API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Sep 22nd, 2017
458
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Diff 23.25 KB | None | 0 0
raw download clone embed print report
  1. From d0974fd9e01cd4be6deb56c53dc867e876945926 Mon Sep 17 00:00:00 2001
  2. From: Jason Garrett-Glaser <jason@x264.com>
  3. Date: Tue, 5 Jul 2011 18:08:22 -0700
  4. Subject: [PATCH 1/6] H.264: av_always_inline some more functions
  5.  
  6. These weren't getting inlined all the time in all gcc versions.
  7. ---
  8. libavcodec/h264.c        |    4 ++--
  9.  libavcodec/h264.h        |    4 ++--
  10.  libavcodec/h264_mvpred.h |    8 ++++----
  11.  3 files changed, 8 insertions(+), 8 deletions(-)
  12.  
  13. diff --git a/libavcodec/h264.c b/libavcodec/h264.c
  14. index dbac4fd..f7c0345 100644
  15. --- a/libavcodec/h264.c
  16. +++ b/libavcodec/h264.c
  17. @@ -1443,7 +1443,7 @@ static void decode_postinit(H264Context *h, int setup_finished){
  18.          ff_thread_finish_setup(s->avctx);
  19.  }
  20.  
  21. -static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int chroma444, int simple){
  22. +static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int chroma444, int simple){
  23.      MpegEncContext * const s = &h->s;
  24.      uint8_t *top_border;
  25.      int top_idx = 1;
  26. @@ -1518,7 +1518,7 @@ static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src
  27.      }
  28.  }
  29.  
  30. -static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
  31. +static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
  32.                                    uint8_t *src_cb, uint8_t *src_cr,
  33.                                    int linesize, int uvlinesize,
  34.                                    int xchg, int chroma444,
  35. diff --git a/libavcodec/h264.h b/libavcodec/h264.h
  36. index dad06e2..a2abab9 100644
  37. --- a/libavcodec/h264.h
  38. +++ b/libavcodec/h264.h
  39. @@ -766,11 +766,11 @@ static av_always_inline uint16_t pack8to16(int a, int b){
  40.  /**
  41.   * gets the chroma qp.
  42.   */
  43. -static inline int get_chroma_qp(H264Context *h, int t, int qscale){
  44. +static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale){
  45.      return h->pps.chroma_qp_table[t][qscale];
  46.  }
  47.  
  48. -static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
  49. +static av_always_inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
  50.  
  51.  static void fill_decode_neighbors(H264Context *h, int mb_type){
  52.      MpegEncContext * const s = &h->s;
  53. diff --git a/libavcodec/h264_mvpred.h b/libavcodec/h264_mvpred.h
  54. index f603e7f..c4e65b5 100644
  55. --- a/libavcodec/h264_mvpred.h
  56. +++ b/libavcodec/h264_mvpred.h
  57. @@ -35,7 +35,7 @@
  58.  //#undef NDEBUG
  59.  #include <assert.h>
  60.  
  61. -static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
  62. +static av_always_inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
  63.      const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
  64.      MpegEncContext *s = &h->s;
  65.  
  66. @@ -92,7 +92,7 @@ static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, in
  67.   * @param mx the x component of the predicted motion vector
  68.   * @param my the y component of the predicted motion vector
  69.   */
  70. -static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
  71. +static av_always_inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
  72.      const int index8= scan8[n];
  73.      const int top_ref=      h->ref_cache[list][ index8 - 8 ];
  74.      const int left_ref=     h->ref_cache[list][ index8 - 1 ];
  75. @@ -147,7 +147,7 @@ static inline void pred_motion(H264Context * const h, int n, int part_width, int
  76.   * @param mx the x component of the predicted motion vector
  77.   * @param my the y component of the predicted motion vector
  78.   */
  79. -static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
  80. +static av_always_inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
  81.      if(n==0){
  82.          const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
  83.          const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
  84. @@ -182,7 +182,7 @@ static inline void pred_16x8_motion(H264Context * const h, int n, int list, int
  85.   * @param mx the x component of the predicted motion vector
  86.   * @param my the y component of the predicted motion vector
  87.   */
  88. -static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
  89. +static av_always_inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
  90.      if(n==0){
  91.          const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
  92.          const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
  93. --
  94. 1.7.3.2.146.gca209
  95.  
  96.  
