From 3e3e2dd51254da177e83e53b1afe3e09a5e821e7 Mon Sep 17 00:00:00 2001 From: Alexander Alekhin Date: Mon, 12 Jun 2017 16:10:13 +0000 Subject: [PATCH] android: make optional "cpufeatures", build fixes for NDK r15 --- 3rdparty/carotene/hal/CMakeLists.txt | 2 +- 3rdparty/carotene/src/channel_extract.cpp | 8 +-- 3rdparty/carotene/src/channels_combine.cpp | 4 +- 3rdparty/carotene/src/colorconvert.cpp | 78 +++++++++++----------- 3rdparty/carotene/src/convert.cpp | 54 +++++++-------- 3rdparty/carotene/src/convert_scale.cpp | 56 ++++++++-------- 3rdparty/carotene/src/gaussian_blur.cpp | 2 +- 3rdparty/carotene/src/opticalflow.cpp | 2 +- 3rdparty/carotene/src/pyramid.cpp | 8 +-- 3rdparty/carotene/src/scharr.cpp | 6 +- 3rdparty/libpng/CMakeLists.txt | 12 ++-- 3rdparty/libwebp/CMakeLists.txt | 8 ++- CMakeLists.txt | 8 ++- cmake/OpenCVCompilerOptimizations.cmake | 25 ++++--- cmake/OpenCVFindLibsGrfmt.cmake | 8 ++- cmake/OpenCVUtils.cmake | 10 +++ cmake/checks/cpu_neon.cpp | 27 ++++++++ modules/core/CMakeLists.txt | 6 +- .../core/include/opencv2/core/cv_cpu_dispatch.h | 6 +- .../core/include/opencv2/core/hal/intrin_neon.hpp | 28 +++++--- modules/core/src/convert.fp16.cpp | 9 +-- modules/core/src/convert.hpp | 8 +-- modules/core/src/system.cpp | 18 ++++- modules/imgcodecs/CMakeLists.txt | 2 +- 24 files changed, 240 insertions(+), 155 deletions(-) create mode 100644 cmake/checks/cpu_neon.cpp diff --git a/3rdparty/carotene/hal/CMakeLists.txt b/3rdparty/carotene/hal/CMakeLists.txt index f2ca5ff..eabde95 100644 --- a/3rdparty/carotene/hal/CMakeLists.txt +++ b/3rdparty/carotene/hal/CMakeLists.txt @@ -67,7 +67,7 @@ function(compile_carotene) endif() check_cxx_compiler_flag("-mfpu=neon" CXX_HAS_MFPU_NEON) check_c_compiler_flag("-mfpu=neon" C_HAS_MFPU_NEON) - if(${CXX_HAS_MFPU_NEON} AND ${C_HAS_MFPU_NEON}) + if(${CXX_HAS_MFPU_NEON} AND ${C_HAS_MFPU_NEON} AND NOT "${CMAKE_CXX_FLAGS} " MATCHES "-mfpu=neon[^ ]*") get_target_property(old_flags "carotene_objs" COMPILE_FLAGS) if(old_flags) set_target_properties("carotene_objs" PROPERTIES COMPILE_FLAGS "${old_flags} -mfpu=neon") diff --git a/3rdparty/carotene/src/channel_extract.cpp b/3rdparty/carotene/src/channel_extract.cpp index fda8f6e..109f74f 100644 --- a/3rdparty/carotene/src/channel_extract.cpp +++ b/3rdparty/carotene/src/channel_extract.cpp @@ -231,7 +231,7 @@ void extract4(const Size2D &size, srcStride == dst2Stride && \ srcStride == dst3Stride && -#if __GNUC__ == 4 && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 #define SPLIT_ASM2(sgn, bits) __asm__ ( \ "vld2." #bits " {d0, d2}, [%[in0]] \n\t" \ @@ -280,7 +280,7 @@ void extract4(const Size2D &size, FILL_LINES##n(VST1Q, sgn##bits) \ } -#endif // __GNUC__ == 4 && __GNUC_MINOR__ < 7 +#endif #define SPLIT(sgn,bits,n) void split##n(const Size2D &_size, \ const sgn##bits * srcBase, ptrdiff_t srcStride \ @@ -351,7 +351,7 @@ void extract4(const Size2D &size, } \ } -#if __GNUC__ == 4 && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 #define ALPHA_QUAD(sgn, bits) { \ internal::prefetch(src + sj); \ @@ -378,7 +378,7 @@ void extract4(const Size2D &size, vst1q_##sgn##bits(dst1 + d1j, vals.v4.val[3]); \ } -#endif // __GNUC__ == 4 && __GNUC_MINOR__ < 7 +#endif #define SPLIT4ALPHA(sgn,bits) void split4(const Size2D &_size, \ const sgn##bits * srcBase, ptrdiff_t srcStride, \ diff --git a/3rdparty/carotene/src/channels_combine.cpp b/3rdparty/carotene/src/channels_combine.cpp index 32b7147..9f0533c 100644 --- a/3rdparty/carotene/src/channels_combine.cpp +++ b/3rdparty/carotene/src/channels_combine.cpp @@ -77,7 +77,7 @@ namespace CAROTENE_NS { dstStride == src2Stride && \ dstStride == src3Stride && -#if __GNUC__ == 4 && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 #define MERGE_ASM2(sgn, bits) __asm__ ( \ "vld1." #bits " {d0-d1}, [%[in0]] \n\t" \ @@ -128,7 +128,7 @@ namespace CAROTENE_NS { vst##n##q_##sgn##bits(dst + dj, v_dst); \ } -#endif // __GNUC__ == 4 && __GNUC_MINOR__ < 7 +#endif #define COMBINE(sgn,bits,n) void combine##n(const Size2D &_size \ FILL_LINES##n(FARG, sgn##bits), \ diff --git a/3rdparty/carotene/src/colorconvert.cpp b/3rdparty/carotene/src/colorconvert.cpp index ea2db60..3ad1ada 100644 --- a/3rdparty/carotene/src/colorconvert.cpp +++ b/3rdparty/carotene/src/colorconvert.cpp @@ -97,7 +97,7 @@ void rgb2gray(const Size2D &size, COLOR_SPACE color_space, const u32 G2Y = color_space == COLOR_SPACE_BT601 ? G2Y_BT601 : G2Y_BT709; const u32 B2Y = color_space == COLOR_SPACE_BT601 ? B2Y_BT601 : B2Y_BT709; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register int16x4_t v_r2y asm ("d31") = vmov_n_s16(R2Y); register int16x4_t v_g2y asm ("d30") = vmov_n_s16(G2Y); register int16x4_t v_b2y asm ("d29") = vmov_n_s16(B2Y); @@ -116,7 +116,7 @@ void rgb2gray(const Size2D &size, COLOR_SPACE color_space, u8 * dst = internal::getRowPtr(dstBase, dstStride, i); size_t sj = 0u, dj = 0u; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 for (; dj < roiw8; sj += 24, dj += 8) { internal::prefetch(src + sj); @@ -198,7 +198,7 @@ void rgbx2gray(const Size2D &size, COLOR_SPACE color_space, const u32 G2Y = color_space == COLOR_SPACE_BT601 ? G2Y_BT601 : G2Y_BT709; const u32 B2Y = color_space == COLOR_SPACE_BT601 ? B2Y_BT601 : B2Y_BT709; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register int16x4_t v_r2y asm ("d31") = vmov_n_s16(R2Y); register int16x4_t v_g2y asm ("d30") = vmov_n_s16(G2Y); register int16x4_t v_b2y asm ("d29") = vmov_n_s16(B2Y); @@ -217,7 +217,7 @@ void rgbx2gray(const Size2D &size, COLOR_SPACE color_space, u8 * dst = internal::getRowPtr(dstBase, dstStride, i); size_t