Skip to content

O
opencv
Commit fd90efc9 authored by Andrey Pavlenko's avatar Andrey Pavlenko Committed by OpenCV Buildbot
Browse files
Options

Merge pull request #2593 from arkunze:pullreq/140319-color

parents 6bfbcf02 785acc18
Show whitespace changes
Inline Side-by-side
Showing with 418 additions and 158 deletions
modules/imgproc/src/color.cpp
View file @ fd90efc9
......@@ -2716,6 +2716,8 @@ struct mRGBA2RGBA
#ifdef HAVE_OPENCL
#define DIVUP(total, grain) (((total) + (grain) - 1) / (grain))
static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
{
bool ok = false;
......@@ -2729,6 +2731,17 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
if (depth != CV_8U && depth != CV_16U && depth != CV_32F)
return false;
cv::String opts = format("-D depth=%d -D scn=%d ", depth, scn);
ocl::Device dev = ocl::Device::getDefault();
int pxPerWIy = 1;
if (dev.isIntel() && (dev.type() & ocl::Device::TYPE_GPU))
{
pxPerWIy = 4;
}
globalsize[1] = DIVUP(globalsize[1], pxPerWIy);
opts += format("-D PIX_PER_WI_Y=%d ", pxPerWIy);
switch (code)
{
case COLOR_BGR2BGRA: case COLOR_RGB2BGRA: case COLOR_BGRA2BGR:
......@@ -2738,7 +2751,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
dcn = code == COLOR_BGR2BGRA || code == COLOR_RGB2BGRA || code == COLOR_BGRA2RGBA ? 4 : 3;
bool reverse = !(code == COLOR_BGR2BGRA || code == COLOR_BGRA2BGR);
k.create("RGB", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=%d -D dcn=%d -D bidx=0 -D %s", depth, scn, dcn,
opts + format("-D dcn=%d -D bidx=0 -D %s", dcn,
reverse ? "REVERSE" : "ORDER"));
break;
}
......@@ -2752,7 +2765,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
int greenbits = code == COLOR_BGR5652BGR || code == COLOR_BGR5652RGB ||
code == COLOR_BGR5652BGRA || code == COLOR_BGR5652RGBA ? 6 : 5;
k.create("RGB5x52RGB", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=2 -D dcn=%d -D bidx=%d -D greenbits=%d", depth, dcn, bidx, greenbits));
opts + format("-D dcn=%d -D bidx=%d -D greenbits=%d", dcn, bidx, greenbits));
break;
}
case COLOR_BGR2BGR565: case COLOR_BGR2BGR555: case COLOR_RGB2BGR565: case COLOR_RGB2BGR555:
......@@ -2765,7 +2778,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
code == COLOR_BGRA2BGR565 || code == COLOR_RGBA2BGR565 ? 6 : 5;
dcn = 2;
k.create("RGB2RGB5x5", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=%d -D dcn=2 -D bidx=%d -D greenbits=%d", depth, scn, bidx, greenbits));
opts + format("-D dcn=2 -D bidx=%d -D greenbits=%d", bidx, greenbits));
break;
}
case COLOR_BGR5652GRAY: case COLOR_BGR5552GRAY:
......@@ -2774,7 +2787,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
dcn = 1;
int greenbits = code == COLOR_BGR5652GRAY ? 6 : 5;
k.create("BGR5x52Gray", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=2 -D dcn=1 -D bidx=0 -D greenbits=%d", depth, greenbits));
opts + format("-D dcn=1 -D bidx=0 -D greenbits=%d", greenbits));
break;
}
case COLOR_GRAY2BGR565: case COLOR_GRAY2BGR555:
......@@ -2783,7 +2796,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
dcn = 2;
int greenbits = code == COLOR_GRAY2BGR565 ? 6 : 5;
k.create("Gray2BGR5x5", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=1 -D dcn=2 -D bidx=0 -D greenbits=%d", depth, greenbits));
opts + format("-D dcn=2 -D bidx=0 -D greenbits=%d", greenbits));
break;
}
case COLOR_BGR2GRAY: case COLOR_BGRA2GRAY:
......@@ -2793,8 +2806,8 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
bidx = code == COLOR_BGR2GRAY || code == COLOR_BGRA2GRAY ? 0 : 2;
dcn = 1;
k.create("RGB2Gray", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=%d -D dcn=1 -D bidx=%d -D STRIPE_SIZE=%d",
depth, scn, bidx, stripeSize));
opts + format("-D dcn=1 -D bidx=%d -D STRIPE_SIZE=%d",
bidx, stripeSize));
globalsize[0] = (src.cols + stripeSize-1)/stripeSize;
break;
}
......@@ -2804,7 +2817,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
CV_Assert(scn == 1);
dcn = code == COLOR_GRAY2BGRA ? 4 : 3;
k.create("Gray2RGB", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D bidx=0 -D scn=1 -D dcn=%d", depth, dcn));
opts + format("-D bidx=0 -D dcn=%d", dcn));
break;
}
case COLOR_BGR2YUV:
......@@ -2814,7 +2827,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
bidx = code == COLOR_RGB2YUV ? 0 : 2;
dcn = 3;
k.create("RGB2YUV", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=%d -D dcn=3 -D bidx=%d", depth, scn, bidx));
opts + format("-D dcn=3 -D bidx=%d", bidx));
break;
}
case COLOR_YUV2BGR:
......@@ -2824,7 +2837,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
CV_Assert(dcn == 3 || dcn == 4);
bidx = code == COLOR_YUV2RGB ? 0 : 2;
k.create("YUV2RGB", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=3 -D dcn=%d -D bidx=%d", depth, dcn, bidx));
