Update x86 H.264 deblock asm

Includes AVX versions from x264.

3 files changed, 277 insertions, 193 deletions

diff --git a/libavcodec/x86/h264_deblock.asm b/libavcodec/x86/h264_deblock.asm
index 01778a45cb..081c0e1aef 100644
--- a/libavcodec/x86/h264_deblock.asm
+++ b/libavcodec/x86/h264_deblock.asm
@@ -1,10 +1,11 @@
 ;*****************************************************************************
-;* MMX/SSE2-optimized H.264 deblocking code
+;* MMX/SSE2/AVX-optimized H.264 deblocking code
 ;*****************************************************************************
-;* Copyright (C) 2005-2008 x264 project
+;* Copyright (C) 2005-2011 x264 project
 ;*
 ;* Authors: Loren Merritt <lorenm@u.washington.edu>
 ;*          Jason Garrett-Glaser <darkshikari@gmail.com>
+;*          Oskar Arvidsson <oskar@irock.se>
 ;*
 ;* This file is part of Libav.
 ;*
@@ -26,96 +27,135 @@
 %include "x86inc.asm"
 %include "x86util.asm"
 
-SECTION_RODATA
+SECTION .text
 
 cextern pb_0
 cextern pb_1
 cextern pb_3
 cextern pb_A1
 
-SECTION .text
-
 ; expands to [base],...,[base+7*stride]
 %define PASS8ROWS(base, base3, stride, stride3) \
     [base], [base+stride], [base+stride*2], [base3], \
     [base3+stride], [base3+stride*2], [base3+stride3], [base3+stride*4]
 
-; in: 8 rows of 4 bytes in %1..%8
+%define PASS8ROWS(base, base3, stride, stride3, offset) \
+    PASS8ROWS(base+offset, base3+offset, stride, stride3)
+
+; in: 8 rows of 4 bytes in %4..%11
 ; out: 4 rows of 8 bytes in m0..m3
-%macro TRANSPOSE4x8_LOAD 8
-    movd       m0, %1
-    movd       m2, %2
-    movd       m1, %3
-    movd       m3, %4
-    punpcklbw  m0, m2
-    punpcklbw  m1, m3
-    movq       m2, m0
-    punpcklwd  m0, m1
-    punpckhwd  m2, m1
-
-    movd       m4, %5
-    movd       m6, %6
-    movd       m5, %7
-    movd       m7, %8
-    punpcklbw  m4, m6
-    punpcklbw  m5, m7
-    movq       m6, m4
-    punpcklwd  m4, m5
-    punpckhwd  m6, m5
-
-    movq       m1, m0
-    movq       m3, m2
-    punpckldq  m0, m4
-    punpckhdq  m1, m4
-    punpckldq  m2, m6
-    punpckhdq  m3, m6
+%macro TRANSPOSE4x8_LOAD 11
+    movh       m0, %4
+    movh       m2, %5
+    movh       m1, %6
+    movh       m3, %7
+    punpckl%1  m0, m2
+    punpckl%1  m1, m3
+    mova       m2, m0
+    punpckl%2  m0, m1
+    punpckh%2  m2, m1
+
+    movh       m4, %8
+    movh       m6, %9
+    movh       m5, %10
+    movh       m7, %11
+    punpckl%1  m4, m6
+    punpckl%1  m5, m7
+    mova       m6, m4
+    punpckl%2  m4, m5
+    punpckh%2  m6, m5
+
+    punpckh%3  m1, m0, m4
+    punpckh%3  m3, m2, m6
+    punpckl%3  m0, m4
+    punpckl%3  m2, m6
 %endmacro
 
 ; in: 4 rows of 8 bytes in m0..m3
 ; out: 8 rows of 4 bytes in %1..%8
-%macro TRANSPOSE8x4_STORE 8
-    movq       m4, m0
-    movq       m5, m1
-    movq       m6, m2
-    punpckhdq  m4, m4
-    punpckhdq  m5, m5
-    punpckhdq  m6, m6
+%macro TRANSPOSE8x4B_STORE 8
+    punpckhdq  m4, m0, m0
+    punpckhdq  m5, m1, m1
+    punpckhdq  m6, m2, m2
 
     punpcklbw  m0, m1
     punpcklbw  m2, m3
-    movq       m1, m0
-    punpcklwd  m0, m2
-    punpckhwd  m1, m2
-    movd       %1, m0
-    punpckhdq  m0, m0
-    movd       %2, m0
-    movd       %3, m1
+    punpcklwd  m1, m0, m2
+    punpckhwd  m0, m2
+    movh       %1, m1
     punpckhdq  m1, m1
-    movd       %4, m1
+    movh       %2, m1
+    movh       %3, m0
+    punpckhdq  m0, m0
+    movh       %4, m0
 