sj = 0u, dj = 0u; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 for (; dj < roiw8; sj += 32, dj += 8) { internal::prefetch(src + sj); @@ -300,7 +300,7 @@ void bgr2gray(const Size2D &size, COLOR_SPACE color_space, const u32 G2Y = color_space == COLOR_SPACE_BT601 ? G2Y_BT601 : G2Y_BT709; const u32 B2Y = color_space == COLOR_SPACE_BT601 ? B2Y_BT601 : B2Y_BT709; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register int16x4_t v_r2y asm ("d31") = vmov_n_s16(R2Y); register int16x4_t v_g2y asm ("d30") = vmov_n_s16(G2Y); register int16x4_t v_b2y asm ("d29") = vmov_n_s16(B2Y); @@ -319,7 +319,7 @@ void bgr2gray(const Size2D &size, COLOR_SPACE color_space, u8 * dst = internal::getRowPtr(dstBase, dstStride, i); size_t sj = 0u, dj = 0u; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 for (; dj < roiw8; sj += 24, dj += 8) { internal::prefetch(src + sj); @@ -402,7 +402,7 @@ void bgrx2gray(const Size2D &size, COLOR_SPACE color_space, const u32 G2Y = color_space == COLOR_SPACE_BT601 ? G2Y_BT601 : G2Y_BT709; const u32 B2Y = color_space == COLOR_SPACE_BT601 ? B2Y_BT601 : B2Y_BT709; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register int16x4_t v_r2y asm ("d31") = vmov_n_s16(R2Y); register int16x4_t v_g2y asm ("d30") = vmov_n_s16(G2Y); register int16x4_t v_b2y asm ("d29") = vmov_n_s16(B2Y); @@ -421,7 +421,7 @@ void bgrx2gray(const Size2D &size, COLOR_SPACE color_space, u8 * dst = internal::getRowPtr(dstBase, dstStride, i); size_t sj = 0u, dj = 0u; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 for (; dj < roiw8; sj += 32, dj += 8) { internal::prefetch(src + sj); @@ -512,7 +512,7 @@ void gray2rgb(const Size2D &size, for (; sj < roiw16; sj += 16, dj += 48) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld1.8 {d0-d1}, [%[in0]] \n\t" "vmov.8 q1, q0 \n\t" @@ -538,7 +538,7 @@ void gray2rgb(const Size2D &size, if (sj < roiw8) { -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld1.8 {d0}, [%[in]] \n\t" "vmov.8 d1, d0 \n\t" @@ -584,7 +584,7 @@ void gray2rgbx(const Size2D &size, size_t roiw16 = size.width >= 15 ? size.width - 15 : 0; size_t roiw8 = size.width >= 7 ? size.width - 7 : 0; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register uint8x16_t vc255 asm ("q4") = vmovq_n_u8(255); #else uint8x16x4_t vRgba; @@ -602,7 +602,7 @@ void gray2rgbx(const Size2D &size, for (; sj < roiw16; sj += 16, dj += 64) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld1.8 {d0-d1}, [%[in0]] \n\t" "vmov.8 q1, q0 \n\t" @@ -628,7 +628,7 @@ void gray2rgbx(const Size2D &size, if (sj < roiw8) { -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld1.8 {d5}, [%[in]] \n\t" "vmov.8 d6, d5 \n\t" @@ -1409,7 +1409,7 @@ inline void convertToHSV(const s32 r, const s32 g, const s32 b, "d24","d25","d26","d27","d28","d29","d30","d31" \ ); -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 #define YCRCB_CONSTS \ register int16x4_t vcYR asm ("d31") = vmov_n_s16(4899); \ @@ -1555,7 +1555,7 @@ inline uint8x8x3_t convertToYCrCb( const int16x8_t& vR, const int16x8_t& vG, con #define COEFF_G ( 8663) #define COEFF_B (-17705) -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 #define YUV420ALPHA3_CONST #define YUV420ALPHA4_CONST register uint8x16_t c255 asm ("q13") = vmovq_n_u8(255); #define YUV420ALPHA3_CONVERT @@ -1852,7 +1852,7 @@ void rgb2hsv(const Size2D &size, #ifdef CAROTENE_NEON size_t roiw8 = size.width >= 7 ? size.width - 7 : 0; const s32 hsv_shift = 12; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register const f32 vsdiv_table = f32(255 << hsv_shift); register f32 vhdiv_table = f32(hrange << hsv_shift); register const s32 vhrange = hrange; @@ -1871,7 +1871,7 @@ void rgb2hsv(const Size2D &size, for (; j < roiw8; sj += 24, dj += 24, j += 8) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERT_TO_HSV_ASM(vld3.8 {d0-d2}, d0, d2) #else uint8x8x3_t vRgb = vld3_u8(src + sj); @@ -1904,7 +1904,7 @@ void rgbx2hsv(const Size2D &size, #ifdef CAROTENE_NEON size_t roiw8 = size.width >= 7 ? size.width - 7 : 0; const s32 hsv_shift = 12; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register const f32 vsdiv_table = f32(255 << hsv_shift); register f32 vhdiv_table = f32(hrange << hsv_shift); register const s32 vhrange = hrange; @@ -1923,7 +1923,7 @@ void rgbx2hsv(const Size2D &size, for (; j < roiw8; sj += 32, dj += 24, j += 8) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERT_TO_HSV_ASM(vld4.8 {d0-d3}, d0, d2) #else uint8x8x4_t vRgb = vld4_u8(src + sj); @@ -1956,7 +1956,7 @@ void bgr2hsv(const Size2D &size, #ifdef CAROTENE_NEON size_t roiw8 = size.width >= 7 ? size.width - 7 : 0; const s32 hsv_shift = 12; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register const f32 vsdiv_table = f32(255 << hsv_shift); register f32 vhdiv_table = f32(hrange << hsv_shift); register const s32 vhrange = hrange; @@ -1975,7 +1975,7 @@ void bgr2hsv(const Size2D &size, for (; j < roiw8; sj += 24, dj += 24, j += 8) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERT_TO_HSV_ASM(vld3.8 {d0-d2}, d2, d0) #else uint8x8x3_t vRgb = vld3_u8(src + sj); @@ -2008,7 +2008,7 @@ void bgrx2hsv(const Size2D &size, #ifdef CAROTENE_NEON size_t roiw8 = size.width >= 7 ? size.width - 7 : 0; const s32 hsv_shift = 12; -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register const f32 vsdiv_table = f32(255 << hsv_shift); register f32 vhdiv_table = f32(hrange << hsv_shift); register const s32 vhrange = hrange; @@ -2027,7 +2027,7 @@ void bgrx2hsv(const Size2D &size, for (; j < roiw8; sj += 32, dj += 24, j += 8) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERT_TO_HSV_ASM(vld4.8 {d0-d3}, d2, d0) #else uint8x8x4_t vRgb = vld4_u8(src + sj); @@ -2068,7 +2068,7 @@ void rgbx2bgr565(const Size2D &size, for (; j < roiw16; sj += 64, dj += 32, j += 16) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld4.8 {d2, d4, d6, d8}, [%[in0]] @ q0 q1 q2 q3 q4 \n\t" "vld4.8 {d3, d5, d7, d9}, [%[in1]] @ xxxxxxxx rrrrRRRR ggggGGGG bbbbBBBB xxxxxxxx \n\t" @@ -2122,7 +2122,7 @@ void rgb2bgr565(const Size2D &size, for (; j < roiw16; sj += 48, dj += 32, j += 16) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld3.8 {d2, d4, d6}, [%[in0]] @ q0 q1 q2 q3 q4 \n\t" "vld3.8 {d3, d5, d7}, [%[in1]] @ xxxxxxxx rrrrRRRR ggggGGGG bbbbBBBB xxxxxxxx \n\t" @@ -2176,7 +2176,7 @@ void rgbx2rgb565(const Size2D &size, for (; j < roiw16; sj += 64, dj += 32, j += 16) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld4.8 {d0, d2, d4, d6}, [%[in0]] @ q0 q1 q2 q3 \n\t" "vld4.8 {d1, d3, d5, d7}, [%[in1]] @ rrrrRRRR ggggGGGG bbbbBBBB aaaaAAAA \n\t" @@ -2230,7 +2230,7 @@ void rgb2rgb565(const Size2D &size, for (; j < roiw16; sj += 48, dj += 32, j += 16) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld3.8 {d0, d2, d4}, [%[in0]] @ q0 q1 q2 q3 \n\t" "vld3.8 {d1, d3, d5}, [%[in1]] @ rrrrRRRR ggggGGGG bbbbBBBB xxxxxxxx \n\t" @@ -2285,7 +2285,7 @@ void rgb2ycrcb(const Size2D &size, for (; j < roiw8; sj += 24, dj += 24, j += 8) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTTOYCRCB(vld3.8 {d0-d2}, d0, d1, d2) #else uint8x8x3_t vRgb = vld3_u8(src + sj); @@ -2329,7 +2329,7 @@ void rgbx2ycrcb(const Size2D &size, for (; j < roiw8; sj += 32, dj += 24, j += 8) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTTOYCRCB(vld4.8 {d0-d3}, d0, d1, d2) #else uint8x8x4_t vRgba = vld4_u8(src + sj); @@ -2373,7 +2373,7 @@ void bgr2ycrcb(const Size2D &size, for (; j < roiw8; sj += 24, dj += 24, j += 8) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTTOYCRCB(vld3.8 {d0-d2}, d2, d1, d0) #else uint8x8x3_t vBgr = vld3_u8(src + sj); @@ -2417,7 +2417,7 @@ void bgrx2ycrcb(const Size2D &size, for (; j < roiw8; sj += 32, dj += 24, j += 8) { internal::prefetch(src + sj); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTTOYCRCB(vld4.8 {d0-d3}, d2, d1, d0) #else uint8x8x4_t vBgra = vld4_u8(src + sj); @@ -2499,7 +2499,7 @@ void yuv420sp2rgb(const Size2D &size, internal::prefetch(uv + j); internal::prefetch(y1 + j); internal::prefetch(y2 + j); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTYUV420TORGB(3, d1, d0, q5, q6) #else convertYUV420.ToRGB(y1 + j, y2 + j, uv + j, dst1 + dj, dst2 + dj); @@ -2545,7 +2545,7 @@ void yuv420sp2rgbx(const Size2D &size, internal::prefetch(uv + j); internal::prefetch(y1 + j); internal::prefetch(y2 + j); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTYUV420TORGB(4, d1, d0, q5, q6) #else convertYUV420.ToRGB(y1 + j, y2 + j, uv + j, dst1 + dj, dst2 + dj); @@ -2591,7 +2591,7 @@ void yuv420i2rgb(const Size2D &size, internal::prefetch(uv + j); internal::prefetch(y1 + j); internal::prefetch(y2 + j); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTYUV420TORGB(3, d0, d1, q5, q6) #else convertYUV420.ToRGB(y1 + j, y2 + j, uv + j, dst1 + dj, dst2 + dj); @@ -2637,7 +2637,7 @@ void yuv420i2rgbx(const Size2D &size, internal::prefetch(uv + j); internal::prefetch(y1 + j); internal::prefetch(y2 + j); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTYUV420TORGB(4, d0, d1, q5, q6) #else convertYUV420.ToRGB(y1 + j, y2 + j, uv + j, dst1 + dj, dst2 + dj); @@ -2683,7 +2683,7 @@ void yuv420sp2bgr(const Size2D &size, internal::prefetch(uv + j); internal::prefetch(y1 + j); internal::prefetch(y2 + j); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTYUV420TORGB(3, d1, d0, q6, q5) #else convertYUV420.ToRGB(y1 + j, y2 + j, uv + j, dst1 + dj, dst2 + dj); @@ -2729,7 +2729,7 @@ void yuv420sp2bgrx(const Size2D &size, internal::prefetch(uv + j); internal::prefetch(y1 + j); internal::prefetch(y2 + j); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTYUV420TORGB(4, d1, d0, q6, q5) #else convertYUV420.ToRGB(y1 + j, y2 + j, uv + j, dst1 + dj, dst2 + dj); @@ -2775,7 +2775,7 @@ void yuv420i2bgr(const Size2D &size, internal::prefetch(uv + j); internal::prefetch(y1 + j); internal::prefetch(y2 + j); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTYUV420TORGB(3, d0, d1, q6, q5) #else convertYUV420.ToRGB(y1 + j, y2 + j, uv + j, dst1 + dj, dst2 + dj); @@ -2821,7 +2821,7 @@ void yuv420i2bgrx(const Size2D &size, internal::prefetch(uv + j); internal::prefetch(y1 + j); internal::prefetch(y2 + j); -#if defined(__GNUC__) && __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CONVERTYUV420TORGB(4, d0, d1, q6, q5) #else convertYUV420.ToRGB(y1 + j, y2 + j, uv + j, dst1 + dj, dst2 + dj); diff --git a/3rdparty/carotene/src/convert.cpp b/3rdparty/carotene/src/convert.cpp index 2f95e29..403f16d 100644 --- a/3rdparty/carotene/src/convert.cpp +++ b/3rdparty/carotene/src/convert.cpp @@ -101,7 +101,7 @@ CVT_FUNC(u8, s8, 16, } }) -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVT_FUNC(u8, u16, 16, register uint8x16_t zero0 asm ("q1") = vmovq_n_u8(0);, { @@ -135,7 +135,7 @@ CVT_FUNC(u8, u16, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVT_FUNC(u8, s32, 16, register uint8x16_t zero0 asm ("q1") = vmovq_n_u8(0); register uint8x16_t zero1 asm ("q2") = vmovq_n_u8(0); @@ -173,7 +173,7 @@ CVT_FUNC(u8, s32, 16, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(u8, f32, 16, , { @@ -248,7 +248,7 @@ CVT_FUNC(s8, u8, 16, } }) -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVT_FUNC(s8, u16, 16, register uint8x16_t zero0 asm ("q1") = vmovq_n_u8(0);, { @@ -284,7 +284,7 @@ CVT_FUNC(s8, u16, 16, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s8, s16, 16, , { @@ -323,7 +323,7 @@ CVT_FUNC(s8, s16, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVT_FUNC(s8, s32, 16, , { @@ -377,7 +377,7 @@ CVT_FUNC(s8, s32, 16, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s8, f32, 16, , { @@ -440,7 +440,7 @@ CVT_FUNC(s8, f32, 16, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(u16, u8, 16, , { @@ -479,7 +479,7 @@ CVT_FUNC(u16, u8, 16, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(u16, s8, 16, register uint8x16_t v127 asm ("q4") = vmovq_n_u8(127);, { @@ -522,7 +522,7 @@ CVT_FUNC(u16, s8, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVT_FUNC(u16, s16, 8, register uint16x8_t v32767 asm ("q4") = vmovq_n_u16(0x7FFF);, { @@ -555,7 +555,7 @@ CVT_FUNC(u16, s16, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVT_FUNC(u16, s32, 8, register uint16x8_t zero0 asm ("q1") = vmovq_n_u16(0);, { @@ -589,7 +589,7 @@ CVT_FUNC(u16, s32, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(u16, f32, 8, , { @@ -633,7 +633,7 @@ CVT_FUNC(u16, f32, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s16, u8, 16, , { @@ -672,7 +672,7 @@ CVT_FUNC(s16, u8, 16, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s16, s8, 16, , { @@ -711,7 +711,7 @@ CVT_FUNC(s16, s8, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVT_FUNC(s16, u16, 8, register int16x8_t vZero asm ("q4") = vmovq_n_s16(0);, { @@ -747,7 +747,7 @@ CVT_FUNC(s16, u16, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s16, s32, 8, , { @@ -786,7 +786,7 @@ CVT_FUNC(s16, s32, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s16, f32, 8, , { @@ -829,7 +829,7 @@ CVT_FUNC(s16, f32, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s32, u8, 8, , { @@ -870,7 +870,7 @@ CVT_FUNC(s32, u8, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s32, s8, 8, , { @@ -911,7 +911,7 @@ CVT_FUNC(s32, s8, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s32, u16, 8, , { @@ -950,7 +950,7 @@ CVT_FUNC(s32, u16, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s32, s16, 8, , { @@ -989,7 +989,7 @@ CVT_FUNC(s32, s16, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(s32, f32, 8, , { @@ -1034,7 +1034,7 @@ CVT_FUNC(s32, f32, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(f32, u8, 8, register float32x4_t vmult asm ("q0") = vdupq_n_f32((float)(1 << 16)); register uint32x4_t vmask asm ("q1") = vdupq_n_u32(1<<16);, @@ -1101,7 +1101,7 @@ CVT_FUNC(f32, u8, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(f32, s8, 8, register float32x4_t vhalf asm ("q0") = vdupq_n_f32(0.5f);, { @@ -1153,7 +1153,7 @@ CVT_FUNC(f32, s8, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(f32, u16, 8, register float32x4_t vhalf asm ("q0") = vdupq_n_f32(0.5f);, { @@ -1212,7 +1212,7 @@ CVT_FUNC(f32, u16, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(f32, s16, 8, register float32x4_t vhalf asm ("q0") = vdupq_n_f32(0.5f);, { @@ -1271,7 +1271,7 @@ CVT_FUNC(f32, s16, 8, }) #endif -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 CVT_FUNC(f32, s32, 8, register float32x4_t vhalf asm ("q0") = vdupq_n_f32(0.5f);, { diff --git a/3rdparty/carotene/src/convert_scale.cpp b/3rdparty/carotene/src/convert_scale.cpp index 50c110b..0a14a80 100644 --- a/3rdparty/carotene/src/convert_scale.cpp +++ b/3rdparty/carotene/src/convert_scale.cpp @@ -473,7 +473,7 @@ CVTS_FUNC(u8, s16, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(u8, s32, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -562,7 +562,7 @@ CVTS_FUNC(u8, s32, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(u8, f32, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta);, @@ -985,7 +985,7 @@ CVTS_FUNC(s8, s16, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s8, s32, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1074,7 +1074,7 @@ CVTS_FUNC(s8, s32, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s8, f32, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta);, @@ -1155,7 +1155,7 @@ CVTS_FUNC(s8, f32, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(u16, u8, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1214,7 +1214,7 @@ CVTS_FUNC(u16, u8, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(u16, s8, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1273,7 +1273,7 @@ CVTS_FUNC(u16, s8, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC1(u16, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1330,7 +1330,7 @@ CVTS_FUNC1(u16, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(u16, s16, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1387,7 +1387,7 @@ CVTS_FUNC(u16, s16, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(u16, s32, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1443,7 +1443,7 @@ CVTS_FUNC(u16, s32, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(u16, f32, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta);, @@ -1495,7 +1495,7 @@ CVTS_FUNC(u16, f32, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s16, u8, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1554,7 +1554,7 @@ CVTS_FUNC(s16, u8, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s16, s8, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1613,7 +1613,7 @@ CVTS_FUNC(s16, s8, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s16, u16, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1670,7 +1670,7 @@ CVTS_FUNC(s16, u16, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC1(s16, 16, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1727,7 +1727,7 @@ CVTS_FUNC1(s16, 16, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s16, s32, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1783,7 +1783,7 @@ CVTS_FUNC(s16, s32, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s16, f32, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta);, @@ -1835,7 +1835,7 @@ CVTS_FUNC(s16, f32, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s32, u8, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1893,7 +1893,7 @@ CVTS_FUNC(s32, u8, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s32, s8, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -1951,7 +1951,7 @@ CVTS_FUNC(s32, s8, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s32, u16, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -2007,7 +2007,7 @@ CVTS_FUNC(s32, u16, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s32, s16, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -2063,7 +2063,7 @@ CVTS_FUNC(s32, s16, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC1(s32, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -2118,7 +2118,7 @@ CVTS_FUNC1(s32, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(s32, f32, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta);, @@ -2169,7 +2169,7 @@ CVTS_FUNC(s32, f32, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(f32, u8, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)((1 << 16)*alpha)); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)((1 << 16)*beta)); @@ -2239,7 +2239,7 @@ CVTS_FUNC(f32, u8, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(f32, s8, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -2293,7 +2293,7 @@ CVTS_FUNC(f32, s8, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(f32, u16, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -2345,7 +2345,7 @@ CVTS_FUNC(f32, u16, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(f32, s16, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -2397,7 +2397,7 @@ CVTS_FUNC(f32, s16, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC(f32, s32, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta + 0.5f);, @@ -2448,7 +2448,7 @@ CVTS_FUNC(f32, s32, 8, }) #endif -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 CVTS_FUNC1(f32, 8, register float32x4_t vscale asm ("q0") = vdupq_n_f32((f32)alpha); register float32x4_t vshift asm ("q1") = vdupq_n_f32((f32)beta);, diff --git a/3rdparty/carotene/src/gaussian_blur.cpp b/3rdparty/carotene/src/gaussian_blur.cpp index 069373e..1b53994 100644 --- a/3rdparty/carotene/src/gaussian_blur.cpp +++ b/3rdparty/carotene/src/gaussian_blur.cpp @@ -327,7 +327,7 @@ void gaussianBlur5x5(const Size2D &size, s32 cn, u16* lidx1 = lane + x - 1*2; u16* lidx3 = lane + x + 1*2; u16* lidx4 = lane + x + 2*2; -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ __volatile__ ( "vld2.16 {d0, d2}, [%[in0]]! \n\t" "vld2.16 {d1, d3}, [%[in0]] \n\t" diff --git a/3rdparty/carotene/src/opticalflow.cpp b/3rdparty/carotene/src/opticalflow.cpp index fa9402a..7b29742 100644 --- a/3rdparty/carotene/src/opticalflow.cpp +++ b/3rdparty/carotene/src/opticalflow.cpp @@ -210,7 +210,7 @@ void pyrLKOptFlowLevel(const Size2D &size, s32 cn, internal::prefetch(dsrc + dstep * 2, 0); for(x = 0; x <= wwcn - 4; x += 4, dsrc += 4*2, dIptr += 4*2 ) { -#if __GNUC_MINOR__ < 0 +#if 0 __asm__ ( "vld2.16 {d0-d1}, [%[dsrc00]] \n\t" "vld2.16 {d2-d3}, [%[dsrc10]] \n\t" diff --git a/3rdparty/carotene/src/pyramid.cpp b/3rdparty/carotene/src/pyramid.cpp index 546ccec..8ef1268 100644 --- a/3rdparty/carotene/src/pyramid.cpp +++ b/3rdparty/carotene/src/pyramid.cpp @@ -331,7 +331,7 @@ void gaussianPyramidDown(const Size2D &srcSize, for (; x < roiw8; x += 8) { internal::prefetch(lane + 2 * x); -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld2.16 {d0-d3}, [%[in0]] \n\t" "vld2.16 {d4-d7}, [%[in4]] \n\t" @@ -538,7 +538,7 @@ void gaussianPyramidDown(const Size2D &srcSize, for (; x < roiw4; x += 4) { internal::prefetch(lane + 2 * x); -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld2.32 {d0-d3}, [%[in0]] \n\t" "vld2.32 {d4-d7}, [%[in4]] \n\t" @@ -672,7 +672,7 @@ void gaussianPyramidDown(const Size2D &srcSize, std::vector _buf(cn*(srcSize.width + 4) + 32/sizeof(f32)); f32* lane = internal::alignPtr(&_buf[2*cn], 32); -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 register float32x4_t vc6d4f32 asm ("q11") = vmovq_n_f32(1.5f); // 6/4 register float32x4_t vc1d4f32 asm ("q12") = vmovq_n_f32(0.25f); // 1/4 @@ -739,7 +739,7 @@ void gaussianPyramidDown(const Size2D &srcSize, for (; x < roiw4; x += 4) { internal::prefetch(lane + 2 * x); -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ __volatile__ ( "vld2.32 {d0-d3}, [%[in0]] \n\t" "vld2.32 {d8-d11}, [%[in4]] \n\t" diff --git a/3rdparty/carotene/src/scharr.cpp b/3rdparty/carotene/src/scharr.cpp index 2c4ba29..5695804 100644 --- a/3rdparty/carotene/src/scharr.cpp +++ b/3rdparty/carotene/src/scharr.cpp @@ -109,7 +109,7 @@ void ScharrDeriv(const Size2D &size, s32 cn, internal::prefetch(srow0 + x); internal::prefetch(srow1 + x); internal::prefetch(srow2 + x); -#if __GNUC_MINOR__ < 7 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 7 __asm__ ( "vld1.8 {d0}, [%[src0]] \n\t" "vld1.8 {d2}, [%[src2]] \n\t" @@ -161,7 +161,7 @@ void ScharrDeriv(const Size2D &size, s32 cn, x = 0; for( ; x < roiw8; x += 8 ) { -#if __GNUC_MINOR__ < 6 +#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ < 6 __asm__ ( "vld1.16 {d4-d5}, [%[s2ptr]] \n\t" "vld1.16 {d8-d9}, [%[s4ptr]] \n\t" @@ -198,7 +198,7 @@ void ScharrDeriv(const Size2D &size, s32 cn, vr.val[1] = vmlaq_s16(s3x10, s24, vc3); vst2q_s16(drow + x*2, vr); -#endif //__GNUC_MINOR__ < 6 +#endif } for( ; x < colsn; x++ ) { diff --git a/3rdparty/libpng/CMakeLists.txt b/3rdparty/libpng/CMakeLists.txt index 919d84e..8d338e7 100644 --- a/3rdparty/libpng/CMakeLists.txt +++ b/3rdparty/libpng/CMakeLists.txt @@ -28,11 +28,13 @@ file(GLOB lib_srcs *.c) file(GLOB lib_hdrs *.h) -if(ENABLE_NEON) - list(APPEND lib_srcs arm/arm_init.c arm/filter_neon.S arm/filter_neon_intrinsics.c) - add_definitions(-DPNG_ARM_NEON_OPT=2) -elseif(AARCH64) - add_definitions(-DPNG_ARM_NEON_OPT=0) # NEON assembler is not supported +if(ARM OR AARCH64) + if(ENABLE_NEON AND NOT AARCH64) + list(APPEND lib_srcs arm/arm_init.c arm/filter_neon.S arm/filter_neon_intrinsics.c) + add_definitions(-DPNG_ARM_NEON_OPT=2) + else() + add_definitions(-DPNG_ARM_NEON_OPT=0) # NEON assembler is not supported + endif() endif() if(ENABLE_SSE diff --git a/3rdparty/libwebp/CMakeLists.txt b/3rdparty/libwebp/CMakeLists.txt index a02e338..2aebc4c 100644 --- a/3rdparty/libwebp/CMakeLists.txt +++ b/3rdparty/libwebp/CMakeLists.txt @@ -5,7 +5,9 @@ project(${WEBP_LIBRARY}) ocv_include_directories(${CMAKE_CURRENT_SOURCE_DIR}) -ocv_include_directories(${CPUFEATURES_INCLUDE_DIRS}) +if(ANDROID) + ocv_include_directories(${CPUFEATURES_INCLUDE_DIRS}) +endif() file(GLOB lib_srcs dec/*.c demux/*.c dsp/*.c enc/*.c mux/*.c utils/*.c webp/*.c) file(GLOB lib_hdrs dec/*.h demux/*.h dsp/*.h enc/*.h mux/*.h utils/*.h webp/*.h) @@ -28,7 +30,9 @@ endif() add_definitions(-DWEBP_USE_THREAD) add_library(${WEBP_LIBRARY} STATIC ${lib_srcs} ${lib_hdrs}) -target_link_libraries(${WEBP_LIBRARY} ${CPUFEATURES_LIBRARIES}) +if(ANDROID) + target_link_libraries(${WEBP_LIBRARY} ${CPUFEATURES_LIBRARIES}) +endif() if(UNIX) if(CMAKE_COMPILER_IS_GNUCXX OR CV_ICC) diff --git a/CMakeLists.txt b/CMakeLists.txt index 16fe3ec..2d2e9fd 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -199,6 +199,7 @@ OCV_OPTION(WITH_1394 "Include IEEE1394 support" ON OCV_OPTION(WITH_AVFOUNDATION "Use AVFoundation for Video I/O (iOS/Mac)" ON IF APPLE) OCV_OPTION(WITH_CARBON "Use Carbon for UI instead of Cocoa" OFF IF APPLE ) OCV_OPTION(WITH_CAROTENE "Use NVidia carotene acceleration library for ARM platform" ON IF (ARM OR AARCH64) AND NOT IOS AND NOT (CMAKE_VERSION VERSION_LESS "2.8.11")) +OCV_OPTION(WITH_CPUFEATURES "Use cpufeatures Android library" ON IF ANDROID) OCV_OPTION(WITH_VTK "Include VTK library support (and build opencv_viz module eiher)" ON IF (NOT ANDROID AND NOT IOS AND NOT WINRT AND NOT CMAKE_CROSSCOMPILING) ) OCV_OPTION(WITH_CUDA "Include NVidia Cuda Runtime support" ON IF (NOT IOS AND NOT WINRT) ) OCV_OPTION(WITH_CUFFT "Include NVidia Cuda Fast Fourier Transform (FFT) library support" ON IF (NOT IOS AND NOT WINRT) ) @@ -302,7 +303,7 @@ OCV_OPTION(ENABLE_COVERAGE "Enable coverage collection with GCov" OCV_OPTION(ENABLE_OMIT_FRAME_POINTER "Enable -fomit-frame-pointer for GCC" ON IF CMAKE_COMPILER_IS_GNUCXX AND NOT (APPLE AND CMAKE_COMPILER_IS_CLANGCXX) ) OCV_OPTION(ENABLE_POWERPC "Enable PowerPC for GCC" ON IF (CMAKE_COMPILER_IS_GNUCXX AND CMAKE_SYSTEM_PROCESSOR MATCHES powerpc.*) ) OCV_OPTION(ENABLE_FAST_MATH "Enable -ffast-math (not recommended for GCC 4.6.x)" OFF IF (CMAKE_COMPILER_IS_GNUCXX AND (X86 OR X86_64)) ) -OCV_OPTION(ENABLE_NEON "Enable NEON instructions" "${NEON}" IF CMAKE_COMPILER_IS_GNUCXX AND (ARM OR AARCH64 OR IOS) ) +OCV_OPTION(ENABLE_NEON "Enable NEON instructions" (NEON OR ANDROID_ARM_NEON OR AARCH64) IF CMAKE_COMPILER_IS_GNUCXX AND (ARM OR AARCH64 OR IOS) ) OCV_OPTION(ENABLE_VFPV3 "Enable VFPv3-D32 instructions" OFF IF CMAKE_COMPILER_IS_GNUCXX AND (ARM OR AARCH64 OR IOS) ) OCV_OPTION(ENABLE_NOISY_WARNINGS "Show all warnings even if they are too noisy" OFF ) OCV_OPTION(OPENCV_WARNINGS_ARE_ERRORS "Treat warnings as errors" OFF ) @@ -578,8 +579,9 @@ endif() # Detect 3rd-party libraries # ---------------------------------------------------------------------------- -if(ANDROID) - add_subdirectory(3rdparty/cpufeatures) +if(ANDROID AND WITH_CPUFEATURES) + add_subdirectory(3rdparty/cpufeatures) + set(HAVE_CPUFEATURES 1) endif() include(cmake/OpenCVFindLibsGrfmt.cmake) diff --git a/cmake/OpenCVCompilerOptimizations.cmake b/cmake/OpenCVCompilerOptimizations.cmake index f6aebc7..f6d4208 100644 --- a/cmake/OpenCVCompilerOptimizations.cmake +++ b/cmake/OpenCVCompilerOptimizations.cmake @@ -122,7 +122,7 @@ endmacro() macro(ocv_append_optimization_flag var OPT) if(CPU_${OPT}_FLAGS_CONFLICT) - string(REGEX REPLACE " ${CPU_${OPT}_FLAGS_CONFLICT}" "" ${var} " ${${var}}") + string(REGEX REPLACE " ${CPU_${OPT}_FLAGS_CONFLICT}" "" ${var} " ${${var}} ") string(REGEX REPLACE "^ +" "" ${var} "${${var}}") endif() set(${var} "${${var}} ${CPU_${OPT}_FLAGS_ON}") @@ -250,17 +250,21 @@ if(X86 OR X86_64) endif() elseif(ARM OR AARCH64) + ocv_update(CPU_NEON_TEST_FILE "${OpenCV_SOURCE_DIR}/cmake/checks/cpu_neon.cpp") ocv_update(CPU_FP16_TEST_FILE "${OpenCV_SOURCE_DIR}/cmake/checks/cpu_fp16.cpp") if(NOT AARCH64) ocv_update(CPU_KNOWN_OPTIMIZATIONS "VFPV3;NEON;FP16") - ocv_update(CPU_NEON_FLAGS_ON "-mfpu=neon") ocv_update(CPU_VFPV3_FLAGS_ON "-mfpu=vfpv3") + ocv_update(CPU_NEON_FLAGS_ON "-mfpu=neon") + ocv_update(CPU_NEON_FLAGS_CONFLICT "-mfpu=[^ ]*") ocv_update(CPU_FP16_FLAGS_ON "-mfpu=neon-fp16") - set(CPU_BASELINE "DETECT" CACHE STRING "${HELP_CPU_BASELINE}") + ocv_update(CPU_FP16_IMPLIES "NEON") + ocv_update(CPU_FP16_FLAGS_CONFLICT "-mfpu=[^ ]*") else() ocv_update(CPU_KNOWN_OPTIMIZATIONS "NEON;FP16") ocv_update(CPU_NEON_FLAGS_ON "") - set(CPU_BASELINE "NEON" CACHE STRING "${HELP_CPU_BASELINE}") + ocv_update(CPU_FP16_IMPLIES "NEON") + set(CPU_BASELINE "NEON;FP16" CACHE STRING "${HELP_CPU_BASELINE}") endif() endif() @@ -270,6 +274,11 @@ set(CPU_DISPATCH "" CACHE STRING "${HELP_CPU_DISPATCH}") set_property(CACHE CPU_BASELINE PROPERTY STRINGS "" ${CPU_KNOWN_OPTIMIZATIONS}) set_property(CACHE CPU_DISPATCH PROPERTY STRINGS "" ${CPU_KNOWN_OPTIMIZATIONS}) +# Update CPU_BASELINE_DETECT flag +if(";${CPU_BASELINE};" MATCHES ";DETECT;") + set(CPU_BASELINE_DETECT ON) +endif() + set(CPU_BASELINE_FLAGS "") set(CPU_BASELINE_FINAL "") @@ -378,7 +387,7 @@ foreach(OPT ${CPU_KNOWN_OPTIMIZATIONS}) if(CPU_${OPT}_SUPPORTED) if(";${CPU_DISPATCH};" MATCHES ";${OPT};" AND NOT __is_from_baseline) list(APPEND CPU_DISPATCH_FINAL ${OPT}) - elseif(__is_from_baseline AND NOT CPU_BASELINE_DETECT) + elseif(__is_from_baseline) list(APPEND CPU_BASELINE_FINAL ${OPT}) ocv_append_optimization_flag(CPU_BASELINE_FLAGS ${OPT}) endif() @@ -483,12 +492,6 @@ macro(ocv_compiler_optimization_options) if(ARM) add_extra_compiler_option("-mfp16-format=ieee") endif(ARM) - if(ENABLE_NEON) - add_extra_compiler_option("-mfpu=neon") - endif() - if(ENABLE_VFPV3 AND NOT ENABLE_NEON) - add_extra_compiler_option("-mfpu=vfpv3") - endif() endmacro() macro(ocv_compiler_optimization_options_finalize) diff --git a/cmake/OpenCVFindLibsGrfmt.cmake b/cmake/OpenCVFindLibsGrfmt.cmake index cc82575..e555d1f 100644 --- a/cmake/OpenCVFindLibsGrfmt.cmake +++ b/cmake/OpenCVFindLibsGrfmt.cmake @@ -96,17 +96,23 @@ if(WITH_WEBP) ocv_clear_vars(WEBP_FOUND WEBP_LIBRARY WEBP_LIBRARIES WEBP_INCLUDE_DIR) else() include(cmake/OpenCVFindWebP.cmake) + if(WEBP_FOUND) + set(HAVE_WEBP 1) + endif() endif() endif() # --- Add libwebp to 3rdparty/libwebp and compile it if not available --- -if(WITH_WEBP AND NOT WEBP_FOUND) +if(WITH_WEBP AND NOT WEBP_FOUND + AND (NOT ANDROID OR HAVE_CPUFEATURES) +) set(WEBP_LIBRARY libwebp) set(WEBP_LIBRARIES ${WEBP_LIBRARY}) add_subdirectory("${OpenCV_SOURCE_DIR}/3rdparty/libwebp") set(WEBP_INCLUDE_DIR "${${WEBP_LIBRARY}_SOURCE_DIR}") + set(HAVE_WEBP 1) endif() if(NOT WEBP_VERSION AND WEBP_INCLUDE_DIR) diff --git a/cmake/OpenCVUtils.cmake b/cmake/OpenCVUtils.cmake index 4879eb6..ae665e8 100644 --- a/cmake/OpenCVUtils.cmake +++ b/cmake/OpenCVUtils.cmake @@ -386,6 +386,16 @@ macro(ocv_warnings_disable) endif(NOT ENABLE_NOISY_WARNINGS) endmacro() +macro(ocv_append_sourge_file_compile_definitions source) + get_source_file_property(_value "${source}" COMPILE_DEFINITIONS) + if(_value) + set(_value ${_value} ${ARGN}) + else() + set(_value ${ARGN}) + endif() + set_source_files_properties("${source}" PROPERTIES COMPILE_DEFINITIONS "${_value}") +endmacro() + macro(add_apple_compiler_options the_module) ocv_check_flag_support(OBJCXX "-fobjc-exceptions" HAVE_OBJC_EXCEPTIONS "") if(HAVE_OBJC_EXCEPTIONS) diff --git a/cmake/checks/cpu_neon.cpp b/cmake/checks/cpu_neon.cpp new file mode 100644 index 0000000..26f48ad --- /dev/null +++ b/cmake/checks/cpu_neon.cpp @@ -0,0 +1,27 @@ +#include + +#if (defined WIN32 || defined _WIN32) && defined(_M_ARM) +# include +# include +# define CV_NEON 1 +#elif defined(__ARM_NEON__) || (defined (__ARM_NEON) && defined(__aarch64__)) +# include +# define CV_NEON 1 +#endif + +#if defined CV_NEON +int test() +{ + const float src[] = { 0.0f, 0.0f, 0.0f, 0.0f }; + float32x4_t val = vld1q_f32((const float32_t*)(src)); + return (int)vgetq_lane_f32(val, 0); +} +#else +#error "NEON is not supported" +#endif + +int main() +{ + printf("%d\n", test()); + return 0; +} diff --git a/modules/core/CMakeLists.txt b/modules/core/CMakeLists.txt index 1ca1daf..e27a93e 100644 --- a/modules/core/CMakeLists.txt +++ b/modules/core/CMakeLists.txt @@ -32,7 +32,11 @@ source_group("Src" FILES "${OPENCV_MODULE_opencv_core_BINARY_DIR}/version_string ocv_glob_module_sources(SOURCES "${OPENCV_MODULE_opencv_core_BINARY_DIR}/version_string.inc" HEADERS ${lib_cuda_hdrs} ${lib_cuda_hdrs_detail}) -ocv_module_include_directories(${the_module} ${ZLIB_INCLUDE_DIRS} ${OPENCL_INCLUDE_DIRS} ${CPUFEATURES_INCLUDE_DIRS}) +ocv_module_include_directories(${the_module} ${ZLIB_INCLUDE_DIRS} ${OPENCL_INCLUDE_DIRS}) +if(ANDROID AND HAVE_CPUFEATURES) + ocv_append_sourge_file_compile_definitions(${CMAKE_CURRENT_SOURCE_DIR}/src/system.cpp "HAVE_CPUFEATURES=1") + ocv_module_include_directories(${CPUFEATURES_INCLUDE_DIRS}) +endif() ocv_create_module(${extra_libs}) ocv_target_link_libraries(${the_module} ${ZLIB_LIBRARIES} "${OPENCL_LIBRARIES}" "${VA_LIBRARIES}" "${LAPACK_LIBRARIES}" "${CPUFEATURES_LIBRARIES}") diff --git a/modules/core/include/opencv2/core/cv_cpu_dispatch.h b/modules/core/include/opencv2/core/cv_cpu_dispatch.h index c6beaca..c41c21d 100644 --- a/modules/core/include/opencv2/core/cv_cpu_dispatch.h +++ b/modules/core/include/opencv2/core/cv_cpu_dispatch.h @@ -71,7 +71,11 @@ # define CV_AVX 1 #endif #ifdef CV_CPU_COMPILE_FP16 -# include +# if defined(__arm__) || defined(__aarch64__) || defined(_M_ARM) +# include +# else +# include +# endif # define CV_FP16 1 #endif #ifdef CV_CPU_COMPILE_AVX2 diff --git a/modules/core/include/opencv2/core/hal/intrin_neon.hpp b/modules/core/include/opencv2/core/hal/intrin_neon.hpp index b126948..ce661c3 100644 --- a/modules/core/include/opencv2/core/hal/intrin_neon.hpp +++ b/modules/core/include/opencv2/core/hal/intrin_neon.hpp @@ -279,15 +279,27 @@ struct v_float64x2 #endif #if CV_FP16 -// Workaround for old comiplers +// Workaround for old compilers template static inline int16x4_t vreinterpret_s16_f16(T a) { return (int16x4_t)a; } template static inline float16x4_t vreinterpret_f16_s16(T a) { return (float16x4_t)a; } -template static inline float16x4_t