opts + format("-D dcn=%d -D bidx=%d", dcn, bidx));
break;
}
case COLOR_YUV2RGB_NV12: case COLOR_YUV2BGR_NV12:
......@@ -2837,7 +2850,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
dstSz = Size(sz.width, sz.height * 2 / 3);
k.create("YUV2RGB_NV12", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=0 -D scn=1 -D dcn=%d -D bidx=%d", dcn, bidx));
opts + format("-D dcn=%d -D bidx=%d", dcn, bidx));
break;
}
case COLOR_BGR2YCrCb:
......@@ -2847,7 +2860,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
bidx = code == COLOR_BGR2YCrCb ? 0 : 2;
dcn = 3;
k.create("RGB2YCrCb", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=%d -D dcn=3 -D bidx=%d", depth, scn, bidx));
opts + format("-D dcn=3 -D bidx=%d", bidx));
break;
}
case COLOR_YCrCb2BGR:
......@@ -2858,7 +2871,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
CV_Assert(scn == 3 && (dcn == 3 || dcn == 4));
bidx = code == COLOR_YCrCb2BGR ? 0 : 2;
k.create("YCrCb2RGB", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=%d -D dcn=%d -D bidx=%d", depth, scn, dcn, bidx));
opts + format("-D dcn=%d -D bidx=%d", dcn, bidx));
break;
}
case COLOR_BGR2XYZ: case COLOR_RGB2XYZ:
......@@ -2904,7 +2917,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
dst = _dst.getUMat();
k.create("RGB2XYZ", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=%d -D dcn=3 -D bidx=%d", depth, scn, bidx));
opts + format("-D dcn=3 -D bidx=%d", bidx));
if (k.empty())
return false;
k.args(ocl::KernelArg::ReadOnlyNoSize(src), ocl::KernelArg::WriteOnly(dst), ocl::KernelArg::PtrReadOnly(c));
......@@ -2955,7 +2968,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
dst = _dst.getUMat();
k.create("XYZ2RGB", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D scn=3 -D dcn=%d -D bidx=%d", depth, dcn, bidx));
opts + format("-D dcn=%d -D bidx=%d", dcn, bidx));
if (k.empty())
return false;
k.args(ocl::KernelArg::ReadOnlyNoSize(src), ocl::KernelArg::WriteOnly(dst), ocl::KernelArg::PtrReadOnly(c));
......@@ -3010,8 +3023,9 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
_dst.create(dstSz, CV_8UC3);
dst = _dst.getUMat();
k.create("RGB2HSV", ocl::imgproc::cvtcolor_oclsrc, format("-D depth=%d -D hrange=%d -D bidx=%d -D dcn=3 -D scn=%d",
depth, hrange, bidx, scn));
k.create("RGB2HSV", ocl::imgproc::cvtcolor_oclsrc,
opts + format("-D hrange=%d -D bidx=%d -D dcn=3",
hrange, bidx));
if (k.empty())
return false;
......@@ -3023,7 +3037,8 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
}
else
k.create(kernelName.c_str(), ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D hscale=%ff -D bidx=%d -D scn=%d -D dcn=3", depth, hrange*(1.f/360.f), bidx, scn));
opts + format("-D hscale=%ff -D bidx=%d -D dcn=3",
hrange*(1.f/360.f), bidx));
break;
}
case COLOR_HSV2BGR: case COLOR_HSV2RGB: case COLOR_HSV2BGR_FULL: case COLOR_HSV2RGB_FULL:
......@@ -3041,8 +3056,8 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
String kernelName = String(is_hsv ? "HSV" : "HLS") + "2RGB";
k.create(kernelName.c_str(), ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D dcn=%d -D scn=3 -D bidx=%d -D hrange=%d -D hscale=%ff",
depth, dcn, bidx, hrange, 6.f/hrange));
opts + format("-D dcn=%d -D bidx=%d -D hrange=%d -D hscale=%ff",
dcn, bidx, hrange, 6.f/hrange));
break;
}
case COLOR_RGBA2mRGBA: case COLOR_mRGBA2RGBA:
......@@ -3051,7 +3066,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
dcn = 4;
k.create(code == COLOR_RGBA2mRGBA ? "RGBA2mRGBA" : "mRGBA2RGBA", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D dcn=4 -D scn=4 -D bidx=3", depth));
opts + "-D dcn=4 -D bidx=3");
break;
}
case CV_BGR2Lab: case CV_RGB2Lab: case CV_LBGR2Lab: case CV_LRGB2Lab:
......@@ -3063,8 +3078,8 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
dcn = 3;
k.create("BGR2Lab", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D dcn=3 -D scn=%d -D bidx=%d%s",
depth, scn, bidx, srgb ? " -D SRGB" : ""));
opts + format("-D dcn=3 -D bidx=%d%s",
bidx, srgb ? " -D SRGB" : ""));
if (k.empty())
return false;
......@@ -3165,8 +3180,8 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
bool srgb = code == CV_Lab2BGR || code == CV_Lab2RGB;
k.create("Lab2BGR", ocl::imgproc::cvtcolor_oclsrc,
format("-D depth=%d -D dcn=%d -D scn=3 -D bidx=%d%s",
depth, dcn, bidx, srgb ? " -D SRGB" : ""));
opts + format("-D dcn=%d -D bidx=%d%s",
dcn, bidx, srgb ? " -D SRGB" : ""));
if (k.empty())
return false;
......
modules/imgproc/src/opencl/cvtcolor.cl
View file @ fd90efc9
......@@ -99,56 +99,70 @@ enum
#define hrange 0
#endif
#if bidx == 0
#define R_COMP z
#define G_COMP y
#define B_COMP x
#elif bidx == 2
#define R_COMP x
#define G_COMP y
#define B_COMP z
#elif bidx == 3
// The only kernel that uses bidx == 3 doesn't use these macros.
