mirror of
https://gitlab.winehq.org/wine/wine-gecko.git
synced 2024-09-13 09:24:08 -07:00
5b18a38cd7
--HG-- rename : media/webrtc/trunk/third_party/libyuv/AUTHORS => media/libyuv/AUTHORS rename : media/webrtc/trunk/third_party/libyuv/Android.mk => media/libyuv/Android.mk rename : media/webrtc/trunk/third_party/libyuv/DEPS => media/libyuv/DEPS rename : media/webrtc/trunk/third_party/libyuv/LICENSE => media/libyuv/LICENSE rename : media/webrtc/trunk/third_party/libyuv/LICENSE_THIRD_PARTY => media/libyuv/LICENSE_THIRD_PARTY rename : media/webrtc/trunk/third_party/libyuv/OWNERS => media/libyuv/OWNERS rename : media/webrtc/trunk/third_party/libyuv/PATENTS => media/libyuv/PATENTS rename : media/webrtc/trunk/third_party/libyuv/README.chromium => media/libyuv/README.chromium rename : media/webrtc/trunk/third_party/libyuv/all.gyp => media/libyuv/all.gyp rename : media/webrtc/trunk/third_party/libyuv/codereview.settings => media/libyuv/codereview.settings rename : media/webrtc/trunk/third_party/libyuv/include/libyuv.h => media/libyuv/include/libyuv.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/basic_types.h => media/libyuv/include/libyuv/basic_types.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/compare.h => media/libyuv/include/libyuv/compare.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/convert.h => media/libyuv/include/libyuv/convert.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/convert_argb.h => media/libyuv/include/libyuv/convert_argb.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/convert_from.h => media/libyuv/include/libyuv/convert_from.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/convert_from_argb.h => media/libyuv/include/libyuv/convert_from_argb.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/cpu_id.h => media/libyuv/include/libyuv/cpu_id.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/format_conversion.h => media/libyuv/include/libyuv/format_conversion.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/mjpeg_decoder.h => media/libyuv/include/libyuv/mjpeg_decoder.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/planar_functions.h => media/libyuv/include/libyuv/planar_functions.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/rotate.h => media/libyuv/include/libyuv/rotate.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/rotate_argb.h => media/libyuv/include/libyuv/rotate_argb.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/row.h => media/libyuv/include/libyuv/row.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/scale.h => media/libyuv/include/libyuv/scale.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/scale_argb.h => media/libyuv/include/libyuv/scale_argb.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/scale_row.h => media/libyuv/include/libyuv/scale_row.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/version.h => media/libyuv/include/libyuv/version.h rename : media/webrtc/trunk/third_party/libyuv/include/libyuv/video_common.h => media/libyuv/include/libyuv/video_common.h rename : media/webrtc/trunk/third_party/libyuv/libyuv.gyp => media/libyuv/libyuv.gyp rename : media/webrtc/trunk/third_party/libyuv/libyuv.gypi => media/libyuv/libyuv.gypi rename : media/webrtc/trunk/third_party/libyuv/libyuv_nacl.gyp => media/libyuv/libyuv_nacl.gyp rename : media/webrtc/trunk/third_party/libyuv/libyuv_test.gyp => media/libyuv/libyuv_test.gyp rename : media/webrtc/trunk/third_party/libyuv/linux.mk => media/libyuv/linux.mk rename : media/webrtc/trunk/third_party/libyuv/public.mk => media/libyuv/public.mk rename : media/webrtc/trunk/third_party/libyuv/source/compare.cc => media/libyuv/source/compare.cc rename : media/webrtc/trunk/third_party/libyuv/source/compare_common.cc => media/libyuv/source/compare_common.cc rename : media/webrtc/trunk/third_party/libyuv/source/compare_neon.cc => media/libyuv/source/compare_neon.cc rename : media/webrtc/trunk/third_party/libyuv/source/compare_posix.cc => media/libyuv/source/compare_posix.cc rename : media/webrtc/trunk/third_party/libyuv/source/compare_win.cc => media/libyuv/source/compare_win.cc rename : media/webrtc/trunk/third_party/libyuv/source/convert.cc => media/libyuv/source/convert.cc rename : media/webrtc/trunk/third_party/libyuv/source/convert_argb.cc => media/libyuv/source/convert_argb.cc rename : media/webrtc/trunk/third_party/libyuv/source/convert_from.cc => media/libyuv/source/convert_from.cc rename : media/webrtc/trunk/third_party/libyuv/source/convert_from_argb.cc => media/libyuv/source/convert_from_argb.cc rename : media/webrtc/trunk/third_party/libyuv/source/convert_jpeg.cc => media/libyuv/source/convert_jpeg.cc rename : media/webrtc/trunk/third_party/libyuv/source/convert_to_argb.cc => media/libyuv/source/convert_to_argb.cc rename : media/webrtc/trunk/third_party/libyuv/source/convert_to_i420.cc => media/libyuv/source/convert_to_i420.cc rename : media/webrtc/trunk/third_party/libyuv/source/cpu_id.cc => media/libyuv/source/cpu_id.cc rename : media/webrtc/trunk/third_party/libyuv/source/format_conversion.cc => media/libyuv/source/format_conversion.cc rename : media/webrtc/trunk/third_party/libyuv/source/mjpeg_decoder.cc => media/libyuv/source/mjpeg_decoder.cc