Skip to content

O
opencv
Commit 7726e273 authored by yao's avatar yao
Browse files
Options

merge add and sub into one set of kernels

parent bee970ab
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 86 additions and 61 deletions
modules/ocl/src/arithm.cpp
View file @ 7726e273
...@@ -92,9 +92,6 @@ namespace cv ...@@ -92,9 +92,6 @@ namespace cv
extern const char *arithm_bitwise_xor_scalar_mask; extern const char *arithm_bitwise_xor_scalar_mask;
extern const char *arithm_compare_eq; extern const char *arithm_compare_eq;
extern const char *arithm_compare_ne; extern const char *arithm_compare_ne;
extern const char *arithm_sub;
extern const char *arithm_sub_scalar;
extern const char *arithm_sub_scalar_mask;
extern const char *arithm_mul; extern const char *arithm_mul;
extern const char *arithm_div; extern const char *arithm_div;
extern const char *arithm_absdiff; extern const char *arithm_absdiff;
...@@ -130,7 +127,8 @@ inline int divUp(int total, int grain) ...@@ -130,7 +127,8 @@ inline int divUp(int total, int grain)
/////////////////////// add subtract multiply divide ///////////////////////// /////////////////////// add subtract multiply divide /////////////////////////
////////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////////
template<typename T> template<typename T>
void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, string kernelName, const char **kernelString, void *_scalar) void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
string kernelName, const char **kernelString, void *_scalar, int op_type = 0)
{ {
if(!src1.clCxt->supportsFeature(Context::CL_DOUBLE) && src1.type() == CV_64F) if(!src1.clCxt->supportsFeature(Context::CL_DOUBLE) && src1.type() == CV_64F)
{ {
...@@ -186,14 +184,25 @@ void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, string ...@@ -186,14 +184,25 @@ void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, string
scalar = (T)scalar1; scalar = (T)scalar1;
args.push_back( make_pair( sizeof(T), (void *)&scalar )); args.push_back( make_pair( sizeof(T), (void *)&scalar ));
} }
switch(op_type)
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, -1, depth); {
case MAT_ADD:
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, -1, depth, "-D ARITHM_ADD");
break;
case MAT_SUB:
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, -1, depth, "-D ARITHM_SUB");
break;
default:
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, -1, depth);
}
} }
static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, string kernelName, const char **kernelString) static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
string kernelName, const char **kernelString, int op_type = 0)
{ {
arithmetic_run<char>(src1, src2, dst, kernelName, kernelString, (void *)NULL); arithmetic_run<char>(src1, src2, dst, kernelName, kernelString, (void *)NULL, op_type);
} }
static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask, string kernelName, const char **kernelString) static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask,
string kernelName, const char **kernelString, int op_type = 0)
{ {
if(!src1.clCxt->supportsFeature(Context::CL_DOUBLE) && src1.type() == CV_64F) if(!src1.clCxt->supportsFeature(Context::CL_DOUBLE) && src1.type() == CV_64F)
{ {
...@@ -248,24 +257,34 @@ static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, ...@@ -248,24 +257,34 @@ static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
args.push_back( make_pair( sizeof(cl_int), (void *)&cols )); args.push_back( make_pair( sizeof(cl_int), (void *)&cols ));
args.push_back( make_pair( sizeof(cl_int), (void *)&dst_step1 )); args.push_back( make_pair( sizeof(cl_int), (void *)&dst_step1 ));
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth); switch (op_type)
{
case MAT_ADD:
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth, "-D ARITHM_ADD");
break;
case MAT_SUB:
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth, "-D ARITHM_SUB");
break;
default:
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth);
}
} }
void cv::ocl::add(const oclMat &src1, const oclMat &src2, oclMat &dst) void cv::ocl::add(const oclMat &src1, const oclMat &src2, oclMat &dst)
{ {
arithmetic_run(src1, src2, dst, "arithm_add", &arithm_add); arithmetic_run(src1, src2, dst, "arithm_add", &arithm_add, MAT_ADD);
} }
void cv::ocl::add(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask) void cv::ocl::add(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask)
{ {
arithmetic_run(src1, src2, dst, mask, "arithm_add_with_mask", &arithm_add); arithmetic_run(src1, src2, dst, mask, "arithm_add_with_mask", &arithm_add, MAT_ADD);
