Skip to content

O
opencv
Unverified Commit 35f66340 authored by Claudio's avatar Claudio
Browse files
Options

Add cuda::Stream capability to cuda::HOG::compute 

In the previous version only the default stream was/could be used, i.e.
cv::cuda::Stream::Null().

With this change, HOG::compute() will now run in parallel over different
cuda::Streams.

The code has been reordered so that all data allocation is completed
first, then all the kernels are run in parallel over streams.

Fix #8177
parent f109c013
Hide whitespace changes
Inline Side-by-side
Showing with 77 additions and 87 deletions
modules/cudaobjdetect/src/cuda/hog.cu
View file @ 35f66340
......@@ -52,6 +52,7 @@ namespace cv { namespace cuda { namespace device
namespace hog
{
__constant__ int cnbins;
__constant__ int cblock_stride_x;
__constant__ int cblock_stride_y;
......@@ -99,27 +100,28 @@ namespace cv { namespace cuda { namespace device
void set_up_constants(int nbins, int block_stride_x, int block_stride_y,
int nblocks_win_x, int nblocks_win_y, int ncells_block_x, int ncells_block_y)
int nblocks_win_x, int nblocks_win_y, int ncells_block_x, int ncells_block_y,
const cudaStream_t& stream)
{
cudaSafeCall( cudaMemcpyToSymbol(cnbins, &nbins, sizeof(nbins)) );
cudaSafeCall( cudaMemcpyToSymbol(cblock_stride_x, &block_stride_x, sizeof(block_stride_x)) );
cudaSafeCall( cudaMemcpyToSymbol(cblock_stride_y, &block_stride_y, sizeof(block_stride_y)) );
cudaSafeCall( cudaMemcpyToSymbol(cnblocks_win_x, &nblocks_win_x, sizeof(nblocks_win_x)) );
cudaSafeCall( cudaMemcpyToSymbol(cnblocks_win_y, &nblocks_win_y, sizeof(nblocks_win_y)) );
cudaSafeCall( cudaMemcpyToSymbol(cncells_block_x, &ncells_block_x, sizeof(ncells_block_x)) );
cudaSafeCall( cudaMemcpyToSymbol(cncells_block_y, &ncells_block_y, sizeof(ncells_block_y)) );
cudaSafeCall(cudaMemcpyToSymbolAsync(cnbins, &nbins, sizeof(nbins), 0, cudaMemcpyHostToDevice, stream));
cudaSafeCall(cudaMemcpyToSymbolAsync(cblock_stride_x, &block_stride_x, sizeof(block_stride_x), 0, cudaMemcpyHostToDevice, stream));
cudaSafeCall(cudaMemcpyToSymbolAsync(cblock_stride_y, &block_stride_y, sizeof(block_stride_y), 0, cudaMemcpyHostToDevice, stream));
cudaSafeCall(cudaMemcpyToSymbolAsync(cnblocks_win_x, &nblocks_win_x, sizeof(nblocks_win_x), 0, cudaMemcpyHostToDevice, stream));
cudaSafeCall(cudaMemcpyToSymbolAsync(cnblocks_win_y, &nblocks_win_y, sizeof(nblocks_win_y), 0, cudaMemcpyHostToDevice, stream));
cudaSafeCall(cudaMemcpyToSymbolAsync(cncells_block_x, &ncells_block_x, sizeof(ncells_block_x), 0, cudaMemcpyHostToDevice, stream));
cudaSafeCall(cudaMemcpyToSymbolAsync(cncells_block_y, &ncells_block_y, sizeof(ncells_block_y), 0, cudaMemcpyHostToDevice, stream));
int block_hist_size = nbins * ncells_block_x * ncells_block_y;
cudaSafeCall( cudaMemcpyToSymbol(cblock_hist_size, &block_hist_size, sizeof(block_hist_size)) );
cudaSafeCall(cudaMemcpyToSymbolAsync(cblock_hist_size, &block_hist_size, sizeof(block_hist_size), 0, cudaMemcpyHostToDevice, stream));
int block_hist_size_2up = power_2up(block_hist_size);
cudaSafeCall( cudaMemcpyToSymbol(cblock_hist_size_2up, &block_hist_size_2up, sizeof(block_hist_size_2up)) );
cudaSafeCall(cudaMemcpyToSymbolAsync(cblock_hist_size_2up, &block_hist_size_2up, sizeof(block_hist_size_2up), 0, cudaMemcpyHostToDevice, stream));
int descr_width = nblocks_win_x * block_hist_size;
cudaSafeCall( cudaMemcpyToSymbol(cdescr_width, &descr_width, sizeof(descr_width)) );
cudaSafeCall(cudaMemcpyToSymbolAsync(cdescr_width, &descr_width, sizeof(descr_width), 0, cudaMemcpyHostToDevice, stream));
int descr_size = descr_width * nblocks_win_y;