     punpckhdq  m3, m3
     punpcklbw  m4, m5
     punpcklbw  m6, m3
-    movq       m5, m4
-    punpcklwd  m4, m6
-    punpckhwd  m5, m6
-    movd       %5, m4
-    punpckhdq  m4, m4
-    movd       %6, m4
-    movd       %7, m5
+    punpcklwd  m5, m4, m6
+    punpckhwd  m4, m6
+    movh       %5, m5
     punpckhdq  m5, m5
-    movd       %8, m5
+    movh       %6, m5
+    movh       %7, m4
+    punpckhdq  m4, m4
+    movh       %8, m4
+%endmacro
+
+%macro TRANSPOSE4x8B_LOAD 8
+    TRANSPOSE4x8_LOAD bw, wd, dq, %1, %2, %3, %4, %5, %6, %7, %8
+%endmacro
+
+%macro TRANSPOSE4x8W_LOAD 8
+%if mmsize==16
+    TRANSPOSE4x8_LOAD wd, dq, qdq, %1, %2, %3, %4, %5, %6, %7, %8
+%else
+    SWAP  1, 4, 2, 3
+    mova  m0, [t5]
+    mova  m1, [t5+r1]
+    mova  m2, [t5+r1*2]
+    mova  m3, [t5+t6]
+    TRANSPOSE4x4W 0, 1, 2, 3, 4
+%endif
+%endmacro
+
+%macro TRANSPOSE8x2W_STORE 8
+    punpckhwd  m0, m1, m2
+    punpcklwd  m1, m2
+%if mmsize==8
+    movd       %3, m0
+    movd       %1, m1
+    psrlq      m1, 32
+    psrlq      m0, 32
+    movd       %2, m1
+    movd       %4, m0
+%else
+    movd       %5, m0
+    movd       %1, m1
+    psrldq     m1, 4
+    psrldq     m0, 4
+    movd       %2, m1
+    movd       %6, m0
+    psrldq     m1, 4
+    psrldq     m0, 4
+    movd       %3, m1
+    movd       %7, m0
+    psrldq     m1, 4
+    psrldq     m0, 4
+    movd       %4, m1
+    movd       %8, m0
+%endif
 %endmacro
 
 %macro SBUTTERFLY3 4
-    movq       %4, %2
+    punpckh%1  %4, %2, %3
     punpckl%1  %2, %3
-    punpckh%1  %4, %3
 %endmacro
 
 ; in: 8 rows of 8 (only the middle 6 pels are used) in %1..%8
 ; out: 6 rows of 8 in [%9+0*16] .. [%9+5*16]
 %macro TRANSPOSE6x8_MEM 9
+    RESET_MM_PERMUTATION
     movq  m0, %1
     movq  m1, %2
     movq  m2, %3
@@ -123,30 +163,32 @@ SECTION .text
     movq  m4, %5
     movq  m5, %6
     movq  m6, %7
-    SBUTTERFLY3 bw, m0, m1, m7
-    SBUTTERFLY3 bw, m2, m3, m1
-    SBUTTERFLY3 bw, m4, m5, m3
-    movq  [%9+0x10], m1
-    SBUTTERFLY3 bw, m6, %8, m5
-    SBUTTERFLY3 wd, m0, m2, m1
-    SBUTTERFLY3 wd, m4, m6, m2
+    SBUTTERFLY bw, 0, 1, 7
+    SBUTTERFLY bw, 2, 3, 7
+    SBUTTERFLY bw, 4, 5, 7
+    movq  [%9+0x10], m3
+    SBUTTERFLY3 bw, m6, %8, m7
+    SBUTTERFLY wd, 0, 2, 3
+    SBUTTERFLY wd, 4, 6, 3
     punpckhdq m0, m4
     movq  [%9+0x00], m0
-    SBUTTERFLY3 wd, m7, [%9+0x10], m6
-    SBUTTERFLY3 wd, m3, m5, m4
-    SBUTTERFLY3 dq, m7, m3, m0
-    SBUTTERFLY3 dq, m1, m2, m5
-    punpckldq m6, m4
-    movq  [%9+0x10], m1
-    movq  [%9+0x20], m5
-    movq  [%9+0x30], m7
-    movq  [%9+0x40], m0
-    movq  [%9+0x50], m6
+    SBUTTERFLY3 wd, m1, [%9+0x10], m3
+    SBUTTERFLY wd, 5, 7, 0
+    SBUTTERFLY dq, 1, 5, 0
+    SBUTTERFLY dq, 2, 6, 0
+    punpckldq m3, m7
+    movq  [%9+0x10], m2
+    movq  [%9+0x20], m6
+    movq  [%9+0x30], m1
+    movq  [%9+0x40], m5
+    movq  [%9+0x50], m3
+    RESET_MM_PERMUTATION
 %endmacro
 