} }
void cv::ocl::subtract(const oclMat &src1, const oclMat &src2, oclMat &dst) void cv::ocl::subtract(const oclMat &src1, const oclMat &src2, oclMat &dst)
{ {
arithmetic_run(src1, src2, dst, "arithm_sub", &arithm_sub); arithmetic_run(src1, src2, dst, "arithm_add", &arithm_add, MAT_SUB);
} }
void cv::ocl::subtract(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask) void cv::ocl::subtract(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask)
{ {
arithmetic_run(src1, src2, dst, mask, "arithm_sub_with_mask", &arithm_sub); arithmetic_run(src1, src2, dst, mask, "arithm_add_with_mask", &arithm_add, MAT_SUB);
} }
typedef void (*MulDivFunc)(const oclMat &src1, const oclMat &src2, oclMat &dst, string kernelName, typedef void (*MulDivFunc)(const oclMat &src1, const oclMat &src2, oclMat &dst, string kernelName,
const char **kernelString, void *scalar); const char **kernelString, void *scalar);
...@@ -351,12 +370,9 @@ void arithmetic_scalar_run(const oclMat &src1, const Scalar &src2, oclMat &dst, ...@@ -351,12 +370,9 @@ void arithmetic_scalar_run(const oclMat &src1, const Scalar &src2, oclMat &dst,
args.push_back( make_pair( sizeof(cl_int) , (void *)&cols )); args.push_back( make_pair( sizeof(cl_int) , (void *)&cols ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&dst_step1 )); args.push_back( make_pair( sizeof(cl_int) , (void *)&dst_step1 ));
if(isMatSubScalar != 0) if(isMatSubScalar != 0)
{ openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth, "-D ARITHM_SUB");
isMatSubScalar = isMatSubScalar > 0 ? 1 : 0; else
args.push_back( make_pair( sizeof(cl_int) , (void *)&isMatSubScalar)); openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth, "-D ARITHM_ADD");
}
openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth);
} }
static void arithmetic_scalar_run(const oclMat &src, oclMat &dst, string kernelName, const char **kernelString, double scalar) static void arithmetic_scalar_run(const oclMat &src, oclMat &dst, string kernelName, const char **kernelString, double scalar)
...@@ -452,14 +468,14 @@ void cv::ocl::add(const oclMat &src1, const Scalar &src2, oclMat &dst, const ocl ...@@ -452,14 +468,14 @@ void cv::ocl::add(const oclMat &src1, const Scalar &src2, oclMat &dst, const ocl
void cv::ocl::subtract(const oclMat &src1, const Scalar &src2, oclMat &dst, const oclMat &mask) void cv::ocl::subtract(const oclMat &src1, const Scalar &src2, oclMat &dst, const oclMat &mask)
{ {
string kernelName = mask.data ? "arithm_s_sub_with_mask" : "arithm_s_sub"; string kernelName = mask.data ? "arithm_s_add_with_mask" : "arithm_s_add";
const char **kernelString = mask.data ? &arithm_sub_scalar_mask : &arithm_sub_scalar; const char **kernelString = mask.data ? &arithm_add_scalar_mask : &arithm_add_scalar;
arithmetic_scalar( src1, src2, dst, mask, kernelName, kernelString, 1); arithmetic_scalar( src1, src2, dst, mask, kernelName, kernelString, 1);
} }
void cv::ocl::subtract(const Scalar &src2, const oclMat &src1, oclMat &dst, const oclMat &mask) void cv::ocl::subtract(const Scalar &src2, const oclMat &src1, oclMat &dst, const oclMat &mask)
{ {
string kernelName = mask.data ? "arithm_s_sub_with_mask" : "arithm_s_sub"; string kernelName = mask.data ? "arithm_s_add_with_mask" : "arithm_s_add";
const char **kernelString = mask.data ? &arithm_sub_scalar_mask : &arithm_sub_scalar; const char **kernelString = mask.data ? &arithm_add_scalar_mask : &arithm_add_scalar;
arithmetic_scalar( src1, src2, dst, mask, kernelName, kernelString, -1); arithmetic_scalar( src1, src2, dst, mask, kernelName, kernelString, -1);
} }
void cv::ocl::divide(double scalar, const oclMat &src, oclMat &dst) void cv::ocl::divide(double scalar, const oclMat &src, oclMat &dst)
......
modules/ocl/src/opencl/arithm_add.cl
View file @ 7726e273
This diff is collapsed.
modules/ocl/src/opencl/arithm_add_scalar.cl
View file @ 7726e273
...@@ -49,7 +49,12 @@ ...@@ -49,7 +49,12 @@
#elif defined (cl_amd_fp64) #elif defined (cl_amd_fp64)
#pragma OPENCL EXTENSION cl_amd_fp64:enable #pragma OPENCL EXTENSION cl_amd_fp64:enable
#endif #endif
#endif
#ifdef ARITHM_ADD
#define ARITHM_OP(A,B) ((A)+(B))
#elif defined ARITHM_SUB
#define ARITHM_OP(A,B) ((A)-(B))
#endif #endif
/**************************************add with scalar without mask**************************************/ /**************************************add with scalar without mask**************************************/
