API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Sep 15th, 2017
455
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Diff 26.81 KB | None | 0 0
raw download clone embed print report
  1. From 8a876dc2921aff14ef3173d75fbb3391c817c259 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 |  278 ++++++++++++++++++++++++++++++++
  12.  3 files changed, 311 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..d68a493 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, sse2)
  43. +CHROMA_MC(avg, 4, 10, sse2)
  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 +2665,10 @@ 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. +            }
  67.  
  68.              c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
  69.  #endif
  70. @@ -2756,6 +2774,12 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
  71.              H264_QPEL_FUNCS(3, 2, sse2);
  72.              H264_QPEL_FUNCS(3, 3, sse2);
  73.              }
  74. +            if (bit_depth == 10) {
  75. +                c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_sse2;
  76. +                c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_sse2;
  77. +                c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_10_sse2;
  78. +                c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_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..c4ad900
  101. --- /dev/null
  102. +++ b/libavcodec/x86/h264_chromamc_10bit.asm
  103. @@ -0,0 +1,278 @@
  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. +%macro MC4_OP 2
  261. +    movq          %1, [r1  ]
  262. +    movq          m1, [r1+2]
  263. +    add           r1, r2
  264. +    pmullw        %1, m4
  265. +    pmullw        m1, m2
  266. +    paddw         m1, %1
  267. +    mova          %1, m1
  268. +
  269. +    pmullw        %2, m5
  270. +    pmullw        m1, m3
  271. +    paddw         %2, [pw_32]
  272. +    paddw         m1, %2
  273. +    psrlw         m1, 6
  274. +    CHROMAMC_AVG4 m1, %2, [r0]
  275. +    movq        [r0], m1
  276. +    add           r0, r2
  277. +%endmacro
  278. +
  279. +%macro CHROMA_MC4 2
  280. +cglobal %1_h264_chroma_mc4_10_%2, 6,6,7
  281. +    movsxdifnidn  r2, r2d
  282. +    movd          m2, r4m         ; x
  283. +    movd          m3, r5m         ; y
  284. +    mova          m4, [pw_8]
  285. +    mova          m5, m4
  286. +    SPLATW        m2, m2
  287. +    SPLATW        m3, m3
  288. +    psubw         m4, m2
  289. +    psubw         m5, m3
  290. +
  291. +    movq          m0, [r1  ]
  292. +    movq          m6, [r1+2]
  293. +    add           r1, r2
  294. +    pmullw        m0, m4
  295. +    pmullw        m6, m2
  296. +    paddw         m6, m0
  297. +
  298. +.next2rows
  299. +    MC4_OP m0, m6
  300. +    MC4_OP m6, m0
  301. +    sub   r3d, 2
  302. +    jnz .next2rows
  303. +    REP_RET
  304. +%endmacro
  305. +
  306. +;-----------------------------------------------------------------------------
  307. +; void put/avg_h264_chroma_mc2(pixel *dst, pixel *src, int stride, int h, int mx, int my)
  308. +;-----------------------------------------------------------------------------
  309. +%macro CHROMA_MC2 2
  310. +cglobal %1_h264_chroma_mc2_10_%2, 6,7
  311. +    movsxdifnidn  r2, r2d
  312. +    mov          r6d, r4d
  313. +    shl          r4d, 16
  314. +    sub          r4d, r6d
  315. +    add          r4d, 8
  316. +    imul         r5d, r4d         ; x*y<<16 | y*(8-x)
  317. +    shl          r4d, 3
  318. +    sub          r4d, r5d         ; x*(8-y)<<16 | (8-x)*(8-y)
  319. +
  320. +    movd          m5, r4d
  321. +    movd          m6, r5d
  322. +    punpckldq     m5, m5          ; mm5 = {A,B,A,B}
  323. +    punpckldq     m6, m6          ; mm6 = {C,D,C,D}
