Make mmx VP8 WHT faster

Avoid pextrw, since it's slow on many older CPUs.
Now it doesn't require mmxext either.

Originally committed as revision 24397 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/x86/vp8dsp-init.c

@@ -220,7 +220,7 @@ HVBILIN(ssse3,  8, 16, 16)
 
 extern void ff_vp8_idct_dc_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
 extern void ff_vp8_idct_dc_add_sse4(uint8_t *dst, DCTELEM block[16], int stride);
-extern void ff_vp8_luma_dc_wht_mmxext(DCTELEM block[4][4][16], DCTELEM dc[16]);
+extern void ff_vp8_luma_dc_wht_mmx(DCTELEM block[4][4][16], DCTELEM dc[16]);
 extern void ff_vp8_idct_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
 
 extern void ff_vp8_v_loop_filter_simple_mmx   (uint8_t *dst, int stride, int flim);
@@ -315,6 +315,7 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
     if (mm_flags & FF_MM_MMX) {
         c->vp8_idct_dc_add                  = ff_vp8_idct_dc_add_mmx;
         c->vp8_idct_add                     = ff_vp8_idct_add_mmx;
+        c->vp8_luma_dc_wht                  = ff_vp8_luma_dc_wht_mmx;
         c->put_vp8_epel_pixels_tab[0][0][0]     =
         c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmx;
         c->put_vp8_epel_pixels_tab[1][0][0]     =
@@ -337,7 +338,6 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
     /* note that 4-tap width=16 functions are missing because w=16
      * is only used for luma, and luma is always a copy or sixtap. */
     if (mm_flags & FF_MM_MMX2) {
-        c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmxext;
         VP8_LUMA_MC_FUNC(0, 16, mmxext);
         VP8_MC_FUNC(1, 8, mmxext);
         VP8_MC_FUNC(2, 4, mmxext);

libavcodec/x86/vp8dsp.asm

@@ -1034,15 +1034,25 @@ cglobal vp8_idct_add_mmx, 3, 3
 ; void vp8_luma_dc_wht_mmxext(DCTELEM block[4][4][16], DCTELEM dc[16])
 ;-----------------------------------------------------------------------------
 
-%macro SCATTER_WHT 1
-    pextrw r1d, m0, %1
-    pextrw r2d, m1, %1
-    mov [r0+2*16*0], r1w
-    mov [r0+2*16*1], r2w
-    pextrw r1d, m2, %1
-    pextrw r2d, m3, %1
-    mov [r0+2*16*2], r1w
-    mov [r0+2*16*3], r2w
+%macro SCATTER_WHT 3
+    movd  r1d, m%1
+    movd  r2d, m%2
+    mov [r0+2*16*(0+%3)], r1w
+    mov [r0+2*16*(1+%3)], r2w
+    shr   r1d, 16
+    shr   r2d, 16
+    psrlq m%1, 32
+    psrlq m%2, 32
+    mov [r0+2*16*(4+%3)], r1w
+    mov [r0+2*16*(5+%3)], r2w
+    movd  r1d, m%1
+    movd  r2d, m%2
+    mov [r0+2*16*(8+%3)], r1w
+    mov [r0+2*16*(9+%3)], r2w
+    shr   r1d, 16
+    shr   r2d, 16
+    mov [r0+2*16*(12+%3)], r1w
+    mov [r0+2*16*(13+%3)], r2w
 %endmacro
 
 %macro HADAMARD4_1D 4
@@ -1052,7 +1062,7 @@ cglobal vp8_idct_add_mmx, 3, 3
 %endmacro
 
 INIT_MMX
-cglobal vp8_luma_dc_wht_mmxext, 2,3
+cglobal vp8_luma_dc_wht_mmx, 2,3
     movq          m0, [r1]
     movq          m1, [r1+8]
     movq          m2, [r1+16]
@@ -1065,13 +1075,8 @@ cglobal vp8_luma_dc_wht_mmxext, 2,3
     psraw         m1, 3
     psraw         m2, 3
     psraw         m3, 3
-    SCATTER_WHT   0
-    add           r0, 2*16*4
-    SCATTER_WHT   1
-    add           r0, 2*16*4
-    SCATTER_WHT   2
-    add           r0, 2*16*4
-    SCATTER_WHT   3
+    SCATTER_WHT   0, 1, 0
+    SCATTER_WHT   2, 3, 2
     RET
 
 ;-----------------------------------------------------------------------------