mmx avg (already existed by not used for bipred)

mmx biweighted avg (3x faster than C)

git-svn-id: svn://svn.videolan.org/x264/trunk@307 df754926-b1dd-0310-bc7b-ec298dee348c

diff --git a/common/amd64/mc-a.asm b/common/amd64/mc-a.asm
index ae4906c288e40578d5322a7c94a69c9d3936b02e..0d4f705deee4e4554acf8ed6d6af39482b874823 100644 (file)
--- a/common/amd64/mc-a.asm
+++ b/common/amd64/mc-a.asm
@@ -74,6 +74,10 @@ cglobal x264_pixel_avg_w8_mmxext
 cglobal x264_pixel_avg_w16_mmxext
 cglobal x264_pixel_avg_w16_sse2
 
+cglobal x264_pixel_avg_weight_4x4_mmxext
+cglobal x264_pixel_avg_weight_w8_mmxext
+cglobal x264_pixel_avg_weight_w16_mmxext
+
 cglobal x264_mc_copy_w4_mmxext
 cglobal x264_mc_copy_w8_mmxext
 cglobal x264_mc_copy_w16_mmxext
@@ -247,6 +251,98 @@ ALIGN 4
 
 
 
+;=============================================================================
+; weighted prediction
+;=============================================================================
+; implicit bipred only:
+; assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64
+
+%macro BIWEIGHT_4P_MMX 2
+    movd      mm0, %1
+    movd      mm1, %2
+    punpcklbw mm0, mm7
+    punpcklbw mm1, mm7
+    pmullw    mm0, mm4
+    pmullw    mm1, mm5
+    paddw     mm0, mm1
+    paddw     mm0, mm6
+    psraw     mm0, 6
+    pmaxsw    mm0, mm7
+    packuswb  mm0, mm0
+    movd      %1,  mm0
+%endmacro
+
+%macro BIWEIGHT_START_MMX 0
+;   mov     rdi, rdi      ; dst
+    movsxd  rsi, esi      ; i_dst
+;   mov     rdx, rdx      ; src
+    movsxd  rcx, ecx      ; i_src
+;   movsxd  r8,  r8d      ; i_weight_dst
+;   movsxd  r9,  r9d      ; i_height
+
+    movd    mm4, r8d
+    pshufw  mm4, mm4, 0   ; weight_dst
+    movq    mm5, [pw_64]
+    psubw   mm5, mm4      ; weight_src
+    movq    mm6, [pw_32]  ; rounding
+    pxor    mm7, mm7
+
+    ALIGN 4
+    .height_loop
+%endmacro
+
+ALIGN 16
+;-----------------------------------------------------------------------------
+;   int __cdecl x264_pixel_avg_weight_w16_mmxext( uint8_t *, int, uint8_t *, int, int, int )
+;-----------------------------------------------------------------------------
+x264_pixel_avg_weight_w16_mmxext:
+    BIWEIGHT_START_MMX
+
+    BIWEIGHT_4P_MMX  [rdi   ], [rdx   ]
+    BIWEIGHT_4P_MMX  [rdi+ 4], [rdx+ 4]
+    BIWEIGHT_4P_MMX  [rdi+ 8], [rdx+ 8]
+    BIWEIGHT_4P_MMX  [rdi+12], [rdx+12]
+
+    add  rdi, rsi
+    add  rdx, rcx
+    dec  r9d
+    jnz  .height_loop
+    ret
+
+ALIGN 16
+;-----------------------------------------------------------------------------
+;   int __cdecl x264_pixel_avg_weight_w8_mmxext( uint8_t *, int, uint8_t *, int, int, int )
+;-----------------------------------------------------------------------------
+x264_pixel_avg_weight_w8_mmxext:
+    BIWEIGHT_START_MMX
+
+    BIWEIGHT_4P_MMX  [rdi      ], [rdx      ]
+    BIWEIGHT_4P_MMX  [rdi+4    ], [rdx+4    ]
+    BIWEIGHT_4P_MMX  [rdi+rsi  ], [rdx+rcx  ]
+    BIWEIGHT_4P_MMX  [rdi+rsi+4], [rdx+rcx+4]
+
+    lea  rdi, [rdi+rsi*2]
+    lea  rdx, [rdx+rcx*2]
+    sub  r9d, byte 2
+    jnz  .height_loop
+    ret
+
+ALIGN 16
+;-----------------------------------------------------------------------------
+;   int __cdecl x264_pixel_avg_weight_4x4_mmxext( uint8_t *, int, uint8_t *, int, int )
+;-----------------------------------------------------------------------------
+x264_pixel_avg_weight_4x4_mmxext:
+    BIWEIGHT_START_MMX
+    BIWEIGHT_4P_MMX  [rdi      ], [rdx      ]
+    BIWEIGHT_4P_MMX  [rdi+rsi  ], [rdx+rcx  ]
+    BIWEIGHT_4P_MMX  [rdi+rsi*2], [rdx+rcx*2]
+    add  rdi, rsi
+    add  rdx, rcx
+    BIWEIGHT_4P_MMX  [rdi+rsi*2], [rdx+rcx*2]
+    ret
+
+
+
 ;=============================================================================
 ; pixel copy
 ;=============================================================================