  324. +    movq          m2, [r1]
  325. +    pxor          m7, m7
  326. +    pshufw        m2, m2, 0x94    ; mm0 = src[0,1,1,2]
  327. +
  328. +.nextrow
  329. +    add           r1, r2
  330. +    movq          m1, m2
  331. +    pmaddwd       m1, m5          ; mm1 = A * src[0,1] + B * src[1,2]
  332. +    movq          m0, [r1]
  333. +    pshufw        m0, m0, 0x94    ; mm0 = src[0,1,1,2]
  334. +    movq          m2, m0
  335. +    pmaddwd       m0, m6
  336. +    paddw         m1, [pw_32]
  337. +    paddw         m1, m0          ; mm1 += C * src[0,1] + D * src[1,2]
  338. +    psrlw         m1, 6
  339. +    packssdw      m1, m7
  340. +    CHROMAMC_AVG4 m1, m3, [r0]
  341. +    movd        [r0], m1
  342. +    add           r0, r2
  343. +    sub          r3d, 1
  344. +    jnz .nextrow
  345. +    REP_RET
  346. +%endmacro
  347. +
  348. +%macro NOTHING 2-3
  349. +%endmacro
  350. +%macro DIRECT_AVG 2
  351. +    PAVG          %1, %2
  352. +%endmacro
  353. +%macro COPY_AVG 3
  354. +    movq          %2, %3
  355. +    PAVG          %1, %2
  356. +%endmacro
  357. +
  358. +%define CHROMAMC_AVG  NOTHING
  359. +%define CHROMAMC_AVG4 NOTHING
  360. +INIT_XMM
  361. +CHROMA_MC8 put, sse2
  362. +%ifdef HAVE_AVX
  363. +CHROMA_MC8 put, avx
  364. +%endif
  365. +INIT_XMM
  366. +CHROMA_MC4 put, sse2
  367. +INIT_MMX
  368. +CHROMA_MC2 put, mmxext
  369. +
  370. +%define CHROMAMC_AVG  DIRECT_AVG
  371. +%define CHROMAMC_AVG4 COPY_AVG
  372. +%define PAVG          pavgw
  373. +INIT_XMM
  374. +CHROMA_MC8 avg, sse2
  375. +%ifdef HAVE_AVX
  376. +CHROMA_MC8 avg, avx
  377. +%endif
  378. +INIT_XMM
  379. +CHROMA_MC4 avg, sse2
  380. +INIT_MMX
  381. +CHROMA_MC2 avg, mmxext
  382. --
  383. 1.7.5.1
  384.  
  385.  
  386. From ab30144e5c598bdf1e058d445dd09ab5633cf5d0 Mon Sep 17 00:00:00 2001
  387. From: Daniel Kang <daniel.d.kang@gmail.com>
  388. Date: Mon, 6 Jun 2011 05:50:49 -0400
  389. Subject: [PATCH 2/2] weight WIP
  390.  
  391. ---
  392. libavcodec/x86/Makefile              |    1 +
  393.  libavcodec/x86/h264_weight_10bit.asm |  301 ++++++++++++++++++++++++++++++++++
  394.  libavcodec/x86/h264dsp_mmx.c         |   72 ++++++++
  395.  3 files changed, 374 insertions(+), 0 deletions(-)
  396.  create mode 100644 libavcodec/x86/h264_weight_10bit.asm
  397.  
  398. diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
  399. index ea57bd1..022ab27 100644
  400. --- a/libavcodec/x86/Makefile
  401. +++ b/libavcodec/x86/Makefile
  402. @@ -15,6 +15,7 @@ YASM-OBJS-$(CONFIG_H264DSP)            += x86/h264_deblock.o            \
  403.                                            x86/h264_idct.o               \
  404.                                            x86/h264_idct_10bit.o         \
  405.                                            x86/h264_weight.o             \
  406. +                                          x86/h264_weight_10bit.o       \
  407.  
  408.  YASM-OBJS-$(CONFIG_H264PRED)           += x86/h264_intrapred.o          \
  409.                                            x86/h264_intrapred_10bit.o
  410. diff --git a/libavcodec/x86/h264_weight_10bit.asm b/libavcodec/x86/h264_weight_10bit.asm
  411. new file mode 100644
  412. index 0000000..55bc8bc
  413. --- /dev/null
  414. +++ b/libavcodec/x86/h264_weight_10bit.asm
  415. @@ -0,0 +1,301 @@
  416. +;*****************************************************************************
  417. +;* MMX/SSE2/AVX-optimized 10-bit H.264 weighted prediction code
  418. +;*****************************************************************************
  419. +;* Copyright (C) 2005-2011 x264 project
  420. +;*
  421. +;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
  422. +;*
  423. +;* This file is part of Libav.
