added CUFFT errors handling into GPU module
This commit is contained in:
parent
8190837dd4
commit
a6d9cce500
@ -247,7 +247,8 @@ enum {
|
||||
CV_StsAssert= -215, /* assertion failed */
|
||||
CV_GpuNotSupported= -216,
|
||||
CV_GpuApiCallError= -217,
|
||||
CV_GpuNppCallError= -218
|
||||
CV_GpuNppCallError= -218,
|
||||
CV_GpuCufftCallError= -219
|
||||
};
|
||||
|
||||
/****************************************************************************************\
|
||||
|
@ -44,13 +44,16 @@
|
||||
#define __OPENCV_CUDA_SAFE_CALL_HPP__
|
||||
|
||||
#include "cuda_runtime_api.h"
|
||||
#include "cufft.h"
|
||||
//#include <nppdefs.h>
|
||||
|
||||
#if defined(__GNUC__)
|
||||
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__, __func__)
|
||||
#define cufftSafeCall(expr) ___cufftSafeCall(expr, __FILE__, __LINE__, __func__)
|
||||
#define nppSafeCall(expr) ___nppSafeCall(expr, __FILE__, __LINE__, __func__)
|
||||
#else /* defined(__CUDACC__) || defined(__MSVC__) */
|
||||
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__)
|
||||
#define cufftSafeCall(expr) ___cufftSafeCall(expr, __FILE__, __LINE__)
|
||||
#define nppSafeCall(expr) ___nppSafeCall(expr, __FILE__, __LINE__)
|
||||
#endif
|
||||
|
||||
@ -59,7 +62,8 @@ namespace cv
|
||||
namespace gpu
|
||||
{
|
||||
void error(const char *error_string, const char *file, const int line, const char *func = "");
|
||||
void nppError( int error, const char *file, const int line, const char *func = "");
|
||||
void nppError(int err, const char *file, const int line, const char *func = "");
|
||||
void cufftError(int err, const char *file, const int line, const char *func = "");
|
||||
|
||||
static inline void ___cudaSafeCall(cudaError_t err, const char *file, const int line, const char *func = "")
|
||||
{
|
||||
@ -67,6 +71,12 @@ namespace cv
|
||||
cv::gpu::error(cudaGetErrorString(err), file, line, func);
|
||||
}
|
||||
|
||||
static inline void ___cufftSafeCall(cufftResult_t err, const char *file, const int line, const char *func = "")
|
||||
{
|
||||
if (CUFFT_SUCCESS != err)
|
||||
cv::gpu::cufftError(err, file, line, func);
|
||||
}
|
||||
|
||||
static inline void ___nppSafeCall(int err, const char *file, const int line, const char *func = "")
|
||||
{
|
||||
if (err < 0)
|
||||
|
@ -56,6 +56,9 @@ namespace
|
||||
{
|
||||
#define error_entry(entry) { entry, #entry }
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////
|
||||
// NPP errors
|
||||
|
||||
struct NppError
|
||||
{
|
||||
int error;
|
||||
@ -116,6 +119,40 @@ namespace
|
||||
bool operator()(const NppError& e) const { return e.error == err; }
|
||||
};
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////
|
||||
// CUFFT errors
|
||||
|
||||
struct CufftError
|
||||
{
|
||||
int code;
|
||||
string message;
|
||||
};
|
||||
|
||||
const CufftError cufft_errors[] =
|
||||
{
|
||||
error_entry(CUFFT_INVALID_PLAN),
|
||||
error_entry(CUFFT_ALLOC_FAILED),
|
||||
error_entry(CUFFT_INVALID_TYPE),
