x86: AVX memzero_aligned

diff --git a/common/common.h b/common/common.h
index 53a6ff03c0b834f592867fbcb348e3a8e8a89649..1732d59b698a90509706412f9b86dd8f2d69ac16 100644 (file)
--- a/common/common.h
+++ b/common/common.h
@@ -770,8 +770,8 @@ struct x264_t
             ALIGNED_16( dctcoef fenc_dct4[16][16] );
 
             /* Psy RD SATD/SA8D scores cache */
-            ALIGNED_16( uint64_t fenc_hadamard_cache[9] );
-            ALIGNED_16( uint32_t fenc_satd_cache[32] );
+            ALIGNED_N( uint64_t fenc_hadamard_cache[9] );
+            ALIGNED_N( uint32_t fenc_satd_cache[32] );
 
             /* pointer over mb of the frame to be compressed */
             pixel *p_fenc[3]; /* y,u,v */

diff --git a/common/x86/mc-a2.asm b/common/x86/mc-a2.asm
index caa9396913686256242e1805f5d9e77087e338cb..27e66b869d9b7f7afcff53ecd97c5b7458bd5f76 100644 (file)
--- a/common/x86/mc-a2.asm
+++ b/common/x86/mc-a2.asm
@@ -1223,7 +1223,7 @@ MEMCPY
 ;-----------------------------------------------------------------------------
 ; void *memzero_aligned( void *dst, size_t n );
 ;-----------------------------------------------------------------------------
-%macro MEMZERO 0
+%macro MEMZERO 1
 cglobal memzero_aligned, 2,2
     add  r0, r1
     neg  r1
@@ -1234,21 +1234,21 @@ cglobal memzero_aligned, 2,2
 %endif
 .loop:
 %assign i 0
-%rep 8
+%rep %1
     mova [r0 + r1 + i], m0
 %assign i i+mmsize
 %endrep
-    add r1, mmsize*8
+    add r1, mmsize*%1
     jl .loop
     RET
 %endmacro
 
 INIT_MMX mmx
-MEMZERO
+MEMZERO 8
 INIT_XMM sse
-MEMZERO
-
-
+MEMZERO 8
+INIT_YMM avx
+MEMZERO 4
 
 %if HIGH_BIT_DEPTH == 0
 ;-----------------------------------------------------------------------------

diff --git a/common/x86/mc-c.c b/common/x86/mc-c.c
index 3bb38838e4cd21a1c42e10b6915f0c1715300f64..198d7e402bc812d79b080cc04304de33e6b92d2b 100644 (file)
--- a/common/x86/mc-c.c
+++ b/common/x86/mc-c.c
@@ -129,6 +129,7 @@ void *x264_memcpy_aligned_mmx( void *dst, const void *src, size_t n );
 void *x264_memcpy_aligned_sse( void *dst, const void *src, size_t n );
 void x264_memzero_aligned_mmx( void *dst, size_t n );
 void x264_memzero_aligned_sse( void *dst, size_t n );
+void x264_memzero_aligned_avx( void *dst, size_t n );
 void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
 void x264_integral_init4h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
 void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
@@ -798,6 +799,7 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 
     if( !(cpu&X264_CPU_AVX) )
         return;
+    pf->memzero_aligned = x264_memzero_aligned_avx;
     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
 
     if( cpu&X264_CPU_FMA4 )

diff --git a/encoder/me.c b/encoder/me.c
index 55896025e59e7a67c17e758d0892238487fed81f..8238b96bb09145eb4f5ed85981c1a3bacb1a4967 100644 (file)
--- a/encoder/me.c
+++ b/encoder/me.c
@@ -1058,7 +1058,7 @@ static void ALWAYS_INLINE x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_m
     uint64_t bcostrd = COST_MAX64;
     uint16_t amvd;
     /* each byte of visited represents 8 possible m1y positions, so a 4D array isn't needed */
-    ALIGNED_ARRAY_16( uint8_t, visited,[8],[8][8] );
+    ALIGNED_ARRAY_N( uint8_t, visited,[8],[8][8] );
     /* all permutations of an offset in up to 2 of the dimensions */
     ALIGNED_4( static const int8_t dia4d[33][4] ) =
     {