API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Sep 19th, 2017
442
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Diff 33.62 KB | None | 0 0
raw download clone embed print report
  1. From 489134cf2343edbb65bb17d2ed99dc2c114657bf Mon Sep 17 00:00:00 2001
  2. From: Daniel Kang <daniel.d.kang@gmail.com>
  3. Date: Wed, 22 Jun 2011 17:40:50 -0400
  4. Subject: [PATCH 1/2] luma mc first pass done \o/
  5.  
  6. ---
  7. libavcodec/x86/Makefile              |    1 +
  8.  libavcodec/x86/dsputil_mmx.c         |   47 ++
  9.  libavcodec/x86/h264_qpel_10bit.asm   |  813 ++++++++++++++++++++++++++++++++++
  10.  libavcodec/x86/h264_qpel_mmx_10bit.c |  141 ++++++
  11.  4 files changed, 1002 insertions(+), 0 deletions(-)
  12.  create mode 100644 libavcodec/x86/h264_qpel_10bit.asm
  13.  create mode 100755 libavcodec/x86/h264_qpel_mmx_10bit.c
  14.  
  15. diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
  16. index 022ab27..d3cf0da 100644
  17. --- a/libavcodec/x86/Makefile
  18. +++ b/libavcodec/x86/Makefile
  19. @@ -46,6 +46,7 @@ MMX-OBJS-$(HAVE_YASM)                  += x86/dsputil_yasm.o            \
  20.                                            x86/fmtconvert.o              \
  21.                                            x86/h264_chromamc.o           \
  22.                                            x86/h264_chromamc_10bit.o     \
  23. +                                          x86/h264_qpel_10bit.o         \
  24.                                            $(YASM-OBJS-yes)
  25.  
  26.  MMX-OBJS-$(CONFIG_FFT)                 += x86/fft.o
  27. diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
  28. index 5c5ecb2..43ff26a 100644
  29. --- a/libavcodec/x86/dsputil_mmx.c
  30. +++ b/libavcodec/x86/dsputil_mmx.c
  31. @@ -1896,6 +1896,7 @@ PREFETCH(prefetch_3dnow, prefetch)
  32.  #undef PREFETCH
  33.  
  34.  #include "h264_qpel_mmx.c"
  35. +#include "h264_qpel_mmx_10bit.c"
  36.  
  37.  void ff_put_h264_chroma_mc8_mmx_rnd   (uint8_t *dst, uint8_t *src,
  38.                                         int stride, int h, int x, int y);
  39. @@ -2649,6 +2650,33 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
  40.              SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2);
  41.              SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2);
  42.              }
  43. +#if HAVE_YASM
  44. +#define SET_QPEL_FUNCS_10(PFX, IDX, SIZE, CPU) \
  45. +            c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## SIZE ## _mc00_10_ ## CPU; \
  46. +            c->PFX ## _pixels_tab[IDX][ 1] = ff_ ## PFX ## SIZE ## _mc10_10_ ## CPU; \
  47. +            c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## SIZE ## _mc20_10_ ## CPU; \
  48. +            c->PFX ## _pixels_tab[IDX][ 3] = ff_ ## PFX ## SIZE ## _mc30_10_ ## CPU; \
  49. +            c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## SIZE ## _mc01_10_ ## CPU; \
  50. +            c->PFX ## _pixels_tab[IDX][ 5] = ff_ ## PFX ## SIZE ## _mc11_10_ ## CPU; \
  51. +            c->PFX ## _pixels_tab[IDX][ 6] = ff_ ## PFX ## SIZE ## _mc21_10_ ## CPU; \
  52. +            c->PFX ## _pixels_tab[IDX][ 7] = ff_ ## PFX ## SIZE ## _mc31_10_ ## CPU; \
  53. +            c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## SIZE ## _mc02_10_ ## CPU; \
  54. +            c->PFX ## _pixels_tab[IDX][ 9] = ff_ ## PFX ## SIZE ## _mc12_10_ ## CPU; \
  55. +            c->PFX ## _pixels_tab[IDX][10] = ff_ ## PFX ## SIZE ## _mc22_10_ ## CPU; \
  56. +            c->PFX ## _pixels_tab[IDX][11] = ff_ ## PFX ## SIZE ## _mc32_10_ ## CPU; \
  57. +            c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## SIZE ## _mc03_10_ ## CPU; \
  58. +            c->PFX ## _pixels_tab[IDX][13] = ff_ ## PFX ## SIZE ## _mc13_10_ ## CPU; \
  59. +            c->PFX ## _pixels_tab[IDX][14] = ff_ ## PFX ## SIZE ## _mc23_10_ ## CPU; \
  60. +            c->PFX ## _pixels_tab[IDX][15] = ff_ ## PFX ## SIZE ## _mc33_10_ ## CPU
  61. +            else if (bit_depth == 10) {
  62. +                SET_QPEL_FUNCS_10(put_h264_qpel, 0, 16, mmxext);
  63. +                SET_QPEL_FUNCS_10(put_h264_qpel, 1, 8,  mmxext);
  64. +                SET_QPEL_FUNCS_10(put_h264_qpel, 2, 4,  mmxext);
  65. +                SET_QPEL_FUNCS_10(avg_h264_qpel, 0, 16, mmxext);
  66. +                SET_QPEL_FUNCS_10(avg_h264_qpel, 1, 8,  mmxext);
  67. +                SET_QPEL_FUNCS_10(avg_h264_qpel, 2, 4,  mmxext);
  68. +            }
  69. +#endif
  70.  
