Optimize predict_8x8_filter and incidentally remove a valgrind false-positive

diff --git a/common/arm/predict-c.c b/common/arm/predict-c.c
index e80b8f09852eddf1edde1ace6c51ac7742151afe..bccdc504b010ec880602c0bdd5654d12a1e53074 100644 (file)
--- a/common/arm/predict-c.c
+++ b/common/arm/predict-c.c
@@ -39,8 +39,8 @@ void x264_predict_8x8c_h_neon( uint8_t *src );
 void x264_predict_8x8c_v_neon( uint8_t *src );
 void x264_predict_8x8c_p_neon( uint8_t *src );
 
-void x264_predict_8x8_dc_neon( uint8_t *src, uint8_t edge[33] );
-void x264_predict_8x8_h_neon( uint8_t *src, uint8_t edge[33] );
+void x264_predict_8x8_dc_neon( uint8_t *src, uint8_t edge[36] );
+void x264_predict_8x8_h_neon( uint8_t *src, uint8_t edge[36] );
 
 void x264_predict_16x16_dc_neon( uint8_t *src );
 void x264_predict_16x16_dc_top_neon( uint8_t *src );

diff --git a/common/macroblock.c b/common/macroblock.c
index 9eb3b59562abdb86e98b5959551481095a552615..7c524ff011fd8f57e621dd5b479b26a515fa3c75 100644 (file)
--- a/common/macroblock.c
+++ b/common/macroblock.c
@@ -350,8 +350,7 @@ int x264_macroblock_thread_allocate( x264_t *h, int b_lookahead )
         for( int i = 0; i <= 4*PARAM_INTERLACED; i++ )
             for( int j = 0; j < (CHROMA444 ? 3 : 2); j++ )
             {
-                /* shouldn't really be initialized, just silences a valgrind false-positive in predict_8x8_filter_mmx */
-                CHECKED_MALLOCZERO( h->intra_border_backup[i][j], (h->sps->i_mb_width*16+32) * sizeof(pixel) );
+                CHECKED_MALLOC( h->intra_border_backup[i][j], (h->sps->i_mb_width*16+32) * sizeof(pixel) );
                 h->intra_border_backup[i][j] += 16;
                 if( !PARAM_INTERLACED )
                     h->intra_border_backup[1][j] = h->intra_border_backup[i][j];

diff --git a/common/pixel.c b/common/pixel.c
index bc8bf975cea88e86d65fba79c93c3d2ffceb046d..7418121a193c66d431532ee9fb900a272dfa1277 100644 (file)
--- a/common/pixel.c
+++ b/common/pixel.c
@@ -499,7 +499,7 @@ SATD_X_DECL7( _neon )
 #endif // !HIGH_BIT_DEPTH
 
 #define INTRA_MBCMP_8x8( mbcmp, cpu )\
-void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[33], int res[3] )\
+void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[36], int res[3] )\
 {\
     ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
     x264_predict_8x8_v_c( pix, edge );\

diff --git a/common/pixel.h b/common/pixel.h
index 862b1424a2b596c830b0f43986dd9b1b622d3c69..6d2fc65cb97c008504f86d941b63d07783721b7f 100644 (file)
--- a/common/pixel.h
+++ b/common/pixel.h
@@ -117,9 +117,9 @@ typedef struct
     void (*intra_mbcmp_x3_4x4)  ( pixel *fenc, pixel *fdec  , int res[3] );
     void (*intra_satd_x3_4x4)   ( pixel *fenc, pixel *fdec  , int res[3] );
     void (*intra_sad_x3_4x4)    ( pixel *fenc, pixel *fdec  , int res[3] );
-    void (*intra_mbcmp_x3_8x8)  ( pixel *fenc, pixel edge[33], int res[3] );
-    void (*intra_sa8d_x3_8x8)   ( pixel *fenc, pixel edge[33], int res[3] );
-    void (*intra_sad_x3_8x8)    ( pixel *fenc, pixel edge[33], int res[3] );
+    void (*intra_mbcmp_x3_8x8)  ( pixel *fenc, pixel edge[36], int res[3] );
+    void (*intra_sa8d_x3_8x8)   ( pixel *fenc, pixel edge[36], int res[3] );
+    void (*intra_sad_x3_8x8)    ( pixel *fenc, pixel edge[36], int res[3] );
 } x264_pixel_function_t;
 
