From: Michael Niedermayer <michaelni@gmx.at>
Date: Fri, 1 Nov 2013 09:26:49 +0000 (+0100)
Subject: Merge commit 'ed1a11ed52bbd1f15bb9b0416d69b7924bee3191'
X-Git-Url: https://git.sesse.net/?a=commitdiff_plain;h=fc5fe4804fd2ee9a29de502e9431b12d027c0c89;p=ffmpeg

Merge commit 'ed1a11ed52bbd1f15bb9b0416d69b7924bee3191'

* commit 'ed1a11ed52bbd1f15bb9b0416d69b7924bee3191':
  gradfun: x86: Factor out common code for some gradfun_filter_line() variants

Conflicts:
	libavfilter/x86/vf_gradfun_init.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
---

fc5fe4804fd2ee9a29de502e9431b12d027c0c89
diff --cc libavfilter/x86/vf_gradfun_init.c
index c638a05e875,3f23bf67994..3f5b842fb92
--- a/libavfilter/x86/vf_gradfun_init.c
+++ b/libavfilter/x86/vf_gradfun_init.c
@@@ -34,21 -35,20 +34,20 @@@ void ff_gradfun_filter_line_ssse3(intpt
                                    const uint16_t *dithers);
  
  void ff_gradfun_blur_line_movdqa_sse2(intptr_t x, uint16_t *buf,
 -                                      uint16_t *buf1, uint16_t *dc,
 -                                      uint8_t *src1, uint8_t *src2);
 +                                      const uint16_t *buf1, uint16_t *dc,
 +                                      const uint8_t *src1, const uint8_t *src2);
  void ff_gradfun_blur_line_movdqu_sse2(intptr_t x, uint16_t *buf,
 -                                      uint16_t *buf1, uint16_t *dc,
 -                                      uint8_t *src1, uint8_t *src2);
 +                                      const uint16_t *buf1, uint16_t *dc,
 +                                      const uint8_t *src1, const uint8_t *src2);
  
  #if HAVE_YASM
- static void gradfun_filter_line_mmxext(uint8_t *dst, const uint8_t *src,
-                                        const uint16_t *dc,
-                                        int width, int thresh,
-                                        const uint16_t *dithers)
+ static void gradfun_filter_line(uint8_t *dst, uint8_t *src, uint16_t *dc,
+                                 int width, int thresh, const uint16_t *dithers,
+                                 int alignment)
  {
      intptr_t x;
-     if (width & 3) {
-         x = width & ~3;
+     if (width & alignment) {
+         x = width & ~alignment;
          ff_gradfun_filter_line_c(dst + x, src + x, dc + x / 2,
                                   width - x, thresh, dithers);
          width = x;
@@@ -58,25 -58,22 +57,23 @@@
                                    thresh, dithers);
  }
  
 -static void gradfun_filter_line_mmxext(uint8_t *dst, uint8_t *src, uint16_t *dc,
++static void gradfun_filter_line_mmxext(uint8_t *dst, const uint8_t *src,
++                                       const uint16_t *dc,
+                                        int width, int thresh,
+                                        const uint16_t *dithers)
+ {
+     gradfun_filter_line(dst, src, dc, width, thresh, dithers, 3);
+ }
+ 
 -static void gradfun_filter_line_ssse3(uint8_t *dst, uint8_t *src, uint16_t *dc,
 +static void gradfun_filter_line_ssse3(uint8_t *dst, const uint8_t *src, const uint16_t *dc,
                                        int width, int thresh,
                                        const uint16_t *dithers)
  {
-     intptr_t x;
-     if (width & 7) {
-         // could be 10% faster if I somehow eliminated this
-         x = width & ~7;
-         ff_gradfun_filter_line_c(dst + x, src + x, dc + x / 2,
-                                  width - x, thresh, dithers);
-         width = x;
-     }
-     x = -width;
-     ff_gradfun_filter_line_ssse3(x, dst + width, src + width, dc + width / 2,
-                                  thresh, dithers);
+     gradfun_filter_line(dst, src, dc, width, thresh, dithers, 7);
  }
  
 -static void gradfun_blur_line_sse2(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
 -                                   uint8_t *src, int src_linesize, int width)
 +static void gradfun_blur_line_sse2(uint16_t *dc, uint16_t *buf, const uint16_t *buf1,
 +                                   const uint8_t *src, int src_linesize, int width)
  {
      intptr_t x = -2 * width;
      if (((intptr_t) src | src_linesize) & 15)