// But we still need to make the compiler happy.
#define R_COMP w
#define G_COMP w
#define B_COMP w
#endif
#define __CAT(x, y) x##y
#define CAT(x, y) __CAT(x, y)
#define DATA_TYPE_4 CAT(DATA_TYPE, 4)
///////////////////////////////////// RGB <-> GRAY //////////////////////////////////////
__kernel void RGB2Gray(__global const uchar* srcptr, int srcstep, int srcoffset,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols)
{
#if 1
const int x = get_global_id(0);
const int y = get_global_id(1);
int x = get_global_id(0);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
__global const DATA_TYPE* src = (__global const DATA_TYPE*)(srcptr + mad24(y, srcstep, srcoffset + x * scnbytes));
__global DATA_TYPE* dst = (__global DATA_TYPE*)(dstptr + mad24(y, dststep, dstoffset + x * dcnbytes));
DATA_TYPE_4 src_pix = vload4(0, src);
#ifdef DEPTH_5
dst[0] = src[bidx] * 0.114f + src[1] * 0.587f + src[(bidx^2)] * 0.299f;
dst[0] = src_pix.B_COMP * 0.114f + src_pix.G_COMP * 0.587f + src_pix.R_COMP * 0.299f;
#else
dst[0] = (DATA_TYPE)CV_DESCALE((src[bidx] * B2Y + src[1] * G2Y + src[(bidx^2)] * R2Y), yuv_shift);
dst[0] = (DATA_TYPE)CV_DESCALE((src_pix.B_COMP * B2Y + src_pix.G_COMP * G2Y + src_pix.R_COMP * R2Y), yuv_shift);
#endif
}
#else
const int x_min = get_global_id(0)*STRIPE_SIZE;
const int x_max = min(x_min + STRIPE_SIZE, cols);
const int y = get_global_id(1);
if( y < rows )
{
__global const DATA_TYPE* src = (__global const DATA_TYPE*)(srcptr +
mad24(y, srcstep, srcoffset)) + x_min*scn;
__global DATA_TYPE* dst = (__global DATA_TYPE*)(dstptr + mad24(y, dststep, dstoffset));
int x;
for( x = x_min; x < x_max; x++, src += scn )
#ifdef DEPTH_5
dst[x] = src[bidx] * 0.114f + src[1] * 0.587f + src[(bidx^2)] * 0.299f;
#else
dst[x] = (DATA_TYPE)(mad24(src[bidx], B2Y, mad24(src[1], G2Y,
mad24(src[(bidx^2)], R2Y, 1 << (yuv_shift-1)))) >> yuv_shift);
#endif
++y;
}
}
#endif
}
__kernel void Gray2RGB(__global const uchar* srcptr, int srcstep, int srcoffset,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols)
{
const int x = get_global_id(0);
const int y = get_global_id(1);
int x = get_global_id(0);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
__global const DATA_TYPE* src = (__global const DATA_TYPE*)(srcptr + mad24(y, srcstep, srcoffset + x * scnbytes));
__global DATA_TYPE* dst = (__global DATA_TYPE*)(dstptr + mad24(y, dststep, dstoffset + x * dcnbytes));
......@@ -158,6 +172,9 @@ __kernel void Gray2RGB(__global const uchar* srcptr, int srcstep, int srcoffset,
dst[3] = MAX_NUM;
#endif
}
++y;
}
}
}
///////////////////////////////////// RGB <-> YUV //////////////////////////////////////
......@@ -170,13 +187,19 @@ __kernel void RGB2YUV(__global const uchar* srcptr, int srcstep, int srcoffset,
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
__global const DATA_TYPE* src = (__global const DATA_TYPE*)(srcptr + mad24(y, srcstep, srcoffset + x * scnbytes));
__global DATA_TYPE* dst = (__global DATA_TYPE*)(dstptr + mad24(y, dststep, dstoffset + x * dcnbytes));
DATA_TYPE b=src[bidx], g=src[1], r=src[bidx^2];
DATA_TYPE_4 src_pix = vload4(0, src);
DATA_TYPE b=src_pix.B_COMP, g=src_pix.G_COMP, r=src_pix.R_COMP;
#ifdef DEPTH_5
__constant float * coeffs = c_RGB2YUVCoeffs_f;
......@@ -195,6 +218,9 @@ __kernel void RGB2YUV(__global const uchar* srcptr, int srcstep, int srcoffset,
dst[1] = SAT_CAST( U );
dst[2] = SAT_CAST( V );
}
++y;
}
}
}
__constant float c_YUV2RGBCoeffs_f[5] = { 2.032f, -0.395f, -0.581f, 1.140f };
......@@ -205,13 +231,19 @@ __kernel void YUV2RGB(__global const uchar* srcptr, int srcstep, int srcoffset,
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
__global const DATA_TYPE* src = (__global const DATA_TYPE*)(srcptr + mad24(y, srcstep, srcoffset + x * scnbytes));
__global DATA_TYPE* dst = (__global DATA_TYPE*)(dstptr + mad24(y, dststep, dstoffset + x * dcnbytes));
DATA_TYPE Y = src[0], U = src[1], V = src[2];
DATA_TYPE_4 src_pix = vload4(0, src);
DATA_TYPE Y = src_pix.x, U = src_pix.y, V = src_pix.z;