 ; in: 8 rows of 8 in %1..%8
 ; out: 8 rows of 8 in %9..%16
 %macro TRANSPOSE8x8_MEM 16
+    RESET_MM_PERMUTATION
     movq  m0, %1
     movq  m1, %2
     movq  m2, %3
@@ -154,38 +196,44 @@ SECTION .text
     movq  m4, %5
     movq  m5, %6
     movq  m6, %7
-    SBUTTERFLY3 bw, m0, m1, m7
-    SBUTTERFLY3 bw, m2, m3, m1
-    SBUTTERFLY3 bw, m4, m5, m3
-    SBUTTERFLY3 bw, m6, %8, m5
-    movq  %9,  m3
-    SBUTTERFLY3 wd, m0, m2, m3
-    SBUTTERFLY3 wd, m4, m6, m2
-    SBUTTERFLY3 wd, m7, m1, m6
-    movq  %11, m2
-    movq  m2,  %9
-    SBUTTERFLY3 wd, m2, m5, m1
-    SBUTTERFLY3 dq, m0, m4, m5
-    SBUTTERFLY3 dq, m7, m2, m4
+    SBUTTERFLY bw, 0, 1, 7
+    SBUTTERFLY bw, 2, 3, 7
+    SBUTTERFLY bw, 4, 5, 7
+    SBUTTERFLY3 bw, m6, %8, m7
+    movq  %9,  m5
+    SBUTTERFLY wd, 0, 2, 5
+    SBUTTERFLY wd, 4, 6, 5
+    SBUTTERFLY wd, 1, 3, 5
+    movq  %11, m6
+    movq  m6,  %9
+    SBUTTERFLY wd, 6, 7, 5
+    SBUTTERFLY dq, 0, 4, 5
+    SBUTTERFLY dq, 1, 6, 5
     movq  %9,  m0
-    movq  %10, m5
-    movq  %13, m7
-    movq  %14, m4
-    SBUTTERFLY3 dq, m3, %11, m0
-    SBUTTERFLY3 dq, m6, m1, m5
-    movq  %11, m3
+    movq  %10, m4
+    movq  %13, m1
+    movq  %14, m6
+    SBUTTERFLY3 dq, m2, %11, m0
+    SBUTTERFLY dq, 3, 7, 4
+    movq  %11, m2
     movq  %12, m0
-    movq  %15, m6
-    movq  %16, m5
+    movq  %15, m3
+    movq  %16, m7
+    RESET_MM_PERMUTATION
 %endmacro
 
 ; out: %4 = |%1-%2|>%3
 ; clobbers: %5
 %macro DIFF_GT 5
+%if avx_enabled == 0
     mova    %5, %2
     mova    %4, %1
     psubusb %5, %1
     psubusb %4, %2
+%else
+    psubusb %5, %2, %1
+    psubusb %4, %1, %2
+%endif
     por     %4, %5
     psubusb %4, %3
 %endmacro
@@ -193,32 +241,28 @@ SECTION .text
 ; out: %4 = |%1-%2|>%3
 ; clobbers: %5
 %macro DIFF_GT2 5
+%ifdef ARCH_X86_64
+    psubusb %5, %2, %1
+    psubusb %4, %1, %2
+%else
     mova    %5, %2
     mova    %4, %1
     psubusb %5, %1
     psubusb %4, %2
+%endif
     psubusb %5, %3
     psubusb %4, %3
     pcmpeqb %4, %5
 %endmacro
 
-%macro SPLATW 1
-%ifidn m0, xmm0
-    pshuflw  %1, %1, 0
-    punpcklqdq %1, %1
-%else
-    pshufw   %1, %1, 0
-%endif
-%endmacro
-
 ; in: m0=p1 m1=p0 m2=q0 m3=q1 %1=alpha-1 %2=beta-1
 ; out: m5=beta-1, m7=mask, %3=alpha-1
 ; clobbers: m4,m6
 %macro LOAD_MASK 2-3
     movd     m4, %1
     movd     m5, %2
-    SPLATW   m4
-    SPLATW   m5
+    SPLATW   m4, m4
+    SPLATW   m5, m5
     packuswb m4, m4  ; 16x alpha-1
     packuswb m5, m5  ; 16x beta-1
 %if %0>2
@@ -237,8 +281,7 @@ SECTION .text
 ; out: m1=p0' m2=q0'
 ; clobbers: m0,3-6
 %macro DEBLOCK_P0_Q0 0
-    mova    m5, m1
-    pxor    m5, m2       ; p0^q0
+    pxor    m5, m1, m2   ; p0^q0
     pand    m5, [pb_1]   ; (p0^q0)&1
     pcmpeqb m4, m4
     pxor    m3, m4
@@ -264,14 +307,12 @@ SECTION .text
 ; out: [q1] = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
 ; clobbers: q2, tmp, tc0
 %macro LUMA_Q1 6
-    mova    %6, m1
-    pavgb   %6, m2
+    pavgb   %6, m1, m2
     pavgb   %2, %6       ; avg(p2,avg(p0,q0))
     pxor    %6, %3
     pand    %6, [pb_1]   ; (p2^avg(p0,q0))&1
     psubusb %2, %6       ; (p2+((p0+q0+1)>>1))>>1
-    mova    %6, %1
-    psubusb %6, %5
+    psubusb %6, %1, %5
     paddusb %5, %1
     pmaxub  %2, %6
     pminub  %2, %5
@@ -280,10 +321,10 @@ SECTION .text
 