 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf );

diff --git a/common/ppc/pixel.c b/common/ppc/pixel.c
index faed77770f9c273fefcf3921ca987bcb5f007c28..c708aa4c3b4933f58ab16be804b2877899cdc74d 100644 (file)
--- a/common/ppc/pixel.c
+++ b/common/ppc/pixel.c
@@ -2009,7 +2009,7 @@ SATD_X( 4x4 )
 
 
 #define INTRA_MBCMP_8x8( mbcmp )\
-void intra_##mbcmp##_x3_8x8_altivec( uint8_t *fenc, uint8_t edge[33], int res[3] )\
+void intra_##mbcmp##_x3_8x8_altivec( uint8_t *fenc, uint8_t edge[36], int res[3] )\
 {\
     ALIGNED_8( uint8_t pix[8*FDEC_STRIDE] );\
     x264_predict_8x8_v_c( pix, edge );\

diff --git a/common/predict.c b/common/predict.c
index cdfb49cc3129dd2c8849e68554aa6ecdddaa85f3..34798c2f01d6b391edc8019e9ea0079f4d519593 100644 (file)
--- a/common/predict.c
+++ b/common/predict.c
@@ -458,7 +458,7 @@ static void x264_predict_4x4_hu_c( pixel *src )
 #define PT(x) \
     edge[16+x] = F2(SRC(x-1,-1), SRC(x,-1), SRC(x+1,-1));
 
-static void x264_predict_8x8_filter_c( pixel *src, pixel edge[33], int i_neighbor, int i_filters )
+static void x264_predict_8x8_filter_c( pixel *src, pixel edge[36], int i_neighbor, int i_filters )
 {
     /* edge[7..14] = l7..l0
      * edge[15] = lt
@@ -525,30 +525,30 @@ static void x264_predict_8x8_filter_c( pixel *src, pixel edge[33], int i_neighbo
         src += FDEC_STRIDE; \
     }
 
-static void x264_predict_8x8_dc_128_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_dc_128_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_DC( PIXEL_SPLAT_X4( 1 << (BIT_DEPTH-1) ) );
 }
-static void x264_predict_8x8_dc_left_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_dc_left_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_LEFT
     pixel4 dc = PIXEL_SPLAT_X4( (l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3 );
     PREDICT_8x8_DC( dc );
 }
-static void x264_predict_8x8_dc_top_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_dc_top_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_TOP
     pixel4 dc = PIXEL_SPLAT_X4( (t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3 );
     PREDICT_8x8_DC( dc );
 }
-void x264_predict_8x8_dc_c( pixel *src, pixel edge[33] )
+void x264_predict_8x8_dc_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_LEFT
     PREDICT_8x8_LOAD_TOP
     pixel4 dc = PIXEL_SPLAT_X4( (l0+l1+l2+l3+l4+l5+l6+l7+t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4 );
     PREDICT_8x8_DC( dc );
 }
-void x264_predict_8x8_h_c( pixel *src, pixel edge[33] )
+void x264_predict_8x8_h_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_LEFT
 #define ROW(y) MPIXEL_X4( src+y*FDEC_STRIDE+0 ) =\
@@ -556,7 +556,7 @@ void x264_predict_8x8_h_c( pixel *src, pixel edge[33] )
     ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
 #undef ROW
 }
-void x264_predict_8x8_v_c( pixel *src, pixel edge[33] )
+void x264_predict_8x8_v_c( pixel *src, pixel edge[36] )
 {
     pixel4 top[2] = { MPIXEL_X4( edge+16 ),
                       MPIXEL_X4( edge+20 ) };
@@ -566,7 +566,7 @@ void x264_predict_8x8_v_c( pixel *src, pixel edge[33] )
         MPIXEL_X4( src+y*FDEC_STRIDE+4 ) = top[1];
     }
 }
-static void x264_predict_8x8_ddl_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_ddl_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_TOP
     PREDICT_8x8_LOAD_TOPRIGHT
@@ -586,7 +586,7 @@ static void x264_predict_8x8_ddl_c( pixel *src, pixel edge[33] )
     SRC(6,7)=SRC(7,6)= F2(t13,t14,t15);
     SRC(7,7)= F2(t14,t15,t15);
 }
-static void x264_predict_8x8_ddr_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_ddr_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_TOP
     PREDICT_8x8_LOAD_LEFT
@@ -608,7 +608,7 @@ static void x264_predict_8x8_ddr_c( pixel *src, pixel edge[33] )
     SRC(7,0)= F2(t5,t6,t7);
 
