falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
11648 Commits
32 Branches
484MB
Tree: bc2bc41bb8
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'bc2bc41bb8'
${ noResults }
FFmpeg/libavcodec/i386/vc1dsp_mmx.c

495 lines
23KB
Raw Blame History 

  1. /*

  2.  * VC-1 and WMV3 - DSP functions MMX-optimized

  3.  * Copyright (c) 2007 Christophe GISQUET <christophe.gisquet@free.fr>

  4.  *

  5.  * Permission is hereby granted, free of charge, to any person

  6.  * obtaining a copy of this software and associated documentation

  7.  * files (the "Software"), to deal in the Software without

  8.  * restriction, including without limitation the rights to use,

  9.  * copy, modify, merge, publish, distribute, sublicense, and/or sell

  10.  * copies of the Software, and to permit persons to whom the

  11.  * Software is furnished to do so, subject to the following

  12.  * conditions:

  13.  *

  14.  * The above copyright notice and this permission notice shall be

  15.  * included in all copies or substantial portions of the Software.

  16.  *

  17.  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,

  18.  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES

  19.  * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND

  20.  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT

  21.  * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,

  22.  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING

  23.  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR

  24.  * OTHER DEALINGS IN THE SOFTWARE.

  25.  */

  26. 

  27. #include "dsputil.h"

  28. #include "dsputil_mmx.h"

  29. #include "x86_cpu.h"

  30. 

  31. /** Add rounder from mm7 to mm3 and pack result at destination */

  32. #define NORMALIZE_MMX(SHIFT)                                    \

  33.      "paddw     %%mm7, %%mm3           \n\t" /* +bias-r */      \

  34.      "paddw     %%mm7, %%mm4           \n\t" /* +bias-r */      \

  35.      "psraw     "SHIFT", %%mm3         \n\t"                    \

  36.      "psraw     "SHIFT", %%mm4         \n\t"

  37. 

  38. #define TRANSFER_DO_PACK                        \

  39.      "packuswb  %%mm4, %%mm3           \n\t"    \

  40.      "movq      %%mm3, (%2)            \n\t"

  41. 

  42. #define TRANSFER_DONT_PACK                      \

  43.      "movq      %%mm3, 0(%2)           \n\t"    \

  44.      "movq      %%mm4, 8(%2)           \n\t"

  45. 

  46. /** @see MSPEL_FILTER13_CORE for use as UNPACK macro */

  47. #define DO_UNPACK(reg)  "punpcklbw %%mm0, " reg "\n\t"

  48. #define DONT_UNPACK(reg)

  49. 

  50. /** Compute the rounder 32-r or 8-r and unpacks it to mm7 */

  51. #define LOAD_ROUNDER_MMX(ROUND)                 \

  52.      "movd      "ROUND", %%mm7         \n\t"    \

  53.      "punpcklwd %%mm7, %%mm7           \n\t"    \

  54.      "punpckldq %%mm7, %%mm7           \n\t"

  55. 

  56. #define SHIFT2_LINE(OFF, R0,R1,R2,R3)           \

  57.     "paddw     %%mm"#R2", %%mm"#R1"    \n\t"    \

  58.     "movd      (%1,%4), %%mm"#R0"      \n\t"    \

  59.     "pmullw    %%mm6, %%mm"#R1"        \n\t"    \

  60.     "punpcklbw %%mm0, %%mm"#R0"        \n\t"    \

  61.     "movd      (%1,%3), %%mm"#R3"      \n\t"    \

  62.     "psubw     %%mm"#R0", %%mm"#R1"    \n\t"    \

  63.     "punpcklbw %%mm0, %%mm"#R3"        \n\t"    \

  64.     "paddw     %%mm7, %%mm"#R1"        \n\t"    \

  65.     "psubw     %%mm"#R3", %%mm"#R1"    \n\t"    \

  66.     "psraw     %5, %%mm"#R1"           \n\t"    \

  67.     "movq      %%mm"#R1", "#OFF"(%2)   \n\t"    \

  68.     "add       %3, %1                  \n\t"

  69. 

  70. DECLARE_ALIGNED_16(static const uint64_t, fact_9) = 0x0009000900090009ULL;

  71. 

  72. /** Sacrifying mm6 allows to pipeline loads from src */

  73. static void vc1_put_ver_16b_shift2_mmx(int16_t *dst,

  74.                                        const uint8_t *src, long int stride,

  75.                                        int rnd, int64_t shift)

  76. {

  77.     int  w = 3;

  78. 