rename : media/webrtc/trunk/third_party/libyuv/source/mjpeg_validate.cc => media/libyuv/source/mjpeg_validate.cc rename : media/webrtc/trunk/third_party/libyuv/source/planar_functions.cc => media/libyuv/source/planar_functions.cc rename : media/webrtc/trunk/third_party/libyuv/source/rotate.cc => media/libyuv/source/rotate.cc rename : media/webrtc/trunk/third_party/libyuv/source/rotate_argb.cc => media/libyuv/source/rotate_argb.cc rename : media/webrtc/trunk/third_party/libyuv/source/rotate_mips.cc => media/libyuv/source/rotate_mips.cc rename : media/webrtc/trunk/third_party/libyuv/source/rotate_neon.cc => media/libyuv/source/rotate_neon.cc rename : media/webrtc/trunk/third_party/libyuv/source/row_any.cc => media/libyuv/source/row_any.cc rename : media/webrtc/trunk/third_party/libyuv/source/row_common.cc => media/libyuv/source/row_common.cc rename : media/webrtc/trunk/third_party/libyuv/source/row_mips.cc => media/libyuv/source/row_mips.cc rename : media/webrtc/trunk/third_party/libyuv/source/row_neon.cc => media/libyuv/source/row_neon.cc rename : media/webrtc/trunk/third_party/libyuv/source/row_posix.cc => media/libyuv/source/row_posix.cc rename : media/webrtc/trunk/third_party/libyuv/source/row_win.cc => media/libyuv/source/row_win.cc rename : media/webrtc/trunk/third_party/libyuv/source/row_x86.asm => media/libyuv/source/row_x86.asm rename : media/webrtc/trunk/third_party/libyuv/source/scale.cc => media/libyuv/source/scale.cc rename : media/webrtc/trunk/third_party/libyuv/source/scale_argb.cc => media/libyuv/source/scale_argb.cc rename : media/webrtc/trunk/third_party/libyuv/source/scale_common.cc => media/libyuv/source/scale_common.cc rename : media/webrtc/trunk/third_party/libyuv/source/scale_mips.cc => media/libyuv/source/scale_mips.cc rename : media/webrtc/trunk/third_party/libyuv/source/scale_neon.cc => media/libyuv/source/scale_neon.cc rename : media/webrtc/trunk/third_party/libyuv/source/scale_posix.cc => media/libyuv/source/scale_posix.cc rename : media/webrtc/trunk/third_party/libyuv/source/scale_win.cc => media/libyuv/source/scale_win.cc rename : media/webrtc/trunk/third_party/libyuv/source/video_common.cc => media/libyuv/source/video_common.cc rename : media/webrtc/trunk/third_party/libyuv/source/x86inc.asm => media/libyuv/source/x86inc.asm rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/libyuv_tests.bat => media/libyuv/tools/valgrind-libyuv/libyuv_tests.bat rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/libyuv_tests.py => media/libyuv/tools/valgrind-libyuv/libyuv_tests.py rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/libyuv_tests.sh => media/libyuv/tools/valgrind-libyuv/libyuv_tests.sh rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/memcheck/OWNERS => media/libyuv/tools/valgrind-libyuv/memcheck/OWNERS rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/memcheck/PRESUBMIT.py => media/libyuv/tools/valgrind-libyuv/memcheck/PRESUBMIT.py rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/memcheck/suppressions.txt => media/libyuv/tools/valgrind-libyuv/memcheck/suppressions.txt rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/memcheck/suppressions_mac.txt => media/libyuv/tools/valgrind-libyuv/memcheck/suppressions_mac.txt rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/memcheck/suppressions_win32.txt => media/libyuv/tools/valgrind-libyuv/memcheck/suppressions_win32.txt rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/tsan/OWNERS => media/libyuv/tools/valgrind-libyuv/tsan/OWNERS rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/tsan/PRESUBMIT.py => media/libyuv/tools/valgrind-libyuv/tsan/PRESUBMIT.py rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/tsan/suppressions.txt => media/libyuv/tools/valgrind-libyuv/tsan/suppressions.txt rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/tsan/suppressions_mac.txt => media/libyuv/tools/valgrind-libyuv/tsan/suppressions_mac.txt rename : media/webrtc/trunk/third_party/libyuv/tools/valgrind-libyuv/tsan/suppressions_win32.txt => media/libyuv/tools/valgrind-libyuv/tsan/suppressions_win32.txt rename : media/webrtc/trunk/third_party/libyuv/unit_test/basictypes_test.cc => media/libyuv/unit_test/basictypes_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/compare_test.cc => media/libyuv/unit_test/compare_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/convert_test.cc => media/libyuv/unit_test/convert_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/cpu_test.cc => media/libyuv/unit_test/cpu_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/math_test.cc => media/libyuv/unit_test/math_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/planar_test.cc => media/libyuv/unit_test/planar_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/rotate_argb_test.cc => media/libyuv/unit_test/rotate_argb_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/rotate_test.cc => media/libyuv/unit_test/rotate_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/scale_argb_test.cc => media/libyuv/unit_test/scale_argb_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/scale_test.cc => media/libyuv/unit_test/scale_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/testdata/arm_v7.txt => media/libyuv/unit_test/testdata/arm_v7.txt rename : media/webrtc/trunk/third_party/libyuv/unit_test/testdata/tegra3.txt => media/libyuv/unit_test/testdata/tegra3.txt rename : media/webrtc/trunk/third_party/libyuv/unit_test/unit_test.cc => media/libyuv/unit_test/unit_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/unit_test.h => media/libyuv/unit_test/unit_test.h rename : media/webrtc/trunk/third_party/libyuv/unit_test/version_test.cc => media/libyuv/unit_test/version_test.cc rename : media/webrtc/trunk/third_party/libyuv/unit_test/video_common_test.cc => media/libyuv/unit_test/video_common_test.cc rename : media/webrtc/trunk/third_party/libyuv/util/Makefile => media/libyuv/util/Makefile rename : media/webrtc/trunk/third_party/libyuv/util/compare.cc => media/libyuv/util/compare.cc rename : media/webrtc/trunk/third_party/libyuv/util/convert.cc => media/libyuv/util/convert.cc rename : media/webrtc/trunk/third_party/libyuv/util/cpuid.c => media/libyuv/util/cpuid.c rename : media/webrtc/trunk/third_party/libyuv/util/psnr.cc => media/libyuv/util/psnr.cc rename : media/webrtc/trunk/third_party/libyuv/util/psnr.h => media/libyuv/util/psnr.h rename : media/webrtc/trunk/third_party/libyuv/util/psnr_main.cc => media/libyuv/util/psnr_main.cc rename : media/webrtc/trunk/third_party/libyuv/util/ssim.cc => media/libyuv/util/ssim.cc rename : media/webrtc/trunk/third_party/libyuv/util/ssim.h => media/libyuv/util/ssim.h rename : media/webrtc/trunk/third_party/libyuv/winarm.mk => media/libyuv/winarm.mk
685 lines
27 KiB
C++
685 lines
27 KiB
C++
/*
|
|
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
|
|
*
|
|
* Use of this source code is governed by a BSD-style license
|
|
* that can be found in the LICENSE file in the root of the source
|
|
* tree. An additional intellectual property rights grant can be found
|
|
* in the file PATENTS. All contributing project authors may
|
|
* be found in the AUTHORS file in the root of the source tree.
|
|
*/
|
|
|
|
#include "libyuv/row.h"
|
|
|
|
#ifdef __cplusplus
|
|
namespace libyuv {
|
|
extern "C" {
|
|
#endif
|
|
|
|
// This module is for GCC Neon.
|
|
#if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__)
|
|
|
|
// NEON downscalers with interpolation.
|
|
// Provided by Fritz Koenig
|
|
|
|
// Read 32x1 throw away even pixels, and write 16x1.
|
|
void ScaleRowDown2_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
|
|
uint8* dst, int dst_width) {
|
|
asm volatile (
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
// load even pixels into q0, odd into q1
|
|
"vld2.8 {q0, q1}, [%0]! \n"
|
|
"subs %2, %2, #16 \n" // 16 processed per loop
|
|
"vst1.8 {q1}, [%1]! \n" // store odd pixels
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst), // %1
|
|
"+r"(dst_width) // %2
|
|
:
|
|
: "q0", "q1" // Clobber List
|
|
);
|
|
}
|
|
|
|
// Read 32x2 average down and write 16x1.
|
|
void ScaleRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
|
|
uint8* dst, int dst_width) {
|
|
asm volatile (
|
|
// change the stride to row 2 pointer
|
|
"add %1, %0 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld1.8 {q0, q1}, [%0]! \n" // load row 1 and post inc
|
|
"vld1.8 {q2, q3}, [%1]! \n" // load row 2 and post inc
|
|
"subs %3, %3, #16 \n" // 16 processed per loop
|
|
"vpaddl.u8 q0, q0 \n" // row 1 add adjacent
|
|
"vpaddl.u8 q1, q1 \n"
|
|
"vpadal.u8 q0, q2 \n" // row 2 add adjacent + row1
|
|
"vpadal.u8 q1, q3 \n"
|
|
"vrshrn.u16 d0, q0, #2 \n" // downshift, round and pack
|
|
"vrshrn.u16 d1, q1, #2 \n"
|
|
"vst1.8 {q0}, [%2]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(src_stride), // %1
|
|
"+r"(dst), // %2
|
|
"+r"(dst_width) // %3
|
|
:
|
|
: "q0", "q1", "q2", "q3" // Clobber List
|
|
);
|
|
}
|
|
|
|
void ScaleRowDown4_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
|
|
uint8* dst_ptr, int dst_width) {
|
|
asm volatile (
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld4.8 {d0, d1, d2, d3}, [%0]! \n" // src line 0
|
|
"subs %2, %2, #8 \n" // 8 processed per loop
|
|
"vst1.8 {d2}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst_ptr), // %1
|
|
"+r"(dst_width) // %2
|
|
:
|
|
: "q0", "q1", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
void ScaleRowDown4Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
|
|
uint8* dst_ptr, int dst_width) {
|
|
asm volatile (
|
|
"add r4, %0, %3 \n"
|
|
"add r5, r4, %3 \n"
|
|
"add %3, r5, %3 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld1.8 {q0}, [%0]! \n" // load up 16x4
|
|
"vld1.8 {q1}, [r4]! \n"
|
|
"vld1.8 {q2}, [r5]! \n"
|
|
"vld1.8 {q3}, [%3]! \n"
|
|
"subs %2, %2, #4 \n"
|
|
"vpaddl.u8 q0, q0 \n"
|
|
"vpadal.u8 q0, q1 \n"
|
|
"vpadal.u8 q0, q2 \n"
|
|
"vpadal.u8 q0, q3 \n"
|
|
"vpaddl.u16 q0, q0 \n"
|
|
"vrshrn.u32 d0, q0, #4 \n" // divide by 16 w/rounding
|
|
"vmovn.u16 d0, q0 \n"
|
|
"vst1.32 {d0[0]}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst_ptr), // %1
|
|
"+r"(dst_width) // %2
|
|
: "r"(src_stride) // %3
|
|
: "r4", "r5", "q0", "q1", "q2", "q3", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
// Down scale from 4 to 3 pixels. Use the neon multilane read/write
|
|
// to load up the every 4th pixel into a 4 different registers.