 %ifdef ARCH_X86_64
 ;-----------------------------------------------------------------------------
-; void x264_deblock_v_luma_sse2( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
+; void deblock_v_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
-INIT_XMM
-cglobal x264_deblock_v_luma_sse2, 5,5,10
+%macro DEBLOCK_LUMA 1
+cglobal deblock_v_luma_%1, 5,5,10
     movd    m8, [r4] ; tc0
     lea     r4, [r1*3]
     dec     r2d        ; alpha-1
@@ -307,8 +348,7 @@ cglobal x264_deblock_v_luma_sse2, 5,5,10
     movdqa  m3, [r4] ; p2
     DIFF_GT2 m1, m3, m5, m6, m7 ; |p2-p0| > beta-1
     pand    m6, m9
-    mova    m7, m8
-    psubb   m7, m6
+    psubb   m7, m8, m6
     pand    m6, m8
     LUMA_Q1 m0, m3, [r4], [r4+r1], m6, m4
 
@@ -326,10 +366,10 @@ cglobal x264_deblock_v_luma_sse2, 5,5,10
     RET
 
 ;-----------------------------------------------------------------------------
-; void x264_deblock_h_luma_sse2( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
+; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
 INIT_MMX
-cglobal x264_deblock_h_luma_sse2, 5,7
+cglobal deblock_h_luma_%1, 5,7
     movsxd r10, r1d
     lea    r11, [r10+r10*2]
     lea    r6,  [r0-4]
@@ -350,13 +390,13 @@ cglobal x264_deblock_h_luma_sse2, 5,7
 
     ; vertical filter
     ; alpha, beta, tc0 are still in r2d, r3d, r4
-    ; don't backup r6, r5, r10, r11 because x264_deblock_v_luma_sse2 doesn't use them
+    ; don't backup r6, r5, r10, r11 because deblock_v_luma_sse2 doesn't use them
     lea    r0, [pix_tmp+0x30]
     mov    r1d, 0x10
 %ifdef WIN64
     mov    [rsp+0x20], r4
 %endif
-    call   x264_deblock_v_luma_sse2
+    call   deblock_v_luma_%1
 
     ; transpose 16x4 -> original space  (only the middle 4 rows were changed by the filter)
     add    r6, 2
@@ -365,7 +405,7 @@ cglobal x264_deblock_h_luma_sse2, 5,7
     movq   m1, [pix_tmp+0x28]
     movq   m2, [pix_tmp+0x38]
     movq   m3, [pix_tmp+0x48]
-    TRANSPOSE8x4_STORE  PASS8ROWS(r6, r5, r10, r11)
+    TRANSPOSE8x4B_STORE  PASS8ROWS(r6, r5, r10, r11)
 
     shl    r10, 3
     sub    r6,  r10
@@ -375,7 +415,7 @@ cglobal x264_deblock_h_luma_sse2, 5,7
     movq   m1, [pix_tmp+0x20]
     movq   m2, [pix_tmp+0x30]
     movq   m3, [pix_tmp+0x40]
-    TRANSPOSE8x4_STORE  PASS8ROWS(r6, r5, r10, r11)
+    TRANSPOSE8x4B_STORE  PASS8ROWS(r6, r5, r10, r11)
 
 %ifdef WIN64
     add    rsp, 0x98
@@ -383,14 +423,20 @@ cglobal x264_deblock_h_luma_sse2, 5,7
     add    rsp, 0x68
 %endif
     RET
+%endmacro
+
+INIT_XMM
+DEBLOCK_LUMA sse2
+INIT_AVX
+DEBLOCK_LUMA avx
 
 %else
 
 %macro DEBLOCK_LUMA 3
 ;-----------------------------------------------------------------------------
-; void x264_deblock_v8_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
+; void deblock_v8_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
-cglobal x264_deblock_%2_luma_%1, 5,5
+cglobal deblock_%2_luma_%1, 5,5
     lea     r4, [r1*3]
     dec     r2     ; alpha-1
     neg     r4
@@ -419,8 +465,7 @@ cglobal x264_deblock_%2_luma_%1, 5,5
     DIFF_GT2 m1, m3, m5, m6, m7 ; |p2-p0| > beta-1
     pand    m6, m4
     pand    m4, [esp+%3] ; tc
-    mova    m7, m4
-    psubb   m7, m6
+    psubb   m7, m4, m6
     pand    m6, m4
     LUMA_Q1 m0, m3, [r4], [r4+r1], m6, m4
 
