API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Sep 17th, 2017
453
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Diff 26.70 KB | None | 0 0
raw download clone embed print report
  1. From dc3ee1e5dcc8cf010223976be504f88bbe4d9889 Mon Sep 17 00:00:00 2001
  2. From: Daniel Kang <daniel.d.kang@gmail.com>
  3. Date: Sun, 5 Jun 2011 18:33:23 -0400
  4. Subject: [PATCH 1/2] H.264: Add x86 assembly for 10-bit MC Chroma H.264
  5.  functions.
  6.  
  7. Mainly ported from 8-bit H.264 MC Chroma.
  8. ---
  9. libavcodec/x86/Makefile                |    1 +
  10.  libavcodec/x86/dsputil_mmx.c           |   32 ++++
  11.  libavcodec/x86/h264_chromamc_10bit.asm |  279 ++++++++++++++++++++++++++++++++
  12.  3 files changed, 312 insertions(+), 0 deletions(-)
  13.  create mode 100644 libavcodec/x86/h264_chromamc_10bit.asm
  14.  
  15. diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
  16. index 1c451c8..ea57bd1 100644
  17. --- a/libavcodec/x86/Makefile
  18. +++ b/libavcodec/x86/Makefile
  19. @@ -44,6 +44,7 @@ MMX-OBJS-$(HAVE_YASM)                  += x86/dsputil_yasm.o            \
  20.                                            x86/deinterlace.o             \
  21.                                            x86/fmtconvert.o              \
  22.                                            x86/h264_chromamc.o           \
  23. +                                          x86/h264_chromamc_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 1cc6991..7ac9679 100644
  29. --- a/libavcodec/x86/dsputil_mmx.c
  30. +++ b/libavcodec/x86/dsputil_mmx.c
  31. @@ -1938,6 +1938,19 @@ void ff_avg_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,
  32.  void ff_avg_h264_chroma_mc4_ssse3     (uint8_t *dst, uint8_t *src,
  33.                                         int stride, int h, int x, int y);
  34.  
  35. +#define CHROMA_MC(OP, NUM, DEPTH, OPT) \
  36. +void ff_ ## OP ## _h264_chroma_mc ## NUM ## _ ## DEPTH ## _ ## OPT \
  37. +                                      (uint8_t *dst, uint8_t *src,\
  38. +                                       int stride, int h, int x, int y);
  39. +
  40. +CHROMA_MC(put, 2, 10, mmxext)
  41. +CHROMA_MC(avg, 2, 10, mmxext)
  42. +CHROMA_MC(put, 4, 10, mmxext)
  43. +CHROMA_MC(avg, 4, 10, mmxext)
  44. +CHROMA_MC(put, 8, 10, sse2)
  45. +CHROMA_MC(avg, 8, 10, sse2)
  46. +CHROMA_MC(put, 8, 10, avx)
  47. +CHROMA_MC(avg, 8, 10, avx)
  48.  
  49.  /* CAVS specific */
  50.  void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
  51. @@ -2420,6 +2433,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
  52.  {
  53.      int mm_flags = av_get_cpu_flags();
  54.      const int high_bit_depth = avctx->codec_id == CODEC_ID_H264 && avctx->bits_per_raw_sample > 8;
  55. +    const int bit_depth = avctx->bits_per_raw_sample;
  56.  
  57.      if (avctx->dsp_mask) {
  58.          if (avctx->dsp_mask & AV_CPU_FLAG_FORCE)
  59. @@ -2651,6 +2665,12 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
  60.              c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_mmx2;
  61.              c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_mmx2;
  62.              }
  63. +            if (bit_depth == 10) {
  64. +                c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_10_mmxext;
  65. +                c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_10_mmxext;
  66. +                c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_10_mmxext;
  67. +                c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_10_mmxext;
  68. +            }
  69.  
  70.              c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
  71.  #endif
  72. @@ -2756,6 +2776,10 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
  73.              H264_QPEL_FUNCS(3, 2, sse2);
  74.              H264_QPEL_FUNCS(3, 3, sse2);
  75.              }
  76. +            if (bit_depth == 10) {
  77. +                c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_sse2;
  78. +                c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_sse2;
  79. +            }
  80.          }
  81.  #if HAVE_SSSE3
  82.          if(mm_flags & AV_CPU_FLAG_SSSE3){
  83. @@ -2854,6 +2878,14 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
  84.              }
  85.  #endif
  86.          }
  87. +#if HAVE_AVX
  88. +        if (mm_flags & AV_CPU_FLAG_AVX) {
  89. +            if (bit_depth == 10) {
  90. +                c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_avx;
  91. +                c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_avx;
  92. +            }
  93. +        }
  94. +#endif
  95.      }
  96.  