|
|
// Point samples 32 pixels to 24 pixels.
|
|
void ScaleRowDown34_NEON(const uint8* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8* dst_ptr, int dst_width) {
|
|
asm volatile (
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld4.8 {d0, d1, d2, d3}, [%0]! \n" // src line 0
|
|
"subs %2, %2, #24 \n"
|
|
"vmov d2, d3 \n" // order d0, d1, d2
|
|
"vst3.8 {d0, d1, d2}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst_ptr), // %1
|
|
"+r"(dst_width) // %2
|
|
:
|
|
: "d0", "d1", "d2", "d3", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
void ScaleRowDown34_0_Box_NEON(const uint8* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8* dst_ptr, int dst_width) {
|
|
asm volatile (
|
|
"vmov.u8 d24, #3 \n"
|
|
"add %3, %0 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld4.8 {d0, d1, d2, d3}, [%0]! \n" // src line 0
|
|
"vld4.8 {d4, d5, d6, d7}, [%3]! \n" // src line 1
|
|
"subs %2, %2, #24 \n"
|
|
|
|
// filter src line 0 with src line 1
|
|
// expand chars to shorts to allow for room
|
|
// when adding lines together
|
|
"vmovl.u8 q8, d4 \n"
|
|
"vmovl.u8 q9, d5 \n"
|
|
"vmovl.u8 q10, d6 \n"
|
|
"vmovl.u8 q11, d7 \n"
|
|
|
|
// 3 * line_0 + line_1
|
|
"vmlal.u8 q8, d0, d24 \n"
|
|
"vmlal.u8 q9, d1, d24 \n"
|
|
"vmlal.u8 q10, d2, d24 \n"
|
|
"vmlal.u8 q11, d3, d24 \n"
|
|
|
|
// (3 * line_0 + line_1) >> 2
|
|
"vqrshrn.u16 d0, q8, #2 \n"
|
|
"vqrshrn.u16 d1, q9, #2 \n"
|
|
"vqrshrn.u16 d2, q10, #2 \n"
|
|
"vqrshrn.u16 d3, q11, #2 \n"
|
|
|
|
// a0 = (src[0] * 3 + s[1] * 1) >> 2
|
|
"vmovl.u8 q8, d1 \n"
|
|
"vmlal.u8 q8, d0, d24 \n"
|
|
"vqrshrn.u16 d0, q8, #2 \n"
|
|
|
|
// a1 = (src[1] * 1 + s[2] * 1) >> 1
|
|
"vrhadd.u8 d1, d1, d2 \n"
|
|
|
|
// a2 = (src[2] * 1 + s[3] * 3) >> 2
|
|
"vmovl.u8 q8, d2 \n"
|
|
"vmlal.u8 q8, d3, d24 \n"
|
|
"vqrshrn.u16 d2, q8, #2 \n"
|
|
|
|
"vst3.8 {d0, d1, d2}, [%1]! \n"
|
|
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst_ptr), // %1
|
|
"+r"(dst_width), // %2
|
|
"+r"(src_stride) // %3
|
|
:
|
|
: "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11", "d24", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
void ScaleRowDown34_1_Box_NEON(const uint8* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8* dst_ptr, int dst_width) {
|
|
asm volatile (
|
|
"vmov.u8 d24, #3 \n"
|
|
"add %3, %0 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld4.8 {d0, d1, d2, d3}, [%0]! \n" // src line 0
|
|
"vld4.8 {d4, d5, d6, d7}, [%3]! \n" // src line 1
|
|
"subs %2, %2, #24 \n"
|
|
// average src line 0 with src line 1
|
|
"vrhadd.u8 q0, q0, q2 \n"
|
|
"vrhadd.u8 q1, q1, q3 \n"
|
|
|
|
// a0 = (src[0] * 3 + s[1] * 1) >> 2
|
|
"vmovl.u8 q3, d1 \n"
|
|
"vmlal.u8 q3, d0, d24 \n"
|
|
"vqrshrn.u16 d0, q3, #2 \n"
|
|
|
|
// a1 = (src[1] * 1 + s[2] * 1) >> 1
|
|
"vrhadd.u8 d1, d1, d2 \n"
|
|
|
|
// a2 = (src[2] * 1 + s[3] * 3) >> 2
|
|
"vmovl.u8 q3, d2 \n"
|
|
"vmlal.u8 q3, d3, d24 \n"
|
|
"vqrshrn.u16 d2, q3, #2 \n"
|
|
|
|
"vst3.8 {d0, d1, d2}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst_ptr), // %1
|
|
"+r"(dst_width), // %2
|
|
"+r"(src_stride) // %3
|
|
:
|
|
: "r4", "q0", "q1", "q2", "q3", "d24", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
#define HAS_SCALEROWDOWN38_NEON
|
|
static uvec8 kShuf38 =
|
|
{ 0, 3, 6, 8, 11, 14, 16, 19, 22, 24, 27, 30, 0, 0, 0, 0 };
|
|
static uvec8 kShuf38_2 =
|
|
{ 0, 8, 16, 2, 10, 17, 4, 12, 18, 6, 14, 19, 0, 0, 0, 0 };
|
|
static vec16 kMult38_Div6 =
|
|