@@ -441,10 +486,10 @@ cglobal x264_deblock_%2_luma_%1, 5,5
     RET
 
 ;-----------------------------------------------------------------------------
-; void x264_deblock_h_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
+; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
 INIT_MMX
-cglobal x264_deblock_h_luma_%1, 0,5
+cglobal deblock_h_luma_%1, 0,5
     mov    r0, r0mp
     mov    r3, r1m
     lea    r4, [r3*3]
@@ -467,11 +512,11 @@ cglobal x264_deblock_h_luma_%1, 0,5
     PUSH   dword r2m
     PUSH   dword 16
     PUSH   dword r0
-    call   x264_deblock_%2_luma_%1
+    call   deblock_%2_luma_%1
 %ifidn %2, v8
     add    dword [esp   ], 8 ; pix_tmp+0x38
     add    dword [esp+16], 2 ; tc0+2
-    call   x264_deblock_%2_luma_%1
+    call   deblock_%2_luma_%1
 %endif
     ADD    esp, 20
 
@@ -484,7 +529,7 @@ cglobal x264_deblock_h_luma_%1, 0,5
     movq   m1, [pix_tmp+0x20]
     movq   m2, [pix_tmp+0x30]
     movq   m3, [pix_tmp+0x40]
-    TRANSPOSE8x4_STORE  PASS8ROWS(r0, r1, r3, r4)
+    TRANSPOSE8x4B_STORE  PASS8ROWS(r0, r1, r3, r4)
 
     lea    r0, [r0+r3*8]
     lea    r1, [r1+r3*8]
@@ -492,7 +537,7 @@ cglobal x264_deblock_h_luma_%1, 0,5
     movq   m1, [pix_tmp+0x28]
     movq   m2, [pix_tmp+0x38]
     movq   m3, [pix_tmp+0x48]
-    TRANSPOSE8x4_STORE  PASS8ROWS(r0, r1, r3, r4)
+    TRANSPOSE8x4B_STORE  PASS8ROWS(r0, r1, r3, r4)
 
     ADD    esp, pad
     RET
@@ -502,22 +547,34 @@ INIT_MMX
 DEBLOCK_LUMA mmxext, v8, 8
 INIT_XMM
 DEBLOCK_LUMA sse2, v, 16
+INIT_AVX
+DEBLOCK_LUMA avx, v, 16
 
 %endif ; ARCH
 
 
 
 %macro LUMA_INTRA_P012 4 ; p0..p3 in memory
+%ifdef ARCH_X86_64
+    pavgb t0, p2, p1
+    pavgb t1, p0, q0
+%else
     mova  t0, p2
     mova  t1, p0
     pavgb t0, p1
     pavgb t1, q0
+%endif
     pavgb t0, t1 ; ((p2+p1+1)/2 + (p0+q0+1)/2 + 1)/2
     mova  t5, t1
+%ifdef ARCH_X86_64
+    paddb t2, p2, p1
+    paddb t3, p0, q0
+%else
     mova  t2, p2
     mova  t3, p0
     paddb t2, p1
     paddb t3, q0
+%endif
     paddb t2, t3
     mova  t3, t2
     mova  t4, t2
@@ -527,10 +584,15 @@ DEBLOCK_LUMA sse2, v, 16
     pand  t2, mpb_1
     psubb t0, t2 ; p1' = (p2+p1+p0+q0+2)/4;
 
+%ifdef ARCH_X86_64
+    pavgb t1, p2, q1
+    psubb t2, p2, q1
+%else
     mova  t1, p2
     mova  t2, p2
     pavgb t1, q1
     psubb t2, q1
+%endif
     paddb t3, t3
     psubb t3, t2 ; p2+2*p1+2*p0+2*q0+q1
     pand  t2, mpb_1
@@ -543,10 +605,8 @@ DEBLOCK_LUMA sse2, v, 16
     pand  t3, mpb_1
     psubb t1, t3 ; p0'a = (p2+2*p1+2*p0+2*q0+q1+4)/8
 