  71.              SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2);
  72.              SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2);
  73. @@ -2777,7 +2805,26 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
  74.              H264_QPEL_FUNCS(3, 3, sse2);
  75.              }
  76.  #if HAVE_YASM
  77. +#define H264_QPEL_FUNCS_10(x, y, CPU)\
  78. +            c->put_h264_qpel_pixels_tab[0][x+y*4] = ff_put_h264_qpel16_mc##x##y##_10_##CPU;\
  79. +            c->put_h264_qpel_pixels_tab[1][x+y*4] = ff_put_h264_qpel8_mc##x##y##_10_##CPU;\
  80. +            c->avg_h264_qpel_pixels_tab[0][x+y*4] = ff_avg_h264_qpel16_mc##x##y##_10_##CPU;\
  81. +            c->avg_h264_qpel_pixels_tab[1][x+y*4] = ff_avg_h264_qpel8_mc##x##y##_10_##CPU;
  82.              if (bit_depth == 10) {
  83. +                H264_QPEL_FUNCS_10(0, 0, sse2);
  84. +                H264_QPEL_FUNCS_10(0, 1, sse2);
  85. +                H264_QPEL_FUNCS_10(0, 2, sse2);
  86. +                H264_QPEL_FUNCS_10(0, 3, sse2);
  87. +                H264_QPEL_FUNCS_10(1, 1, sse2);
  88. +                H264_QPEL_FUNCS_10(1, 2, sse2);
  89. +                H264_QPEL_FUNCS_10(1, 3, sse2);
  90. +                H264_QPEL_FUNCS_10(2, 1, sse2);
  91. +                H264_QPEL_FUNCS_10(2, 2, sse2);
  92. +                H264_QPEL_FUNCS_10(2, 3, sse2);
  93. +                H264_QPEL_FUNCS_10(3, 1, sse2);
  94. +                H264_QPEL_FUNCS_10(3, 2, sse2);
  95. +                H264_QPEL_FUNCS_10(3, 3, sse2);
  96. +
  97.                  c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_sse2;
  98.                  c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_sse2;
  99.              }
  100. diff --git a/libavcodec/x86/h264_qpel_10bit.asm b/libavcodec/x86/h264_qpel_10bit.asm
  101. new file mode 100644
  102. index 0000000..fc44e85
  103. --- /dev/null
  104. +++ b/libavcodec/x86/h264_qpel_10bit.asm
  105. @@ -0,0 +1,813 @@
  106. +;*****************************************************************************
  107. +;* MMX/SSE2/AVX-optimized 10-bit H.264 qpel code
  108. +;*****************************************************************************
  109. +;* Copyright (C) 2005-2011 x264 project
  110. +;*
  111. +;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
  112. +;*
  113. +;* This file is part of Libav.
  114. +;*
  115. +;* Libav is free software; you can redistribute it and/or
  116. +;* modify it under the terms of the GNU Lesser General Public
  117. +;* License as published by the Free Software Foundation; either
  118. +;* version 2.1 of the License, or (at your option) any later version.
  119. +;*
  120. +;* Libav is distributed in the hope that it will be useful,
  121. +;* but WITHOUT ANY WARRANTY; without even the implied warranty of
  122. +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  123. +;* Lesser General Public License for more details.
  124. +;*
  125. +;* You should have received a copy of the GNU Lesser General Public
  126. +;* License along with Libav; if not, write to the Free Software
  127. +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  128. +;******************************************************************************
  129. +
  130. +%include "x86inc.asm"
  131. +%include "x86util.asm"
  132. +
  133. +SECTION_RODATA 32
  134. +
  135. +cextern pw_16
  136. +cextern pw_1
  137. +cextern pb_0
  138. +
  139. +pw_pixel_max: times 8 dw ((1 << 10)-1)
  140. +
  141. +pad10: times 8 dw 10*1023
  142. +pad20: times 8 dw 20*1023
  143. +pad30: times 8 dw 30*1023
  144. +depad: times 4 dd 32*20*1023 + 512
  145. +depad2: times 8 dw 20*1023 + 16*1022 + 16
  146. +unpad: times 8 dw 16*1022/32 ; needs to be mod 16
  147. +
  148. +tap1: times 4 dw  1, -5
  149. +tap2: times 4 dw 20, 20
  150. +tap3: times 4 dw -5,  1
  151. +pd_0f: times 4 dd 0xffff
  152. +
  153. +SECTION .text
  154. +
  155. +; All of  the 2x2 functions are probably no faster than the C version.
  156. +
  157. +;-----------------------------------------------------------------------------
  158. +; void h264_qpel_mc00(uint8_t *dst, uint8_t *src, int stride)
  159. +;-----------------------------------------------------------------------------
  160. +%macro MC00 1
  161. +INIT_MMX
  162. +cglobal %1_h264_qpel4_mc00_10_mmxext, 3,4
  163. +    lea           r3, [r2*3]
  164. +    movq          m0, [r1     ]
  165. +    OP_MOV [r0     ], m0
  166. +    movq          m0, [r1+r2  ]
  167. +    OP_MOV [r0+r2  ], m0
  168. +    movq          m0, [r1+r2*2]
  169. +    OP_MOV [r0+r2*2], m0
  170. +    movq          m0, [r1+r3  ]
  171. +    OP_MOV [r0+r3  ], m0
  172. +    RET
  173. +
  174. +INIT_XMM
  175. +cglobal %1_h264_qpel8_mc00_10_sse2, 3,3
  176. +%rep 4
  177. +    movu        m0, [r1   ]
  178. +    OP_MOV [r0   ], m0
  179. +    movu        m0, [r1+r2]
  180. +    OP_MOV [r0+r2], m0
  181. +    lea         r0, [r0+r2*2]
  182. +    lea         r1, [r1+r2*2]
  183. +%endrep
  184. +    RET
  185. +
  186. +cglobal %1_h264_qpel16_mc00_10_sse2, 3,3
  187. +%rep 8
  188. +    movu           m0, [r1      ]
  189. +    movu           m1, [r1   +16]
  190. +    OP_MOV [r0      ], m0
  191. +    OP_MOV [r0   +16], m1
  192. +    movu           m0, [r1+r2   ]
  193. +    movu           m1, [r1+r2+16]
  194. +    OP_MOV [r0+r2   ], m0
  195. +    OP_MOV [r0+r2+16], m1
  196. +    lea            r0, [r0+r2*2]
  197. +    lea            r1, [r1+r2*2]
  198. +%endrep
  199. +    RET
  200. +%endmacro
  201. +
  202. +%macro AVG_MOV 2
  203. +    pavgw %2, %1
  204. +    mova  %1, %2
  205. +%endmacro
  206. +