cudaSafeCall( cudaMemcpyToSymbol(cdescr_size, &descr_size, sizeof(descr_size)) );
cudaSafeCall(cudaMemcpyToSymbolAsync(cdescr_size, &descr_size, sizeof(descr_size), 0, cudaMemcpyHostToDevice, stream));
}
......@@ -233,7 +235,8 @@ namespace cv { namespace cuda { namespace device
void compute_hists(int nbins, int block_stride_x, int block_stride_y,
int height, int width, const PtrStepSzf& grad,
const PtrStepSzb& qangle, float sigma, float* block_hists,
int cell_size_x, int cell_size_y, int ncells_block_x, int ncells_block_y)
int cell_size_x, int cell_size_y, int ncells_block_x, int ncells_block_y,
const cudaStream_t& stream)
{
const int ncells_block = ncells_block_x * ncells_block_y;
const int patch_side = cell_size_x / 4;
......@@ -259,20 +262,15 @@ namespace cv { namespace cuda { namespace device
int final_hists_size = (nbins * ncells_block * nblocks) * sizeof(float);
int smem = hists_size + final_hists_size;
if (nblocks == 4)
compute_hists_kernel_many_blocks<4><<<grid, threads, smem>>>(
img_block_width, grad, qangle, scale, block_hists, cell_size_x, patch_size, block_patch_size, threads_cell, threads_block, half_cell_size);
compute_hists_kernel_many_blocks<4><<<grid, threads, smem, stream>>>(img_block_width, grad, qangle, scale, block_hists, cell_size_x, patch_size, block_patch_size, threads_cell, threads_block, half_cell_size);
else if (nblocks == 3)
compute_hists_kernel_many_blocks<3><<<grid, threads, smem>>>(
img_block_width, grad, qangle, scale, block_hists, cell_size_x, patch_size, block_patch_size, threads_cell, threads_block, half_cell_size);
compute_hists_kernel_many_blocks<3><<<grid, threads, smem, stream>>>(img_block_width, grad, qangle, scale, block_hists, cell_size_x, patch_size, block_patch_size, threads_cell, threads_block, half_cell_size);
else if (nblocks == 2)
compute_hists_kernel_many_blocks<2><<<grid, threads, smem>>>(
img_block_width, grad, qangle, scale, block_hists, cell_size_x, patch_size, block_patch_size, threads_cell, threads_block, half_cell_size);
compute_hists_kernel_many_blocks<2><<<grid, threads, smem, stream>>>(img_block_width, grad, qangle, scale, block_hists, cell_size_x, patch_size, block_patch_size, threads_cell, threads_block, half_cell_size);
else
compute_hists_kernel_many_blocks<1><<<grid, threads, smem>>>(
img_block_width, grad, qangle, scale, block_hists, cell_size_x, patch_size, block_patch_size, threads_cell, threads_block, half_cell_size);
cudaSafeCall( cudaGetLastError() );
compute_hists_kernel_many_blocks<1><<<grid, threads, smem, stream>>>(img_block_width, grad, qangle, scale, block_hists, cell_size_x, patch_size, block_patch_size, threads_cell, threads_block, half_cell_size);
cudaSafeCall( cudaDeviceSynchronize() );
cudaSafeCall( cudaGetLastError() );
}
......@@ -348,7 +346,8 @@ namespace cv { namespace cuda { namespace device
void normalize_hists(int nbins, int block_stride_x, int block_stride_y,
int height, int width, float* block_hists, float threshold, int cell_size_x, int cell_size_y, int ncells_block_x, int ncells_block_y)
int height, int width, float* block_hists, float threshold, int cell_size_x, int cell_size_y, int ncells_block_x, int ncells_block_y,
const cudaStream_t& stream)
{
const int nblocks = 1;
......@@ -361,21 +360,19 @@ namespace cv { namespace cuda { namespace device
dim3 grid(divUp(img_block_width, nblocks), img_block_height);
if (nthreads == 32)
normalize_hists_kernel_many_blocks<32, nblocks><<<grid, threads>>>(block_hist_size, img_block_width, block_hists, threshold);
normalize_hists_kernel_many_blocks<32, nblocks><<<grid, threads, 0, stream>>>(block_hist_size, img_block_width, block_hists, threshold);