  97.      if (CONFIG_ENCODERS)
  98. diff --git a/libavcodec/x86/h264_chromamc_10bit.asm b/libavcodec/x86/h264_chromamc_10bit.asm
  99. new file mode 100644
  100. index 0000000..757d99f
  101. --- /dev/null
  102. +++ b/libavcodec/x86/h264_chromamc_10bit.asm
  103. @@ -0,0 +1,279 @@
  104. +;*****************************************************************************
  105. +;* MMX/SSE2/AVX-optimized 10-bit H.264 chroma MC code
  106. +;*****************************************************************************
  107. +;* Copyright (C) 2005-2011 x264 project
  108. +;*
  109. +;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
  110. +;*
  111. +;* This file is part of Libav.
  112. +;*
  113. +;* Libav is free software; you can redistribute it and/or
  114. +;* modify it under the terms of the GNU Lesser General Public
  115. +;* License as published by the Free Software Foundation; either
  116. +;* version 2.1 of the License, or (at your option) any later version.
  117. +;*
  118. +;* Libav is distributed in the hope that it will be useful,
  119. +;* but WITHOUT ANY WARRANTY; without even the implied warranty of
  120. +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  121. +;* Lesser General Public License for more details.
  122. +;*
  123. +;* You should have received a copy of the GNU Lesser General Public
  124. +;* License along with Libav; if not, write to the Free Software
  125. +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  126. +;******************************************************************************
  127. +
  128. +%include "x86inc.asm"
  129. +%include "x86util.asm"
  130. +
  131. +SECTION_RODATA
  132. +
  133. +cextern pw_4
  134. +cextern pw_8
  135. +cextern pw_32
  136. +cextern pw_64
  137. +
  138. +SECTION .text
  139. +
  140. +
  141. +%macro mv0_pixels_mc8 0
  142. +    lea           r4, [r2*2 ]
  143. +.next4rows
  144. +    movu          m0, [r1   ]
  145. +    movu          m1, [r1+r2]
  146. +    CHROMAMC_AVG  m0, [r0   ]
  147. +    CHROMAMC_AVG  m1, [r0+r2]
  148. +    mova     [r0   ], m0
  149. +    mova     [r0+r2], m1
  150. +    add           r0, r4
  151. +    add           r1, r4
  152. +    movu          m0, [r1   ]
  153. +    movu          m1, [r1+r2]
  154. +    CHROMAMC_AVG  m0, [r0   ]
  155. +    CHROMAMC_AVG  m1, [r0+r2]
  156. +    add           r1, r4
  157. +    mova     [r0   ], m0
  158. +    mova     [r0+r2], m1
  159. +    add           r0, r4
  160. +    sub          r3d, 4
  161. +    jne .next4rows
  162. +%endmacro
  163. +
  164. +;-----------------------------------------------------------------------------
  165. +; void put/avg_h264_chroma_mc8(pixel *dst, pixel *src, int stride, int h, int mx, int my)
  166. +;-----------------------------------------------------------------------------
  167. +%macro CHROMA_MC8 2
  168. +; put/avg_h264_chroma_mc8_*(uint8_t *dst /*align 8*/, uint8_t *src /*align 1*/,
  169. +;                              int stride, int h, int mx, int my)
  170. +cglobal %1_h264_chroma_mc8_10_%2, 6,7,8
  171. +    movsxdifnidn  r2, r2d
  172. +    mov          r6d, r5d
  173. +    or           r6d, r4d
  174. +    jne .at_least_one_non_zero
  175. +    ; mx == 0 AND my == 0 - no filter needed
  176. +    mv0_pixels_mc8
  177. +    REP_RET
  178. +
  179. +.at_least_one_non_zero
  180. +    mov          r6d, 2
  181. +    test         r5d, r5d
  182. +    je .my_is_zero
  183. +    mov           r6, r2        ; dxy = x ? 1 : stride
  184. +    test         r4d, r4d
  185. +    jne .both_non_zero
  186. +.my_is_zero
  187. +    ; mx == 0 XOR my == 0 - 1 dimensional filter only
  188. +    or           r4d, r5d       ; x + y
  189. +    movd          m5, r4d
  190. +    mova          m4, [pw_8]