|
||||
error_entry(CUFFT_INVALID_VALUE),
|
||||
error_entry(CUFFT_INTERNAL_ERROR),
|
||||
error_entry(CUFFT_EXEC_FAILED),
|
||||
error_entry(CUFFT_SETUP_FAILED),
|
||||
error_entry(CUFFT_INVALID_SIZE),
|
||||
error_entry(CUFFT_UNALIGNED_DATA)
|
||||
};
|
||||
|
||||
struct CufftErrorComparer
|
||||
{
|
||||
CufftErrorComparer(int code_): code(code_) {}
|
||||
bool operator()(const CufftError& other) const
|
||||
{
|
||||
return other.code == code;
|
||||
}
|
||||
int code;
|
||||
};
|
||||
|
||||
const int cufft_error_num = sizeof(cufft_errors) / sizeof(cufft_errors[0]);
|
||||
|
||||
}
|
||||
|
||||
namespace cv
|
||||
@ -138,6 +175,26 @@ namespace cv
|
||||
cv::error( cv::Exception(CV_GpuNppCallError, getNppErrorString(err), func, file, line) );
|
||||
}
|
||||
|
||||
const string getCufftErrorString(int err_code)
|
||||
{
|
||||
const CufftError* cufft_error = std::find_if(
|
||||
cufft_errors, cufft_errors + cufft_error_num,
|
||||
CufftErrorComparer(err_code));
|
||||
|
||||
bool found = cufft_error != cufft_errors + cufft_error_num;
|
||||
|
||||
std::stringstream ss;
|
||||
ss << (found ? cufft_error->message : "Unknown error code");
|
||||
ss << " [Code = " << err_code << "]";
|
||||
|
||||
return ss.str();
|
||||
}
|
||||
|
||||
void cufftError(int err, const char *file, const int line, const char *func)
|
||||
{
|
||||
cv::error(cv::Exception(CV_GpuCufftCallError, getCufftErrorString(err), func, file, line));
|
||||
}
|
||||
|
||||
void error(const char *error_string, const char *file, const int line, const char *func)
|
||||
{
|
||||
int code = CV_GpuApiCallError;
|
||||
|
@ -52,8 +52,6 @@ void cv::gpu::matchTemplate(const GpuMat&, const GpuMat&, GpuMat&, int) { throw_
|
||||
|
||||
#else
|
||||
|
||||
#include <cufft.h>
|
||||
|
||||
namespace cv { namespace gpu { namespace imgproc
|
||||
{
|
||||
void multiplyAndNormalizeSpects(int n, float scale, const cufftComplex* a,
|
||||
@ -271,27 +269,27 @@ namespace
|
||||
cufftReal* image_data;
|
||||
cufftReal* templ_data;
|
||||
cufftReal* result_data;
|
||||
cudaMalloc((void**)&image_data, sizeof(cufftReal) * dft_size.area());
|
||||
cudaMalloc((void**)&templ_data, sizeof(cufftReal) * dft_size.area());
|
||||
cudaMalloc((void**)&result_data, sizeof(cufftReal) * dft_size.area());
|
||||
cudaSafeCall(cudaMalloc((void**)&image_data, sizeof(cufftReal) * dft_size.area()));
|
||||
cudaSafeCall(cudaMalloc((void**)&templ_data, sizeof(cufftReal) * dft_size.area()));
|
||||
cudaSafeCall(cudaMalloc((void**)&result_data, sizeof(cufftReal) * dft_size.area()));
|
||||
|
||||
int spect_len = dft_size.height * (dft_size.width / 2 + 1);
|
||||
cufftComplex* image_spect;
|
||||
cufftComplex* templ_spect;
|
||||
cufftComplex* result_spect;
|
||||
cudaMalloc((void**)&image_spect, sizeof(cufftComplex) * spect_len);
|
||||
cudaMalloc((void**)&templ_spect, sizeof(cufftComplex) * spect_len);
|
||||
cudaMalloc((void**)&result_spect, sizeof(cufftComplex) * spect_len);
