API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Jul 1st, 2017
531
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Diff 38.36 KB | None | 0 0
raw download clone embed print report
  1. From 7fde0735a8676fc36efc054abb55b4a3c9580773 Mon Sep 17 00:00:00 2001
  2. From: Daniel Kang <daniel.d.kang@gmail.com>
  3. Date: Thu, 25 Nov 2010 19:44:56 -0500
  4. Subject: [PATCH 1/8] predict_4x4_ddl
  5.  
  6. ---
  7. common/x86/predict-a.asm |  108 +++++++++++++++++++++++++--------------------
  8.  common/x86/predict-c.c   |    9 +++-
  9.  2 files changed, 67 insertions(+), 50 deletions(-)
  10.  
  11. diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
  12. index a05e91b..739ce35 100644
  13. --- a/common/x86/predict-a.asm
  14. +++ b/common/x86/predict-a.asm
  15. @@ -40,6 +40,7 @@ SECTION .text
  16.  
  17.  cextern pb_1
  18.  cextern pb_3
  19. +cextern pw_1
  20.  cextern pw_2
  21.  cextern pw_4
  22.  cextern pw_8
  23. @@ -98,19 +99,16 @@ cextern pb_reverse
  24.  ; dest, left, right, src, tmp
  25.  ; output: %1 = (t[n-1] + t[n]*2 + t[n+1] + 2) >> 2
  26.  %macro PRED8x8_LOWPASS0 6
  27. -    mov%6       %5, %2
  28. -    pavgb       %2, %3
  29. -    pxor        %3, %5
  30. -    mov%6       %1, %4
  31. -    pand        %3, [pb_1]
  32. -    psubusb     %2, %3
  33. -    pavgb       %1, %2
  34. +    mova        %6, %3
  35. +    pavg%1      %3, %4
  36. +    pxor        %4, %6
  37. +    mova        %2, %5
  38. +    pand        %4, [p%1_1]
  39. +    psubus%1    %3, %4
  40. +    pavg%1      %2, %3
  41.  %endmacro
  42. -%macro PRED8x8_LOWPASS 5
  43. -    PRED8x8_LOWPASS0 %1, %2, %3, %4, %5, q
  44. -%endmacro
  45. -%macro PRED8x8_LOWPASS_XMM 5
  46. -    PRED8x8_LOWPASS0 %1, %2, %3, %4, %5, dqa
  47. +%macro PRED8x8_LOWPASS 6
  48. +    PRED8x8_LOWPASS0 %1, %2, %3, %4, %5, %6
  49.  %endmacro
  50.  
  51.  %macro LOAD_PLANE_ARGS 0
  52. @@ -129,27 +127,37 @@ cextern pb_reverse
  53.  %endmacro
  54.  
  55.  ;-----------------------------------------------------------------------------
  56. -; void predict_4x4_ddl( uint8_t *src )
  57. +; void predict_4x4_ddl( pixel *src )
  58.  ;-----------------------------------------------------------------------------
  59. -cglobal predict_4x4_ddl_mmxext, 1,1
  60. -    movq    mm1, [r0-FDEC_STRIDE]
  61. -    movq    mm2, mm1
  62. -    movq    mm3, mm1
  63. -    movq    mm4, mm1
  64. -    psllq   mm1, 8
  65. -    pxor    mm2, mm1
  66. -    psrlq   mm2, 8
  67. -    pxor    mm3, mm2
  68. -    PRED8x8_LOWPASS mm0, mm1, mm3, mm4, mm5
  69. +%macro PREDICT_4x4_DDL 4
  70. +cglobal predict_4x4_ddl_%1, 1,1
  71. +    mova    m1, [r0-SIZEOF_PIXEL*FDEC_STRIDE]
  72. +    mova    m2, m1
  73. +    mova    m3, m1
  74. +    mova    m4, m1
  75. +    psll%2  m1, %3
  76. +    pxor    m2, m1
  77. +    psrl%2  m2, %3
  78. +    pxor    m3, m2
  79. +    PRED8x8_LOWPASS %4, m0, m1, m3, m4, m5
  80.  
  81.  %assign Y 0
  82.  %rep 4
  83. -    psrlq       mm0, 8
  84. -    movd        [r0+Y*FDEC_STRIDE], mm0
  85. +    psrl%2      m0, %3
  86. +    movh        [r0+SIZEOF_PIXEL*Y*FDEC_STRIDE], m0
  87.  %assign Y (Y+1)
  88.  %endrep
  89.  
  90.      RET
  91. +%endmacro
  92. +
  93. +%ifdef HIGH_BIT_DEPTH
  94. +INIT_XMM
  95. +PREDICT_4x4_DDL sse2  , dq, 2, w
  96. +%else
  97. +INIT_MMX
  98. +PREDICT_4x4_DDL mmxext, q , 8, b
  99. +%endif
  100.  
  101.  ;-----------------------------------------------------------------------------
  102.  ; void predict_4x4_ddr( uint8_t *src )
  103. @@ -166,7 +174,7 @@ cglobal predict_4x4_ddr_%1, 1,1
  104.      PALIGNR   mm3, [r0+2*FDEC_STRIDE-8], 7, mm4
  105.      movq      mm2, mm3
  106.      PALIGNR   mm3, [r0+3*FDEC_STRIDE-8], 7, mm4
  107. -    PRED8x8_LOWPASS mm0, mm3, mm1, mm2, mm4
  108. +    PRED8x8_LOWPASS b, mm0, mm3, mm1, mm2, mm4
  109.  %assign Y 3
  110.      movd    [r0+Y*FDEC_STRIDE], mm0
  111.  %rep 3
  112. @@ -186,7 +194,7 @@ cglobal predict_4x4_vr_%1, 1,1
  113.      PALIGNR mm0, [r0+1*FDEC_STRIDE-8], 7, mm2    ; ..t3t2t1t0ltl0l1
  114.      movq    mm2, mm0
  115.      PALIGNR mm0, [r0+2*FDEC_STRIDE-8], 7, mm3    ; t3t2t1t0ltl0l1l2
  116. -    PRED8x8_LOWPASS mm3, mm1, mm0, mm2, mm4
  117. +    PRED8x8_LOWPASS b, mm3, mm1, mm0, mm2, mm4
  118.      movq    mm1, mm3
  119.      psrlq   mm3, 16
  120.      psllq   mm1, 48
  121. @@ -215,7 +223,7 @@ cglobal predict_4x4_hd_%1, 1,1
  122.      psrlq     mm0, 16                   ; .. .. t2 t1 t0 lt l0 l1
  123.      psrlq     mm2, 8                    ; .. t2 t1 t0 lt l0 l1 l2
  124.      pavgb     mm7, mm2
  125. -    PRED8x8_LOWPASS mm3, mm1, mm0, mm2, mm4
  126. +    PRED8x8_LOWPASS b, mm3, mm1, mm0, mm2, mm4
  127.      punpcklbw mm7, mm3
  128.      psrlq     mm3, 32
  129.      PALIGNR   mm3, mm7, 6, mm6
  130. @@ -230,6 +238,7 @@ cglobal predict_4x4_hd_%1, 1,1
  131.      RET
  132.  %endmacro
  133.  
  134. +INIT_MMX
  135.  %define PALIGNR PALIGNR_MMX
  136.  PREDICT_4x4 mmxext
  137.  %define PALIGNR PALIGNR_SSSE3
  138. @@ -254,7 +263,7 @@ cglobal predict_4x4_hu_mmxext, 1,1
  139.      psrlq     mm2, 16
  140.      psrlq     mm3, 8
  141.      pavgb     mm7, mm3
  142. -    PRED8x8_LOWPASS mm4, mm0, mm2, mm3, mm5
  143. +    PRED8x8_LOWPASS b, mm4, mm0, mm2, mm3, mm5
  144.      punpcklbw mm7, mm4
  145.  %assign Y 0
  146.      movd    [r0+Y*FDEC_STRIDE], mm7
  147. @@ -278,7 +287,7 @@ cglobal predict_4x4_vl_mmxext, 1,1
  148.      movq        mm4, mm3
  149.      pavgb       mm4, mm1
  150.  