  191. +    mova          m6, [pw_4]    ; mm6 = rnd >> 3
  192. +    SPLATW        m5, m5        ; mm5 = B = x
  193. +    psubw         m4, m5        ; mm4 = A = 8-x
  194. +
  195. +.next1drow
  196. +    movu          m0, [r1   ]   ; mm0 = src[0..7]
  197. +    movu          m2, [r1+r6]   ; mm2 = src[1..8]
  198. +
  199. +    pmullw        m0, m4        ; mm0 = A * src[0..7]
  200. +    pmullw        m2, m5        ; mm2 = B * src[1..8]
  201. +
  202. +    paddw         m0, m6
  203. +    paddw         m0, m2
  204. +    psrlw         m0, 3
  205. +    CHROMAMC_AVG  m0, [r0]
  206. +    mova        [r0], m0        ; dst[0..7] = (A * src[0..7] + B * src[1..8] + (rnd >> 3)) >> 3
  207. +
  208. +    add           r0, r2
  209. +    add           r1, r2
  210. +    dec           r3d
  211. +    jne .next1drow
  212. +    REP_RET
  213. +
  214. +.both_non_zero ; general case, bilinear
  215. +    movd          m4, r4m         ; x
  216. +    movd          m6, r5m         ; y
  217. +
  218. +    SPLATW        m4, m4          ; mm4 = x words
  219. +    SPLATW        m6, m6          ; mm6 = y words
  220. +    psllw         m5, m4, 3       ; mm5 = 8x
  221. +    pmullw        m4, m6          ; mm4 = x * y
  222. +    psllw         m6, 3           ; mm6 = 8y
  223. +    paddw         m1, m5, m6      ; mm7 = 8x+8y
  224. +    mova          m7, m4          ; DD = x * y
  225. +    psubw         m5, m4          ; mm5 = B = 8x - xy
  226. +    psubw         m6, m4          ; mm6 = C = 8y - xy
  227. +    paddw         m4, [pw_64]
  228. +    psubw         m4, m1          ; mm4 = A = xy - (8x+8y) + 64
  229. +
  230. +    movu          m0, [r1  ]      ; mm0 = src[0..7]
  231. +    movu          m1, [r1+2]      ; mm1 = src[1..8]
  232. +.next2drow
  233. +    add           r1, r2
  234. +
  235. +    pmullw        m2, m0, m4
  236. +    pmullw        m1, m5
  237. +    paddw         m2, m1          ; mm2 = A * src[0..7] + B * src[1..8]
  238. +
  239. +    movu          m0, [r1]
  240. +    movu          m1, [r1+2]
  241. +    pmullw        m3, m0, m6
  242. +    paddw         m2, m3          ; mm2 += C * src[0..7+strde]
  243. +    pmullw        m3, m1, m7
  244. +    paddw         m2, m3          ; mm2 += D * src[1..8+strde]
  245. +
  246. +    paddw         m2, [pw_32]
  247. +    psrlw         m2, 6
  248. +    CHROMAMC_AVG  m2, [r0]
  249. +    mova        [r0], m2          ; dst[0..7] = (mm2 + 32) >> 6
  250. +
  251. +    add           r0, r2
  252. +    dec          r3d
  253. +    jne .next2drow
  254. +    REP_RET
  255. +%endmacro
  256. +
  257. +;-----------------------------------------------------------------------------
  258. +; void put/avg_h264_chroma_mc4(pixel *dst, pixel *src, int stride, int h, int mx, int my)
  259. +;-----------------------------------------------------------------------------
  260. +;TODO: xmm mc4
  261. +%macro MC4_OP 2
  262. +    movq          %1, [r1  ]
  263. +    movq          m1, [r1+2]
  264. +    add           r1, r2
  265. +    pmullw        %1, m4
  266. +    pmullw        m1, m2
  267. +    paddw         m1, %1
  268. +    mova          %1, m1
  269. +
  270. +    pmullw        %2, m5
  271. +    pmullw        m1, m3
  272. +    paddw         %2, [pw_32]
  273. +    paddw         m1, %2
  274. +    psrlw         m1, 6
  275. +    CHROMAMC_AVG4 m1, %2, [r0]
  276. +    movq        [r0], m1
  277. +    add           r0, r2
  278. +%endmacro
  279. +
  280. +%macro CHROMA_MC4 2
  281. +cglobal %1_h264_chroma_mc4_10_%2, 6,6,7
  282. +    movsxdifnidn  r2, r2d
  283. +    movd          m2, r4m         ; x
  284. +    movd          m3, r5m         ; y
  285. +    mova          m4, [pw_8]
  286. +    mova          m5, m4
  287. +    SPLATW        m2, m2
  288. +    SPLATW        m3, m3