-    mova  t3, p0
-    mova  t2, p0
-    pxor  t3, q1
-    pavgb t2, q1
+    pxor  t3, p0, q1
+    pavgb t2, p0, q1
     pand  t3, mpb_1
     psubb t2, t3
     pavgb t2, p1 ; p0'b = (2*p1+p0+q0+2)/4
@@ -560,9 +620,8 @@ DEBLOCK_LUMA sse2, v, 16
     mova  %1, t1 ; store p0
 
     mova  t1, %4 ; p3
-    mova  t2, t1
+    paddb t2, t1, p2
     pavgb t1, p2
-    paddb t2, p2
     pavgb t1, t0 ; (p3+p2+1)/2 + (p2+p1+p0+q0+2)/4
     paddb t2, t2
     paddb t2, t4 ; 2*p3+3*p2+p1+p0+q0
@@ -624,9 +683,9 @@ DEBLOCK_LUMA sse2, v, 16
 %endif
 
 ;-----------------------------------------------------------------------------
-; void x264_deblock_v_luma_intra_sse2( uint8_t *pix, int stride, int alpha, int beta )
+; void deblock_v_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal x264_deblock_%2_luma_intra_%1, 4,6,16
+cglobal deblock_%2_luma_intra_%1, 4,6,16
 %ifndef ARCH_X86_64
     sub     esp, 0x60
 %endif
@@ -686,9 +745,9 @@ cglobal x264_deblock_%2_luma_intra_%1, 4,6,16
 INIT_MMX
 %ifdef ARCH_X86_64
 ;-----------------------------------------------------------------------------
-; void x264_deblock_h_luma_intra_sse2( uint8_t *pix, int stride, int alpha, int beta )
+; void deblock_h_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal x264_deblock_h_luma_intra_%1, 4,7
+cglobal deblock_h_luma_intra_%1, 4,7
     movsxd r10, r1d
     lea    r11, [r10*3]
     lea    r6,  [r0-4]
@@ -704,7 +763,7 @@ cglobal x264_deblock_h_luma_intra_%1, 4,7
 
     lea    r0,  [pix_tmp+0x40]
     mov    r1,  0x10
-    call   x264_deblock_v_luma_intra_%1
+    call   deblock_v_luma_intra_%1
 
     ; transpose 16x6 -> original space (but we can't write only 6 pixels, so really 16x8)
     lea    r5, [r6+r11]
@@ -717,7 +776,7 @@ cglobal x264_deblock_h_luma_intra_%1, 4,7
     add    rsp, 0x88
     RET
 %else
-cglobal x264_deblock_h_luma_intra_%1, 2,4
+cglobal deblock_h_luma_intra_%1, 2,4
     lea    r3,  [r1*3]
     sub    r0,  4
     lea    r2,  [r0+r3]
@@ -736,10 +795,10 @@ cglobal x264_deblock_h_luma_intra_%1, 2,4
     PUSH   dword r2m
     PUSH   dword 16
     PUSH   r0
-    call   x264_deblock_%2_luma_intra_%1
+    call   deblock_%2_luma_intra_%1
 %ifidn %2, v8
     add    dword [rsp], 8 ; pix_tmp+8
-    call   x264_deblock_%2_luma_intra_%1
+    call   deblock_%2_luma_intra_%1
 %endif
     ADD    esp, 16
 
@@ -760,13 +819,13 @@ cglobal x264_deblock_h_luma_intra_%1, 2,4
 
 INIT_XMM
 DEBLOCK_LUMA_INTRA sse2, v
+INIT_AVX
+DEBLOCK_LUMA_INTRA avx , v
 %ifndef ARCH_X86_64
 INIT_MMX
 DEBLOCK_LUMA_INTRA mmxext, v8
 %endif
 
-
-
 INIT_MMX
 
 %macro CHROMA_V_START 0
@@ -790,23 +849,23 @@ INIT_MMX
 %define t6 r6
 
 ;-----------------------------------------------------------------------------
-; void x264_deblock_v_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
+; void ff_deblock_v_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
-cglobal x264_deblock_v_chroma_mmxext, 5,6
+cglobal deblock_v_chroma_mmxext, 5,6
     CHROMA_V_START
     movq  m0, [t5]
     movq  m1, [t5+r1]
     movq  m2, [r0]
     movq  m3, [r0+r1]
-    call x264_chroma_inter_body_mmxext
+    call ff_chroma_inter_body_mmxext
     movq  [t5+r1], m1
     movq  [r0], m2
     RET
 
 ;-----------------------------------------------------------------------------
-; void x264_deblock_h_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
+; void ff_deblock_h_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
-cglobal x264_deblock_h_chroma_mmxext, 5,7
+cglobal deblock_h_chroma_mmxext, 5,7
 %ifdef ARCH_X86_64
     %define buf0 [rsp-24]
     %define buf1 [rsp-16]
@@ -815,17 +874,17 @@ cglobal x264_deblock_h_chroma_mmxext, 5,7
     %define buf1 r2m
 %endif
     CHROMA_H_START
-    TRANSPOSE4x8_LOAD  PASS8ROWS(t5, r0, r1, t6)
+    TRANSPOSE4x8_LOAD  bw, wd, dq, PASS8ROWS(t5, r0, r1, t6)
     movq  buf0, m0
     movq  buf1, m3
-    call x264_chroma_inter_body_mmxext
+    call ff_chroma_inter_body_mmxext
     movq  m0, buf0
     movq  m3, buf1
-    TRANSPOSE8x4_STORE PASS8ROWS(t5, r0, r1, t6)
+    TRANSPOSE8x4B_STORE PASS8ROWS(t5, r0, r1, t6)
     RET
 