diff --git a/common/i386/mc-a.asm b/common/i386/mc-a.asm
index db50294ddb470999f54910feba1452b9ff383998..55c383fe26deebfc28ca9032d3868245cdb19e2e 100644 (file)
--- a/common/i386/mc-a.asm
+++ b/common/i386/mc-a.asm
@@ -72,6 +72,10 @@ cglobal x264_pixel_avg_w8_mmxext
 cglobal x264_pixel_avg_w16_mmxext
 cglobal x264_pixel_avg_w16_sse2
 
+cglobal x264_pixel_avg_weight_4x4_mmxext
+cglobal x264_pixel_avg_weight_w8_mmxext
+cglobal x264_pixel_avg_weight_w16_mmxext
+
 cglobal x264_mc_copy_w4_mmxext
 cglobal x264_mc_copy_w8_mmxext
 cglobal x264_mc_copy_w16_mmxext
@@ -244,6 +248,105 @@ ALIGN 4
     ret
 
 
+;=============================================================================
+; weighted prediction
+;=============================================================================
+; implicit bipred only:
+; assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64
+
+%macro BIWEIGHT_4P_MMX 2
+    movd      mm0, %1
+    movd      mm1, %2
+    punpcklbw mm0, mm7
+    punpcklbw mm1, mm7
+    pmullw    mm0, mm4
+    pmullw    mm1, mm5
+    paddw     mm0, mm1
+    paddw     mm0, mm6
+    psraw     mm0, 6
+    pmaxsw    mm0, mm7
+    packuswb  mm0, mm0
+    movd      %1,  mm0
+%endmacro
+
+%macro BIWEIGHT_START_MMX 0
+    push    edi
+    push    esi
+    mov     edi, [esp+12] ; dst
+    mov     esi, [esp+16] ; i_dst
+    mov     edx, [esp+20] ; src
+    mov     ecx, [esp+24] ; i_src
+
+    pshufw  mm4, [esp+28], 0 ; weight_dst
+    movq    mm5, [pw_64]
+    psubw   mm5, mm4      ; weight_src
+    movq    mm6, [pw_32]  ; rounding
+    pxor    mm7, mm7
+%endmacro
+%macro BIWEIGHT_END_MMX 0
+    pop     esi
+    pop     edi
+    ret
+%endmacro
+
+ALIGN 16
+;-----------------------------------------------------------------------------
+;   int __cdecl x264_pixel_avg_weight_w16_mmxext( uint8_t *, int, uint8_t *, int, int, int )
+;-----------------------------------------------------------------------------
+x264_pixel_avg_weight_w16_mmxext:
+    BIWEIGHT_START_MMX
+    mov     eax, [esp+32] ; i_height
+    ALIGN 4
+    .height_loop
+
+    BIWEIGHT_4P_MMX  [edi   ], [edx   ]
+    BIWEIGHT_4P_MMX  [edi+ 4], [edx+ 4]
+    BIWEIGHT_4P_MMX  [edi+ 8], [edx+ 8]
+    BIWEIGHT_4P_MMX  [edi+12], [edx+12]
+
+    add  edi, esi
+    add  edx, ecx
+    dec  eax
+    jnz  .height_loop
+    BIWEIGHT_END_MMX
+
+ALIGN 16
+;-----------------------------------------------------------------------------
+;   int __cdecl x264_pixel_avg_weight_w8_mmxext( uint8_t *, int, uint8_t *, int, int, int )
+;-----------------------------------------------------------------------------
+x264_pixel_avg_weight_w8_mmxext:
+    BIWEIGHT_START_MMX
+    mov     eax, [esp+32]
+    ALIGN 4
+    .height_loop
+
+    BIWEIGHT_4P_MMX  [edi      ], [edx      ]
+    BIWEIGHT_4P_MMX  [edi+4    ], [edx+4    ]
+    BIWEIGHT_4P_MMX  [edi+esi  ], [edx+ecx  ]
+    BIWEIGHT_4P_MMX  [edi+esi+4], [edx+ecx+4]
+
+    lea  edi, [edi+esi*2]
+    lea  edx, [edi+ecx*2]
+    sub  eax, byte 2
+    jnz  .height_loop
+    BIWEIGHT_END_MMX
+
+ALIGN 16
+;-----------------------------------------------------------------------------
+;   int __cdecl x264_pixel_avg_weight_4x4_mmxext( uint8_t *, int, uint8_t *, int, int )
+;-----------------------------------------------------------------------------
+x264_pixel_avg_weight_4x4_mmxext:
+    BIWEIGHT_START_MMX
+    BIWEIGHT_4P_MMX  [edi      ], [edx      ]
+    BIWEIGHT_4P_MMX  [edi+esi  ], [edx+ecx  ]
+    BIWEIGHT_4P_MMX  [edi+esi*2], [edx+ecx*2]
+    add  edi, esi
+    add  edx, ecx
+    BIWEIGHT_4P_MMX  [edi+esi*2], [edx+ecx*2]
+    BIWEIGHT_END_MMX
+
+
+
 ;=============================================================================
 ; pixel copy
 ;=============================================================================