  97. From e61600fa054188eea481a55233b67b2d33f41892 Mon Sep 17 00:00:00 2001
  98. From: Jason Garrett-Glaser <jason@x264.com>
  99. Date: Tue, 5 Jul 2011 17:55:14 -0700
  100. Subject: [PATCH 2/6] H.264: faster P-SKIP decoding
  101.  
  102. Inline the relevant parts of fill_decode_caches into P-SKIP mv prediction to
  103. avoid calling the whole thing.
  104. ---
  105. libavcodec/h264.h        |    1 -
  106.  libavcodec/h264_mvpred.h |  108 ++++++++++++++++++++++++++++++++++++++++++----
  107.  2 files changed, 99 insertions(+), 10 deletions(-)
  108.  
  109. diff --git a/libavcodec/h264.h b/libavcodec/h264.h
  110. index a2abab9..d34e6db 100644
  111. --- a/libavcodec/h264.h
  112. +++ b/libavcodec/h264.h
  113. @@ -1331,7 +1331,6 @@ static void av_unused decode_mb_skip(H264Context *h){
  114.          mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
  115.  
  116.          fill_decode_neighbors(h, mb_type);
  117. -        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
  118.          pred_pskip_motion(h, &mx, &my);
  119.          fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
  120.          fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
  121. diff --git a/libavcodec/h264_mvpred.h b/libavcodec/h264_mvpred.h
  122. index c4e65b5..27adac6 100644
  123. --- a/libavcodec/h264_mvpred.h
  124. +++ b/libavcodec/h264_mvpred.h
  125. @@ -213,21 +213,111 @@ static av_always_inline void pred_8x16_motion(H264Context * const h, int n, int
  126.      pred_motion(h, n, 2, list, ref, mx, my);
  127.  }
  128.  
  129. -static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
  130. -    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
  131. -    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
  132. -
  133. -    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
  134. +#define FIX_MV_MBAFF(type, refn, mvn, idx)\
  135. +    if(FRAME_MBAFF){\
  136. +        if(MB_FIELD){\
  137. +            if(!IS_INTERLACED(type)){\
  138. +                refn <<= 1;\
  139. +                AV_COPY32(mvbuf[idx], mvn);\
  140. +                mvbuf[idx][1] /= 2;\
  141. +                mvn = mvbuf[idx];\
  142. +            }\
  143. +        }else{\
  144. +            if(IS_INTERLACED(type)){\
  145. +                refn >>= 1;\
  146. +                AV_COPY32(mvbuf[idx], mvn);\
  147. +                mvbuf[idx][1] <<= 1;\
  148. +                mvn = mvbuf[idx];\
  149. +            }\
  150. +        }\
  151. +    }
  152.  
  153. -    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
  154. -       || !( top_ref | AV_RN32A(h->mv_cache[0][ scan8[0] - 8 ]))
  155. -       || !(left_ref | AV_RN32A(h->mv_cache[0][ scan8[0] - 1 ]))){
  156. +static av_always_inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
  157. +    DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = {0};
  158. +    DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2];
  159. +    MpegEncContext * const s = &h->s;
  160. +    int8_t *ref = s->current_picture.ref_index[0];
  161. +    int16_t (*mv)[2] = s->current_picture.motion_val[0];
  162. +    int top_ref, left_ref, diagonal_ref, match_count;
  163. +    const int16_t *A, *B, *C;
  164. +    int b_stride = h->b_stride;
  165. +
  166. +    /* To avoid doing an entire fill_decode_caches, we inline the relevant parts here. */
  167. +    if(USES_LIST(h->left_type[LTOP], 0)){
  168. +        left_ref = ref[4*h->left_mb_xy[LTOP] + 1 + (h->left_block[0]&~1)];
  169. +        A = mv[h->mb2b_xy[h->left_mb_xy[LTOP]] + 3 + b_stride*h->left_block[0]];
  170. +        FIX_MV_MBAFF(h->left_type[LTOP], left_ref, A, 0);
  171. +        if(!(left_ref | AV_RN32A(A))){
  172. +            *mx = *my = 0;
  173. +            return;
  174. +        }
  175. +    }else if(h->left_type[LTOP]){
  176. +        left_ref = LIST_NOT_USED;
  177. +        A = zeromv;
  178. +    }else{
  179. +        *mx = *my = 0;
  180. +        return;
  181. +    }
  182.  