  207. +%define OP_MOV mova
  208. +MC00 put
  209. +
  210. +%define OP_MOV AVG_MOV
  211. +MC00 avg
  212. +
  213. +;-----------------------------------------------------------------------------
  214. +; void h264_qpel_mc20(uint8_t *dst, uint8_t *src, int stride)
  215. +;-----------------------------------------------------------------------------
  216. +%macro FILT_H 4
  217. +    paddw  %1, %4
  218. +    psubw  %1, %2  ; a-b
  219. +    psraw  %1, 2   ; (a-b)/4
  220. +    psubw  %1, %2  ; (a-b)/4-b
  221. +    paddw  %1, %3  ; (a-b)/4-b+c
  222. +    psraw  %1, 2   ; ((a-b)/4-b+c)/4
  223. +    paddw  %1, %3  ; ((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16
  224. +%endmacro
  225. +
  226. +%macro ADDW 3
  227. +%if mmsize == 8
  228. +    paddw %1, %2
  229. +%else
  230. +    movu  %3, %2
  231. +    paddw %1, %3
  232. +%endif
  233. +%endmacro
  234. +
  235. +%macro MC20 3
  236. +cglobal %2_h264_qpel%3_mc20_10_%1, 3,4,7
  237. +    mov     r3d, %3
  238. +    pxor     m0, m0
  239. +    mova     m1, [pw_pixel_max]
  240. +    mova     m6, [pw_16]
  241. +.nextrow
  242. +    movu     m2, [r1-4]
  243. +    movu     m3, [r1-2]
  244. +    movu     m4, [r1+0]
  245. +    ADDW     m4, [r1+2], m5
  246. +    ADDW     m3, [r1+4], m5
  247. +    ADDW     m2, [r1+6], m5
  248. +
  249. +    FILT_H   m2, m3, m4, m6
  250. +    psraw    m2, 1
  251. +    CLIPW    m2, m0, m1
  252. +    OP_MOV [r0], m2
  253. +    add      r0, r2
  254. +    add      r1, r2
  255. +    dec     r3d
  256. +    jg .nextrow
  257. +    REP_RET
  258. +%endmacro
  259. +
  260. +%define OP_MOV mova
  261. +INIT_MMX
  262. +MC20 mmxext, put, 4
  263. +INIT_XMM
  264. +MC20 sse2  , put, 8
  265. +
  266. +%define OP_MOV AVG_MOV
  267. +INIT_MMX
  268. +MC20 mmxext, avg, 4
  269. +INIT_XMM
  270. +MC20 sse2  , avg, 8
  271. +
  272. +;-----------------------------------------------------------------------------
  273. +; void h264_qpel_mc30(uint8_t *dst, uint8_t *src, int stride)
  274. +;-----------------------------------------------------------------------------
  275. +%macro MC30 3
  276. +cglobal %2_h264_qpel%3_mc30_10_%1, 3,5,6
  277. +    lea r4, [r1+2]
  278. +    jmp mangle(ff_%2_h264_qpel%3_mc10_10_%1.body)
  279. +%endmacro
  280. +
  281. +%define OP_MOV mova
  282. +INIT_MMX
  283. +MC30 mmxext, put, 4
  284. +INIT_XMM
  285. +MC30 sse2  , put, 8
  286. +
  287. +%define OP_MOV AVG_MOV
  288. +INIT_MMX
  289. +MC30 mmxext, avg, 4
  290. +INIT_XMM
  291. +MC30 sse2  , avg, 8
  292. +
  293. +;-----------------------------------------------------------------------------
  294. +; void h264_qpel_mc10(uint8_t *dst, uint8_t *src, int stride)
  295. +;-----------------------------------------------------------------------------
  296. +%macro MC10 3
  297. +cglobal %2_h264_qpel%3_mc10_10_%1, 3,5,7
  298. +    mov      r4, r1
  299. +.body
  300. +    mov     r3d, %3
  301. +    pxor     m0, m0
  302. +    mova     m1, [pw_pixel_max]
  303. +    mova     m6, [pw_16]
  304. +.nextrow
  305. +    movu     m2, [r1-4]
  306. +    movu     m3, [r1-2]
  307. +    movu     m4, [r1+0]
  308. +    ADDW     m4, [r1+2], m5
  309. +    ADDW     m3, [r1+4], m5
  310. +    ADDW     m2, [r1+6], m5
  311. +
  312. +    FILT_H   m2, m3, m4, m6
  313. +    psraw    m2, 1
  314. +    CLIPW    m2, m0, m1
  315. +    movu     m3, [r4]
  316. +    pavgw    m2, m3
  317. +    OP_MOV [r0], m2
  318. +    add      r0, r2
  319. +    add      r1, r2
  320. +    add      r4, r2
  321. +    dec     r3d
  322. +    jg .nextrow
  323. +    REP_RET
  324. +%endmacro
  325. +
  326. +%define OP_MOV mova
  327. +INIT_MMX
  328. +MC10 mmxext, put, 4
  329. +INIT_XMM
  330. +MC10 sse2  , put, 8
  331. +
  332. +%define OP_MOV AVG_MOV
  333. +INIT_MMX
  334. +MC10 mmxext, avg, 4
  335. +INIT_XMM
  336. +MC10 sse2  , avg, 8
  337. +
  338. +;-----------------------------------------------------------------------------
  339. +; void h264_qpel_mc02(uint8_t *dst, uint8_t *src, int stride)
  340. +;-----------------------------------------------------------------------------
  341. +%macro FILT_V 8
  342. +    movu     %6, [r1]
  343. +    paddw    %1, %6
  344. +    mova     %7, %2
  345. +    paddw    %7, %5
  346. +    mova     %8, %3
  347. +    paddw    %8, %4
  348. +    FILT_H   %1, %7, %8, [pw_16]
  349. +    psraw    %1, 1
  350. +    CLIPW    %1, [pb_0], [pw_pixel_max]
  351. +%endmacro
  352. +
  353. +%macro MC02 3
  354. +cglobal %2_h264_qpel%3_mc02_10_%1, 3,4,8
  355. +    lea      r3, [r2*2]
  356. +    sub      r1, r3
  357. +    movu     m0, [r1]
  358. +    movu     m1, [r1+r2]
  359. +    add      r1, r3
  360. +    movu     m2, [r1]
  361. +    movu     m3, [r1+r2]
  362. +    add      r1, r3
  363. +    movu     m4, [r1]
  364. +    add      r1, r2
  365. +
  366. +%rep %3-1
  367. +    FILT_V   m0, m1, m2, m3, m4, m5, m6, m7
  368. +    OP_MOV [r0], m0
  369. +    add      r1, r2
  370. +    add      r0, r2
  371. +    SWAP 0,1,2,3,4,5
  372. +%endrep
  373. +    FILT_V   m0, m1, m2, m3, m4, m5, m6, m7
  374. +    OP_MOV [r0], m0
  375. +    RET
  376. +%endmacro
  377. +
  378. +%define OP_MOV mova
  379. +INIT_MMX
  380. +MC02 mmxext, put, 4
  381. +INIT_XMM
  382. +MC02 sse2  , put, 8
  383. +
  384. +%define OP_MOV AVG_MOV
  385. +INIT_MMX
  386. +MC02 mmxext, avg, 4
  387. +INIT_XMM
  388. +MC02 sse2  , avg, 8
  389. +
  390. +;-----------------------------------------------------------------------------
  391. +; void h264_qpel_mc01(uint8_t *dst, uint8_t *src, int stride)
  392. +;-----------------------------------------------------------------------------