  79.     asm volatile(

  80.         LOAD_ROUNDER_MMX("%6")

  81.         "movq      %7, %%mm6               \n\t"

  82.         "1:                                \n\t"

  83.         "movd      (%1), %%mm2             \n\t"

  84.         "add       %3, %1                  \n\t"

  85.         "movd      (%1), %%mm3             \n\t"

  86.         "punpcklbw %%mm0, %%mm2            \n\t"

  87.         "punpcklbw %%mm0, %%mm3            \n\t"

  88.         SHIFT2_LINE(  0, 1, 2, 3, 4)

  89.         SHIFT2_LINE( 24, 2, 3, 4, 1)

  90.         SHIFT2_LINE( 48, 3, 4, 1, 2)

  91.         SHIFT2_LINE( 72, 4, 1, 2, 3)

  92.         SHIFT2_LINE( 96, 1, 2, 3, 4)

  93.         SHIFT2_LINE(120, 2, 3, 4, 1)

  94.         SHIFT2_LINE(144, 3, 4, 1, 2)

  95.         SHIFT2_LINE(168, 4, 1, 2, 3)

  96.         "sub       %8, %1                  \n\t"

  97.         "add       $8, %2                  \n\t"

  98.         "dec       %0                      \n\t"

  99.         "jnz 1b                            \n\t"

  100.         : "+g"(w), "+r"(src), "+r"(dst)

  101.         : "r"(stride), "r"(-2*stride), "m"(shift),

  102.           "m"(rnd), "m"(fact_9), "g"(9*stride-4)

  103.         : "memory"

  104.     );

  105. }

  106. 

  107. /**

  108.  * Data is already unpacked, so some operations can directly be made from

  109.  * memory.

  110.  */

  111. static void vc1_put_hor_16b_shift2_mmx(uint8_t *dst, long int stride,

  112.                                        const int16_t *src, int rnd)

  113. {

  114.     int h = 8;

  115. 

  116.     src -= 1;

  117.     rnd -= (-1+9+9-1)*1024; /* Add -1024 bias */

  118.     asm volatile(

  119.         LOAD_ROUNDER_MMX("%4")

  120.         "movq      %6, %%mm6               \n\t"

  121.         "movq      %5, %%mm5               \n\t"

  122.         "1:                                \n\t"

  123.         "movq      2*0+0(%1), %%mm1        \n\t"

  124.         "movq      2*0+8(%1), %%mm2        \n\t"

  125.         "movq      2*1+0(%1), %%mm3        \n\t"

  126.         "movq      2*1+8(%1), %%mm4        \n\t"

  127.         "paddw     2*3+0(%1), %%mm1        \n\t"

  128.         "paddw     2*3+8(%1), %%mm2        \n\t"

  129.         "paddw     2*2+0(%1), %%mm3        \n\t"

  130.         "paddw     2*2+8(%1), %%mm4        \n\t"

  131.         "pmullw    %%mm5, %%mm3            \n\t"

  132.         "pmullw    %%mm5, %%mm4            \n\t"

  133.         "psubw     %%mm1, %%mm3            \n\t"

  134.         "psubw     %%mm2, %%mm4            \n\t"

  135.         NORMALIZE_MMX("$7")

  136.         /* Remove bias */

  137.         "paddw     %%mm6, %%mm3            \n\t"

  138.         "paddw     %%mm6, %%mm4            \n\t"

  139.         TRANSFER_DO_PACK

  140.         "add       $24, %1                 \n\t"

  141.         "add       %3, %2                  \n\t"

  142.         "dec       %0                      \n\t"

  143.         "jnz 1b                            \n\t"

  144.         : "+g"(h), "+r" (src),  "+r" (dst)

  145.         : "g"(stride), "m"(rnd), "m"(fact_9), "m"(ff_pw_128)

  146.         : "memory"

  147.     );

  148. }

  149. 

  150. 

  151. /**

  152.  * Purely vertical or horizontal 1/2 shift interpolation.

  153.  * Sacrify mm6 for *9 factor.

  154.  */

  155. static void vc1_put_shift2_mmx(uint8_t *dst, const uint8_t *src,

  156.                                long int stride, int rnd, long int offset)

  157. {

  158.     int h = 8;

  159. 

