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

Fix arithm's substract mismatch caused by incomplete merge

Browse Source
This commit is contained in:
yao 2013-04-13 15:57:12 +08:00
parent 3b364330ad
commit 2f781a53a9
3 changed files with 90 additions and 634 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

53
modules/ocl/src/arithm.cpp
View File

@ -126,7 +126,8 @@ 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)
{ {
@ -182,14 +183,25 @@ void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, String
scalar = (T)scalar1; scalar = (T)scalar1;
args.push_back( std::make_pair( sizeof(T), (void *)&scalar )); args.push_back( std::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)
{ {
@ -244,24 +256,34 @@ static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
args.push_back( std::make_pair( sizeof(cl_int), (void *)&cols )); args.push_back( std::make_pair( sizeof(cl_int), (void *)&cols ));
args.push_back( std::make_pair( sizeof(cl_int), (void *)&dst_step1 )); args.push_back( std::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_add", &arithm_add); 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_add_with_mask", &arithm_add); 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);
@ -347,12 +369,9 @@ void arithmetic_scalar_run(const oclMat &src1, const Scalar &src2, oclMat &dst,
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&cols )); args.push_back( std::make_pair( sizeof(cl_int) , (void *)&cols ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&dst_step1 )); args.push_back( std::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( std::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)

316
modules/ocl/src/opencl/arithm_add_scalar.cl
View File

@ -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,
@ -62,7 +67,7 @@ __kernel void arithm_s_add_C1_D0 (__global uchar *src1, int src1_step, int src
if (x < cols && y < rows) if (x < cols && y < rows)
{ {
x = x << 2; x = x << 2;
#ifdef dst_align #ifdef dst_align
#undef dst_align #undef dst_align
#endif #endif
@ -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;
@ -105,7 +110,7 @@ __kernel void arithm_s_add_C1_D2 (__global ushort *src1, int src1_step, int sr
if (x < cols && y < rows) if (x < cols && y < rows)
{ {
x = x << 1; x = x << 1;
#ifdef dst_align #ifdef dst_align
#undef dst_align #undef dst_align
#endif #endif
@ -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;
@ -140,7 +145,7 @@ __kernel void arithm_s_add_C1_D3 (__global short *src1, int src1_step, int src
if (x < cols && y < rows) if (x < cols && y < rows)
{ {
x = x << 1; x = x << 1;
#ifdef dst_align #ifdef dst_align
#undef dst_align #undef dst_align
#endif #endif
@ -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_
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
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
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;
} }
@ -245,7 +250,7 @@ __kernel void arithm_s_add_C2_D0 (__global uchar *src1, int src1_step, int src
if (x < cols && y < rows) if (x < cols && y < rows)
{ {
x = x << 1; x = x << 1;
#ifdef dst_align #ifdef dst_align
#undef dst_align #undef dst_align
#endif #endif
@ -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
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
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_
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
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;
} }
} }
@ -376,280 +381,13 @@ __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;
} }
} }
#endif #endif
__kernel void arithm_s_add_C3_D0 (__global uchar *src1, int src1_step, int src1_offset,
__global uchar *dst, int dst_step, int dst_offset,
int4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 2;
#ifdef dst_align
#undef dst_align
#endif
#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
int dst_start = mad24(y, dst_step, dst_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
uchar4 src1_data_0 = vload4(0, src1 + src1_index + 0);
uchar4 src1_data_1 = vload4(0, src1 + src1_index + 4);
uchar4 src1_data_2 = vload4(0, src1 + src1_index + 8);
int4 src2_data_0 = (int4)(src2.x, src2.y, src2.z, src2.x);
int4 src2_data_1 = (int4)(src2.y, src2.z, src2.x, src2.y);
int4 src2_data_2 = (int4)(src2.z, src2.x, src2.y, src2.z);
uchar4 data_0 = *((__global uchar4 *)(dst + dst_index + 0));
uchar4 data_1 = *((__global uchar4 *)(dst + dst_index + 4));