{ 65536 / 12, 65536 / 12, 65536 / 12, 65536 / 12,
|
|
65536 / 12, 65536 / 12, 65536 / 12, 65536 / 12 };
|
|
static vec16 kMult38_Div9 =
|
|
{ 65536 / 18, 65536 / 18, 65536 / 18, 65536 / 18,
|
|
65536 / 18, 65536 / 18, 65536 / 18, 65536 / 18 };
|
|
|
|
// 32 -> 12
|
|
void ScaleRowDown38_NEON(const uint8* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8* dst_ptr, int dst_width) {
|
|
asm volatile (
|
|
"vld1.8 {q3}, [%3] \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld1.8 {d0, d1, d2, d3}, [%0]! \n"
|
|
"subs %2, %2, #12 \n"
|
|
"vtbl.u8 d4, {d0, d1, d2, d3}, d6 \n"
|
|
"vtbl.u8 d5, {d0, d1, d2, d3}, d7 \n"
|
|
"vst1.8 {d4}, [%1]! \n"
|
|
"vst1.32 {d5[0]}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst_ptr), // %1
|
|
"+r"(dst_width) // %2
|
|
: "r"(&kShuf38) // %3
|
|
: "d0", "d1", "d2", "d3", "d4", "d5", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
// 32x3 -> 12x1
|
|
void OMITFP ScaleRowDown38_3_Box_NEON(const uint8* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8* dst_ptr, int dst_width) {
|
|
asm volatile (
|
|
"vld1.16 {q13}, [%4] \n"
|
|
"vld1.8 {q14}, [%5] \n"
|
|
"vld1.8 {q15}, [%6] \n"
|
|
"add r4, %0, %3, lsl #1 \n"
|
|
"add %3, %0 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
|
|
// d0 = 00 40 01 41 02 42 03 43
|
|
// d1 = 10 50 11 51 12 52 13 53
|
|
// d2 = 20 60 21 61 22 62 23 63
|
|
// d3 = 30 70 31 71 32 72 33 73
|
|
"vld4.8 {d0, d1, d2, d3}, [%0]! \n"
|
|
"vld4.8 {d4, d5, d6, d7}, [%3]! \n"
|
|
"vld4.8 {d16, d17, d18, d19}, [r4]! \n"
|
|
"subs %2, %2, #12 \n"
|
|
|
|
// Shuffle the input data around to get align the data
|
|
// so adjacent data can be added. 0,1 - 2,3 - 4,5 - 6,7
|
|
// d0 = 00 10 01 11 02 12 03 13
|
|
// d1 = 40 50 41 51 42 52 43 53
|
|
"vtrn.u8 d0, d1 \n"
|
|
"vtrn.u8 d4, d5 \n"
|
|
"vtrn.u8 d16, d17 \n"
|
|
|
|
// d2 = 20 30 21 31 22 32 23 33
|
|
// d3 = 60 70 61 71 62 72 63 73
|
|
"vtrn.u8 d2, d3 \n"
|
|
"vtrn.u8 d6, d7 \n"
|
|
"vtrn.u8 d18, d19 \n"
|
|
|
|
// d0 = 00+10 01+11 02+12 03+13
|
|
// d2 = 40+50 41+51 42+52 43+53
|
|
"vpaddl.u8 q0, q0 \n"
|
|
"vpaddl.u8 q2, q2 \n"
|
|
"vpaddl.u8 q8, q8 \n"
|
|
|
|
// d3 = 60+70 61+71 62+72 63+73
|
|
"vpaddl.u8 d3, d3 \n"
|
|
"vpaddl.u8 d7, d7 \n"
|
|
"vpaddl.u8 d19, d19 \n"
|
|
|
|
// combine source lines
|
|
"vadd.u16 q0, q2 \n"
|
|
"vadd.u16 q0, q8 \n"
|
|
"vadd.u16 d4, d3, d7 \n"
|
|
"vadd.u16 d4, d19 \n"
|
|
|
|
// dst_ptr[3] = (s[6 + st * 0] + s[7 + st * 0]
|
|
// + s[6 + st * 1] + s[7 + st * 1]
|
|
// + s[6 + st * 2] + s[7 + st * 2]) / 6
|
|
"vqrdmulh.s16 q2, q2, q13 \n"
|
|
"vmovn.u16 d4, q2 \n"
|
|
|
|
// Shuffle 2,3 reg around so that 2 can be added to the
|
|
// 0,1 reg and 3 can be added to the 4,5 reg. This
|
|
// requires expanding from u8 to u16 as the 0,1 and 4,5
|
|
// registers are already expanded. Then do transposes
|
|
// to get aligned.
|
|
// q2 = xx 20 xx 30 xx 21 xx 31 xx 22 xx 32 xx 23 xx 33
|
|
"vmovl.u8 q1, d2 \n"
|
|
"vmovl.u8 q3, d6 \n"
|
|
"vmovl.u8 q9, d18 \n"
|
|
|
|
// combine source lines
|
|
"vadd.u16 q1, q3 \n"
|
|
"vadd.u16 q1, q9 \n"
|
|
|
|
// d4 = xx 20 xx 30 xx 22 xx 32
|
|
// d5 = xx 21 xx 31 xx 23 xx 33
|
|
"vtrn.u32 d2, d3 \n"
|
|
|
|
// d4 = xx 20 xx 21 xx 22 xx 23
|
|
// d5 = xx 30 xx 31 xx 32 xx 33
|
|
"vtrn.u16 d2, d3 \n"
|
|
|
|
// 0+1+2, 3+4+5
|
|
"vadd.u16 q0, q1 \n"
|
|
|
|
// Need to divide, but can't downshift as the the value
|
|
// isn't a power of 2. So multiply by 65536 / n
|
|
// and take the upper 16 bits.