  289. +    psubw         m4, m2
  290. +    psubw         m5, m3
  291. +
  292. +    movq          m0, [r1  ]
  293. +    movq          m6, [r1+2]
  294. +    add           r1, r2
  295. +    pmullw        m0, m4
  296. +    pmullw        m6, m2
  297. +    paddw         m6, m0
  298. +
  299. +.next2rows
  300. +    MC4_OP m0, m6
  301. +    MC4_OP m6, m0
  302. +    sub   r3d, 2
  303. +    jnz .next2rows
  304. +    REP_RET
  305. +%endmacro
  306. +
  307. +;-----------------------------------------------------------------------------
  308. +; void put/avg_h264_chroma_mc2(pixel *dst, pixel *src, int stride, int h, int mx, int my)
  309. +;-----------------------------------------------------------------------------
  310. +%macro CHROMA_MC2 2
  311. +cglobal %1_h264_chroma_mc2_10_%2, 6,7
  312. +    movsxdifnidn  r2, r2d
  313. +    mov          r6d, r4d
  314. +    shl          r4d, 16
  315. +    sub          r4d, r6d
  316. +    add          r4d, 8
  317. +    imul         r5d, r4d         ; x*y<<16 | y*(8-x)
  318. +    shl          r4d, 3
  319. +    sub          r4d, r5d         ; x*(8-y)<<16 | (8-x)*(8-y)
  320. +
  321. +    movd          m5, r4d
  322. +    movd          m6, r5d
  323. +    punpckldq     m5, m5          ; mm5 = {A,B,A,B}
  324. +    punpckldq     m6, m6          ; mm6 = {C,D,C,D}
  325. +    movq          m2, [r1]
  326. +    pxor          m7, m7
  327. +    pshufw        m2, m2, 0x94    ; mm0 = src[0,1,1,2]
  328. +
  329. +.nextrow
  330. +    add           r1, r2
  331. +    movq          m1, m2
  332. +    pmaddwd       m1, m5          ; mm1 = A * src[0,1] + B * src[1,2]
  333. +    movq          m0, [r1]
  334. +    pshufw        m0, m0, 0x94    ; mm0 = src[0,1,1,2]
  335. +    movq          m2, m0
  336. +    pmaddwd       m0, m6
  337. +    paddw         m1, [pw_32]
  338. +    paddw         m1, m0          ; mm1 += C * src[0,1] + D * src[1,2]
  339. +    psrlw         m1, 6
  340. +    packssdw      m1, m7
  341. +    CHROMAMC_AVG4 m1, m3, [r0]
  342. +    movd        [r0], m1
  343. +    add           r0, r2
  344. +    sub          r3d, 1
  345. +    jnz .nextrow
  346. +    REP_RET
  347. +%endmacro
  348. +
  349. +%macro NOTHING 2-3
  350. +%endmacro
  351. +%macro DIRECT_AVG 2
  352. +    PAVG          %1, %2
  353. +%endmacro
  354. +%macro COPY_AVG 3
  355. +    movq          %2, %3
  356. +    PAVG          %1, %2
  357. +%endmacro
  358. +
  359. +%define CHROMAMC_AVG  NOTHING
  360. +%define CHROMAMC_AVG4 NOTHING
  361. +INIT_XMM
  362. +CHROMA_MC8 put, sse2
  363. +%ifdef HAVE_AVX
  364. +CHROMA_MC8 put, avx
  365. +%endif
  366. +INIT_XMM
  367. +CHROMA_MC4 put, mmxext
  368. +INIT_MMX
  369. +CHROMA_MC2 put, mmxext
  370. +
  371. +%define CHROMAMC_AVG  DIRECT_AVG
  372. +%define CHROMAMC_AVG4 COPY_AVG
  373. +%define PAVG          pavgw
  374. +INIT_XMM
  375. +CHROMA_MC8 avg, sse2
  376. +%ifdef HAVE_AVX
  377. +CHROMA_MC8 avg, avx
  378. +%endif
  379. +INIT_XMM
  380. +CHROMA_MC4 avg, mmxext
  381. +INIT_MMX
  382. +CHROMA_MC2 avg, mmxext
  383. --
  384. 1.7.5.1
  385.  
  386.  
  387. From 068d2421c97a6f34d47275fe193a55629051fee0 Mon Sep 17 00:00:00 2001
  388. From: Daniel Kang <daniel.d.kang@gmail.com>
  389. Date: Sun, 12 Jun 2011 18:06:56 -0400
  390. Subject: [PATCH 2/2] H.264: Add x86 assembly for 10-bit weight/biweight H.264
  391.  functions.
  392.  
  393. Mainly ported from 8-bit H.264 weight/biweight.
  394. ---
  395. libavcodec/x86/Makefile              |    1 +
  396.  libavcodec/x86/h264_weight_10bit.asm |  298 ++++++++++++++++++++++++++++++++++
  397.  libavcodec/x86/h264dsp_mmx.c         |   72 ++++++++
  398.  3 files changed, 371 insertions(+), 0 deletions(-)
  399.  create mode 100644 libavcodec/x86/h264_weight_10bit.asm
  400.  