 }
-static void x264_predict_8x8_vr_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_vr_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_TOP
     PREDICT_8x8_LOAD_LEFT
@@ -636,7 +636,7 @@ static void x264_predict_8x8_vr_c( pixel *src, pixel edge[33] )
     SRC(7,1)= F2(t5,t6,t7);
     SRC(7,0)= F1(t6,t7);
 }
-static void x264_predict_8x8_hd_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_hd_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_TOP
     PREDICT_8x8_LOAD_LEFT
@@ -663,7 +663,7 @@ static void x264_predict_8x8_hd_c( pixel *src, pixel edge[33] )
     SRC_X4(4,1)= pack_pixel_2to4(p9,p10);
     SRC_X4(4,0)= pack_pixel_2to4(p10,p11);
 }
-static void x264_predict_8x8_vl_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_vl_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_TOP
     PREDICT_8x8_LOAD_TOPRIGHT
@@ -690,7 +690,7 @@ static void x264_predict_8x8_vl_c( pixel *src, pixel edge[33] )
     SRC(7,6)= F1(t10,t11);
     SRC(7,7)= F2(t10,t11,t12);
 }
-static void x264_predict_8x8_hu_c( pixel *src, pixel edge[33] )
+static void x264_predict_8x8_hu_c( pixel *src, pixel edge[36] )
 {
     PREDICT_8x8_LOAD_LEFT
     int p1 = pack_pixel_1to2(F1(l0,l1), F2(l0,l1,l2));

diff --git a/common/predict.h b/common/predict.h
index aa7673f0802007551e21dd1098b66515dfc50bfd..23330f51744746c8d86bd9fdf4411f36100e30e5 100644 (file)
--- a/common/predict.h
+++ b/common/predict.h
@@ -28,8 +28,8 @@
 #define X264_PREDICT_H
 
 typedef void (*x264_predict_t)( pixel *src );
-typedef void (*x264_predict8x8_t)( pixel *src, pixel edge[33] );
-typedef void (*x264_predict_8x8_filter_t) ( pixel *src, pixel edge[33], int i_neighbor, int i_filters );
+typedef void (*x264_predict8x8_t)( pixel *src, pixel edge[36] );
+typedef void (*x264_predict_8x8_filter_t) ( pixel *src, pixel edge[36], int i_neighbor, int i_filters );
 
 enum intra_chroma_pred_e
 {
@@ -109,9 +109,9 @@ enum intra8x8_pred_e
     I_PRED_8x8_DC_128  = 11,
 };
 
-void x264_predict_8x8_dc_c  ( pixel *src, pixel edge[33] );
-void x264_predict_8x8_h_c   ( pixel *src, pixel edge[33] );
-void x264_predict_8x8_v_c   ( pixel *src, pixel edge[33] );
+void x264_predict_8x8_dc_c  ( pixel *src, pixel edge[36] );
+void x264_predict_8x8_h_c   ( pixel *src, pixel edge[36] );
+void x264_predict_8x8_v_c   ( pixel *src, pixel edge[36] );
 void x264_predict_4x4_dc_c  ( pixel *src );
 void x264_predict_4x4_h_c   ( pixel *src );
 void x264_predict_4x4_v_c   ( pixel *src );

diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
index 9fe26e9fdce5b04cdbe1bf6b8f585cb1960c2ff6..0bfc2db207a02facf60db21a48b82c831aec4776 100644 (file)
--- a/common/x86/predict-a.asm
+++ b/common/x86/predict-a.asm
@@ -37,6 +37,8 @@ pw_m3:       times 8 dw -3
 pb_00s_ff:   times 8 db 0
 pb_0s_ff:    times 7 db 0
              db 0xff
+shuf_fixtr:  db 0, 1, 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7
+shuf_nop:    db 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
 