|
|
"vqrdmulh.s16 q0, q0, q15 \n"
|
|
|
|
// Align for table lookup, vtbl requires registers to
|
|
// be adjacent
|
|
"vmov.u8 d2, d4 \n"
|
|
|
|
"vtbl.u8 d3, {d0, d1, d2}, d28 \n"
|
|
"vtbl.u8 d4, {d0, d1, d2}, d29 \n"
|
|
|
|
"vst1.8 {d3}, [%1]! \n"
|
|
"vst1.32 {d4[0]}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst_ptr), // %1
|
|
"+r"(dst_width), // %2
|
|
"+r"(src_stride) // %3
|
|
: "r"(&kMult38_Div6), // %4
|
|
"r"(&kShuf38_2), // %5
|
|
"r"(&kMult38_Div9) // %6
|
|
: "r4", "q0", "q1", "q2", "q3", "q8", "q9",
|
|
"q13", "q14", "q15", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
// 32x2 -> 12x1
|
|
void ScaleRowDown38_2_Box_NEON(const uint8* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8* dst_ptr, int dst_width) {
|
|
asm volatile (
|
|
"vld1.16 {q13}, [%4] \n"
|
|
"vld1.8 {q14}, [%5] \n"
|
|
"add %3, %0 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
|
|
// d0 = 00 40 01 41 02 42 03 43
|
|
// d1 = 10 50 11 51 12 52 13 53
|
|
// d2 = 20 60 21 61 22 62 23 63
|
|
// d3 = 30 70 31 71 32 72 33 73
|
|
"vld4.8 {d0, d1, d2, d3}, [%0]! \n"
|
|
"vld4.8 {d4, d5, d6, d7}, [%3]! \n"
|
|
"subs %2, %2, #12 \n"
|
|
|
|
// Shuffle the input data around to get align the data
|
|
// so adjacent data can be added. 0,1 - 2,3 - 4,5 - 6,7
|
|
// d0 = 00 10 01 11 02 12 03 13
|
|
// d1 = 40 50 41 51 42 52 43 53
|
|
"vtrn.u8 d0, d1 \n"
|
|
"vtrn.u8 d4, d5 \n"
|
|
|
|
// d2 = 20 30 21 31 22 32 23 33
|
|
// d3 = 60 70 61 71 62 72 63 73
|
|
"vtrn.u8 d2, d3 \n"
|
|
"vtrn.u8 d6, d7 \n"
|
|
|
|
// d0 = 00+10 01+11 02+12 03+13
|
|
// d2 = 40+50 41+51 42+52 43+53
|
|
"vpaddl.u8 q0, q0 \n"
|
|
"vpaddl.u8 q2, q2 \n"
|
|
|
|
// d3 = 60+70 61+71 62+72 63+73
|
|
"vpaddl.u8 d3, d3 \n"
|
|
"vpaddl.u8 d7, d7 \n"
|
|
|
|
// combine source lines
|
|
"vadd.u16 q0, q2 \n"
|
|
"vadd.u16 d4, d3, d7 \n"
|
|
|
|
// dst_ptr[3] = (s[6] + s[7] + s[6+st] + s[7+st]) / 4
|
|
"vqrshrn.u16 d4, q2, #2 \n"
|
|
|
|
// Shuffle 2,3 reg around so that 2 can be added to the
|
|
// 0,1 reg and 3 can be added to the 4,5 reg. This
|
|
// requires expanding from u8 to u16 as the 0,1 and 4,5
|
|
// registers are already expanded. Then do transposes
|
|
// to get aligned.
|
|
// q2 = xx 20 xx 30 xx 21 xx 31 xx 22 xx 32 xx 23 xx 33
|
|
"vmovl.u8 q1, d2 \n"
|
|
"vmovl.u8 q3, d6 \n"
|
|
|
|
// combine source lines
|
|
"vadd.u16 q1, q3 \n"
|
|
|
|
// d4 = xx 20 xx 30 xx 22 xx 32
|
|
// d5 = xx 21 xx 31 xx 23 xx 33
|
|
"vtrn.u32 d2, d3 \n"
|
|
|
|
// d4 = xx 20 xx 21 xx 22 xx 23
|
|
// d5 = xx 30 xx 31 xx 32 xx 33
|
|
"vtrn.u16 d2, d3 \n"
|
|
|
|
// 0+1+2, 3+4+5
|
|
"vadd.u16 q0, q1 \n"
|
|
|
|
// Need to divide, but can't downshift as the the value
|
|
// isn't a power of 2. So multiply by 65536 / n
|
|
// and take the upper 16 bits.