  183. +    if(USES_LIST(h->top_type, 0)){
  184. +        top_ref = ref[4*h->top_mb_xy + 2];
  185. +        B = mv[h->mb2b_xy[h->top_mb_xy] + 3*b_stride];
  186. +        FIX_MV_MBAFF(h->top_type, top_ref, B, 1);
  187. +        if(!(top_ref | AV_RN32A(B))){
  188. +            *mx = *my = 0;
  189. +            return;
  190. +        }
  191. +    }else if(h->top_type){
  192. +        top_ref = LIST_NOT_USED;
  193. +        B = zeromv;
  194. +    }else{
  195.          *mx = *my = 0;
  196.          return;
  197.      }
  198.  
  199. -    pred_motion(h, 0, 4, 0, 0, mx, my);
  200. +    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
  201. +
  202. +    if(USES_LIST(h->topright_type, 0)){
  203. +        diagonal_ref = ref[4*h->topright_mb_xy + 2];
  204. +        C = mv[h->mb2b_xy[h->topright_mb_xy] + 3*b_stride];
  205. +        FIX_MV_MBAFF(h->topright_type, diagonal_ref, C, 2);
  206. +    }else if(h->topright_type){
  207. +        diagonal_ref = LIST_NOT_USED;
  208. +        C = zeromv;
  209. +    }else{
  210. +        if(USES_LIST(h->topleft_type, 0)){
  211. +            diagonal_ref = ref[4*h->topleft_mb_xy + 1 + (h->topleft_partition & 2)];
  212. +            C = mv[h->mb2b_xy[h->topleft_mb_xy] + 3 + b_stride + (h->topleft_partition & 2*b_stride)];
  213. +            FIX_MV_MBAFF(h->topleft_type, diagonal_ref, C, 2);
  214. +        }else if(h->topleft_type){
  215. +            diagonal_ref = LIST_NOT_USED;
  216. +            C = zeromv;
  217. +        }else{
  218. +            diagonal_ref = PART_NOT_AVAILABLE;
  219. +            C = zeromv;
  220. +        }
  221. +    }
  222. +
  223. +    match_count= !diagonal_ref + !top_ref + !left_ref;
  224. +    tprintf(h->s.avctx, "pred_pskip_motion match_count=%d\n", match_count);
  225. +    if(match_count > 1){
  226. +        *mx= mid_pred(A[0], B[0], C[0]);
  227. +        *my= mid_pred(A[1], B[1], C[1]);
  228. +    }else if(match_count==1){
  229. +        if(!left_ref){
  230. +            *mx= A[0];
  231. +            *my= A[1];
  232. +        }else if(!top_ref){
  233. +            *mx= B[0];
  234. +            *my= B[1];
  235. +        }else{
  236. +            *mx= C[0];
  237. +            *my= C[1];
  238. +        }
  239. +    }else{
  240. +        *mx= mid_pred(A[0], B[0], C[0]);
  241. +        *my= mid_pred(A[1], B[1], C[1]);
  242. +    }
  243.  
  244.      return;
  245.  }
  246. --
  247. 1.7.3.2.146.gca209
  248.  
  249.  
  250. From d35a5612ef3aa50d7b91d495c0318639d2415a62 Mon Sep 17 00:00:00 2001
  251. From: Jason Garrett-Glaser <jason@x264.com>
  252. Date: Wed, 6 Jul 2011 07:58:50 -0700
  253. Subject: [PATCH 3/6] H.264: merge fill_rectangle into P-SKIP MV prediction, to match B-SKIP
  254.  
  255. ---
  256. libavcodec/h264.h        |    7 ++-----
  257.  libavcodec/h264_mvpred.h |   42 ++++++++++++++++++++++--------------------
  258.  2 files changed, 24 insertions(+), 25 deletions(-)
  259.  
  260. diff --git a/libavcodec/h264.h b/libavcodec/h264.h
  261. index d34e6db..e88005a 100644
  262. --- a/libavcodec/h264.h
  263. +++ b/libavcodec/h264.h
  264. @@ -770,7 +770,7 @@ static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale){
  265.      return h->pps.chroma_qp_table[t][qscale];
  266.  }
  267.  
  268. -static av_always_inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
  269. +static av_always_inline void pred_pskip_motion(H264Context * const h);
  270.  