 SECTION .text
 
@@ -596,9 +598,9 @@ cglobal predict_4x4_dc_mmx2, 1,4
 
 %macro PREDICT_FILTER 5
 ;-----------------------------------------------------------------------------
-;void predict_8x8_filter( pixel *src, pixel edge[33], int i_neighbor, int i_filters )
+;void predict_8x8_filter( pixel *src, pixel edge[36], int i_neighbor, int i_filters )
 ;-----------------------------------------------------------------------------
-cglobal predict_8x8_filter, 4,5,7
+cglobal predict_8x8_filter, 4,6,6
     add          r0, 0x58*SIZEOF_PIXEL
 %define src r0-0x58*SIZEOF_PIXEL
 %ifndef ARCH_X86_64
@@ -609,10 +611,13 @@ cglobal predict_8x8_filter, 4,5,7
 %define t1 r1
 %define t4 r4
 %endif
-    test       r3b, 0x01
+    test       r3b, 1
     je .check_top
+    mov        t4d, r2d
+    and        t4d, 8
+    neg         t4
     mova        m0, [src+0*FDEC_STRIDEB-8*SIZEOF_PIXEL]
-    punpckh%1%2 m0, [src-1*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+    punpckh%1%2 m0, [src+0*FDEC_STRIDEB-8*SIZEOF_PIXEL+t4*(FDEC_STRIDEB/8)]
     mova        m1, [src+2*FDEC_STRIDEB-8*SIZEOF_PIXEL]
     punpckh%1%2 m1, [src+1*FDEC_STRIDEB-8*SIZEOF_PIXEL]
     punpckh%2%3 m1, m0
@@ -628,74 +633,79 @@ cglobal predict_8x8_filter, 4,5,7
     mova        m2, m3
     PALIGNR     m4, m0, 7*SIZEOF_PIXEL, m0
     PALIGNR     m1, m2, 1*SIZEOF_PIXEL, m2
-    test       r2b, 0x08
-    je .fix_lt_1
-.do_left:
-    mova        m0, m4
     PRED8x8_LOWPASS %1, m2, m1, m4, m3, m5
     mova        [t1+8*SIZEOF_PIXEL], m2
-    mova        m4, m0
-    PRED8x8_LOWPASS %1, m1, m3, m0, m4, m5
-    movd        t4, m1
+    movzx      t4d, pixel [src+7*FDEC_STRIDEB-1*SIZEOF_PIXEL]
+    movzx      r5d, pixel [src+6*FDEC_STRIDEB-1*SIZEOF_PIXEL]
+    lea        t4d, [t4*3+2]
+    add        t4d, r5d
+    shr        t4d, 2
     mov         [t1+7*SIZEOF_PIXEL], t4%1
-.check_top:
-    test        r3b, 0x02
+    test       r3b, 2
     je .done
+.check_top:
+%if SIZEOF_PIXEL==1 && cpuflag(ssse3)
+INIT_XMM cpuname
+    movu        m3, [src-1*FDEC_STRIDEB]
+    movhps      m0, [src-1*FDEC_STRIDEB-8]
+    test       r2b, 8
+    je .fix_lt_2
+.do_top:
+    and        r2d, 4
+%ifdef PIC
+    lea         r3, [shuf_fixtr]
+    pshufb      m3, [r3+r2*4]
+%else
+    pshufb      m3, [shuf_fixtr+r2*4] ; neighbor&MB_TOPRIGHT ? shuf_nop : shuf_fixtr
+%endif
+    psrldq      m1, m3, 15