  393. +%macro MC01 3
  394. +cglobal %2_h264_qpel%3_mc01_10_%1, 3,5,8
  395. +    mov      r4, r1
  396. +.body
  397. +    lea      r3, [r2*2]
  398. +    sub      r1, r3
  399. +    movu     m0, [r1]
  400. +    movu     m1, [r1+r2]
  401. +    add      r1, r3
  402. +    movu     m2, [r1]
  403. +    movu     m3, [r1+r2]
  404. +    add      r1, r3
  405. +    movu     m4, [r1]
  406. +    add      r1, r2
  407. +
  408. +%rep %3-1
  409. +    FILT_V   m0, m1, m2, m3, m4, m5, m6, m7
  410. +    movu     m7, [r4]
  411. +    pavgw    m0, m7
  412. +    OP_MOV [r0], m0
  413. +    add      r4, r2
  414. +    add      r1, r2
  415. +    add      r0, r2
  416. +    SWAP 0,1,2,3,4,5
  417. +%endrep
  418. +    FILT_V   m0, m1, m2, m3, m4, m5, m6, m7
  419. +    movu     m7, [r4]
  420. +    pavgw    m0, m7
  421. +    OP_MOV [r0], m0
  422. +    RET
  423. +%endmacro
  424. +
  425. +%define OP_MOV mova
  426. +INIT_MMX
  427. +MC01 mmxext, put, 4
  428. +INIT_XMM
  429. +MC01 sse2  , put, 8
  430. +
  431. +%define OP_MOV AVG_MOV
  432. +INIT_MMX
  433. +MC01 mmxext, avg, 4
  434. +INIT_XMM
  435. +MC01 sse2  , avg, 8
  436. +
  437. +;-----------------------------------------------------------------------------
  438. +; void h264_qpel_mc03(uint8_t *dst, uint8_t *src, int stride)
  439. +;-----------------------------------------------------------------------------
  440. +%macro MC03 3
  441. +cglobal %2_h264_qpel%3_mc03_10_%1, 3,5,8
  442. +    lea r4, [r1+r2]
  443. +    jmp mangle(ff_%2_h264_qpel%3_mc01_10_%1.body)
  444. +%endmacro
  445. +
  446. +%define OP_MOV mova
  447. +INIT_MMX
  448. +MC03 mmxext, put, 4
  449. +INIT_XMM
  450. +MC03 sse2  , put, 8
  451. +
  452. +%define OP_MOV AVG_MOV
  453. +INIT_MMX
  454. +MC03 mmxext, avg, 4
  455. +INIT_XMM
  456. +MC03 sse2  , avg, 8
  457. +
  458. +;-----------------------------------------------------------------------------
  459. +; void h264_qpel_mc11(uint8_t *dst, uint8_t *src, int stride)
  460. +;-----------------------------------------------------------------------------
  461. +%macro MC11 3
  462. +; this REALLY needs x86_64
  463. +cglobal %2_h264_qpel%3_mc11_10_%1, 3,5,8
  464. +    mov      r4, r1
  465. +.body
  466. +    lea      r3, [r2*2]
  467. +    sub      r1, r3
  468. +    movu     m0, [r1]
  469. +    movu     m1, [r1+r2]
  470. +    add      r1, r3
  471. +    movu     m2, [r1]
  472. +    movu     m3, [r1+r2]
  473. +    add      r1, r3
  474. +    movu     m4, [r1]
  475. +    add      r1, r2
  476. +
  477. +%assign i 0
  478. +%rep %3
  479. +%assign i i+1
  480. +    FILT_V   m0, m1, m2, m3, m4, m5, m6, m7
  481. +;now do FILT_H with fewer registers. probably faster than doing FILT_V then FILT_H
  482. +;m6,m7 are tmp registers, m0 is the FILT_V result, the rest are to be used next in the next iteration
  483. +;unfortunately I need three registers, so m5 will have to be re-read from memory
  484. +    movu     m5, [r4-4]
  485. +    ADDW     m5, [r4+6], m7
  486. +    movu     m6, [r4-2]
  487. +    ADDW     m6, [r4+4], m7
  488. +    paddw    m5, [pw_16]
  489. +    psubw    m5, m6  ; a-b
  490. +    psraw    m5, 2   ; (a-b)/4
  491. +    psubw    m5, m6  ; (a-b)/4-b
  492. +;now I need to load c...
  493. +    movu     m6, [r4+0]
  494. +    ADDW     m6, [r4+2], m7
  495. +    paddw    m5, m6  ; (a-b)/4-b+c
  496. +    psraw    m5, 2   ; ((a-b)/4-b+c)/4
  497. +    paddw    m5, m6  ; ((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16
  498. +    psraw    m5, 1
  499. +    CLIPW    m5, [pb_0], [pw_pixel_max]
  500. +;avg FILT_V, FILT_H and reload m5
  501. +    pavgw    m0, m5
  502. +    OP_MOV [r0], m0
  503. +%if i<%3
  504. +    movu     m5, [r1]
  505. +    add      r4, r2
  506. +    add      r1, r2
  507. +    add      r0, r2
  508. +    SWAP 0,1,2,3,4,5
  509. +%endif
  510. +%endrep
  511. +    RET
  512. +%endmacro
  513. +
  514. +%define OP_MOV mova
  515. +INIT_MMX
  516. +MC11 mmxext, put, 4
  517. +INIT_XMM
  518. +MC11 sse2  , put, 8
  519. +
  520. +%define OP_MOV AVG_MOV
  521. +INIT_MMX
  522. +MC11 mmxext, avg, 4
  523. +INIT_XMM
  524. +MC11 sse2  , avg, 8
  525. +
  526. +;-----------------------------------------------------------------------------
  527. +; void h264_qpel_mc31(uint8_t *dst, uint8_t *src, int stride)
  528. +;-----------------------------------------------------------------------------
  529. +%macro MC31 3
  530. +cglobal %2_h264_qpel%3_mc31_10_%1, 3,5,8
  531. +    mov r4, r1
  532. +    add r1, 2
  533. +    jmp mangle(ff_%2_h264_qpel%3_mc11_10_%1.body)
  534. +%endmacro
  535. +
  536. +%define OP_MOV mova
  537. +INIT_MMX
  538. +MC31 mmxext, put, 4
  539. +INIT_XMM
  540. +MC31 sse2  , put, 8
  541. +
  542. +%define OP_MOV AVG_MOV
  543. +INIT_MMX
  544. +MC31 mmxext, avg, 4
  545. +INIT_XMM
  546. +MC31 sse2  , avg, 8
  547. +
  548. +;-----------------------------------------------------------------------------
  549. +; void h264_qpel_mc13(uint8_t *dst, uint8_t *src, int stride)
  550. +;-----------------------------------------------------------------------------
  551. +%macro MC13 3
  552. +cglobal %2_h264_qpel%3_mc13_10_%1, 3,5,8
  553. +    lea r4, [r1+r2]
  554. +    jmp mangle(ff_%2_h264_qpel%3_mc11_10_%1.body)
  555. +%endmacro
  556. +
  557. +%define OP_MOV mova
  558. +INIT_MMX
  559. +MC13 mmxext, put, 4
  560. +INIT_XMM
  561. +MC13 sse2  , put, 8
  562. +
  563. +%define OP_MOV AVG_MOV
  564. +INIT_MMX
  565. +MC13 mmxext, avg, 4
  566. +INIT_XMM
  567. +MC13 sse2  , avg, 8
  568. +
  569. +;-----------------------------------------------------------------------------
