Line data    Source code

       1             : /*
       2             :  * Copyright (c) 2016, Alliance for Open Media. All rights reserved
       3             :  *
       4             :  * This source code is subject to the terms of the BSD 2 Clause License and
       5             :  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
       6             :  * was not distributed with this source code in the LICENSE file, you can
       7             :  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
       8             :  * Media Patent License 1.0 was not distributed with this source code in the
       9             :  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
      10             :  */
      11             : #include "EbDefinitions.h"
      12             : #include <assert.h>
      13             : #include <emmintrin.h>  // SSE2
      14             : #include "aom_dsp_rtcd.h"
      15             : #include "EbVariance_SSE2.h"
      16             : #include "synonyms.h"
      17             : 
      18             : #ifdef __cplusplus
      19             : extern "C" {
      20             : #endif
      21             : 
      22             : #ifdef __cplusplus
      23             : }
      24             : #endif
      25             : 
      26           0 : uint32_t eb_aom_get_mb_ss_sse2(const int16_t *src) {
      27           0 :     __m128i vsum = _mm_setzero_si128();
      28             :     int32_t i;
      29             : 
      30           0 :     for (i = 0; i < 32; ++i) {
      31           0 :         const __m128i v = xx_loadu_128(src);
      32           0 :         vsum = _mm_add_epi32(vsum, _mm_madd_epi16(v, v));
      33           0 :         src += 8;
      34             :     }
      35             : 
      36           0 :     vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
      37           0 :     vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
      38           0 :     return _mm_cvtsi128_si32(vsum);
      39             : }
      40             : 
      41             : // Can handle 128 pixels' diff sum (such as 8x16 or 16x8)
      42             : // Slightly faster than variance_final_256_pel_sse2()
      43             : // diff sum of 128 pixels can still fit in 16bit integer
      44      436567 : static INLINE void variance_final_128_pel_sse2(__m128i vsse, __m128i vsum,
      45             :     unsigned int *const sse,
      46             :     int *const sum) {
      47      436567 :     *sse = add32x4_sse2(vsse);
      48             : 
      49      436563 :     vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
      50      436563 :     vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
      51      436563 :     vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 2));
      52      436563 :     *sum = (int16_t)_mm_extract_epi16(vsum, 0);
      53      436563 : }
      54             : 
      55             : // Can handle 256 pixels' diff sum (such as 16x16)
      56       33831 : static INLINE void variance_final_256_pel_sse2(__m128i vsse, __m128i vsum,
      57             :     unsigned int *const sse,
      58             :     int *const sum) {
      59       33831 :     *sse = add32x4_sse2(vsse);
      60             : 
      61       33831 :     vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
      62       33831 :     vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
      63       33831 :     *sum = (int16_t)_mm_extract_epi16(vsum, 0);
      64       33831 :     *sum += (int16_t)_mm_extract_epi16(vsum, 1);
      65       33831 : }
      66             : 
      67     4461300 : static INLINE void variance_kernel_sse2(const __m128i src, const __m128i ref,
      68             :     __m128i *const sse,
      69             :     __m128i *const sum) {
      70     4461300 :     const __m128i diff = _mm_sub_epi16(src, ref);
      71     4461300 :     *sse = _mm_add_epi32(*sse, _mm_madd_epi16(diff, diff));
      72     4461300 :     *sum = _mm_add_epi16(*sum, diff);
      73     4461300 : }
      74             : 
      75      170835 : static INLINE void variance4_sse2(const uint8_t *src, const int src_stride,
      76             :     const uint8_t *ref, const int ref_stride,
      77             :     const int h, __m128i *const sse,
      78             :     __m128i *const sum) {
      79      170835 :     assert(h <= 256);  // May overflow for larger height.
