intra_sad_x3_8x8 assembly

diff --git a/common/pixel.c b/common/pixel.c
index 72b5e6fd3002dcedc171b93e314d92df58c963ec..5932f07f6c16f1aa19a5fa3cff702ea56a5b5be8 100644 (file)
--- a/common/pixel.c
+++ b/common/pixel.c
@@ -665,6 +665,7 @@ void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmxext;
         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmxext;
+        pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmxext;
         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmxext;
     }

diff --git a/common/pixel.h b/common/pixel.h
index 9ce368be4eceb2796b72cd8f79ab70b79dad12d9..207c74f2105842de53eeaef981e5b70a0aa1c7a3 100644 (file)
--- a/common/pixel.h
+++ b/common/pixel.h
@@ -103,7 +103,9 @@ typedef struct
     void (*intra_mbcmp_x3_4x4)  ( uint8_t *fenc, uint8_t *fdec  , int res[3] );
     void (*intra_satd_x3_4x4)   ( uint8_t *fenc, uint8_t *fdec  , int res[3] );
     void (*intra_sad_x3_4x4)    ( uint8_t *fenc, uint8_t *fdec  , int res[3] );
+    void (*intra_mbcmp_x3_8x8)  ( uint8_t *fenc, uint8_t edge[33], int res[3] );
     void (*intra_sa8d_x3_8x8)   ( uint8_t *fenc, uint8_t edge[33], int res[3] );
+    void (*intra_sad_x3_8x8)    ( uint8_t *fenc, uint8_t edge[33], int res[3] );
 } x264_pixel_function_t;
 
 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf );

diff --git a/common/x86/pixel.h b/common/x86/pixel.h
index dbc04fe2fb16c20d0380d4fec85573aacc939043..312aca8f54b7d2e3bbe7c0c5e0a7c8b316815a3d 100644 (file)
--- a/common/x86/pixel.h
+++ b/common/x86/pixel.h
@@ -92,6 +92,7 @@ void x264_intra_sad_x3_16x16_ssse3  ( uint8_t *, uint8_t *, int * );
 void x264_intra_sa8d_x3_8x8_mmxext  ( uint8_t *, uint8_t *, int * );
 void x264_intra_sa8d_x3_8x8_sse2    ( uint8_t *, uint8_t *, int * );
 void x264_intra_sa8d_x3_8x8_ssse3   ( uint8_t *, uint8_t *, int * );
+void x264_intra_sad_x3_8x8_mmxext   ( uint8_t *, uint8_t *, int * );
 void x264_intra_sa8d_x3_8x8_core_mmxext( uint8_t *, int16_t [2][8], int * );
 void x264_intra_sa8d_x3_8x8_core_sse2  ( uint8_t *, int16_t [2][8], int * );
 void x264_intra_sa8d_x3_8x8_core_ssse3 ( uint8_t *, int16_t [2][8], int * );

diff --git a/common/x86/sad-a.asm b/common/x86/sad-a.asm
index ba1cbe4829ebfa83c471438513dda12351782e8b..843065c3c663b79f6e41c9b12513657594f73105 100644 (file)
--- a/common/x86/sad-a.asm
+++ b/common/x86/sad-a.asm
@@ -28,6 +28,7 @@
 
 SECTION_RODATA
 pb_3: times 16 db 3
+pw_8: times 4 dw 8
 pb_shuf8x8c0: db 0,0,0,0,2,2,2,2
 pb_shuf8x8c1: db 4,4,4,4,6,6,6,6
 sw_64: dd 64
@@ -304,6 +305,73 @@ cglobal x264_intra_sad_x3_4x4_mmxext, 3,3
     movd   [r2+4], mm1 ;H prediction cost
     RET
 
