API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Jan 17th, 2018
100
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Diff 8.64 KB | None | 0 0
raw download clone embed print report
  1. diff --git a/common/common.h b/common/common.h
  2. index 1406255..994be69 100644
  3. --- a/common/common.h
  4. +++ b/common/common.h
  5. @@ -424,6 +424,8 @@ struct x264_t
  6.      struct
  7.      {
  8.          ALIGNED_64( dctcoef luma16x16_dc[3][16] );
  9. +        ALIGNED_64( int misalign_pad_align_1 );
  10. +        int misalign_pad_1;
  11.          ALIGNED_16( dctcoef chroma_dc[2][8] );
  12.          // FIXME share memory?
  13.          ALIGNED_64( dctcoef luma8x8[12][64] );
  14. @@ -569,7 +571,11 @@ struct x264_t
  15.              ALIGNED_64( pixel fdec_buf[54*FDEC_STRIDE] );
  16.  
  17.              /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */
  18. +            ALIGNED_64( int misalign_pad_align_1 );
  19. +            int misalign_pad_1;
  20.              ALIGNED_16( pixel i4x4_fdec_buf[16*16] );
  21. +            ALIGNED_64( int misalign_pad_align_2 );
  22. +            int misalign_pad_2;
  23.              ALIGNED_16( pixel i8x8_fdec_buf[16*16] );
  24.              ALIGNED_64( dctcoef i8x8_dct_buf[3][64] );
  25.              ALIGNED_64( dctcoef i4x4_dct_buf[15][16] );
  26. @@ -577,11 +583,17 @@ struct x264_t
  27.              uint32_t i8x8_nnz_buf[4];
  28.  
  29.              /* Psy trellis DCT data */
  30. +            ALIGNED_64( int misalign_pad_align_3 );
  31. +            int misalign_pad_3;
  32.              ALIGNED_64( dctcoef fenc_dct8[4][64] );
  33. +            ALIGNED_64( int misalign_pad_align_4 );
  34. +            int misalign_pad_4;
  35.              ALIGNED_64( dctcoef fenc_dct4[16][16] );
  36.  
  37.              /* Psy RD SATD/SA8D scores cache */
  38.              ALIGNED_64( uint32_t fenc_satd_cache[32] );
  39. +            ALIGNED_64( int misalign_pad_align_5 );
  40. +            int misalign_pad_5;
  41.              ALIGNED_16( uint64_t fenc_hadamard_cache[9] );
  42.  
  43.              int i4x4_cbp;
  44. @@ -610,6 +622,8 @@ struct x264_t
  45.          struct
  46.          {
  47.              /* real intra4x4_pred_mode if I_4X4 or I_8X8, I_PRED_4x4_DC if mb available, -1 if not */
  48. +            ALIGNED_64( int misalign_pad_align_1 );
  49. +            int misalign_pad_1;
  50.              ALIGNED_16( int8_t intra4x4_pred_mode[X264_SCAN8_LUMA_SIZE] );
  51.  
  52.              /* i_non_zero_count if available else 0x80. intentionally misaligned by 8 for asm */
  53. @@ -619,6 +633,8 @@ struct x264_t
  54.              ALIGNED_4( int8_t ref[2][X264_SCAN8_LUMA_SIZE] );
  55.  
  56.              /* 0 if not available */
  57. +            ALIGNED_64( int misalign_pad_align_2 );
  58. +            int misalign_pad_2;
  59.              ALIGNED_16( int16_t mv[2][X264_SCAN8_LUMA_SIZE][2] );
  60.              ALIGNED_8( uint8_t mvd[2][X264_SCAN8_LUMA_SIZE][2] );
  61.  
  62. @@ -718,7 +734,11 @@ struct x264_t
  63.      uint32_t (*nr_residual_sum)[64];
  64.      uint32_t *nr_count;
  65.  
  66. +    ALIGNED_64( int misalign_pad_align_1 );
  67. +    int misalign_pad_1;
  68.      ALIGNED_32( udctcoef nr_offset_denoise[4][64] );
  69. +    ALIGNED_64( int misalign_pad_align_2 );
  70. +    int misalign_pad_2;
  71.      ALIGNED_32( uint32_t nr_residual_sum_buf[2][4][64] );
  72.      uint32_t nr_count_buf[2][4];
  73.  
  74. diff --git a/common/osdep.h b/common/osdep.h
  75. index fbc4801..31afba6 100644
  76. --- a/common/osdep.h
  77. +++ b/common/osdep.h
  78. @@ -123,17 +123,29 @@ int x264_is_pipe( const char *path );
  79.  // - Apple gcc only maintains 4 byte alignment
  80.  // - llvm can align the stack, but only in svn and (unrelated) it exposes bugs in all released GNU binutils...