#ifdef DEPTH_5
__constant float * coeffs = c_YUV2RGBCoeffs_f;
......@@ -232,6 +264,9 @@ __kernel void YUV2RGB(__global const uchar* srcptr, int srcstep, int srcoffset,
dst[3] = MAX_NUM;
#endif
}
++y;
}
}
}
__constant int ITUR_BT_601_CY = 1220542;
......@@ -246,9 +281,14 @@ __kernel void YUV2RGB_NV12(__global const uchar* srcptr, int srcstep, int srcoff
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows / 2 && x < cols / 2 )
if (x < cols / 2)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows / 2 )
{
__global const uchar* ysrc = srcptr + mad24(y << 1, srcstep, (x << 1) + srcoffset);
__global const uchar* usrc = srcptr + mad24(rows + y, srcstep, (x << 1) + srcoffset);
......@@ -299,6 +339,9 @@ __kernel void YUV2RGB_NV12(__global const uchar* srcptr, int srcstep, int srcoff
dst2[7] = 255;
#endif
}
++y;
}
}
}
///////////////////////////////////// RGB <-> YCrCb //////////////////////////////////////
......@@ -311,13 +354,19 @@ __kernel void RGB2YCrCb(__global const uchar* srcptr, int srcstep, int srcoffset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
__global const DATA_TYPE* src = (__global const DATA_TYPE*)(srcptr + mad24(y, srcstep, srcoffset + x * scnbytes));
__global DATA_TYPE* dst = (__global DATA_TYPE*)(dstptr + mad24(y, dststep, dstoffset + x * dcnbytes));
DATA_TYPE b=src[bidx], g=src[1], r=src[bidx^2];
DATA_TYPE_4 src_pix = vload4(0, src);
DATA_TYPE b=src_pix.B_COMP, g=src_pix.G_COMP, r=src_pix.R_COMP;
#ifdef DEPTH_5
__constant float * coeffs = c_RGB2YCrCbCoeffs_f;
......@@ -336,6 +385,9 @@ __kernel void RGB2YCrCb(__global const uchar* srcptr, int srcstep, int srcoffset
dst[1] = SAT_CAST( Cr );
dst[2] = SAT_CAST( Cb );
}
++y;
}
}
}
__constant float c_YCrCb2RGBCoeffs_f[4] = { 1.403f, -0.714f, -0.344f, 1.773f };
......@@ -346,16 +398,22 @@ __kernel void YCrCb2RGB(__global const uchar* src, int src_step, int src_offset,
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
__global const DATA_TYPE * srcptr = (__global const DATA_TYPE*)(src + src_idx);
__global DATA_TYPE * dstptr = (__global DATA_TYPE*)(dst + dst_idx);
DATA_TYPE y = srcptr[0], cr = srcptr[1], cb = srcptr[2];
DATA_TYPE_4 src_pix = vload4(0, srcptr);
DATA_TYPE y = src_pix.x, cr = src_pix.y, cb = src_pix.z;
#ifdef DEPTH_5
__constant float * coeff = c_YCrCb2RGBCoeffs_f;
......@@ -376,6 +434,9 @@ __kernel void YCrCb2RGB(__global const uchar* src, int src_step, int src_offset,
dstptr[3] = MAX_NUM;
#endif
}
++y;
}
}
}
///////////////////////////////////// RGB <-> XYZ //////////////////////////////////////
......@@ -385,9 +446,14 @@ __kernel void RGB2XYZ(__global const uchar * srcptr, int src_step, int src_offse
int rows, int cols, __constant COEFF_TYPE * coeffs)
{
int dx = get_global_id(0);
int dy = get_global_id(1);
int dy = get_global_id(1) * PIX_PER_WI_Y;
if (dy < rows && dx < cols)
if (dx < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (dy < rows)
{
int src_idx = mad24(dy, src_step, src_offset + dx * scnbytes);
int dst_idx = mad24(dy, dst_step, dst_offset + dx * dcnbytes);
......@@ -395,7 +461,8 @@ __kernel void RGB2XYZ(__global const uchar * srcptr, int src_step, int src_offse
__global const DATA_TYPE * src = (__global const DATA_TYPE *)(srcptr + src_idx);
__global DATA_TYPE * dst = (__global DATA_TYPE *)(dstptr + dst_idx);
DATA_TYPE r = src[0], g = src[1], b = src[2];
DATA_TYPE_4 src_pix = vload4(0, src);
DATA_TYPE r = src_pix.x, g = src_pix.y, b = src_pix.z;
#ifdef DEPTH_5
float x = r * coeffs[0] + g * coeffs[1] + b * coeffs[2];
......@@ -410,6 +477,9 @@ __kernel void RGB2XYZ(__global const uchar * srcptr, int src_step, int src_offse
dst[1] = SAT_CAST(y);
dst[2] = SAT_CAST(z);
}
++dy;
}
}
}
__kernel void XYZ2RGB(__global const uchar * srcptr, int src_step, int src_offset,
......@@ -417,9 +487,14 @@ __kernel void XYZ2RGB(__global const uchar * srcptr, int src_step, int src_offse
int rows, int cols, __constant COEFF_TYPE * coeffs)
{
int dx = get_global_id(0);
int dy = get_global_id(1);