diff --git a/common/i386/mc-c.c b/common/i386/mc-c.c
index 9000f10317a25b93a847342cfae0a0f762765775..7af01c46f520e43a6871936a1f93fbf353328a93 100644 (file)
--- a/common/i386/mc-c.c
+++ b/common/i386/mc-c.c
@@ -42,11 +42,39 @@ extern void x264_pixel_avg_w4_mmxext( uint8_t *,  int, uint8_t *, int, uint8_t *
 extern void x264_pixel_avg_w8_mmxext( uint8_t *,  int, uint8_t *, int, uint8_t *, int, int );
 extern void x264_pixel_avg_w16_mmxext( uint8_t *,  int, uint8_t *, int, uint8_t *, int, int );
 extern void x264_pixel_avg_w16_sse2( uint8_t *,  int, uint8_t *, int, uint8_t *, int, int );
+extern void x264_pixel_avg_weight_4x4_mmxext( uint8_t *, int, uint8_t *, int, int );
+extern void x264_pixel_avg_weight_w8_mmxext( uint8_t *, int, uint8_t *, int, int, int );
+extern void x264_pixel_avg_weight_w16_mmxext( uint8_t *, int, uint8_t *, int, int, int );
 extern void x264_mc_copy_w4_mmxext( uint8_t *, int, uint8_t *, int, int );
 extern void x264_mc_copy_w8_mmxext( uint8_t *, int, uint8_t *, int, int );
 extern void x264_mc_copy_w16_mmxext( uint8_t *, int, uint8_t *, int, int );
 extern void x264_mc_copy_w16_sse2( uint8_t *, int, uint8_t *, int, int );
 
+#define AVG(W,H) \
+static void x264_pixel_avg_ ## W ## x ## H ## _mmxext( uint8_t *dst, int i_dst, uint8_t *src, int i_src ) \
+{ \
+    x264_pixel_avg_w ## W ## _mmxext( dst, i_dst, dst, i_dst, src, i_src, H ); \
+}
+AVG(16,16)
+AVG(16,8)
+AVG(8,16)
+AVG(8,8)
+AVG(8,4)
+AVG(4,8)
+AVG(4,4)
+AVG(4,2)
+
+#define AVG_WEIGHT(W,H) \
+void x264_pixel_avg_weight_ ## W ## x ## H ## _mmxext( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int i_weight_dst ) \
+{ \
+    x264_pixel_avg_weight_w ## W ## _mmxext( dst, i_dst, src, i_src, i_weight_dst, H ); \
+}
+AVG_WEIGHT(16,16)
+AVG_WEIGHT(16,8)
+AVG_WEIGHT(8,16)
+AVG_WEIGHT(8,8)
+AVG_WEIGHT(8,4)
+
 #if 0
 
 #if __GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 3)
@@ -1128,6 +1156,23 @@ void x264_mc_mmxext_init( x264_mc_functions_t *pf )
 {
     pf->mc_luma   = mc_luma_mmx;
     pf->get_ref   = get_ref_mmx;
+
+    pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmxext;
+    pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmxext;
+    pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmxext;
+    pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmxext;
+    pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmxext;
+    pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmxext;
+    pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmxext;
+    pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmxext;
+    
+    pf->avg_weight[PIXEL_16x16] = x264_pixel_avg_weight_16x16_mmxext;
+    pf->avg_weight[PIXEL_16x8]  = x264_pixel_avg_weight_16x8_mmxext;
+    pf->avg_weight[PIXEL_8x16]  = x264_pixel_avg_weight_8x16_mmxext;
+    pf->avg_weight[PIXEL_8x8]   = x264_pixel_avg_weight_8x8_mmxext;
+    pf->avg_weight[PIXEL_8x4]   = x264_pixel_avg_weight_8x4_mmxext;
+    pf->avg_weight[PIXEL_4x4]   = x264_pixel_avg_weight_4x4_mmxext;
+    // avg_weight_4x8 is rare and 4x2 is not used
 }
 void x264_mc_sse2_init( x264_mc_functions_t *pf )
 {