chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/loopfilter_sse2.c

/*
 *  Copyright (c) 2010 The WebM 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 <emmintrin.h>  // SSE2

#include "./vpx_dsp_rtcd.h"
#include "vpx_ports/mem.h"
#include "vpx_ports/emmintrin_compat.h"
#include "vpx_dsp/x86/mem_sse2.h"

static INLINE __m128i abs_diff(__m128i a, __m128i b) { … }

// filter_mask and hev_mask
#define FILTER_HEV_MASK …

#define FILTER4 …

void vpx_lpf_horizontal_4_sse2(uint8_t *s, int pitch, const uint8_t *blimit,
                               const uint8_t *limit, const uint8_t *thresh) { … }

void vpx_lpf_vertical_4_sse2(uint8_t *s, int pitch, const uint8_t *blimit,
                             const uint8_t *limit, const uint8_t *thresh) { … }

void vpx_lpf_horizontal_16_sse2(unsigned char *s, int pitch,
                                const unsigned char *blimit,
                                const unsigned char *limit,
                                const unsigned char *thresh) { … }

static INLINE __m128i filter_add2_sub2(const __m128i *const total,
                                       const __m128i *const a1,
                                       const __m128i *const a2,
                                       const __m128i *const s1,
                                       const __m128i *const s2) { … }

static INLINE __m128i filter8_mask(const __m128i *const flat,
                                   const __m128i *const other_filt,
                                   const __m128i *const f8_lo,
                                   const __m128i *const f8_hi) { … }

static INLINE __m128i filter16_mask(const __m128i *const flat,
                                    const __m128i *const other_filt,
                                    const __m128i *const f_lo,
                                    const __m128i *const f_hi) { … }

void vpx_lpf_horizontal_16_dual_sse2(unsigned char *s, int pitch,
                                     const unsigned char *blimit,
                                     const unsigned char *limit,
                                     const unsigned char *thresh) { … }

void vpx_lpf_horizontal_8_sse2(unsigned char *s, int pitch,
                               const unsigned char *blimit,
                               const unsigned char *limit,
                               const unsigned char *thresh) { … }

void vpx_lpf_horizontal_8_dual_sse2(
    uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0,
    const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
    const uint8_t *thresh1) { … }

void vpx_lpf_horizontal_4_dual_sse2(unsigned char *s, int pitch,
                                    const unsigned char *blimit0,
                                    const unsigned char *limit0,
                                    const unsigned char *thresh0,
                                    const unsigned char *blimit1,
                                    const unsigned char *limit1,
                                    const unsigned char *thresh1) { … }

static INLINE void transpose8x16(unsigned char *in0, unsigned char *in1,
                                 int in_p, unsigned char *out, int out_p) { … }

static INLINE void transpose(unsigned char *src[], int in_p,
                             unsigned char *dst[], int out_p,
                             int num_8x8_to_transpose) { … }

void vpx_lpf_vertical_4_dual_sse2(uint8_t *s, int pitch, const uint8_t *blimit0,
                                  const uint8_t *limit0, const uint8_t *thresh0,
                                  const uint8_t *blimit1, const uint8_t *limit1,
                                  const uint8_t *thresh1) { … }

void vpx_lpf_vertical_8_sse2(unsigned char *s, int pitch,
                             const unsigned char *blimit,
                             const unsigned char *limit,
                             const unsigned char *thresh) { … }

void vpx_lpf_vertical_8_dual_sse2(uint8_t *s, int pitch, const uint8_t *blimit0,
                                  const uint8_t *limit0, const uint8_t *thresh0,
                                  const uint8_t *blimit1, const uint8_t *limit1,
                                  const uint8_t *thresh1) { … }

void vpx_lpf_vertical_16_sse2(unsigned char *s, int pitch,
                              const unsigned char *blimit,
                              const unsigned char *limit,
                              const unsigned char *thresh) { … }

void vpx_lpf_vertical_16_dual_sse2(unsigned char *s, int pitch,
                                   const uint8_t *blimit, const uint8_t *limit,
                                   const uint8_t *thresh) { … }