  570. +; void h264_qpel_mc33(uint8_t *dst, uint8_t *src, int stride)
  571. +;-----------------------------------------------------------------------------
  572. +%macro MC33 3
  573. +cglobal %2_h264_qpel%3_mc33_10_%1, 3,5,8
  574. +    lea r4, [r1+r2]
  575. +    add r1, 2
  576. +    jmp mangle(ff_%2_h264_qpel%3_mc11_10_%1.body)
  577. +%endmacro
  578. +
  579. +%define OP_MOV mova
  580. +INIT_MMX
  581. +MC33 mmxext, put, 4
  582. +INIT_XMM
  583. +MC33 sse2  , put, 8
  584. +
  585. +%define OP_MOV AVG_MOV
  586. +INIT_MMX
  587. +MC33 mmxext, avg, 4
  588. +INIT_XMM
  589. +MC33 sse2  , avg, 8
  590. +
  591. +
  592. +
  593. +;-----------------------------------------------------------------------------
  594. +; void h264_qpel_mc22(uint8_t *dst, uint8_t *src, int stride)
  595. +;-----------------------------------------------------------------------------
  596. +%macro FILT_H2 3
  597. +    psubw  %1, %2  ; a-b
  598. +    psubw  %2, %3  ; b-c
  599. +    psllw  %2, 2
  600. +    psubw  %1, %2  ; a-5*b+4*c
  601. +    psllw  %3, 4
  602. +    paddw  %1, %3  ; a-5*b+20*c
  603. +%endmacro
  604. +
  605. +%macro FILT_VNRD 8
  606. +    movu     %6, [r1]
  607. +    paddw    %1, %6
  608. +    mova     %7, %2
  609. +    paddw    %7, %5
  610. +    mova     %8, %3
  611. +    paddw    %8, %4
  612. +    FILT_H2  %1, %7, %8
  613. +%endmacro
  614. +
  615. +%macro MC22 3
  616. +%2_hv%3_10_%1:
  617. +    add     rsp, gprsize
  618. +    neg      r2           ; This actually saves instructions
  619. +    lea      r1, [r1+r2*2]
  620. +    sub      r1, mmsize
  621. +    xor      r4, r4
  622. +    mov      r3, 3
  623. +.v_loop:
  624. +    movu     m0, [r1]
  625. +    sub      r1, r2
  626. +    movu     m1, [r1]
  627. +    sub      r1, r2
  628. +    movu     m2, [r1]
  629. +    sub      r1, r2
  630. +    movu     m3, [r1]
  631. +    sub      r1, r2
  632. +    movu     m4, [r1]
  633. +    sub      r1, r2
  634. +%assign i 0
  635. +%rep %3-1
  636. +    FILT_VNRD m0, m1, m2, m3, m4, m5, m6, m7
  637. +    psubw    m0, [pad20]
  638. +    mova     [rsp+r4+i*mmsize*3], m0
  639. +    sub      r1, r2
  640. +    SWAP 0,1,2,3,4,5
  641. +%assign i i+1
  642. +%endrep
  643. +    FILT_VNRD m0, m1, m2, m3, m4, m5, m6, m7
  644. +    psubw    m0, [pad20]
  645. +    mova     [rsp+r4+i*mmsize*3], m0
  646. +    add      r4, mmsize
  647. +    lea      r1, [r1+r2*8+mmsize]
  648. +%if %3==8
  649. +    lea      r1, [r1+r2*4]
  650. +%endif
  651. +    dec      r3
  652. +    jg .v_loop
  653. +    sub     rsp, gprsize
  654. +    neg      r2
  655. +    ret
  656. +
  657. +cglobal %2_h264_qpel%3_mc22_10_%1, 3,7,10
  658. +    mov      r6, rsp          ; backup stack pointer
  659. +    and     rsp, ~(mmsize-1)  ; align stack
  660. +    sub     rsp, 4096         ; TODO: calculate this correctly
  661. +
  662. +    call %2_hv%3_10_%1
  663. +
  664. +    mov        r4, mmsize
  665. +    mov       r3d, %3
  666. +    mova       m0, [tap1]
  667. +    mova       m7, [tap3]
  668. +%if num_mmregs > 8
  669. +    mova       m8, [tap2]
  670. +    mova       m9, [depad]
  671. +    %define s1 m8
  672. +    %define s2 m9
  673. +%else
  674. +    %define s1 [tap2]
  675. +    %define s2 [depad]
  676. +%endif
  677. +.h_loop:
  678. +    movu       m1, [rsp+r4-4]
  679. +    movu       m2, [rsp+r4-2]
  680. +    mova       m3, [rsp+r4+0]
  681. +    movu       m4, [rsp+r4+2]
  682. +    movu       m5, [rsp+r4+4]
  683. +    movu       m6, [rsp+r4+6]
  684. +    pmaddwd    m1, m0
  685. +    pmaddwd    m2, m0
  686. +    pmaddwd    m3, s1
  687. +    pmaddwd    m4, s1
  688. +    pmaddwd    m5, m7
  689. +    pmaddwd    m6, m7
  690. +    paddd      m1, s2
  691. +    paddd      m2, s2
  692. +    paddd      m3, m5
  693. +    paddd      m4, m6
  694. +    paddd      m1, m3
  695. +    paddd      m2, m4
  696. +    psrad      m1, 10
  697. +    psrad      m2, 10
  698. +    pslld      m2, 16
  699. +    pand       m1, [pd_0f]
  700. +    por        m1, m2
  701. +    CLIPW      m1, [pb_0], [pw_pixel_max]
  702. +    OP_MOV   [r0], m1
  703. +    add        r4, mmsize*3
  704. +    add        r0, r2
  705. +    dec       r3d
  706. +    jg .h_loop
  707. +
  708. +    mov     rsp, r6          ; restore stack pointer
  709. +    RET
  710. +%endmacro
  711. +
  712. +%define OP_MOV mova
  713. +INIT_MMX
  714. +MC22 mmxext, put, 4
  715. +INIT_XMM
  716. +MC22 sse2  , put, 8
  717. +
  718. +%define OP_MOV AVG_MOV
  719. +INIT_MMX
  720. +MC22 mmxext, avg, 4
  721. +INIT_XMM
  722. +MC22 sse2  , avg, 8
  723. +
  724. +;-----------------------------------------------------------------------------
  725. +; void h264_qpel_mc12(uint8_t *dst, uint8_t *src, int stride)
  726. +;-----------------------------------------------------------------------------
  727. +%macro MC12 3
  728. +cglobal %2_h264_qpel%3_mc12_10_%1, 3,7,10
  729. +    mov      r6, rsp          ; backup stack pointer
  730. +    and     rsp, ~(mmsize-1)  ; align stack
  731. +    sub     rsp, 4096         ; TODO: calculate this correctly
  732. +
  733. +    call %2_hv%3_10_%1
  734. +
  735. +    xor        r4, r4
  736. +.body
  737. +    mov       r3d, %3
  738. +    mova       m0, [tap1]
  739. +    mova       m7, [tap3]
  740. +%if num_mmregs > 8
  741. +    mova       m8, [tap2]
  742. +    mova       m9, [depad]
  743. +    %define s1 m8
  744. +    %define s2 m9
  745. +%else
  746. +    %define s1 [tap2]
  747. +    %define s2 [depad]
  748. +%endif
  749. +.h_loop:
  750. +    movu       m1, [rsp+mmsize-4]
  751. +    movu       m2, [rsp+mmsize-2]
  752. +    mova       m3, [rsp+mmsize+0]