else if (nthreads == 64)
normalize_hists_kernel_many_blocks<64, nblocks><<<grid, threads>>>(block_hist_size, img_block_width, block_hists, threshold);
normalize_hists_kernel_many_blocks<64, nblocks><<<grid, threads, 0, stream>>>(block_hist_size, img_block_width, block_hists, threshold);
else if (nthreads == 128)
normalize_hists_kernel_many_blocks<128, nblocks><<<grid, threads>>>(block_hist_size, img_block_width, block_hists, threshold);
normalize_hists_kernel_many_blocks<128, nblocks><<<grid, threads, 0, stream>>>(block_hist_size, img_block_width, block_hists, threshold);
else if (nthreads == 256)
normalize_hists_kernel_many_blocks<256, nblocks><<<grid, threads>>>(block_hist_size, img_block_width, block_hists, threshold);
normalize_hists_kernel_many_blocks<256, nblocks><<<grid, threads, 0, stream>>>(block_hist_size, img_block_width, block_hists, threshold);
else if (nthreads == 512)
normalize_hists_kernel_many_blocks<512, nblocks><<<grid, threads>>>(block_hist_size, img_block_width, block_hists, threshold);
normalize_hists_kernel_many_blocks<512, nblocks><<<grid, threads, 0, stream>>>(block_hist_size, img_block_width, block_hists, threshold);
else
CV_Error(cv::Error::StsBadArg, "normalize_hists: histogram's size is too big, try to decrease number of bins");
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
......@@ -581,7 +578,8 @@ namespace cv { namespace cuda { namespace device
void extract_descrs_by_cols(int win_height, int win_width, int block_stride_y, int block_stride_x,
int win_stride_y, int win_stride_x, int height, int width, float* block_hists, int cell_size_x, int ncells_block_x,
PtrStepSzf descriptors)
PtrStepSzf descriptors,
const cudaStream_t& stream)
{
const int nthreads = 256;
......@@ -593,11 +591,9 @@ namespace cv { namespace cuda { namespace device
dim3 grid(img_win_width, img_win_height);
int img_block_width = (width - ncells_block_x * cell_size_x + block_stride_x) / block_stride_x;
extract_descrs_by_cols_kernel<nthreads><<<grid, threads>>>(
img_block_width, win_block_stride_x, win_block_stride_y, block_hists, descriptors);
cudaSafeCall( cudaGetLastError() );
extract_descrs_by_cols_kernel<nthreads><<<grid, threads, 0, stream>>>(img_block_width, win_block_stride_x, win_block_stride_y, block_hists, descriptors);
cudaSafeCall( cudaDeviceSynchronize() );
cudaSafeCall( cudaGetLastError() );
}
//----------------------------------------------------------------------------
......@@ -708,7 +704,8 @@ namespace cv { namespace cuda { namespace device
void compute_gradients_8UC4(int nbins, int height, int width, const PtrStepSzb& img,
float angle_scale, PtrStepSzf grad, PtrStepSzb qangle, bool correct_gamma)
float angle_scale, PtrStepSzf grad, PtrStepSzb qangle, bool correct_gamma,
const cudaStream_t& stream)
{
(void)nbins;
const int nthreads = 256;
......@@ -717,13 +714,11 @@ namespace cv { namespace cuda { namespace device
dim3 gdim(divUp(width, bdim.x), divUp(height, bdim.y));
if (correct_gamma)
compute_gradients_8UC4_kernel<nthreads, 1><<<gdim, bdim>>>(height, width, img, angle_scale, grad, qangle);
compute_gradients_8UC4_kernel<nthreads, 1><<<gdim, bdim, 0, stream>>>(height, width, img, angle_scale, grad, qangle);
else
compute_gradients_8UC4_kernel<nthreads, 0><<<gdim, bdim>>>(height, width, img, angle_scale, grad, qangle);
compute_gradients_8UC4_kernel<nthreads, 0><<<gdim, bdim, 0, stream>>>(height, width, img, angle_scale, grad, qangle);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
template <int nthreads, int correct_gamma>
......