int dy = get_global_id(1) * PIX_PER_WI_Y;
if (dy < rows && dx < cols)
if (dx < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (dy < rows)
{
int src_idx = mad24(dy, src_step, src_offset + dx * scnbytes);
int dst_idx = mad24(dy, dst_step, dst_offset + dx * dcnbytes);
......@@ -427,7 +502,8 @@ __kernel void XYZ2RGB(__global const uchar * srcptr, int src_step, int src_offse
__global const DATA_TYPE * src = (__global const DATA_TYPE *)(srcptr + src_idx);
__global DATA_TYPE * dst = (__global DATA_TYPE *)(dstptr + dst_idx);
DATA_TYPE x = src[0], y = src[1], z = src[2];
DATA_TYPE_4 src_pix = vload4(0, src);
DATA_TYPE x = src_pix.x, y = src_pix.y, z = src_pix.z;
#ifdef DEPTH_5
float b = x * coeffs[0] + y * coeffs[1] + z * coeffs[2];
......@@ -445,6 +521,9 @@ __kernel void XYZ2RGB(__global const uchar * srcptr, int src_step, int src_offse
dst[3] = MAX_NUM;
#endif
}
++dy;
}
}
}
///////////////////////////////////// RGB[A] <-> BGR[A] //////////////////////////////////////
......@@ -454,24 +533,30 @@ __kernel void RGB(__global const uchar* srcptr, int src_step, int src_offset,
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
__global const DATA_TYPE * src = (__global const DATA_TYPE *)(srcptr + src_idx);
__global DATA_TYPE * dst = (__global DATA_TYPE *)(dstptr + dst_idx);
DATA_TYPE_4 src_pix = vload4(0, src);
#ifdef REVERSE
dst[0] = src[2];
dst[1] = src[1];
dst[2] = src[0];
dst[0] = src_pix.z;
dst[1] = src_pix.y;
dst[2] = src_pix.x;
#else
dst[0] = src[0];
dst[1] = src[1];
dst[2] = src[2];
dst[0] = src_pix.x;
dst[1] = src_pix.y;
dst[2] = src_pix.z;
#endif
#if dcn == 4
......@@ -482,6 +567,9 @@ __kernel void RGB(__global const uchar* srcptr, int src_step, int src_offset,
#endif
#endif
}
++y;
}
}
}
///////////////////////////////////// RGB5x5 <-> RGB //////////////////////////////////////
......@@ -491,9 +579,14 @@ __kernel void RGB5x52RGB(__global const uchar* src, int src_step, int src_offset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
......@@ -517,6 +610,9 @@ __kernel void RGB5x52RGB(__global const uchar* src, int src_step, int src_offset
#endif
#endif
}
++y;
}
}
}
__kernel void RGB2RGB5x5(__global const uchar* src, int src_step, int src_offset,
......@@ -524,22 +620,31 @@ __kernel void RGB2RGB5x5(__global const uchar* src, int src_step, int src_offset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
uchar4 src_pix = vload4(0, src + src_idx);
#if greenbits == 6
*((__global ushort*)(dst + dst_idx)) = (ushort)((src[src_idx + bidx] >> 3)|((src[src_idx + 1]&~3) << 3)|((src[src_idx + (bidx^2)]&~7) << 8));
*((__global ushort*)(dst + dst_idx)) = (ushort)((src_pix.B_COMP >> 3)|((src_pix.G_COMP&~3) << 3)|((src_pix.R_COMP&~7) << 8));
#elif scn == 3
*((__global ushort*)(dst + dst_idx)) = (ushort)((src[src_idx + bidx] >> 3)|((src[src_idx + 1]&~7) << 2)|((src[src_idx + (bidx^2)]&~7) << 7));
*((__global ushort*)(dst + dst_idx)) = (ushort)((src_pix.B_COMP >> 3)|((src_pix.G_COMP&~7) << 2)|((src_pix.R_COMP&~7) << 7));
#else
*((__global ushort*)(dst + dst_idx)) = (ushort)((src[src_idx + bidx] >> 3)|((src[src_idx + 1]&~7) << 2)|
((src[src_idx + (bidx^2)]&~7) << 7)|(src[src_idx + 3] ? 0x8000 : 0));
*((__global ushort*)(dst + dst_idx)) = (ushort)((src_pix.B_COMP >> 3)|((src_pix.G_COMP&~7) << 2)|
((src_pix.R_COMP&~7) << 7)|(src_pix.w ? 0x8000 : 0));
#endif
}
++y;
}
}
}
///////////////////////////////////// RGB5x5 <-> Gray //////////////////////////////////////
......@@ -549,9 +654,14 @@ __kernel void BGR5x52Gray(__global const uchar* src, int src_step, int src_offse
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x);
......@@ -567,6 +677,9 @@ __kernel void BGR5x52Gray(__global const uchar* src, int src_step, int src_offse
((t >> 7) & 0xf8)*R2Y, yuv_shift);
#endif
}
++y;
}
}
}
__kernel void Gray2BGR5x5(__global const uchar* src, int src_step, int src_offset,
......@@ -574,9 +687,14 @@ __kernel void Gray2BGR5x5(__global const uchar* src, int src_step, int src_offse
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
