potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 15:53:03 +04:00
Code Packages Releases Wiki Activity

OCL: fix incompatibility with Mali ruintime

Browse Source
This commit is contained in:
Maksim Shabunin
2023-12-21 00:30:44 +03:00
parent 9434c89ba0
commit adde942e34
22 changed files with 404 additions and 404 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/src/arithm.cpp
+1 -1
View File
@@ -1643,7 +1643,7 @@ static bool ocl_inRange( InputArray _src, InputArray _lowerb,
if (kercn % cn != 0)
kercn = cn;
int colsPerWI = kercn / cn;
String opts = format("%s-D cn=%d -D srcT=%s -D srcT1=%s -D dstT=%s -D kercn=%d -D depth=%d%s -D colsPerWI=%d",
String opts = format("%s-D CN=%d -D SRC_T=%s -D SRC_T1=%s -D DST_T=%s -D KERCN=%d -D DEPTH=%d%s -D COLS_PER_WI=%d",
haveScalar ? "-D HAVE_SCALAR " : "", cn, ocl::typeToStr(CV_MAKE_TYPE(sdepth, kercn)),
ocl::typeToStr(sdepth), ocl::typeToStr(CV_8UC(colsPerWI)), kercn, sdepth,
doubleSupport ? " -D DOUBLE_SUPPORT" : "", colsPerWI);
modules/core/src/opencl/inrange.cl
+24 -24
View File
@@ -52,7 +52,7 @@
__kernel void inrange(__global const uchar * src1ptr, int src1_step, int src1_offset,
__global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
#ifdef HAVE_SCALAR
__global const srcT1 * src2, __global const srcT1 * src3,
__global const SRC_T1 * src2, __global const SRC_T1 * src3,
#else
__global const uchar * src2ptr, int src2_step, int src2_offset,
__global const uchar * src3ptr, int src3_step, int src3_offset,
@@ -64,56 +64,56 @@ __kernel void inrange(__global const uchar * src1ptr, int src1_step, int src1_of
if (x < dst_cols)
{
int src1_index = mad24(y0, src1_step, mad24(x, (int)sizeof(srcT1) * kercn, src1_offset));
int dst_index = mad24(y0, dst_step, mad24(x, colsPerWI, dst_offset));
int src1_index = mad24(y0, src1_step, mad24(x, (int)sizeof(SRC_T1) * KERCN, src1_offset));
int dst_index = mad24(y0, dst_step, mad24(x, COLS_PER_WI, dst_offset));
#ifndef HAVE_SCALAR
int src2_index = mad24(y0, src2_step, mad24(x, (int)sizeof(srcT1) * kercn, src2_offset));
int src3_index = mad24(y0, src3_step, mad24(x, (int)sizeof(srcT1) * kercn, src3_offset));
int src2_index = mad24(y0, src2_step, mad24(x, (int)sizeof(SRC_T1) * KERCN, src2_offset));
int src3_index = mad24(y0, src3_step, mad24(x, (int)sizeof(SRC_T1) * KERCN, src3_offset));
#endif
for (int y = y0, y1 = min(dst_rows, y0 + rowsPerWI); y < y1; ++y, src1_index += src1_step, dst_index += dst_step)
{
#if kercn >= cn && kercn == 4 && depth <= 4 && !defined HAVE_SCALAR
srcT src1 = *(__global const srcT *)(src1ptr + src1_index);
srcT src2 = *(__global const srcT *)(src2ptr + src2_index);
srcT src3 = *(__global const srcT *)(src3ptr + src3_index);
__global dstT * dst = (__global dstT *)(dstptr + dst_index);
#if cn == 1
dst[0] = src2 > src1 || src3 < src1 ? (dstT)(0) : (dstT)(255);
#elif cn == 2
dst[0] = (dstT)(src2.xy > src1.xy || src3.xy < src1.xy ||
src2.zw > src1.zw || src3.zw < src1.zw ? (dstT)(0) : (dstT)(255);
#elif cn == 4
dst[0] = (dstT)(src2.x > src1.x || src3.x < src1.x ||
#if KERCN >= CN && KERCN == 4 && DEPTH <= 4 && !defined HAVE_SCALAR
SRC_T src1 = *(__global const SRC_T *)(src1ptr + src1_index);
SRC_T src2 = *(__global const SRC_T *)(src2ptr + src2_index);
SRC_T src3 = *(__global const SRC_T *)(src3ptr + src3_index);
__global DST_T * dst = (__global DST_T *)(dstptr + dst_index);
#if CN == 1
dst[0] = src2 > src1 || src3 < src1 ? (DST_T)(0) : (DST_T)(255);
#elif CN == 2
dst[0] = (DST_T)(src2.xy > src1.xy || src3.xy < src1.xy ||
src2.zw > src1.zw || src3.zw < src1.zw ? (DST_T)(0) : (DST_T)(255);
#elif CN == 4
dst[0] = (DST_T)(src2.x > src1.x || src3.x < src1.x ||
src2.y > src1.y || src3.y < src1.y ||
src2.z > src1.z || src3.z < src1.z ||
src2.w > src1.w || src3.w < src1.w ? 0 : 255);
#endif
#else
__global const srcT1 * src1 = (__global const srcT1 *)(src1ptr + src1_index);
__global const SRC_T1 * src1 = (__global const SRC_T1 *)(src1ptr + src1_index);
__global uchar * dst = dstptr + dst_index;
#ifndef HAVE_SCALAR
__global const srcT1 * src2 = (__global const srcT1 *)(src2ptr + src2_index);
__global const srcT1 * src3 = (__global const srcT1 *)(src3ptr + src3_index);
__global const SRC_T1 * src2 = (__global const SRC_T1 *)(src2ptr + src2_index);
__global const SRC_T1 * src3 = (__global const SRC_T1 *)(src3ptr + src3_index);
#endif
#pragma unroll
for (int px = 0; px < colsPerWI; ++px, src1 += cn
for (int px = 0; px < COLS_PER_WI; ++px, src1 += CN
#ifndef HAVE_SCALAR
, src2 += cn, src3 += cn
, src2 += CN, src3 += CN
#endif
)
{
dst[px] = 255;
for (int c = 0; c < cn; ++c)
for (int c = 0; c < CN; ++c)
if (src2[c] > src1[c] || src3[c] < src1[c])
{
dst[px] = 0;
break;
}
}
#endif // kercn >= cn
#endif // KERCN >= CN
#ifndef HAVE_SCALAR
src2_index += src2_step;
src3_index += src3_step;