), cudaMemcpyHostToDevice);
cufftHandle plan;
cufftPlan1d(&plan, NY, CUFFT_C2C, NX);
cufftExecC2C(plan, gpu_initial_array, gpu_transformed_array, CUFFT_FORWARD);
cudaMemcpy(transformed_array, gpu_transformed_array, NX*NY*sizeof(cufftComplex), cudaMemcpyDeviceToHost);
printComplexArray(transformed_array);
*/
/************************************************************ C2C ************************************************************/
/************************************************************ R2C ************************************************************/
float *initial_array = (float *)malloc(sizeof(float) * NX * NY);
for (int h = 0; h < NX; h++){
for (int w = 0; w < NY; w++)
initial_array[NY * h + w] = 0;
}
initial_array[NY*3 + 0] = 1;
printArray(initial_array);
float2 *transformed_array= (float2 *)malloc(sizeof(float2) * (NY/2+1) * NX);
cufftReal *gpu_initial_array;
cufftComplex *gpu_transformed_array;
cudaMalloc((void **)&gpu_initial_array, NX*NY*sizeof(cufftReal));
cudaMalloc((void **)&gpu_transformed_array, (NY/2+1)*NX*sizeof(cufftComplex));
cudaMemcpy(gpu_initial_array, initial_array, NX*NY*sizeof(float), cudaMemcpyHostToDevice);
cufftHandle plan;
cufftPlan1d(&plan, NY, CUFFT_R2C, NX);
// ***** cufftPlanMany *****
//int n[2] = {NX, NY};
//cufftPlanMany(&plan,1,n,NULL,1,0,NULL,1,0,CUFFT_R2C,NX);
cufftExecR2C(plan, gpu_initial_array, gpu_transformed_array);
cudaMemcpy(transformed_array, gpu_transformed_array, NX*(NY/2+1)*sizeof(cufftComplex), cudaMemcpyDeviceToHost);
printComplexArray(transformed_array);
/************************************************************ R2C ************************************************************/
cufftDestroy(plan);
free(initial_array);
free(transformed_array);
cudaFree(gpu_initial_array);
cudaFree(gpu_transformed_array);
std::system("pause");
return 0;
}
void printArray(float *my_array){
for (int h = 0; h < NX; h++){
for (int w = 0; w < NY; w++)
std::cout << my_array[NY * h + w] << " | ";
std::cout << std::endl;
}
std::cout << std::endl;
}
void printComplexArray(float2 *my_array){
for (int h = 0; h < NX; h++){
for (int w = 0; w < NY; w++)
std::cout << my_array[NY * h + w].x << " + " << my_array[NY * h + w].y << " | ";
std::cout << std::endl;
}
std::cout << std::endl;
}
答案 0 :(得分：3)
您的问题似乎与您打印结果的方式有关。对于CUFFT_R2C和CUFFT_C2C这两种情况，您无法使用相同的例程进行打印。在前一种情况下，您有一个(NY/2+1)*NX大小的输出，而在后一种情况下，您有一个NY*NX大小的输出。下面的固定代码应该有效。
此外，添加适当的CUDA error check和CUFFT error check也很不错，我也将其添加到下面的代码中。
#include <stdio.h>
#include <stdlib.h>
#include <cufft.h>
#include <cuda_runtime.h>
#include <assert.h>
#include <iostream>
#define NX 6
#define NY 5
void printArray(float *my_array);
void printComplexSymmetricArray(float2 *my_array);
/********************/
/* CUDA ERROR CHECK */
/********************/
#define gpuErrchk(ans) { gpuAssert((ans), __FILE__, __LINE__); }
inline void gpuAssert(cudaError_t code, char *file, int line, bool abort=true)
{
if (code != cudaSuccess)
{
fprintf(stderr,"GPUassert: %s %s %d\n", cudaGetErrorString(code), file, line);