uchar4 data_2 = *((__global uchar4 *)(dst + dst_index + 8));
uchar4 tmp_data_0 = convert_uchar4_sat(convert_int4_sat(src1_data_0) + src2_data_0);
uchar4 tmp_data_1 = convert_uchar4_sat(convert_int4_sat(src1_data_1) + src2_data_1);
uchar4 tmp_data_2 = convert_uchar4_sat(convert_int4_sat(src1_data_2) + src2_data_2);
data_0.xyz = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
data_0.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
? tmp_data_0.w : data_0.w;
data_1.xy = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
? tmp_data_1.xy : data_1.xy;
data_1.zw = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_1.zw : data_1.zw;
data_2.x = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_2.x : data_2.x;
data_2.yzw = ((dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
? tmp_data_2.yzw : data_2.yzw;
*((__global uchar4 *)(dst + dst_index + 0)) = data_0;
*((__global uchar4 *)(dst + dst_index + 4)) = data_1;
*((__global uchar4 *)(dst + dst_index + 8)) = data_2;
}
}
__kernel void arithm_s_add_C3_D2 (__global ushort *src1, int src1_step, int src1_offset,
__global ushort *dst, int dst_step, int dst_offset,
int4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 1;
#ifdef dst_align
#undef dst_align
#endif
#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
int dst_start = mad24(y, dst_step, dst_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
ushort2 src1_data_0 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 0));
ushort2 src1_data_1 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 4));
ushort2 src1_data_2 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 8));
int2 src2_data_0 = (int2)(src2.x, src2.y);
int2 src2_data_1 = (int2)(src2.z, src2.x);
int2 src2_data_2 = (int2)(src2.y, src2.z);
ushort2 data_0 = *((__global ushort2 *)((__global char *)dst + dst_index + 0));
ushort2 data_1 = *((__global ushort2 *)((__global char *)dst + dst_index + 4));
ushort2 data_2 = *((__global ushort2 *)((__global char *)dst + dst_index + 8));
ushort2 tmp_data_0 = convert_ushort2_sat(convert_int2_sat(src1_data_0) + src2_data_0);
ushort2 tmp_data_1 = convert_ushort2_sat(convert_int2_sat(src1_data_1) + src2_data_1);
ushort2 tmp_data_2 = convert_ushort2_sat(convert_int2_sat(src1_data_2) + src2_data_2);
data_0.xy = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
data_1.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
? tmp_data_1.x : data_1.x;
data_1.y = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_1.y : data_1.y;
data_2.xy = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_2.xy : data_2.xy;
*((__global ushort2 *)((__global char *)dst + dst_index + 0))= data_0;
*((__global ushort2 *)((__global char *)dst + dst_index + 4))= data_1;
*((__global ushort2 *)((__global char *)dst + dst_index + 8))= data_2;
}
}
__kernel void arithm_s_add_C3_D3 (__global short *src1, int src1_step, int src1_offset,
__global short *dst, int dst_step, int dst_offset,
int4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 1;
#ifdef dst_align
#undef dst_align
#endif
#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
int dst_start = mad24(y, dst_step, dst_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
short2 src1_data_0 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 0));
short2 src1_data_1 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 4));
short2 src1_data_2 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 8));
int2 src2_data_0 = (int2)(src2.x, src2.y);
int2 src2_data_1 = (int2)(src2.z, src2.x);
int2 src2_data_2 = (int2)(src2.y, src2.z);
short2 data_0 = *((__global short2 *)((__global char *)dst + dst_index + 0));
short2 data_1 = *((__global short2 *)((__global char *)dst + dst_index + 4));
short2 data_2 = *((__global short2 *)((__global char *)dst + dst_index + 8));
short2 tmp_data_0 = convert_short2_sat(convert_int2_sat(src1_data_0) + src2_data_0);
short2 tmp_data_1 = convert_short2_sat(convert_int2_sat(src1_data_1) + src2_data_1);
short2 tmp_data_2 = convert_short2_sat(convert_int2_sat(src1_data_2) + src2_data_2);
data_0.xy = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
data_1.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
? tmp_data_1.x : data_1.x;
data_1.y = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_1.y : data_1.y;
data_2.xy = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_2.xy : data_2.xy;