  151. -    PRED8x8_LOWPASS mm0, mm1, mm2, mm3, mm5
  152. +    PRED8x8_LOWPASS b, mm0, mm1, mm2, mm3, mm5
  153.  
  154.      movd        [r0+0*FDEC_STRIDE], mm4
  155.      movd        [r0+1*FDEC_STRIDE], mm0
  156. @@ -353,10 +362,10 @@ cglobal predict_8x8_filter_%1, 4,5
  157.      je .fix_lt_1
  158.  .do_left:
  159.      movq        mm0, mm4
  160. -    PRED8x8_LOWPASS mm2, mm1, mm4, mm3, mm5
  161. +    PRED8x8_LOWPASS b, mm2, mm1, mm4, mm3, mm5
  162.      movq     [t1+8], mm2
  163.      movq        mm4, mm0
  164. -    PRED8x8_LOWPASS mm1, mm3, mm0, mm4, mm5
  165. +    PRED8x8_LOWPASS b, mm1, mm3, mm0, mm4, mm5
  166.      movd         t4, mm1
  167.      mov      [t1+7], t4b
  168.  .check_top:
  169. @@ -374,7 +383,7 @@ cglobal predict_8x8_filter_%1, 4,5
  170.      test        r2b, 0x04
  171.      je .fix_tr_1
  172.  .do_top:
  173. -    PRED8x8_LOWPASS mm4, mm2, mm1, mm3, mm5
  174. +    PRED8x8_LOWPASS b, mm4, mm2, mm1, mm3, mm5
  175.      movq    [t1+16], mm4
  176.      test        r3b, 0x04
  177.      je .done
  178. @@ -387,7 +396,7 @@ cglobal predict_8x8_filter_%1, 4,5
  179.      psrlq       mm5, 56
  180.      PALIGNR     mm2, mm3, 7, mm3
  181.      PALIGNR     mm5, mm4, 1, mm4
  182. -    PRED8x8_LOWPASS mm1, mm2, mm5, mm0, mm4
  183. +    PRED8x8_LOWPASS b, mm1, mm2, mm5, mm0, mm4
  184.      jmp .do_topright
  185.  .fix_tr_2:
  186.      punpckhbw   mm3, mm3
  187. @@ -424,6 +433,7 @@ cglobal predict_8x8_filter_%1, 4,5
  188.  %endmacro
  189.  
  190.  %define PALIGNR PALIGNR_MMX
  191. +INIT_MMX
  192.  PREDICT_FILTER mmxext
  193.  %define PALIGNR PALIGNR_SSSE3
  194.  PREDICT_FILTER ssse3
  195. @@ -511,8 +521,8 @@ cglobal predict_8x8_ddl_mmxext, 2,2
  196.      movq        mm1, mm5
  197.      psllq       mm1, 8
  198.      add          r0, FDEC_STRIDE*4
  199. -    PRED8x8_LOWPASS mm0, mm1, mm2, mm5, mm7
  200. -    PRED8x8_LOWPASS mm1, mm3, mm4, [r1+24], mm6
  201. +    PRED8x8_LOWPASS b, mm0, mm1, mm2, mm5, mm7
  202. +    PRED8x8_LOWPASS b, mm1, mm3, mm4, [r1+24], mm6
  203.  %assign Y 3
  204.  %rep 6
  205.      movq        [r0+Y*FDEC_STRIDE], mm1
  206. @@ -540,8 +550,8 @@ cglobal predict_8x8_ddr_mmxext, 2,2
  207.      movq        mm3, [r1+15]
  208.      movq        mm4, [r1+17]
  209.      add          r0, FDEC_STRIDE*4
  210. -    PRED8x8_LOWPASS mm0, mm1, mm2, [r1+8], mm7
  211. -    PRED8x8_LOWPASS mm1, mm3, mm4, [r1+16], mm6
  212. +    PRED8x8_LOWPASS b, mm0, mm1, mm2, [r1+8], mm7
  213. +    PRED8x8_LOWPASS b, mm1, mm3, mm4, [r1+16], mm6
  214.  %assign Y 3
  215.  %rep 6
  216.      movq        [r0+Y*FDEC_STRIDE], mm0
  217. @@ -582,7 +592,7 @@ cglobal predict_8x8_hu_mmxext, 2,2
  218.      punpckhbw mm1, mm1
  219.      por     mm3, mm1            ; l7 l7 l7 l6 l5 l4 l3 l2
  220.      pavgb   mm4, mm2
  221. -    PRED8x8_LOWPASS mm1, mm3, mm5, mm2, mm6
  222. +    PRED8x8_LOWPASS b, mm1, mm3, mm5, mm2, mm6
  223.      movq    mm5, mm4
  224.      punpcklbw mm4, mm1          ; p4 p3 p2 p1
  225.      punpckhbw mm5, mm1          ; p8 p7 p6 p5
  226. @@ -627,7 +637,7 @@ cglobal predict_8x8_vr_core_mmxext, 2,2
  227.      movq        mm4, mm3
  228.      pavgb       mm3, mm2
  229.      add          r0, FDEC_STRIDE*4
  230. -    PRED8x8_LOWPASS mm0, mm1, mm2, mm4, mm7
  231. +    PRED8x8_LOWPASS b, mm0, mm1, mm2, mm4, mm7
  232.  
  233.  %assign Y -4
  234.  %rep 3
  235. @@ -714,6 +724,7 @@ ALIGN 4
  236.  
  237.  %endif ; !ARCH_X86_64
  238.  
  239. +INIT_XMM
  240.  ;-----------------------------------------------------------------------------
  241.  ; void predict_8x8_ddl( uint8_t *src, uint8_t *edge )
  242.  ;-----------------------------------------------------------------------------
  243. @@ -723,7 +734,7 @@ cglobal predict_8x8_ddl_sse2, 2,2
  244.      movdqa      xmm1, xmm3
  245.      pslldq      xmm1, 1
  246.      add          r0, FDEC_STRIDE*4
  247. -    PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm4
  248. +    PRED8x8_LOWPASS b, xmm0, xmm1, xmm2, xmm3, xmm4
  249.  
  250.  %assign Y -4
  251.  %rep 8
  252. @@ -742,7 +753,7 @@ cglobal predict_8x8_ddr_sse2, 2,2
  253.      movdqa      xmm2, xmm3
  254.      psrldq      xmm2, 1
  255.      add           r0, FDEC_STRIDE*4
  256. -    PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm4
  257. +    PRED8x8_LOWPASS b, xmm0, xmm1, xmm2, xmm3, xmm4
  258.  
  259.      movdqa      xmm1, xmm0
  260.      psrldq      xmm1, 1
  261. @@ -771,7 +782,7 @@ cglobal predict_8x8_vl_sse2, 2,2
  262.      pslldq      xmm1, 1
  263.      pavgb       xmm3, xmm2
  264.      add           r0, FDEC_STRIDE*4
  265. -    PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm4, xmm5
  266. +    PRED8x8_LOWPASS b, xmm0, xmm1, xmm2, xmm4, xmm5
  267.  ; xmm0: (t0 + 2*t1 + t2 + 2) >> 2
  268.  ; xmm3: (t0 + t1 + 1) >> 1
  269.  
  270. @@ -802,7 +813,7 @@ cglobal predict_8x8_vr_sse2, 2,2,7
  271.      pslldq      xmm0, 1
  272.      pslldq      xmm1, 2
  273.      pavgb       xmm2, xmm0
  274. -    PRED8x8_LOWPASS_XMM xmm4, xmm3, xmm1, xmm0, xmm5
  275. +    PRED8x8_LOWPASS b, xmm4, xmm3, xmm1, xmm0, xmm5
  276.      pandn       xmm6, xmm4
  277.      movdqa      xmm5, xmm4
  278.      psrlw       xmm4, 8
  279. @@ -824,6 +835,7 @@ cglobal predict_8x8_vr_sse2, 2,2,7
  280.  %endrep
  281.      RET
  282.  
