Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge pull request #1146 from jet47:fix-gpu-without-cufft-cublas

Browse Source
pull/1138/merge
Roman Donchenko 11 years ago committed by OpenCV Buildbot
parent c48d3ad71e ebe7ff991f
commit 12f304ec54
4 changed files with 114 additions and 28 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 8
      modules/gpu/src/cuda/imgproc.cu
  2. 95
      modules/gpu/src/cuda/safe_call.hpp
  3. 8
      modules/gpu/src/error.cpp
  4. 31
      modules/gpu/src/imgproc.cpp

8
modules/gpu/src/cuda/imgproc.cu
Unescape View File

@ -619,6 +619,7 @@ namespace cv { namespace gpu { namespace device
//////////////////////////////////////////////////////////////////////////
// mulSpectrums
#ifdef HAVE_CUFFT
__global__ void mulSpectrumsKernel(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, PtrStepSz<cufftComplex> c)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
@ -642,11 +643,13 @@ namespace cv { namespace gpu { namespace device
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
#endif
//////////////////////////////////////////////////////////////////////////
// mulSpectrums_CONJ
#ifdef HAVE_CUFFT
__global__ void mulSpectrumsKernel_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, PtrStepSz<cufftComplex> c)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
@ -670,11 +673,13 @@ namespace cv { namespace gpu { namespace device
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
#endif
//////////////////////////////////////////////////////////////////////////
// mulAndScaleSpectrums
#ifdef HAVE_CUFFT
__global__ void mulAndScaleSpectrumsKernel(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, float scale, PtrStepSz<cufftComplex> c)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
@ -699,11 +704,13 @@ namespace cv { namespace gpu { namespace device
if (stream)
cudaSafeCall( cudaDeviceSynchronize() );
}
#endif
//////////////////////////////////////////////////////////////////////////
// mulAndScaleSpectrums_CONJ
#ifdef HAVE_CUFFT
__global__ void mulAndScaleSpectrumsKernel_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, float scale, PtrStepSz<cufftComplex> c)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
@ -728,6 +735,7 @@ namespace cv { namespace gpu { namespace device
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
#endif
//////////////////////////////////////////////////////////////////////////
// buildWarpMaps

95
modules/gpu/src/cuda/safe_call.hpp
Unescape View File

@ -43,53 +43,94 @@
#ifndef __OPENCV_CUDA_SAFE_CALL_HPP__
#define __OPENCV_CUDA_SAFE_CALL_HPP__
#include "cvconfig.h"
#include <cuda_runtime_api.h>
#include <cufft.h>
#include <cublas.h>
#include "NCV.hpp"
#if defined(__GNUC__)
#define nppSafeCall(expr) ___nppSafeCall(expr, __FILE__, __LINE__, __func__)
#define ncvSafeCall(expr) ___ncvSafeCall(expr, __FILE__, __LINE__, __func__)
#define cufftSafeCall(expr) ___cufftSafeCall(expr, __FILE__, __LINE__, __func__)
#define cublasSafeCall(expr) ___cublasSafeCall(expr, __FILE__, __LINE__, __func__)
#else /* defined(__CUDACC__) || defined(__MSVC__) */
#define nppSafeCall(expr) ___nppSafeCall(expr, __FILE__, __LINE__)
#define ncvSafeCall(expr) ___ncvSafeCall(expr, __FILE__, __LINE__)
#define cufftSafeCall(expr) ___cufftSafeCall(expr, __FILE__, __LINE__)
#define cublasSafeCall(expr) ___cublasSafeCall(expr, __FILE__, __LINE__)
#ifdef HAVE_CUFFT
# include <cufft.h>
#endif
namespace cv { namespace gpu
{
void nppError(int err, const char *file, const int line, const char *func = "");
void ncvError(int err, const char *file, const int line, const char *func = "");
#ifdef HAVE_CUBLAS
# include <cublas.h>
#endif
#include "NCV.hpp"
namespace cv { namespace gpu {
void nppError(int err, const char *file, const int line, const char *func = "");
void ncvError(int err, const char *file, const int line, const char *func = "");
#ifdef HAVE_CUFFT
void cufftError(int err, const char *file, const int line, const char *func = "");
#endif
#ifdef HAVE_CUBLAS
void cublasError(int err, const char *file, const int line, const char *func = "");
#endif
}}
// nppSafeCall
static inline void ___nppSafeCall(int err, const char *file, const int line, const char *func = "")
{
if (err < 0)
cv::gpu::nppError(err, file, line, func);
}
#if defined(__GNUC__)
#define nppSafeCall(expr) ___nppSafeCall(expr, __FILE__, __LINE__, __func__)
#else
#define nppSafeCall(expr) ___nppSafeCall(expr, __FILE__, __LINE__)
#endif
// ncvSafeCall
static inline void ___ncvSafeCall(int err, const char *file, const int line, const char *func = "")
{
if (NCV_SUCCESS != err)
cv::gpu::ncvError(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);
}
#if defined(__GNUC__)
#define ncvSafeCall(expr) ___ncvSafeCall(expr, __FILE__, __LINE__, __func__)
#else
#define ncvSafeCall(expr) ___ncvSafeCall(expr, __FILE__, __LINE__)
#endif
static inline void ___cublasSafeCall(cublasStatus_t err, const char *file, const int line, const char *func = "")
{
if (CUBLAS_STATUS_SUCCESS != err)
cv::gpu::cublasError(err, file, line, func);
}
// cufftSafeCall
#ifdef HAVE_CUFFT
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);
}
#if defined(__GNUC__)
#define cufftSafeCall(expr) ___cufftSafeCall(expr, __FILE__, __LINE__, __func__)
#else
#define cufftSafeCall(expr) ___cufftSafeCall(expr, __FILE__, __LINE__)
#endif
#endif
// cublasSafeCall
#ifdef HAVE_CUBLAS
static inline void ___cublasSafeCall(cublasStatus_t err, const char *file, const int line, const char *func = "")
{
if (CUBLAS_STATUS_SUCCESS != err)
cv::gpu::cublasError(err, file, line, func);
}
#if defined(__GNUC__)
#define cublasSafeCall(expr) ___cublasSafeCall(expr, __FILE__, __LINE__, __func__)
#else
#define cublasSafeCall(expr) ___cublasSafeCall(expr, __FILE__, __LINE__)
#endif
#endif
#endif /* __OPENCV_CUDA_SAFE_CALL_HPP__ */