  424. +;*
  425. +;* Libav is free software; you can redistribute it and/or
  426. +;* modify it under the terms of the GNU Lesser General Public
  427. +;* License as published by the Free Software Foundation; either
  428. +;* version 2.1 of the License, or (at your option) any later version.
  429. +;*
  430. +;* Libav is distributed in the hope that it will be useful,
  431. +;* but WITHOUT ANY WARRANTY; without even the implied warranty of
  432. +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  433. +;* Lesser General Public License for more details.
  434. +;*
  435. +;* You should have received a copy of the GNU Lesser General Public
  436. +;* License along with Libav; if not, write to the Free Software
  437. +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  438. +;******************************************************************************
  439. +
  440. +%include "x86inc.asm"
  441. +%include "x86util.asm"
  442. +
  443. +SECTION_RODATA 32
  444. +
  445. +pw_pixel_max: times 8 dw ((1 << 10)-1)
  446. +sq_1: dq 1
  447. +      dq 0
  448. +
  449. +cextern pw_1
  450. +
  451. +SECTION .text
  452. +
  453. +;-----------------------------------------------------------------------------
  454. +; void h264_weight(uint8_t *dst, int stride, int log2_denom,
  455. +;                  int weight, int offset);
  456. +;-----------------------------------------------------------------------------
  457. +
  458. +%macro WEIGHT_SETUP 1
  459. +    mova       m0, [pw_1]
  460. +    movd       m2, r2d
  461. +    pslld      m0, m2       ; 1<<log2_denom
  462. +    lea        r4, [r4*8+1] ; 1+(offset<<3)
  463. +    shl        r3, 1        ; weight << 1
  464. +    shl        r4, 16
  465. +    or         r4, r3
  466. +    movd       m3, r4d      ; weight<<1 | 1+(offset<<(3))
  467. +    pshufd     m3, m3, 0
  468. +    mova       m4, [pw_pixel_max]
  469. +    paddw      m2, [sq_1]   ; denom+1
  470. +%ifnidn %1, sse4
  471. +    pxor       m7, m7
  472. +%endif
  473. +%endmacro
  474. +
  475. +%macro WEIGHT_OP 3
  476. +;;;;;;;;;;;;;;;; can be optimized
  477. +    movq        m5, [r0+%2]
  478. +    movq        m6, [r0+%3]
  479. +    punpcklwd   m5, m0
  480. +    punpcklwd   m6, m0
  481. +    pmaddwd     m5, m3
  482. +    pmaddwd     m6, m3
  483. +    psrad       m5, m2
  484. +    psrad       m6, m2
  485. +%ifidn %1, sse4
  486. +    packusdw    m5, m6
  487. +    pminsw      m5, m4
  488. +%else
  489. +    packssdw    m5, m6
  490. +    CLIPW       m5, m7, m4
  491. +%endif
  492. +%endmacro
  493. +
  494. +%macro WEIGHT_FUNC_DBL 1
  495. +cglobal h264_weight_16x16_10_%1, 5,5,8
  496. +    WEIGHT_SETUP %1
  497. +    mov        r2, 16
  498. +.nextrow
  499. +    WEIGHT_OP %1,  0,  8
  500. +    mova     [r0   ], m5
  501. +    WEIGHT_OP %1, 16, 24
  502. +    mova     [r0+16], m5
  503. +    add        r0, r1
  504. +    dec        r2
  505. +    jnz .nextrow
  506. +    REP_RET
  507. +
  508. +cglobal h264_weight_16x8_10_%1, 5,5,8