  283. +INIT_MMX
  284.  ;-----------------------------------------------------------------------------
  285.  ; void predict_8x8_hd( uint8_t *src, uint8_t *edge )
  286.  ;-----------------------------------------------------------------------------
  287. @@ -840,12 +852,12 @@ cglobal predict_8x8_hd_mmxext, 2,2
  288.      PALIGNR mm4, mm3, 1, mm7    ; t0 lt l0 l1 l2 l3 l4 l5
  289.      movq    mm5, mm3
  290.      pavgb   mm3, mm1
  291. -    PRED8x8_LOWPASS mm0, mm4, mm1, mm5, mm7
  292. +    PRED8x8_LOWPASS b, mm0, mm4, mm1, mm5, mm7
  293.      movq    mm4, mm2
  294.      movq    mm1, mm2            ; t6 t5 t4 t3 t2 t1 t0 lt
  295.      psrlq   mm4, 16             ; .. .. t6 t5 t4 t3 t2 t1
  296.      psrlq   mm1, 8              ; .. t6 t5 t4 t3 t2 t1 t0
  297. -    PRED8x8_LOWPASS mm6, mm4, mm2, mm1, mm5
  298. +    PRED8x8_LOWPASS b, mm6, mm4, mm2, mm1, mm5
  299.                                  ; .. p11 p10 p9
  300.      movq    mm7, mm3
  301.      punpcklbw mm3, mm0          ; p4 p3 p2 p1
  302. @@ -886,7 +898,7 @@ cglobal predict_8x8_hd_%1, 2,2
  303.      PALIGNR xmm3, xmm0, 8, xmm0
  304.      movdqa  xmm4, xmm1
  305.      pavgb   xmm4, xmm3
  306. -    PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm5
  307. +    PRED8x8_LOWPASS b, xmm0, xmm1, xmm2, xmm3, xmm5
  308.      punpcklbw xmm4, xmm0
  309.      movhlps xmm0, xmm4
  310.  
  311. @@ -946,7 +958,7 @@ cglobal predict_8x8_hu_%1, 2,2
  312.      por       mm3, mm1              ; l7 l7 l7 l6 l5 l4 l3 l2
  313.  %endif
  314.      pavgb     mm4, mm2
  315. -    PRED8x8_LOWPASS mm1, mm3, mm5, mm2, mm6
  316. +    PRED8x8_LOWPASS b, mm1, mm3, mm5, mm2, mm6
  317.  
  318.      movq2dq   xmm0, mm4
  319.      movq2dq   xmm1, mm1
  320. diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
  321. index 8eafcc0..077f2c7 100644
  322. --- a/common/x86/predict-c.c
  323. +++ b/common/x86/predict-c.c
  324. @@ -64,6 +64,7 @@
  325.   void x264_predict_8x8_filter_mmxext( uint8_t *src, uint8_t edge[33], int i_neighbor, int i_filters );
  326.   void x264_predict_8x8_filter_ssse3( uint8_t *src, uint8_t edge[33], int i_neighbor, int i_filters );
  327.   void x264_predict_4x4_ddl_mmxext( uint8_t *src );
  328. + void x264_predict_4x4_ddl_sse2( uint16_t *src );
  329.   void x264_predict_4x4_ddr_mmxext( uint8_t *src );
  330.   void x264_predict_4x4_vl_mmxext( uint8_t *src );
  331.   void x264_predict_4x4_vr_mmxext( uint8_t *src );
  332. @@ -474,7 +475,11 @@ void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
  333.  {
  334.      if( !(cpu&X264_CPU_MMXEXT) )
  335.          return;
  336. -#if !HIGH_BIT_DEPTH
  337. +#if HIGH_BIT_DEPTH
  338. +    if( !(cpu&X264_CPU_SSE2) )
  339. +        return;
  340. +    pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_sse2;
  341. +#else
  342.      pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_mmxext;
  343.      pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_mmxext;
  344.      pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_mmxext;
  345. @@ -487,5 +492,5 @@ void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
  346.      pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_ssse3;
  347.      pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_ssse3;
  348.      pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_ssse3;
  349. -#endif // !HIGH_BIT_DEPTH
  350. +#endif // HIGH_BIT_DEPTH
  351.  }
  352. --
  353. 1.7.2.3
  354.  
  355.  
  356. From ad274f6257b3205448a2dbb4ad77a6d0a51b722b Mon Sep 17 00:00:00 2001
  357. From: Daniel Kang <daniel.d.kang@gmail.com>
  358. Date: Thu, 25 Nov 2010 19:57:30 -0500
  359. Subject: [PATCH 2/8] predict_4x4_hu
  360.  
  361. ---
  362. common/x86/predict-a.asm |   36 +++++++++++++++++++++++++++++++++++-
  363.  common/x86/predict-c.c   |    2 ++
  364.  2 files changed, 37 insertions(+), 1 deletions(-)
  365.  
  366. diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
  367. index 739ce35..f2ab91e 100644
  368. --- a/common/x86/predict-a.asm
  369. +++ b/common/x86/predict-a.asm
  370. @@ -245,8 +245,40 @@ PREDICT_4x4 mmxext
  371.  PREDICT_4x4 ssse3
  372.  
  373.  ;-----------------------------------------------------------------------------
  374. -; void predict_4x4_hu( uint8_t *src )
  375. +; void predict_4x4_hu( pixel *src )
  376.  ;-----------------------------------------------------------------------------
  377. +%ifdef HIGH_BIT_DEPTH
  378. +INIT_XMM
  379. +cglobal predict_4x4_hu_sse2, 1,1,7
  380. +    mova       m0, [r0+2*0*FDEC_STRIDE-8*2]
  381. +    punpckhwd  m0, [r0+2*1*FDEC_STRIDE-8*2]
  382. +    mova       m1, [r0+2*2*FDEC_STRIDE-8*2]
  383. +    punpckhwd  m1, [r0+2*3*FDEC_STRIDE-8*2]
  384. +    punpckhdq  m0, m1
  385. +    mova       m1, m0
  386. +    pshufhw    m1, m1, 0xFF
  387. +    punpckhqdq m1, m1
  388. +    punpckhqdq m0, m1
  389. +    mova       m2, m0
  390. +    mova       m3, m0
  391. +    mova       m6, m0
  392. +    psrldq     m2, 4
  393. +    psrldq     m3, 2
  394. +    pavgw      m6, m3
  395. +    PRED8x8_LOWPASS w, m4, m0, m2, m3, m5
  396. +    punpcklwd  m6, m4
  397. +%assign Y 0
  398. +    movq    [r0+Y*2*FDEC_STRIDE], m6
  399. +%rep 2
  400. +%assign Y (Y+1)
  401. +    psrldq   m6, 4
  402. +    movq    [r0+2*Y*FDEC_STRIDE], m6
  403. +%endrep
  404. +    movq    [r0+2*3*FDEC_STRIDE], m1
  405. +    RET
  406. +
  407. +%else
  408. +INIT_MMX
  409.  cglobal predict_4x4_hu_mmxext, 1,1
  410.      movq      mm0, [r0+0*FDEC_STRIDE-8]
  411.      punpckhbw mm0, [r0+1*FDEC_STRIDE-8]
  412. @@ -274,10 +306,12 @@ cglobal predict_4x4_hu_mmxext, 1,1
  413.  %endrep
  414.      movd    [r0+3*FDEC_STRIDE], mm1
  415.      RET
  416. +%endif ;HIGH_BIT_DEPTH
  417.  
