Skip to content
This repository was archived by the owner on Jan 25, 2023. It is now read-only.

Multi-threading with OpenMP for some statements in own_FilterEdge #2

Merged
samir-nasibli merged 4 commits into from
Aug 18, 2020
Merged

Multi-threading with OpenMP for some statements in own_FilterEdge #2

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
43abcd9
Multi-threading with OpenMP for some statements in own_FilterEdge
samir-nasibli Aug 11, 2020
bd31f0a
added USE_OPENMP directive
samir-nasibli Aug 18, 2020
973c0d5
Update own_filters.c
samir-nasibli Aug 18, 2020
1a44d30
Merge branch 'master' into feature/edge_filters_optimize
samir-nasibli Aug 18, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
101 changes: 89 additions & 12 deletions skipp/filters/src/own_filters.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -325,6 +325,7 @@ own_FilterEdge(
int dstStep;
IppiSize roiSize = { img_width, img_height }; // Size of source and
// destination ROI in pixels
Ipp32f value;
IppiMaskSize maskSize = ippMskSize3x3;
IppNormType normType = ippNormL2; // As is in skimage.filters.prewitt

Expand All @@ -345,6 +346,22 @@ own_FilterEdge(
// currently Intel(R) IPP supports only 1C images for filtering by prewitt kernels
srcStep = numChannels * img_width * sizeof_src;
dstStep = numChannels * img_width * sizeof_dst;
#ifdef USE_OPENMP
IppStatus * pStatus = NULL;
int max_num_threads;
int numThreads, slice, tail;
IppiSize dstTileSize, dstLastTileSize;

#ifdef MAX_NUM_THREADS
max_num_threads = MAX_NUM_THREADS;
#else
max_num_threads = omp_get_max_threads();
if(roiSize.height / max_num_threads < 2)
{
max_num_threads = 1;
}
#endif
#endif

switch (edgeKernel)
{
Expand Down Expand Up @@ -438,8 +455,7 @@ own_FilterEdge(
// `convolve` func. `convolve` uses `reflect` border mode.
// `reflect` border mode is equivalent of Intel IPP ippBorderMirrorR border type
// ippiFilterSobelVertBorder_<mode> doesn't supports this border type
Ipp32f value = (Ipp32f)(-4.0);
status = ippiDivC_32f_C1IR(value, pDst, dstStep, roiSize);
value = (Ipp32f)(-4.0);
break;
}
case own_filterSobelHoriz:
Expand All @@ -449,8 +465,7 @@ own_FilterEdge(
// `convolve` func. `convolve` uses `reflect` border mode.
// `reflect` border mode is equivalent of Intel IPP ippBorderMirrorR border type
// ippiFilterSobelHorizBorder_<mode> doesn't supports this border type
Ipp32f value = (Ipp32f)4.0;
status = ippiDivC_32f_C1IR(value, pDst, dstStep, roiSize);
value = (Ipp32f)4.0;
break;
}
case own_filterSobel:
Expand All @@ -460,9 +475,7 @@ own_FilterEdge(
// `convolve` func. `convolve` uses `reflect` border mode.
// `reflect` border mode is equivalent of Intel IPP ippBorderMirrorR border type
// ippiFilterSobelBorder_<mode> doesn't supports this border type
Ipp32f sqrt2 = (Ipp32f)IPP_SQRT2;
Ipp32f value = (Ipp32f)(4.0 * sqrt2);
status = ippiDivC_32f_C1IR(value, pDst, dstStep, roiSize);
value = (Ipp32f)(4.0 * (Ipp32f)IPP_SQRT2);
break;
}
case own_filterPrewittVert:
Expand All @@ -472,8 +485,7 @@ own_FilterEdge(
// `convolve` func. `convolve` uses `reflect` border mode.
// `reflect` border mode is equivalent of Intel IPP ippBorderMirrorR border type
// ippiFilterPrewittVertBorder_<mode> doesn't supports this border type
Ipp32f value = (Ipp32f)-3.0;
status = ippiDivC_32f_C1IR(value, pDst, dstStep, roiSize);
value = (Ipp32f)-3.0;
break;
}
case own_filterPrewittHoriz:
Expand All @@ -483,17 +495,82 @@ own_FilterEdge(
// `convolve` func. `convolve` uses `reflect` border mode.
// `reflect` border mode is equivalent of Intel IPP ippBorderMirrorR border type
// ippiFilterPrewittHorizBorder_<mode> doesn't supports this border type
Ipp32f value = (Ipp32f)3.0;
status = ippiDivC_32f_C1IR(value, pDst, dstStep, roiSize);
value = (Ipp32f)3.0;
break;
}
default:
{
status = ippStsErr;
}
}
check_sts(status);
#ifdef USE_OPENMP
if (max_num_threads != 1)
{
// Parallelized only by Y-direction here
#pragma omp parallel num_threads(max_num_threads)
{
#pragma omp master
{
numThreads = omp_get_num_threads();

pStatus = (IppStatus*)ippsMalloc_8u(sizeof(IppStatus) * numThreads);
if (pStatus == NULL)
{
status = ippStsMemAllocErr;
}
if(status == ippStsNoErr)
{
for (int i = 0; i < numThreads; ++i) pStatus[i] = ippStsNoErr;

slice = roiSize.height / numThreads;
tail = roiSize.height % numThreads;

dstTileSize.width = roiSize.width;
dstTileSize.height = slice;
dstLastTileSize.width = roiSize.width;
dstLastTileSize.height = slice + tail;
}
}
#pragma omp barrier
{
if (status == ippStsNoErr)
{
Ipp32u i;
void * pDstT = NULL;
i = omp_get_thread_num();
IppiPoint dstOffset = { 0, 0 };
IppiSize dstSizeT = dstTileSize;

dstSizeT.height = slice;
dstOffset.y += i * slice;

if (i == numThreads - 1) dstSizeT = dstLastTileSize;

pDstT = (void*)((Ipp8u*)pDst + dstOffset.y * (Ipp32s)dstStep);
pStatus[i] = ippiDivC_32f_C1IR(value, pDstT, dstStep, dstSizeT);
}
}
}
check_sts(status);
// Checking status for slices and tile
for (Ipp32u i = 0; i < numThreads; ++i)
{
status = pStatus[i];
check_sts(status);
}
}
else
{
#endif
status = ippiDivC_32f_C1IR(value, pDst, dstStep, roiSize);
check_sts(status);
#ifdef USE_OPENMP
}
#endif
EXIT_FUNC
#ifdef USE_OPENMP
ippsFree(pStatus);
#endif
ippsFree(pBuffer);
return status;
}
Expand Down