  271.  static void fill_decode_neighbors(H264Context *h, int mb_type){
  272.      MpegEncContext * const s = &h->s;
  273. @@ -1327,13 +1327,10 @@ static void av_unused decode_mb_skip(H264Context *h){
  274.      }
  275.      else
  276.      {
  277. -        int mx, my;
  278.          mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
  279.  
  280.          fill_decode_neighbors(h, mb_type);
  281. -        pred_pskip_motion(h, &mx, &my);
  282. -        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
  283. -        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
  284. +        pred_pskip_motion(h);
  285.      }
  286.  
  287.      write_back_motion(h, mb_type);
  288. diff --git a/libavcodec/h264_mvpred.h b/libavcodec/h264_mvpred.h
  289. index 27adac6..178eb24 100644
  290. --- a/libavcodec/h264_mvpred.h
  291. +++ b/libavcodec/h264_mvpred.h
  292. @@ -232,31 +232,31 @@ static av_always_inline void pred_8x16_motion(H264Context * const h, int n, int
  293.          }\
  294.      }
  295.  
  296. -static av_always_inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
  297. +static av_always_inline void pred_pskip_motion(H264Context * const h){
  298.      DECLARE_ALIGNED(4, static const int16_t, zeromv)[2] = {0};
  299.      DECLARE_ALIGNED(4, int16_t, mvbuf)[3][2];
  300.      MpegEncContext * const s = &h->s;
  301.      int8_t *ref = s->current_picture.ref_index[0];
  302.      int16_t (*mv)[2] = s->current_picture.motion_val[0];
  303. -    int top_ref, left_ref, diagonal_ref, match_count;
  304. +    int top_ref, left_ref, diagonal_ref, match_count, mx, my;
  305.      const int16_t *A, *B, *C;
  306.      int b_stride = h->b_stride;
  307.  
  308. +    fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
  309. +
  310.      /* To avoid doing an entire fill_decode_caches, we inline the relevant parts here. */
  311.      if(USES_LIST(h->left_type[LTOP], 0)){
  312.          left_ref = ref[4*h->left_mb_xy[LTOP] + 1 + (h->left_block[0]&~1)];
  313.          A = mv[h->mb2b_xy[h->left_mb_xy[LTOP]] + 3 + b_stride*h->left_block[0]];
  314.          FIX_MV_MBAFF(h->left_type[LTOP], left_ref, A, 0);
  315.          if(!(left_ref | AV_RN32A(A))){
  316. -            *mx = *my = 0;
  317. -            return;
  318. +            goto zeromv;
  319.          }
  320.      }else if(h->left_type[LTOP]){
  321.          left_ref = LIST_NOT_USED;
  322.          A = zeromv;
  323.      }else{
  324. -        *mx = *my = 0;
  325. -        return;
  326. +        goto zeromv;
  327.      }
  328.  
  329.      if(USES_LIST(h->top_type, 0)){
  330. @@ -264,15 +264,13 @@ static av_always_inline void pred_pskip_motion(H264Context * const h, int * cons
  331.          B = mv[h->mb2b_xy[h->top_mb_xy] + 3*b_stride];
  332.          FIX_MV_MBAFF(h->top_type, top_ref, B, 1);
  333.          if(!(top_ref | AV_RN32A(B))){
  334. -            *mx = *my = 0;
  335. -            return;
  336. +            goto zeromv;
  337.          }
  338.      }else if(h->top_type){
  339.          top_ref = LIST_NOT_USED;
  340.          B = zeromv;
  341.      }else{
  342. -        *mx = *my = 0;
  343. -        return;
  344. +        goto zeromv;
  345.      }
  346.  
  347.      tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
  348. @@ -301,24 +299,28 @@ static av_always_inline void pred_pskip_motion(H264Context * const h, int * cons
  349.      match_count= !diagonal_ref + !top_ref + !left_ref;
  350.      tprintf(h->s.avctx, "pred_pskip_motion match_count=%d\n", match_count);
  351.      if(match_count > 1){
  352. -        *mx= mid_pred(A[0], B[0], C[0]);
  353. -        *my= mid_pred(A[1], B[1], C[1]);
  354. +        mx = mid_pred(A[0], B[0], C[0]);
  355. +        my = mid_pred(A[1], B[1], C[1]);
  356.      }else if(match_count==1){
  357.          if(!left_ref){
  358. -            *mx= A[0];
  359. -            *my= A[1];
  360. +            mx = A[0];
  361. +            my = A[1];
  362.          }else if(!top_ref){
  363. -            *mx= B[0];
  364. -            *my= B[1];
  365. +            mx = B[0];
  366. +            my = B[1];
  367.          }else{
  368. -            *mx= C[0];
  369. -            *my= C[1];
  370. +            mx = C[0];
  371. +            my = C[1];
  372.          }
  373.      }else{
  374. -        *mx= mid_pred(A[0], B[0], C[0]);
  375. -        *my= mid_pred(A[1], B[1], C[1]);
  376. +        mx = mid_pred(A[0], B[0], C[0]);
  377. +        my = mid_pred(A[1], B[1], C[1]);
  378.      }
  379.  