+    PALIGNR     m2, m3, m0, 15, m0
+    PALIGNR     m1, m3, 1, m5
+    PRED8x8_LOWPASS %1, m0, m2, m1, m3, m5
+    mova        [t1+16*SIZEOF_PIXEL], m0
+    psrldq      m0, 15
+    movd        [t1+32*SIZEOF_PIXEL], m0
+.done:
+    REP_RET
+.fix_lt_2:
+    pslldq      m0, m3, 15
+    jmp .do_top
+
+%else
     mova        m0, [src-1*FDEC_STRIDEB-8*SIZEOF_PIXEL]
     mova        m3, [src-1*FDEC_STRIDEB]
     mova        m1, [src-1*FDEC_STRIDEB+8*SIZEOF_PIXEL]
-    mova        m2, m3
-    mova        m4, m3
-    PALIGNR     m2, m0, 7*SIZEOF_PIXEL, m0
-    PALIGNR     m1, m4, 1*SIZEOF_PIXEL, m4
-    test        r2b, 0x08
+    test       r2b, 8
     je .fix_lt_2
-    test        r2b, 0x04
+    test       r2b, 4
     je .fix_tr_1
 .do_top:
-    PRED8x8_LOWPASS %1, m4, m2, m1, m3, m5
+    PALIGNR     m2, m3, m0, 7*SIZEOF_PIXEL, m0
+    PALIGNR     m0, m1, m3, 1*SIZEOF_PIXEL, m5
+    PRED8x8_LOWPASS %1, m4, m2, m0, m3, m5
     mova        [t1+16*SIZEOF_PIXEL], m4
-    test        r3b, 0x04
+    test       r3b, 4
     je .done
-    test        r2b, 0x04
-    je .fix_tr_2
-    mova        m0, [src-1*FDEC_STRIDEB+8*SIZEOF_PIXEL]
-    mova        m2, m0
-    mova        m4, m0
-    psrl%4      m5, m0, 7*%5
+    mova        m2, m1
+    mova        m4, m1
+    psrl%4      m5, m1, 7*%5
     PALIGNR     m2, m3, 7*SIZEOF_PIXEL, m3
     PALIGNR     m5, m4, 1*SIZEOF_PIXEL, m4
-    PRED8x8_LOWPASS %1, m1, m2, m5, m0, m4
-    jmp .do_topright
-.fix_tr_2:
-    punpckh%1%2 m3, m3
-    pshuf%2     m1, m3, q3333
-.do_topright:
-    mova        [t1+24*SIZEOF_PIXEL], m1
-    psrl%4      m1, 7*%5
-    movd        t4, m1
-    mov         [t1+32*SIZEOF_PIXEL], t4%1
+    PRED8x8_LOWPASS %1, m0, m2, m5, m1, m4
+    mova        [t1+24*SIZEOF_PIXEL], m0
+    psrl%4      m0, 7*%5
+    movd        [t1+32*SIZEOF_PIXEL], m0
 .done:
     REP_RET
-.fix_lt_1:
-    pxor        m5, m3, m4
-    psrl%4      m5, 7*%5
-    psll%4      m5, 6*%5
-    pxor        m1, m5
-    jmp .do_left
 .fix_lt_2:
-    pxor        m5, m3, m2
-    psll%4      m5, 7*%5
-    psrl%4      m5, 7*%5
-    pxor        m2, m5
-    test       r2b, 0x04
+    psll%4      m0, m3, 7*%5
+    test       r2b, 4
     jne .do_top
 .fix_tr_1:
-    pxor        m5, m3, m1
-    psrl%4      m5, 7*%5
-    psll%4      m5, 7*%5
-    pxor        m1, m5
+    punpckh%1%2 m1, m3, m3
+    pshuf%2     m1, m1, q3333
     jmp .do_top
+%endif
 %endmacro
 
