Various VP8 x86 deblocking speedups

SSSE3 versions, improve SSE2 versions a bit.
SSE2/SSSE3 mbedge h functions are currently broken, so explicitly disable them.

Originally committed as revision 24403 to svn://svn.ffmpeg.org/ffmpeg/trunk

1 files changed, 67 insertions, 32 deletions

diff --git a/libavcodec/x86/vp8dsp.asm b/libavcodec/x86/vp8dsp.asm
index 1d9d8c5b6c..0358781438 100644
--- a/libavcodec/x86/vp8dsp.asm
+++ b/libavcodec/x86/vp8dsp.asm
@@ -1229,18 +1229,22 @@ cglobal vp8_luma_dc_wht_mmx, 2,3
     movd    [%7+%9*2], m%4
 %endmacro
 
-%macro SPLATB_REG 3
+%macro SPLATB_REG 3-4
     movd           %1, %2
+%ifidn %3, ssse3
+    pshufb         %1, %4
+%else
     punpcklbw      %1, %1
 %if mmsize == 16 ; sse2
-    punpcklwd      %1, %1
-    pshufd         %1, %1, 0x0
+    pshuflw        %1, %1, 0x0
+    punpcklqdq     %1, %1
 %elifidn %3, mmx
     punpcklwd      %1, %1
     punpckldq      %1, %1
 %else ; mmxext
     pshufw         %1, %1, 0x0
 %endif
+%endif
 %endmacro
 
 %macro SIMPLE_LOOPFILTER 3
@@ -1252,7 +1256,10 @@ cglobal vp8_%2_loop_filter_simple_%1, 3, %3
 %if mmsize == 8 ; mmx/mmxext
     mov            r3, 2
 %endif
-    SPLATB_REG     m7, r2, %1       ; splat "flim" into register
+%ifidn %1, ssse3
+    pxor           m0, m0
+%endif
+    SPLATB_REG     m7, r2, %1, m0   ; splat "flim" into register
 
     ; set up indexes to address 4 rows
     mov            r2, r1
@@ -1398,6 +1405,8 @@ SIMPLE_LOOPFILTER mmxext, h, 6
 INIT_XMM
 SIMPLE_LOOPFILTER sse2,   v, 3
 SIMPLE_LOOPFILTER sse2,   h, 6
+SIMPLE_LOOPFILTER ssse3,  v, 3
+SIMPLE_LOOPFILTER ssse3,  h, 6
 
 ;-----------------------------------------------------------------------------
 ; void vp8_h/v_loop_filter<size>_inner_<opt>(uint8_t *dst, [uint8_t *v,] int stride,
@@ -1433,11 +1442,15 @@ cglobal vp8_%2_loop_filter16y_inner_%1, 5, %3, %5
 %define stack_reg   hev_thr_reg
 %endif
 
+%ifidn %1, ssse3
+    pxor             m7, m7
+%endif
+
 %ifndef m8 ; mmx/mmxext or sse2 on x86-32
     ; splat function arguments
-    SPLATB_REG       m0, E_reg, %1   ; E
-    SPLATB_REG       m1, I_reg, %1   ; I
-    SPLATB_REG       m2, hev_thr_reg, %1 ; hev_thresh
+    SPLATB_REG       m0, E_reg, %1, m7 ; E
+    SPLATB_REG       m1, I_reg, %1, m7 ; I
+    SPLATB_REG       m2, hev_thr_reg, %1, m7 ; hev_thresh
 
     ; align stack
     mov       stack_reg, rsp         ; backup stack pointer
@@ -1470,9 +1483,9 @@ cglobal vp8_%2_loop_filter16y_inner_%1, 5, %3, %5
 %define q0backup m8
 
     ; splat function arguments
-    SPLATB_REG   flim_E, E_reg, %1   ; E
-    SPLATB_REG   flim_I, I_reg, %1   ; I
-    SPLATB_REG  hev_thr, hev_thr_reg, %1 ; hev_thresh
+    SPLATB_REG   flim_E, E_reg, %1, m7 ; E
+    SPLATB_REG   flim_I, I_reg, %1, m7 ; I
+    SPLATB_REG  hev_thr, hev_thr_reg, %1, m7 ; hev_thresh
 %endif
 
 %if mmsize == 8 && %4 == 16 ; mmx/mmxext
@@ -1884,15 +1897,15 @@ cglobal vp8_%2_loop_filter16y_inner_%1, 5, %3, %5
 %endmacro
 
 INIT_MMX
-INNER_LOOPFILTER mmx,    v, 6, 16, 8
-INNER_LOOPFILTER mmx,    h, 6, 16, 8
-INNER_LOOPFILTER mmxext, v, 6, 16, 8
-INNER_LOOPFILTER mmxext, h, 6, 16, 8
+INNER_LOOPFILTER mmx,    v, 6, 16, 0
+INNER_LOOPFILTER mmx,    h, 6, 16, 0
+INNER_LOOPFILTER mmxext, v, 6, 16, 0
+INNER_LOOPFILTER mmxext, h, 6, 16, 0
 
-INNER_LOOPFILTER mmx,    v, 6,  8, 8
-INNER_LOOPFILTER mmx,    h, 6,  8, 8
-INNER_LOOPFILTER mmxext, v, 6,  8, 8
-INNER_LOOPFILTER mmxext, h, 6,  8, 8
+INNER_LOOPFILTER mmx,    v, 6,  8, 0
+INNER_LOOPFILTER mmx,    h, 6,  8, 0
+INNER_LOOPFILTER mmxext, v, 6,  8, 0
+INNER_LOOPFILTER mmxext, h, 6,  8, 0
 