  380. +    fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
  381. +    return;
  382. +zeromv:
  383. +    fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
  384.      return;
  385.  }
  386.  
  387. --
  388. 1.7.3.2.146.gca209
  389.  
  390.  
  391. From eda7f1ca1b9748a2027dfe81259b9140bddbdd61 Mon Sep 17 00:00:00 2001
  392. From: Jason Garrett-Glaser <jason@x264.com>
  393. Date: Wed, 6 Jul 2011 12:26:04 -0700
  394. Subject: [PATCH 4/6] H.264: Remove redundant hl_motion_16/8 code
  395.  
  396. ---
  397. libavcodec/h264.c |   52 ++++++++++------------------------------------------
  398.  1 files changed, 10 insertions(+), 42 deletions(-)
  399.  
  400. diff --git a/libavcodec/h264.c b/libavcodec/h264.c
  401. index f7c0345..f79f4d1 100644
  402. --- a/libavcodec/h264.c
  403. +++ b/libavcodec/h264.c
  404. @@ -778,24 +778,6 @@ static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t
  405.      prefetch_motion(h, 1, pixel_shift, chroma444);
  406.  }
  407.  
  408. -#define hl_motion_fn(sh, bits) \
  409. -static av_always_inline void hl_motion_ ## bits(H264Context *h, \
  410. -                                       uint8_t *dest_y, \
  411. -                                       uint8_t *dest_cb, uint8_t *dest_cr, \
  412. -                                       qpel_mc_func (*qpix_put)[16], \
  413. -                                       h264_chroma_mc_func (*chroma_put), \
  414. -                                       qpel_mc_func (*qpix_avg)[16], \
  415. -                                       h264_chroma_mc_func (*chroma_avg), \
  416. -                                       h264_weight_func *weight_op, \
  417. -                                       h264_biweight_func *weight_avg, \
  418. -                                       int chroma444) \
  419. -{ \
  420. -    hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put, \
  421. -              qpix_avg, chroma_avg, weight_op, weight_avg, sh, chroma444); \
  422. -}
  423. -hl_motion_fn(0, 8);
  424. -hl_motion_fn(1, 16);
  425. -
  426.  static void free_tables(H264Context *h, int free_rbsp){
  427.      int i;
  428.      H264Context *hx;
  429. @@ -1876,18 +1858,11 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
  430.              if(h->deblocking_filter)
  431.                  xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
  432.          }else if(is_h264){
  433. -            if (pixel_shift) {
  434. -                hl_motion_16(h, dest_y, dest_cb, dest_cr,
  435. -                             s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
  436. -                             s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
  437. -                             h->h264dsp.weight_h264_pixels_tab,
  438. -                             h->h264dsp.biweight_h264_pixels_tab, 0);
  439. -            } else
  440. -                hl_motion_8(h, dest_y, dest_cb, dest_cr,
  441. -                            s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
  442. -                            s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
  443. -                            h->h264dsp.weight_h264_pixels_tab,
  444. -                            h->h264dsp.biweight_h264_pixels_tab, 0);
  445. +            hl_motion(h, dest_y, dest_cb, dest_cr,
  446. +                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
  447. +                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
  448. +                      h->h264dsp.weight_h264_pixels_tab,
  449. +                      h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 0);
  450.          }
  451.  