  753. +    movu       m4, [rsp+mmsize+2]
  754. +    movu       m5, [rsp+mmsize+4]
  755. +    movu       m6, [rsp+mmsize+6]
  756. +    pmaddwd    m1, m0
  757. +    pmaddwd    m2, m0
  758. +    pmaddwd    m3, s1
  759. +    pmaddwd    m4, s1
  760. +    pmaddwd    m5, m7
  761. +    pmaddwd    m6, m7
  762. +    paddd      m1, s2
  763. +    paddd      m2, s2
  764. +    paddd      m3, m5
  765. +    paddd      m4, m6
  766. +    paddd      m1, m3
  767. +    paddd      m2, m4
  768. +    psrad      m1, 10
  769. +    psrad      m2, 10
  770. +    pslld      m2, 16
  771. +    pand       m1, [pd_0f]
  772. +    por        m1, m2
  773. +    CLIPW      m1, [pw_0], [pw_pixel_max]
  774. +
  775. +    movu       m3, [rsp+r4+mmsize] ; movu needed for mc32
  776. +    paddw      m3, [depad2]
  777. +    psrlw      m3, 5
  778. +    psubw      m3, [unpad]
  779. +    CLIPW      m3, [pw_0], [pw_pixel_max]
  780. +    pavgw      m1, m3
  781. +
  782. +    OP_MOV   [r0], m1
  783. +    add       rsp, mmsize*3
  784. +    add        r0, r2
  785. +    dec       r3d
  786. +    jg .h_loop
  787. +
  788. +    mov     rsp, r6          ; restore stack pointer
  789. +    RET
  790. +%endmacro
  791. +
  792. +%define OP_MOV mova
  793. +INIT_MMX
  794. +MC12 mmxext, put, 4
  795. +INIT_XMM
  796. +MC12 sse2  , put, 8
  797. +
  798. +%define OP_MOV AVG_MOV
  799. +INIT_MMX
  800. +MC12 mmxext, avg, 4
  801. +INIT_XMM
  802. +MC12 sse2  , avg, 8
  803. +
  804. +;-----------------------------------------------------------------------------
  805. +; void h264_qpel_mc32(uint8_t *dst, uint8_t *src, int stride)
  806. +;-----------------------------------------------------------------------------
  807. +%macro MC32 3
  808. +cglobal %2_h264_qpel%3_mc32_10_%1, 3,7,10
  809. +    mov  r6, rsp          ; backup stack pointer
  810. +    and rsp, ~(mmsize-1)  ; align stack
  811. +    sub rsp, 4096         ; TODO: calculate this correctly
  812. +
  813. +    call %2_hv%3_10_%1
  814. +
  815. +    mov  r4, 2            ; sizeof(pixel)
  816. +    jmp mangle(ff_%2_h264_qpel%3_mc12_10_%1.body)
  817. +%endmacro
  818. +
  819. +%define OP_MOV mova
  820. +INIT_MMX
  821. +MC32 mmxext, put, 4
  822. +INIT_XMM
  823. +MC32 sse2  , put, 8
  824. +
  825. +%define OP_MOV AVG_MOV
  826. +INIT_MMX
  827. +MC32 mmxext, avg, 4
  828. +INIT_XMM
  829. +MC32 sse2  , avg, 8
  830. +
  831. +;-----------------------------------------------------------------------------
  832. +; void h264_qpel_mc21(uint8_t *dst, uint8_t *src, int stride)
  833. +;-----------------------------------------------------------------------------
  834. +%macro MC21 3
  835. +%2_h%3_10_%1:
  836. +    add       rsp, gprsize
  837. +    mov       r3d, %3
  838. +    xor        r4, r4
  839. +    mova       m6, [pad20]
  840. +.nextrow
  841. +    movu       m2, [r5-4]
  842. +    movu       m3, [r5-2]
  843. +    movu       m4, [r5+0]
  844. +    ADDW       m4, [r5+2], m5
  845. +    ADDW       m3, [r5+4], m5
  846. +    ADDW       m2, [r5+6], m5
  847. +
  848. +    FILT_H2    m2, m3, m4
  849. +    psubw      m2, m6
  850. +    mova [rsp+r4], m2
  851. +    add        r4, mmsize*3
  852. +    add        r5, r2
  853. +    dec       r3d
  854. +    jg .nextrow
  855. +    sub       rsp, gprsize
  856. +    ret
  857. +
  858. +cglobal %2_h264_qpel%3_mc21_10_%1, 3,7,10
  859. +    mov   r6, rsp          ; backup stack pointer
  860. +    and  rsp, ~(mmsize-1)  ; align stack
  861. +    sub  rsp, 4096         ; TODO: calculate this correctly
  862. +
  863. +    mov   r5, r1
  864. +    call  %2_hv%3_10_%1
  865. +
  866. +%define PAD mmsize*16*3*2     ; SIZE*16*3*sizeof(pixel)
  867. +    add  rsp, PAD
  868. +    call %2_h%3_10_%1
  869. +    sub  rsp, PAD
  870. +
  871. +    mov  r4, PAD-mmsize            ; H buffer
  872. +    jmp mangle(ff_%2_h264_qpel%3_mc12_10_%1.body)
  873. +%endmacro
  874. +
  875. +%define OP_MOV mova
  876. +INIT_MMX
  877. +MC21 mmxext, put, 4
  878. +INIT_XMM
  879. +MC21 sse2  , put, 8
  880. +
  881. +%define OP_MOV AVG_MOV
  882. +INIT_MMX
  883. +MC21 mmxext, avg, 4
  884. +INIT_XMM
  885. +MC21 sse2  , avg, 8
  886. +
  887. +;-----------------------------------------------------------------------------
  888. +; void h264_qpel_mc23(uint8_t *dst, uint8_t *src, int stride)
  889. +;-----------------------------------------------------------------------------
  890. +%macro MC23 3
  891. +cglobal %2_h264_qpel%3_mc23_10_%1, 3,7,10
  892. +    mov   r6, rsp          ; backup stack pointer
  893. +    and  rsp, ~(mmsize-1)  ; align stack
  894. +    sub  rsp, 4096         ; TODO: calculate this correctly
  895. +
  896. +    lea   r5, [r1+r2]
  897. +    call  %2_hv%3_10_%1
  898. +
  899. +%define PAD mmsize*16*3*2     ; SIZE*16*3*sizeof(pixel)
  900. +    add  rsp, PAD
  901. +    call %2_h%3_10_%1
  902. +    sub  rsp, PAD
  903. +
  904. +    mov  r4, PAD-mmsize            ; H buffer
  905. +    jmp mangle(ff_%2_h264_qpel%3_mc12_10_%1.body)
  906. +%endmacro
  907. +
  908. +%define OP_MOV mova
  909. +INIT_MMX
  910. +MC23 mmxext, put, 4
  911. +INIT_XMM
  912. +MC23 sse2  , put, 8
  913. +
  914. +%define OP_MOV AVG_MOV
  915. +INIT_MMX
  916. +MC23 mmxext, avg, 4
  917. +INIT_XMM
  918. +MC23 sse2  , avg, 8
  919. diff --git a/libavcodec/x86/h264_qpel_mmx_10bit.c b/libavcodec/x86/h264_qpel_mmx_10bit.c
  920. new file mode 100755
  921. index 0000000..98cf6da
  922. --- /dev/null
  923. +++ b/libavcodec/x86/h264_qpel_mmx_10bit.c
  924. @@ -0,0 +1,141 @@
  925. +/*
  926. + * Copyright (c) 2011 Daniel Kang
  927. + *
  928. + * This file is part of Libav.
