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

Merge pull request #2041 from ilya-lavrenov:tapi_mul_div

Browse Source
This commit is contained in:
Andrey Pavlenko 2013-12-23 19:09:59 +04:00 committed by OpenCV Buildbot
commit 6a2fb8e5d1
3 changed files with 90 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
modules/core/src/arithm.cpp
View File

@ -915,11 +915,12 @@ void convertAndUnrollScalar( const Mat& sc, int buftype, uchar* scbuf, size_t bl
enum { OCL_OP_ADD=0, OCL_OP_SUB=1, OCL_OP_RSUB=2, OCL_OP_ABSDIFF=3, OCL_OP_MUL=4,
OCL_OP_MUL_SCALE=5, OCL_OP_DIV_SCALE=6, OCL_OP_RECIP_SCALE=7, OCL_OP_ADDW=8,
OCL_OP_AND=9, OCL_OP_OR=10, OCL_OP_XOR=11, OCL_OP_NOT=12, OCL_OP_MIN=13, OCL_OP_MAX=14 };
OCL_OP_AND=9, OCL_OP_OR=10, OCL_OP_XOR=11, OCL_OP_NOT=12, OCL_OP_MIN=13, OCL_OP_MAX=14,
OCL_OP_RDIV_SCALE=15 };
static const char* oclop2str[] = { "OP_ADD", "OP_SUB", "OP_RSUB", "OP_ABSDIFF",
"OP_MUL", "OP_MUL_SCALE", "OP_DIV_SCALE", "OP_RECIP_SCALE",
"OP_ADDW", "OP_AND", "OP_OR", "OP_XOR", "OP_NOT", "OP_MIN", "OP_MAX", 0 };
"OP_ADDW", "OP_AND", "OP_OR", "OP_XOR", "OP_NOT", "OP_MIN", "OP_MAX", "OP_RDIV_SCALE", 0 };
static bool ocl_binary_op(InputArray _src1, InputArray _src2, OutputArray _dst,
InputArray _mask, bool bitwise, int oclop, bool haveScalar )
@ -1301,25 +1302,27 @@ static bool ocl_arithm_op(InputArray _src1, InputArray _src2, OutputArray _dst,
int kercn = haveMask || haveScalar ? cn : 1;
char cvtstr[3][32], opts[1024];
char cvtstr[4][32], opts[1024];
sprintf(opts, "-D %s%s -D %s -D srcT1=%s -D srcT2=%s "
"-D dstT=%s -D workT=%s -D convertToWT1=%s "
"-D dstT=%s -D workT=%s -D scaleT=%s -D convertToWT1=%s "
"-D convertToWT2=%s -D convertToDT=%s%s",
(haveMask ? "MASK_" : ""), (haveScalar ? "UNARY_OP" : "BINARY_OP"),
oclop2str[oclop], ocl::typeToStr(CV_MAKETYPE(depth1, kercn)),
ocl::typeToStr(CV_MAKETYPE(depth2, kercn)),
ocl::typeToStr(CV_MAKETYPE(ddepth, kercn)),
ocl::typeToStr(CV_MAKETYPE(wdepth, kercn)),
ocl::typeToStr(CV_MAKETYPE(wdepth, 1)),
ocl::convertTypeStr(depth1, wdepth, kercn, cvtstr[0]),
ocl::convertTypeStr(depth2, wdepth, kercn, cvtstr[1]),
ocl::convertTypeStr(wdepth, ddepth, kercn, cvtstr[2]),
doubleSupport ? " -D DOUBLE_SUPPORT" : "");
size_t usrdata_esz = CV_ELEM_SIZE(wdepth);
const uchar* usrdata_p = (const uchar*)usrdata;
const double* usrdata_d = (const double*)usrdata;
float usrdata_f[3];
int i, n = oclop == OCL_OP_MUL_SCALE || oclop == OCL_OP_DIV_SCALE ||
oclop == OCL_OP_RECIP_SCALE ? 1 : oclop == OCL_OP_ADDW ? 3 : 0;
oclop == OCL_OP_RDIV_SCALE || oclop == OCL_OP_RECIP_SCALE ? 1 : oclop == OCL_OP_ADDW ? 3 : 0;
if( n > 0 && wdepth == CV_32F )
{
for( i = 0; i < n; i++ )
@ -1352,13 +1355,20 @@ static bool ocl_arithm_op(InputArray _src1, InputArray _src2, OutputArray _dst,
ocl::KernelArg scalararg = ocl::KernelArg(0, 0, 0, buf, esz);
if( !haveMask )
k.args(src1arg, dstarg, scalararg);
{
if(n == 0)
k.args(src1arg, dstarg, scalararg);
else if(n == 1)
k.args(src1arg, dstarg, scalararg,
ocl::KernelArg(0, 0, 0, usrdata_p, usrdata_esz));
else
CV_Error(Error::StsNotImplemented, "unsupported number of extra parameters");
}
else
k.args(src1arg, maskarg, dstarg, scalararg);
}
else
{
size_t usrdata_esz = CV_ELEM_SIZE(wdepth);
src2 = _src2.getUMat();
ocl::KernelArg src2arg = ocl::KernelArg::ReadOnlyNoSize(src2, cscale);
@ -1439,6 +1449,8 @@ static void arithm_op(InputArray _src1, InputArray _src2, OutputArray _dst,
swapped12 = true;
if( oclop == OCL_OP_SUB )
oclop = OCL_OP_RSUB;
if ( oclop == OCL_OP_DIV_SCALE )
oclop = OCL_OP_RDIV_SCALE;
}
else if( !checkScalar(*psrc2, type1, kind2, kind1) )
CV_Error( CV_StsUnmatchedSizes,