  401. diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
  402. index ea57bd1..022ab27 100644
  403. --- a/libavcodec/x86/Makefile
  404. +++ b/libavcodec/x86/Makefile
  405. @@ -15,6 +15,7 @@ YASM-OBJS-$(CONFIG_H264DSP)            += x86/h264_deblock.o            \
  406.                                            x86/h264_idct.o               \
  407.                                            x86/h264_idct_10bit.o         \
  408.                                            x86/h264_weight.o             \
  409. +                                          x86/h264_weight_10bit.o       \
  410.  
  411.  YASM-OBJS-$(CONFIG_H264PRED)           += x86/h264_intrapred.o          \
  412.                                            x86/h264_intrapred_10bit.o
  413. diff --git a/libavcodec/x86/h264_weight_10bit.asm b/libavcodec/x86/h264_weight_10bit.asm
  414. new file mode 100644
  415. index 0000000..ea6ed83
  416. --- /dev/null
  417. +++ b/libavcodec/x86/h264_weight_10bit.asm
  418. @@ -0,0 +1,298 @@
  419. +;*****************************************************************************
  420. +;* MMX/SSE2/AVX-optimized 10-bit H.264 weighted prediction code
  421. +;*****************************************************************************
  422. +;* Copyright (C) 2005-2011 x264 project
  423. +;*
  424. +;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
  425. +;*
  426. +;* This file is part of Libav.
  427. +;*
  428. +;* Libav is free software; you can redistribute it and/or
  429. +;* modify it under the terms of the GNU Lesser General Public
  430. +;* License as published by the Free Software Foundation; either
  431. +;* version 2.1 of the License, or (at your option) any later version.
  432. +;*
  433. +;* Libav is distributed in the hope that it will be useful,
  434. +;* but WITHOUT ANY WARRANTY; without even the implied warranty of
  435. +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  436. +;* Lesser General Public License for more details.
  437. +;*
  438. +;* You should have received a copy of the GNU Lesser General Public
  439. +;* License along with Libav; if not, write to the Free Software
  440. +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  441. +;******************************************************************************
  442. +
  443. +%include "x86inc.asm"
  444. +%include "x86util.asm"
  445. +
  446. +SECTION_RODATA 32
  447. +
  448. +pw_pixel_max: times 8 dw ((1 << 10)-1)
  449. +sq_1: dq 1
  450. +      dq 0
  451. +
  452. +cextern pw_1
  453. +
  454. +SECTION .text
  455. +
  456. +;-----------------------------------------------------------------------------
  457. +; void h264_weight(uint8_t *dst, int stride, int log2_denom,
  458. +;                  int weight, int offset);
  459. +;-----------------------------------------------------------------------------
  460. +
  461. +%macro WEIGHT_SETUP 1
  462. +    mova       m0, [pw_1]
  463. +    movd       m2, r2d
  464. +    pslld      m0, m2       ; 1<<log2_denom
  465. +    SPLATW     m0, m0
  466. +    shl        r4, 19       ; *8, move to upper half of dword
  467. +    lea        r4, [r4+r3*2+0x10000]
  468. +    movd       m3, r4d      ; weight<<1 | 1+(offset<<(3))
  469. +    pshufd     m3, m3, 0
  470. +    mova       m4, [pw_pixel_max]
  471. +    paddw      m2, [sq_1]   ; log2_denom+1
  472. +%ifnidn %1, sse4
  473. +    pxor       m7, m7
  474. +%endif
  475. +%endmacro
  476. +
  477. +%macro WEIGHT_OP 2-3
  478. +%if %0==2
  479. +    mova        m5, [r0+%2]
  480. +    punpckhwd   m6, m5, m0