__kernel void arithm_s_add_C1_D0 (__global uchar *src1, int src1_step, int src1_offset, __kernel void arithm_s_add_C1_D0 (__global uchar *src1, int src1_step, int src1_offset,
...@@ -83,7 +88,7 @@ __kernel void arithm_s_add_C1_D0 (__global uchar *src1, int src1_step, int src ...@@ -83,7 +88,7 @@ __kernel void arithm_s_add_C1_D0 (__global uchar *src1, int src1_step, int src
} }
uchar4 data = *((__global uchar4 *)(dst + dst_index)); uchar4 data = *((__global uchar4 *)(dst + dst_index));
int4 tmp = convert_int4_sat(src1_data) + src2_data; int4 tmp = ARITHM_OP(convert_int4_sat(src1_data), src2_data);
uchar4 tmp_data = convert_uchar4_sat(tmp); uchar4 tmp_data = convert_uchar4_sat(tmp);
data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x; data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
...@@ -120,7 +125,7 @@ __kernel void arithm_s_add_C1_D2 (__global ushort *src1, int src1_step, int sr ...@@ -120,7 +125,7 @@ __kernel void arithm_s_add_C1_D2 (__global ushort *src1, int src1_step, int sr
int2 src2_data = (int2)(src2.x, src2.x); int2 src2_data = (int2)(src2.x, src2.x);
ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index)); ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index));
int2 tmp = convert_int2_sat(src1_data) + src2_data; int2 tmp = ARITHM_OP(convert_int2_sat(src1_data), src2_data);
ushort2 tmp_data = convert_ushort2_sat(tmp); ushort2 tmp_data = convert_ushort2_sat(tmp);
data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x; data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x;
...@@ -155,7 +160,7 @@ __kernel void arithm_s_add_C1_D3 (__global short *src1, int src1_step, int src ...@@ -155,7 +160,7 @@ __kernel void arithm_s_add_C1_D3 (__global short *src1, int src1_step, int src
int2 src2_data = (int2)(src2.x, src2.x); int2 src2_data = (int2)(src2.x, src2.x);
short2 data = *((__global short2 *)((__global uchar *)dst + dst_index)); short2 data = *((__global short2 *)((__global uchar *)dst + dst_index));
int2 tmp = convert_int2_sat(src1_data) + src2_data; int2 tmp = ARITHM_OP(convert_int2_sat(src1_data), src2_data);
short2 tmp_data = convert_short2_sat(tmp); short2 tmp_data = convert_short2_sat(tmp);
data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x; data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x;
...@@ -181,7 +186,7 @@ __kernel void arithm_s_add_C1_D4 (__global int *src1, int src1_step, int src1_ ...@@ -181,7 +186,7 @@ __kernel void arithm_s_add_C1_D4 (__global int *src1, int src1_step, int src1_
int src_data2 = src2.x; int src_data2 = src2.x;
int dst_data = *((__global int *)((__global char *)dst + dst_index)); int dst_data = *((__global int *)((__global char *)dst + dst_index));
int data = convert_int_sat((long)src_data1 + (long)src_data2); int data = convert_int_sat(ARITHM_OP((long)src_data1, (long)src_data2));
*((__global int *)((__global char *)dst + dst_index)) = data; *((__global int *)((__global char *)dst + dst_index)) = data;
} }
...@@ -203,7 +208,7 @@ __kernel void arithm_s_add_C1_D5 (__global float *src1, int src1_step, int src ...@@ -203,7 +208,7 @@ __kernel void arithm_s_add_C1_D5 (__global float *src1, int src1_step, int src
float src_data2 = src2.x; float src_data2 = src2.x;
float dst_data = *((__global float *)((__global char *)dst + dst_index)); float dst_data = *((__global float *)((__global char *)dst + dst_index));
float data = src_data1 + src_data2; float data = ARITHM_OP(src_data1, src_data2);
*((__global float *)((__global char *)dst + dst_index)) = data; *((__global float *)((__global char *)dst + dst_index)) = data;
} }
...@@ -227,7 +232,7 @@ __kernel void arithm_s_add_C1_D6 (__global double *src1, int src1_step, int sr ...@@ -227,7 +232,7 @@ __kernel void arithm_s_add_C1_D6 (__global double *src1, int src1_step, int sr
double src2_data = src2.x; double src2_data = src2.x;
double dst_data = *((__global double *)((__global char *)dst + dst_index)); double dst_data = *((__global double *)((__global char *)dst + dst_index));
double data = src_data1 + src2_data; double data = ARITHM_OP(src_data1, src2_data);
*((__global double *)((__global char *)dst + dst_index)) = data; *((__global double *)((__global char *)dst + dst_index)) = data;
} }
...@@ -260,7 +265,7 @@ __kernel void arithm_s_add_C2_D0 (__global uchar *src1, int src1_step, int src ...@@ -260,7 +265,7 @@ __kernel void arithm_s_add_C2_D0 (__global uchar *src1, int src1_step, int src
int4 src2_data = (int4)(src2.x, src2.y, src2.x, src2.y); int4 src2_data = (int4)(src2.x, src2.y, src2.x, src2.y);
uchar4 data = *((__global uchar4 *)(dst + dst_index)); uchar4 data = *((__global uchar4 *)(dst + dst_index));
