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

fixes

Browse Source
This commit is contained in:
Ilya Lavrenov
2015-01-12 10:59:30 +03:00
parent 25e99c453f
commit f2cd65cf1e
7 changed files with 78 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
cmake/OpenCVCompilerOptions.cmake
View File

@@ -224,7 +224,7 @@ if(MSVC)
set(OPENCV_EXTRA_FLAGS_RELEASE "${OPENCV_EXTRA_FLAGS_RELEASE} /Zi")
endif()
if(ENABLE_AVX2 AND NOT MSVC_VERSION LESS 1600)
if(ENABLE_AVX2 AND NOT MSVC_VERSION LESS 1800)
set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:AVX2")
endif()
if(ENABLE_AVX AND NOT MSVC_VERSION LESS 1600 AND NOT OPENCV_EXTRA_FLAGS MATCHES "/arch:")
@@ -309,7 +309,7 @@ if(MSVC)
string(REPLACE "/W3" "/W4" CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG}")
if(NOT ENABLE_NOISY_WARNINGS AND MSVC_VERSION EQUAL 1400)
ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4510 /wd4610 /wd4312 /wd4201 /wd4244 /wd4328 /wd4267 /wd4324)
ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4510 /wd4610 /wd4312 /wd4201 /wd4244 /wd4328 /wd4267)
endif()
# allow extern "C" functions throw exceptions
@@ -321,6 +321,7 @@ if(MSVC)
endforeach()
if(NOT ENABLE_NOISY_WARNINGS)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd4251") #class 'std::XXX' needs to have dll-interface to be used by clients of YYY
ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4251) # class 'std::XXX' needs to have dll-interface to be used by clients of YYY
ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4324) # 'struct_name' : structure was padded due to __declspec(align())
endif()
endif()

8
modules/core/include/opencv2/core/cvdef.h
View File

@@ -159,12 +159,14 @@
# define CV_SSE4_2 1
# endif
# if defined __POPCNT__ || (defined _MSC_VER && _MSC_VER >= 1500)
# ifndef _MSC_VER
# ifdef _MSC_VER
# include <nmmintrin.h>
# else
# include <popcntintrin.h>
# endif
# define CV_POPCNT 1
# endif
# if defined __AVX__ || (defined _MSC_FULL_VER && _MSC_FULL_VER >= 160040219)
# if defined __AVX__ || (defined _MSC_VER && _MSC_VER >= 1600)
// MS Visual Studio 2010 (2012?) has no macro pre-defined to identify the use of /arch:AVX
// See: http://connect.microsoft.com/VisualStudio/feedback/details/605858/arch-avx-should-define-a-predefined-macro-in-x64-and-set-a-unique-value-for-m-ix86-fp-in-win32
# include <immintrin.h>
@@ -175,7 +177,7 @@
# define __xgetbv() 0
# endif
# endif
# if defined __AVX2__ || (defined _MSC_FULL_VER && _MSC_FULL_VER >= 160040219)
# if defined __AVX2__ || (defined _MSC_VER && _MSC_VER >= 1800)
# include <immintrin.h>
# define CV_AVX2 1
# if defined __FMA__