vld1_f16(const T* ptr) -{ return vreinterpret_f16_s16(vld1_s16((const short*)ptr)); } -template static inline void vst1_f16(T* ptr, float16x4_t a) -{ vst1_s16((short*)ptr, vreinterpret_s16_f16(a)); } +template static inline float16x4_t cv_vld1_f16(const T* ptr) +{ +#ifndef vld1_f16 // APPLE compiler defines vld1_f16 as macro + return vreinterpret_f16_s16(vld1_s16((const short*)ptr)); +#else + return vld1_f16((const __fp16*)ptr); +#endif +} +template static inline void cv_vst1_f16(T* ptr, float16x4_t a) +{ +#ifndef vst1_f16 // APPLE compiler defines vst1_f16 as macro + vst1_s16((short*)ptr, vreinterpret_s16_f16(a)); +#else + vst1_f16((__fp16*)ptr, a); +#endif +} struct v_float16x4 { @@ -299,7 +311,7 @@ struct v_float16x4 v_float16x4(short v0, short v1, short v2, short v3) { short v[] = {v0, v1, v2, v3}; - val = vld1_f16(v); + val = cv_vld1_f16(v); } short get0() const { @@ -778,9 +790,9 @@ OPENCV_HAL_IMPL_NEON_LOADSTORE_OP(v_float64x2, double, f64) #if CV_FP16 // Workaround for old comiplers inline v_float16x4 v_load_f16(const short* ptr) -{ return v_float16x4(vld1_f16(ptr)); } +{ return v_float16x4(cv_vld1_f16(ptr)); } inline void v_store_f16(short* ptr, v_float16x4& a) -{ vst1_f16(ptr, a.val); } +{ cv_vst1_f16(ptr, a.val); } #endif #define OPENCV_HAL_IMPL_NEON_REDUCE_OP_8(_Tpvec, _Tpnvec, scalartype, func, vectorfunc, suffix) \ diff --git a/modules/core/src/convert.fp16.cpp b/modules/core/src/convert.fp16.cpp index f5ad29b..927fb62 100644 --- a/modules/core/src/convert.fp16.cpp +++ b/modules/core/src/convert.fp16.cpp @@ -108,11 +108,6 @@ void cvtScaleHalf_SIMD16f32f( const short* src, size_t sstep, float* dst, size_t #elif CV_NEON const static int cVectorWidth = 4; -template static inline float16x4_t vld1_f16(const T* ptr) -{ return (float16x4_t)vld1_s16((const short*)ptr); } -template static inline void vst1_f16(T* ptr, float16x4_t a) -{ vst1_s16((short*)ptr, a); } - void cvtScaleHalf_SIMD32f16f( const float* src, size_t sstep, short* dst, size_t dstep, cv::Size size ) { CV_INSTRUMENT_REGION() @@ -129,7 +124,7 @@ void cvtScaleHalf_SIMD32f16f( const float* src, size_t sstep, short* dst, size_t float16x4_t v_dst = vcvt_f16_f32(v_src); - vst1_f16((__fp16*)dst + x, v_dst); + cv_vst1_f16((__fp16*)dst + x, v_dst); } for ( ; x < size.width; x++ ) @@ -151,7 +146,7 @@ void cvtScaleHalf_SIMD16f32f( const short* src, size_t sstep, float* dst, size_t int x = 0; for ( ; x <= size.width - cVectorWidth ; x += cVectorWidth ) { - float16x4_t v_src = vld1_f16((__fp16*)src + x); + float16x4_t v_src = cv_vld1_f16((__fp16*)src + x); float32x4_t v_dst = vcvt_f32_f16(v_src); diff --git a/modules/core/src/convert.hpp b/modules/core/src/convert.hpp index 936d1ca..c149d21 100644 --- a/modules/core/src/convert.hpp +++ b/modules/core/src/convert.hpp @@ -13,7 +13,7 @@ const unsigned int kBiasFp32Exponent = 127; #endif #if CV_FP16_TYPE -float convertFp16SW(short fp16) +inline float convertFp16SW(short fp16) { // Fp16 -> Fp32 Cv16suf a; @@ -21,7 +21,7 @@ float convertFp16SW(short fp16) return (float)a.h; } #else -float convertFp16SW(short fp16) +inline float convertFp16SW(short fp16) { // Fp16 -> Fp32 Cv16suf b; @@ -75,7 +75,7 @@ float convertFp16SW(short fp16) #endif #if CV_FP16_TYPE -short convertFp16SW(float fp32) +inline short convertFp16SW(float fp32) { // Fp32 -> Fp16 Cv16suf a; @@ -83,7 +83,7 @@ short convertFp16SW(float fp32) return a.i; } #else -short convertFp16SW(float fp32) +inline short convertFp16SW(float fp32) { // Fp32 -> Fp16 Cv32suf a; diff --git a/modules/core/src/system.cpp b/modules/core/src/system.cpp index b864ba4..c3317dd 100644 --- a/modules/core/src/system.cpp +++ b/modules/core/src/system.cpp @@ -73,7 +73,7 @@ Mutex* __initialization_mutex_initializer = &getInitializationMutex(); #endif #endif -#if defined ANDROID +#if defined ANDROID && defined HAVE_CPUFEATURES # include #endif @@ -450,11 +450,27 @@ struct HWFeatures have[CV_CPU_NEON] = true; have[CV_CPU_FP16] = true; #elif defined __arm__ && defined __ANDROID__ + #if defined HAVE_CPUFEATURES __android_log_print(ANDROID_LOG_INFO, "OpenCV", "calling android_getCpuFeatures() ..."); uint64_t features = android_getCpuFeatures(); __android_log_print(ANDROID_LOG_INFO, "OpenCV", "calling android_getCpuFeatures() ... Done (%llx)", features); have[CV_CPU_NEON] = (features & ANDROID_CPU_ARM_FEATURE_NEON) != 0; have[CV_CPU_FP16] = (features & ANDROID_CPU_ARM_FEATURE_VFP_FP16) != 0; + #else + __android_log_print(ANDROID_LOG_INFO, "OpenCV", "cpufeatures library is not avaialble for CPU detection"); + #if CV_NEON + __android_log_print(ANDROID_LOG_INFO, "OpenCV", "- NEON instructions is enabled via build flags"); + have[CV_CPU_NEON] = true; + #else + __android_log_print(ANDROID_LOG_INFO, "OpenCV", "- NEON instructions is NOT enabled via build flags"); + #endif + #if CV_FP16 + __android_log_print(ANDROID_LOG_INFO, "OpenCV", "- FP16 instructions is enabled via build flags"); + have[CV_CPU_FP16] = true; + #else + __android_log_print(ANDROID_LOG_INFO, "OpenCV", "- FP16 instructions is NOT enabled via build flags"); + #endif + #endif #elif defined __arm__ int cpufile = open("/proc/self/auxv", O_RDONLY); diff --git a/modules/imgcodecs/CMakeLists.txt b/modules/imgcodecs/CMakeLists.txt index 546d956..0096e77 100644 --- a/modules/imgcodecs/CMakeLists.txt +++ b/modules/imgcodecs/CMakeLists.txt @@ -23,7 +23,7 @@ if(HAVE_JPEG) list(APPEND GRFMT_LIBS ${JPEG_LIBRARIES}) endif() -if(WITH_WEBP) +if(HAVE_WEBP) add_definitions(-DHAVE_WEBP) ocv_include_directories(${WEBP_INCLUDE_DIR}) list(APPEND GRFMT_LIBS ${WEBP_LIBRARIES}) -- 2.7.4