Commit ee3ef107 authored by Alexander Alekhin's avatar Alexander Alekhin
Browse files

core(SIMD): allow to build modules with SIMD emulator

parent 91b39aa0

modules/optflow/src/sparse_matching_gpc.cpp

+1 −0
Original line number Diff line number Diff line
@@ -41,6 +41,7 @@ 
 //M*/



#include "precomp.hpp"

#undef CV_FORCE_SIMD128_CPP  // mixed HAL SIMD/SSE code

#include "opencv2/core/core_c.h"

#include "opencv2/core/private.hpp"

#include "opencv2/flann/miniflann.hpp"

modules/xphoto/src/learning_based_color_balance.cpp

+18 −14
Original line number Diff line number Diff line
@@ -262,26 +262,26 @@ void LearningBasedWBImpl::getAverageAndBrightestColorChromaticity(Vec2f &average 
        uint sumB = 0, sumG = 0, sumR = 0;

        uchar *src_ptr = src.ptr<uchar>();

#if CV_SIMD128

        v_uint8x16 v_inB, v_inG, v_inR, v_mask;

        v_uint16x8 v_sR1, v_sR2, v_sG1, v_sG2, v_sB1, v_sB2, v_sum;

        v_uint16x8 v_max_sum = v_setall_u16(0), v_max_mask, v_brightestR, v_brightestG, v_brightestB;

        v_uint32x4 v_uint1, v_uint2, v_SB = v_setzero_u32(), v_SG = v_setzero_u32(), v_SR = v_setzero_u32();

        v_uint16x8 v_max_sum = v_setall_u16(0), v_brightestR = v_setall_u16(0), v_brightestG = v_setall_u16(0), v_brightestB = v_setall_u16(0);

        v_uint32x4 v_SB = v_setzero_u32(), v_SG = v_setzero_u32(), v_SR = v_setzero_u32();

        for (; i < src_len - 15; i += 16)

        {

            v_uint8x16 v_inB, v_inG, v_inR;

            v_load_deinterleave(src_ptr + 3 * i, v_inB, v_inG, v_inR);

            v_mask = v_load(mask_ptr + i);

            v_uint8x16 v_mask = v_load(mask_ptr + i);



            v_inB &= v_mask;

            v_inG &= v_mask;

            v_inR &= v_mask;



            v_uint16x8 v_sR1, v_sR2, v_sG1, v_sG2, v_sB1, v_sB2;

            v_expand(v_inB, v_sB1, v_sB2);

            v_expand(v_inG, v_sG1, v_sG2);

            v_expand(v_inR, v_sR1, v_sR2);



            // update the brightest (R,G,B) tuple (process left half):

            v_sum = v_sB1 + v_sG1 + v_sR1;

            v_max_mask = (v_sum > v_max_sum);

            v_uint16x8 v_sum = v_sB1 + v_sG1 + v_sR1;

            v_uint16x8 v_max_mask = (v_sum > v_max_sum);

            v_max_sum = v_max(v_sum, v_max_sum);

            v_brightestB = (v_sB1 & v_max_mask) + (v_brightestB & (~v_max_mask));

            v_brightestG = (v_sG1 & v_max_mask) + (v_brightestG & (~v_max_mask));
@@ -299,6 +299,8 @@ void LearningBasedWBImpl::getAverageAndBrightestColorChromaticity(Vec2f &average 
            v_sB1 = v_sB1 + v_sB2;

            v_sG1 = v_sG1 + v_sG2;

            v_sR1 = v_sR1 + v_sR2;



            v_uint32x4 v_uint1, v_uint2;

            v_expand(v_sB1, v_uint1, v_uint2);

            v_SB += v_uint1 + v_uint2;

            v_expand(v_sG1, v_uint1, v_uint2);
@@ -351,27 +353,28 @@ void LearningBasedWBImpl::getAverageAndBrightestColorChromaticity(Vec2f &average 
        uint64 sumB = 0, sumG = 0, sumR = 0;

        ushort *src_ptr = src.ptr<ushort>();

#if CV_SIMD128

        v_uint16x8 v_inB, v_inG, v_inR, v_mask, v_mask_lower = v_setall_u16(255);

        v_uint32x4 v_iR1, v_iR2, v_iG1, v_iG2, v_iB1, v_iB2, v_sum;

        v_uint32x4 v_max_sum = v_setall_u32(0), v_max_mask, v_brightestR, v_brightestG, v_brightestB;

        v_uint64x2 v_uint64_1, v_uint64_2, v_SB = v_setzero_u64(), v_SG = v_setzero_u64(), v_SR = v_setzero_u64();

        const v_uint16x8 v_mask_lower = v_setall_u16(255);

        v_uint32x4 v_max_sum = v_setall_u32(0), v_brightestR = v_setall_u32(0), v_brightestG = v_setall_u32(0), v_brightestB = v_setall_u32(0);

        v_uint64x2 v_SB = v_setzero_u64(), v_SG = v_setzero_u64(), v_SR = v_setzero_u64();

        for (; i < src_len - 7; i += 8)

        {

            v_uint16x8 v_inB, v_inG, v_inR;

            v_load_deinterleave(src_ptr + 3 * i, v_inB, v_inG, v_inR);

            v_mask = v_load_expand(mask_ptr + i);

            v_uint16x8 v_mask = v_load_expand(mask_ptr + i);

            v_mask = v_mask | ((v_mask & v_mask_lower) << 8);



            v_inB &= v_mask;

            v_inG &= v_mask;

            v_inR &= v_mask;



            v_uint32x4 v_iR1, v_iR2, v_iG1, v_iG2, v_iB1, v_iB2;

            v_expand(v_inB, v_iB1, v_iB2);

            v_expand(v_inG, v_iG1, v_iG2);

            v_expand(v_inR, v_iR1, v_iR2);



            // update the brightest (R,G,B) tuple (process left half):

            v_sum = v_iB1 + v_iG1 + v_iR1;

            v_max_mask = (v_sum > v_max_sum);

            v_uint32x4 v_sum = v_iB1 + v_iG1 + v_iR1;

            v_uint32x4 v_max_mask = (v_sum > v_max_sum);

            v_max_sum = v_max(v_sum, v_max_sum);

            v_brightestB = (v_iB1 & v_max_mask) + (v_brightestB & (~v_max_mask));

            v_brightestG = (v_iG1 & v_max_mask) + (v_brightestG & (~v_max_mask));
@@ -389,6 +392,7 @@ void LearningBasedWBImpl::getAverageAndBrightestColorChromaticity(Vec2f &average 
            v_iB1 = v_iB1 + v_iB2;

            v_iG1 = v_iG1 + v_iG2;

            v_iR1 = v_iR1 + v_iR2;

            v_uint64x2 v_uint64_1, v_uint64_2;

            v_expand(v_iB1, v_uint64_1, v_uint64_2);

            v_SB += v_uint64_1 + v_uint64_2;

            v_expand(v_iG1, v_uint64_1, v_uint64_2);