modules/cudaobjdetect/src/hog.cpp
View file @ 35f66340
......@@ -66,15 +66,18 @@ namespace cv { namespace cuda { namespace device
{
void set_up_constants(int nbins, int block_stride_x, int block_stride_y,
int nblocks_win_x, int nblocks_win_y,
int ncells_block_x, int ncells_block_y);
int ncells_block_x, int ncells_block_y,
const cudaStream_t& stream);
void compute_hists(int nbins, int block_stride_x, int block_stride_y,
int height, int width, const PtrStepSzf& grad,
const PtrStepSzb& qangle, float sigma, float* block_hists,
int cell_size_x, int cell_size_y, int ncells_block_x, int ncells_block_y);
int cell_size_x, int cell_size_y, int ncells_block_x, int ncells_block_y,
const cudaStream_t& stream);
void normalize_hists(int nbins, int block_stride_x, int block_stride_y,
int height, int width, float* block_hists, float threshold, int cell_size_x, int cell_size_y, int ncells_block_x, int ncells_block_y);
int height, int width, float* block_hists, float threshold, int cell_size_x, int cell_size_y, int ncells_block_x, int ncells_block_y,
const cudaStream_t& stream);
void classify_hists(int win_height, int win_width, int block_stride_y,
int block_stride_x, int win_stride_y, int win_stride_x, int height,
......@@ -90,12 +93,14 @@ namespace cv { namespace cuda { namespace device
cv::cuda::PtrStepSzf descriptors);
void extract_descrs_by_cols(int win_height, int win_width, int block_stride_y, int block_stride_x,
int win_stride_y, int win_stride_x, int height, int width, float* block_hists, int cell_size_x, int ncells_block_x,
cv::cuda::PtrStepSzf descriptors);
cv::cuda::PtrStepSzf descriptors,
const cudaStream_t& stream);
void compute_gradients_8UC1(int nbins, int height, int width, const cv::cuda::PtrStepSzb& img,
float angle_scale, cv::cuda::PtrStepSzf grad, cv::cuda::PtrStepSzb qangle, bool correct_gamma);
void compute_gradients_8UC4(int nbins, int height, int width, const cv::cuda::PtrStepSzb& img,
float angle_scale, cv::cuda::PtrStepSzf grad, cv::cuda::PtrStepSzb qangle, bool correct_gamma);
float angle_scale, cv::cuda::PtrStepSzf grad, cv::cuda::PtrStepSzb qangle, bool correct_gamma,
const cudaStream_t& stream);
void resize_8UC1(const cv::cuda::PtrStepSzb& src, cv::cuda::PtrStepSzb dst);
void resize_8UC4(const cv::cuda::PtrStepSzb& src, cv::cuda::PtrStepSzb dst);
......@@ -182,8 +187,8 @@ namespace
private:
int getTotalHistSize(Size img_size) const;
void computeBlockHistograms(const GpuMat& img, GpuMat& block_hists);
void computeGradient(const GpuMat& img, GpuMat& grad, GpuMat& qangle);
void computeBlockHistograms(const GpuMat& img, GpuMat& block_hists, Stream& stream);
// void computeGradient(const GpuMat& img, GpuMat& grad, GpuMat& qangle, Stream& stream);
// Coefficients of the separating plane
float free_coef_;
......@@ -310,7 +315,7 @@ namespace
BufferPool pool(Stream::Null());
GpuMat block_hists = pool.getBuffer(1, getTotalHistSize(img.size()), CV_32FC1);
computeBlockHistograms(img, block_hists);
computeBlockHistograms(img, block_hists, Stream::Null());
Size wins_per_img = numPartsWithin(img.size(), win_size_, win_stride_);
......@@ -458,19 +463,16 @@ namespace
CV_Assert( img.type() == CV_8UC1 || img.type() == CV_8UC4 );
CV_Assert( win_stride_.width % block_stride_.width == 0 && win_stride_.height % block_stride_.height == 0 );
CV_Assert( !stream );
BufferPool pool(stream);
GpuMat block_hists = pool.getBuffer(1, getTotalHistSize(img.size()), CV_32FC1);
computeBlockHistograms(img, block_hists);
BufferPool pool(stream);
GpuMat block_hists = pool.getBuffer(1, getTotalHistSize(img.size()), CV_32FC1);