  160.     rnd = 8-rnd;

  161.     asm volatile(

  162.         LOAD_ROUNDER_MMX("%6")

  163.         "movq      %8, %%mm6               \n\t"

  164.         "1:                                \n\t"

  165.         "movd      0(%1   ), %%mm3         \n\t"

  166.         "movd      4(%1   ), %%mm4         \n\t"

  167.         "movd      0(%1,%3), %%mm1         \n\t"

  168.         "movd      4(%1,%3), %%mm2         \n\t"

  169.         "add       %3, %1                  \n\t"

  170.         "punpcklbw %%mm0, %%mm3            \n\t"

  171.         "punpcklbw %%mm0, %%mm4            \n\t"

  172.         "punpcklbw %%mm0, %%mm1            \n\t"

  173.         "punpcklbw %%mm0, %%mm2            \n\t"

  174.         "paddw     %%mm1, %%mm3            \n\t"

  175.         "paddw     %%mm2, %%mm4            \n\t"

  176.         "movd      0(%1,%4), %%mm1         \n\t"

  177.         "movd      4(%1,%4), %%mm2         \n\t"

  178.         "pmullw    %%mm6, %%mm3            \n\t" /* 0,9,9,0*/

  179.         "pmullw    %%mm6, %%mm4            \n\t" /* 0,9,9,0*/

  180.         "punpcklbw %%mm0, %%mm1            \n\t"

  181.         "punpcklbw %%mm0, %%mm2            \n\t"

  182.         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,0*/

  183.         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,0*/

  184.         "movd      0(%1,%3), %%mm1         \n\t"

  185.         "movd      4(%1,%3), %%mm2         \n\t"

  186.         "punpcklbw %%mm0, %%mm1            \n\t"

  187.         "punpcklbw %%mm0, %%mm2            \n\t"

  188.         "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,-1*/

  189.         "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,-1*/

  190.         NORMALIZE_MMX("$4")

  191.         TRANSFER_DO_PACK

  192.         "add       %7, %1                  \n\t"

  193.         "add       %5, %2                  \n\t"

  194.         "dec       %0                      \n\t"

  195.         "jnz 1b                            \n\t"

  196.         : "+g"(h), "+r"(src),  "+r"(dst)

  197.         : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),

  198.           "g"(stride-offset), "m"(fact_9)

  199.         : "memory"

  200.     );

  201. }

  202. 

  203. /**

  204.  * Filter coefficients made global to allow access by all 1 or 3 quarter shift

  205.  * interpolation functions.

  206.  */

  207. DECLARE_ALIGNED_16(static const uint64_t, fact_53) = 0x0035003500350035ULL;

  208. DECLARE_ALIGNED_16(static const uint64_t, fact_18) = 0x0012001200120012ULL;

  209. 

  210. /**

  211.  * Core of the 1/4 and 3/4 shift bicubic interpolation.

  212.  *

  213.  * @param UNPACK  Macro unpacking arguments from 8 to 16bits (can be empty).

  214.  * @param MOVQ    "movd 1" or "movq 2", if data read is already unpacked.

  215.  * @param A1      Address of 1st tap (beware of unpacked/packed).

  216.  * @param A2      Address of 2nd tap

  217.  * @param A3      Address of 3rd tap

  218.  * @param A4      Address of 4th tap

  219.  */

  220. #define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4, POS)  \

  221.      MOVQ "*0+"A1", %%mm1       \n\t"                           \

  222.      MOVQ "*4+"A1", %%mm2       \n\t"                           \

  223.      UNPACK("%%mm1")                                            \

  224.      UNPACK("%%mm2")                                            \

  225.      "pmullw    "POS", %%mm1    \n\t"                           \

  226.      "pmullw    "POS", %%mm2    \n\t"                           \

  227.      MOVQ "*0+"A2", %%mm3       \n\t"                           \

  228.      MOVQ "*4+"A2", %%mm4       \n\t"                           \

  229.      UNPACK("%%mm3")                                            \

  230.      UNPACK("%%mm4")                                            \

  231.      "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                 \

  232.      "pmullw    %%mm6, %%mm4    \n\t" /* *18 */                 \

  233.      "psubw     %%mm1, %%mm3    \n\t" /* 18,-3 */               \

  234.      "psubw     %%mm2, %%mm4    \n\t" /* 18,-3 */               \

  235.      MOVQ "*0+"A4", %%mm1       \n\t"                           \

  236.      MOVQ "*4+"A4", %%mm2       \n\t"                           \

  237.      UNPACK("%%mm1")                                            \

  238.      UNPACK("%%mm2")                                            \

  239.      "psllw     $2, %%mm1       \n\t" /* 4* */                  \

  240.      "psllw     $2, %%mm2       \n\t" /* 4* */                  \

  241.      "psubw     %%mm1, %%mm3    \n\t" /* -4,18,-3 */            \

  242.      "psubw     %%mm2, %%mm4    \n\t" /* -4,18,-3 */            \

  243.      MOVQ "*0+"A3", %%mm1       \n\t"                           \

  244.      MOVQ "*4+"A3", %%mm2       \n\t"                           \

  245.      UNPACK("%%mm1")                                            \

  246.      UNPACK("%%mm2")                                            \

  247.      "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                 \