      80      170835 :     *sum = _mm_setzero_si128();
      81             : 
      82     1096600 :     for (int i = 0; i < h; i += 2) {
      83      925766 :         const __m128i s = load4x2_sse2(src, src_stride);
      84      925765 :         const __m128i r = load4x2_sse2(ref, ref_stride);
      85             : 
      86      925767 :         variance_kernel_sse2(s, r, sse, sum);
      87      925768 :         src += 2 * src_stride;
      88      925768 :         ref += 2 * ref_stride;
      89             :     }
      90      170837 : }
      91             : 
      92      299509 : static INLINE void variance8_sse2(const uint8_t *src, const int src_stride,
      93             :     const uint8_t *ref, const int ref_stride,
      94             :     const int h, __m128i *const sse,
      95             :     __m128i *const sum) {
      96      299509 :     assert(h <= 128);  // May overflow for larger height.
      97      299509 :     *sum = _mm_setzero_si128();
      98     3835210 :     for (int i = 0; i < h; i++) {
      99     3535650 :         const __m128i s = load8_8to16_sse2(src);
     100     3535610 :         const __m128i r = load8_8to16_sse2(ref);
     101             : 
     102     3535550 :         variance_kernel_sse2(s, r, sse, sum);
     103     3535700 :         src += src_stride;
     104     3535700 :         ref += ref_stride;
     105             :     }
     106      299563 : }
     107             : 
     108             : #define AOM_VAR_NO_LOOP_SSE2(bw, bh, bits, max_pixels)                        \
     109             :   unsigned int eb_aom_variance##bw##x##bh##_sse2(                                \
     110             :       const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \
     111             :       unsigned int *sse) {                                                    \
     112             :     __m128i vsse = _mm_setzero_si128();                                       \
     113             :     __m128i vsum;                                                             \
     114             :     int sum = 0;                                                              \
     115             :     variance##bw##_sse2(src, src_stride, ref, ref_stride, bh, &vsse, &vsum);  \
     116             :     variance_final_##max_pixels##_pel_sse2(vsse, vsum, sse, &sum);            \
     117             :     assert(sum <= 255 * bw * bh);                                             \
     118             :     assert(sum >= -255 * bw * bh);                                            \
     119             :     return *sse - (uint32_t)(((int64_t)sum * sum) >> bits);                   \
     120             :   }
     121             : 
     122       66998 : AOM_VAR_NO_LOOP_SSE2(4, 4, 4, 128);
     123      119964 : AOM_VAR_NO_LOOP_SSE2(4, 8, 5, 128);
     124      154712 : AOM_VAR_NO_LOOP_SSE2(4, 16, 6, 128);
     125             : 
     126      120116 : AOM_VAR_NO_LOOP_SSE2(8, 4, 5, 128);
     127      269242 : AOM_VAR_NO_LOOP_SSE2(8, 8, 6, 128);
     128      142004 : AOM_VAR_NO_LOOP_SSE2(8, 16, 7, 128);
     129       67662 : AOM_VAR_NO_LOOP_SSE2(8, 32, 8, 256);
     130             : #if OBMC_FLAG
     131             : 
     132    51623100 :     static INLINE const int16_t *av1_get_interp_filter_subpel_kernel(
     133             :         const InterpFilterParams filter_params, const int32_t subpel) {
     134    51623100 :         return filter_params.filter_ptr + filter_params.taps * subpel;
     135             :     }
     136             :     DECLARE_ALIGNED(256, static const InterpKernel,
     137             :                 av1_bilinear_filters[SUBPEL_SHIFTS]) = {
     138             :   { 0, 0, 0, 128, 0, 0, 0, 0 },  { 0, 0, 0, 120, 8, 0, 0, 0 },
     139             :   { 0, 0, 0, 112, 16, 0, 0, 0 }, { 0, 0, 0, 104, 24, 0, 0, 0 },
     140             :   { 0, 0, 0, 96, 32, 0, 0, 0 },  { 0, 0, 0, 88, 40, 0, 0, 0 },
     141             :   { 0, 0, 0, 80, 48, 0, 0, 0 },  { 0, 0, 0, 72, 56, 0, 0, 0 },
     142             :   { 0, 0, 0, 64, 64, 0, 0, 0 },  { 0, 0, 0, 56, 72, 0, 0, 0 },
     143             :   { 0, 0, 0, 48, 80, 0, 0, 0 },  { 0, 0, 0, 40, 88, 0, 0, 0 },
     144             :   { 0, 0, 0, 32, 96, 0, 0, 0 },  { 0, 0, 0, 24, 104, 0, 0, 0 },
     145             :   { 0, 0, 0, 16, 112, 0, 0, 0 }, { 0, 0, 0, 8, 120, 0, 0, 0 }
     146             : };
     147             : 
     148             :     DECLARE_ALIGNED(256, static const InterpKernel,
     149             :                 av1_sub_pel_filters_4[SUBPEL_SHIFTS]) = {
     150             :   { 0, 0, 0, 128, 0, 0, 0, 0 },     { 0, 0, -4, 126, 8, -2, 0, 0 },
     151             :   { 0, 0, -8, 122, 18, -4, 0, 0 },  { 0, 0, -10, 116, 28, -6, 0, 0 },
     152             :   { 0, 0, -12, 110, 38, -8, 0, 0 }, { 0, 0, -12, 102, 48, -10, 0, 0 },
     153             :   { 0, 0, -14, 94, 58, -10, 0, 0 }, { 0, 0, -12, 84, 66, -10, 0, 0 },
     154             :   { 0, 0, -12, 76, 76, -12, 0, 0 }, { 0, 0, -10, 66, 84, -12, 0, 0 },
     155             :   { 0, 0, -10, 58, 94, -14, 0, 0 }, { 0, 0, -10, 48, 102, -12, 0, 0 },
     156             :   { 0, 0, -8, 38, 110, -12, 0, 0 }, { 0, 0, -6, 28, 116, -10, 0, 0 },
     157             :   { 0, 0, -4, 18, 122, -8, 0, 0 },  { 0, 0, -2, 8, 126, -4, 0, 0 }
     158             : };
     159             : DECLARE_ALIGNED(256, static const InterpKernel,
     160             :                 av1_sub_pel_filters_4smooth[SUBPEL_SHIFTS]) = {
     161             :   { 0, 0, 0, 128, 0, 0, 0, 0 },   { 0, 0, 30, 62, 34, 2, 0, 0 },
     162             :   { 0, 0, 26, 62, 36, 4, 0, 0 },  { 0, 0, 22, 62, 40, 4, 0, 0 },
     163             :   { 0, 0, 20, 60, 42, 6, 0, 0 },  { 0, 0, 18, 58, 44, 8, 0, 0 },
     164             :   { 0, 0, 16, 56, 46, 10, 0, 0 }, { 0, 0, 14, 54, 48, 12, 0, 0 },
     165             :   { 0, 0, 12, 52, 52, 12, 0, 0 }, { 0, 0, 12, 48, 54, 14, 0, 0 },
     166             :   { 0, 0, 10, 46, 56, 16, 0, 0 }, { 0, 0, 8, 44, 58, 18, 0, 0 },
     167             :   { 0, 0, 6, 42, 60, 20, 0, 0 },  { 0, 0, 4, 40, 62, 22, 0, 0 },
     168             :   { 0, 0, 4, 36, 62, 26, 0, 0 },  { 0, 0, 2, 34, 62, 30, 0, 0 }
     169             : };