  418.  ;-----------------------------------------------------------------------------
  419.  ; void predict_4x4_vl( uint8_t *src )
  420.  ;-----------------------------------------------------------------------------
  421. +INIT_MMX
  422.  cglobal predict_4x4_vl_mmxext, 1,1
  423.      movq        mm1, [r0-FDEC_STRIDE]
  424.      movq        mm3, mm1
  425. diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
  426. index 077f2c7..8f15e4e 100644
  427. --- a/common/x86/predict-c.c
  428. +++ b/common/x86/predict-c.c
  429. @@ -74,6 +74,7 @@
  430.   void x264_predict_4x4_dc_mmxext( uint8_t *src );
  431.   void x264_predict_4x4_ddr_ssse3( uint8_t *src );
  432.   void x264_predict_4x4_hu_mmxext( uint8_t *src );
  433. + void x264_predict_4x4_hu_sse2( uint16_t *src );
  434.   void x264_predict_16x16_dc_top_sse2( uint8_t *src );
  435.   void x264_predict_16x16_dc_core_sse2( uint8_t *src, int i_dc_left );
  436.   void x264_predict_16x16_dc_left_core_sse2( uint8_t *src, int i_dc_left );
  437. @@ -479,6 +480,7 @@ void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
  438.      if( !(cpu&X264_CPU_SSE2) )
  439.          return;
  440.      pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_sse2;
  441. +    pf[I_PRED_4x4_HU]  = x264_predict_4x4_hu_sse2;
  442.  #else
  443.      pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_mmxext;
  444.      pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_mmxext;
  445. --
  446. 1.7.2.3
  447.  
  448.  
  449. From e5ff6fb0be3ac1a19b9414db5e83589d6c5a0b4a Mon Sep 17 00:00:00 2001
  450. From: Daniel Kang <daniel.d.kang@gmail.com>
  451. Date: Fri, 26 Nov 2010 00:24:28 -0500
  452. Subject: [PATCH 3/8] predict_4x4_vl
  453.  
  454. ---
  455. common/x86/predict-a.asm |   43 ++++++++++++++++++++++++++-----------------
  456.  common/x86/predict-c.c   |    2 ++
  457.  2 files changed, 28 insertions(+), 17 deletions(-)
  458.  
  459. diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
  460. index f2ab91e..acdcca3 100644
  461. --- a/common/x86/predict-a.asm
  462. +++ b/common/x86/predict-a.asm
  463. @@ -309,28 +309,37 @@ cglobal predict_4x4_hu_mmxext, 1,1
  464.  %endif ;HIGH_BIT_DEPTH
  465.  
  466.  ;-----------------------------------------------------------------------------
  467. -; void predict_4x4_vl( uint8_t *src )
  468. -;-----------------------------------------------------------------------------
  469. -INIT_MMX
  470. -cglobal predict_4x4_vl_mmxext, 1,1
  471. -    movq        mm1, [r0-FDEC_STRIDE]
  472. -    movq        mm3, mm1
  473. -    movq        mm2, mm1
  474. -    psrlq       mm3, 8
  475. -    psrlq       mm2, 16
  476. -    movq        mm4, mm3
  477. -    pavgb       mm4, mm1
  478. -
  479. -    PRED8x8_LOWPASS b, mm0, mm1, mm2, mm3, mm5
  480. -
  481. -    movd        [r0+0*FDEC_STRIDE], mm4
  482. -    movd        [r0+1*FDEC_STRIDE], mm0
  483. -    psrlq       mm4, 8
  484. -    psrlq       mm0, 8
  485. -    movd        [r0+2*FDEC_STRIDE], mm4
  486. -    movd        [r0+3*FDEC_STRIDE], mm0
  487. +; void predict_4x4_vl( pixel *src )
  488. +;-----------------------------------------------------------------------------
  489. +%macro PREDICT_4x4_V1 4
  490. +cglobal predict_4x4_vl_%1, 1,1
  491. +    mova        m1, [r0-SIZEOF_PIXEL*FDEC_STRIDE]
  492. +    mova        m3, m1
  493. +    mova        m2, m1
  494. +    psrl%2      m3, %3
  495. +    psrl%2      m2, %3*2
  496. +    mova        m4, m3
  497. +    pavg%4      m4, m1
  498. +
  499. +    PRED8x8_LOWPASS %4, m0, m1, m2, m3, m5
  500. +
  501. +    movh        [r0+SIZEOF_PIXEL*0*FDEC_STRIDE], m4
  502. +    movh        [r0+SIZEOF_PIXEL*1*FDEC_STRIDE], m0
  503. +    psrl%2      m4, %3
  504. +    psrl%2      m0, %3
  505. +    movh        [r0+SIZEOF_PIXEL*2*FDEC_STRIDE], m4
  506. +    movh        [r0+SIZEOF_PIXEL*3*FDEC_STRIDE], m0
  507.  
  508.      RET
  509. +%endmacro
  510. +
  511. +%ifdef HIGH_BIT_DEPTH
  512. +INIT_XMM
  513. +PREDICT_4x4_V1 sse2  , dq, 2, w
  514. +%else
  515. +INIT_MMX
  516. +PREDICT_4x4_V1 mmxext, q , 8, b
  517. +%endif
  518.  
  519.  ;-----------------------------------------------------------------------------
  520.  ; void predict_4x4_dc( uint8_t *src )
  521. diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
  522. index 8f15e4e..75843a0 100644
  523. --- a/common/x86/predict-c.c
  524. +++ b/common/x86/predict-c.c
  525. @@ -67,6 +67,7 @@
  526.   void x264_predict_4x4_ddl_sse2( uint16_t *src );
  527.   void x264_predict_4x4_ddr_mmxext( uint8_t *src );
  528.   void x264_predict_4x4_vl_mmxext( uint8_t *src );
  529. + void x264_predict_4x4_vl_sse2( uint16_t *src );
  530.   void x264_predict_4x4_vr_mmxext( uint8_t *src );
  531.   void x264_predict_4x4_vr_ssse3( uint8_t *src );
  532.   void x264_predict_4x4_hd_mmxext( uint8_t *src );
  533. @@ -481,6 +482,7 @@ void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
  534.          return;
  535.      pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_sse2;
  536.      pf[I_PRED_4x4_HU]  = x264_predict_4x4_hu_sse2;
  537. +    pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_sse2;
  538.  #else
  539.      pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_mmxext;
  540.      pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_mmxext;
  541. --
  542. 1.7.2.3
  543.  
  544.  
  545. From d9df9a5c8ae6d55cf5bb814b99ab98f6ba517b0d Mon Sep 17 00:00:00 2001
  546. From: Daniel Kang <daniel.d.kang@gmail.com>
  547. Date: Fri, 26 Nov 2010 02:29:40 -0500
  548. Subject: [PATCH 4/8] predict_8x8_v
  549.  
  550. ---
  551. common/x86/predict-a.asm |   36 +++++++++++++++++++++++-------------
  552.  common/x86/predict-c.c   |    9 +++++++--
  553.  2 files changed, 30 insertions(+), 15 deletions(-)
  554.  
  555. diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
  556. index acdcca3..8d0ffd7 100644
  557. --- a/common/x86/predict-a.asm
  558. +++ b/common/x86/predict-a.asm
  559. @@ -48,15 +48,15 @@ cextern pw_ff00
  560.  cextern pb_reverse
  561.  