  929. + *
  930. + * Libav is free software; you can redistribute it and/or
  931. + * modify it under the terms of the GNU Lesser General Public
  932. + * License as published by the Free Software Foundation; either
  933. + * version 2.1 of the License, or (at your option) any later version.
  934. + *
  935. + * Libav is distributed in the hope that it will be useful,
  936. + * but WITHOUT ANY WARRANTY; without even the implied warranty of
  937. + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  938. + * Lesser General Public License for more details.
  939. + *
  940. + * You should have received a copy of the GNU Lesser General Public
  941. + * License along with Libav; if not, write to the Free Software
  942. + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  943. + */
  944. +
  945. +#include "dsputil_mmx.h"
  946. +
  947. +#define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
  948. +void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
  949. +    (uint8_t *dst, uint8_t *src, int stride);
  950. +
  951. +#define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
  952. +    LUMA_MC_OP(put,  4, DEPTH, TYPE, OPT) \
  953. +    LUMA_MC_OP(avg,  4, DEPTH, TYPE, OPT) \
  954. +    LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
  955. +    LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
  956. +    LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
  957. +    LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
  958. +
  959. +#define LUMA_MC_816(DEPTH, TYPE, OPT) \
  960. +    LUMA_MC_OP(put,  8, DEPTH, TYPE, OPT) \
  961. +    LUMA_MC_OP(avg,  8, DEPTH, TYPE, OPT) \
  962. +    LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
  963. +    LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
  964. +
  965. +LUMA_MC_ALL(10, mc00, mmxext)
  966. +LUMA_MC_ALL(10, mc10, mmxext)
  967. +LUMA_MC_ALL(10, mc20, mmxext)
  968. +LUMA_MC_ALL(10, mc30, mmxext)
  969. +LUMA_MC_ALL(10, mc01, mmxext)
  970. +LUMA_MC_ALL(10, mc11, mmxext)
  971. +LUMA_MC_ALL(10, mc21, mmxext)
  972. +LUMA_MC_ALL(10, mc31, mmxext)
  973. +LUMA_MC_ALL(10, mc02, mmxext)
  974. +LUMA_MC_ALL(10, mc12, mmxext)
  975. +LUMA_MC_ALL(10, mc22, mmxext)
  976. +LUMA_MC_ALL(10, mc32, mmxext)
  977. +LUMA_MC_ALL(10, mc03, mmxext)
  978. +LUMA_MC_ALL(10, mc13, mmxext)
  979. +LUMA_MC_ALL(10, mc23, mmxext)
  980. +LUMA_MC_ALL(10, mc33, mmxext)
  981. +
  982. +LUMA_MC_816(10, mc00, sse2)
  983. +LUMA_MC_816(10, mc10, sse2)
  984. +LUMA_MC_816(10, mc20, sse2)
  985. +LUMA_MC_816(10, mc30, sse2)
  986. +LUMA_MC_816(10, mc01, sse2)
  987. +LUMA_MC_816(10, mc11, sse2)
  988. +LUMA_MC_816(10, mc21, sse2)
  989. +LUMA_MC_816(10, mc31, sse2)
  990. +LUMA_MC_816(10, mc02, sse2)
  991. +LUMA_MC_816(10, mc12, sse2)
  992. +LUMA_MC_816(10, mc22, sse2)
  993. +LUMA_MC_816(10, mc32, sse2)
  994. +LUMA_MC_816(10, mc03, sse2)
  995. +LUMA_MC_816(10, mc13, sse2)
  996. +LUMA_MC_816(10, mc23, sse2)
  997. +LUMA_MC_816(10, mc33, sse2)
  998. +
  999. +#define QPEL8_OPMC(OP, MC, MMX)\
  1000. +void ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
  1001. +    ff_ ## OP ## _h264_qpel4_ ## MC ## _10_ ## MMX(dst  , src  , stride);\
  1002. +    ff_ ## OP ## _h264_qpel4_ ## MC ## _10_ ## MMX(dst+8, src+8, stride);\
  1003. +    src += 4*stride;\
  1004. +    dst += 4*stride;\
  1005. +    ff_ ## OP ## _h264_qpel4_ ## MC ## _10_ ## MMX(dst  , src  , stride);\
  1006. +    ff_ ## OP ## _h264_qpel4_ ## MC ## _10_ ## MMX(dst+8, src+8, stride);\
  1007. +}
  1008. +
  1009. +#define QPEL8_OP(MC, MMX)\
  1010. +QPEL8_OPMC(put, MC, MMX)\
  1011. +QPEL8_OPMC(avg, MC, MMX)
  1012. +
  1013. +#define QPEL8(MMX)\
  1014. +QPEL8_OP(mc00, MMX)\
  1015. +QPEL8_OP(mc01, MMX)\
  1016. +QPEL8_OP(mc02, MMX)\
  1017. +QPEL8_OP(mc03, MMX)\
  1018. +QPEL8_OP(mc10, MMX)\
  1019. +QPEL8_OP(mc11, MMX)\
  1020. +QPEL8_OP(mc12, MMX)\
  1021. +QPEL8_OP(mc13, MMX)\
  1022. +QPEL8_OP(mc20, MMX)\
  1023. +QPEL8_OP(mc21, MMX)\
  1024. +QPEL8_OP(mc22, MMX)\
  1025. +QPEL8_OP(mc23, MMX)\
  1026. +QPEL8_OP(mc30, MMX)\
  1027. +QPEL8_OP(mc31, MMX)\
  1028. +QPEL8_OP(mc32, MMX)\
  1029. +QPEL8_OP(mc33, MMX)
  1030. +
  1031. +#define QPEL16_OPMC(OP, MC, MMX)\
  1032. +void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
  1033. +    ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
  1034. +    ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
  1035. +    src += 8*stride;\
  1036. +    dst += 8*stride;\
  1037. +    ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst   , src   , stride);\
  1038. +    ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
  1039. +}
  1040. +
  1041. +#define QPEL16_OP(MC, MMX)\
  1042. +QPEL16_OPMC(put, MC, MMX)\
  1043. +QPEL16_OPMC(avg, MC, MMX)
  1044. +
  1045. +#define QPEL16(MMX)\
  1046. +QPEL16_OP(mc01, MMX)\
  1047. +QPEL16_OP(mc02, MMX)\
  1048. +QPEL16_OP(mc03, MMX)\
  1049. +QPEL16_OP(mc10, MMX)\
  1050. +QPEL16_OP(mc11, MMX)\
  1051. +QPEL16_OP(mc12, MMX)\
  1052. +QPEL16_OP(mc13, MMX)\
  1053. +QPEL16_OP(mc20, MMX)\
  1054. +QPEL16_OP(mc21, MMX)\
  1055. +QPEL16_OP(mc22, MMX)\
  1056. +QPEL16_OP(mc23, MMX)\
  1057. +QPEL16_OP(mc30, MMX)\
  1058. +QPEL16_OP(mc31, MMX)\
  1059. +QPEL16_OP(mc32, MMX)\
  1060. +QPEL16_OP(mc33, MMX)
  1061. +
  1062. +QPEL8(mmxext)
  1063. +QPEL16_OP(mc00, mmxext)
  1064. +QPEL16(mmxext)
  1065. +QPEL16(sse2)
  1066. --
  1067. 1.7.5.1
  1068.  