*((__global short2 *)((__global char *)dst + dst_index + 0))= data_0;
*((__global short2 *)((__global char *)dst + dst_index + 4))= data_1;
*((__global short2 *)((__global char *)dst + dst_index + 8))= data_2;
}
}
__kernel void arithm_s_add_C3_D4 (__global int *src1, int src1_step, int src1_offset,
__global int *dst, int dst_step, int dst_offset,
int4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
int dst_index = mad24(y, dst_step, dst_offset + (x * 12));
int src1_data_0 = *((__global int *)((__global char *)src1 + src1_index + 0));
int src1_data_1 = *((__global int *)((__global char *)src1 + src1_index + 4));
int src1_data_2 = *((__global int *)((__global char *)src1 + src1_index + 8));
int src2_data_0 = src2.x;
int src2_data_1 = src2.y;
int src2_data_2 = src2.z;
int data_0 = *((__global int *)((__global char *)dst + dst_index + 0));
int data_1 = *((__global int *)((__global char *)dst + dst_index + 4));
int data_2 = *((__global int *)((__global char *)dst + dst_index + 8));
int tmp_data_0 = convert_int_sat((long)src1_data_0 + (long)src2_data_0);
int tmp_data_1 = convert_int_sat((long)src1_data_1 + (long)src2_data_1);
int tmp_data_2 = convert_int_sat((long)src1_data_2 + (long)src2_data_2);
*((__global int *)((__global char *)dst + dst_index + 0))= tmp_data_0;
*((__global int *)((__global char *)dst + dst_index + 4))= tmp_data_1;
*((__global int *)((__global char *)dst + dst_index + 8))= tmp_data_2;
}
}
__kernel void arithm_s_add_C3_D5 (__global float *src1, int src1_step, int src1_offset,
__global float *dst, int dst_step, int dst_offset,
float4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
int dst_index = mad24(y, dst_step, dst_offset + (x * 12));
float src1_data_0 = *((__global float *)((__global char *)src1 + src1_index + 0));
float src1_data_1 = *((__global float *)((__global char *)src1 + src1_index + 4));
float src1_data_2 = *((__global float *)((__global char *)src1 + src1_index + 8));
float src2_data_0 = src2.x;
float src2_data_1 = src2.y;
float src2_data_2 = src2.z;
float data_0 = *((__global float *)((__global char *)dst + dst_index + 0));
float data_1 = *((__global float *)((__global char *)dst + dst_index + 4));
float data_2 = *((__global float *)((__global char *)dst + dst_index + 8));
float tmp_data_0 = src1_data_0 + src2_data_0;
float tmp_data_1 = src1_data_1 + src2_data_1;
float tmp_data_2 = src1_data_2 + src2_data_2;
*((__global float *)((__global char *)dst + dst_index + 0))= tmp_data_0;
*((__global float *)((__global char *)dst + dst_index + 4))= tmp_data_1;
*((__global float *)((__global char *)dst + dst_index + 8))= tmp_data_2;
}
}
#if defined (DOUBLE_SUPPORT)
__kernel void arithm_s_add_C3_D6 (__global double *src1, int src1_step, int src1_offset,
__global double *dst, int dst_step, int dst_offset,
double4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, src1_step, (x * 24) + src1_offset);
int dst_index = mad24(y, dst_step, dst_offset + (x * 24));
double src1_data_0 = *((__global double *)((__global char *)src1 + src1_index + 0 ));
double src1_data_1 = *((__global double *)((__global char *)src1 + src1_index + 8 ));
double src1_data_2 = *((__global double *)((__global char *)src1 + src1_index + 16));
double src2_data_0 = src2.x;
double src2_data_1 = src2.y;
double src2_data_2 = src2.z;
double data_0 = *((__global double *)((__global char *)dst + dst_index + 0 ));
double data_1 = *((__global double *)((__global char *)dst + dst_index + 8 ));
double data_2 = *((__global double *)((__global char *)dst + dst_index + 16));
double tmp_data_0 = src1_data_0 + src2_data_0;
double tmp_data_1 = src1_data_1 + src2_data_1;
double tmp_data_2 = src1_data_2 + src2_data_2;
*((__global double *)((__global char *)dst + dst_index + 0 ))= tmp_data_0;
*((__global double *)((__global char *)dst + dst_index + 8 ))= tmp_data_1;
*((__global double *)((__global char *)dst + dst_index + 16))= tmp_data_2;
}
}
#endif
__kernel void arithm_s_add_C4_D0 (__global uchar *src1, int src1_step, int src1_offset, __kernel void arithm_s_add_C4_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,
int4 src2, int rows, int cols, int dst_step1) int4 src2, int rows, int cols, int dst_step1)
@ -665,7 +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;
} }
@ -685,7 +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;
} }
@ -705,7 +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;
