mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-06-12 20:42:53 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #9833 from tomoaki0705:universalMathFuncs

Browse Source
This commit is contained in:
Vadim Pisarevsky 2017-10-16 10:46:55 +00:00
commit 1563300197
1 changed files with 19 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
modules/core/src/mathfuncs.cpp
View File

@ -605,24 +605,18 @@ void polarToCart( InputArray src1, InputArray src2,
{ {
k = 0; k = 0;
#if CV_NEON #if CV_SIMD128
for( ; k <= len - 4; k += 4 ) if( hasSIMD128() )
{ {
float32x4_t v_m = vld1q_f32(mag + k); int cWidth = v_float32x4::nlanes;
vst1q_f32(x + k, vmulq_f32(vld1q_f32(x + k), v_m)); for( ; k <= len - cWidth; k += cWidth )
vst1q_f32(y + k, vmulq_f32(vld1q_f32(y + k), v_m));
}
#elif CV_SSE2
if (USE_SSE2)
{
for( ; k <= len - 4; k += 4 )
{ {
__m128 v_m = _mm_loadu_ps(mag + k); v_float32x4 v_m = v_load(mag + k);
_mm_storeu_ps(x + k, _mm_mul_ps(_mm_loadu_ps(x + k), v_m)); v_store(x + k, v_load(x + k) * v_m);
_mm_storeu_ps(y + k, _mm_mul_ps(_mm_loadu_ps(y + k), v_m)); v_store(y + k, v_load(y + k) * v_m);
} }
} }
#endif #endif
for( ; k < len; k++ ) for( ; k < len; k++ )
{ {
@ -1599,12 +1593,9 @@ void patchNaNs( InputOutputArray _a, double _val )
Cv32suf val; Cv32suf val;
val.f = (float)_val; val.f = (float)_val;
#if CV_SSE2 #if CV_SIMD128
__m128i v_mask1 = _mm_set1_epi32(0x7fffffff), v_mask2 = _mm_set1_epi32(0x7f800000); v_int32x4 v_mask1 = v_setall_s32(0x7fffffff), v_mask2 = v_setall_s32(0x7f800000);
__m128i v_val = _mm_set1_epi32(val.i); v_int32x4 v_val = v_setall_s32(val.i);
#elif CV_NEON
int32x4_t v_mask1 = vdupq_n_s32(0x7fffffff), v_mask2 = vdupq_n_s32(0x7f800000),
v_val = vdupq_n_s32(val.i);
#endif #endif
for( size_t i = 0; i < it.nplanes; i++, ++it ) for( size_t i = 0; i < it.nplanes; i++, ++it )
@ -1612,25 +1603,18 @@ void patchNaNs( InputOutputArray _a, double _val )
int* tptr = ptrs[0]; int* tptr = ptrs[0];
size_t j = 0; size_t j = 0;
#if CV_SSE2 #if CV_SIMD128
if (USE_SSE2) if( hasSIMD128() )
{ {
for ( ; j + 4 <= len; j += 4) size_t cWidth = (size_t)v_int32x4::nlanes;
for ( ; j + cWidth <= len; j += cWidth)
{ {
__m128i v_src = _mm_loadu_si128((__m128i const *)(tptr + j)); v_int32x4 v_src = v_load(tptr + j);
__m128i v_cmp_mask = _mm_cmplt_epi32(v_mask2, _mm_and_si128(v_src, v_mask1)); v_int32x4 v_cmp_mask = v_mask2 < (v_src & v_mask1);
__m128i v_res = _mm_or_si128(_mm_andnot_si128(v_cmp_mask, v_src), _mm_and_si128(v_cmp_mask, v_val)); v_int32x4 v_dst = v_select(v_cmp_mask, v_val, v_src);
_mm_storeu_si128((__m128i *)(tptr + j), v_res); v_store(tptr + j, v_dst);
} }
} }
#elif CV_NEON
for ( ; j + 4 <= len; j += 4)
{
int32x4_t v_src = vld1q_s32(tptr + j);
uint32x4_t v_cmp_mask = vcltq_s32(v_mask2, vandq_s32(v_src, v_mask1));
int32x4_t v_dst = vbslq_s32(v_cmp_mask, v_val, v_src);
vst1q_s32(tptr + j, v_dst);
}
#endif #endif
for( ; j < len; j++ ) for( ; j < len; j++ )