/* * Copyright (c) 2018, 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 "config/aom_config.h" #include "config/av1_rtcd.h" #include "av1/common/av1_inv_txfm1d_cfg.h" #include "av1/common/x86/av1_txfm_sse2.h" #include "av1/common/x86/av1_inv_txfm_avx2.h" #include "av1/common/x86/av1_inv_txfm_ssse3.h" // TODO([email protected]): move this to header file // Sqrt2, Sqrt2^2, Sqrt2^3, Sqrt2^4, Sqrt2^5 static int32_t NewSqrt2list[TX_SIZES] = …; static inline void idct16_stage5_avx2(__m256i *x1, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct16_stage6_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct16_stage7_avx2(__m256i *output, __m256i *x1) { … } static void idct16_avx2(const __m256i *input, __m256i *output) { … } static void idct16_low8_avx2(const __m256i *input, __m256i *output) { … } static void idct16_low1_avx2(const __m256i *input, __m256i *output) { … } static inline void iadst16_stage3_avx2(__m256i *x) { … } static inline void iadst16_stage4_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void iadst16_stage5_avx2(__m256i *x) { … } static inline void iadst16_stage6_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void iadst16_stage7_avx2(__m256i *x) { … } static inline void iadst16_stage8_avx2(__m256i *x1, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void iadst16_stage9_avx2(__m256i *output, __m256i *x1) { … } static void iadst16_avx2(const __m256i *input, __m256i *output) { … } static void iadst16_low8_avx2(const __m256i *input, __m256i *output) { … } static void iadst16_low1_avx2(const __m256i *input, __m256i *output) { … } static inline void idct32_high16_stage3_avx2(__m256i *x) { … } static inline void idct32_high16_stage4_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct32_high24_stage5_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct32_high28_stage6_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct32_stage7_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct32_stage8_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct32_stage9_avx2(__m256i *output, __m256i *x) { … } static void idct32_low1_avx2(const __m256i *input, __m256i *output) { … } static void idct32_low8_avx2(const __m256i *input, __m256i *output) { … } static void idct32_low16_avx2(const __m256i *input, __m256i *output) { … } static void idct32_avx2(const __m256i *input, __m256i *output) { … } static inline void idct64_stage4_high32_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct64_stage5_high48_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct64_stage6_high32_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct64_stage6_high48_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct64_stage7_high48_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct64_stage8_high48_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct64_stage9_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct64_stage10_avx2(__m256i *x, const int32_t *cospi, const __m256i _r, int8_t cos_bit) { … } static inline void idct64_stage11_avx2(__m256i *output, __m256i *x) { … } static void idct64_low1_avx2(const __m256i *input, __m256i *output) { … } static void idct64_low8_avx2(const __m256i *input, __m256i *output) { … } static void idct64_low16_avx2(const __m256i *input, __m256i *output) { … } static void idct64_low32_avx2(const __m256i *input, __m256i *output) { … } transform_1d_avx2; // 1D functions process 16 pixels at one time. static const transform_1d_avx2 lowbd_txfm_all_1d_zeros_w16_arr[TX_SIZES][ITX_TYPES_1D][4] = …; // only process w >= 16 h >= 16 static inline void lowbd_inv_txfm2d_add_no_identity_avx2( const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob) { … } static inline void iidentity_row_16xn_avx2(__m256i *out, const int32_t *input, int stride, int shift, int height, int txw_idx, int rect_type) { … } static inline void iidentity_col_16xn_avx2(uint8_t *output, int stride, __m256i *buf, int shift, int height, int txh_idx) { … } static inline void lowbd_inv_txfm2d_add_idtx_avx2(const int32_t *input, uint8_t *output, int stride, TX_SIZE tx_size, int32_t eob) { … } static inline void lowbd_inv_txfm2d_add_h_identity_avx2( const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob) { … } static inline void lowbd_inv_txfm2d_add_v_identity_avx2( const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob) { … } static const transform_1d_ssse3 lowbd_txfm_all_1d_zeros_8x8_arr[2][2] = …; static inline void load_buffer_avx2(const int32_t *in, int stride, __m128i *out) { … } static inline void round_and_transpose_avx2(const __m128i *const in, __m128i *const out, int bit, int *lr_flip) { … } static inline void round_shift_lowbd_write_buffer_avx2(__m128i *in, int bit, uint8_t *output, int stride, int flipud) { … } // AVX2 implementation has the advantage when combined multiple operations // together. static inline void lowbd_inv_txfm2d_8x8_no_identity_avx2( const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob) { … } // AVX2 implementation of 8x8 inverse transform. Observed that coding AVX2 for // tx_type with identity in either of the direction has no advantage. static void lowbd_inv_txfm2d_add_8x8_avx2(const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob) { … } // for 32x32,32x64,64x32,64x64,16x32,32x16,64x16,16x64 static inline void lowbd_inv_txfm2d_add_universe_avx2( const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob) { … } void av1_lowbd_inv_txfm2d_add_avx2(const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob) { … } void av1_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param) { … }
Codebase Browser
chromium