} }
@ -725,7 +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;
} }
@ -745,7 +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;
} }
@ -767,7 +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;
} }

355
modules/ocl/src/opencl/arithm_add_scalar_mask.cl
View File

@ -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,
@ -64,7 +69,7 @@ __kernel void arithm_s_add_with_mask_C1_D0 (__global uchar *src1, int src1_ste
if (x < cols && y < rows) if (x < cols && y < rows)
{ {
x = x << 2; x = x << 2;
#ifdef dst_align #ifdef dst_align
#undef dst_align #undef dst_align
#endif #endif
@ -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;
@ -117,7 +122,7 @@ __kernel void arithm_s_add_with_mask_C1_D2 (__global ushort *src1, int src1_st
if (x < cols && y < rows) if (x < cols && y < rows)
{ {
x = x << 1; x = x << 1;
#ifdef dst_align #ifdef dst_align
#undef dst_align #undef dst_align
#endif #endif
@ -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;
@ -155,7 +160,7 @@ __kernel void arithm_s_add_with_mask_C1_D3 (__global short *src1, int src1_ste
if (x < cols && y < rows) if (x < cols && y < rows)
{ {
x = x << 1; x = x << 1;
#ifdef dst_align #ifdef dst_align
#undef dst_align #undef dst_align
#endif #endif
@ -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
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
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_
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;
@ -279,7 +284,7 @@ __kernel void arithm_s_add_with_mask_C2_D0 (__global uchar *src1, int src1_ste
if (x < cols && y < rows) if (x < cols && y < rows)
{ {
x = x << 1; x = x << 1;
#ifdef dst_align #ifdef dst_align
#undef dst_align #undef dst_align
#endif #endif
@ -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
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
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,
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
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
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;
@ -446,317 +451,11 @@ __kernel void arithm_s_add_with_mask_C2_D6 (__global double *src1, int src1_st
} }
#endif #endif
__kernel void arithm_s_add_with_mask_C3_D0 (__global uchar *src1, int src1_step, int src1_offset,
__global uchar *dst, int dst_step, int dst_offset,
__global uchar *mask, int mask_step, int mask_offset,
int4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 2;
#ifdef dst_align
#undef dst_align
#endif
#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
int dst_start = mad24(y, dst_step, dst_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
uchar4 src1_data_0 = vload4(0, src1 + src1_index + 0);
uchar4 src1_data_1 = vload4(0, src1 + src1_index + 4);
uchar4 src1_data_2 = vload4(0, src1 + src1_index + 8);
int4 src2_data_0 = (int4)(src2.x, src2.y, src2.z, src2.x);
int4 src2_data_1 = (int4)(src2.y, src2.z, src2.x, src2.y);
int4 src2_data_2 = (int4)(src2.z, src2.x, src2.y, src2.z);
uchar4 mask_data = vload4(0, mask + mask_index);
uchar4 data_0 = *((__global uchar4 *)(dst + dst_index + 0));
uchar4 data_1 = *((__global uchar4 *)(dst + dst_index + 4));
uchar4 data_2 = *((__global uchar4 *)(dst + dst_index + 8));
uchar4 tmp_data_0 = convert_uchar4_sat(convert_int4_sat(src1_data_0) + src2_data_0);
uchar4 tmp_data_1 = convert_uchar4_sat(convert_int4_sat(src1_data_1) + src2_data_1);
uchar4 tmp_data_2 = convert_uchar4_sat(convert_int4_sat(src1_data_2) + src2_data_2);
data_0.xyz = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
data_0.w = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
? tmp_data_0.w : data_0.w;
data_1.xy = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
? tmp_data_1.xy : data_1.xy;
data_1.zw = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_1.zw : data_1.zw;
data_2.x = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_2.x : data_2.x;
data_2.yzw = ((mask_data.w) && (dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
? tmp_data_2.yzw : data_2.yzw;
*((__global uchar4 *)(dst + dst_index + 0)) = data_0;
*((__global uchar4 *)(dst + dst_index + 4)) = data_1;
*((__global uchar4 *)(dst + dst_index + 8)) = data_2;
}
}
__kernel void arithm_s_add_with_mask_C3_D2 (__global ushort *src1, int src1_step, int src1_offset,
__global ushort *dst, int dst_step, int dst_offset,
__global uchar *mask, int mask_step, int mask_offset,
int4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 1;