......@@ -589,6 +707,9 @@ __kernel void Gray2BGR5x5(__global const uchar* src, int src_step, int src_offse
*((__global ushort*)(dst + dst_idx)) = (ushort)(t|(t << 5)|(t << 10));
#endif
}
++y;
}
}
}
//////////////////////////////////// RGB <-> HSV //////////////////////////////////////
......@@ -608,14 +729,20 @@ __kernel void RGB2HSV(__global const uchar* src, int src_step, int src_offset,
__constant int * sdiv_table, __constant int * hdiv_table)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
uchar4 src_pix = vload4(0, src + src_idx);
int b = src[src_idx + bidx], g = src[src_idx + 1], r = src[src_idx + (bidx^2)];
int b = src_pix.B_COMP, g = src_pix.G_COMP, r = src_pix.R_COMP;
int h, s, v = b;
int vmin = b, diff;
int vr, vg;
......@@ -639,6 +766,9 @@ __kernel void RGB2HSV(__global const uchar* src, int src_step, int src_offset,
dst[dst_idx + 1] = (uchar)s;
dst[dst_idx + 2] = (uchar)v;
}
++y;
}
}
}
__kernel void HSV2RGB(__global const uchar* src, int src_step, int src_offset,
......@@ -646,14 +776,20 @@ __kernel void HSV2RGB(__global const uchar* src, int src_step, int src_offset,
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
uchar4 src_pix = vload4(0, src + src_idx);
float h = src[src_idx], s = src[src_idx + 1]*(1/255.f), v = src[src_idx + 2]*(1/255.f);
float h = src_pix.x, s = src_pix.y*(1/255.f), v = src_pix.z*(1/255.f);
float b, g, r;
if (s != 0)
......@@ -692,6 +828,9 @@ __kernel void HSV2RGB(__global const uchar* src, int src_step, int src_offset,
dst[dst_idx + 3] = MAX_NUM;
#endif
}
++y;
}
}
}
#elif defined DEPTH_5
......@@ -701,17 +840,23 @@ __kernel void RGB2HSV(__global const uchar* srcptr, int src_step, int src_offset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
__global const float * src = (__global const float *)(srcptr + src_idx);
__global float * dst = (__global float *)(dstptr + dst_idx);
float4 src_pix = vload4(0, src);
float b = src[bidx], g = src[1], r = src[bidx^2];
float b = src_pix.B_COMP, g = src_pix.G_COMP, r = src_pix.R_COMP;
float h, s, v;
float vmin, diff;
......@@ -738,6 +883,9 @@ __kernel void RGB2HSV(__global const uchar* srcptr, int src_step, int src_offset
dst[1] = s;
dst[2] = v;
}
++y;
}
}
}
__kernel void HSV2RGB(__global const uchar* srcptr, int src_step, int src_offset,
......@@ -745,17 +893,23 @@ __kernel void HSV2RGB(__global const uchar* srcptr, int src_step, int src_offset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
__global const float * src = (__global const float *)(srcptr + src_idx);
__global float * dst = (__global float *)(dstptr + dst_idx);
float4 src_pix = vload4(0, src);
float h = src[0], s = src[1], v = src[2];
float h = src_pix.x, s = src_pix.y, v = src_pix.z;
float b, g, r;
if (s != 0)
......@@ -794,6 +948,9 @@ __kernel void HSV2RGB(__global const uchar* srcptr, int src_step, int src_offset
dst[3] = MAX_NUM;
#endif
}
++y;
}
}
}
#endif
......@@ -807,14 +964,20 @@ __kernel void RGB2HLS(__global const uchar* src, int src_step, int src_offset,
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
uchar4 src_pix = vload4(0, src + src_idx);
float b = src[src_idx + bidx]*(1/255.f), g = src[src_idx + 1]*(1/255.f), r = src[src_idx + (bidx^2)]*(1/255.f);
float b = src_pix.B_COMP*(1/255.f), g = src_pix.G_COMP*(1/255.f), r = src_pix.R_COMP*(1/255.f);
float h = 0.f, s = 0.f, l;
float vmin, vmax, diff;
......@@ -846,6 +1009,9 @@ __kernel void RGB2HLS(__global const uchar* src, int src_step, int src_offset,
dst[dst_idx + 1] = convert_uchar_sat_rte(l*255.f);
dst[dst_idx + 2] = convert_uchar_sat_rte(s*255.f);
}
++y;
}
}
}
__kernel void HLS2RGB(__global const uchar* src, int src_step, int src_offset,
......@@ -853,14 +1019,20 @@ __kernel void HLS2RGB(__global const uchar* src, int src_step, int src_offset,
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
uchar4 src_pix = vload4(0, src + src_idx);
float h = src[src_idx], l = src[src_idx + 1]*(1.f/255.f), s = src[src_idx + 2]*(1.f/255.f);