82
modules/core/include/opencv2/core/sse_utils.hpp
View File

@@ -43,7 +43,7 @@
#define __OPENCV_CORE_SSE_UTILS_HPP__
#ifndef __cplusplus
# error base.hpp header must be compiled as C++
# error sse_utils.hpp header must be compiled as C++
#endif
#if CV_SSE2
@@ -117,7 +117,7 @@ inline void _mm_deinterleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0
inline void _mm_deinterleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1,
__m128i & v_b0, __m128i & v_b1, __m128i & v_a0, __m128i & v_a1)
{
{
__m128i layer1_chunk0 = _mm_unpacklo_epi8(v_r0, v_b0);
__m128i layer1_chunk1 = _mm_unpackhi_epi8(v_r0, v_b0);
__m128i layer1_chunk2 = _mm_unpacklo_epi8(v_r1, v_b1);
@@ -165,9 +165,9 @@ inline void _mm_deinterleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0
}
inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1)
{
{
__m128i v_mask = _mm_set1_epi16(0x00ff);
__m128i layer4_chunk0 = _mm_packus_epi16(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
__m128i layer4_chunk2 = _mm_packus_epi16(_mm_srli_epi16(v_r0, 8), _mm_srli_epi16(v_r1, 8));
__m128i layer4_chunk1 = _mm_packus_epi16(_mm_and_si128(v_g0, v_mask), _mm_and_si128(v_g1, v_mask));
@@ -177,28 +177,28 @@ inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
__m128i layer3_chunk2 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk0, 8), _mm_srli_epi16(layer4_chunk1, 8));
__m128i layer3_chunk1 = _mm_packus_epi16(_mm_and_si128(layer4_chunk2, v_mask), _mm_and_si128(layer4_chunk3, v_mask));
__m128i layer3_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk2, 8), _mm_srli_epi16(layer4_chunk3, 8));
__m128i layer2_chunk0 = _mm_packus_epi16(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
__m128i layer2_chunk2 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk0, 8), _mm_srli_epi16(layer3_chunk1, 8));
__m128i layer2_chunk1 = _mm_packus_epi16(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
__m128i layer2_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk2, 8), _mm_srli_epi16(layer3_chunk3, 8));
__m128i layer1_chunk0 = _mm_packus_epi16(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
__m128i layer1_chunk2 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk0, 8), _mm_srli_epi16(layer2_chunk1, 8));
__m128i layer1_chunk1 = _mm_packus_epi16(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
__m128i layer1_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk2, 8), _mm_srli_epi16(layer2_chunk3, 8));
v_r0 = _mm_packus_epi16(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
v_g0 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk0, 8), _mm_srli_epi16(layer1_chunk1, 8));
v_r1 = _mm_packus_epi16(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
v_g1 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk2, 8), _mm_srli_epi16(layer1_chunk3, 8));
}
inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
__m128i & v_g1, __m128i & v_b0, __m128i & v_b1)
{
{
__m128i v_mask = _mm_set1_epi16(0x00ff);
__m128i layer4_chunk0 = _mm_packus_epi16(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
__m128i layer4_chunk3 = _mm_packus_epi16(_mm_srli_epi16(v_r0, 8), _mm_srli_epi16(v_r1, 8));
__m128i layer4_chunk1 = _mm_packus_epi16(_mm_and_si128(v_g0, v_mask), _mm_and_si128(v_g1, v_mask));
@@ -237,7 +237,7 @@ inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1,
__m128i & v_b0, __m128i & v_b1, __m128i & v_a0, __m128i & v_a1)
{
{
__m128i v_mask = _mm_set1_epi16(0x00ff);
__m128i layer4_chunk0 = _mm_packus_epi16(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
@@ -286,8 +286,8 @@ inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
v_a1 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk6, 8), _mm_srli_epi16(layer1_chunk7, 8));
}
inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1)
{
inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1)
{
__m128i layer1_chunk0 = _mm_unpacklo_epi16(v_r0, v_g0);
__m128i layer1_chunk1 = _mm_unpackhi_epi16(v_r0, v_g0);
__m128i layer1_chunk2 = _mm_unpacklo_epi16(v_r1, v_g1);
@@ -310,8 +310,8 @@ inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g
}
inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
__m128i & v_g1, __m128i & v_b0, __m128i & v_b1)
{
__m128i & v_g1, __m128i & v_b0, __m128i & v_b1)
{
__m128i layer1_chunk0 = _mm_unpacklo_epi16(v_r0, v_g1);
__m128i layer1_chunk1 = _mm_unpackhi_epi16(v_r0, v_g1);
__m128i layer1_chunk2 = _mm_unpacklo_epi16(v_r1, v_b0);
@@ -342,7 +342,7 @@ inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g
}
inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1,