 INIT_XMM
 INNER_LOOPFILTER sse2,   v, 5, 16, 13
@@ -1904,6 +1917,15 @@ INNER_LOOPFILTER sse2,   h, 6, 16, 13
 INNER_LOOPFILTER sse2,   v, 6,  8, 13
 INNER_LOOPFILTER sse2,   h, 6,  8, 13
 
+INNER_LOOPFILTER ssse3,  v, 5, 16, 13
+%ifdef m8
+INNER_LOOPFILTER ssse3,  h, 5, 16, 13
+%else
+INNER_LOOPFILTER ssse3,  h, 6, 16, 13
+%endif
+INNER_LOOPFILTER ssse3,  v, 6,  8, 13
+INNER_LOOPFILTER ssse3,  h, 6,  8, 13
+
 ;-----------------------------------------------------------------------------
 ; void vp8_h/v_loop_filter<size>_mbedge_<opt>(uint8_t *dst, [uint8_t *v,] int stride,
 ;                                            int flimE, int flimI, int hev_thr);
@@ -1984,11 +2006,15 @@ cglobal vp8_%2_loop_filter16y_mbedge_%1, 5, %3, %5
 %define stack_reg   hev_thr_reg
 %endif
 
+%ifidn %1, ssse3
+    pxor             m7, m7
+%endif
+
 %ifndef m8 ; mmx/mmxext or sse2 on x86-32
     ; splat function arguments
-    SPLATB_REG       m0, E_reg, %1   ; E
-    SPLATB_REG       m1, I_reg, %1   ; I
-    SPLATB_REG       m2, hev_thr_reg, %1 ; hev_thresh
+    SPLATB_REG       m0, E_reg, %1, m7 ; E
+    SPLATB_REG       m1, I_reg, %1, m7 ; I
+    SPLATB_REG       m2, hev_thr_reg, %1, m7 ; hev_thresh
 
     ; align stack
     mov       stack_reg, rsp         ; backup stack pointer
@@ -2028,9 +2054,9 @@ cglobal vp8_%2_loop_filter16y_mbedge_%1, 5, %3, %5
 %define lim_sign m15
 
     ; splat function arguments
-    SPLATB_REG   flim_E, E_reg, %1   ; E
-    SPLATB_REG   flim_I, I_reg, %1   ; I
-    SPLATB_REG  hev_thr, hev_thr_reg, %1 ; hev_thresh
+    SPLATB_REG   flim_E, E_reg, %1, m7 ; E
+    SPLATB_REG   flim_I, I_reg, %1, m7 ; I
+    SPLATB_REG  hev_thr, hev_thr_reg, %1, m7 ; hev_thresh
 %endif
 
 %if mmsize == 8 && %4 == 16 ; mmx/mmxext
@@ -2521,15 +2547,15 @@ cglobal vp8_%2_loop_filter16y_mbedge_%1, 5, %3, %5
 %endmacro
 
 INIT_MMX
-MBEDGE_LOOPFILTER mmx,    v, 6, 16, 8
-MBEDGE_LOOPFILTER mmx,    h, 6, 16, 8
-MBEDGE_LOOPFILTER mmxext, v, 6, 16, 8
-MBEDGE_LOOPFILTER mmxext, h, 6, 16, 8
+MBEDGE_LOOPFILTER mmx,    v, 6, 16, 0
+MBEDGE_LOOPFILTER mmx,    h, 6, 16, 0
+MBEDGE_LOOPFILTER mmxext, v, 6, 16, 0
+MBEDGE_LOOPFILTER mmxext, h, 6, 16, 0
 
-MBEDGE_LOOPFILTER mmx,    v, 6,  8, 8
-MBEDGE_LOOPFILTER mmx,    h, 6,  8, 8
-MBEDGE_LOOPFILTER mmxext, v, 6,  8, 8
-MBEDGE_LOOPFILTER mmxext, h, 6,  8, 8
+MBEDGE_LOOPFILTER mmx,    v, 6,  8, 0
+MBEDGE_LOOPFILTER mmx,    h, 6,  8, 0
+MBEDGE_LOOPFILTER mmxext, v, 6,  8, 0
+MBEDGE_LOOPFILTER mmxext, h, 6,  8, 0
 
 INIT_XMM
 MBEDGE_LOOPFILTER sse2,   v, 5, 16, 16
@@ -2540,3 +2566,12 @@ MBEDGE_LOOPFILTER sse2,   h, 6, 16, 16
 %endif
 MBEDGE_LOOPFILTER sse2,   v, 6,  8, 16
 MBEDGE_LOOPFILTER sse2,   h, 6,  8, 16
+
+MBEDGE_LOOPFILTER ssse3,  v, 5, 16, 16
+%ifdef m8
+MBEDGE_LOOPFILTER ssse3,  h, 5, 16, 16
+%else
+MBEDGE_LOOPFILTER ssse3,  h, 6, 16, 16
+%endif
+MBEDGE_LOOPFILTER ssse3,  v, 6,  8, 16
+MBEDGE_LOOPFILTER ssse3,  h, 6,  8, 16