  509. +    WEIGHT_SETUP %1
  510. +    mov        r2, 8
  511. +    jmp mangle(ff_h264_weight_16x16_10_%1.nextrow)
  512. +%endmacro
  513. +
  514. +INIT_XMM
  515. +WEIGHT_FUNC_DBL sse2
  516. +WEIGHT_FUNC_DBL sse4
  517. +
  518. +
  519. +%macro WEIGHT_FUNC_MM 1
  520. +cglobal h264_weight_8x16_10_%1, 7,7,8
  521. +    WEIGHT_SETUP %1
  522. +    mov        r2, 16
  523. +.nextrow
  524. +    WEIGHT_OP  %1, 0, 8
  525. +    mova     [r0], m5
  526. +    add        r0, r1
  527. +    dec        r2
  528. +    jnz .nextrow
  529. +    REP_RET
  530. +
  531. +cglobal h264_weight_8x8_10_%1, 7,7,8
  532. +    WEIGHT_SETUP %1
  533. +    mov        r2, 8
  534. +    jmp mangle(ff_h264_weight_8x16_10_%1.nextrow)
  535. +
  536. +cglobal h264_weight_8x4_10_%1, 7,7,8
  537. +    WEIGHT_SETUP %1
  538. +    mov        r2, 4
  539. +    jmp mangle(ff_h264_weight_8x16_10_%1.nextrow)
  540. +%endmacro
  541. +
  542. +INIT_XMM
  543. +WEIGHT_FUNC_MM sse2
  544. +WEIGHT_FUNC_MM sse4
  545. +
  546. +
  547. +%macro WEIGHT_FUNC_HALF_MM 1
  548. +cglobal h264_weight_4x8_10_%1, 5,5,8
  549. +    WEIGHT_SETUP %1
  550. +    mov        r2, 4
  551. +    lea        r3, [r1*2]
  552. +.nextrow
  553. +    WEIGHT_OP   %1, 0, r1
  554. +    movh      [r0], m5
  555. +    movhps [r0+r1], m5
  556. +    add         r0, r3
  557. +    dec         r2
  558. +    jnz .nextrow
  559. +    REP_RET
  560. +
  561. +cglobal h264_weight_4x4_10_%1, 5,5,8
  562. +    WEIGHT_SETUP %1
  563. +    mov        r2, 2
  564. +    lea        r3, [r1*2]
  565. +    jmp mangle(ff_h264_weight_4x8_10_%1.nextrow)
  566. +
  567. +;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; NEEDS TO BE TESTED
  568. +cglobal h264_weight_4x2_10_%1, 5,5,8
  569. +int 3
  570. +    WEIGHT_SETUP %1
  571. +    mov        r2, 1
  572. +    lea        r3, [r1*2]
  573. +    jmp mangle(ff_h264_weight_4x8_10_%1.nextrow)
  574. +%endmacro
  575. +
  576. +INIT_XMM
  577. +WEIGHT_FUNC_HALF_MM sse2
  578. +WEIGHT_FUNC_HALF_MM sse4
  579. +
  580. +
  581. +;-----------------------------------------------------------------------------
  582. +; void h264_biweight(uint8_t *dst, uint8_t *src, int stride, int log2_denom,
  583. +;                    int weightd, int weights, int offset);
  584. +;-----------------------------------------------------------------------------
  585. +%macro BIWEIGHT_SETUP 1
  586. +    add        r6, 1
  587. +    or         r6, 1
  588. +    shl        r5, 16
  589. +    or         r4, r5
  590. +    movd       m4, r4d ; weightd | weights
  591. +    movd       m5, r6d ; (offset+1)|1
  592. +    movd       m6, r3d ; log2_denom
  593. +    pslld      m5, m6  ; ((offset+1)|1)<<log2_denom
  594. +    paddd      m6, [sq_1]
  595. +    pshufd     m4, m4, 0
  596. +    pshufd     m5, m5, 0
  597. +    mova       m3, [pw_pixel_max]
  598. +%ifnidn %1, sse4
  599. +    pxor       m7, m7
  600. +%endif
  601. +%endmacro
  602. +
  603. +%macro BIWEIGHT 2-3
  604. +%if %0==2
  605. +    mova       m0, [r0+%2]