int4 tmp = convert_int4_sat(src1_data) + src2_data; int4 tmp = ARITHM_OP(convert_int4_sat(src1_data), src2_data);
uchar4 tmp_data = convert_uchar4_sat(tmp); uchar4 tmp_data = convert_uchar4_sat(tmp);
data.xy = (dst_index + 0 >= dst_start) ? tmp_data.xy : data.xy; data.xy = (dst_index + 0 >= dst_start) ? tmp_data.xy : data.xy;
...@@ -286,7 +291,7 @@ __kernel void arithm_s_add_C2_D2 (__global ushort *src1, int src1_step, int sr ...@@ -286,7 +291,7 @@ __kernel void arithm_s_add_C2_D2 (__global ushort *src1, int src1_step, int sr
int2 src_data2 = (int2)(src2.x, src2.y); int2 src_data2 = (int2)(src2.x, src2.y);
ushort2 dst_data = *((__global ushort2 *)((__global char *)dst + dst_index)); ushort2 dst_data = *((__global ushort2 *)((__global char *)dst + dst_index));
int2 tmp = convert_int2_sat(src_data1) + src_data2; int2 tmp = ARITHM_OP(convert_int2_sat(src_data1), src_data2);
ushort2 data = convert_ushort2_sat(tmp); ushort2 data = convert_ushort2_sat(tmp);
*((__global ushort2 *)((__global char *)dst + dst_index)) = data; *((__global ushort2 *)((__global char *)dst + dst_index)) = data;
...@@ -309,7 +314,7 @@ __kernel void arithm_s_add_C2_D3 (__global short *src1, int src1_step, int src ...@@ -309,7 +314,7 @@ __kernel void arithm_s_add_C2_D3 (__global short *src1, int src1_step, int src
int2 src_data2 = (int2)(src2.x, src2.y); int2 src_data2 = (int2)(src2.x, src2.y);
short2 dst_data = *((__global short2 *)((__global char *)dst + dst_index)); short2 dst_data = *((__global short2 *)((__global char *)dst + dst_index));
int2 tmp = convert_int2_sat(src_data1) + src_data2; int2 tmp = ARITHM_OP(convert_int2_sat(src_data1), src_data2);
short2 data = convert_short2_sat(tmp); short2 data = convert_short2_sat(tmp);
*((__global short2 *)((__global char *)dst + dst_index)) = data; *((__global short2 *)((__global char *)dst + dst_index)) = data;
...@@ -332,7 +337,7 @@ __kernel void arithm_s_add_C2_D4 (__global int *src1, int src1_step, int src1_ ...@@ -332,7 +337,7 @@ __kernel void arithm_s_add_C2_D4 (__global int *src1, int src1_step, int src1_
int2 src_data2 = (int2)(src2.x, src2.y); int2 src_data2 = (int2)(src2.x, src2.y);
int2 dst_data = *((__global int2 *)((__global char *)dst + dst_index)); int2 dst_data = *((__global int2 *)((__global char *)dst + dst_index));
int2 data = convert_int2_sat(convert_long2_sat(src_data1) + convert_long2_sat(src_data2)); int2 data = convert_int2_sat(ARITHM_OP(convert_long2_sat(src_data1), convert_long2_sat(src_data2)));
*((__global int2 *)((__global char *)dst + dst_index)) = data; *((__global int2 *)((__global char *)dst + dst_index)) = data;
} }
} }
...@@ -353,7 +358,7 @@ __kernel void arithm_s_add_C2_D5 (__global float *src1, int src1_step, int src ...@@ -353,7 +358,7 @@ __kernel void arithm_s_add_C2_D5 (__global float *src1, int src1_step, int src
float2 src_data2 = (float2)(src2.x, src2.y); float2 src_data2 = (float2)(src2.x, src2.y);
float2 dst_data = *((__global float2 *)((__global char *)dst + dst_index)); float2 dst_data = *((__global float2 *)((__global char *)dst + dst_index));
float2 data = src_data1 + src_data2; float2 data = ARITHM_OP(src_data1, src_data2);
*((__global float2 *)((__global char *)dst + dst_index)) = data; *((__global float2 *)((__global char *)dst + dst_index)) = data;
} }
} }
...@@ -376,7 +381,7 @@ __kernel void arithm_s_add_C2_D6 (__global double *src1, int src1_step, int sr ...@@ -376,7 +381,7 @@ __kernel void arithm_s_add_C2_D6 (__global double *src1, int src1_step, int sr
double2 src_data2 = (double2)(src2.x, src2.y); double2 src_data2 = (double2)(src2.x, src2.y);
double2 dst_data = *((__global double2 *)((__global char *)dst + dst_index)); double2 dst_data = *((__global double2 *)((__global char *)dst + dst_index));
double2 data = src_data1 + src_data2; double2 data = ARITHM_OP(src_data1, src_data2);
*((__global double2 *)((__global char *)dst + dst_index)) = data; *((__global double2 *)((__global char *)dst + dst_index)) = data;
} }
...@@ -398,7 +403,7 @@ __kernel void arithm_s_add_C4_D0 (__global uchar *src1, int src1_step, int src ...@@ -398,7 +403,7 @@ __kernel void arithm_s_add_C4_D0 (__global uchar *src1, int src1_step, int src
uchar4 src_data1 = *((__global uchar4 *)(src1 + src1_index)); uchar4 src_data1 = *((__global uchar4 *)(src1 + src1_index));
uchar4 data = convert_uchar4_sat(convert_int4_sat(src_data1) + src2); uchar4 data = convert_uchar4_sat(ARITHM_OP(convert_int4_sat(src_data1), src2));
*((__global uchar4 *)(dst + dst_index)) = data; *((__global uchar4 *)(dst + dst_index)) = data;
} }