  481. +    punpcklwd   m5, m0
  482. +%else
  483. +    movq        m5, [r0+%2]
  484. +    movq        m6, [r0+%3]
  485. +    punpcklwd   m5, m0
  486. +    punpcklwd   m6, m0
  487. +%endif
  488. +    pmaddwd     m5, m3
  489. +    pmaddwd     m6, m3
  490. +    psrad       m5, m2
  491. +    psrad       m6, m2
  492. +%ifidn %1, sse4
  493. +    packusdw    m5, m6
  494. +    pminsw      m5, m4
  495. +%else
  496. +    packssdw    m5, m6
  497. +    CLIPW       m5, m7, m4
  498. +%endif
  499. +%endmacro
  500. +
  501. +%macro WEIGHT_FUNC_DBL 1
  502. +cglobal h264_weight_16x16_10_%1, 5,6,8
  503. +    mov        r5, 16
  504. +.body
  505. +    WEIGHT_SETUP %1
  506. +.nextrow
  507. +    WEIGHT_OP %1,  0
  508. +    mova     [r0   ], m5
  509. +    WEIGHT_OP %1, 16
  510. +    mova     [r0+16], m5
  511. +    add        r0, r1
  512. +    dec        r6
  513. +    jnz .nextrow
  514. +    REP_RET
  515. +
  516. +cglobal h264_weight_16x8_10_%1, 5,6,8
  517. +    mov r6, 8
  518. +    jmp mangle(ff_h264_weight_16x16_10_%1.body)
  519. +%endmacro
  520. +
  521. +INIT_XMM
  522. +WEIGHT_FUNC_DBL sse2
  523. +WEIGHT_FUNC_DBL sse4
  524. +
  525. +
  526. +%macro WEIGHT_FUNC_MM 1
  527. +cglobal h264_weight_8x16_10_%1, 5,6,8
  528. +    mov        r6, 16
  529. +.body
  530. +    WEIGHT_SETUP %1
  531. +.nextrow
  532. +    WEIGHT_OP  %1, 0
  533. +    mova     [r0], m5
  534. +    add        r0, r1
  535. +    dec        r2
  536. +    jnz .nextrow
  537. +    REP_RET
  538. +
  539. +cglobal h264_weight_8x8_10_%1, 5,6,8
  540. +    mov r6, 8
  541. +    jmp mangle(ff_h264_weight_8x16_10_%1.body)
  542. +
  543. +cglobal h264_weight_8x4_10_%1, 5,6,8
  544. +    mov r6, 4
  545. +    jmp mangle(ff_h264_weight_8x16_10_%1.body)
  546. +%endmacro
  547. +
  548. +INIT_XMM
  549. +WEIGHT_FUNC_MM sse2
  550. +WEIGHT_FUNC_MM sse4
  551. +
  552. +
  553. +%macro WEIGHT_FUNC_HALF_MM 1
  554. +cglobal h264_weight_4x8_10_%1, 5,6,8
  555. +    mov        r6, 4
  556. +.body
  557. +    WEIGHT_SETUP %1
  558. +    lea        r3, [r1*2]
  559. +.nextrow
  560. +    WEIGHT_OP   %1, 0, r1
  561. +    movh      [r0], m5
  562. +    movhps [r0+r1], m5
  563. +    add         r0, r3
  564. +    dec         r6
  565. +    jnz .nextrow
  566. +    REP_RET
  567. +
  568. +cglobal h264_weight_4x4_10_%1, 5,6,8
  569. +    mov r6, 2
  570. +    jmp mangle(ff_h264_weight_4x8_10_%1.body)
  571. +
  572. +cglobal h264_weight_4x2_10_%1, 5,6,8
  573. +    mov r6, 1
  574. +    jmp mangle(ff_h264_weight_4x8_10_%1.body)
  575. +%endmacro
  576. +
  577. +INIT_XMM
  578. +WEIGHT_FUNC_HALF_MM sse2
  579. +WEIGHT_FUNC_HALF_MM sse4
  580. +
  581. +
  582. +;-----------------------------------------------------------------------------
  583. +; void h264_biweight(uint8_t *dst, uint8_t *src, int stride, int log2_denom,
  584. +;                    int weightd, int weights, int offset);
  585. +;-----------------------------------------------------------------------------
  586. +%macro BIWEIGHT_SETUP 1
  587. +    lea        r6, [r6*4+1] ; (offset<<2)+1
  588. +    or         r6, 1
  589. +    shl        r5, 16
  590. +    or         r4, r5
  591. +    movd       m4, r4d      ; weightd | weights
  592. +    movd       m5, r6d      ; (offset+1)|1
  593. +    movd       m6, r3d      ; log2_denom
  594. +    pslld      m5, m6       ; (((offset<<2)+1)|1)<<log2_denom
  595. +    paddd      m6, [sq_1]
  596. +    pshufd     m4, m4, 0
  597. +    pshufd     m5, m5, 0
  598. +    mova       m3, [pw_pixel_max]
  599. +%ifnidn %1, sse4
  600. +    pxor       m7, m7