__m128i & v_b0, __m128i & v_b1, __m128i & v_a0, __m128i & v_a1)
__m128i & v_b0, __m128i & v_b1, __m128i & v_a0, __m128i & v_a1)
{
__m128i layer1_chunk0 = _mm_unpacklo_epi16(v_r0, v_b0);
__m128i layer1_chunk1 = _mm_unpackhi_epi16(v_r0, v_b0);
@@ -352,7 +352,7 @@ inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g
__m128i layer1_chunk5 = _mm_unpackhi_epi16(v_g0, v_a0);
__m128i layer1_chunk6 = _mm_unpacklo_epi16(v_g1, v_a1);
__m128i layer1_chunk7 = _mm_unpackhi_epi16(v_g1, v_a1);
__m128i layer2_chunk0 = _mm_unpacklo_epi16(layer1_chunk0, layer1_chunk4);
__m128i layer2_chunk1 = _mm_unpackhi_epi16(layer1_chunk0, layer1_chunk4);
__m128i layer2_chunk2 = _mm_unpacklo_epi16(layer1_chunk1, layer1_chunk5);
@@ -393,14 +393,14 @@ inline void _mm_interleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
__m128i layer3_chunk3 = _mm_packus_epi32(_mm_srli_epi32(v_g0, 16), _mm_srli_epi32(v_g1, 16));
__m128i layer2_chunk0 = _mm_packus_epi32(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
__m128i layer2_chunk2 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
__m128i layer2_chunk2 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
__m128i layer2_chunk1 = _mm_packus_epi32(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
__m128i layer2_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
__m128i layer2_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
__m128i layer1_chunk0 = _mm_packus_epi32(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
__m128i layer1_chunk2 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
__m128i layer1_chunk2 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
__m128i layer1_chunk1 = _mm_packus_epi32(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
__m128i layer1_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
__m128i layer1_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
v_r0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
v_g0 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk0, 16), _mm_srli_epi32(layer1_chunk1, 16));
@@ -421,18 +421,18 @@ inline void _mm_interleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
__m128i layer3_chunk5 = _mm_packus_epi32(_mm_srli_epi32(v_b0, 16), _mm_srli_epi32(v_b1, 16));
__m128i layer2_chunk0 = _mm_packus_epi32(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
__m128i layer2_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
__m128i layer2_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
__m128i layer2_chunk1 = _mm_packus_epi32(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
__m128i layer2_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
__m128i layer2_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
__m128i layer2_chunk2 = _mm_packus_epi32(_mm_and_si128(layer3_chunk4, v_mask), _mm_and_si128(layer3_chunk5, v_mask));
__m128i layer2_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk4, 16), _mm_srli_epi32(layer3_chunk5, 16));
__m128i layer2_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk4, 16), _mm_srli_epi32(layer3_chunk5, 16));
__m128i layer1_chunk0 = _mm_packus_epi32(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
__m128i layer1_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
__m128i layer1_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
__m128i layer1_chunk1 = _mm_packus_epi32(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
__m128i layer1_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
__m128i layer1_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
__m128i layer1_chunk2 = _mm_packus_epi32(_mm_and_si128(layer2_chunk4, v_mask), _mm_and_si128(layer2_chunk5, v_mask));
__m128i layer1_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk4, 16), _mm_srli_epi32(layer2_chunk5, 16));
__m128i layer1_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk4, 16), _mm_srli_epi32(layer2_chunk5, 16));
v_r0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
v_g1 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk0, 16), _mm_srli_epi32(layer1_chunk1, 16));
@@ -457,26 +457,26 @@ inline void _mm_interleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