...@@ -418,7 +423,7 @@ __kernel void arithm_s_add_C4_D2 (__global ushort *src1, int src1_step, int sr ...@@ -418,7 +423,7 @@ __kernel void arithm_s_add_C4_D2 (__global ushort *src1, int src1_step, int sr
ushort4 src_data1 = *((__global ushort4 *)((__global char *)src1 + src1_index)); ushort4 src_data1 = *((__global ushort4 *)((__global char *)src1 + src1_index));
ushort4 data = convert_ushort4_sat(convert_int4_sat(src_data1) + src2); ushort4 data = convert_ushort4_sat(ARITHM_OP(convert_int4_sat(src_data1), src2));
*((__global ushort4 *)((__global char *)dst + dst_index)) = data; *((__global ushort4 *)((__global char *)dst + dst_index)) = data;
} }
...@@ -438,7 +443,7 @@ __kernel void arithm_s_add_C4_D3 (__global short *src1, int src1_step, int src ...@@ -438,7 +443,7 @@ __kernel void arithm_s_add_C4_D3 (__global short *src1, int src1_step, int src
short4 src_data1 = *((__global short4 *)((__global char *)src1 + src1_index)); short4 src_data1 = *((__global short4 *)((__global char *)src1 + src1_index));
short4 data = convert_short4_sat(convert_int4_sat(src_data1) + src2); short4 data = convert_short4_sat(ARITHM_OP(convert_int4_sat(src_data1), src2));
*((__global short4 *)((__global char *)dst + dst_index)) = data; *((__global short4 *)((__global char *)dst + dst_index)) = data;
} }
...@@ -458,7 +463,7 @@ __kernel void arithm_s_add_C4_D4 (__global int *src1, int src1_step, int src1_ ...@@ -458,7 +463,7 @@ __kernel void arithm_s_add_C4_D4 (__global int *src1, int src1_step, int src1_
int4 src_data1 = *((__global int4 *)((__global char *)src1 + src1_index)); int4 src_data1 = *((__global int4 *)((__global char *)src1 + src1_index));
int4 data = convert_int4_sat(convert_long4_sat(src_data1) + convert_long4_sat(src2)); int4 data = convert_int4_sat(ARITHM_OP(convert_long4_sat(src_data1), convert_long4_sat(src2)));
*((__global int4 *)((__global char *)dst + dst_index)) = data; *((__global int4 *)((__global char *)dst + dst_index)) = data;
} }
...@@ -478,7 +483,7 @@ __kernel void arithm_s_add_C4_D5 (__global float *src1, int src1_step, int src ...@@ -478,7 +483,7 @@ __kernel void arithm_s_add_C4_D5 (__global float *src1, int src1_step, int src
float4 src_data1 = *((__global float4 *)((__global char *)src1 + src1_index)); float4 src_data1 = *((__global float4 *)((__global char *)src1 + src1_index));
float4 data = src_data1 + src2; float4 data = ARITHM_OP(src_data1, src2);
*((__global float4 *)((__global char *)dst + dst_index)) = data; *((__global float4 *)((__global char *)dst + dst_index)) = data;
} }
...@@ -500,7 +505,7 @@ __kernel void arithm_s_add_C4_D6 (__global double *src1, int src1_step, int sr ...@@ -500,7 +505,7 @@ __kernel void arithm_s_add_C4_D6 (__global double *src1, int src1_step, int sr
double4 src_data1 = *((__global double4 *)((__global char *)src1 + src1_index)); double4 src_data1 = *((__global double4 *)((__global char *)src1 + src1_index));
double4 data = src_data1 + src2; double4 data = ARITHM_OP(src_data1, src2);
*((__global double4 *)((__global char *)dst + dst_index)) = data; *((__global double4 *)((__global char *)dst + dst_index)) = data;
} }
......
modules/ocl/src/opencl/arithm_add_scalar_mask.cl
View file @ 7726e273
...@@ -51,6 +51,11 @@ ...@@ -51,6 +51,11 @@
#endif #endif
#endif #endif
#ifdef ARITHM_ADD
#define ARITHM_OP(A,B) ((A)+(B))
#elif defined ARITHM_SUB
#define ARITHM_OP(A,B) ((A)-(B))
#endif
/**************************************add with scalar with mask**************************************/ /**************************************add with scalar with mask**************************************/
__kernel void arithm_s_add_with_mask_C1_D0 (__global uchar *src1, int src1_step, int src1_offset, __kernel void arithm_s_add_with_mask_C1_D0 (__global uchar *src1, int src1_step, int src1_offset,
__global uchar *dst, int dst_step, int dst_offset, __global uchar *dst, int dst_step, int dst_offset,
...@@ -94,7 +99,7 @@ __kernel void arithm_s_add_with_mask_C1_D0 (__global uchar *src1, int src1_ste ...@@ -94,7 +99,7 @@ __kernel void arithm_s_add_with_mask_C1_D0 (__global uchar *src1, int src1_ste
} }
uchar4 data = *((__global uchar4 *)(dst + dst_index)); uchar4 data = *((__global uchar4 *)(dst + dst_index));
int4 tmp = convert_int4_sat(src1_data) + src2_data; int4 tmp = ARITHM_OP(convert_int4_sat(src1_data), src2_data);
uchar4 tmp_data = convert_uchar4_sat(tmp); uchar4 tmp_data = convert_uchar4_sat(tmp);
data.x = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x; data.x = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