  248.      "pmullw    %%mm5, %%mm2    \n\t" /* *53 */                 \

  249.      "paddw     %%mm1, %%mm3    \n\t" /* 4,53,18,-3 */          \

  250.      "paddw     %%mm2, %%mm4    \n\t" /* 4,53,18,-3 */

  251. 

  252. /**

  253.  * Macro to build the vertical 16bits version of vc1_put_shift[13].

  254.  * Here, offset=src_stride. Parameters passed A1 to A4 must use

  255.  * %3 (src_stride) and %4 (3*src_stride).

  256.  *

  257.  * @param  NAME   Either 1 or 3

  258.  * @see MSPEL_FILTER13_CORE for information on A1->A4

  259.  */

  260. #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4)                    \

  261. static void                                                             \

  262. vc1_put_ver_16b_ ## NAME ## _mmx(int16_t *dst, const uint8_t *src,      \

  263.                                  long int src_stride,                   \

  264.                                  int rnd, int64_t shift)                \

  265. {                                                                       \

  266.     int h = 8;                                                          \

  267.     src -= src_stride;                                                  \

  268.     asm volatile(                                                       \

  269.         LOAD_ROUNDER_MMX("%5")                                          \

  270.         "movq      %7, %%mm5       \n\t"                                \

  271.         "movq      %8, %%mm6       \n\t"                                \

  272.         ASMALIGN(3)                                                     \

  273.         "1:                        \n\t"                                \

  274.         MSPEL_FILTER13_CORE(DO_UNPACK, "movd  1", A1, A2, A3, A4, "%9") \

  275.         NORMALIZE_MMX("%6")                                             \

  276.         TRANSFER_DONT_PACK                                              \

  277.         /* Last 3 (in fact 4) bytes on the line */                      \

  278.         "movd      8+"A1", %%mm1   \n\t"                                \

  279.         DO_UNPACK("%%mm1")                                              \

  280.         "movq      %%mm1, %%mm3    \n\t"                                \

  281.         "paddw     %%mm1, %%mm1    \n\t"                                \

  282.         "paddw     %%mm3, %%mm1    \n\t" /* 3* */                       \

  283.         "movd      8+"A2", %%mm3   \n\t"                                \

  284.         DO_UNPACK("%%mm3")                                              \

  285.         "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                      \

  286.         "psubw     %%mm1, %%mm3    \n\t" /*18,-3 */                     \

  287.         "movd      8+"A3", %%mm1   \n\t"                                \

  288.         DO_UNPACK("%%mm1")                                              \

  289.         "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                      \

  290.         "paddw     %%mm1, %%mm3    \n\t" /*53,18,-3 */                  \

  291.         "movd      8+"A4", %%mm1   \n\t"                                \

  292.         DO_UNPACK("%%mm1")                                              \

  293.         "psllw     $2, %%mm1       \n\t" /* 4* */                       \

  294.         "psubw     %%mm1, %%mm3    \n\t"                                \

  295.         "paddw     %%mm7, %%mm3    \n\t"                                \

  296.         "psraw     %6, %%mm3       \n\t"                                \

  297.         "movq      %%mm3, 16(%2)   \n\t"                                \

  298.         "add       %3, %1          \n\t"                                \

  299.         "add       $24, %2         \n\t"                                \

  300.         "dec       %0              \n\t"                                \

  301.         "jnz 1b                    \n\t"                                \

  302.         : "+g"(h), "+r" (src),  "+r" (dst)                              \

  303.         : "r"(src_stride), "r"(3*src_stride),                           \

  304.           "m"(rnd), "m"(shift),                                         \

  305.           "m"(fact_53), "m"(fact_18), "m"(ff_pw_3)                      \

  306.         : "memory"                                                      \

  307.     );                                                                  \

  308. }

  309. 

  310. /**

  311.  * Macro to build the horizontal 16bits version of vc1_put_shift[13].

  312.  * Here, offset=16bits, so parameters passed A1 to A4 should be simple.

