LCOV - code coverage report
Current view: top level - libavcodec/x86 - idctdsp_init.c (source / functions) Hit Total Coverage
Test: coverage.info Lines: 66 88 75.0 %
Date: 2017-10-18 21:45:51 Functions: 2 2 100.0 %

          Line data    Source code
       1             : /*
       2             :  * This file is part of FFmpeg.
       3             :  *
       4             :  * FFmpeg is free software; you can redistribute it and/or
       5             :  * modify it under the terms of the GNU Lesser General Public
       6             :  * License as published by the Free Software Foundation; either
       7             :  * version 2.1 of the License, or (at your option) any later version.
       8             :  *
       9             :  * FFmpeg is distributed in the hope that it will be useful,
      10             :  * but WITHOUT ANY WARRANTY; without even the implied warranty of
      11             :  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
      12             :  * Lesser General Public License for more details.
      13             :  *
      14             :  * You should have received a copy of the GNU Lesser General Public
      15             :  * License along with FFmpeg; if not, write to the Free Software
      16             :  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
      17             :  */
      18             : 
      19             : #include "config.h"
      20             : #include "libavutil/attributes.h"
      21             : #include "libavutil/cpu.h"
      22             : #include "libavutil/x86/cpu.h"
      23             : #include "libavcodec/avcodec.h"
      24             : #include "libavcodec/idctdsp.h"
      25             : #include "idctdsp.h"
      26             : #include "simple_idct.h"
      27             : 
      28             : /* Input permutation for the simple_idct_mmx */
      29             : static const uint8_t simple_mmx_permutation[64] = {
      30             :     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
      31             :     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
      32             :     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
      33             :     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
      34             :     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
      35             :     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
      36             :     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
      37             :     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
      38             : };
      39             : 
      40             : static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };
      41             : 
      42        1652 : av_cold int ff_init_scantable_permutation_x86(uint8_t *idct_permutation,
      43             :                                               enum idct_permutation_type perm_type)
      44             : {
      45             :     int i;
      46             : 
      47        1652 :     switch (perm_type) {
      48           0 :     case FF_IDCT_PERM_SIMPLE:
      49           0 :         for (i = 0; i < 64; i++)
      50           0 :             idct_permutation[i] = simple_mmx_permutation[i];
      51           0 :         return 1;
      52           4 :     case FF_IDCT_PERM_SSE2:
      53         260 :         for (i = 0; i < 64; i++)
      54         256 :             idct_permutation[i] = (i & 0x38) | idct_sse2_row_perm[i & 7];
      55           4 :         return 1;
      56             :     }
      57             : 
      58        1648 :     return 0;
      59             : }
      60             : 
      61        1503 : av_cold void ff_idctdsp_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
      62             :                                  unsigned high_bit_depth)
      63             : {
      64        1503 :     int cpu_flags = av_get_cpu_flags();
      65             : 
      66        1503 :     if (EXTERNAL_MMX(cpu_flags)) {
      67         265 :         c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
      68         265 :         c->put_pixels_clamped        = ff_put_pixels_clamped_mmx;
      69         265 :         c->add_pixels_clamped        = ff_add_pixels_clamped_mmx;
      70             : 
      71         530 :         if (!high_bit_depth &&
      72         530 :             avctx->lowres == 0 &&
      73         268 :             (avctx->idct_algo == FF_IDCT_AUTO ||
      74           6 :                 avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
      75           3 :                 avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
      76         262 :                 c->idct_put  = ff_simple_idct_put_mmx;
      77         262 :                 c->idct_add  = ff_simple_idct_add_mmx;
      78         262 :                 c->idct      = ff_simple_idct_mmx;
      79         262 :                 c->perm_type = FF_IDCT_PERM_SIMPLE;
      80             :         }
      81             :     }
      82             : 