  562.  %macro STORE8x8 2
  563. -    add r0, 4*FDEC_STRIDE
  564. -    movq        [r0 + -4*FDEC_STRIDE], %1
  565. -    movq        [r0 + -3*FDEC_STRIDE], %1
  566. -    movq        [r0 + -2*FDEC_STRIDE], %1
  567. -    movq        [r0 + -1*FDEC_STRIDE], %1
  568. -    movq        [r0 +  0*FDEC_STRIDE], %2
  569. -    movq        [r0 +  1*FDEC_STRIDE], %2
  570. -    movq        [r0 +  2*FDEC_STRIDE], %2
  571. -    movq        [r0 +  3*FDEC_STRIDE], %2
  572. +    add r0, 4*SIZEOF_PIXEL*FDEC_STRIDE
  573. +    mova        [r0 + -4*SIZEOF_PIXEL*FDEC_STRIDE], %1
  574. +    mova        [r0 + -3*SIZEOF_PIXEL*FDEC_STRIDE], %1
  575. +    mova        [r0 + -2*SIZEOF_PIXEL*FDEC_STRIDE], %1
  576. +    mova        [r0 + -1*SIZEOF_PIXEL*FDEC_STRIDE], %1
  577. +    mova        [r0 +  0*SIZEOF_PIXEL*FDEC_STRIDE], %2
  578. +    mova        [r0 +  1*SIZEOF_PIXEL*FDEC_STRIDE], %2
  579. +    mova        [r0 +  2*SIZEOF_PIXEL*FDEC_STRIDE], %2
  580. +    mova        [r0 +  3*SIZEOF_PIXEL*FDEC_STRIDE], %2
  581.  %endmacro
  582.  
  583.  %macro STORE16x16 2
  584. @@ -482,12 +482,22 @@ PREDICT_FILTER mmxext
  585.  PREDICT_FILTER ssse3
  586.  
  587.  ;-----------------------------------------------------------------------------
  588. -; void predict_8x8_v( uint8_t *src, uint8_t *edge )
  589. +; void predict_8x8_v( pixel *src, pixel *edge )
  590.  ;-----------------------------------------------------------------------------
  591. -cglobal predict_8x8_v_mmxext, 2,2
  592. -    movq        mm0, [r1+16]
  593. -    STORE8x8    mm0, mm0
  594. +%macro PREDICT_8x8_V 1
  595. +cglobal predict_8x8_v_%1, 2,2
  596. +    mova        m0, [r1+SIZEOF_PIXEL*16]
  597. +    STORE8x8    m0, m0
  598.      RET
  599. +%endmacro
  600. +
  601. +%ifdef HIGH_BIT_DEPTH
  602. +INIT_XMM
  603. +PREDICT_8x8_V sse2
  604. +%else
  605. +INIT_MMX
  606. +PREDICT_8x8_V mmxext
  607. +%endif
  608.  
  609.  ;-----------------------------------------------------------------------------
  610.  ; void predict_8x8_h( uint8_t *src, uint8_t edge[33] )
  611. diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
  612. index 75843a0..90fd0c7 100644
  613. --- a/common/x86/predict-c.c
  614. +++ b/common/x86/predict-c.c
  615. @@ -44,6 +44,7 @@
  616.   void x264_predict_8x8c_h_mmxext( uint8_t *src );
  617.   void x264_predict_8x8c_h_ssse3( uint8_t *src );
  618.   void x264_predict_8x8_v_mmxext( uint8_t *src, uint8_t edge[33] );
  619. + void x264_predict_8x8_v_sse2( uint16_t *src, uint16_t edge[33] );
  620.   void x264_predict_8x8_h_mmxext( uint8_t *src, uint8_t edge[33] );
  621.   void x264_predict_8x8_hd_mmxext( uint8_t *src, uint8_t edge[33] );
  622.   void x264_predict_8x8_hu_mmxext( uint8_t *src, uint8_t edge[33] );
  623. @@ -443,7 +444,11 @@ void x264_predict_8x8_init_mmx( int cpu, x264_predict8x8_t pf[12], x264_predict_
  624.  {
  625.      if( !(cpu&X264_CPU_MMXEXT) )
  626.          return;
  627. -#if !HIGH_BIT_DEPTH
  628. +#if HIGH_BIT_DEPTH
  629. +    if( !(cpu&X264_CPU_SSE2) )
  630. +        return;
  631. +    pf[I_PRED_8x8_V]      = x264_predict_8x8_v_sse2;
  632. +#else
  633.      pf[I_PRED_8x8_V]      = x264_predict_8x8_v_mmxext;
  634.      pf[I_PRED_8x8_H]      = x264_predict_8x8_h_mmxext;
  635.      pf[I_PRED_8x8_DC]     = x264_predict_8x8_dc_mmxext;
  636. @@ -470,7 +475,7 @@ void x264_predict_8x8_init_mmx( int cpu, x264_predict8x8_t pf[12], x264_predict_
  637.      pf[I_PRED_8x8_HD]   = x264_predict_8x8_hd_ssse3;
  638.      pf[I_PRED_8x8_HU]   = x264_predict_8x8_hu_ssse3;
  639.      *predict_8x8_filter = x264_predict_8x8_filter_ssse3;
  640. -#endif // !HIGH_BIT_DEPTH
  641. +#endif // HIGH_BIT_DEPTH
  642.  }
  643.  
  644.  void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
  645. --
  646. 1.7.2.3
  647.  
  648.  
  649. From 44ec73f438c89ad86c2c73706045ea92da967ec1 Mon Sep 17 00:00:00 2001
  650. From: Daniel Kang <daniel.d.kang@gmail.com>
  651. Date: Fri, 26 Nov 2010 03:30:32 -0500
  652. Subject: [PATCH 5/8] predict_8x8_h
  653.  
  654. ---
  655. common/x86/predict-a.asm |   43 ++++++++++++++++++++++++++-----------------
  656.  common/x86/predict-c.c   |    2 ++
  657.  2 files changed, 28 insertions(+), 17 deletions(-)
  658.  
  659. diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
  660. index 8d0ffd7..cedc3dd 100644
  661. --- a/common/x86/predict-a.asm
  662. +++ b/common/x86/predict-a.asm
  663. @@ -500,31 +500,40 @@ PREDICT_8x8_V mmxext
  664.  %endif
  665.  
  666.  ;-----------------------------------------------------------------------------
  667. -; void predict_8x8_h( uint8_t *src, uint8_t edge[33] )
  668. +; void predict_8x8_h( pixel *src, pixel edge[33] )
  669.  ;-----------------------------------------------------------------------------
  670. -
  671. -INIT_MMX
  672. -cglobal predict_8x8_h_mmxext, 2,2
  673. -    movu   m3, [r1+7]
  674. -    add    r0, FDEC_STRIDE*4
  675. +%macro PREDICT_8x8_H 3
  676. +cglobal predict_8x8_h_%1, 2, 2
  677. +    movu   m3, [r1+SIZEOF_PIXEL*7]
  678. +    add    r0, SIZEOF_PIXEL*FDEC_STRIDE*4
  679.      mova   m7, m3
  680. -    punpckhbw m3, m3
  681. -    punpcklbw m7, m7
  682. -    pshufw m0, m3, 0xff
  683. -    pshufw m1, m3, 0xaa
  684. -    pshufw m2, m3, 0x55
  685. -    pshufw m3, m3, 0x00
  686. -    pshufw m4, m7, 0xff
  687. -    pshufw m5, m7, 0xaa
  688. -    pshufw m6, m7, 0x55
  689. -    pshufw m7, m7, 0x00
  690. +    punpckh%2 m3, m3
  691. +    punpckl%2 m7, m7
  692. +    pshuf%3 m0, m3, 0xff
  693. +    pshuf%3 m1, m3, 0xaa
  694. +    pshuf%3 m2, m3, 0x55
  695. +    pshuf%3 m3, m3, 0x00
  696. +    pshuf%3 m4, m7, 0xff
  697. +    pshuf%3 m5, m7, 0xaa
  698. +    pshuf%3 m6, m7, 0x55
  699. +    pshuf%3 m7, m7, 0x00
  700.  %assign n 0
  701.  %rep 8
  702. -    mova [r0+(n-4)*FDEC_STRIDE], m %+ n
  703. +    mova [r0+(n-4)*SIZEOF_PIXEL*FDEC_STRIDE], m %+ n
  704.  %assign n n+1
  705.  %endrep
  706.      RET
  707. +%endmacro
  708.  