 %ifdef HIGH_BIT_DEPTH
@@ -731,7 +741,7 @@ PREDICT_8x8_V
 %endif
 
 ;-----------------------------------------------------------------------------
-; void predict_8x8_h( pixel *src, pixel edge[33] )
+; void predict_8x8_h( pixel *src, pixel edge[36] )
 ;-----------------------------------------------------------------------------
 %macro PREDICT_8x8_H 2
 cglobal predict_8x8_h, 2,2

diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
index abc5cdf4c285c29045b3f2b212aa2b26b748b74b..e37511cb8d93fe957cb08e5fa95b24b672d64da1 100644 (file)
--- a/common/x86/predict-c.c
+++ b/common/x86/predict-c.c
@@ -55,40 +55,40 @@
  void x264_predict_8x8c_h_mmx2( uint8_t *src );
  void x264_predict_8x8c_h_sse2( pixel *src );
  void x264_predict_8x8c_h_ssse3( uint8_t *src );
- void x264_predict_8x8_v_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_v_sse2( uint16_t *src, uint16_t edge[33] );
- void x264_predict_8x8_h_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_h_sse2( uint16_t *src, uint16_t edge[33] );
- void x264_predict_8x8_hd_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_hu_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_dc_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_dc_sse2( uint16_t *src, uint16_t edge[33] );
- void x264_predict_8x8_dc_top_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_dc_top_sse2( uint16_t *src, uint16_t edge[33] );
- void x264_predict_8x8_dc_left_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_dc_left_sse2( uint16_t *src, uint16_t edge[33] );
- void x264_predict_8x8_ddl_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_ddl_sse2( pixel *src, pixel edge[33] );
- void x264_predict_8x8_ddl_avx( pixel *src, pixel edge[33] );
- void x264_predict_8x8_ddr_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_ddr_sse2( pixel *src, pixel edge[33] );
- void x264_predict_8x8_ddr_avx( pixel *src, pixel edge[33] );
- void x264_predict_8x8_vl_sse2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_vl_avx( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_vr_mmx2( uint8_t *src, uint8_t edge[33] );
- void x264_predict_8x8_vr_sse2( pixel *src, pixel edge[33] );
- void x264_predict_8x8_vr_ssse3( uint16_t *src, uint16_t edge[33] );
- void x264_predict_8x8_vr_avx( pixel *src, pixel edge[33] );
- void x264_predict_8x8_hu_sse2( pixel *src, pixel edge[33] );
- void x264_predict_8x8_hu_ssse3( pixel *src, pixel edge[33] );
- void x264_predict_8x8_hu_avx( pixel *src, pixel edge[33] );
- void x264_predict_8x8_hd_sse2( pixel *src, pixel edge[33] );
- void x264_predict_8x8_hd_ssse3( pixel *src, pixel edge[33] );
- void x264_predict_8x8_hd_avx( pixel *src, pixel edge[33] );
- void x264_predict_8x8_filter_mmx2( uint8_t *src, uint8_t edge[33], int i_neighbor, int i_filters );
- void x264_predict_8x8_filter_sse2( uint16_t *src, uint16_t edge[33], int i_neighbor, int i_filters );
- void x264_predict_8x8_filter_ssse3( pixel *src, pixel edge[33], int i_neighbor, int i_filters );
- void x264_predict_8x8_filter_avx( uint16_t *src, uint16_t edge[33], int i_neighbor, int i_filters );
+ void x264_predict_8x8_v_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_v_sse2( uint16_t *src, uint16_t edge[36] );
+ void x264_predict_8x8_h_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_h_sse2( uint16_t *src, uint16_t edge[36] );
+ void x264_predict_8x8_hd_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_hu_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_dc_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_dc_sse2( uint16_t *src, uint16_t edge[36] );
+ void x264_predict_8x8_dc_top_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_dc_top_sse2( uint16_t *src, uint16_t edge[36] );
+ void x264_predict_8x8_dc_left_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_dc_left_sse2( uint16_t *src, uint16_t edge[36] );
+ void x264_predict_8x8_ddl_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_ddl_sse2( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_ddl_avx( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_ddr_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_ddr_sse2( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_ddr_avx( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_vl_sse2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_vl_avx( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_vr_mmx2( uint8_t *src, uint8_t edge[36] );
+ void x264_predict_8x8_vr_sse2( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_vr_ssse3( uint16_t *src, uint16_t edge[36] );
+ void x264_predict_8x8_vr_avx( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_hu_sse2( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_hu_ssse3( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_hu_avx( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_hd_sse2( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_hd_ssse3( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_hd_avx( pixel *src, pixel edge[36] );
+ void x264_predict_8x8_filter_mmx2( uint8_t *src, uint8_t edge[36], int i_neighbor, int i_filters );
+ void x264_predict_8x8_filter_sse2( uint16_t *src, uint16_t edge[36], int i_neighbor, int i_filters );
+ void x264_predict_8x8_filter_ssse3( pixel *src, pixel edge[36], int i_neighbor, int i_filters );
+ void x264_predict_8x8_filter_avx( uint16_t *src, uint16_t edge[36], int i_neighbor, int i_filters );
  void x264_predict_4x4_ddl_mmx2( pixel *src );
  void x264_predict_4x4_ddl_sse2( uint16_t *src );
  void x264_predict_4x4_ddl_avx( uint16_t *src );
@@ -368,7 +368,7 @@ static void x264_predict_8x8c_dc_left( uint8_t *src )
     t=g; g+=h; h-=t;
 