  81.  
  82. +#if 0
  83.  #define ALIGNED_ARRAY_EMU( mask, type, name, sub1, ... )\
  84.      uint8_t name##_u [sizeof(type sub1 __VA_ARGS__) + mask]; \
  85.      type (*name) __VA_ARGS__ = (void*)((intptr_t)(name##_u+mask) & ~mask)
  86. +#else
  87. +#define ALIGNED_ARRAY_EMU( mask, type, name, sub1, ... )\
  88. +    uint8_t name##_u [sizeof(type sub1 __VA_ARGS__) + (mask*3+2)]; \
  89. +    type (*name) __VA_ARGS__ = (void*)(((intptr_t)(name##_u+(mask*2+1)) & ~(mask*2+1)) + (mask+1))
  90. +#define ALIGNED_ARRAY_EMU_ZERO( mask, attr, type, name, sub1, ... )\
  91. +    attr uint8_t name##_u [sizeof(type sub1 __VA_ARGS__) + (mask*3+2)] = {0}; \
  92. +    type (*name) __VA_ARGS__ = (void*)(((intptr_t)(name##_u+(mask*2+1)) & ~(mask*2+1)) + (mask+1))
  93. +#define ALIGNED_ZERO_16( ... ) EXPAND( ALIGNED_ARRAY_EMU_ZERO( 15, __VA_ARGS__ ) )
  94. +#define ALIGNED_ZERO_32( ... ) EXPAND( ALIGNED_ARRAY_EMU_ZERO( 31, __VA_ARGS__ ) )
  95. +#define ALIGNED_ZERO_64( ... ) EXPAND( ALIGNED_ARRAY_EMU_ZERO( 63, __VA_ARGS__ ) )
  96. +#endif
  97.  
  98. -#if ARCH_ARM && SYS_MACOSX
  99. +#if 1 || ARCH_ARM && SYS_MACOSX
  100.  #define ALIGNED_ARRAY_8( ... ) EXPAND( ALIGNED_ARRAY_EMU( 7, __VA_ARGS__ ) )
  101.  #else
  102.  #define ALIGNED_ARRAY_8( type, name, sub1, ... ) ALIGNED_8( type name sub1 __VA_ARGS__ )
  103.  #endif
  104.  
  105. -#if ARCH_ARM
  106. +#if 1 || ARCH_ARM
  107.  #define ALIGNED_ARRAY_16( ... ) EXPAND( ALIGNED_ARRAY_EMU( 15, __VA_ARGS__ ) )
  108.  #else
  109.  #define ALIGNED_ARRAY_16( type, name, sub1, ... ) ALIGNED_16( type name sub1 __VA_ARGS__ )
  110. @@ -145,12 +157,12 @@ int x264_is_pipe( const char *path );
  111.  #define NATIVE_ALIGN 64
  112.  #define ALIGNED_32( var ) DECLARE_ALIGNED( var, 32 )
  113.  #define ALIGNED_64( var ) DECLARE_ALIGNED( var, 64 )
  114. -#if STACK_ALIGNMENT >= 32
  115. +#if 0 && STACK_ALIGNMENT >= 32
  116.  #define ALIGNED_ARRAY_32( type, name, sub1, ... ) ALIGNED_32( type name sub1 __VA_ARGS__ )
  117.  #else
  118.  #define ALIGNED_ARRAY_32( ... ) EXPAND( ALIGNED_ARRAY_EMU( 31, __VA_ARGS__ ) )
  119.  #endif
  120. -#if STACK_ALIGNMENT >= 64
  121. +#if 0 && STACK_ALIGNMENT >= 64
  122.  #define ALIGNED_ARRAY_64( type, name, sub1, ... ) ALIGNED_64( type name sub1 __VA_ARGS__ )
  123.  #else
  124.  #define ALIGNED_ARRAY_64( ... ) EXPAND( ALIGNED_ARRAY_EMU( 63, __VA_ARGS__ ) )
  125. diff --git a/encoder/analyse.c b/encoder/analyse.c
  126. index a9d7dc2..bcb6911 100644
  127. --- a/encoder/analyse.c
  128. +++ b/encoder/analyse.c
  129. @@ -78,6 +78,8 @@ typedef struct
  130.  