34
modules/core/src/opencl/arithm.cl
View File

@ -136,8 +136,12 @@
#elif defined OP_MUL_SCALE
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT scale
#define PROCESS_ELEM dstelem = convertToDT(srcelem1 * srcelem2 * scale)
#ifdef UNARY_OP
#define EXTRA_PARAMS , workT srcelem2, scaleT scale
#else
#define EXTRA_PARAMS , scaleT scale
#endif
#define PROCESS_ELEM dstelem = convertToDT(srcelem1 * scale * srcelem2)
#elif defined OP_DIV
#define PROCESS_ELEM \
@ -146,21 +150,36 @@
#elif defined OP_DIV_SCALE
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT scale
#ifdef UNARY_OP
#define EXTRA_PARAMS , workT srcelem2, scaleT scale
#else
#define EXTRA_PARAMS , scaleT scale
#endif
#define PROCESS_ELEM \
workT e2 = srcelem2, zero = (workT)(0); \
dstelem = convertToDT(e2 != zero ? srcelem1 * scale / e2 : zero)
dstelem = convertToDT(e2 == zero ? zero : (srcelem1 * (workT)(scale) / e2))
#elif defined OP_RDIV_SCALE
#undef EXTRA_PARAMS
#ifdef UNARY_OP
#define EXTRA_PARAMS , workT srcelem2, scaleT scale
#else
#define EXTRA_PARAMS , scaleT scale
#endif
#define PROCESS_ELEM \
workT e1 = srcelem1, zero = (workT)(0); \
dstelem = convertToDT(e1 == zero ? zero : (srcelem2 * (workT)(scale) / e1))
#elif defined OP_RECIP_SCALE
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT scale
#define EXTRA_PARAMS , scaleT scale
#define PROCESS_ELEM \
workT e1 = srcelem1, zero = (workT)(0); \
dstelem = convertToDT(e1 != zero ? scale / e1 : zero)
#elif defined OP_ADDW
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT alpha, workT beta, workT gamma
#define EXTRA_PARAMS , scaleT alpha, scaleT beta, scaleT gamma
#define PROCESS_ELEM dstelem = convertToDT(srcelem1*alpha + srcelem2*beta + gamma)
#elif defined OP_MAG
@ -260,7 +279,8 @@ dstelem = v > (dstT)(0) ? log(v) : log(-v)
#undef srcelem2
#if defined OP_AND || defined OP_OR || defined OP_XOR || defined OP_ADD || defined OP_SAT_ADD || \
defined OP_SUB || defined OP_SAT_SUB || defined OP_RSUB || defined OP_SAT_RSUB || \
defined OP_ABSDIFF || defined OP_CMP || defined OP_MIN || defined OP_MAX || defined OP_POW
defined OP_ABSDIFF || defined OP_CMP || defined OP_MIN || defined OP_MAX || defined OP_POW || \
defined OP_MUL || defined OP_DIV
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT srcelem2
#endif