  709. +%ifdef HIGH_BIT_DEPTH
  710. +INIT_XMM
  711. +PREDICT_8x8_H sse2  , wd, d
  712. +%else
  713. +INIT_MMX
  714. +PREDICT_8x8_H mmxext, bw, w
  715. +%endif
  716. +
  717. +INIT_MMX
  718.  ;-----------------------------------------------------------------------------
  719.  ; void predict_8x8_dc( uint8_t *src, uint8_t *edge );
  720.  ;-----------------------------------------------------------------------------
  721. diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
  722. index 90fd0c7..e5a3fa7 100644
  723. --- a/common/x86/predict-c.c
  724. +++ b/common/x86/predict-c.c
  725. @@ -46,6 +46,7 @@
  726.   void x264_predict_8x8_v_mmxext( uint8_t *src, uint8_t edge[33] );
  727.   void x264_predict_8x8_v_sse2( uint16_t *src, uint16_t edge[33] );
  728.   void x264_predict_8x8_h_mmxext( uint8_t *src, uint8_t edge[33] );
  729. + void x264_predict_8x8_h_sse2( uint16_t *src, uint16_t edge[33] );
  730.   void x264_predict_8x8_hd_mmxext( uint8_t *src, uint8_t edge[33] );
  731.   void x264_predict_8x8_hu_mmxext( uint8_t *src, uint8_t edge[33] );
  732.   void x264_predict_8x8_dc_mmxext( uint8_t *src, uint8_t edge[33] );
  733. @@ -448,6 +449,7 @@ void x264_predict_8x8_init_mmx( int cpu, x264_predict8x8_t pf[12], x264_predict_
  734.      if( !(cpu&X264_CPU_SSE2) )
  735.          return;
  736.      pf[I_PRED_8x8_V]      = x264_predict_8x8_v_sse2;
  737. +    pf[I_PRED_8x8_H]      = x264_predict_8x8_h_sse2;
  738.  #else
  739.      pf[I_PRED_8x8_V]      = x264_predict_8x8_v_mmxext;
  740.      pf[I_PRED_8x8_H]      = x264_predict_8x8_h_mmxext;
  741. --
  742. 1.7.2.3
  743.  
  744.  
  745. From 0933e04c859c9f4bcd93dd6e03984d09dde314b0 Mon Sep 17 00:00:00 2001
  746. From: Daniel Kang <daniel.d.kang@gmail.com>
  747. Date: Fri, 26 Nov 2010 13:57:21 -0500
  748. Subject: [PATCH 6/8] predict_4x4_ddr, predict_4x4_vr, predict_4x4_hd
  749.  
  750. ---
  751. common/x86/predict-a.asm |  132 ++++++++++++++++++++++++---------------------
  752.  common/x86/predict-c.c   |    5 ++
  753.  2 files changed, 75 insertions(+), 62 deletions(-)
  754.  
  755. diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
  756. index cedc3dd..6bd845d 100644
  757. --- a/common/x86/predict-a.asm
  758. +++ b/common/x86/predict-a.asm
  759. @@ -160,89 +160,97 @@ PREDICT_4x4_DDL mmxext, q , 8, b
  760.  %endif
  761.  
  762.  ;-----------------------------------------------------------------------------
  763. -; void predict_4x4_ddr( uint8_t *src )
  764. +; void predict_4x4_ddr( pixel *src )
  765.  ;-----------------------------------------------------------------------------
  766. -%macro PREDICT_4x4 1
  767. +%macro PREDICT_4x4 7
  768.  cglobal predict_4x4_ddr_%1, 1,1
  769. -    movq      mm1, [r0+1*FDEC_STRIDE-8]
  770. -    movq      mm2, [r0+0*FDEC_STRIDE-8]
  771. -    punpckhbw mm2, [r0-1*FDEC_STRIDE-8]
  772. -    movd      mm3, [r0-1*FDEC_STRIDE]
  773. -    punpckhwd mm1, mm2
  774. -    PALIGNR   mm3, mm1, 5, mm1
  775. -    movq      mm1, mm3
  776. -    PALIGNR   mm3, [r0+2*FDEC_STRIDE-8], 7, mm4
  777. -    movq      mm2, mm3
  778. -    PALIGNR   mm3, [r0+3*FDEC_STRIDE-8], 7, mm4
  779. -    PRED8x8_LOWPASS b, mm0, mm3, mm1, mm2, mm4
  780. +    mova      m1, [r0+1*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL]
  781. +    mova      m2, [r0+0*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL]
  782. +    punpckh%2 m2, [r0-1*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL]
  783. +    movh      m3, [r0-1*SIZEOF_PIXEL*FDEC_STRIDE]
  784. +    punpckh%3 m1, m2
  785. +    PALIGNR   m3, m1, 5*SIZEOF_PIXEL, m1
  786. +    mova      m1, m3
  787. +    PALIGNR   m3, [r0+2*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m4
  788. +    mova      m2, m3
  789. +    PALIGNR   m3, [r0+3*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m4
  790. +    PRED8x8_LOWPASS %5, m0, m3, m1, m2, m4
  791.  %assign Y 3
  792. -    movd    [r0+Y*FDEC_STRIDE], mm0
  793. +    movh    [r0+Y*SIZEOF_PIXEL*FDEC_STRIDE], m0
  794.  %rep 3
  795.  %assign Y (Y-1)
  796. -    psrlq    mm0, 8
  797. -    movd    [r0+Y*FDEC_STRIDE], mm0
  798. +    psrl%4   m0, %7
  799. +    movh    [r0+Y*SIZEOF_PIXEL*FDEC_STRIDE], m0
  800.  %endrep
  801.      RET
  802.  
  803.  cglobal predict_4x4_vr_%1, 1,1
  804. -    movd    mm0, [r0-1*FDEC_STRIDE]              ; ........t3t2t1t0
  805. -    movq    mm7, mm0
  806. -    PALIGNR mm0, [r0-1*FDEC_STRIDE-8], 7, mm1    ; ......t3t2t1t0lt
  807. -    pavgb   mm7, mm0
  808. -    PALIGNR mm0, [r0+0*FDEC_STRIDE-8], 7, mm1    ; ....t3t2t1t0ltl0
  809. -    movq    mm1, mm0
  810. -    PALIGNR mm0, [r0+1*FDEC_STRIDE-8], 7, mm2    ; ..t3t2t1t0ltl0l1
  811. -    movq    mm2, mm0
  812. -    PALIGNR mm0, [r0+2*FDEC_STRIDE-8], 7, mm3    ; t3t2t1t0ltl0l1l2
  813. -    PRED8x8_LOWPASS b, mm3, mm1, mm0, mm2, mm4
  814. -    movq    mm1, mm3
  815. -    psrlq   mm3, 16
  816. -    psllq   mm1, 48
  817. -    movd   [r0+0*FDEC_STRIDE], mm7
  818. -    movd   [r0+1*FDEC_STRIDE], mm3
  819. -    PALIGNR mm7, mm1, 7, mm2
  820. -    psllq   mm1, 8
  821. -    movd   [r0+2*FDEC_STRIDE], mm7
  822. -    PALIGNR mm3, mm1, 7, mm1
  823. -    movd   [r0+3*FDEC_STRIDE], mm3
  824. +    movh    m0, [r0-1*SIZEOF_PIXEL*FDEC_STRIDE]              ; ........t3t2t1t0
  825. +    mova    m5, m0
  826. +    PALIGNR m0, [r0-1*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m1    ; ......t3t2t1t0lt
  827. +    pavg%5  m5, m0
  828. +    PALIGNR m0, [r0+0*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m1    ; ....t3t2t1t0ltl0
  829. +    mova    m1, m0
  830. +    PALIGNR m0, [r0+1*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m2    ; ..t3t2t1t0ltl0l1
  831. +    mova    m2, m0
  832. +    PALIGNR m0, [r0+2*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m3    ; t3t2t1t0ltl0l1l2
  833. +    PRED8x8_LOWPASS %5, m3, m1, m0, m2, m4
  834. +    mova    m1, m3
  835. +    psrl%4  m3, %7*2
  836. +    psll%4  m1, %7*6
  837. +    movh   [r0+0*SIZEOF_PIXEL*FDEC_STRIDE], m5
  838. +    movh   [r0+1*SIZEOF_PIXEL*FDEC_STRIDE], m3
  839. +    PALIGNR m5, m1, 7*SIZEOF_PIXEL, m2
  840. +    psll%4  m1, %7
  841. +    movh   [r0+2*SIZEOF_PIXEL*FDEC_STRIDE], m5
  842. +    PALIGNR m3, m1, 7*SIZEOF_PIXEL, m1
  843. +    movh   [r0+3*SIZEOF_PIXEL*FDEC_STRIDE], m3
  844.      RET
  845.  