...@@ -134,7 +139,7 @@ __kernel void arithm_s_add_with_mask_C1_D2 (__global ushort *src1, int src1_st ...@@ -134,7 +139,7 @@ __kernel void arithm_s_add_with_mask_C1_D2 (__global ushort *src1, int src1_st
uchar2 mask_data = vload2(0, mask + mask_index); uchar2 mask_data = vload2(0, mask + mask_index);
ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index)); ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index));
int2 tmp = convert_int2_sat(src1_data) + src2_data; int2 tmp = ARITHM_OP(convert_int2_sat(src1_data), src2_data);
ushort2 tmp_data = convert_ushort2_sat(tmp); ushort2 tmp_data = convert_ushort2_sat(tmp);
data.x = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.x : data.x; data.x = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.x : data.x;
...@@ -172,7 +177,7 @@ __kernel void arithm_s_add_with_mask_C1_D3 (__global short *src1, int src1_ste ...@@ -172,7 +177,7 @@ __kernel void arithm_s_add_with_mask_C1_D3 (__global short *src1, int src1_ste
uchar2 mask_data = vload2(0, mask + mask_index); uchar2 mask_data = vload2(0, mask + mask_index);
short2 data = *((__global short2 *)((__global uchar *)dst + dst_index)); short2 data = *((__global short2 *)((__global uchar *)dst + dst_index));
int2 tmp = convert_int2_sat(src1_data) + src2_data; int2 tmp = ARITHM_OP(convert_int2_sat(src1_data), src2_data);
short2 tmp_data = convert_short2_sat(tmp); short2 tmp_data = convert_short2_sat(tmp);
data.x = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.x : data.x; data.x = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.x : data.x;
...@@ -202,7 +207,7 @@ __kernel void arithm_s_add_with_mask_C1_D4 (__global int *src1, int src1_ste ...@@ -202,7 +207,7 @@ __kernel void arithm_s_add_with_mask_C1_D4 (__global int *src1, int src1_ste
int src_data2 = src2.x; int src_data2 = src2.x;
int dst_data = *((__global int *)((__global char *)dst + dst_index)); int dst_data = *((__global int *)((__global char *)dst + dst_index));
int data = convert_int_sat((long)src_data1 + (long)src_data2); int data = convert_int_sat(ARITHM_OP((long)src_data1, (long)src_data2));
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global int *)((__global char *)dst + dst_index)) = data; *((__global int *)((__global char *)dst + dst_index)) = data;
...@@ -230,7 +235,7 @@ __kernel void arithm_s_add_with_mask_C1_D5 (__global float *src1, int src1_s ...@@ -230,7 +235,7 @@ __kernel void arithm_s_add_with_mask_C1_D5 (__global float *src1, int src1_s
float src_data2 = src2.x; float src_data2 = src2.x;
float dst_data = *((__global float *)((__global char *)dst + dst_index)); float dst_data = *((__global float *)((__global char *)dst + dst_index));
float data = src_data1 + src_data2; float data = ARITHM_OP(src_data1, src_data2);
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global float *)((__global char *)dst + dst_index)) = data; *((__global float *)((__global char *)dst + dst_index)) = data;
...@@ -260,7 +265,7 @@ __kernel void arithm_s_add_with_mask_C1_D6 (__global double *src1, int src1_ ...@@ -260,7 +265,7 @@ __kernel void arithm_s_add_with_mask_C1_D6 (__global double *src1, int src1_
double src_data2 = src2.x; double src_data2 = src2.x;
double dst_data = *((__global double *)((__global char *)dst + dst_index)); double dst_data = *((__global double *)((__global char *)dst + dst_index));
double data = src_data1 + src_data2; double data = ARITHM_OP(src_data1, src_data2);
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global double *)((__global char *)dst + dst_index)) = data; *((__global double *)((__global char *)dst + dst_index)) = data;
...@@ -296,7 +301,7 @@ __kernel void arithm_s_add_with_mask_C2_D0 (__global uchar *src1, int src1_ste ...@@ -296,7 +301,7 @@ __kernel void arithm_s_add_with_mask_C2_D0 (__global uchar *src1, int src1_ste
uchar2 mask_data = vload2(0, mask + mask_index); uchar2 mask_data = vload2(0, mask + mask_index);
uchar4 data = *((__global uchar4 *)(dst + dst_index)); uchar4 data = *((__global uchar4 *)(dst + dst_index));
int4 tmp = convert_int4_sat(src1_data) + src2_data; int4 tmp = ARITHM_OP(convert_int4_sat(src1_data), src2_data);
uchar4 tmp_data = convert_uchar4_sat(tmp); uchar4 tmp_data = convert_uchar4_sat(tmp);
data.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.xy : data.xy; data.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.xy : data.xy;
...@@ -326,7 +331,7 @@ __kernel void arithm_s_add_with_mask_C2_D2 (__global ushort *src1, int src1_st ...@@ -326,7 +331,7 @@ __kernel void arithm_s_add_with_mask_C2_D2 (__global ushort *src1, int src1_st
int2 src_data2 = (int2)(src2.x, src2.y); int2 src_data2 = (int2)(src2.x, src2.y);