  606. +    mova       m1, [r1+%2]
  607. +    punpckhwd  m2, m0, m1
  608. +    punpcklwd  m0, m1
  609. +%else
  610. +    movq       m0, [r0+%2]
  611. +    movq       m1, [r1+%2]
  612. +    punpcklwd  m0, m1
  613. +    movq       m2, [r0+%3]
  614. +    movq       m1, [r1+%3]
  615. +    punpcklwd  m2, m1
  616. +%endif
  617. +    pmaddwd    m0, m4
  618. +    pmaddwd    m2, m4
  619. +    paddd      m0, m5
  620. +    paddd      m2, m5
  621. +    psrad      m0, m6
  622. +    psrad      m2, m6
  623. +%ifidn %1, sse4
  624. +    packusdw   m0, m2
  625. +    pminsw     m0, m3
  626. +%else
  627. +    packssdw   m0, m2
  628. +    CLIPW      m0, m7, m3
  629. +%endif
  630. +%endmacro
  631. +
  632. +%macro BIWEIGHT_FUNC_DBL 1
  633. +cglobal h264_biweight_16x16_10_%1, 7,7,8
  634. +    BIWEIGHT_SETUP %1
  635. +    mov        r3, 16
  636. +.nextrow
  637. +    BIWEIGHT   %1,  0
  638. +    mova  [r0   ], m0
  639. +    BIWEIGHT   %1, 16
  640. +    mova  [r0+16], m0
  641. +    add        r0, r2
  642. +    add        r1, r2
  643. +    dec        r3
  644. +    jnz .nextrow
  645. +    REP_RET
  646. +
  647. +cglobal h264_biweight_16x8_10_%1, 7,7,8
  648. +    BIWEIGHT_SETUP %1
  649. +    mov r3, 8
  650. +    jmp mangle(ff_h264_biweight_16x16_10_%1.nextrow)
  651. +%endmacro
  652. +
  653. +INIT_XMM
  654. +BIWEIGHT_FUNC_DBL sse2
  655. +BIWEIGHT_FUNC_DBL sse4
  656. +
  657. +%macro BIWEIGHT_FUNC 1
  658. +cglobal h264_biweight_8x16_10_%1, 7,7,8
  659. +    BIWEIGHT_SETUP %1
  660. +    mov      r3, 16
  661. +.nextrow
  662. +    BIWEIGHT %1, 0
  663. +    mova   [r0], m0
  664. +    add      r0, r2
  665. +    add      r1, r2
  666. +    dec      r3
  667. +    jnz .nextrow
  668. +    REP_RET
  669. +
  670. +cglobal h264_biweight_8x8_10_%1, 7,7,8
  671. +    BIWEIGHT_SETUP %1
  672. +    mov r3, 8
  673. +    jmp mangle(ff_h264_biweight_8x16_10_%1.nextrow)
  674. +
  675. +cglobal h264_biweight_8x4_10_%1, 7,7,8
  676. +    BIWEIGHT_SETUP %1
  677. +    mov r3, 4
  678. +    jmp mangle(ff_h264_biweight_8x16_10_%1.nextrow)
  679. +%endmacro
  680. +
  681. +INIT_XMM
  682. +BIWEIGHT_FUNC sse2
  683. +BIWEIGHT_FUNC sse4
  684. +
  685. +%macro BIWEIGHT_FUNC_HALF 1
  686. +cglobal h264_biweight_4x8_10_%1, 7,7,8
  687. +    BIWEIGHT_SETUP %1
  688. +    mov        r3, 4
  689. +    lea        r4, [r2*2]
  690. +.nextrow
  691. +    BIWEIGHT    %1, 0, r2
  692. +    movh   [r0   ], m0
  693. +    movhps [r0+r2], m0
  694. +    add         r0, r4
  695. +    add         r1, r4
  696. +    dec         r3
  697. +    jnz .nextrow
  698. +    REP_RET
  699. +
  700. +cglobal h264_biweight_4x4_10_%1, 7,7,8
  701. +    BIWEIGHT_SETUP %1
  702. +    mov        r3, 2
  703. +    lea        r4, [r2*2]
  704. +    jmp mangle(ff_h264_biweight_4x8_10_%1.nextrow)
  705. +
  706. +cglobal h264_biweight_4x2_10_%1, 7,7,8
  707. +;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; NEEDS TO BE TESTED
  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!