if (abort) exit(code);
}
}
/*********************/
/* CUFFT ERROR CHECK */
/*********************/
static const char *_cudaGetErrorEnum(cufftResult error)
{
switch (error)
{
case CUFFT_SUCCESS:
return "CUFFT_SUCCESS";
case CUFFT_INVALID_PLAN:
return "CUFFT_INVALID_PLAN";
case CUFFT_ALLOC_FAILED:
return "CUFFT_ALLOC_FAILED";
case CUFFT_INVALID_TYPE:
return "CUFFT_INVALID_TYPE";
case CUFFT_INVALID_VALUE:
return "CUFFT_INVALID_VALUE";
case CUFFT_INTERNAL_ERROR:
return "CUFFT_INTERNAL_ERROR";
case CUFFT_EXEC_FAILED:
return "CUFFT_EXEC_FAILED";
case CUFFT_SETUP_FAILED:
return "CUFFT_SETUP_FAILED";
case CUFFT_INVALID_SIZE:
return "CUFFT_INVALID_SIZE";
case CUFFT_UNALIGNED_DATA:
return "CUFFT_UNALIGNED_DATA";
}
return "<unknown>";
}
#define cufftSafeCall(err) __cufftSafeCall(err, __FILE__, __LINE__)
inline void __cufftSafeCall(cufftResult err, const char *file, const int line)
{
if( CUFFT_SUCCESS != err) {
fprintf(stderr, "CUFFT error in file '%s', line %d\n %s\nerror %d: %s\nterminating!\n",__FILE__, __LINE__,err, \
_cudaGetErrorEnum(err)); \
cudaDeviceReset(); assert(0); \
}
}
/********/
/* MAIN */
/********/
int main(){
float *initial_array = (float *)malloc(sizeof(float) * NX * NY);
for (int h = 0; h < NX; h++){
for (int w = 0; w < NY; w++)
initial_array[NY * h + w] = 0;
}
initial_array[NY*3 + 0] = 1;
printArray(initial_array);
float2 *transformed_array= (float2 *)malloc(sizeof(float2) * (NY/2+1) * NX);
cufftReal *gpu_initial_array;
cufftComplex *gpu_transformed_array;
gpuErrchk(cudaMalloc((void **)&gpu_initial_array, NX*NY*sizeof(cufftReal)));
gpuErrchk(cudaMalloc((void **)&gpu_transformed_array, (NY/2+1)*NX*sizeof(cufftComplex)));
gpuErrchk(cudaMemcpy(gpu_initial_array, initial_array, NX*NY*sizeof(float), cudaMemcpyHostToDevice));
cufftHandle plan;
cufftSafeCall(cufftPlan1d(&plan, NY, CUFFT_R2C, NX));
cufftSafeCall(cufftExecR2C(plan, (cufftReal*)gpu_initial_array, (cufftComplex*)gpu_transformed_array));
gpuErrchk(cudaMemcpy(transformed_array, gpu_transformed_array, NX*(NY/2+1)*sizeof(cufftComplex), cudaMemcpyDeviceToHost));
printComplexSymmetricArray(transformed_array);
cufftSafeCall(cufftDestroy(plan));
free(initial_array);
free(transformed_array);
gpuErrchk(cudaFree(gpu_initial_array));
gpuErrchk(cudaFree(gpu_transformed_array));
std::system("pause");
return 0;
}
/***********************/
/* PRINTOUT REAL ARRAY */
/***********************/
void printArray(float *my_array){
for (int h = 0; h < NX; h++){
for (int w = 0; w < NY; w++)
std::cout << my_array[NY * h + w] << " | ";
std::cout << std::endl;
}
std::cout << std::endl;
}
/************************************/
/* PRINTOUT COMPLEX SYMMETRIC ARRAY */
/************************************/
void printComplexSymmetricArray(float2 *my_array){
for (int h = 0; h < NX; h++){
for (int w = 0; w < NY/2+1; w++)
std::cout << my_array[(NY/2+1) * h + w].x << " + " << my_array[(NY/2+1) * h + w].y << " | ";
std::cout << std::endl;
}
std::cout << std::endl;
}