  601. +%endif
  602. +%endmacro
  603. +
  604. +%macro BIWEIGHT 2-3
  605. +%if %0==2
  606. +    mova       m0, [r0+%2]
  607. +    mova       m1, [r1+%2]
  608. +    punpckhwd  m2, m0, m1
  609. +    punpcklwd  m0, m1
  610. +%else
  611. +    movq       m0, [r0+%2]
  612. +    movq       m1, [r1+%2]
  613. +    punpcklwd  m0, m1
  614. +    movq       m2, [r0+%3]
  615. +    movq       m1, [r1+%3]
  616. +    punpcklwd  m2, m1
  617. +%endif
  618. +    pmaddwd    m0, m4
  619. +    pmaddwd    m2, m4
  620. +    paddd      m0, m5
  621. +    paddd      m2, m5
  622. +    psrad      m0, m6
  623. +    psrad      m2, m6
  624. +%ifidn %1, sse4
  625. +    packusdw   m0, m2
  626. +    pminsw     m0, m3
  627. +%else
  628. +    packssdw   m0, m2
  629. +    CLIPW      m0, m7, m3
  630. +%endif
  631. +%endmacro
  632. +
  633. +%macro BIWEIGHT_FUNC_DBL 1
  634. +cglobal h264_biweight_16x16_10_%1, 7,7,8
  635. +    BIWEIGHT_SETUP %1
  636. +    mov        r3, 16
  637. +.nextrow
  638. +    BIWEIGHT   %1,  0
  639. +    mova  [r0   ], m0
  640. +    BIWEIGHT   %1, 16
  641. +    mova  [r0+16], m0
  642. +    add        r0, r2
  643. +    add        r1, r2
  644. +    dec        r3
  645. +    jnz .nextrow
  646. +    REP_RET
  647. +
  648. +cglobal h264_biweight_16x8_10_%1, 7,7,8
  649. +    BIWEIGHT_SETUP %1
  650. +    mov r3, 8
  651. +    jmp mangle(ff_h264_biweight_16x16_10_%1.nextrow)
  652. +%endmacro
  653. +
  654. +INIT_XMM
  655. +BIWEIGHT_FUNC_DBL sse2
  656. +BIWEIGHT_FUNC_DBL sse4
  657. +
  658. +%macro BIWEIGHT_FUNC 1
  659. +cglobal h264_biweight_8x16_10_%1, 7,7,8
  660. +    BIWEIGHT_SETUP %1
  661. +    mov      r3, 16
  662. +.nextrow
  663. +    BIWEIGHT %1, 0
  664. +    mova   [r0], m0
  665. +    add      r0, r2
  666. +    add      r1, r2
  667. +    dec      r3
  668. +    jnz .nextrow
  669. +    REP_RET
  670. +
  671. +cglobal h264_biweight_8x8_10_%1, 7,7,8
  672. +    BIWEIGHT_SETUP %1
  673. +    mov r3, 8
  674. +    jmp mangle(ff_h264_biweight_8x16_10_%1.nextrow)
  675. +
  676. +cglobal h264_biweight_8x4_10_%1, 7,7,8
  677. +    BIWEIGHT_SETUP %1
  678. +    mov r3, 4
  679. +    jmp mangle(ff_h264_biweight_8x16_10_%1.nextrow)
  680. +%endmacro
  681. +
  682. +INIT_XMM
  683. +BIWEIGHT_FUNC sse2
  684. +BIWEIGHT_FUNC sse4
  685. +
  686. +%macro BIWEIGHT_FUNC_HALF 1
  687. +cglobal h264_biweight_4x8_10_%1, 7,7,8
  688. +    BIWEIGHT_SETUP %1
  689. +    mov        r3, 4
  690. +    lea        r4, [r2*2]
  691. +.nextrow
  692. +    BIWEIGHT    %1, 0, r2
  693. +    movh   [r0   ], m0
  694. +    movhps [r0+r2], m0
  695. +    add         r0, r4
  696. +    add         r1, r4
  697. +    dec         r3
  698. +    jnz .nextrow
  699. +    REP_RET
  700. +
  701. +cglobal h264_biweight_4x4_10_%1, 7,7,8
  702. +    BIWEIGHT_SETUP %1
  703. +    mov        r3, 2
  704. +    lea        r4, [r2*2]
  705. +    jmp mangle(ff_h264_biweight_4x8_10_%1.nextrow)
  706. +
  707. +cglobal h264_biweight_4x2_10_%1, 7,7,8
  708. +    BIWEIGHT_SETUP %1
  709. +    mov        r3, 2
  710. +    lea        r4, [r2*2]
  711. +    jmp mangle(ff_h264_biweight_4x8_10_%1.nextrow)
  712. +%endmacro
  713. +
  714. +INIT_XMM
  715. +BIWEIGHT_FUNC_HALF sse2
  716. +BIWEIGHT_FUNC_HALF sse4
  717. diff --git a/libavcodec/x86/h264dsp_mmx.c b/libavcodec/x86/h264dsp_mmx.c
  718. index 3fccd08..d57314f 100644
  719. --- a/libavcodec/x86/h264dsp_mmx.c
  720. +++ b/libavcodec/x86/h264dsp_mmx.c
  721. @@ -326,6 +326,41 @@ H264_BIWEIGHT_MMX    ( 4,  8)
  722.  H264_BIWEIGHT_MMX    ( 4,  4)
  723.  H264_BIWEIGHT_MMX    ( 4,  2)
  724.  
