potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 07:13:02 +04:00
Code Packages Releases Wiki Activity

Merge pull request #24918 from opencv-pushbot:gitee/alalek/core_convertfp16_replacement

Browse Source 
core(OpenCL): optimize convertTo() with CV_16F (convertFp16() replacement) #24918

relates #24909
relates #24917
relates #24892

Performance changes:

- [x] 12700K (1 thread) + Intel iGPU

|Name of Test|noOCL|convertFp16|convertTo BASE|convertTo PATCH|
|---|:-:|:-:|:-:|:-:|
|ConvertFP16FP32MatMat::OCL_Core|3.130|3.152|3.127|3.136|
|ConvertFP16FP32MatUMat::OCL_Core|3.030|3.996|3.007|2.671|
|ConvertFP16FP32UMatMat::OCL_Core|3.010|3.101|3.056|2.854|
|ConvertFP16FP32UMatUMat::OCL_Core|3.016|3.298|2.072|2.061|
|ConvertFP32FP16MatMat::OCL_Core|2.697|2.652|2.723|2.721|
|ConvertFP32FP16MatUMat::OCL_Core|2.752|4.268|2.662|2.947|
|ConvertFP32FP16UMatMat::OCL_Core|2.706|2.601|2.603|2.528|
|ConvertFP32FP16UMatUMat::OCL_Core|2.704|3.215|1.999|1.988|

Patched version is not worse than convertFp16 and convertTo baseline (except MatUMat 32->16, baseline uses CPU code+dst buffer map).
There are still gaps against noOpenCL(CPU only) mode due to T-API implementation issues (unnecessary synchronization).


- [x] 12700K + AMD dGPU

|Name of Test|noOCL|convertFp16 dGPU|convertTo BASE dGPU|convertTo PATCH dGPU|
|---|:-:|:-:|:-:|:-:|
|ConvertFP16FP32MatMat::OCL_Core|3.130|3.133|3.172|3.087|
|ConvertFP16FP32MatUMat::OCL_Core|3.030|1.713|9.559|1.729|
|ConvertFP16FP32UMatMat::OCL_Core|3.010|6.515|6.309|4.452|
|ConvertFP16FP32UMatUMat::OCL_Core|3.016|0.242|23.597|0.170|
|ConvertFP32FP16MatMat::OCL_Core|2.697|2.641|2.713|2.689|
|ConvertFP32FP16MatUMat::OCL_Core|2.752|4.076|6.483|4.191|
|ConvertFP32FP16UMatMat::OCL_Core|2.706|9.042|16.481|1.834|
|ConvertFP32FP16UMatUMat::OCL_Core|2.704|0.229|15.730|0.176|

convertTo-baseline can't compile OpenCL kernel for FP16 properly - FIXED.
dGPU has much more power, so results are x16-17 better than single cpu core. 
Patched version is not worse than convertFp16 and convertTo baseline.
There are still gaps against noOpenCL(CPU only) mode due to T-API implementation issues (unnecessary synchronization) and required memory transfers.

Co-authored-by: Alexander Alekhin <alexander.a.alekhin@gmail.com>
This commit is contained in:
Alexander Alekhin
2024-01-26 12:56:52 +03:00
committed by GitHub
parent ae21368eb9
commit 40533dbf69
9 changed files with 355 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/src/umatrix.cpp
+3 -63
View File
@@ -1233,70 +1233,10 @@ void UMat::copyTo(OutputArray _dst, InputArray _mask) const
src.copyTo(_dst, _mask);
}
void UMat::convertTo(OutputArray _dst, int _type, double alpha, double beta) const
{
CV_INSTRUMENT_REGION();
bool noScale = std::fabs(alpha - 1) < DBL_EPSILON && std::fabs(beta) < DBL_EPSILON;
int stype = type(), cn = CV_MAT_CN(stype);
if( _type < 0 )
_type = _dst.fixedType() ? _dst.type() : stype;
else
_type = CV_MAKETYPE(CV_MAT_DEPTH(_type), cn);
int sdepth = CV_MAT_DEPTH(stype), ddepth = CV_MAT_DEPTH(_type);
if( sdepth == ddepth && noScale )
{
copyTo(_dst);
return;
}
#ifdef HAVE_OPENCL
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
bool needDouble = sdepth == CV_64F || ddepth == CV_64F;
if( dims <= 2 && cn && _dst.isUMat() && ocl::useOpenCL() &&
((needDouble && doubleSupport) || !needDouble) )
{
int wdepth = std::max(CV_32F, sdepth), rowsPerWI = 4;
char cvt[2][50];
ocl::Kernel k("convertTo", ocl::core::convert_oclsrc,
format("-D srcT=%s -D WT=%s -D dstT=%s -D convertToWT=%s -D convertToDT=%s%s%s",
ocl::typeToStr(sdepth), ocl::typeToStr(wdepth), ocl::typeToStr(ddepth),
ocl::convertTypeStr(sdepth, wdepth, 1, cvt[0], sizeof(cvt[0])),
ocl::convertTypeStr(wdepth, ddepth, 1, cvt[1], sizeof(cvt[1])),
doubleSupport ? " -D DOUBLE_SUPPORT" : "", noScale ? " -D NO_SCALE" : ""));
if (!k.empty())
{
UMat src = *this;
_dst.create( size(), _type );
UMat dst = _dst.getUMat();
float alphaf = (float)alpha, betaf = (float)beta;
ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
dstarg = ocl::KernelArg::WriteOnly(dst, cn);
if (noScale)
k.args(srcarg, dstarg, rowsPerWI);
else if (wdepth == CV_32F)
k.args(srcarg, dstarg, alphaf, betaf, rowsPerWI);
else
k.args(srcarg, dstarg, alpha, beta, rowsPerWI);
size_t globalsize[2] = { (size_t)dst.cols * cn, ((size_t)dst.rows + rowsPerWI - 1) / rowsPerWI };
if (k.run(2, globalsize, NULL, false))
{
CV_IMPL_ADD(CV_IMPL_OCL);
return;
}
}
}
#endif
UMat src = *this; // Fake reference to itself.
// Resolves issue 8693 in case of src == dst.
Mat m = getMat(ACCESS_READ);
m.convertTo(_dst, _type, alpha, beta);
}
//
// void UMat::convertTo moved to convert.dispatch.cpp
//
UMat& UMat::setTo(InputArray _value, InputArray _mask)
{