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

Merge pull request #1579 from ilya-lavrenov:ocl_norm

Browse Source
This commit is contained in:
Andrey Pavlenko 2013-10-09 13:52:20 +04:00 committed by OpenCV Buildbot
commit 6b69a66c3c
1 changed files with 15 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
modules/ocl/src/arithm.cpp
View File

@ -364,6 +364,7 @@ Scalar cv::ocl::sum(const oclMat &src)
if (!src.clCxt->supportsFeature(FEATURE_CL_DOUBLE) && src.depth() == CV_64F) if (!src.clCxt->supportsFeature(FEATURE_CL_DOUBLE) && src.depth() == CV_64F)
{ {
CV_Error(CV_GpuNotSupported, "Selected device doesn't support double"); CV_Error(CV_GpuNotSupported, "Selected device doesn't support double");
return Scalar::all(0);
} }
static sumFunc functab[3] = static sumFunc functab[3] =
{ {
@ -372,11 +373,7 @@ Scalar cv::ocl::sum(const oclMat &src)
arithmetic_sum<double> arithmetic_sum<double>
}; };
bool hasDouble = src.clCxt->supportsFeature(FEATURE_CL_DOUBLE);
int ddepth = std::max(src.depth(), CV_32S); int ddepth = std::max(src.depth(), CV_32S);
if (!hasDouble && ddepth == CV_64F)
ddepth = CV_32F;
sumFunc func = functab[ddepth - CV_32S]; sumFunc func = functab[ddepth - CV_32S];
return func(src, SUM, ddepth); return func(src, SUM, ddepth);
} }
@ -386,7 +383,9 @@ Scalar cv::ocl::absSum(const oclMat &src)
if (!src.clCxt->supportsFeature(FEATURE_CL_DOUBLE) && src.depth() == CV_64F) if (!src.clCxt->supportsFeature(FEATURE_CL_DOUBLE) && src.depth() == CV_64F)
{ {
CV_Error(CV_GpuNotSupported, "Selected device doesn't support double"); CV_Error(CV_GpuNotSupported, "Selected device doesn't support double");
return cv::Scalar::all(0);
} }
static sumFunc functab[3] = static sumFunc functab[3] =
{ {
arithmetic_sum<int>, arithmetic_sum<int>,
@ -394,11 +393,7 @@ Scalar cv::ocl::absSum(const oclMat &src)
arithmetic_sum<double> arithmetic_sum<double>
}; };
bool hasDouble = src.clCxt->supportsFeature(FEATURE_CL_DOUBLE);
int ddepth = std::max(src.depth(), CV_32S); int ddepth = std::max(src.depth(), CV_32S);
if (!hasDouble && ddepth == CV_64F)
ddepth = CV_32F;
sumFunc func = functab[ddepth - CV_32S]; sumFunc func = functab[ddepth - CV_32S];
return func(src, ABS_SUM, ddepth); return func(src, ABS_SUM, ddepth);
} }
@ -408,17 +403,16 @@ Scalar cv::ocl::sqrSum(const oclMat &src)
if (!src.clCxt->supportsFeature(FEATURE_CL_DOUBLE) && src.depth() == CV_64F) if (!src.clCxt->supportsFeature(FEATURE_CL_DOUBLE) && src.depth() == CV_64F)
{ {
CV_Error(CV_GpuNotSupported, "Selected device doesn't support double"); CV_Error(CV_GpuNotSupported, "Selected device doesn't support double");
return cv::Scalar::all(0);
} }
static sumFunc functab[3] = static sumFunc functab[3] =
{ {
arithmetic_sum<int>, arithmetic_sum<int>,
arithmetic_sum<double>, arithmetic_sum<float>,
arithmetic_sum<double> arithmetic_sum<double>
}; };
bool hasDouble = src.clCxt->supportsFeature(FEATURE_CL_DOUBLE); int ddepth = src.depth() <= CV_32S ? CV_32S : CV_64F;
int ddepth = src.depth() <= CV_32S ? CV_32S : (hasDouble ? CV_64F : CV_32F);
sumFunc func = functab[ddepth - CV_32S]; sumFunc func = functab[ddepth - CV_32S];
return func(src, SQR_SUM, ddepth); return func(src, SQR_SUM, ddepth);
} }
@ -539,6 +533,7 @@ void cv::ocl::minMax(const oclMat &src, double *minVal, double *maxVal, const oc
if (!src.clCxt->supportsFeature(FEATURE_CL_DOUBLE) && src.depth() == CV_64F) if (!src.clCxt->supportsFeature(FEATURE_CL_DOUBLE) && src.depth() == CV_64F)
{ {
CV_Error(CV_GpuNotSupported, "Selected device doesn't support double"); CV_Error(CV_GpuNotSupported, "Selected device doesn't support double");
return;
} }
static minMaxFunc functab[] = static minMaxFunc functab[] =
@ -569,13 +564,17 @@ double cv::ocl::norm(const oclMat &src1, int normType)
return norm(src1, oclMat(), normType); return norm(src1, oclMat(), normType);
} }
static void arithm_absdiff_nonsaturate_run(const oclMat & src1, const oclMat & src2, oclMat & diff) static void arithm_absdiff_nonsaturate_run(const oclMat & src1, const oclMat & src2, oclMat & diff, int ntype)
{ {
CV_Assert(src1.step % src1.elemSize() == 0 && (src2.empty() || src2.step % src2.elemSize() == 0)); CV_Assert(src1.step % src1.elemSize() == 0 && (src2.empty() || src2.step % src2.elemSize() == 0));
Context *clCxt = src1.clCxt; Context *clCxt = src1.clCxt;
int ddepth = CV_64F; int ddepth = std::max(src1.depth(), CV_32S);
if (ntype == NORM_L2)
ddepth = std::max<int>(CV_32F, ddepth);
diff.create(src1.size(), CV_MAKE_TYPE(ddepth, src1.channels())); diff.create(src1.size(), CV_MAKE_TYPE(ddepth, src1.channels()));
CV_Assert(diff.step % diff.elemSize() == 0);
int oclChannels = src1.oclchannels(), sdepth = src1.depth(); int oclChannels = src1.oclchannels(), sdepth = src1.depth();
int src1step1 = src1.step / src1.elemSize(), src1offset1 = src1.offset / src1.elemSize(); int src1step1 = src1.step / src1.elemSize(), src1offset1 = src1.offset / src1.elemSize();
@ -638,7 +637,8 @@ double cv::ocl::norm(const oclMat &src1, const oclMat &src2, int normType)
int cn = src1.channels(); int cn = src1.channels();
double r = 0; double r = 0;
oclMat diff; oclMat diff;
arithm_absdiff_nonsaturate_run(src1, src2, diff);
arithm_absdiff_nonsaturate_run(src1, src2, diff, normType);
switch (normType) switch (normType)
{ {