  725. +#define H264_WEIGHT_10(W, H, DEPTH, OPT) \
  726. +void ff_h264_weight_ ## W ## x ## H ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
  727. +    int stride, int log2_denom, int weight, int offset);
  728. +
  729. +#define H264_BIWEIGHT_10(W, H, DEPTH, OPT) \
  730. +void ff_h264_biweight_ ## W ## x ## H ## _ ## DEPTH ## _ ## OPT \
  731. +    (uint8_t *dst, uint8_t *src, int stride, int log2_denom, \
  732. +   int weightd, int weights, int offset);
  733. +
  734. +#define H264_WEIGHT_10_SSE(W, H, DEPTH) \
  735. +H264_WEIGHT_10(W, H, DEPTH, sse2) \
  736. +H264_WEIGHT_10(W, H, DEPTH, sse4)
  737. +
  738. +#define H264_BIWEIGHT_10_SSE(W, H, DEPTH) \
  739. +H264_BIWEIGHT_10(W, H, DEPTH, sse2) \
  740. +H264_BIWEIGHT_10(W, H, DEPTH, sse4)
  741. +
  742. +H264_WEIGHT_10_SSE(16, 16, 10)
  743. +H264_WEIGHT_10_SSE(16,  8, 10)
  744. +H264_WEIGHT_10_SSE( 8, 16, 10)
  745. +H264_WEIGHT_10_SSE( 8,  8, 10)
  746. +H264_WEIGHT_10_SSE( 8,  4, 10)
  747. +H264_WEIGHT_10_SSE( 4,  8, 10)
  748. +H264_WEIGHT_10_SSE( 4,  4, 10)
  749. +H264_WEIGHT_10_SSE( 4,  2, 10)
  750. +
  751. +H264_BIWEIGHT_10_SSE(16, 16, 10)
  752. +H264_BIWEIGHT_10_SSE(16,  8, 10)
  753. +H264_BIWEIGHT_10_SSE( 8, 16, 10)
  754. +H264_BIWEIGHT_10_SSE( 8,  8, 10)
  755. +H264_BIWEIGHT_10_SSE( 8,  4, 10)
  756. +H264_BIWEIGHT_10_SSE( 4,  8, 10)
  757. +H264_BIWEIGHT_10_SSE( 4,  4, 10)
  758. +H264_BIWEIGHT_10_SSE( 4,  2, 10)
  759. +
  760.  void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
  761.  {
  762.      int mm_flags = av_get_cpu_flags();
  763. @@ -454,6 +489,24 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
  764.                  c->h264_idct8_add4     = ff_h264_idct8_add4_10_sse2;
  765.  #endif
  766.  
  767. +                c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_10_sse2;
  768. +                c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_10_sse2;
  769. +                c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_10_sse2;
  770. +                c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_10_sse2;
  771. +                c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_10_sse2;
  772. +                c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_10_sse2;
  773. +                c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_10_sse2;
  774. +                c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_10_sse2;
  775. +
  776. +                c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_10_sse2;
  777. +                c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_10_sse2;
  778. +                c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_10_sse2;
  779. +                c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_10_sse2;
  780. +                c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_10_sse2;
  781. +                c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_10_sse2;
  782. +                c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_10_sse2;
  783. +                c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_10_sse2;
  784. +
  785.                  c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_sse2;
  786.                  c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_sse2;
  787.  #if HAVE_ALIGNED_STACK
  788. @@ -463,6 +516,25 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
  789.                  c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
  790.  #endif
  791.              }
  792. +            if (mm_flags&AV_CPU_FLAG_SSE4) {
  793. +                c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_10_sse4;
  794. +                c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_10_sse4;
  795. +                c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_10_sse4;
  796. +                c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_10_sse4;
  797. +                c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_10_sse4;
  798. +                c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_10_sse4;
  799. +                c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_10_sse4;
  800. +                c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_10_sse4;
  801. +
  802. +                c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_10_sse4;
  803. +                c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_10_sse4;
  804. +                c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_10_sse4;
  805. +                c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_10_sse4;
  806. +                c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_10_sse4;
  807. +                c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_10_sse4;
  808. +                c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_10_sse4;
  809. +                c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_10_sse4;
  810. +            }
  811.  #if HAVE_AVX
  812.              if (mm_flags&AV_CPU_FLAG_AVX) {
  813.                  c->h264_idct_dc_add    =
  814. --
  815. 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!