49
modules/core/test/ocl/test_arithm.cpp
View File

@ -293,7 +293,7 @@ OCL_TEST_P(Mul, Mat)
}
}
OCL_TEST_P(Mul, DISABLED_Scalar)
OCL_TEST_P(Mul, Scalar)
{
for (int j = 0; j < test_loop_times; j++)
{
@ -306,7 +306,7 @@ OCL_TEST_P(Mul, DISABLED_Scalar)
}
}
OCL_TEST_P(Mul, DISABLED_Mat_Scale)
OCL_TEST_P(Mul, Mat_Scale)
{
for (int j = 0; j < test_loop_times; j++)
{
@ -319,6 +319,20 @@ OCL_TEST_P(Mul, DISABLED_Mat_Scale)
}
}
OCL_TEST_P(Mul, Mat_Scalar_Scale)
{
for (int j = 0; j < test_loop_times; j++)
{
generateTestData();
OCL_OFF(cv::multiply(src1_roi, val, dst1_roi, val[0]));
OCL_ON(cv::multiply(usrc1_roi, val, udst1_roi, val[0]));
Near(udst1_roi.depth() >= CV_32F ? 1e-2 : 1);
}
}
//////////////////////////////// Div /////////////////////////////////////////////////
typedef ArithmTestBase Div;
@ -335,7 +349,7 @@ OCL_TEST_P(Div, Mat)
}
}
OCL_TEST_P(Div, DISABLED_Scalar)
OCL_TEST_P(Div, Scalar)
{
for (int j = 0; j < test_loop_times; j++)
{
@ -348,6 +362,19 @@ OCL_TEST_P(Div, DISABLED_Scalar)
}
}
OCL_TEST_P(Div, Scalar2)
{
for (int j = 0; j < test_loop_times; j++)
{
generateTestData();
OCL_OFF(cv::divide(src1_roi, val, dst1_roi));
OCL_ON(cv::divide(usrc1_roi, val, udst1_roi));
Near(udst1_roi.depth() >= CV_32F ? 1e-3 : 1);
}
}
OCL_TEST_P(Div, Mat_Scale)
{
for (int j = 0; j < test_loop_times; j++)
@ -361,8 +388,7 @@ OCL_TEST_P(Div, Mat_Scale)
}
}
OCL_TEST_P(Div, DISABLED_Mat_Scalar_Scale)
OCL_TEST_P(Div, Mat_Scalar_Scale)
{
for (int j = 0; j < test_loop_times; j++)
{
@ -375,6 +401,19 @@ OCL_TEST_P(Div, DISABLED_Mat_Scalar_Scale)
}
}
OCL_TEST_P(Div, Recip)
{
for (int j = 0; j < test_loop_times; j++)
{
generateTestData();
OCL_OFF(cv::divide(val[0], src1_roi, dst1_roi));
OCL_ON(cv::divide(val[0], usrc1_roi, udst1_roi));
Near(udst1_roi.depth() >= CV_32F ? 1e-3 : 1);
}
}
//////////////////////////////// Min/Max /////////////////////////////////////////////////
typedef ArithmTestBase Min;