|
||||
cudaSafeCall(cudaMalloc((void**)&image_spect, sizeof(cufftComplex) * spect_len));
|
||||
cudaSafeCall(cudaMalloc((void**)&templ_spect, sizeof(cufftComplex) * spect_len));
|
||||
cudaSafeCall(cudaMalloc((void**)&result_spect, sizeof(cufftComplex) * spect_len));
|
||||
|
||||
cufftHandle planR2C, planC2R;
|
||||
CV_Assert(cufftPlan2d(&planC2R, dft_size.height, dft_size.width, CUFFT_C2R) == CUFFT_SUCCESS);
|
||||
CV_Assert(cufftPlan2d(&planR2C, dft_size.height, dft_size.width, CUFFT_R2C) == CUFFT_SUCCESS);
|
||||
cufftSafeCall(cufftPlan2d(&planC2R, dft_size.height, dft_size.width, CUFFT_C2R));
|
||||
cufftSafeCall(cufftPlan2d(&planR2C, dft_size.height, dft_size.width, CUFFT_R2C));
|
||||
|
||||
GpuMat templ_roi(templ.size(), CV_32S, templ.data, templ.step);
|
||||
GpuMat templ_block(dft_size, CV_32S, templ_data, dft_size.width * sizeof(cufftReal));
|
||||
copyMakeBorder(templ_roi, templ_block, 0, templ_block.rows - templ_roi.rows, 0,
|
||||
templ_block.cols - templ_roi.cols, 0);
|
||||
CV_Assert(cufftExecR2C(planR2C, templ_data, templ_spect) == CUFFT_SUCCESS);
|
||||
cufftSafeCall(cufftExecR2C(planR2C, templ_data, templ_spect));
|
||||
|
||||
GpuMat image_block(dft_size, CV_32S, image_data, dft_size.width * sizeof(cufftReal));
|
||||
|
||||
@ -306,10 +304,10 @@ namespace
|
||||
copyMakeBorder(image_roi, image_block, 0, image_block.rows - image_roi.rows, 0,
|
||||
image_block.cols - image_roi.cols, 0);
|
||||
|
||||
CV_Assert(cufftExecR2C(planR2C, image_data, image_spect) == CUFFT_SUCCESS);
|
||||
cufftSafeCall(cufftExecR2C(planR2C, image_data, image_spect));
|
||||
imgproc::multiplyAndNormalizeSpects(spect_len, 1.f / dft_size.area(),
|
||||
image_spect, templ_spect, result_spect);
|
||||
CV_Assert(cufftExecC2R(planC2R, result_spect, result_data) == CUFFT_SUCCESS);
|
||||
cufftSafeCall(cufftExecC2R(planC2R, result_spect, result_data));
|
||||
|
||||
Size result_roi_size;
|
||||
result_roi_size.width = min(x + block_size.width, result.cols) - x;
|
||||
@ -320,15 +318,15 @@ namespace
|
||||
}
|
||||
}
|
||||
|
||||
cufftDestroy(planR2C);
|
||||
cufftDestroy(planC2R);
|
||||
cufftSafeCall(cufftDestroy(planR2C));
|
||||
cufftSafeCall(cufftDestroy(planC2R));
|
||||
|
||||
cudaFree(image_spect);
|
||||
cudaFree(templ_spect);
|
||||
cudaFree(result_spect);
|
||||
cudaFree(image_data);
|
||||
cudaFree(templ_data);
|
||||
cudaFree(result_data);
|
||||
cudaSafeCall(cudaFree(image_spect));
|
||||
cudaSafeCall(cudaFree(templ_spect));
|
||||
cudaSafeCall(cudaFree(result_spect));
|
||||
cudaSafeCall(cudaFree(image_data));
|
||||
cudaSafeCall(cudaFree(templ_data));
|
||||
cudaSafeCall(cudaFree(result_data));
|
||||
}
|
||||
|
||||
|
||||
|
@ -64,6 +64,7 @@
|
||||
|
||||
#include "internal_shared.hpp"
|
||||
#include "cuda_runtime_api.h"
|
||||
#include "cufft.h"
|
||||
#include "opencv2/gpu/stream_accessor.hpp"
|
||||
#include "npp.h"
|
||||
#include "npp_staging.h"
|
||||
|
Loading…
Reference in New Issue
Block a user