#ifdef dst_align
#undef dst_align
#endif
#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
int dst_start = mad24(y, dst_step, dst_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
ushort2 src1_data_0 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 0));
ushort2 src1_data_1 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 4));
ushort2 src1_data_2 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 8));
int2 src2_data_0 = (int2)(src2.x, src2.y);
int2 src2_data_1 = (int2)(src2.z, src2.x);
int2 src2_data_2 = (int2)(src2.y, src2.z);
uchar2 mask_data = vload2(0, mask + mask_index);
ushort2 data_0 = *((__global ushort2 *)((__global char *)dst + dst_index + 0));
ushort2 data_1 = *((__global ushort2 *)((__global char *)dst + dst_index + 4));
ushort2 data_2 = *((__global ushort2 *)((__global char *)dst + dst_index + 8));
ushort2 tmp_data_0 = convert_ushort2_sat(convert_int2_sat(src1_data_0) + src2_data_0);
ushort2 tmp_data_1 = convert_ushort2_sat(convert_int2_sat(src1_data_1) + src2_data_1);
ushort2 tmp_data_2 = convert_ushort2_sat(convert_int2_sat(src1_data_2) + src2_data_2);
data_0.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
data_1.x = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
? tmp_data_1.x : data_1.x;
data_1.y = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_1.y : data_1.y;
data_2.xy = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_2.xy : data_2.xy;
*((__global ushort2 *)((__global char *)dst + dst_index + 0))= data_0;
*((__global ushort2 *)((__global char *)dst + dst_index + 4))= data_1;
*((__global ushort2 *)((__global char *)dst + dst_index + 8))= data_2;
}
}
__kernel void arithm_s_add_with_mask_C3_D3 (__global short *src1, int src1_step, int src1_offset,
__global short *dst, int dst_step, int dst_offset,
__global uchar *mask, int mask_step, int mask_offset,
int4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 1;
#ifdef dst_align
#undef dst_align
#endif
#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
int dst_start = mad24(y, dst_step, dst_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
short2 src1_data_0 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 0));
short2 src1_data_1 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 4));
short2 src1_data_2 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 8));
int2 src2_data_0 = (int2)(src2.x, src2.y);
int2 src2_data_1 = (int2)(src2.z, src2.x);
int2 src2_data_2 = (int2)(src2.y, src2.z);
uchar2 mask_data = vload2(0, mask + mask_index);
short2 data_0 = *((__global short2 *)((__global char *)dst + dst_index + 0));
short2 data_1 = *((__global short2 *)((__global char *)dst + dst_index + 4));
short2 data_2 = *((__global short2 *)((__global char *)dst + dst_index + 8));
short2 tmp_data_0 = convert_short2_sat(convert_int2_sat(src1_data_0) + src2_data_0);
short2 tmp_data_1 = convert_short2_sat(convert_int2_sat(src1_data_1) + src2_data_1);
short2 tmp_data_2 = convert_short2_sat(convert_int2_sat(src1_data_2) + src2_data_2);
data_0.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
data_1.x = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
? tmp_data_1.x : data_1.x;
data_1.y = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_1.y : data_1.y;
data_2.xy = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
? tmp_data_2.xy : data_2.xy;
*((__global short2 *)((__global char *)dst + dst_index + 0))= data_0;
*((__global short2 *)((__global char *)dst + dst_index + 4))= data_1;
*((__global short2 *)((__global char *)dst + dst_index + 8))= data_2;
}
}
__kernel void arithm_s_add_with_mask_C3_D4 (__global int *src1, int src1_step, int src1_offset,
__global int *dst, int dst_step, int dst_offset,
__global uchar *mask, int mask_step, int mask_offset,
int4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
int mask_index = mad24(y, mask_step, x + mask_offset);
int dst_index = mad24(y, dst_step, dst_offset + (x * 12));
int src1_data_0 = *((__global int *)((__global char *)src1 + src1_index + 0));
int src1_data_1 = *((__global int *)((__global char *)src1 + src1_index + 4));
int src1_data_2 = *((__global int *)((__global char *)src1 + src1_index + 8));
int src2_data_0 = src2.x;
int src2_data_1 = src2.y;
int src2_data_2 = src2.z;
uchar mask_data = * (mask + mask_index);
int data_0 = *((__global int *)((__global char *)dst + dst_index + 0));