     170             : DECLARE_ALIGNED(256, static const InterpKernel,
     171             :                 av1_sub_pel_filters_8[SUBPEL_SHIFTS]) = {
     172             :   { 0, 0, 0, 128, 0, 0, 0, 0 },      { 0, 2, -6, 126, 8, -2, 0, 0 },
     173             :   { 0, 2, -10, 122, 18, -4, 0, 0 },  { 0, 2, -12, 116, 28, -8, 2, 0 },
     174             :   { 0, 2, -14, 110, 38, -10, 2, 0 }, { 0, 2, -14, 102, 48, -12, 2, 0 },
     175             :   { 0, 2, -16, 94, 58, -12, 2, 0 },  { 0, 2, -14, 84, 66, -12, 2, 0 },
     176             :   { 0, 2, -14, 76, 76, -14, 2, 0 },  { 0, 2, -12, 66, 84, -14, 2, 0 },
     177             :   { 0, 2, -12, 58, 94, -16, 2, 0 },  { 0, 2, -12, 48, 102, -14, 2, 0 },
     178             :   { 0, 2, -10, 38, 110, -14, 2, 0 }, { 0, 2, -8, 28, 116, -12, 2, 0 },
     179             :   { 0, 0, -4, 18, 122, -10, 2, 0 },  { 0, 0, -2, 8, 126, -6, 2, 0 }
     180             : };
     181             : 
     182             : DECLARE_ALIGNED(256, static const InterpKernel,
     183             :                 av1_sub_pel_filters_8sharp[SUBPEL_SHIFTS]) = {
     184             :   { 0, 0, 0, 128, 0, 0, 0, 0 },         { -2, 2, -6, 126, 8, -2, 2, 0 },
     185             :   { -2, 6, -12, 124, 16, -6, 4, -2 },   { -2, 8, -18, 120, 26, -10, 6, -2 },
     186             :   { -4, 10, -22, 116, 38, -14, 6, -2 }, { -4, 10, -22, 108, 48, -18, 8, -2 },
     187             :   { -4, 10, -24, 100, 60, -20, 8, -2 }, { -4, 10, -24, 90, 70, -22, 10, -2 },
     188             :   { -4, 12, -24, 80, 80, -24, 12, -4 }, { -2, 10, -22, 70, 90, -24, 10, -4 },
     189             :   { -2, 8, -20, 60, 100, -24, 10, -4 }, { -2, 8, -18, 48, 108, -22, 10, -4 },
     190             :   { -2, 6, -14, 38, 116, -22, 10, -4 }, { -2, 6, -10, 26, 120, -18, 8, -2 },
     191             :   { -2, 4, -6, 16, 124, -12, 6, -2 },   { 0, 2, -2, 8, 126, -6, 2, -2 }
     192             : };
     193             : 
     194             : DECLARE_ALIGNED(256, static const InterpKernel,
     195             :                 av1_sub_pel_filters_8smooth[SUBPEL_SHIFTS]) = {
     196             :   { 0, 0, 0, 128, 0, 0, 0, 0 },     { 0, 2, 28, 62, 34, 2, 0, 0 },
     197             :   { 0, 0, 26, 62, 36, 4, 0, 0 },    { 0, 0, 22, 62, 40, 4, 0, 0 },
     198             :   { 0, 0, 20, 60, 42, 6, 0, 0 },    { 0, 0, 18, 58, 44, 8, 0, 0 },
     199             :   { 0, 0, 16, 56, 46, 10, 0, 0 },   { 0, -2, 16, 54, 48, 12, 0, 0 },
     200             :   { 0, -2, 14, 52, 52, 14, -2, 0 }, { 0, 0, 12, 48, 54, 16, -2, 0 },
     201             :   { 0, 0, 10, 46, 56, 16, 0, 0 },   { 0, 0, 8, 44, 58, 18, 0, 0 },
     202             :   { 0, 0, 6, 42, 60, 20, 0, 0 },    { 0, 0, 4, 40, 62, 22, 0, 0 },
     203             :   { 0, 0, 4, 36, 62, 26, 0, 0 },    { 0, 0, 2, 34, 62, 28, 2, 0 }
     204             : };
     205             :   // For w<=4, MULTITAP_SHARP is the same as EIGHTTAP_REGULAR
     206             : static const InterpFilterParams av1_interp_4tap[SWITCHABLE_FILTERS + 1] = {
     207             :   { (const int16_t *)av1_sub_pel_filters_4, SUBPEL_TAPS, SUBPEL_SHIFTS,
     208             :     EIGHTTAP_REGULAR },