 #define INTRA_SA8D_X3(cpu)\
-void x264_intra_sa8d_x3_8x8_##cpu( uint8_t *fenc, uint8_t edge[33], int res[3] )\
+void x264_intra_sa8d_x3_8x8_##cpu( uint8_t *fenc, uint8_t edge[36], int res[3] )\
 {\
     PREDICT_8x8_LOAD_TOP\
     PREDICT_8x8_LOAD_LEFT\

diff --git a/common/x86/sad-a.asm b/common/x86/sad-a.asm
index 36efa979252dcb31c43098e3058ae120fc6a37a8..743afcda460b1ff8e1ba5c64d6a8e4f22f5daedf 100644 (file)
--- a/common/x86/sad-a.asm
+++ b/common/x86/sad-a.asm
@@ -424,7 +424,7 @@ INIT_XMM avx
 INTRA_SADx3_4x4
 
 ;-----------------------------------------------------------------------------
-; void intra_sad_x3_8x8( uint8_t *fenc, uint8_t edge[33], int res[3]);
+; void intra_sad_x3_8x8( uint8_t *fenc, uint8_t edge[36], int res[3]);
 ;-----------------------------------------------------------------------------
 
 ;m0 = DC

diff --git a/encoder/analyse.c b/encoder/analyse.c
index ea9c7174b549544b770a1e81f9cd2a54fcbcb1ee..2352c3e20a69000a4f7f1f5c401bb75ff6bde8a5 100644 (file)
--- a/encoder/analyse.c
+++ b/encoder/analyse.c
@@ -834,7 +834,7 @@ static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_satd_
     /* 8x8 prediction selection */
     if( flags & X264_ANALYSE_I8x8 )
     {
-        ALIGNED_ARRAY_16( pixel, edge,[33] );
+        ALIGNED_ARRAY_16( pixel, edge,[36] );
         x264_pixel_cmp_t sa8d = (h->pixf.mbcmp[0] == h->pixf.satd[0]) ? h->pixf.sa8d[PIXEL_8x8] : h->pixf.mbcmp[PIXEL_8x8];
         int i_satd_thresh = a->i_mbrd ? COST_MAX : X264_MIN( i_satd_inter, a->i_satd_i16x16 );
 

diff --git a/encoder/macroblock.c b/encoder/macroblock.c
index 261e39bfdd3d1be7cd431279f83dbfa7c92eef8e..adee51de8f49d123dabd37f3c0ec35c12ec3471e 100644 (file)
--- a/encoder/macroblock.c
+++ b/encoder/macroblock.c
@@ -171,7 +171,7 @@ void x264_mb_encode_i8x8( x264_t *h, int p, int idx, int i_qp, int i_mode, pixel
     pixel *p_src = &h->mb.pic.p_fenc[p][8*x + 8*y*FENC_STRIDE];
     pixel *p_dst = &h->mb.pic.p_fdec[p][8*x + 8*y*FDEC_STRIDE];
     ALIGNED_ARRAY_16( dctcoef, dct8x8,[64] );
-    ALIGNED_ARRAY_16( pixel, edge_buf,[33] );
+    ALIGNED_ARRAY_16( pixel, edge_buf,[36] );
 
     if( !edge )
     {
@@ -519,7 +519,7 @@ void x264_predict_lossless_4x4( x264_t *h, pixel *p_dst, int p, int idx, int i_m
         h->predict_4x4[i_mode]( p_dst );
 }
 
-void x264_predict_lossless_8x8( x264_t *h, pixel *p_dst, int p, int idx, int i_mode, pixel edge[33] )
+void x264_predict_lossless_8x8( x264_t *h, pixel *p_dst, int p, int idx, int i_mode, pixel edge[36] )
 {
     int stride = h->fenc->i_stride[p] << MB_INTERLACED;
     pixel *p_src = h->mb.pic.p_fenc_plane[p] + (idx&1)*8 + (idx>>1)*8*stride;

diff --git a/encoder/macroblock.h b/encoder/macroblock.h
index 3da0e0d2d42ec333cb15b6bd5cd5ad5bacc5bff2..5e3b188d09dfd07a419175672b767bd854d98b2d 100644 (file)
--- a/encoder/macroblock.h
+++ b/encoder/macroblock.h
@@ -43,7 +43,7 @@ int x264_macroblock_probe_skip( x264_t *h, int b_bidir );
 