|
|
"vqrdmulh.s16 q0, q0, q13 \n"
|
|
|
|
// Align for table lookup, vtbl requires registers to
|
|
// be adjacent
|
|
"vmov.u8 d2, d4 \n"
|
|
|
|
"vtbl.u8 d3, {d0, d1, d2}, d28 \n"
|
|
"vtbl.u8 d4, {d0, d1, d2}, d29 \n"
|
|
|
|
"vst1.8 {d3}, [%1]! \n"
|
|
"vst1.32 {d4[0]}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst_ptr), // %1
|
|
"+r"(dst_width), // %2
|
|
"+r"(src_stride) // %3
|
|
: "r"(&kMult38_Div6), // %4
|
|
"r"(&kShuf38_2) // %5
|
|
: "q0", "q1", "q2", "q3", "q13", "q14", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
// 16x2 -> 16x1
|
|
void ScaleFilterRows_NEON(uint8* dst_ptr,
|
|
const uint8* src_ptr, ptrdiff_t src_stride,
|
|
int dst_width, int source_y_fraction) {
|
|
asm volatile (
|
|
"cmp %4, #0 \n"
|
|
"beq 100f \n"
|
|
"add %2, %1 \n"
|
|
"cmp %4, #64 \n"
|
|
"beq 75f \n"
|
|
"cmp %4, #128 \n"
|
|
"beq 50f \n"
|
|
"cmp %4, #192 \n"
|
|
"beq 25f \n"
|
|
|
|
"vdup.8 d5, %4 \n"
|
|
"rsb %4, #256 \n"
|
|
"vdup.8 d4, %4 \n"
|
|
// General purpose row blend.
|
|
"1: \n"
|
|
"vld1.8 {q0}, [%1]! \n"
|
|
"vld1.8 {q1}, [%2]! \n"
|
|
"subs %3, %3, #16 \n"
|
|
"vmull.u8 q13, d0, d4 \n"
|
|
"vmull.u8 q14, d1, d4 \n"
|
|
"vmlal.u8 q13, d2, d5 \n"
|
|
"vmlal.u8 q14, d3, d5 \n"
|
|
"vrshrn.u16 d0, q13, #8 \n"
|
|
"vrshrn.u16 d1, q14, #8 \n"
|
|
"vst1.8 {q0}, [%0]! \n"
|
|
"bgt 1b \n"
|
|
"b 99f \n"
|
|
|
|
// Blend 25 / 75.
|
|
"25: \n"
|
|
"vld1.8 {q0}, [%1]! \n"
|
|
"vld1.8 {q1}, [%2]! \n"
|
|
"subs %3, %3, #16 \n"
|
|
"vrhadd.u8 q0, q1 \n"
|
|
"vrhadd.u8 q0, q1 \n"
|
|
"vst1.8 {q0}, [%0]! \n"
|
|
"bgt 25b \n"
|
|
"b 99f \n"
|
|
|
|
// Blend 50 / 50.
|
|
"50: \n"
|
|
"vld1.8 {q0}, [%1]! \n"
|
|
"vld1.8 {q1}, [%2]! \n"
|
|
"subs %3, %3, #16 \n"
|
|
"vrhadd.u8 q0, q1 \n"
|
|
"vst1.8 {q0}, [%0]! \n"
|
|
"bgt 50b \n"
|
|
"b 99f \n"
|
|
|
|
// Blend 75 / 25.
|
|
"75: \n"
|
|
"vld1.8 {q1}, [%1]! \n"
|
|
"vld1.8 {q0}, [%2]! \n"
|
|
"subs %3, %3, #16 \n"
|
|
"vrhadd.u8 q0, q1 \n"
|
|
"vrhadd.u8 q0, q1 \n"
|
|
"vst1.8 {q0}, [%0]! \n"
|
|
"bgt 75b \n"
|
|
"b 99f \n"
|
|
|
|
// Blend 100 / 0 - Copy row unchanged.
|
|
"100: \n"
|
|
"vld1.8 {q0}, [%1]! \n"
|
|
"subs %3, %3, #16 \n"
|
|
"vst1.8 {q0}, [%0]! \n"
|
|
"bgt 100b \n"
|
|
|
|
"99: \n"
|
|
"vst1.8 {d1[7]}, [%0] \n"
|
|
: "+r"(dst_ptr), // %0
|
|
"+r"(src_ptr), // %1
|
|
"+r"(src_stride), // %2
|
|
"+r"(dst_width), // %3
|
|
"+r"(source_y_fraction) // %4
|
|
:
|
|
: "q0", "q1", "d4", "d5", "q13", "q14", "memory", "cc"
|
|
);
|
|
}
|
|
|
|
void ScaleARGBRowDown2_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
|
|
uint8* dst, int dst_width) {
|
|
asm volatile (
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
// load even pixels into q0, odd into q1
|
|
"vld2.32 {q0, q1}, [%0]! \n"
|
|
"vld2.32 {q2, q3}, [%0]! \n"
|
|
"subs %2, %2, #8 \n" // 8 processed per loop
|
|
"vst1.8 {q1}, [%1]! \n" // store odd pixels
|
|
"vst1.8 {q3}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(dst), // %1
|
|
"+r"(dst_width) // %2
|
|
:
|
|
: "memory", "cc", "q0", "q1", "q2", "q3" // Clobber List
|
|
);
|
|
}
|
|
|
|
void ScaleARGBRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
|
|
uint8* dst, int dst_width) {
|
|
asm volatile (
|
|
// change the stride to row 2 pointer
|
|
"add %1, %1, %0 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld4.8 {d0, d2, d4, d6}, [%0]! \n" // load 8 ARGB pixels.
|
|
"vld4.8 {d1, d3, d5, d7}, [%0]! \n" // load next 8 ARGB pixels.