     209             :   { (const int16_t *)av1_sub_pel_filters_4smooth, SUBPEL_TAPS, SUBPEL_SHIFTS,
     210             :     EIGHTTAP_SMOOTH },
     211             :   { (const int16_t *)av1_sub_pel_filters_4, SUBPEL_TAPS, SUBPEL_SHIFTS,
     212             :     EIGHTTAP_REGULAR },
     213             :   { (const int16_t *)av1_bilinear_filters, SUBPEL_TAPS, SUBPEL_SHIFTS,
     214             :     BILINEAR },
     215             : };
     216             : static const InterpFilterParams
     217             :     av1_interp_filter_params_list[SWITCHABLE_FILTERS + 1] = {
     218             :       { (const int16_t *)av1_sub_pel_filters_8, SUBPEL_TAPS, SUBPEL_SHIFTS,
     219             :         EIGHTTAP_REGULAR },
     220             :       { (const int16_t *)av1_sub_pel_filters_8smooth, SUBPEL_TAPS,
     221             :         SUBPEL_SHIFTS, EIGHTTAP_SMOOTH },
     222             :       { (const int16_t *)av1_sub_pel_filters_8sharp, SUBPEL_TAPS, SUBPEL_SHIFTS,
     223             :         MULTITAP_SHARP },
     224             :       { (const int16_t *)av1_bilinear_filters, SUBPEL_TAPS, SUBPEL_SHIFTS,
     225             :         BILINEAR }
     226             :     };
     227           0 : static INLINE const InterpFilterParams *get_4tap_interp_filter_params(
     228             :     const InterpFilter interp_filter) {
     229           0 :   return &av1_interp_4tap[interp_filter];
     230             : }
     231    36553500 : static INLINE const InterpFilterParams *av1_get_filter(int subpel_search) {
     232    36553500 :   assert(subpel_search >= USE_2_TAPS);
     233             : 
     234    36553500 :   switch (subpel_search) {
     235           0 :     case USE_2_TAPS: return get_4tap_interp_filter_params(BILINEAR);
     236           0 :     case USE_4_TAPS: return get_4tap_interp_filter_params(EIGHTTAP_REGULAR);
     237    36554000 :     case USE_8_TAPS: return &av1_interp_filter_params_list[EIGHTTAP_REGULAR];
     238           0 :     default: assert(0); return NULL;
     239             :   }
     240             : }
     241             : 
     242             : 
     243    36553600 : void aom_upsampled_pred_sse2(MacroBlockD *xd, const struct AV1Common *const cm,
     244             :                              int mi_row, int mi_col, const MV *const mv,
     245             :                              uint8_t *comp_pred, int width, int height,
     246             :                              int subpel_x_q3, int subpel_y_q3,
     247             :                              const uint8_t *ref, int ref_stride,
     248             :                              int subpel_search) {
     249             :     (void)xd;
     250             :     (void)cm;
     251             :     (void)mi_row;
     252             :     (void)mi_col;
     253             :     (void)mv;
     254    36553600 :   const InterpFilterParams *filter = av1_get_filter(subpel_search);
     255             :   // (TODO:yunqing) 2-tap case uses 4-tap functions since there is no SIMD for
     256             :   // 2-tap yet.