8
modules/gpu/src/error.cpp
Unescape View File

@ -224,6 +224,7 @@ namespace
//////////////////////////////////////////////////////////////////////////
// CUFFT errors
#ifdef HAVE_CUFFT
const ErrorEntry cufft_errors[] =
{
error_entry( CUFFT_INVALID_PLAN ),
@ -238,10 +239,12 @@ namespace
};
const int cufft_error_num = sizeof(cufft_errors) / sizeof(cufft_errors[0]);
#endif
//////////////////////////////////////////////////////////////////////////
// CUBLAS errors
#ifdef HAVE_CUBLAS
const ErrorEntry cublas_errors[] =
{
error_entry( CUBLAS_STATUS_SUCCESS ),
@ -255,6 +258,7 @@ namespace
};
const int cublas_error_num = sizeof(cublas_errors) / sizeof(cublas_errors[0]);
#endif
}
namespace cv
@ -273,17 +277,21 @@ namespace cv
cv::gpu::error(msg.c_str(), file, line, func);
}
#ifdef HAVE_CUFFT
void cufftError(int code, const char *file, const int line, const char *func)
{
string msg = getErrorString(code, cufft_errors, cufft_error_num);
cv::gpu::error(msg.c_str(), file, line, func);
}
#endif
#ifdef HAVE_CUBLAS
void cublasError(int code, const char *file, const int line, const char *func)
{
string msg = getErrorString(code, cublas_errors, cublas_error_num);
cv::gpu::error(msg.c_str(), file, line, func);
}
#endif
}
}

31
modules/gpu/src/imgproc.cpp
Unescape View File

@ -1136,6 +1136,8 @@ void cv::gpu::cornerMinEigenVal(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuM
//////////////////////////////////////////////////////////////////////////////
// mulSpectrums
#ifdef HAVE_CUFFT
namespace cv { namespace gpu { namespace device
{
namespace imgproc
@ -1146,9 +1148,20 @@ namespace cv { namespace gpu { namespace device
}
}}}
#endif
void cv::gpu::mulSpectrums(const GpuMat& a, const GpuMat& b, GpuMat& c, int flags, bool conjB, Stream& stream)
{
(void)flags;
#ifndef HAVE_CUFFT
(void) a;
(void) b;
(void) c;
(void) flags;
(void) conjB;
(void) stream;
throw_nogpu();
#else
(void) flags;
using namespace ::cv::gpu::device::imgproc;
typedef void (*Caller)(const PtrStep<cufftComplex>, const PtrStep<cufftComplex>, PtrStepSz<cufftComplex>, cudaStream_t stream);
@ -1162,11 +1175,14 @@ void cv::gpu::mulSpectrums(const GpuMat& a, const GpuMat& b, GpuMat& c, int flag
Caller caller = callers[(int)conjB];
caller(a, b, c, StreamAccessor::getStream(stream));
#endif
}
//////////////////////////////////////////////////////////////////////////////
// mulAndScaleSpectrums
#ifdef HAVE_CUFFT
namespace cv { namespace gpu { namespace device
{
namespace imgproc
@ -1177,8 +1193,20 @@ namespace cv { namespace gpu { namespace device
}
}}}
#endif
void cv::gpu::mulAndScaleSpectrums(const GpuMat& a, const GpuMat& b, GpuMat& c, int flags, float scale, bool conjB, Stream& stream)
{
#ifndef HAVE_CUFFT
(void) a;
(void) b;
(void) c;
(void) flags;
(void) scale;
(void) conjB;
(void) stream;
throw_nogpu();
#else
(void)flags;
using namespace ::cv::gpu::device::imgproc;
@ -1192,6 +1220,7 @@ void cv::gpu::mulAndScaleSpectrums(const GpuMat& a, const GpuMat& b, GpuMat& c,
Caller caller = callers[(int)conjB];
caller(a, b, scale, c, StreamAccessor::getStream(stream));
#endif
}
//////////////////////////////////////////////////////////////////////////////

Write Preview
Loading…
Cancel
Save