  846. -cglobal predict_4x4_hd_%1, 1,1
  847. -    movd      mm0, [r0-1*FDEC_STRIDE-4] ; lt ..
  848. -    punpckldq mm0, [r0-1*FDEC_STRIDE]   ; t3 t2 t1 t0 lt .. .. ..
  849. -    psllq     mm0, 8                    ; t2 t1 t0 lt .. .. .. ..
  850. -    movq      mm1, [r0+3*FDEC_STRIDE-8] ; l3
  851. -    punpckhbw mm1, [r0+2*FDEC_STRIDE-8] ; l2 l3
  852. -    movq      mm2, [r0+1*FDEC_STRIDE-8] ; l1
  853. -    punpckhbw mm2, [r0+0*FDEC_STRIDE-8] ; l0 l1
  854. -    punpckhwd mm1, mm2                  ; l0 l1 l2 l3
  855. -    punpckhdq mm1, mm0                  ; t2 t1 t0 lt l0 l1 l2 l3
  856. -    movq      mm0, mm1
  857. -    movq      mm2, mm1
  858. -    movq      mm7, mm1
  859. -    psrlq     mm0, 16                   ; .. .. t2 t1 t0 lt l0 l1
  860. -    psrlq     mm2, 8                    ; .. t2 t1 t0 lt l0 l1 l2
  861. -    pavgb     mm7, mm2
  862. -    PRED8x8_LOWPASS b, mm3, mm1, mm0, mm2, mm4
  863. -    punpcklbw mm7, mm3
  864. -    psrlq     mm3, 32
  865. -    PALIGNR   mm3, mm7, 6, mm6
  866. +cglobal predict_4x4_hd_%1, 1,1,7
  867. +    movh      m0, [r0-1*SIZEOF_PIXEL*FDEC_STRIDE-4*SIZEOF_PIXEL] ; lt ..
  868. +    punpckl%6 m0, [r0-1*SIZEOF_PIXEL*FDEC_STRIDE]   ; t3 t2 t1 t0 lt .. .. ..
  869. +    psll%4    m0, %7                    ; t2 t1 t0 lt .. .. .. ..
  870. +    mova      m1, [r0+3*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL] ; l3
  871. +    punpckh%2 m1, [r0+2*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL] ; l2 l3
  872. +    mova      m2, [r0+1*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL] ; l1
  873. +    punpckh%2 m2, [r0+0*SIZEOF_PIXEL*FDEC_STRIDE-8*SIZEOF_PIXEL] ; l0 l1
  874. +    punpckh%3 m1, m2                  ; l0 l1 l2 l3
  875. +    punpckh%6 m1, m0                  ; t2 t1 t0 lt l0 l1 l2 l3
  876. +    mova      m0, m1
  877. +    mova      m2, m1
  878. +    mova      m6, m1
  879. +    psrl%4    m0, %7*2                   ; .. .. t2 t1 t0 lt l0 l1
  880. +    psrl%4    m2, %7                     ; .. t2 t1 t0 lt l0 l1 l2
  881. +    pavg%5    m6, m2
  882. +    PRED8x8_LOWPASS %5, m3, m1, m0, m2, m4
  883. +    punpckl%2 m6, m3
  884. +    psrl%4    m3, %7*4
  885. +    PALIGNR   m3, m6, 6*SIZEOF_PIXEL, m5
  886.  %assign Y 3
  887. -    movd     [r0+Y*FDEC_STRIDE], mm7
  888. +    movh     [r0+Y*SIZEOF_PIXEL*FDEC_STRIDE], m6
  889.  %rep 2
  890.  %assign Y (Y-1)
  891. -    psrlq     mm7, 16
  892. -    movd     [r0+Y*FDEC_STRIDE], mm7
  893. +    psrl%4    m6, %7*2
  894. +    movh     [r0+Y*SIZEOF_PIXEL*FDEC_STRIDE], m6
  895.  %endrep
  896. -    movd     [r0+0*FDEC_STRIDE], mm3
  897. +    movh     [r0+0*SIZEOF_PIXEL*FDEC_STRIDE], m3
  898.      RET
  899.  %endmacro
  900.  
  901. +%ifdef HIGH_BIT_DEPTH
  902. +INIT_XMM
  903. +%define PALIGNR PALIGNR_SSSE3
  904. +PREDICT_4x4 ssse3 , wd, dq, dq, w, qdq, 2
  905. +
  906. +%else
  907.  INIT_MMX
  908.  %define PALIGNR PALIGNR_MMX
  909. -PREDICT_4x4 mmxext
  910. +PREDICT_4x4 mmxext, bw, wd, q , b, dq , 8
  911.  %define PALIGNR PALIGNR_SSSE3
  912. -PREDICT_4x4 ssse3
  913. +PREDICT_4x4 ssse3 , bw, wd, q , b, dq , 8
  914. +
  915. +%endif
  916.  
  917.  ;-----------------------------------------------------------------------------
  918.  ; void predict_4x4_hu( pixel *src )
  919. diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
  920. index e5a3fa7..0d6f7f2 100644
  921. --- a/common/x86/predict-c.c
  922. +++ b/common/x86/predict-c.c
  923. @@ -490,6 +490,11 @@ void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
  924.      pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_sse2;
  925.      pf[I_PRED_4x4_HU]  = x264_predict_4x4_hu_sse2;
  926.      pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_sse2;
  927. +    if( !(cpu&X264_CPU_SSSE3) )
  928. +        return;
  929. +    pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_ssse3;
  930. +    pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_ssse3;
  931. +    pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_ssse3;
  932.  #else
  933.      pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_mmxext;
  934.      pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_mmxext;
  935. --
  936. 1.7.2.3
  937.  
  938.  
  939. From d7d01cb8124f447e05d29c3db0ec40ae456de87c Mon Sep 17 00:00:00 2001
  940. From: Daniel Kang <daniel.d.kang@gmail.com>
  941. Date: Fri, 26 Nov 2010 15:54:40 -0500
  942. Subject: [PATCH 7/8] predict_4x4_dc
  943.  
  944. ---
  945. common/x86/predict-a.asm |   26 ++++++++++++++++++++++++--
  946.  common/x86/predict-c.c   |    4 ++--
  947.  2 files changed, 26 insertions(+), 4 deletions(-)
  948.  
  949. diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
  950. index 6bd845d..fd322f3 100644
  951. --- a/common/x86/predict-a.asm
  952. +++ b/common/x86/predict-a.asm
  953. @@ -350,9 +350,30 @@ PREDICT_4x4_V1 mmxext, q , 8, b
  954.  %endif
  955.  