     257    36554700 :   int filter_taps = (subpel_search <= USE_4_TAPS) ? 4 : SUBPEL_TAPS;
     258             : 
     259    36554700 :   if (!subpel_x_q3 && !subpel_y_q3) {
     260     3082220 :     if (width >= 16) {
     261             :       int i;
     262     1750220 :       assert(!(width & 15));
     263             :       /*Read 16 pixels one row at a time.*/
     264    40672600 :       for (i = 0; i < height; i++) {
     265             :         int j;
     266   109480000 :         for (j = 0; j < width; j += 16) {
     267    70557900 :           xx_storeu_128(comp_pred, xx_loadu_128(ref));
     268    70557600 :           comp_pred += 16;
     269    70557600 :           ref += 16;
     270             :         }
     271    38922400 :         ref += ref_stride - width;
     272             :       }
     273     1332000 :     } else if (width >= 8) {
     274             :       int i;
     275     1332140 :       assert(!(width & 7));
     276     1332140 :       assert(!(height & 1));
     277             :       /*Read 8 pixels two rows at a time.*/
     278    11107500 :       for (i = 0; i < height; i += 2) {
     279     9775420 :         __m128i s0 = xx_loadl_64(ref + 0 * ref_stride);
     280     9775420 :         __m128i s1 = xx_loadl_64(ref + 1 * ref_stride);
     281     9775390 :         xx_storeu_128(comp_pred, _mm_unpacklo_epi64(s0, s1));
     282     9775390 :         comp_pred += 16;
     283     9775390 :         ref += 2 * ref_stride;
     284             :       }
     285             :     } else {
     286             :       int i;
     287           0 :       assert(!(width & 3));
     288           0 :       assert(!(height & 3));
     289             :       /*Read 4 pixels four rows at a time.*/
     290           0 :       for (i = 0; i < height; i++) {
     291           0 :         const __m128i row0 = xx_loadl_64(ref + 0 * ref_stride);
     292           0 :         const __m128i row1 = xx_loadl_64(ref + 1 * ref_stride);
     293           0 :         const __m128i row2 = xx_loadl_64(ref + 2 * ref_stride);
     294           0 :         const __m128i row3 = xx_loadl_64(ref + 3 * ref_stride);
     295           0 :         const __m128i reg = _mm_unpacklo_epi64(_mm_unpacklo_epi32(row0, row1),
     296             :                                                _mm_unpacklo_epi32(row2, row3));
     297           0 :         xx_storeu_128(comp_pred, reg);
     298           0 :         comp_pred += 16;
     299           0 :         ref += 4 * ref_stride;
     300             :       }
     301             :     }
     302    33472400 :   } else if (!subpel_y_q3) {
     303             :     const int16_t *const kernel =
     304     7618850 :         av1_get_interp_filter_subpel_kernel(*filter, subpel_x_q3 << 1);
     305     7618870 :     aom_convolve8_horiz(ref, ref_stride, comp_pred, width, kernel, 16, NULL, -1,
     306             :                         width, height);
     307    25853600 :   } else if (!subpel_x_q3) {
     308             :     const int16_t *const kernel =
     309     7715320 :         av1_get_interp_filter_subpel_kernel(*filter, subpel_y_q3 << 1);
     310     7715260 :     aom_convolve8_vert(ref, ref_stride, comp_pred, width, NULL, -1, kernel, 16,
     311             :                        width, height);
     312             :   } else {
     313             :     DECLARE_ALIGNED(16, uint8_t,
     314             :                     temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
     315             :     const int16_t *const kernel_x =
     316    18138300 :         av1_get_interp_filter_subpel_kernel(*filter, subpel_x_q3 << 1);
     317             :     const int16_t *const kernel_y =
     318    18173000 :         av1_get_interp_filter_subpel_kernel(*filter, subpel_y_q3 << 1);
     319    18169900 :     const uint8_t *ref_start = ref - ref_stride * ((filter_taps >> 1) - 1);
     320    18169900 :     uint8_t *temp_start_horiz = (subpel_search <= USE_4_TAPS)
     321           0 :                                     ? temp + (filter_taps >> 1) * MAX_SB_SIZE
     322    18169900 :                                     : temp;
     323    18169900 :     uint8_t *temp_start_vert = temp + MAX_SB_SIZE * ((filter->taps >> 1) - 1);
     324    18169900 :     int intermediate_height =
     325    18169900 :         (((height - 1) * 8 + subpel_y_q3) >> 3) + filter_taps;
     326    18169900 :     assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
     327    18169900 :     aom_convolve8_horiz(ref_start, ref_stride, temp_start_horiz, MAX_SB_SIZE,
     328             :                         kernel_x, 16, NULL, -1, width, intermediate_height);
     329    18181400 :     aom_convolve8_vert(temp_start_vert, MAX_SB_SIZE, comp_pred, width, NULL, -1,
     330             :                        kernel_y, 16, width, height);
     331             :   }
     332    36597100 : }
     333             : #endif