diff x86/vp8dsp-init.c @ 12235:e08d65897115 libavcodec

VP8: clear DCT blocks in iDCT instead of using clear_blocks. ~0.3% faster overall.
author darkshikari
date Fri, 23 Jul 2010 00:07:16 +0000
parents d07e6037846d
children 1a7903913e9b
line wrap: on
line diff
--- a/x86/vp8dsp-init.c	Fri Jul 23 00:05:44 2010 +0000
+++ b/x86/vp8dsp-init.c	Fri Jul 23 00:07:16 2010 +0000
@@ -222,6 +222,7 @@
 extern void ff_vp8_idct_dc_add_sse4(uint8_t *dst, DCTELEM block[16], int stride);
 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_idct_add_sse(uint8_t *dst, DCTELEM block[16], int stride);
 
 #define DECLARE_LOOP_FILTER(NAME)\
 extern void ff_vp8_v_loop_filter_simple_ ## NAME(uint8_t *dst, int stride, int flim);\
@@ -328,6 +329,7 @@
     }
 
     if (mm_flags & FF_MM_SSE) {
+        c->vp8_idct_add                         = ff_vp8_idct_add_sse;
         c->put_vp8_epel_pixels_tab[0][0][0]     =
         c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_sse;
     }