diff x86/vp8dsp-init.c @ 12209:9eef00a43280 libavcodec

Make mmx VP8 WHT faster Avoid pextrw, since it's slow on many older CPUs. Now it doesn't require mmxext either.

--- a/x86/vp8dsp-init.c	Wed Jul 21 12:37:37 2010 +0000
+++ b/x86/vp8dsp-init.c	Wed Jul 21 20:51:01 2010 +0000
@@ -220,7 +220,7 @@
 
 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 @@
     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 @@
     /* 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);