  131.      int i_satd_i8x8;
  132.      int i_cbp_i8x8_luma;
  133. +    ALIGNED_64( int misalign_pad_align_1 );
  134. +    int misalign_pad_1;
  135.      ALIGNED_16( uint16_t i_satd_i8x8_dir[4][16] );
  136.      int i_predict8x8[4];
  137.  
  138. @@ -558,7 +560,7 @@ static ALWAYS_INLINE const int8_t *predict_4x4_mode_available( int force_intra,
  139.  /* For trellis=2, we need to do this for both sizes of DCT, for trellis=1 we only need to use it on the chosen mode. */
  140.  static void inline psy_trellis_init( x264_t *h, int do_both_dct )
  141.  {
  142. -    ALIGNED_16( static pixel zero[16*FDEC_STRIDE] ) = {0};
  143. +    ALIGNED_ZERO_16( static, pixel, zero,[16*FDEC_STRIDE] );
  144.  
  145.      if( do_both_dct || h->mb.b_transform_8x8 )
  146.          h->dctf.sub16x16_dct8( h->mb.pic.fenc_dct8, h->mb.pic.p_fenc[0], zero );
  147. diff --git a/encoder/me.c b/encoder/me.c
  148. index 9cb5a84..7babd73 100644
  149. --- a/encoder/me.c
  150. +++ b/encoder/me.c
  151. @@ -633,7 +633,7 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc,
  152.              /* successive elimination by comparing DC before a full SAD,
  153.               * because sum(abs(diff)) >= abs(diff(sum)). */
  154.              uint16_t *sums_base = m->integral;
  155. -            ALIGNED_16( static pixel zero[8*FENC_STRIDE] ) = {0};
  156. +            ALIGNED_ZERO_16( static, pixel, zero,[8*FENC_STRIDE] );
  157.              ALIGNED_ARRAY_16( int, enc_dc,[4] );
  158.              int sad_size = i_pixel <= PIXEL_8x8 ? PIXEL_8x8 : PIXEL_4x4;
  159.              int delta = x264_pixel_size[sad_size].w;
  160. diff --git a/encoder/rdo.c b/encoder/rdo.c
  161. index 68ba0a2..a83d719 100644
  162. --- a/encoder/rdo.c
  163. +++ b/encoder/rdo.c
  164. @@ -96,7 +96,7 @@ static ALWAYS_INLINE int cached_satd( x264_t *h, int size, int x, int y )
  165.      static const uint8_t satd_shift_x[3] = {3,   2,   2};
  166.      static const uint8_t satd_shift_y[3] = {2-1, 3-2, 2-2};
  167.      static const uint8_t  satd_offset[3] = {0,   8,   16};
  168. -    ALIGNED_16( static pixel zero[16] ) = {0};
  169. +    ALIGNED_ZERO_16( static, pixel, zero,[16] );
  170.      int cache_index = (x >> satd_shift_x[size - PIXEL_8x4]) + (y >> satd_shift_y[size - PIXEL_8x4])
  171.                      + satd_offset[size - PIXEL_8x4];
  172.      int res = h->mb.pic.fenc_satd_cache[cache_index];
  173. @@ -123,7 +123,7 @@ static ALWAYS_INLINE int cached_satd( x264_t *h, int size, int x, int y )
  174.  
  175.  static inline int ssd_plane( x264_t *h, int size, int p, int x, int y )
  176.  {
  177. -    ALIGNED_16( static pixel zero[16] ) = {0};
  178. +    ALIGNED_ZERO_16( static, pixel, zero,[16] );
  179.      int satd = 0;
  180.      pixel *fdec = h->mb.pic.p_fdec[p] + x + y*FDEC_STRIDE;
  181.      pixel *fenc = h->mb.pic.p_fenc[p] + x + y*FENC_STRIDE;
  182. @@ -912,8 +912,8 @@ int quant_trellis_cavlc( x264_t *h, dctcoef *dct,
  183.                           const uint8_t *zigzag, int ctx_block_cat, int lambda2, int b_ac,
  184.                           int b_chroma, int dc, int num_coefs, int idx, int b_8x8 )
  185.  {
  186. -    ALIGNED_16( dctcoef quant_coefs[2][16] );
  187. -    ALIGNED_16( dctcoef coefs[16] ) = {0};
  188. +    ALIGNED_ARRAY_16( dctcoef, quant_coefs,[2],[16] );
  189. +    ALIGNED_ZERO_16( ,dctcoef, coefs,[16] );
  190.      const uint32_t *coef_weight1 = b_8x8 ? x264_dct8_weight_tab : x264_dct4_weight_tab;
  191.      const uint32_t *coef_weight2 = b_8x8 ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
  192.      int delta_distortion[16];
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!