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

added norm_inf support to minmaxloc kernel

Browse Source
This commit is contained in:
Ilya Lavrenov
2014-06-09 15:32:35 +04:00
parent 7f2662b310
commit 634da9f3bf
4 changed files with 26 additions and 67 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
modules/core/src/opencl/minmaxloc.cl
View File

@@ -41,10 +41,15 @@
#if kercn != 3
#define loadpix(addr) *(__global const srcT *)(addr)
#define srcTSIZE (int)sizeof(srcT1)
#define srcTSIZE (int)sizeof(srcT)
#else
#define loadpix(addr) vload3(0, (__global const srcT1 *)(addr))
#define srcTSIZE ((int)sizeof(srcT1))
#define srcTSIZE ((int)sizeof(srcT1) * 3)
#endif
#ifndef HAVE_MASK
#undef srcTSIZE
#define srcTSIZE (int)sizeof(srcT1)
#endif
#ifdef NEED_MINLOC
@@ -106,7 +111,12 @@ __kernel void minmaxloc(__global const uchar * srcptr, int src_step, int src_off
{
int lid = get_local_id(0);
int gid = get_group_id(0);
int id = get_global_id(0) * kercn;
int id = get_global_id(0)
#ifndef HAVE_MASK
* kercn;
#else
;
#endif
srcptr += src_offset;
#ifdef HAVE_MASK
@@ -150,7 +160,11 @@ __kernel void minmaxloc(__global const uchar * srcptr, int src_step, int src_off
dstT temp2;
#endif
for (int grain = groupnum * WGS * kercn; id < total; id += grain)
for (int grain = groupnum * WGS
#ifndef HAVE_MASK
* kercn
#endif
; id < total; id += grain)
{
#ifdef HAVE_MASK
#ifdef HAVE_MASK_CONT

25
modules/core/src/opencl/reduce.cl
View File

@@ -543,32 +543,9 @@
#define CALC_RESULT \
storepix(localmem[0], dstptr + dstTSIZE * gid)
// norm (NORM_INF) with cn > 1 and mask
#elif defined OP_NORM_INF_MASK
#define DECLARE_LOCAL_MEM \
__local srcT localmem_max[WGS2_ALIGNED]
#define DEFINE_ACCUMULATOR \
srcT maxval = MIN_VAL, temp
#define REDUCE_GLOBAL \
MASK_INDEX; \
if (mask[mask_index]) \
{ \
temp = loadpix(srcptr + src_index); \
maxval = max(maxval, (srcT)(temp >= (srcT)(0) ? temp : -temp)); \
}
#define SET_LOCAL_1 \
localmem_max[lid] = maxval
#define REDUCE_LOCAL_1 \
localmem_max[lid - WGS2_ALIGNED] = max(maxval, localmem_max[lid - WGS2_ALIGNED])
#define REDUCE_LOCAL_2 \
localmem_max[lid] = max(localmem_max[lid], localmem_max[lid2])
#define CALC_RESULT \
storepix(localmem_max[0], dstptr + dstTSIZE * gid)
#else
#error "No operation"
#endif // end of norm (NORM_INF) with cn > 1 and mask
#endif
#ifdef OP_DOT
#undef EXTRA_PARAMS