|
|
"subs %3, %3, #8 \n" // 8 processed per loop.
|
|
"vpaddl.u8 q0, q0 \n" // B 16 bytes -> 8 shorts.
|
|
"vpaddl.u8 q1, q1 \n" // G 16 bytes -> 8 shorts.
|
|
"vpaddl.u8 q2, q2 \n" // R 16 bytes -> 8 shorts.
|
|
"vpaddl.u8 q3, q3 \n" // A 16 bytes -> 8 shorts.
|
|
"vld4.8 {d16, d18, d20, d22}, [%1]! \n" // load 8 more ARGB pixels.
|
|
"vld4.8 {d17, d19, d21, d23}, [%1]! \n" // load last 8 ARGB pixels.
|
|
"vpadal.u8 q0, q8 \n" // B 16 bytes -> 8 shorts.
|
|
"vpadal.u8 q1, q9 \n" // G 16 bytes -> 8 shorts.
|
|
"vpadal.u8 q2, q10 \n" // R 16 bytes -> 8 shorts.
|
|
"vpadal.u8 q3, q11 \n" // A 16 bytes -> 8 shorts.
|
|
"vrshrn.u16 d0, q0, #2 \n" // downshift, round and pack
|
|
"vrshrn.u16 d1, q1, #2 \n"
|
|
"vrshrn.u16 d2, q2, #2 \n"
|
|
"vrshrn.u16 d3, q3, #2 \n"
|
|
"vst4.8 {d0, d1, d2, d3}, [%2]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_ptr), // %0
|
|
"+r"(src_stride), // %1
|
|
"+r"(dst), // %2
|
|
"+r"(dst_width) // %3
|
|
:
|
|
: "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11"
|
|
);
|
|
}
|
|
|
|
// Reads 4 pixels at a time.
|
|
// Alignment requirement: src_argb 4 byte aligned.
|
|
void ScaleARGBRowDownEven_NEON(const uint8* src_argb, ptrdiff_t src_stride,
|
|
int src_stepx, uint8* dst_argb, int dst_width) {
|
|
asm volatile (
|
|
"mov r12, %3, lsl #2 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld1.32 {d0[0]}, [%0], r12 \n"
|
|
"vld1.32 {d0[1]}, [%0], r12 \n"
|
|
"vld1.32 {d1[0]}, [%0], r12 \n"
|
|
"vld1.32 {d1[1]}, [%0], r12 \n"
|
|
"subs %2, %2, #4 \n" // 4 pixels per loop.
|
|
"vst1.8 {q0}, [%1]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_argb), // %0
|
|
"+r"(dst_argb), // %1
|
|
"+r"(dst_width) // %2
|
|
: "r"(src_stepx) // %3
|
|
: "memory", "cc", "r12", "q0"
|
|
);
|
|
}
|
|
|
|
// Reads 4 pixels at a time.
|
|
// Alignment requirement: src_argb 4 byte aligned.
|
|
void ScaleARGBRowDownEvenBox_NEON(const uint8* src_argb, ptrdiff_t src_stride,
|
|
int src_stepx,
|
|
uint8* dst_argb, int dst_width) {
|
|
asm volatile (
|
|
"mov r12, %4, lsl #2 \n"
|
|
"add %1, %1, %0 \n"
|
|
".p2align 2 \n"
|
|
"1: \n"
|
|
"vld1.8 {d0}, [%0], r12 \n" // Read 4 2x2 blocks -> 2x1
|
|
"vld1.8 {d1}, [%1], r12 \n"
|
|
"vld1.8 {d2}, [%0], r12 \n"
|
|
"vld1.8 {d3}, [%1], r12 \n"
|
|
"vld1.8 {d4}, [%0], r12 \n"
|
|
"vld1.8 {d5}, [%1], r12 \n"
|
|
"vld1.8 {d6}, [%0], r12 \n"
|
|
"vld1.8 {d7}, [%1], r12 \n"
|
|
"vaddl.u8 q0, d0, d1 \n"
|
|
"vaddl.u8 q1, d2, d3 \n"
|
|
"vaddl.u8 q2, d4, d5 \n"
|
|
"vaddl.u8 q3, d6, d7 \n"
|
|
"vswp.8 d1, d2 \n" // ab_cd -> ac_bd
|
|
"vswp.8 d5, d6 \n" // ef_gh -> eg_fh
|
|
"vadd.u16 q0, q0, q1 \n" // (a+b)_(c+d)
|
|
"vadd.u16 q2, q2, q3 \n" // (e+f)_(g+h)
|
|
"vrshrn.u16 d0, q0, #2 \n" // first 2 pixels.
|
|
"vrshrn.u16 d1, q2, #2 \n" // next 2 pixels.
|
|
"subs %3, %3, #4 \n" // 4 pixels per loop.
|
|
"vst1.8 {q0}, [%2]! \n"
|
|
"bgt 1b \n"
|
|
: "+r"(src_argb), // %0
|
|
"+r"(src_stride), // %1
|
|
"+r"(dst_argb), // %2
|
|
"+r"(dst_width) // %3
|
|
: "r"(src_stepx) // %4
|
|
: "memory", "cc", "r12", "q0", "q1", "q2", "q3"
|
|
);
|
|
}
|
|
|
|
#endif // __ARM_NEON__
|
|
|
|
#ifdef __cplusplus
|
|
} // extern "C"
|
|
} // namespace libyuv
|
|
#endif
|