generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

fixed implementation gpumat::setTo()

Browse Source
This commit is contained in:
Andrey Morozov 2010-07-19 14:36:23 +00:00
parent c3f1a6e78a
commit 2be1c7d13f
2 changed files with 38 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
modules/gpu/src/cuda/matrix_operations.cu
View File

@ -41,6 +41,7 @@
//M*/ //M*/
#include <stddef.h> #include <stddef.h>
#include <iostream>
#include "cuda_shared.hpp" #include "cuda_shared.hpp"
#include "cuda_runtime.h" #include "cuda_runtime.h"
@ -88,42 +89,47 @@ namespace mat_operators
} }
extern "C" void cv::gpu::impl::set_to_with_mask(const DevMem2D& mat, const double * scalar, const DevMem2D& mask, int depth, int channels) extern "C" void cv::gpu::impl::set_to_with_mask(const DevMem2D& mat, const double * scalar, const DevMem2D& mask, int elemSize1, int channels)
{ {
scalar_d[0] = scalar[0]; // download scalar to constant memory
scalar_d[1] = scalar[1]; float data[4];
scalar_d[2] = scalar[2]; data[0] = scalar[0];
scalar_d[3] = scalar[3]; data[1] = scalar[1];
data[2] = scalar[2];
data[3] = scalar[3];
cudaMemcpyToSymbol(scalar_d, data, sizeof(data));
dim3 numBlocks(mat.rows * mat.step / 256, 1, 1); dim3 numBlocks(mat.rows * mat.step / 256, 1, 1);
dim3 threadsPerBlock(256); dim3 threadsPerBlock(256);
if (channels == 1) if (channels == 1)
{ {
if (depth == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 1><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr); if (elemSize1 == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 1><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 1><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr); if (elemSize1 == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 1><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 1><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr); if (elemSize1 == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 1><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr);
} }
if (channels == 2) if (channels == 2)
{ {
if (depth == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 2><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr); if (elemSize1 == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 2><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 2><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr); if (elemSize1 == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 2><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 2><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr); if (elemSize1 == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 2><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr);
} }
if (channels == 3) if (channels == 3)
{ {
if (depth == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 3><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr); if (elemSize1 == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 3><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 3><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr); if (elemSize1 == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 3><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 3><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr); if (elemSize1 == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 3><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr);
} }
} }
extern "C" void cv::gpu::impl::set_to_without_mask(const DevMem2D& mat, const double * scalar, int depth, int channels) extern "C" void cv::gpu::impl::set_to_without_mask(const DevMem2D& mat, const double * scalar, int elemSize1, int channels)
{ {
scalar_d[0] = scalar[0]; float data[4];
scalar_d[1] = scalar[1]; data[0] = scalar[0];
scalar_d[2] = scalar[2]; data[1] = scalar[1];
scalar_d[3] = scalar[3]; data[2] = scalar[2];
data[3] = scalar[3];
cudaMemcpyToSymbol(scalar_d, data, sizeof(data));
int numBlocks = mat.rows * mat.step / 256; int numBlocks = mat.rows * mat.step / 256;
@ -131,20 +137,20 @@ extern "C" void cv::gpu::impl::set_to_without_mask(const DevMem2D& mat, const do
if (channels == 1) if (channels == 1)
{ {
if (depth == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 1><<<numBlocks,threadsPerBlock>>>(mat.ptr); if (elemSize1 == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 1><<<numBlocks,threadsPerBlock>>>(mat.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 1><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr); if (elemSize1 == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 1><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 1><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr); if (elemSize1 == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 1><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr);
} }
if (channels == 2) if (channels == 2)
{ {
if (depth == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 2><<<numBlocks,threadsPerBlock>>>(mat.ptr); if (elemSize1 == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 2><<<numBlocks,threadsPerBlock>>>(mat.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 2><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr); if (elemSize1 == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 2><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 2><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr); if (elemSize1 == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 2><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr);
} }
if (channels == 3) if (channels == 3)
{ {
if (depth == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 3><<<numBlocks,threadsPerBlock>>>(mat.ptr); if (elemSize1 == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 3><<<numBlocks,threadsPerBlock>>>(mat.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 3><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr); if (elemSize1 == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 3><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 3><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr); if (elemSize1 == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 3><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr);
} }
} }

8
modules/gpu/src/matrix_operations.cpp
View File

@ -111,23 +111,23 @@ void cv::gpu::GpuMat::convertTo( GpuMat& /*m*/, int /*rtype*/, double /*alpha*/,
GpuMat& GpuMat::operator = (const Scalar& s) GpuMat& GpuMat::operator = (const Scalar& s)
{ {
cv::gpu::impl::set_to_without_mask(*this, s.val, this->depth(), this->channels()); cv::gpu::impl::set_to_without_mask(*this, s.val, this->elemSize1(), this->channels());
return *this; return *this;
} }
GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask) GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask)
{ {
CV_Assert(mask.type() == CV_8U); //CV_Assert(mask.type() == CV_8U);
CV_DbgAssert(!this->empty()); CV_DbgAssert(!this->empty());
if (mask.empty()) if (mask.empty())
{ {
cv::gpu::impl::set_to_without_mask(*this, s.val, this->depth(), this->channels()); cv::gpu::impl::set_to_without_mask(*this, s.val, this->elemSize1(), this->channels());
} }
else else
{ {
cv::gpu::impl::set_to_with_mask(*this, s.val, mask, this->depth(), this->channels()); cv::gpu::impl::set_to_with_mask(*this, s.val, mask, this->elemSize1(), this->channels());
} }
return *this; return *this;