float h = src_pix.x, l = src_pix.y*(1.f/255.f), s = src_pix.z*(1.f/255.f);
float b, g, r;
if (s != 0)
......@@ -898,6 +1070,9 @@ __kernel void HLS2RGB(__global const uchar* src, int src_step, int src_offset,
dst[dst_idx + 3] = MAX_NUM;
#endif
}
++y;
}
}
}
#elif defined DEPTH_5
......@@ -907,17 +1082,23 @@ __kernel void RGB2HLS(__global const uchar* srcptr, int src_step, int src_offset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
__global const float * src = (__global const float *)(srcptr + src_idx);
__global float * dst = (__global float *)(dstptr + dst_idx);
float4 src_pix = vload4(0, src);
float b = src[bidx], g = src[1], r = src[bidx^2];
float b = src_pix.B_COMP, g = src_pix.G_COMP, r = src_pix.R_COMP;
float h = 0.f, s = 0.f, l;
float vmin, vmax, diff;
......@@ -949,6 +1130,9 @@ __kernel void RGB2HLS(__global const uchar* srcptr, int src_step, int src_offset
dst[1] = l;
dst[2] = s;
}
++y;
}
}
}
__kernel void HLS2RGB(__global const uchar* srcptr, int src_step, int src_offset,
......@@ -956,17 +1140,23 @@ __kernel void HLS2RGB(__global const uchar* srcptr, int src_step, int src_offset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
__global const float * src = (__global const float *)(srcptr + src_idx);
__global float * dst = (__global float *)(dstptr + dst_idx);
float4 src_pix = vload4(0, src);
float h = src[0], l = src[1], s = src[2];
float h = src_pix.x, l = src_pix.y, s = src_pix.z;
float b, g, r;
if (s != 0)
......@@ -1005,6 +1195,9 @@ __kernel void HLS2RGB(__global const uchar* srcptr, int src_step, int src_offset
dst[3] = MAX_NUM;
#endif
}
++y;
}
}
}
#endif
......@@ -1018,22 +1211,30 @@ __kernel void RGBA2mRGBA(__global const uchar* src, int src_step, int src_offset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
x <<= 2;
int src_idx = mad24(y, src_step, src_offset + x);
int dst_idx = mad24(y, dst_step, dst_offset + x);
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + (x << 2));
int dst_idx = mad24(y, dst_step, dst_offset + (x << 2));
uchar4 src_pix = vload4(0, src + src_idx);
uchar v0 = src[src_idx], v1 = src[src_idx + 1];
uchar v2 = src[src_idx + 2], v3 = src[src_idx + 3];
uchar v0 = src_pix.x, v1 = src_pix.y;
uchar v2 = src_pix.z, v3 = src_pix.w;
dst[dst_idx] = (v0 * v3 + HALF_MAX) / MAX_NUM;
dst[dst_idx + 1] = (v1 * v3 + HALF_MAX) / MAX_NUM;
dst[dst_idx + 2] = (v2 * v3 + HALF_MAX) / MAX_NUM;
dst[dst_idx + 3] = v3;
}
++y;
}
}
}
__kernel void mRGBA2RGBA(__global const uchar* src, int src_step, int src_offset,
......@@ -1041,16 +1242,21 @@ __kernel void mRGBA2RGBA(__global const uchar* src, int src_step, int src_offset
int rows, int cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
x <<= 2;
int src_idx = mad24(y, src_step, src_offset + x);
int dst_idx = mad24(y, dst_step, dst_offset + x);
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + (x << 2));
int dst_idx = mad24(y, dst_step, dst_offset + (x << 2));
uchar4 src_pix = vload4(0, src + src_idx);
uchar v0 = src[src_idx], v1 = src[src_idx + 1];
uchar v2 = src[src_idx + 2], v3 = src[src_idx + 3];
uchar v0 = src_pix.x, v1 = src_pix.y;
uchar v2 = src_pix.z, v3 = src_pix.w;
uchar v3_half = v3 / 2;
dst[dst_idx] = v3 == 0 ? 0 : (v0 * MAX_NUM + v3_half) / v3;
......@@ -1058,6 +1264,9 @@ __kernel void mRGBA2RGBA(__global const uchar* src, int src_step, int src_offset
dst[dst_idx + 2] = v3 == 0 ? 0 : (v2 * MAX_NUM + v3_half) / v3;
dst[dst_idx + 3] = v3;
}
++y;
}
}
}
#endif
......@@ -1086,21 +1295,27 @@ __kernel void BGR2Lab(__global const uchar * src, int src_step, int src_offset,
__constant int * coeffs, int Lscale, int Lshift)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
src += src_idx;
dst += dst_idx;
__global const uchar* src_ptr = src + src_idx;
__global uchar* dst_ptr = dst + dst_idx;
uchar4 src_pix = vload4(0, src_ptr);
int C0 = coeffs[0], C1 = coeffs[1], C2 = coeffs[2],
C3 = coeffs[3], C4 = coeffs[4], C5 = coeffs[5],
C6 = coeffs[6], C7 = coeffs[7], C8 = coeffs[8];
int R = gammaTab[src[0]], G = gammaTab[src[1]], B = gammaTab[src[2]];
