/* * Copyright (c) 2016, 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 <assert.h> #include <smmintrin.h> /* SSE4.1 */ #include "config/aom_config.h" #include "config/av1_rtcd.h" #include "av1/common/av1_inv_txfm1d_cfg.h" #include "av1/common/idct.h" #include "av1/common/x86/av1_inv_txfm_ssse3.h" #include "av1/common/x86/av1_txfm_sse2.h" #include "av1/common/x86/av1_txfm_sse4.h" #include "av1/common/x86/highbd_txfm_utility_sse4.h" static inline __m128i highbd_clamp_epi16(__m128i u, int bd) { … } static inline void round_shift_4x4(__m128i *in, int shift) { … } static void round_shift_8x8(__m128i *in, int shift) { … } static void highbd_clamp_epi32_sse4_1(__m128i *in, __m128i *out, const __m128i *clamp_lo, const __m128i *clamp_hi, int size) { … } static inline __m128i highbd_get_recon_8x8_sse4_1(const __m128i pred, __m128i res0, __m128i res1, const int bd) { … } static inline __m128i highbd_get_recon_4xn_sse4_1(const __m128i pred, __m128i res0, const int bd) { … } static inline void highbd_write_buffer_4xn_sse4_1(__m128i *in, uint16_t *output, int stride, int flipud, int height, const int bd) { … } static inline void highbd_write_buffer_8xn_sse4_1(__m128i *in, uint16_t *output, int stride, int flipud, int height, const int bd) { … } static inline void load_buffer_32bit_input(const int32_t *in, int stride, __m128i *out, int out_size) { … } static inline void load_buffer_4x4(const int32_t *coeff, __m128i *in) { … } void av1_highbd_iwht4x4_16_add_sse4_1(const tran_low_t *input, uint8_t *dest8, int stride, int bd) { … } static void addsub_sse4_1(const __m128i in0, const __m128i in1, __m128i *out0, __m128i *out1, const __m128i *clamp_lo, const __m128i *clamp_hi) { … } static void shift_and_clamp_sse4_1(__m128i *in0, __m128i *in1, const __m128i *clamp_lo, const __m128i *clamp_hi, int shift) { … } static inline void idct32_stage4_sse4_1( __m128i *bf1, const __m128i *cospim8, const __m128i *cospi56, const __m128i *cospi8, const __m128i *cospim56, const __m128i *cospim40, const __m128i *cospi24, const __m128i *cospi40, const __m128i *cospim24, const __m128i *rounding, int bit) { … } static inline void idct32_stage5_sse4_1( __m128i *bf1, const __m128i *cospim16, const __m128i *cospi48, const __m128i *cospi16, const __m128i *cospim48, const __m128i *clamp_lo, const __m128i *clamp_hi, const __m128i *rounding, int bit) { … } static inline void idct32_stage6_sse4_1( __m128i *bf1, const __m128i *cospim32, const __m128i *cospi32, const __m128i *cospim16, const __m128i *cospi48, const __m128i *cospi16, const __m128i *cospim48, const __m128i *clamp_lo, const __m128i *clamp_hi, const __m128i *rounding, int bit) { … } static inline void idct32_stage7_sse4_1(__m128i *bf1, const __m128i *cospim32, const __m128i *cospi32, const __m128i *clamp_lo, const __m128i *clamp_hi, const __m128i *rounding, int bit) { … } static inline void idct32_stage8_sse4_1(__m128i *bf1, const __m128i *cospim32, const __m128i *cospi32, const __m128i *clamp_lo, const __m128i *clamp_hi, const __m128i *rounding, int bit) { … } static inline void idct32_stage9_sse4_1(__m128i *bf1, __m128i *out, const int do_cols, const int bd, const int out_shift, const __m128i *clamp_lo, const __m128i *clamp_hi) { … } static void neg_shift_sse4_1(const __m128i in0, const __m128i in1, __m128i *out0, __m128i *out1, const __m128i *clamp_lo, const __m128i *clamp_hi, int shift) { … } static void idct4x4_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iadst4x4_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void write_buffer_4x4(__m128i *in, uint16_t *output, int stride, int fliplr, int flipud, int shift, int bd) { … } static void iidentity4_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } void av1_inv_txfm2d_add_4x4_sse4_1(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd) { … } // 8x8 static void load_buffer_8x8(const int32_t *coeff, __m128i *in) { … } static void idct8x8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iadst8x8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iidentity8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static __m128i get_recon_8x8(const __m128i pred, __m128i res_lo, __m128i res_hi, int fliplr, int bd) { … } static void write_buffer_8x8(__m128i *in, uint16_t *output, int stride, int fliplr, int flipud, int shift, int bd) { … } void av1_inv_txfm2d_add_8x8_sse4_1(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd) { … } static void idct8x8_low1_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct8x8_new_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iadst8x8_low1_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iadst8x8_new_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct16x16_low1_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct16x16_low8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iadst16x16_low1_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iadst16x16_low8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct16x16_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iadst16x16_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void iidentity16_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static inline void idct64_stage8_sse4_1( __m128i *u, const __m128i *cospim32, const __m128i *cospi32, const __m128i *cospim16, const __m128i *cospi48, const __m128i *cospi16, const __m128i *cospim48, const __m128i *clamp_lo, const __m128i *clamp_hi, const __m128i *rnding, int bit) { … } static inline void idct64_stage9_sse4_1(__m128i *u, const __m128i *cospim32, const __m128i *cospi32, const __m128i *clamp_lo, const __m128i *clamp_hi, const __m128i *rnding, int bit) { … } static inline void idct64_stage10_sse4_1(__m128i *u, const __m128i *cospim32, const __m128i *cospi32, const __m128i *clamp_lo, const __m128i *clamp_hi, const __m128i *rnding, int bit) { … } static inline void idct64_stage11_sse4_1(__m128i *u, __m128i *out, int do_cols, int bd, int out_shift, const __m128i *clamp_lo, const __m128i *clamp_hi) { … } static void idct64x64_low1_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct64x64_low8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct64x64_low16_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct64x64_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct32x32_low1_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct32x32_low8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct32x32_low16_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void idct32x32_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param) { … } static void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param) { … } static void iidentity32_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols, int bd, int out_shift) { … } static const transform_1d_sse4_1 highbd_txfm_all_1d_zeros_w8_arr[TX_SIZES][ITX_TYPES_1D][4] = …; static void highbd_inv_txfm2d_add_h_identity_ssse41(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } static void highbd_inv_txfm2d_add_v_identity_ssse41(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } static void highbd_inv_txfm2d_add_idtx_ssse41(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } static void highbd_inv_txfm2d_add_no_identity_sse41(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } static void highbd_inv_txfm2d_add_4x8_sse41(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } static void highbd_inv_txfm2d_add_8x4_sse41(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } static void highbd_inv_txfm2d_add_4x16_sse4_1(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } static void highbd_inv_txfm2d_add_16x4_sse4_1(const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } void av1_highbd_inv_txfm2d_add_universe_sse4_1(const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type, TX_SIZE tx_size, int eob, const int bd) { … } static void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param) { … } static void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param) { … } static void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param) { … } static void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param) { … } void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param) { … }
Codebase Browser
chromium