Size wins_per_img = numPartsWithin(img.size(), win_size_, win_stride_);
Size blocks_per_win = numPartsWithin(win_size_, block_size_, block_stride_);
const size_t block_hist_size = getBlockHistogramSize();
Size blocks_per_win = numPartsWithin(win_size_, block_size_, block_stride_);
Size wins_per_img = numPartsWithin(img.size(), win_size_, win_stride_);
_descriptors.create(wins_per_img.area(), static_cast<int>(blocks_per_win.area() * block_hist_size), CV_32FC1);
GpuMat descriptors = _descriptors.getGpuMat();
GpuMat descriptors = _descriptors.getGpuMat();
computeBlockHistograms(img, block_hists, stream);
switch (descr_format_)
{
......@@ -490,7 +492,8 @@ namespace
img.rows, img.cols,
block_hists.ptr<float>(),
cell_size_.width, cells_per_block_.width,
descriptors);
descriptors,
StreamAccessor::getStream(stream));
break;
default:
CV_Error(cv::Error::StsBadArg, "Unknown descriptor format");
......@@ -504,18 +507,25 @@ namespace
return static_cast<int>(block_hist_size * blocks_per_img.area());
}
void HOG_Impl::computeBlockHistograms(const GpuMat& img, GpuMat& block_hists)
void HOG_Impl::computeBlockHistograms(const GpuMat& img, GpuMat& block_hists, Stream& stream)
{
BufferPool pool(stream);
cv::Size blocks_per_win = numPartsWithin(win_size_, block_size_, block_stride_);
hog::set_up_constants(nbins_, block_stride_.width, block_stride_.height, blocks_per_win.width, blocks_per_win.height, cells_per_block_.width, cells_per_block_.height);
BufferPool pool(Stream::Null());
float angleScale = static_cast<float>(nbins_ / CV_PI);
GpuMat grad = pool.getBuffer(img.size(), CV_32FC2);
GpuMat qangle = pool.getBuffer(img.size(), CV_8UC2);
GpuMat grad = pool.getBuffer(img.size(), CV_32FC2);
GpuMat qangle = pool.getBuffer(img.size(), CV_8UC2);
computeGradient(img, grad, qangle);
hog::set_up_constants(nbins_, block_stride_.width, block_stride_.height, blocks_per_win.width, blocks_per_win.height, cells_per_block_.width, cells_per_block_.height, StreamAccessor::getStream(stream));
block_hists.create(1, getTotalHistSize(img.size()), CV_32FC1);
switch (img.type())
{
case CV_8UC1:
hog::compute_gradients_8UC1(nbins_, img.rows, img.cols, img, angleScale, grad, qangle, gamma_correction_);
break;
case CV_8UC4:
hog::compute_gradients_8UC4(nbins_, img.rows, img.cols, img, angleScale, grad, qangle, gamma_correction_, StreamAccessor::getStream(stream));
break;
}
hog::compute_hists(nbins_,
block_stride_.width, block_stride_.height,
......@@ -524,7 +534,8 @@ namespace
(float)getWinSigma(),
block_hists.ptr<float>(),
cell_size_.width, cell_size_.height,
cells_per_block_.width, cells_per_block_.height);
cells_per_block_.width, cells_per_block_.height,
StreamAccessor::getStream(stream));
hog::normalize_hists(nbins_,
block_stride_.width, block_stride_.height,
......@@ -532,24 +543,8 @@ namespace
block_hists.ptr<float>(),
(float)threshold_L2hys_,
cell_size_.width, cell_size_.height,
cells_per_block_.width, cells_per_block_.height);
}
void HOG_Impl::computeGradient(const GpuMat& img, GpuMat& grad, GpuMat& qangle)
{
grad.create(img.size(), CV_32FC2);
qangle.create(img.size(), CV_8UC2);
float angleScale = (float)(nbins_ / CV_PI);
switch (img.type())
{
case CV_8UC1:
hog::compute_gradients_8UC1(nbins_, img.rows, img.cols, img, angleScale, grad, qangle, gamma_correction_);
break;
case CV_8UC4:
hog::compute_gradients_8UC4(nbins_, img.rows, img.cols, img, angleScale, grad, qangle, gamma_correction_);
break;
}
cells_per_block_.width, cells_per_block_.height,
StreamAccessor::getStream(stream));
}
}
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment