/* * Copyright (c) 2014 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>#include "./vpx_dsp_rtcd.h" #include "vpx/vpx_integer.h" #include "vpx_dsp/x86/bitdepth_conversion_sse2.h" #include "vpx_ports/mem.h" static INLINE void sign_extend_16bit_to_32bit_sse2(__m128i in, __m128i zero,const __m128i sign_bits = _mm_cmplt_epi16(in, zero);void vpx_minmax_8x8_sse2(const uint8_t *s, int p, const uint8_t *d, int dp,int *min, int *max) {// Row 0 const __m128i *)(s)), u0);const __m128i *)(d)), u0);// Row 1 const __m128i *)(s + p)), u0);const __m128i *)(d + dp)), u0);// Row 2 const __m128i *)(s + 2 * p)), u0);const __m128i *)(d + 2 * dp)), u0);// Row 3 const __m128i *)(s + 3 * p)), u0);const __m128i *)(d + 3 * dp)), u0);// Row 4 const __m128i *)(s + 4 * p)), u0);const __m128i *)(d + 4 * dp)), u0);// Row 5 const __m128i *)(s + 5 * p)), u0);const __m128i *)(d + 5 * dp)), u0);// Row 6 const __m128i *)(s + 6 * p)), u0);const __m128i *)(d + 6 * dp)), u0);// Row 7 const __m128i *)(s + 7 * p)), u0);const __m128i *)(d + 7 * dp)), u0);unsigned int vpx_avg_8x8_sse2(const uint8_t *s, int p) {unsigned int avg = 0;const __m128i *)(s)), u0);const __m128i *)(s + p)), u0);const __m128i *)(s + 2 * p)), u0);const __m128i *)(s + 3 * p)), u0);const __m128i *)(s + 4 * p)), u0);const __m128i *)(s + 5 * p)), u0);const __m128i *)(s + 6 * p)), u0);const __m128i *)(s + 7 * p)), u0);return (avg + 32) >> 6;unsigned int vpx_avg_4x4_sse2(const uint8_t *s, int p) {unsigned int avg = 0;const __m128i *)(s)), u0);const __m128i *)(s + p)), u0);const __m128i *)(s + 2 * p)), u0);const __m128i *)(s + 3 * p)), u0);return (avg + 8) >> 4;#if CONFIG_VP9_HIGHBITDEPTHunsigned int vpx_highbd_avg_8x8_sse2(const uint8_t *s8, int p) {unsigned int avg;const uint16_t *s = CONVERT_TO_SHORTPTR(s8);const __m128i zero = _mm_setzero_si128();const __m128i *)(s));const __m128i *)(s + p));const __m128i *)(s + 2 * p));const __m128i *)(s + 3 * p));const __m128i *)(s + 4 * p));const __m128i *)(s + 5 * p));const __m128i *)(s + 6 * p));const __m128i *)(s + 7 * p));unsigned int )_mm_cvtsi128_si32(s0);return (avg + 32) >> 6;unsigned int vpx_highbd_avg_4x4_sse2(const uint8_t *s8, int p) {unsigned int avg;const uint16_t *s = CONVERT_TO_SHORTPTR(s8);const __m128i *)(s));const __m128i *)(s + p));const __m128i *)(s + 2 * p));const __m128i *)(s + 3 * p));return (avg + 8) >> 4;#endif // CONFIG_VP9_HIGHBITDEPTH static void hadamard_col8_sse2(__m128i *in, int iter) {if (iter == 0) {else {static INLINE void hadamard_8x8_sse2(const int16_t *src_diff,int is_final) {const __m128i *)src_diff);const __m128i *)(src_diff += src_stride));const __m128i *)(src_diff += src_stride));const __m128i *)(src_diff += src_stride));const __m128i *)(src_diff += src_stride));const __m128i *)(src_diff += src_stride));const __m128i *)(src_diff += src_stride));const __m128i *)(src_diff + src_stride));if (is_final) {else {void vpx_hadamard_8x8_sse2(const int16_t *src_diff, ptrdiff_t src_stride,static INLINE void hadamard_16x16_sse2(const int16_t *src_diff,int is_final) {#if CONFIG_VP9_HIGHBITDEPTH// For high bitdepths, it is unnecessary to store_tran_low // (mult/unpack/store), then load_tran_low (load/pack) the same memory in the // next stage. Output to an intermediate buffer first, then store_tran_low() // in the final stage. #else #endif int idx;for (idx = 0; idx < 4; ++idx) {const int16_t *src_ptr =for (idx = 0; idx < 64; idx += 8) {const __m128i *)t_coeff);const __m128i *)(t_coeff + 64));const __m128i *)(t_coeff + 128));const __m128i *)(t_coeff + 192));if (is_final) {else {void vpx_hadamard_16x16_sse2(const int16_t *src_diff, ptrdiff_t src_stride,void vpx_hadamard_32x32_sse2(const int16_t *src_diff, ptrdiff_t src_stride,#if CONFIG_VP9_HIGHBITDEPTH// For high bitdepths, it is unnecessary to store_tran_low // (mult/unpack/store), then load_tran_low (load/pack) the same memory in the // next stage. Output to an intermediate buffer first, then store_tran_low() // in the final stage. #else #endif int idx;const __m128i zero = _mm_setzero_si128();for (idx = 0; idx < 4; ++idx) {const int16_t *src_ptr =for (idx = 0; idx < 256; idx += 8) {const __m128i *)t_coeff);const __m128i *)(t_coeff + 256));const __m128i *)(t_coeff + 512));const __m128i *)(t_coeff + 768));// Sign extend 16 bit to 32 bit. int vpx_satd_sse2(const tran_low_t *coeff, int length) {int i;const __m128i zero = _mm_setzero_si128();for (i = 0; i < length; i += 8) {const __m128i src_line = load_tran_low(coeff);const __m128i inv = _mm_sub_epi16(zero, src_line);const __m128i abs = _mm_max_epi16(src_line, inv); // abs(src_line) const __m128i abs_lo = _mm_unpacklo_epi16(abs, zero);const __m128i abs_hi = _mm_unpackhi_epi16(abs, zero);const __m128i sum = _mm_add_epi32(abs_lo, abs_hi);// cascading summation of accum return _mm_cvtsi128_si32(accum);void vpx_int_pro_row_sse2(int16_t hbuf[16], const uint8_t *ref,const int ref_stride, const int height) {int idx;const __m128i *)ref);int height_1 = height - 1;for (idx = 1; idx < height_1; idx += 2) {const __m128i *)ref);const __m128i *)ref);const __m128i *)ref);if (height == 64) {else if (height == 32) {else {const uint8_t *ref, const int width) {const __m128i *)ref);int i;for (i = 16; i < width; i += 16) {const __m128i *)ref);return _mm_extract_epi16(s0, 0);int vpx_vector_var_sse2(const int16_t *ref, const int16_t *src, const int bwl) {int idx;int width = 4 << bwl;const __m128i *)ref);const __m128i *)src);for (idx = 8; idx < width; idx += 8) {const __m128i *)ref);const __m128i *)src);return _mm_cvtsi128_si32(sse) - ((mean * mean) >> (bwl + 2));
Messung V0.5 C=96 H=95 G=95
¤ Dauer der Verarbeitung: 0.14 Sekunden
(vorverarbeitet)
¤ *© Formatika GbR, Deutschland