  313.  *

  314.  * @param  NAME   Either 1 or 3

  315.  * @see MSPEL_FILTER13_CORE for information on A1->A4

  316.  */

  317. #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4)                    \

  318. static void                                                             \

  319. vc1_put_hor_16b_ ## NAME ## _mmx(uint8_t *dst, long int stride,         \

  320.                                  const int16_t *src, int rnd)           \

  321. {                                                                       \

  322.     int h = 8;                                                          \

  323.     src -= 1;                                                           \

  324.     rnd -= (-4+58+13-3)*256; /* Add -256 bias */                        \

  325.     asm volatile(                                                       \

  326.         LOAD_ROUNDER_MMX("%4")                                          \

  327.         "movq      %6, %%mm6       \n\t"                                \

  328.         "movq      %5, %%mm5       \n\t"                                \

  329.         ASMALIGN(3)                                                     \

  330.         "1:                        \n\t"                                \

  331.         MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4, "%8")\

  332.         NORMALIZE_MMX("$7")                                             \

  333.         /* Remove bias */                                               \

  334.         "paddw     %7, %%mm3       \n\t"                                \

  335.         "paddw     %7, %%mm4       \n\t"                                \

  336.         TRANSFER_DO_PACK                                                \

  337.         "add       $24, %1         \n\t"                                \

  338.         "add       %3, %2          \n\t"                                \

  339.         "dec       %0              \n\t"                                \

  340.         "jnz 1b                    \n\t"                                \

  341.         : "+g"(h), "+r" (src),  "+r" (dst)                              \

  342.         : "g"(stride), "m"(rnd), "m"(fact_53), "m"(fact_18),            \

  343.           "m"(ff_pw_128), "m"(ff_pw_3)                                  \

  344.         : "memory"                                                      \

  345.     );                                                                  \

  346. }

  347. 

  348. /**

  349.  * Macro to build the 8bits, any direction, version of vc1_put_shift[13].

  350.  * Here, offset=src_stride. Parameters passed A1 to A4 must use

  351.  * %3 (offset) and %4 (3*offset).

  352.  *

  353.  * @param  NAME   Either 1 or 3

  354.  * @see MSPEL_FILTER13_CORE for information on A1->A4

  355.  */

  356. #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4)                         \

  357. static void                                                             \

  358. vc1_put_## NAME ## _mmx(uint8_t *dst, const uint8_t *src,               \

  359.                         long int stride, int rnd, long int offset)      \

  360. {                                                                       \

  361.     int h = 8;                                                          \

  362.     src -= offset;                                                      \

  363.     rnd = 32-rnd;                                                       \

  364.     asm volatile (                                                      \

  365.         LOAD_ROUNDER_MMX("%6")                                          \

  366.         "movq      %7, %%mm5       \n\t"                                \

  367.         "movq      %8, %%mm6       \n\t"                                \

  368.         ASMALIGN(3)                                                     \

  369.         "1:                        \n\t"                                \

  370.         MSPEL_FILTER13_CORE(DO_UNPACK, "movd   1", A1, A2, A3, A4, "%9")\

  371.         NORMALIZE_MMX("$6")                                             \

  372.         TRANSFER_DO_PACK                                                \

  373.         "add       %5, %1          \n\t"                                \

  374.         "add       %5, %2          \n\t"                                \

  375.         "dec       %0              \n\t"                                \

  376.         "jnz 1b                    \n\t"                                \

  377.         : "+g"(h), "+r" (src),  "+r" (dst)                              \

  378.         : "r"(offset), "r"(3*offset), "g"(stride), "m"(rnd),            \

  379.           "m"(fact_53), "m"(fact_18), "m"(ff_pw_3)                      \

  380.         : "memory"                                                      \

  381.     );                                                                  \

  382. }

  383. 

  384. /** 1/4 shift bicubic interpolation */

  385. MSPEL_FILTER13_8B     (shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )")

  386. MSPEL_FILTER13_VER_16B(shift1, "0(%1,%4  )", "0(%1,%3,2)", "0(%1,%3  )", "0(%1     )")

  387. MSPEL_FILTER13_HOR_16B(shift1, "2*3(%1)", "2*2(%1)", "2*1(%1)", "2*0(%1)")

  388. 

  389. /** 3/4 shift bicubic interpolation */

  390. MSPEL_FILTER13_8B     (shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )")

  391. MSPEL_FILTER13_VER_16B(shift3, "0(%1     )", "0(%1,%3  )", "0(%1,%3,2)", "0(%1,%4  )")

  392. MSPEL_FILTER13_HOR_16B(shift3, "2*0(%1)", "2*1(%1)", "2*2(%1)", "2*3(%1)")

  393. 