+;-----------------------------------------------------------------------------
+; void intra_sad_x3_8x8 ( uint8_t *fenc, uint8_t edge[33], int res[3]);
+;-----------------------------------------------------------------------------
+
+;m0 = DC
+;m6 = V
+;m7 = H
+;m1 = DC score
+;m2 = V score
+;m3 = H score
+;m5 = pixel row
+;m4 = temp
+
+%macro INTRA_SAD_HVDC_ITER 2
+    movq      m5, [r0+FENC_STRIDE*%1]
+    movq      m4, m5
+    psadbw    m4, m0
+%if %1
+    paddw     m1, m4
+%else
+    SWAP      m1, m4
+%endif
+    movq      m4, m5
+    psadbw    m4, m6
+%if %1
+    paddw     m2, m4
+%else
+    SWAP      m2, m4
+%endif
+    pshufw    m4, m7, %2
+    psadbw    m5, m4
+%if %1
+    paddw     m3, m5
+%else
+    SWAP      m3, m5
+%endif
+%endmacro
+
+INIT_MMX
+cglobal x264_intra_sad_x3_8x8_mmxext, 3,3
+    movq      m7, [r1+7]
+    pxor      m0, m0
+    movq      m6, [r1+16]  ;V prediction
+    pxor      m1, m1
+    psadbw    m0, m7
+    psadbw    m1, m6
+    paddw     m0, m1
+    paddw     m0, [pw_8 GLOBAL]
+    psrlw     m0, 4
+    punpcklbw m0, m0
+    pshufw    m0, m0, 0x0 ;DC prediction
+    punpckhbw m7, m7
+    INTRA_SAD_HVDC_ITER 0, 0xff
+    INTRA_SAD_HVDC_ITER 1, 0xaa
+    INTRA_SAD_HVDC_ITER 2, 0x55
+    INTRA_SAD_HVDC_ITER 3, 0x00
+    movq      m7, [r1+7]
+    punpcklbw m7, m7
+    INTRA_SAD_HVDC_ITER 4, 0xff
+    INTRA_SAD_HVDC_ITER 5, 0xaa
+    INTRA_SAD_HVDC_ITER 6, 0x55
+    INTRA_SAD_HVDC_ITER 7, 0x00
+    movd  [r2+0], m2
+    movd  [r2+4], m3
+    movd  [r2+8], m1
+    RET
+
 ;-----------------------------------------------------------------------------
 ; void intra_sad_x3_8x8c ( uint8_t *fenc, uint8_t *fdec, int res[3] );
 ;-----------------------------------------------------------------------------

diff --git a/encoder/analyse.c b/encoder/analyse.c
index 3bcde1191b42e60c30f4d1f5e9d78555d63c9e95..50178e8cf36b93c1870a7037f0521c5ea4c47781 100644 (file)
--- a/encoder/analyse.c
+++ b/encoder/analyse.c
@@ -668,7 +668,7 @@ static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_satd_
         int i_satd_thresh = a->i_mbrd ? COST_MAX : X264_MIN( i_satd_inter, a->i_satd_i16x16 );
         int i_cost = 0;
         h->mb.i_cbp_luma = 0;
-        b_merged_satd = h->pixf.intra_sa8d_x3_8x8 && h->pixf.mbcmp[0] == h->pixf.satd[0];
+        b_merged_satd = h->pixf.intra_mbcmp_x3_8x8 && !h->mb.b_lossless;
 
         // FIXME some bias like in i4x4?
         if( h->sh.i_type == SLICE_TYPE_B )
@@ -689,7 +689,7 @@ static void x264_mb_analyse_intra( x264_t *h, x264_mb_analysis_t *a, int i_satd_
             if( b_merged_satd && i_max == 9 )
             {
                 int satd[9];
-                h->pixf.intra_sa8d_x3_8x8( p_src_by, edge, satd );
+                h->pixf.intra_mbcmp_x3_8x8( p_src_by, edge, satd );
                 satd[i_pred_mode] -= 3 * a->i_lambda;
                 for( i=2; i>=0; i-- )
                 {

diff --git a/tools/checkasm.c b/tools/checkasm.c
index 85a65c237da214209019c2a7f2778d7f806e01d9..b842ac9803f82ce041370701087522f4178c4f0c 100644 (file)
--- a/tools/checkasm.c
+++ b/tools/checkasm.c
@@ -406,6 +406,7 @@ static int check_pixel( int cpu_ref, int cpu_new )
     report( "intra satd_x3 :" );
     TEST_INTRA_MBCMP( intra_sad_x3_16x16 , predict_16x16, sad [PIXEL_16x16], 0 );
     TEST_INTRA_MBCMP( intra_sad_x3_8x8c  , predict_8x8c , sad [PIXEL_8x8]  , 0 );
+    TEST_INTRA_MBCMP( intra_sad_x3_8x8   , predict_8x8  , sad [PIXEL_8x8]  , 1, edge );
     TEST_INTRA_MBCMP( intra_sad_x3_4x4   , predict_4x4  , sad [PIXEL_4x4]  , 0 );
     report( "intra sad_x3 :" );