__m128i layer3_chunk7 = _mm_packus_epi32(_mm_srli_epi32(v_a0, 16), _mm_srli_epi32(v_a1, 16));
__m128i layer2_chunk0 = _mm_packus_epi32(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
__m128i layer2_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
__m128i layer2_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
__m128i layer2_chunk1 = _mm_packus_epi32(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
__m128i layer2_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
__m128i layer2_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
__m128i layer2_chunk2 = _mm_packus_epi32(_mm_and_si128(layer3_chunk4, v_mask), _mm_and_si128(layer3_chunk5, v_mask));
__m128i layer2_chunk6 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk4, 16), _mm_srli_epi32(layer3_chunk5, 16));
__m128i layer2_chunk6 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk4, 16), _mm_srli_epi32(layer3_chunk5, 16));
__m128i layer2_chunk3 = _mm_packus_epi32(_mm_and_si128(layer3_chunk6, v_mask), _mm_and_si128(layer3_chunk7, v_mask));
__m128i layer2_chunk7 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk6, 16), _mm_srli_epi32(layer3_chunk7, 16));
__m128i layer2_chunk7 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk6, 16), _mm_srli_epi32(layer3_chunk7, 16));
__m128i layer1_chunk0 = _mm_packus_epi32(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
__m128i layer1_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
__m128i layer1_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
__m128i layer1_chunk1 = _mm_packus_epi32(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
__m128i layer1_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
__m128i layer1_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
__m128i layer1_chunk2 = _mm_packus_epi32(_mm_and_si128(layer2_chunk4, v_mask), _mm_and_si128(layer2_chunk5, v_mask));
__m128i layer1_chunk6 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk4, 16), _mm_srli_epi32(layer2_chunk5, 16));
__m128i layer1_chunk6 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk4, 16), _mm_srli_epi32(layer2_chunk5, 16));
__m128i layer1_chunk3 = _mm_packus_epi32(_mm_and_si128(layer2_chunk6, v_mask), _mm_and_si128(layer2_chunk7, v_mask));
__m128i layer1_chunk7 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk6, 16), _mm_srli_epi32(layer2_chunk7, 16));
__m128i layer1_chunk7 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk6, 16), _mm_srli_epi32(layer2_chunk7, 16));
v_r0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
v_b0 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk0, 16), _mm_srli_epi32(layer1_chunk1, 16));
v_r0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
v_r1 = _mm_packus_epi32(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
v_b1 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk2, 16), _mm_srli_epi32(layer1_chunk3, 16));
v_g0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk4, v_mask), _mm_and_si128(layer1_chunk5, v_mask));
v_a0 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk4, 16), _mm_srli_epi32(layer1_chunk5, 16));
@@ -487,12 +487,12 @@ inline void _mm_interleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
#endif // CV_SSE4_1
inline void _mm_deinterleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0, __m128 & v_g1)
{
{
__m128 layer1_chunk0 = _mm_unpacklo_ps(v_r0, v_g0);
__m128 layer1_chunk1 = _mm_unpackhi_ps(v_r0, v_g0);
__m128 layer1_chunk2 = _mm_unpacklo_ps(v_r1, v_g1);
__m128 layer1_chunk3 = _mm_unpackhi_ps(v_r1, v_g1);
__m128 layer2_chunk0 = _mm_unpacklo_ps(layer1_chunk0, layer1_chunk2);
__m128 layer2_chunk1 = _mm_unpackhi_ps(layer1_chunk0, layer1_chunk2);
__m128 layer2_chunk2 = _mm_unpacklo_ps(layer1_chunk1, layer1_chunk3);
@@ -506,14 +506,14 @@ inline void _mm_deinterleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0, __m
inline void _mm_deinterleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0,
__m128 & v_g1, __m128 & v_b0, __m128 & v_b1)
{
{
__m128 layer1_chunk0 = _mm_unpacklo_ps(v_r0, v_g1);
__m128 layer1_chunk1 = _mm_unpackhi_ps(v_r0, v_g1);
__m128 layer1_chunk2 = _mm_unpacklo_ps(v_r1, v_b0);
__m128 layer1_chunk3 = _mm_unpackhi_ps(v_r1, v_b0);
__m128 layer1_chunk4 = _mm_unpacklo_ps(v_g0, v_b1);
__m128 layer1_chunk5 = _mm_unpackhi_ps(v_g0, v_b1);
__m128 layer2_chunk0 = _mm_unpacklo_ps(layer1_chunk0, layer1_chunk3);
__m128 layer2_chunk1 = _mm_unpackhi_ps(layer1_chunk0, layer1_chunk3);
__m128 layer2_chunk2 = _mm_unpacklo_ps(layer1_chunk1, layer1_chunk4);
@@ -531,7 +531,7 @@ inline void _mm_deinterleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0,
inline void _mm_deinterleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0, __m128 & v_g1,
__m128 & v_b0, __m128 & v_b1, __m128 & v_a0, __m128 & v_a1)
{
{
__m128 layer1_chunk0 = _mm_unpacklo_ps(v_r0, v_b0);
__m128 layer1_chunk1 = _mm_unpackhi_ps(v_r0, v_b0);
__m128 layer1_chunk2 = _mm_unpacklo_ps(v_r1, v_b1);