 ALIGN 16
-x264_chroma_inter_body_mmxext:
+ff_chroma_inter_body_mmxext:
     LOAD_MASK  r2d, r3d
     movd       m6, [r4] ; tc0
     punpcklbw  m6, m6
@@ -850,31 +909,31 @@ x264_chroma_inter_body_mmxext:
 %define t6 r5
 
 ;-----------------------------------------------------------------------------
-; void x264_deblock_v_chroma_intra( uint8_t *pix, int stride, int alpha, int beta )
+; void ff_deblock_v_chroma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal x264_deblock_v_chroma_intra_mmxext, 4,5
+cglobal deblock_v_chroma_intra_mmxext, 4,5
     CHROMA_V_START
     movq  m0, [t5]
     movq  m1, [t5+r1]
     movq  m2, [r0]
     movq  m3, [r0+r1]
-    call x264_chroma_intra_body_mmxext
+    call ff_chroma_intra_body_mmxext
     movq  [t5+r1], m1
     movq  [r0], m2
     RET
 
 ;-----------------------------------------------------------------------------
-; void x264_deblock_h_chroma_intra( uint8_t *pix, int stride, int alpha, int beta )
+; void ff_deblock_h_chroma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal x264_deblock_h_chroma_intra_mmxext, 4,6
+cglobal deblock_h_chroma_intra_mmxext, 4,6
     CHROMA_H_START
-    TRANSPOSE4x8_LOAD  PASS8ROWS(t5, r0, r1, t6)
-    call x264_chroma_intra_body_mmxext
-    TRANSPOSE8x4_STORE PASS8ROWS(t5, r0, r1, t6)
+    TRANSPOSE4x8_LOAD  bw, wd, dq, PASS8ROWS(t5, r0, r1, t6)
+    call ff_chroma_intra_body_mmxext
+    TRANSPOSE8x4B_STORE PASS8ROWS(t5, r0, r1, t6)
     RET
 
 ALIGN 16
-x264_chroma_intra_body_mmxext:
+ff_chroma_intra_body_mmxext:
     LOAD_MASK r2d, r3d
     movq   m5, m1
     movq   m6, m2
diff --git a/libavcodec/x86/h264dsp_mmx.c b/libavcodec/x86/h264dsp_mmx.c
index 3a783a39ab..7d27c02ea2 100644
--- a/libavcodec/x86/h264dsp_mmx.c
+++ b/libavcodec/x86/h264dsp_mmx.c
@@ -219,11 +219,11 @@ static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40]
 }
 
 #define LF_FUNC(DIR, TYPE, OPT) \
-void ff_x264_deblock_ ## DIR ## _ ## TYPE ## _ ## OPT (uint8_t *pix, int stride, \
-                                               int alpha, int beta, int8_t *tc0);
+void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## OPT (uint8_t *pix, int stride, \
+                                                  int alpha, int beta, int8_t *tc0);
 #define LF_IFUNC(DIR, TYPE, OPT) \
-void ff_x264_deblock_ ## DIR ## _ ## TYPE ## _ ## OPT (uint8_t *pix, int stride, \
-                                               int alpha, int beta);
+void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## OPT (uint8_t *pix, int stride, \
+                                                  int alpha, int beta);
 
 LF_FUNC (h,  chroma,       mmxext)
 LF_IFUNC(h,  chroma_intra, mmxext)
@@ -234,18 +234,18 @@ LF_FUNC (h,  luma,         mmxext)
 LF_IFUNC(h,  luma_intra,   mmxext)
 #if HAVE_YASM && ARCH_X86_32
 LF_FUNC (v8, luma,         mmxext)
-static void ff_x264_deblock_v_luma_mmxext(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
+static void ff_deblock_v_luma_mmxext(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
     if((tc0[0] & tc0[1]) >= 0)
-        ff_x264_deblock_v8_luma_mmxext(pix+0, stride, alpha, beta, tc0);
+        ff_deblock_v8_luma_mmxext(pix+0, stride, alpha, beta, tc0);
     if((tc0[2] & tc0[3]) >= 0)
-        ff_x264_deblock_v8_luma_mmxext(pix+8, stride, alpha, beta, tc0+2);
+        ff_deblock_v8_luma_mmxext(pix+8, stride, alpha, beta, tc0+2);
 }
 LF_IFUNC(v8, luma_intra,   mmxext)
-static void ff_x264_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta)
+static void ff_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta)
 {
-    ff_x264_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta);
-    ff_x264_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta);
+    ff_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta);
+    ff_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta);
 }
 #endif
 