 void x264_predict_lossless_8x8_chroma( x264_t *h, int i_mode );
 void x264_predict_lossless_4x4( x264_t *h, pixel *p_dst, int p, int idx, int i_mode );
-void x264_predict_lossless_8x8( x264_t *h, pixel *p_dst, int p, int idx, int i_mode, pixel edge[33] );
+void x264_predict_lossless_8x8( x264_t *h, pixel *p_dst, int p, int idx, int i_mode, pixel edge[36] );
 void x264_predict_lossless_16x16( x264_t *h, int p, int i_mode );
 
 void x264_macroblock_encode      ( x264_t *h );

diff --git a/encoder/slicetype.c b/encoder/slicetype.c
index 240c7fd9efbe953a71d32134f00b8c346bbbea3f..5a91c167f33d27194631df9fb20fc72edc21937e 100644 (file)
--- a/encoder/slicetype.c
+++ b/encoder/slicetype.c
@@ -622,7 +622,7 @@ skip_motionest:
 lowres_intra_mb:
     if( !fenc->b_intra_calculated )
     {
-        ALIGNED_ARRAY_16( pixel, edge,[33] );
+        ALIGNED_ARRAY_16( pixel, edge,[36] );
         pixel *pix = &pix1[8+FDEC_STRIDE - 1];
         pixel *src = &fenc->lowres[0][i_pel_offset - 1];
         const int intra_penalty = 5 * a->i_lambda;

diff --git a/tools/checkasm.c b/tools/checkasm.c
index 8ec773c2064e014640d524ad6f021a35cf5d04be..03d7f054a45849467b901aab4e1cb2e65e029b57 100644 (file)
--- a/tools/checkasm.c
+++ b/tools/checkasm.c
@@ -244,7 +244,7 @@ static int check_pixel( int cpu_ref, int cpu_new )
     x264_pixel_function_t pixel_asm;
     x264_predict8x8_t predict_8x8[9+3];
     x264_predict_8x8_filter_t predict_8x8_filter;
-    ALIGNED_16( pixel edge[33] );
+    ALIGNED_16( pixel edge[36] );
     uint16_t cost_mv[32];
     int ret = 0, ok, used_asm;
 
@@ -1784,8 +1784,8 @@ static int check_quant( int cpu_ref, int cpu_new )
 static int check_intra( int cpu_ref, int cpu_new )
 {
     int ret = 0, ok = 1, used_asm = 0;
-    ALIGNED_16( pixel edge[33] );
-    ALIGNED_16( pixel edge2[33] );
+    ALIGNED_16( pixel edge[36] );
+    ALIGNED_16( pixel edge2[36] );
     ALIGNED_16( pixel fdec[FDEC_STRIDE*20] );
     struct
     {
@@ -1864,10 +1864,16 @@ static int check_intra( int cpu_ref, int cpu_new )
         used_asm = 1;
         for( int i = 0; i < 32; i++ )
         {
-            memcpy( edge2, edge, 33 * sizeof(pixel) );
-            call_c(ip_c.predict_8x8_filter, pbuf1+48, edge, (i&24)>>1, i&7);
-            call_a(ip_a.predict_8x8_filter, pbuf1+48, edge2, (i&24)>>1, i&7);
-            if( memcmp( edge, edge2, 33 * sizeof(pixel) ) )
+            if( !(i&7) || ((i&MB_TOPRIGHT) && !(i&MB_TOP)) )
+                continue;
+            int neighbor = (i&24)>>1;
+            memset( edge,  0, sizeof(edge) );
+            memset( edge2, 0, sizeof(edge2) );
+            call_c( ip_c.predict_8x8_filter, pbuf1+48, edge,  neighbor, i&7 );
+            call_a( ip_a.predict_8x8_filter, pbuf1+48, edge2, neighbor, i&7 );
+            if( !(neighbor&MB_TOPLEFT) )
+                edge[15] = edge2[15] = 0;
+            if( memcmp( edge+7, edge2+7, (i&MB_TOPRIGHT ? 26 : i&MB_TOP ? 17 : 8) * sizeof(pixel) ) )
             {
                 fprintf( stderr, "predict_8x8_filter :  [FAILED] %d %d\n", (i&24)>>1, i&7);
                 ok = 0;