2
modules/core/src/arithm.cpp
View File

@@ -3476,7 +3476,7 @@ struct Cmp_SIMD<schar>
haveSSE = checkHardwareSupport(CV_CPU_SSE2);
v_mask = _mm_set1_epi8(0xff);
v_mask = _mm_set1_epi8(-1);
}
int operator () (const schar * src1, const schar * src2, uchar * dst, int width) const

7
modules/core/src/convert.cpp
View File

@@ -616,18 +616,17 @@ struct VMerge4<data_type>
bool support; \
}
MERGE2_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_deinterleave_epi8, si128);
MERGE2_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128);
MERGE3_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128);
MERGE4_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128);
MERGE2_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_deinterleave_epi16, si128);
#if CV_SSE4_1
MERGE2_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128);
MERGE3_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128);
MERGE4_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128);
#endif
MERGE2_KERNEL_TEMPLATE( int, __m128, float, _mm_deinterleave_ps, ps);
MERGE2_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps);
MERGE3_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps);
MERGE4_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps);

6
modules/imgproc/src/color.cpp
View File

@@ -1460,9 +1460,9 @@ struct RGB2Gray<ushort>
if( blueIdx == 0 )
std::swap(coeffs[0], coeffs[2]);
v_cb = _mm_set1_epi16(coeffs[0]);
v_cg = _mm_set1_epi16(coeffs[1]);
v_cr = _mm_set1_epi16(coeffs[2]);
v_cb = _mm_set1_epi16((short)coeffs[0]);
v_cg = _mm_set1_epi16((short)coeffs[1]);
v_cr = _mm_set1_epi16((short)coeffs[2]);
v_delta = _mm_set1_epi32(1 << (yuv_shift - 1));
}