@@ -253,6 +253,10 @@ LF_FUNC (h,  luma,         sse2)
 LF_IFUNC(h,  luma_intra,   sse2)
 LF_FUNC (v,  luma,         sse2)
 LF_IFUNC(v,  luma_intra,   sse2)
+LF_FUNC (h,  luma,         avx)
+LF_IFUNC(h,  luma_intra,   avx)
+LF_FUNC (v,  luma,         avx)
+LF_IFUNC(v,  luma_intra,   avx)
 
 /***********************************/
 /* weighted prediction */
@@ -314,15 +318,15 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
             c->h264_idct_add8      = ff_h264_idct_add8_mmx2;
             c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx2;
 
-            c->h264_v_loop_filter_chroma= ff_x264_deblock_v_chroma_mmxext;
-            c->h264_h_loop_filter_chroma= ff_x264_deblock_h_chroma_mmxext;
-            c->h264_v_loop_filter_chroma_intra= ff_x264_deblock_v_chroma_intra_mmxext;
-            c->h264_h_loop_filter_chroma_intra= ff_x264_deblock_h_chroma_intra_mmxext;
+            c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_mmxext;
+            c->h264_h_loop_filter_chroma= ff_deblock_h_chroma_mmxext;
+            c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_mmxext;
+            c->h264_h_loop_filter_chroma_intra= ff_deblock_h_chroma_intra_mmxext;
 #if ARCH_X86_32
-            c->h264_v_loop_filter_luma= ff_x264_deblock_v_luma_mmxext;
-            c->h264_h_loop_filter_luma= ff_x264_deblock_h_luma_mmxext;
-            c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_mmxext;
-            c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_mmxext;
+            c->h264_v_loop_filter_luma= ff_deblock_v_luma_mmxext;
+            c->h264_h_loop_filter_luma= ff_deblock_h_luma_mmxext;
+            c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_mmxext;
+            c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_mmxext;
 #endif
             c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
             c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
@@ -360,10 +364,10 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
                 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_sse2;
 
 #if HAVE_ALIGNED_STACK
-                c->h264_v_loop_filter_luma = ff_x264_deblock_v_luma_sse2;
-                c->h264_h_loop_filter_luma = ff_x264_deblock_h_luma_sse2;
-                c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_sse2;
-                c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_sse2;
+                c->h264_v_loop_filter_luma = ff_deblock_v_luma_sse2;
+                c->h264_h_loop_filter_luma = ff_deblock_h_luma_sse2;
+                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_sse2;
+                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_sse2;
 #endif
 
                 c->h264_idct_add16 = ff_h264_idct_add16_sse2;
@@ -377,6 +381,14 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
                 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_ssse3;
                 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_ssse3;
             }
+            if (mm_flags&AV_CPU_FLAG_AVX) {
+#if HAVE_ALIGNED_STACK
+                c->h264_v_loop_filter_luma = ff_deblock_v_luma_avx;
+                c->h264_h_loop_filter_luma = ff_deblock_h_luma_avx;
+                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_avx;
+                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_avx;
+#endif
+            }
         }
     }
 #endif
diff --git a/libavcodec/x86/x86util.asm b/libavcodec/x86/x86util.asm
index f731616270..28baf7a96a 100644
--- a/libavcodec/x86/x86util.asm
+++ b/libavcodec/x86/x86util.asm
@@ -24,16 +24,20 @@
 ;******************************************************************************
 
 %macro SBUTTERFLY 4
+%if avx_enabled == 0
     mova      m%4, m%2
     punpckl%1 m%2, m%3
     punpckh%1 m%4, m%3
+%else
+    punpckh%1 m%4, m%2, m%3
+    punpckl%1 m%2, m%3
+%endif
     SWAP %3, %4
 %endmacro
 
 %macro SBUTTERFLY2 4
-    mova      m%4, m%2
-    punpckh%1 m%2, m%3
-    punpckl%1 m%4, m%3
+    punpckl%1 m%4, m%2, m%3
+    punpckh%1 m%2, m%2, m%3
     SWAP %2, %4, %3
 %endmacro
 
@@ -444,3 +448,12 @@
 %macro PMINUB_MMXEXT 3 ; dst, src, ignored
     pminub   %1, %2
 %endmacro
+
+%macro SPLATW 2-3 0
+%if mmsize == 16
+    pshuflw    %1, %2, (%3)*0x55
+    punpcklqdq %1, %1
+%else
+    pshufw     %1, %2, (%3)*0x55
+%endif
+%endmacro