  1069.  
  1070. From 970a7a164d0ae110821c174a7b9081f93172c095 Mon Sep 17 00:00:00 2001
  1071. From: Daniel Kang <daniel.d.kang@gmail.com>
  1072. Date: Sat, 25 Jun 2011 13:28:49 -0400
  1073. Subject: [PATCH 2/2] improvement?
  1074.  
  1075. ---
  1076. libavcodec/x86/h264_qpel_10bit.asm |   40 ++++++++++++++++++++---------------
  1077.  1 files changed, 23 insertions(+), 17 deletions(-)
  1078.  
  1079. diff --git a/libavcodec/x86/h264_qpel_10bit.asm b/libavcodec/x86/h264_qpel_10bit.asm
  1080. index fc44e85..bc0e78d 100644
  1081. --- a/libavcodec/x86/h264_qpel_10bit.asm
  1082. +++ b/libavcodec/x86/h264_qpel_10bit.asm
  1083. @@ -620,7 +620,7 @@ MC22 sse2  , avg, 8
  1084.  ; void h264_qpel_mc12(uint8_t *dst, uint8_t *src, int stride)
  1085.  ;-----------------------------------------------------------------------------
  1086.  %macro MC12 3
  1087. -cglobal %2_h264_qpel%3_mc12_10_%1, 3,7,10
  1088. +cglobal %2_h264_qpel%3_mc12_10_%1, 3,7,12
  1089.      mov      r6, rsp          ; backup stack pointer
  1090.      and     rsp, ~(mmsize-1)  ; align stack
  1091.      sub     rsp, 4096         ; TODO: calculate this correctly
  1092. @@ -630,16 +630,22 @@ cglobal %2_h264_qpel%3_mc12_10_%1, 3,7,10
  1093.      xor        r4, r4
  1094.  .body
  1095.      mov       r3d, %3
  1096. -    mova       m0, [tap1]
  1097. -    mova       m7, [tap3]
  1098. +    pxor       m0, m0
  1099. +    mova       m7, [pw_pixel_max]
  1100.  %if num_mmregs > 8
  1101. -    mova       m8, [tap2]
  1102. -    mova       m9, [depad]
  1103. +    mova       m8, [tap1]
  1104. +    mova       m9, [tap2]
  1105. +    mova      m10, [tap3]
  1106. +    mova      m11, [depad]
  1107.      %define s1 m8
  1108.      %define s2 m9
  1109. +    %define s3 m10
  1110. +    %define s1 m11
  1111.  %else
  1112. -    %define s1 [tap2]
  1113. -    %define s2 [depad]
  1114. +    %define s1 [tap1]
  1115. +    %define s2 [tap2]
  1116. +    %define s3 [tap3]
  1117. +    %define d1 [depad]
  1118.  %endif
  1119.  .h_loop:
  1120.      movu       m1, [rsp+mmsize-4]
  1121. @@ -648,14 +654,14 @@ cglobal %2_h264_qpel%3_mc12_10_%1, 3,7,10
  1122.      movu       m4, [rsp+mmsize+2]
  1123.      movu       m5, [rsp+mmsize+4]
  1124.      movu       m6, [rsp+mmsize+6]
  1125. -    pmaddwd    m1, m0
  1126. -    pmaddwd    m2, m0
  1127. -    pmaddwd    m3, s1
  1128. -    pmaddwd    m4, s1
  1129. -    pmaddwd    m5, m7
  1130. -    pmaddwd    m6, m7
  1131. -    paddd      m1, s2
  1132. -    paddd      m2, s2
  1133. +    pmaddwd    m1, s1
  1134. +    pmaddwd    m2, s1
  1135. +    pmaddwd    m3, s2
  1136. +    pmaddwd    m4, s2
  1137. +    pmaddwd    m5, s3
  1138. +    pmaddwd    m6, s3
  1139. +    paddd      m1, d1
  1140. +    paddd      m2, d1
  1141.      paddd      m3, m5
  1142.      paddd      m4, m6
  1143.      paddd      m1, m3
  1144. @@ -665,13 +671,13 @@ cglobal %2_h264_qpel%3_mc12_10_%1, 3,7,10
  1145.      pslld      m2, 16
  1146.      pand       m1, [pd_0f]
  1147.      por        m1, m2
  1148. -    CLIPW      m1, [pw_0], [pw_pixel_max]
  1149. +    CLIPW      m1, m0, m7
  1150.  
  1151.      movu       m3, [rsp+r4+mmsize] ; movu needed for mc32
  1152.      paddw      m3, [depad2]
  1153.      psrlw      m3, 5
  1154.      psubw      m3, [unpad]
  1155. -    CLIPW      m3, [pw_0], [pw_pixel_max]
  1156. +    CLIPW      m3, m0, m7
  1157.      pavgw      m1, m3
  1158.  
  1159.      OP_MOV   [r0], m1
  1160. --
  1161. 1.7.5.1
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!