  956.  ;-----------------------------------------------------------------------------
  957. -; void predict_4x4_dc( uint8_t *src )
  958. +; void predict_4x4_dc( pixel *src )
  959.  ;-----------------------------------------------------------------------------
  960. +%ifdef HIGH_BIT_DEPTH
  961. +INIT_MMX
  962. +cglobal predict_4x4_dc_mmxext, 1,1
  963. +    movq   m0, [r0-SIZEOF_PIXEL*FDEC_STRIDE]
  964. +    HADDW  m0, m1
  965. +%assign n 0
  966. +%rep 4
  967. +    movd   m1, [r0+SIZEOF_PIXEL*(FDEC_STRIDE*n-1)]
  968. +    paddw  m0, m1
  969. +%assign n n+1
  970. +%endrep
  971. +    paddw  m0, [pw_4]
  972. +    psrlw  m0, 3
  973. +    SPLATW m0, m0
  974. +    movq   [r0+SIZEOF_PIXEL*FDEC_STRIDE*0], m0
  975. +    movq   [r0+SIZEOF_PIXEL*FDEC_STRIDE*1], m0
  976. +    movq   [r0+SIZEOF_PIXEL*FDEC_STRIDE*2], m0
  977. +    movq   [r0+SIZEOF_PIXEL*FDEC_STRIDE*3], m0
  978. +    RET
  979.  
  980. +%else
  981. +INIT_MMX
  982.  cglobal predict_4x4_dc_mmxext, 1,4
  983.      pxor   mm7, mm7
  984.      movd   mm0, [r0-FDEC_STRIDE]
  985. @@ -373,12 +394,13 @@ cglobal predict_4x4_dc_mmxext, 1,4
  986.      mov   [r0+FDEC_STRIDE*2], r1d
  987.      mov   [r0+FDEC_STRIDE*3], r1d
  988.      RET
  989. +%endif
  990.  
  991.  %macro PREDICT_FILTER 1
  992.  ;-----------------------------------------------------------------------------
  993.  ;void predict_8x8_filter( uint8_t *src, uint8_t edge[33], int i_neighbor, int i_filters )
  994.  ;-----------------------------------------------------------------------------
  995. -
  996. +INIT_MMX
  997.  cglobal predict_8x8_filter_%1, 4,5
  998.      add          r0, 0x58
  999.  %define src r0-0x58
  1000. diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
  1001. index 0d6f7f2..1a44f3a 100644
  1002. --- a/common/x86/predict-c.c
  1003. +++ b/common/x86/predict-c.c
  1004. @@ -74,7 +74,7 @@
  1005.   void x264_predict_4x4_vr_ssse3( uint8_t *src );
  1006.   void x264_predict_4x4_hd_mmxext( uint8_t *src );
  1007.   void x264_predict_4x4_hd_ssse3( uint8_t *src );
  1008. - void x264_predict_4x4_dc_mmxext( uint8_t *src );
  1009. + void x264_predict_4x4_dc_mmxext( pixel *src );
  1010.   void x264_predict_4x4_ddr_ssse3( uint8_t *src );
  1011.   void x264_predict_4x4_hu_mmxext( uint8_t *src );
  1012.   void x264_predict_4x4_hu_sse2( uint16_t *src );
  1013. @@ -484,6 +484,7 @@ void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
  1014.  {
  1015.      if( !(cpu&X264_CPU_MMXEXT) )
  1016.          return;
  1017. +    pf[I_PRED_4x4_DC]  = x264_predict_4x4_dc_mmxext;
  1018.  #if HIGH_BIT_DEPTH
  1019.      if( !(cpu&X264_CPU_SSE2) )
  1020.          return;
  1021. @@ -499,7 +500,6 @@ void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
  1022.      pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_mmxext;
  1023.      pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_mmxext;
  1024.      pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_mmxext;
  1025. -    pf[I_PRED_4x4_DC]  = x264_predict_4x4_dc_mmxext;
  1026.      pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_mmxext;
  1027.      pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_mmxext;
  1028.      pf[I_PRED_4x4_HU]  = x264_predict_4x4_hu_mmxext;
  1029. --
  1030. 1.7.2.3
  1031.  
  1032.  
  1033. From 51d57538ea8ea9ee729cae9890bcd5164ca16e6d Mon Sep 17 00:00:00 2001
  1034. From: Daniel Kang <daniel.d.kang@gmail.com>
  1035. Date: Fri, 26 Nov 2010 19:41:53 -0500
  1036. Subject: [PATCH 8/8] predict_8x8_dc
  1037.  
  1038. ---
  1039. common/x86/predict-a.asm |   23 +++++++++++++++++++++--
  1040.  common/x86/predict-c.c   |    2 ++
  1041.  2 files changed, 23 insertions(+), 2 deletions(-)
  1042.  
  1043. diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
  1044. index fd322f3..82a4cf7 100644
  1045. --- a/common/x86/predict-a.asm
  1046. +++ b/common/x86/predict-a.asm
  1047. @@ -563,10 +563,27 @@ INIT_MMX
  1048.  PREDICT_8x8_H mmxext, bw, w
  1049.  %endif
  1050.  
  1051. -INIT_MMX
  1052.  ;-----------------------------------------------------------------------------
  1053. -; void predict_8x8_dc( uint8_t *src, uint8_t *edge );
  1054. +; void predict_8x8_dc( pixel *src, pixel *edge );
  1055.  ;-----------------------------------------------------------------------------
  1056. +%ifdef HIGH_BIT_DEPTH
  1057. +INIT_XMM
  1058. +cglobal predict_8x8_dc_sse2, 2,2,3
  1059. +    pxor        m0, m0
  1060. +    movu        m1, [r1+14]
  1061. +    HADDW       m1, m2
  1062. +    paddd       m0, m1
  1063. +    mova        m3, [r1+32]
  1064. +    HADDW       m3, m4
  1065. +    paddd       m0, m3
  1066. +    paddw       m0, [pw_8]
  1067. +    psrlw       m0, 4
  1068. +    SPLATW      m0, m0
  1069. +    STORE8x8    m0, m0
  1070. +    RET
  1071. +
  1072. +%else
  1073. +INIT_MMX
  1074.  cglobal predict_8x8_dc_mmxext, 2,2
  1075.      pxor        mm0, mm0
  1076.      pxor        mm1, mm1
  1077. @@ -579,7 +596,9 @@ cglobal predict_8x8_dc_mmxext, 2,2
  1078.      packuswb    mm0, mm0
  1079.      STORE8x8    mm0, mm0
  1080.      RET
  1081. +%endif
  1082.  
  1083. +INIT_MMX
  1084.  ;-----------------------------------------------------------------------------
  1085.  ; void predict_8x8_dc_top( uint8_t *src, uint8_t *edge );
  1086.  ;-----------------------------------------------------------------------------
  1087. diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
  1088. index 1a44f3a..39cebd9 100644
  1089. --- a/common/x86/predict-c.c
  1090. +++ b/common/x86/predict-c.c
  1091. @@ -50,6 +50,7 @@
  1092.   void x264_predict_8x8_hd_mmxext( uint8_t *src, uint8_t edge[33] );
  1093.   void x264_predict_8x8_hu_mmxext( uint8_t *src, uint8_t edge[33] );
  1094.   void x264_predict_8x8_dc_mmxext( uint8_t *src, uint8_t edge[33] );
  1095. + void x264_predict_8x8_dc_sse2( uint16_t *src, uint16_t edge[33] );
  1096.   void x264_predict_8x8_dc_top_mmxext( uint8_t *src, uint8_t edge[33] );
  1097.   void x264_predict_8x8_dc_left_mmxext( uint8_t *src, uint8_t edge[33] );
  1098.   void x264_predict_8x8_ddl_mmxext( uint8_t *src, uint8_t edge[33] );
  1099. @@ -450,6 +451,7 @@ void x264_predict_8x8_init_mmx( int cpu, x264_predict8x8_t pf[12], x264_predict_
  1100.          return;
  1101.      pf[I_PRED_8x8_V]      = x264_predict_8x8_v_sse2;
  1102.      pf[I_PRED_8x8_H]      = x264_predict_8x8_h_sse2;
  1103. +    pf[I_PRED_8x8_DC]     = x264_predict_8x8_dc_sse2;
  1104.  #else
  1105.      pf[I_PRED_8x8_V]      = x264_predict_8x8_v_mmxext;
  1106.      pf[I_PRED_8x8_H]      = x264_predict_8x8_h_mmxext;
  1107. --
  1108. 1.7.2.3
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!