int data_1 = *((__global int *)((__global char *)dst + dst_index + 4));
int data_2 = *((__global int *)((__global char *)dst + dst_index + 8));
int tmp_data_0 = convert_int_sat((long)src1_data_0 + (long)src2_data_0);
int tmp_data_1 = convert_int_sat((long)src1_data_1 + (long)src2_data_1);
int tmp_data_2 = convert_int_sat((long)src1_data_2 + (long)src2_data_2);
data_0 = mask_data ? tmp_data_0 : data_0;
data_1 = mask_data ? tmp_data_1 : data_1;
data_2 = mask_data ? tmp_data_2 : data_2;
*((__global int *)((__global char *)dst + dst_index + 0))= data_0;
*((__global int *)((__global char *)dst + dst_index + 4))= data_1;
*((__global int *)((__global char *)dst + dst_index + 8))= data_2;
}
}
__kernel void arithm_s_add_with_mask_C3_D5 (__global float *src1, int src1_step, int src1_offset,
__global float *dst, int dst_step, int dst_offset,
__global uchar *mask, int mask_step, int mask_offset,
float4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
int mask_index = mad24(y, mask_step, x + mask_offset);
int dst_index = mad24(y, dst_step, dst_offset + (x * 12));
float src1_data_0 = *((__global float *)((__global char *)src1 + src1_index + 0));
float src1_data_1 = *((__global float *)((__global char *)src1 + src1_index + 4));
float src1_data_2 = *((__global float *)((__global char *)src1 + src1_index + 8));
float src2_data_0 = src2.x;
float src2_data_1 = src2.y;
float src2_data_2 = src2.z;
uchar mask_data = * (mask + mask_index);
float data_0 = *((__global float *)((__global char *)dst + dst_index + 0));
float data_1 = *((__global float *)((__global char *)dst + dst_index + 4));
float data_2 = *((__global float *)((__global char *)dst + dst_index + 8));
float tmp_data_0 = src1_data_0 + src2_data_0;
float tmp_data_1 = src1_data_1 + src2_data_1;
float tmp_data_2 = src1_data_2 + src2_data_2;
data_0 = mask_data ? tmp_data_0 : data_0;
data_1 = mask_data ? tmp_data_1 : data_1;
data_2 = mask_data ? tmp_data_2 : data_2;
*((__global float *)((__global char *)dst + dst_index + 0))= data_0;
*((__global float *)((__global char *)dst + dst_index + 4))= data_1;
*((__global float *)((__global char *)dst + dst_index + 8))= data_2;
}
}
#if defined (DOUBLE_SUPPORT)
__kernel void arithm_s_add_with_mask_C3_D6 (__global double *src1, int src1_step, int src1_offset,
__global double *dst, int dst_step, int dst_offset,
__global uchar *mask, int mask_step, int mask_offset,
double4 src2, int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, src1_step, (x * 24) + src1_offset);
int mask_index = mad24(y, mask_step, x + mask_offset);
int dst_index = mad24(y, dst_step, dst_offset + (x * 24));
double src1_data_0 = *((__global double *)((__global char *)src1 + src1_index + 0 ));
double src1_data_1 = *((__global double *)((__global char *)src1 + src1_index + 8 ));
double src1_data_2 = *((__global double *)((__global char *)src1 + src1_index + 16));
double src2_data_0 = src2.x;
double src2_data_1 = src2.y;
double src2_data_2 = src2.z;
uchar mask_data = * (mask + mask_index);
double data_0 = *((__global double *)((__global char *)dst + dst_index + 0 ));
double data_1 = *((__global double *)((__global char *)dst + dst_index + 8 ));
double data_2 = *((__global double *)((__global char *)dst + dst_index + 16));
double tmp_data_0 = src1_data_0 + src2_data_0;
double tmp_data_1 = src1_data_1 + src2_data_1;
double tmp_data_2 = src1_data_2 + src2_data_2;
data_0 = mask_data ? tmp_data_0 : data_0;
data_1 = mask_data ? tmp_data_1 : data_1;
data_2 = mask_data ? tmp_data_2 : data_2;
*((__global double *)((__global char *)dst + dst_index + 0 ))= data_0;
*((__global double *)((__global char *)dst + dst_index + 8 ))= data_1;
*((__global double *)((__global char *)dst + dst_index + 16))= data_2;
}
}
#endif
__kernel void arithm_s_add_with_mask_C4_D0 (__global uchar *src1, int src1_step, int src1_offset, __kernel void arithm_s_add_with_mask_C4_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,
__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);
@ -771,7 +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;
@ -797,7 +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;
@ -823,7 +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;
@ -849,7 +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;
@ -875,7 +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;
@ -903,7 +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;