28
modules/photo/test/test_cloning.cpp
View File

@@ -64,6 +64,7 @@ TEST(Photo_SeamlessClone_normal, regression)
string original_path1 = folder + "source1.png";
string original_path2 = folder + "destination1.png";
string original_path3 = folder + "mask.png";
string reference_path = folder + "reference.png";
Mat source = imread(original_path1, IMREAD_COLOR);
Mat destination = imread(original_path2, IMREAD_COLOR);
@@ -79,8 +80,8 @@ TEST(Photo_SeamlessClone_normal, regression)
p.y = destination.size().height/2;
seamlessClone(source, destination, mask, p, result, 1);
Mat reference = imread(folder + "reference.png");
Mat reference = imread(reference_path);
ASSERT_FALSE(reference.empty()) << "Could not load reference image " << reference_path;
SAVE(result);
@@ -94,6 +95,7 @@ TEST(Photo_SeamlessClone_mixed, regression)
string original_path1 = folder + "source1.png";
string original_path2 = folder + "destination1.png";
string original_path3 = folder + "mask.png";
string reference_path = folder + "reference.png";
Mat source = imread(original_path1, IMREAD_COLOR);
Mat destination = imread(original_path2, IMREAD_COLOR);
@@ -111,7 +113,9 @@ TEST(Photo_SeamlessClone_mixed, regression)
SAVE(result);
Mat reference = imread(folder + "reference.png");
Mat reference = imread(reference_path);
ASSERT_FALSE(reference.empty()) << "Could not load reference image " << reference_path;
double error = cvtest::norm(reference, result, NORM_L1);
EXPECT_LE(error, numerical_precision);
@@ -123,6 +127,7 @@ TEST(Photo_SeamlessClone_featureExchange, regression)
string original_path1 = folder + "source1.png";
string original_path2 = folder + "destination1.png";
string original_path3 = folder + "mask.png";
string reference_path = folder + "reference.png";
Mat source = imread(original_path1, IMREAD_COLOR);
Mat destination = imread(original_path2, IMREAD_COLOR);
@@ -140,7 +145,9 @@ TEST(Photo_SeamlessClone_featureExchange, regression)
SAVE(result);
Mat reference = imread(folder + "reference.png");
Mat reference = imread(reference_path);
ASSERT_FALSE(reference.empty()) << "Could not load reference image " << reference_path;
double error = cvtest::norm(reference, result, NORM_L1);
EXPECT_LE(error, numerical_precision);
@@ -151,6 +158,7 @@ TEST(Photo_SeamlessClone_colorChange, regression)
string folder = string(cvtest::TS::ptr()->get_data_path()) + "cloning/color_change/";
string original_path1 = folder + "source1.png";
string original_path2 = folder + "mask.png";
string reference_path = folder + "reference.png";
Mat source = imread(original_path1, IMREAD_COLOR);
Mat mask = imread(original_path2, IMREAD_COLOR);
@@ -163,7 +171,9 @@ TEST(Photo_SeamlessClone_colorChange, regression)
SAVE(result);
Mat reference = imread(folder + "reference.png");
Mat reference = imread(reference_path);
ASSERT_FALSE(reference.empty()) << "Could not load reference image " << reference_path;
double error = cvtest::norm(reference, result, NORM_L1);
EXPECT_LE(error, numerical_precision);
@@ -174,6 +184,7 @@ TEST(Photo_SeamlessClone_illuminationChange, regression)
string folder = string(cvtest::TS::ptr()->get_data_path()) + "cloning/Illumination_Change/";
string original_path1 = folder + "source1.png";
string original_path2 = folder + "mask.png";
string reference_path = folder + "reference.png";
Mat source = imread(original_path1, IMREAD_COLOR);
Mat mask = imread(original_path2, IMREAD_COLOR);
@@ -186,7 +197,7 @@ TEST(Photo_SeamlessClone_illuminationChange, regression)
SAVE(result);
Mat reference = imread(folder + "reference.png");
Mat reference = imread(reference_path);
double error = cvtest::norm(reference, result, NORM_L1);
EXPECT_LE(error, numerical_precision);
@@ -197,6 +208,7 @@ TEST(Photo_SeamlessClone_textureFlattening, regression)
string folder = string(cvtest::TS::ptr()->get_data_path()) + "cloning/Texture_Flattening/";
string original_path1 = folder + "source1.png";
string original_path2 = folder + "mask.png";
string reference_path = folder + "reference.png";
Mat source = imread(original_path1, IMREAD_COLOR);
Mat mask = imread(original_path2, IMREAD_COLOR);
@@ -209,7 +221,9 @@ TEST(Photo_SeamlessClone_textureFlattening, regression)
SAVE(result);
Mat reference = imread(folder + "reference.png");
Mat reference = imread(reference_path);
ASSERT_FALSE(reference.empty()) << "Could not load reference image " << reference_path;
double error = cvtest::norm(reference, result, NORM_L1);
EXPECT_LE(error, numerical_precision);