      83        1503 :     if (EXTERNAL_SSE2(cpu_flags)) {
      84         265 :         c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
      85         265 :         c->put_pixels_clamped        = ff_put_pixels_clamped_sse2;
      86         265 :         c->add_pixels_clamped        = ff_add_pixels_clamped_sse2;
      87             : 
      88         530 :         if (!high_bit_depth &&
      89         530 :             avctx->lowres == 0 &&
      90         268 :             (avctx->idct_algo == FF_IDCT_AUTO ||
      91           6 :                 avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
      92           3 :                 avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
      93         262 :                 c->idct_put  = ff_simple_idct_put_sse2;
      94         262 :                 c->idct_add  = ff_simple_idct_add_sse2;
      95         262 :                 c->perm_type = FF_IDCT_PERM_SIMPLE;
      96             :         }
      97             : 
      98         265 :         if (ARCH_X86_64 &&
      99         265 :             !high_bit_depth &&
     100         530 :             avctx->lowres == 0 &&
     101         268 :             (avctx->idct_algo == FF_IDCT_AUTO ||
     102           6 :                 avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
     103           6 :                 avctx->idct_algo == FF_IDCT_SIMPLEMMX ||
     104           3 :                 avctx->idct_algo == FF_IDCT_SIMPLE)) {
     105         263 :                 c->idct      = ff_simple_idct8_sse2;
     106         263 :                 c->idct_put  = ff_simple_idct8_put_sse2;
     107         263 :                 c->idct_add  = ff_simple_idct8_add_sse2;
     108         263 :                 c->perm_type = FF_IDCT_PERM_TRANSPOSE;
     109             :         }
     110             :     }
     111             : 
     112        1503 :     if (ARCH_X86_64 && avctx->lowres == 0) {
     113        1501 :         if (EXTERNAL_AVX(cpu_flags) &&
     114         265 :             !high_bit_depth &&
     115         268 :             (avctx->idct_algo == FF_IDCT_AUTO ||
     116           6 :                 avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
     117           6 :                 avctx->idct_algo == FF_IDCT_SIMPLEMMX ||
     118           3 :                 avctx->idct_algo == FF_IDCT_SIMPLE)) {
     119         263 :                 c->idct      = ff_simple_idct8_avx;
     120         263 :                 c->idct_put  = ff_simple_idct8_put_avx;
     121         263 :                 c->idct_add  = ff_simple_idct8_add_avx;
     122         263 :                 c->perm_type = FF_IDCT_PERM_TRANSPOSE;
     123             :         }
     124             : 
     125        1524 :         if (avctx->bits_per_raw_sample == 10 &&
     126          46 :             (avctx->idct_algo == FF_IDCT_AUTO ||
     127          46 :              avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
     128          23 :              avctx->idct_algo == FF_IDCT_SIMPLE)) {
     129          23 :             if (EXTERNAL_SSE2(cpu_flags)) {
     130           0 :                 c->idct_put  = ff_simple_idct10_put_sse2;
     131           0 :                 c->idct_add  = NULL;
     132           0 :                 c->idct      = ff_simple_idct10_sse2;
     133           0 :                 c->perm_type = FF_IDCT_PERM_TRANSPOSE;
     134             : 
     135             :             }
     136          23 :             if (EXTERNAL_AVX(cpu_flags)) {
     137           0 :                 c->idct_put  = ff_simple_idct10_put_avx;
     138           0 :                 c->idct_add  = NULL;
     139           0 :                 c->idct      = ff_simple_idct10_avx;
     140           0 :                 c->perm_type = FF_IDCT_PERM_TRANSPOSE;
     141             :             }
     142             :         }
     143             : 
     144        1505 :         if (avctx->bits_per_raw_sample == 12 &&
     145           8 :             (avctx->idct_algo == FF_IDCT_AUTO ||
     146           4 :              avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
     147           0 :             if (EXTERNAL_SSE2(cpu_flags)) {
     148           0 :                 c->idct_put  = ff_simple_idct12_put_sse2;
     149           0 :                 c->idct_add  = NULL;
     150           0 :                 c->idct      = ff_simple_idct12_sse2;
     151           0 :                 c->perm_type = FF_IDCT_PERM_TRANSPOSE;
     152             :             }
     153           0 :             if (EXTERNAL_AVX(cpu_flags)) {
     154           0 :                 c->idct_put  = ff_simple_idct12_put_avx;
     155           0 :                 c->idct_add  = NULL;
     156           0 :                 c->idct      = ff_simple_idct12_avx;
     157           0 :                 c->perm_type = FF_IDCT_PERM_TRANSPOSE;
     158             :             }
     159             :         }
     160             :     }
     161        1503 : }

Generated by: LCOV version 1.13