  394. typedef void (*vc1_mspel_mc_filter_ver_16bits)(int16_t *dst, const uint8_t *src, long int src_stride, int rnd, int64_t shift);

  395. typedef void (*vc1_mspel_mc_filter_hor_16bits)(uint8_t *dst, long int dst_stride, const int16_t *src, int rnd);

  396. typedef void (*vc1_mspel_mc_filter_8bits)(uint8_t *dst, const uint8_t *src, long int stride, int rnd, long int offset);

  397. 

  398. /**

  399.  * Interpolates fractional pel values by applying proper vertical then

  400.  * horizontal filter.

  401.  *

  402.  * @param  dst     Destination buffer for interpolated pels.

  403.  * @param  src     Source buffer.

  404.  * @param  stride  Stride for both src and dst buffers.

  405.  * @param  hmode   Horizontal filter (expressed in quarter pixels shift).

  406.  * @param  hmode   Vertical filter.

  407.  * @param  rnd     Rounding bias.

  408.  */

  409. static void vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,

  410.                          int hmode, int vmode, int rnd)

  411. {

  412.     static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =

  413.          { NULL, vc1_put_ver_16b_shift1_mmx, vc1_put_ver_16b_shift2_mmx, vc1_put_ver_16b_shift3_mmx };

  414.     static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =

  415.          { NULL, vc1_put_hor_16b_shift1_mmx, vc1_put_hor_16b_shift2_mmx, vc1_put_hor_16b_shift3_mmx };

  416.     static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =

  417.          { NULL, vc1_put_shift1_mmx, vc1_put_shift2_mmx, vc1_put_shift3_mmx };

  418. 

  419.     asm volatile(

  420.         "pxor %%mm0, %%mm0         \n\t"

  421.         ::: "memory"

  422.     );

  423. 

  424.     if (vmode) { /* Vertical filter to apply */

  425.         if (hmode) { /* Horizontal filter to apply, output to tmp */

  426.             static const int shift_value[] = { 0, 5, 1, 5 };

  427.             int              shift = (shift_value[hmode]+shift_value[vmode])>>1;

  428.             int              r;

  429.             DECLARE_ALIGNED_16(int16_t, tmp[12*8]);

  430. 

  431.             r = (1<<(shift-1)) + rnd-1;

  432.             vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);

  433. 

  434.             vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);

  435.             return;

  436.         }

  437.         else { /* No horizontal filter, output 8 lines to dst */

  438.             vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);

  439.             return;

  440.         }

  441.     }

  442. 

  443.     /* Horizontal mode with no vertical mode */

  444.     vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);

  445. }

  446. 

  447. void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd);

  448. 

  449. /** Macro to ease bicubic filter interpolation functions declarations */

  450. #define DECLARE_FUNCTION(a, b)                                          \

  451. static void put_vc1_mspel_mc ## a ## b ## _mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \

  452.      vc1_mspel_mc(dst, src, stride, a, b, rnd);                         \

  453. }

  454. 

  455. DECLARE_FUNCTION(0, 1)

  456. DECLARE_FUNCTION(0, 2)

  457. DECLARE_FUNCTION(0, 3)

  458. 

  459. DECLARE_FUNCTION(1, 0)

  460. DECLARE_FUNCTION(1, 1)

  461. DECLARE_FUNCTION(1, 2)

  462. DECLARE_FUNCTION(1, 3)

  463. 

  464. DECLARE_FUNCTION(2, 0)

  465. DECLARE_FUNCTION(2, 1)

  466. DECLARE_FUNCTION(2, 2)

  467. DECLARE_FUNCTION(2, 3)

  468. 

  469. DECLARE_FUNCTION(3, 0)

  470. DECLARE_FUNCTION(3, 1)

  471. DECLARE_FUNCTION(3, 2)

  472. DECLARE_FUNCTION(3, 3)

  473. 

  474. void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx) {

  475.     dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_mmx;

  476.     dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_mmx;

  477.     dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_mmx;

  478.     dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_mmx;

  479. 

  480.     dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_mmx;

  481.     dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_mmx;

  482.     dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_mmx;

  483.     dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_mmx;

  484. 

  485.     dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_mmx;

  486.     dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_mmx;

  487.     dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_mmx;

  488.     dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_mmx;

  489. 

  490.     dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_mmx;

  491.     dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_mmx;

  492.     dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_mmx;

  493.     dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_mmx;

  494. }