int R = gammaTab[src_pix.x], G = gammaTab[src_pix.y], B = gammaTab[src_pix.z];
int fX = LabCbrtTab_b[CV_DESCALE(R*C0 + G*C1 + B*C2, lab_shift)];
int fY = LabCbrtTab_b[CV_DESCALE(R*C3 + G*C4 + B*C5, lab_shift)];
int fZ = LabCbrtTab_b[CV_DESCALE(R*C6 + G*C7 + B*C8, lab_shift)];
......@@ -1109,9 +1324,12 @@ __kernel void BGR2Lab(__global const uchar * src, int src_step, int src_offset,
int a = CV_DESCALE( 500*(fX - fY) + 128*(1 << lab_shift2), lab_shift2 );
int b = CV_DESCALE( 200*(fY - fZ) + 128*(1 << lab_shift2), lab_shift2 );
dst[0] = SAT_CAST(L);
dst[1] = SAT_CAST(a);
dst[2] = SAT_CAST(b);
dst_ptr[0] = SAT_CAST(L);
dst_ptr[1] = SAT_CAST(a);
dst_ptr[2] = SAT_CAST(b);
}
++y;
}
}
}
......@@ -1125,23 +1343,29 @@ __kernel void BGR2Lab(__global const uchar * srcptr, int src_step, int src_offse
__constant float * coeffs, float _1_3, float _a)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
__global const float * src = (__global const float *)(srcptr + src_idx);
__global float * dst = (__global float *)(dstptr + dst_idx);
float4 src_pix = vload4(0, src);
float C0 = coeffs[0], C1 = coeffs[1], C2 = coeffs[2],
C3 = coeffs[3], C4 = coeffs[4], C5 = coeffs[5],
C6 = coeffs[6], C7 = coeffs[7], C8 = coeffs[8];
float R = clamp(src[0], 0.0f, 1.0f);
float G = clamp(src[1], 0.0f, 1.0f);
float B = clamp(src[2], 0.0f, 1.0f);
float R = clamp(src_pix.x, 0.0f, 1.0f);
float G = clamp(src_pix.y, 0.0f, 1.0f);
float B = clamp(src_pix.z, 0.0f, 1.0f);
#ifdef SRGB
R = splineInterpolate(R * GammaTabScale, gammaTab, GAMMA_TAB_SIZE);
......@@ -1165,6 +1389,9 @@ __kernel void BGR2Lab(__global const uchar * srcptr, int src_step, int src_offse
dst[1] = a;
dst[2] = b;
}
++y;
}
}
}
#endif
......@@ -1225,20 +1452,26 @@ __kernel void Lab2BGR(__global const uchar * src, int src_step, int src_offset,
__constant float * coeffs, float lThresh, float fThresh)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
src += src_idx;
dst += dst_idx;
__global const uchar* src_ptr = src + src_idx;
__global uchar* dst_ptr = dst + dst_idx;
uchar4 src_pix = vload4(0, src_ptr);
float srcbuf[3], dstbuf[3];
srcbuf[0] = src[0]*(100.f/255.f);
srcbuf[1] = convert_float(src[1] - 128);
srcbuf[2] = convert_float(src[2] - 128);
srcbuf[0] = src_pix.x*(100.f/255.f);
srcbuf[1] = convert_float(src_pix.y - 128);
srcbuf[2] = convert_float(src_pix.z - 128);
Lab2BGR_f(&srcbuf[0], &dstbuf[0],
#ifdef SRGB
......@@ -1246,13 +1479,16 @@ __kernel void Lab2BGR(__global const uchar * src, int src_step, int src_offset,
#endif
coeffs, lThresh, fThresh);
dst[0] = SAT_CAST(dstbuf[0] * 255.0f);
dst[1] = SAT_CAST(dstbuf[1] * 255.0f);
dst[2] = SAT_CAST(dstbuf[2] * 255.0f);
dst_ptr[0] = SAT_CAST(dstbuf[0] * 255.0f);
dst_ptr[1] = SAT_CAST(dstbuf[1] * 255.0f);
dst_ptr[2] = SAT_CAST(dstbuf[2] * 255.0f);
#if dcn == 4
dst[3] = MAX_NUM;
dst_ptr[3] = MAX_NUM;
#endif
}
++y;
}
}
}
#elif defined DEPTH_5
......@@ -1265,18 +1501,24 @@ __kernel void Lab2BGR(__global const uchar * srcptr, int src_step, int src_offse
__constant float * coeffs, float lThresh, float fThresh)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y = get_global_id(1) * PIX_PER_WI_Y;
if (y < rows && x < cols)
if (x < cols)
{
#pragma unroll
for (int cy = 0; cy < PIX_PER_WI_Y; ++cy)
{
if (y < rows)
{
int src_idx = mad24(y, src_step, src_offset + x * scnbytes);
int dst_idx = mad24(y, dst_step, dst_offset + x * dcnbytes);
__global const float * src = (__global const float *)(srcptr + src_idx);
__global float * dst = (__global float *)(dstptr + dst_idx);
float4 src_pix = vload4(0, src);
float srcbuf[3], dstbuf[3];
srcbuf[0] = src[0], srcbuf[1] = src[1], srcbuf[2] = src[2];
srcbuf[0] = src_pix.x, srcbuf[1] = src_pix.y, srcbuf[2] = src_pix.z;
Lab2BGR_f(&srcbuf[0], &dstbuf[0],
#ifdef SRGB
......@@ -1289,6 +1531,9 @@ __kernel void Lab2BGR(__global const uchar * srcptr, int src_step, int src_offse
dst[3] = MAX_NUM;
#endif
}
++y;
}
}
}
#endif
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