  452.          hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
  453. @@ -2017,18 +1992,11 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
  454.              if(h->deblocking_filter)
  455.                  xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
  456.          }else{
  457. -            if (pixel_shift) {
  458. -                hl_motion_16(h, dest[0], dest[1], dest[2],
  459. -                             s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
  460. -                             s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
  461. -                             h->h264dsp.weight_h264_pixels_tab,
  462. -                             h->h264dsp.biweight_h264_pixels_tab, 1);
  463. -            } else
  464. -                hl_motion_8(h, dest[0], dest[1], dest[2],
  465. -                            s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
  466. -                            s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
  467. -                            h->h264dsp.weight_h264_pixels_tab,
  468. -                            h->h264dsp.biweight_h264_pixels_tab, 1);
  469. +            hl_motion(h, dest[0], dest[1], dest[2],
  470. +                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
  471. +                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
  472. +                      h->h264dsp.weight_h264_pixels_tab,
  473. +                      h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 1);
  474.          }
  475.  
  476.          for (p = 0; p < plane_count; p++)
  477. --
  478. 1.7.3.2.146.gca209
  479.  
  480.  
  481. From 50f29cfa15d6ddafd7ebc41f91b9131deebdfa03 Mon Sep 17 00:00:00 2001
  482. From: Jason Garrett-Glaser <jason@x264.com>
  483. Date: Wed, 6 Jul 2011 13:25:13 -0700
  484. Subject: [PATCH 5/6] H.264: use fill_rectangle in CABAC decoding
  485.  
  486. ---
  487. libavcodec/h264_cabac.c |    3 +--
  488.  1 files changed, 1 insertions(+), 2 deletions(-)
  489.  
  490. diff --git a/libavcodec/h264_cabac.c b/libavcodec/h264_cabac.c
  491. index 6dacf7a..390a7b6 100644
  492. --- a/libavcodec/h264_cabac.c
  493. +++ b/libavcodec/h264_cabac.c
  494. @@ -1818,8 +1818,7 @@ static av_always_inline void decode_cabac_luma_residual( H264Context *h, const u
  495.                      }
  496.                  }
  497.              } else {
  498. -                uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8+16*p] ];
  499. -                nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
  500. +                fill_rectangle(&h->non_zero_count_cache[scan8[4*i8x8+16*p]], 2, 2, 8, 0, 1);
  501.              }
  502.          }
  503.      }
  504. --
  505. 1.7.3.2.146.gca209
  506.  
  507.  
  508. From 1644142d17d231cd9a077db600587785dd8a354c Mon Sep 17 00:00:00 2001
  509. From: Jason Garrett-Glaser <jason@x264.com>
  510. Date: Wed, 6 Jul 2011 14:08:30 -0700
  511. Subject: [PATCH 6/6] H.264: improve qp_thresh check
  512.  
  513. Eliminate redundant check in filter_mb_fast, consider bit depth in calculating qp_thresh.
  514. ---
  515. libavcodec/h264.c            |    4 +++-
  516.  libavcodec/h264_loopfilter.c |    6 +-----
  517.  2 files changed, 4 insertions(+), 6 deletions(-)
  518.  
  519. diff --git a/libavcodec/h264.c b/libavcodec/h264.c
  520. index f79f4d1..33c9527 100644
  521. --- a/libavcodec/h264.c
  522. +++ b/libavcodec/h264.c
  523. @@ -2931,7 +2931,9 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
  524.              }
  525.          }
  526.      }
  527. -    h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
  528. +    h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
  529. +                 - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
  530. +                 + 6 * (h->sps.bit_depth_luma - 8);
  531.  
  532.  #if 0 //FMO
  533.      if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
  534. diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c
  535. index c716347..5de9f78 100644
  536. --- a/libavcodec/h264_loopfilter.c
  537. +++ b/libavcodec/h264_loopfilter.c
  538. @@ -216,7 +216,7 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
  539.      MpegEncContext * const s = &h->s;
  540.      int mb_xy;
  541.      int mb_type, left_type, top_type;
  542. -    int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
  543. +    int qp, qp0, qp1, qpc, qpc0, qpc1;
  544.      int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
  545.      int chroma444 = CHROMA444;
  546.  
  547. @@ -241,10 +241,6 @@ void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
  548.      qp1 = (qp + qp1 + 1) >> 1;
  549.      qpc0 = (qpc + qpc0 + 1) >> 1;
  550.      qpc1 = (qpc + qpc1 + 1) >> 1;
  551. -    qp_thresh = 15+52 - h->slice_alpha_c0_offset;
  552. -    if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
  553. -       qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
  554. -        return;
  555.  
  556.      if( IS_INTRA(mb_type) ) {
  557.          static const int16_t bS4[4] = {4,4,4,4};
  558. --
  559. 1.7.3.2.146.gca209
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!