/* * Copyright (c) 2017, Alliance for Open Media. All rights reserved. * * This source code is subject to the terms of the BSD 2 Clause License and * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License * was not distributed with this source code in the LICENSE file, you can * obtain it at www.aomedia.org/license/software. If the Alliance for Open * Media Patent License 1.0 was not distributed with this source code in the * PATENTS file, you can obtain it at www.aomedia.org/license/patent. */ #include <tmmintrin.h> #include "config/aom_dsp_rtcd.h" #include "aom_dsp/intrapred_common.h" // ----------------------------------------------------------------------------- // PAETH_PRED // Return 8 16-bit pixels in one row static inline __m128i paeth_8x1_pred(const __m128i *left, const __m128i *top, const __m128i *topleft) { … } void aom_paeth_predictor_4x4_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_4x8_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_4x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_8x4_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_8x8_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_8x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_8x32_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } // Return 16 8-bit pixels in one row static inline __m128i paeth_16x1_pred(const __m128i *left, const __m128i *top0, const __m128i *top1, const __m128i *topleft) { … } void aom_paeth_predictor_16x4_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_16x8_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_16x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_16x32_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_16x64_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_32x8_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_32x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_32x32_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_32x64_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_64x32_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_64x64_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_paeth_predictor_64x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } // ----------------------------------------------------------------------------- // SMOOTH_PRED // pixels[0]: above and below_pred interleave vector // pixels[1]: left vector // pixels[2]: right_pred vector static inline void load_pixel_w4(const uint8_t *above, const uint8_t *left, int height, __m128i *pixels) { … } // weight_h[0]: weight_h vector // weight_h[1]: scale - weight_h vector // weight_h[2]: same as [0], second half for height = 16 only // weight_h[3]: same as [1], second half for height = 16 only // weight_w[0]: weights_w and scale - weights_w interleave vector static inline void load_weight_w4(int height, __m128i *weight_h, __m128i *weight_w) { … } static inline void smooth_pred_4xh(const __m128i *pixel, const __m128i *wh, const __m128i *ww, int h, uint8_t *dst, ptrdiff_t stride, int second_half) { … } void aom_smooth_predictor_4x4_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_4x8_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_4x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } // pixels[0]: above and below_pred interleave vector, first half // pixels[1]: above and below_pred interleave vector, second half // pixels[2]: left vector // pixels[3]: right_pred vector // pixels[4]: above and below_pred interleave vector, first half // pixels[5]: above and below_pred interleave vector, second half // pixels[6]: left vector + 16 // pixels[7]: right_pred vector static inline void load_pixel_w8(const uint8_t *above, const uint8_t *left, int height, __m128i *pixels) { … } // weight_h[0]: weight_h vector // weight_h[1]: scale - weight_h vector // weight_h[2]: same as [0], offset 8 // weight_h[3]: same as [1], offset 8 // weight_h[4]: same as [0], offset 16 // weight_h[5]: same as [1], offset 16 // weight_h[6]: same as [0], offset 24 // weight_h[7]: same as [1], offset 24 // weight_w[0]: weights_w and scale - weights_w interleave vector, first half // weight_w[1]: weights_w and scale - weights_w interleave vector, second half static inline void load_weight_w8(int height, __m128i *weight_h, __m128i *weight_w) { … } static inline void smooth_pred_8xh(const __m128i *pixels, const __m128i *wh, const __m128i *ww, int h, uint8_t *dst, ptrdiff_t stride, int second_half) { … } void aom_smooth_predictor_8x4_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_8x8_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_8x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_8x32_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } // TODO(slavarnway): Visual Studio only supports restrict when /std:c11 // (available in 2019+) or greater is specified; __restrict can be used in that // case. This should be moved to rtcd and used consistently between the // function declarations and definitions to avoid warnings in Visual Studio // when defining LIBAOM_RESTRICT to restrict or __restrict. #if defined(_MSC_VER) #define LIBAOM_RESTRICT #else #define LIBAOM_RESTRICT … #endif static AOM_FORCE_INLINE __m128i Load4(const void *src) { … } static AOM_FORCE_INLINE __m128i LoadLo8(const void *a) { … } static AOM_FORCE_INLINE __m128i LoadUnaligned16(const void *a) { … } static AOM_FORCE_INLINE void Store4(void *dst, const __m128i x) { … } static AOM_FORCE_INLINE void StoreLo8(void *a, const __m128i v) { … } static AOM_FORCE_INLINE void StoreUnaligned16(void *a, const __m128i v) { … } static AOM_FORCE_INLINE __m128i cvtepu8_epi16(__m128i x) { … } static AOM_FORCE_INLINE __m128i cvtepu8_epi32(__m128i x) { … } static AOM_FORCE_INLINE __m128i cvtepu16_epi32(__m128i x) { … } static void smooth_predictor_wxh(uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column, int width, int height) { … } void aom_smooth_predictor_16x4_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_16x8_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_16x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_16x32_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_16x64_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_32x8_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_32x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_32x32_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_32x64_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_64x16_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_64x32_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } void aom_smooth_predictor_64x64_ssse3(uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left) { … } // ----------------------------------------------------------------------------- // Smooth horizontal/vertical helper functions. // For Horizontal, pixels1 and pixels2 are the same repeated value. For // Vertical, weights1 and weights2 are the same, and scaled_corner1 and // scaled_corner2 are the same. static AOM_FORCE_INLINE void write_smooth_directional_sum16( uint8_t *LIBAOM_RESTRICT dst, const __m128i pixels1, const __m128i pixels2, const __m128i weights1, const __m128i weights2, const __m128i scaled_corner1, const __m128i scaled_corner2, const __m128i round) { … } static AOM_FORCE_INLINE __m128i smooth_directional_sum8( const __m128i pixels, const __m128i weights, const __m128i scaled_corner) { … } static AOM_FORCE_INLINE void write_smooth_directional_sum8( uint8_t *LIBAOM_RESTRICT dst, const __m128i *pixels, const __m128i *weights, const __m128i *scaled_corner, const __m128i *round) { … } // ----------------------------------------------------------------------------- // SMOOTH_V_PRED static AOM_FORCE_INLINE void load_smooth_vertical_pixels4( const uint8_t *LIBAOM_RESTRICT above, const uint8_t *LIBAOM_RESTRICT left, const int height, __m128i *pixels) { … } // |weight_array| alternates weight vectors from the table with their inverted // (256-w) counterparts. This is precomputed by the compiler when the weights // table is visible to this module. Removing this visibility can cut speed by up // to half in both 4xH and 8xH transforms. static AOM_FORCE_INLINE void load_smooth_vertical_weights4( const uint8_t *LIBAOM_RESTRICT weight_array, const int height, __m128i *weights) { … } static AOM_FORCE_INLINE void write_smooth_vertical4xh( const __m128i *pixel, const __m128i *weight, const int height, uint8_t *LIBAOM_RESTRICT dst, const ptrdiff_t stride) { … } void aom_smooth_v_predictor_4x4_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_4x8_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_4x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_8x4_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_8x8_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_8x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_8x32_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_16x4_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_16x8_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_16x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_16x32_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_16x64_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_32x8_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_32x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_32x32_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_32x64_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_64x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_64x32_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_v_predictor_64x64_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } // ----------------------------------------------------------------------------- // SMOOTH_H_PRED static AOM_FORCE_INLINE void write_smooth_horizontal_sum4( uint8_t *LIBAOM_RESTRICT dst, const __m128i *left_y, const __m128i *weights, const __m128i *scaled_top_right, const __m128i *round) { … } void aom_smooth_h_predictor_4x4_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_4x8_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_4x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } // For SMOOTH_H, |pixels| is the repeated left value for the row. For SMOOTH_V, // |pixels| is a segment of the top row or the whole top row, and |weights| is // repeated. void aom_smooth_h_predictor_8x4_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_8x8_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_8x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_8x32_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_16x4_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_16x8_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_16x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_16x32_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_16x64_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_32x8_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_32x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_32x32_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_32x64_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_64x16_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_64x32_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … } void aom_smooth_h_predictor_64x64_ssse3( uint8_t *LIBAOM_RESTRICT dst, ptrdiff_t stride, const uint8_t *LIBAOM_RESTRICT top_row, const uint8_t *LIBAOM_RESTRICT left_column) { … }
Codebase Browser
chromium