ushort2 dst_data = *((__global ushort2 *)((__global char *)dst + dst_index)); ushort2 dst_data = *((__global ushort2 *)((__global char *)dst + dst_index));
int2 tmp = convert_int2_sat(src_data1) + src_data2; int2 tmp = ARITHM_OP(convert_int2_sat(src_data1), src_data2);
ushort2 data = convert_ushort2_sat(tmp); ushort2 data = convert_ushort2_sat(tmp);
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
...@@ -354,7 +359,7 @@ __kernel void arithm_s_add_with_mask_C2_D3 (__global short *src1, int src1_ste ...@@ -354,7 +359,7 @@ __kernel void arithm_s_add_with_mask_C2_D3 (__global short *src1, int src1_ste
int2 src_data2 = (int2)(src2.x, src2.y); int2 src_data2 = (int2)(src2.x, src2.y);
short2 dst_data = *((__global short2 *)((__global char *)dst + dst_index)); short2 dst_data = *((__global short2 *)((__global char *)dst + dst_index));
int2 tmp = convert_int2_sat(src_data1) + src_data2; int2 tmp = ARITHM_OP(convert_int2_sat(src_data1), src_data2);
short2 data = convert_short2_sat(tmp); short2 data = convert_short2_sat(tmp);
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
...@@ -382,7 +387,7 @@ __kernel void arithm_s_add_with_mask_C2_D4 (__global int *src1, int src1_step, ...@@ -382,7 +387,7 @@ __kernel void arithm_s_add_with_mask_C2_D4 (__global int *src1, int src1_step,
int2 src_data2 = (int2)(src2.x, src2.y); int2 src_data2 = (int2)(src2.x, src2.y);
int2 dst_data = *((__global int2 *)((__global char *)dst + dst_index)); int2 dst_data = *((__global int2 *)((__global char *)dst + dst_index));
int2 data = convert_int2_sat(convert_long2_sat(src_data1) + convert_long2_sat(src_data2)); int2 data = convert_int2_sat(ARITHM_OP(convert_long2_sat(src_data1), convert_long2_sat(src_data2)));
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global int2 *)((__global char *)dst + dst_index)) = data; *((__global int2 *)((__global char *)dst + dst_index)) = data;
...@@ -409,7 +414,7 @@ __kernel void arithm_s_add_with_mask_C2_D5 (__global float *src1, int src1_ste ...@@ -409,7 +414,7 @@ __kernel void arithm_s_add_with_mask_C2_D5 (__global float *src1, int src1_ste
float2 src_data2 = (float2)(src2.x, src2.y); float2 src_data2 = (float2)(src2.x, src2.y);
float2 dst_data = *((__global float2 *)((__global char *)dst + dst_index)); float2 dst_data = *((__global float2 *)((__global char *)dst + dst_index));
float2 data = src_data1 + src_data2; float2 data = ARITHM_OP(src_data1, src_data2);
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global float2 *)((__global char *)dst + dst_index)) = data; *((__global float2 *)((__global char *)dst + dst_index)) = data;
...@@ -438,7 +443,7 @@ __kernel void arithm_s_add_with_mask_C2_D6 (__global double *src1, int src1_st ...@@ -438,7 +443,7 @@ __kernel void arithm_s_add_with_mask_C2_D6 (__global double *src1, int src1_st
double2 src_data2 = (double2)(src2.x, src2.y); double2 src_data2 = (double2)(src2.x, src2.y);
double2 dst_data = *((__global double2 *)((__global char *)dst + dst_index)); double2 dst_data = *((__global double2 *)((__global char *)dst + dst_index));
double2 data = src_data1 + src_data2; double2 data = ARITHM_OP(src_data1, src_data2);
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global double2 *)((__global char *)dst + dst_index)) = data; *((__global double2 *)((__global char *)dst + dst_index)) = data;
...@@ -451,7 +456,6 @@ __kernel void arithm_s_add_with_mask_C4_D0 (__global uchar *src1, int src1_ste ...@@ -451,7 +456,6 @@ __kernel void arithm_s_add_with_mask_C4_D0 (__global uchar *src1, int src1_ste
__global uchar *mask, int mask_step, int mask_offset, __global uchar *mask, int mask_step, int mask_offset,
int4 src2, int rows, int cols, int dst_step1) int4 src2, int rows, int cols, int dst_step1)
{ {
int x = get_global_id(0); int x = get_global_id(0);
int y = get_global_id(1); int y = get_global_id(1);
...@@ -466,7 +470,7 @@ __kernel void arithm_s_add_with_mask_C4_D0 (__global uchar *src1, int src1_ste ...@@ -466,7 +470,7 @@ __kernel void arithm_s_add_with_mask_C4_D0 (__global uchar *src1, int src1_ste
uchar4 src_data1 = *((__global uchar4 *)(src1 + src1_index)); uchar4 src_data1 = *((__global uchar4 *)(src1 + src1_index));
uchar4 dst_data = *((__global uchar4 *)(dst + dst_index)); uchar4 dst_data = *((__global uchar4 *)(dst + dst_index));
uchar4 data = convert_uchar4_sat(convert_int4_sat(src_data1) + src2); uchar4 data = convert_uchar4_sat(ARITHM_OP(convert_int4_sat(src_data1), src2));
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global uchar4 *)(dst + dst_index)) = data; *((__global uchar4 *)(dst + dst_index)) = data;
...@@ -492,7 +496,7 @@ __kernel void arithm_s_add_with_mask_C4_D2 (__global ushort *src1, int src1_st ...@@ -492,7 +496,7 @@ __kernel void arithm_s_add_with_mask_C4_D2 (__global ushort *src1, int src1_st
ushort4 src_data1 = *((__global ushort4 *)((__global char *)src1 + src1_index)); ushort4 src_data1 = *((__global ushort4 *)((__global char *)src1 + src1_index));
ushort4 dst_data = *((__global ushort4 *)((__global char *)dst + dst_index)); ushort4 dst_data = *((__global ushort4 *)((__global char *)dst + dst_index));
ushort4 data = convert_ushort4_sat(convert_int4_sat(src_data1) + src2); ushort4 data = convert_ushort4_sat(ARITHM_OP(convert_int4_sat(src_data1), src2));
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global ushort4 *)((__global char *)dst + dst_index)) = data; *((__global ushort4 *)((__global char *)dst + dst_index)) = data;
...@@ -518,7 +522,7 @@ __kernel void arithm_s_add_with_mask_C4_D3 (__global short *src1, int src1_ste ...@@ -518,7 +522,7 @@ __kernel void arithm_s_add_with_mask_C4_D3 (__global short *src1, int src1_ste
short4 src_data1 = *((__global short4 *)((__global char *)src1 + src1_index)); short4 src_data1 = *((__global short4 *)((__global char *)src1 + src1_index));
short4 dst_data = *((__global short4 *)((__global char *)dst + dst_index)); short4 dst_data = *((__global short4 *)((__global char *)dst + dst_index));
short4 data = convert_short4_sat(convert_int4_sat(src_data1) + src2); short4 data = convert_short4_sat(ARITHM_OP(convert_int4_sat(src_data1), src2));
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global short4 *)((__global char *)dst + dst_index)) = data; *((__global short4 *)((__global char *)dst + dst_index)) = data;
...@@ -544,7 +548,7 @@ __kernel void arithm_s_add_with_mask_C4_D4 (__global int *src1, int src1_step, ...@@ -544,7 +548,7 @@ __kernel void arithm_s_add_with_mask_C4_D4 (__global int *src1, int src1_step,
int4 src_data1 = *((__global int4 *)((__global char *)src1 + src1_index)); int4 src_data1 = *((__global int4 *)((__global char *)src1 + src1_index));
int4 dst_data = *((__global int4 *)((__global char *)dst + dst_index)); int4 dst_data = *((__global int4 *)((__global char *)dst + dst_index));
int4 data = convert_int4_sat(convert_long4_sat(src_data1) + convert_long4_sat(src2)); int4 data = convert_int4_sat(ARITHM_OP(convert_long4_sat(src_data1), convert_long4_sat(src2)));
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global int4 *)((__global char *)dst + dst_index)) = data; *((__global int4 *)((__global char *)dst + dst_index)) = data;
...@@ -570,7 +574,7 @@ __kernel void arithm_s_add_with_mask_C4_D5 (__global float *src1, int src1_ste ...@@ -570,7 +574,7 @@ __kernel void arithm_s_add_with_mask_C4_D5 (__global float *src1, int src1_ste
float4 src_data1 = *((__global float4 *)((__global char *)src1 + src1_index)); float4 src_data1 = *((__global float4 *)((__global char *)src1 + src1_index));
float4 dst_data = *((__global float4 *)((__global char *)dst + dst_index)); float4 dst_data = *((__global float4 *)((__global char *)dst + dst_index));
float4 data = src_data1 + src2; float4 data = ARITHM_OP(src_data1, src2);
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global float4 *)((__global char *)dst + dst_index)) = data; *((__global float4 *)((__global char *)dst + dst_index)) = data;
...@@ -598,7 +602,7 @@ __kernel void arithm_s_add_with_mask_C4_D6 (__global double *src1, int src1_st ...@@ -598,7 +602,7 @@ __kernel void arithm_s_add_with_mask_C4_D6 (__global double *src1, int src1_st
double4 src_data1 = *((__global double4 *)((__global char *)src1 + src1_index)); double4 src_data1 = *((__global double4 *)((__global char *)src1 + src1_index));
double4 dst_data = *((__global double4 *)((__global char *)dst + dst_index)); double4 dst_data = *((__global double4 *)((__global char *)dst + dst_index));
double4 data = src_data1 + src2; double4 data = ARITHM_OP(src_data1, src2);
data = mask_data ? data : dst_data; data = mask_data ? data : dst_data;
*((__global double4 *)((__global char *)dst + dst_index)) = data; *((__global double4 *)((__global char *)dst + dst_index)) = data;
......
modules/ocl/src/opencl/arithm_sub.cl deleted 100644 → 0
View file @ bee970ab
This diff is collapsed.
modules/ocl/src/opencl/arithm_sub_scalar.cl deleted 100644 → 0
View file @ bee970ab
This diff is collapsed.
modules/ocl/src/opencl/arithm_sub_scalar_mask.cl deleted 100644 → 0
View file @ bee970ab
This diff is collapsed.
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