[FFmpeg-cvslog] Merge commit 'c3e83ad3b7d75f3597f47ada2616ba4479665009'

[James Almer]

ffmpeg | branch: master | James Almer <jamrial at gmail.com> | Tue Jan 31 14:53:27 2017 -0300| [4d0e89ce27bc8d995bf9be831e62ea0b3893e375] | committer: James Almer

Merge commit 'c3e83ad3b7d75f3597f47ada2616ba4479665009'

* commit 'c3e83ad3b7d75f3597f47ada2616ba4479665009':
  x86: hpeldsp: Use EXTERNAL_SSE2_FAST where appropriate

Merged-by: James Almer <jamrial at gmail.com>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=4d0e89ce27bc8d995bf9be831e62ea0b3893e375
---

 libavcodec/x86/hpeldsp_init.c | 27 ++++++++++++---------------
 1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/libavcodec/x86/hpeldsp_init.c b/libavcodec/x86/hpeldsp_init.c
index e8da184..a192924 100644
--- a/libavcodec/x86/hpeldsp_init.c
+++ b/libavcodec/x86/hpeldsp_init.c
@@ -264,21 +264,18 @@ static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int cpu_flags)
 #endif /* HAVE_AMD3DNOW_EXTERNAL */
 }
 
-static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int cpu_flags)
+static void hpeldsp_init_sse2_fast(HpelDSPContext *c, int flags, int cpu_flags)
 {
 #if HAVE_SSE2_EXTERNAL
-    if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {
-        // these functions are slower than mmx on AMD, but faster on Intel
-        c->put_pixels_tab[0][0]        = ff_put_pixels16_sse2;
-        c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
-        c->put_pixels_tab[0][1]        = ff_put_pixels16_x2_sse2;
-        c->put_pixels_tab[0][2]        = ff_put_pixels16_y2_sse2;
-        c->put_pixels_tab[0][3]        = ff_put_pixels16_xy2_sse2;
-        c->avg_pixels_tab[0][0]        = ff_avg_pixels16_sse2;
-        c->avg_pixels_tab[0][1]        = ff_avg_pixels16_x2_sse2;
-        c->avg_pixels_tab[0][2]        = ff_avg_pixels16_y2_sse2;
-        c->avg_pixels_tab[0][3]        = ff_avg_pixels16_xy2_sse2;
-    }
+    c->put_pixels_tab[0][0]        = ff_put_pixels16_sse2;
+    c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
+    c->put_pixels_tab[0][1]        = ff_put_pixels16_x2_sse2;
+    c->put_pixels_tab[0][2]        = ff_put_pixels16_y2_sse2;
+    c->put_pixels_tab[0][3]        = ff_put_pixels16_xy2_sse2;
+    c->avg_pixels_tab[0][0]        = ff_avg_pixels16_sse2;
+    c->avg_pixels_tab[0][1]        = ff_avg_pixels16_x2_sse2;
+    c->avg_pixels_tab[0][2]        = ff_avg_pixels16_y2_sse2;
+    c->avg_pixels_tab[0][3]        = ff_avg_pixels16_xy2_sse2;
 #endif /* HAVE_SSE2_EXTERNAL */
 }
 
@@ -305,8 +302,8 @@ av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
     if (EXTERNAL_MMXEXT(cpu_flags))
         hpeldsp_init_mmxext(c, flags, cpu_flags);
 
-    if (EXTERNAL_SSE2(cpu_flags))
-        hpeldsp_init_sse2(c, flags, cpu_flags);
+    if (EXTERNAL_SSE2_FAST(cpu_flags))
+        hpeldsp_init_sse2_fast(c, flags, cpu_flags);
 
     if (EXTERNAL_SSSE3(cpu_flags))
         hpeldsp_init_ssse3(c, flags, cpu_flags);


======================================================================

diff --cc libavcodec/x86/hpeldsp_init.c
index e8da184,a7a4515..a192924
--- a/libavcodec/x86/hpeldsp_init.c
+++ b/libavcodec/x86/hpeldsp_init.c
@@@ -264,21 -229,12 +264,18 @@@ static void hpeldsp_init_3dnow(HpelDSPC
  #endif /* HAVE_AMD3DNOW_EXTERNAL */
  }
  
- static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int cpu_flags)
+ static void hpeldsp_init_sse2_fast(HpelDSPContext *c, int flags, int cpu_flags)
  {
  #if HAVE_SSE2_EXTERNAL
-     if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {
-         // these functions are slower than mmx on AMD, but faster on Intel
-         c->put_pixels_tab[0][0]        = ff_put_pixels16_sse2;
-         c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
-         c->put_pixels_tab[0][1]        = ff_put_pixels16_x2_sse2;
-         c->put_pixels_tab[0][2]        = ff_put_pixels16_y2_sse2;
-         c->put_pixels_tab[0][3]        = ff_put_pixels16_xy2_sse2;
-         c->avg_pixels_tab[0][0]        = ff_avg_pixels16_sse2;
-         c->avg_pixels_tab[0][1]        = ff_avg_pixels16_x2_sse2;
-         c->avg_pixels_tab[0][2]        = ff_avg_pixels16_y2_sse2;
-         c->avg_pixels_tab[0][3]        = ff_avg_pixels16_xy2_sse2;
-     }
+     c->put_pixels_tab[0][0]        = ff_put_pixels16_sse2;
+     c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
++    c->put_pixels_tab[0][1]        = ff_put_pixels16_x2_sse2;
++    c->put_pixels_tab[0][2]        = ff_put_pixels16_y2_sse2;
++    c->put_pixels_tab[0][3]        = ff_put_pixels16_xy2_sse2;
+     c->avg_pixels_tab[0][0]        = ff_avg_pixels16_sse2;
++    c->avg_pixels_tab[0][1]        = ff_avg_pixels16_x2_sse2;
++    c->avg_pixels_tab[0][2]        = ff_avg_pixels16_y2_sse2;
++    c->avg_pixels_tab[0][3]        = ff_avg_pixels16_xy2_sse2;
  #endif /* HAVE_SSE2_EXTERNAL */
  }
  
@@@ -305,12 -251,9 +302,12 @@@ av_cold void ff_hpeldsp_init_x86(HpelDS
      if (EXTERNAL_MMXEXT(cpu_flags))
          hpeldsp_init_mmxext(c, flags, cpu_flags);
  
-     if (EXTERNAL_SSE2(cpu_flags))
-         hpeldsp_init_sse2(c, flags, cpu_flags);
+     if (EXTERNAL_SSE2_FAST(cpu_flags))
+         hpeldsp_init_sse2_fast(c, flags, cpu_flags);
  
 +    if (EXTERNAL_SSSE3(cpu_flags))
 +        hpeldsp_init_ssse3(c, flags, cpu_flags);
 +
      if (CONFIG_VP3_DECODER)
          ff_hpeldsp_vp3_init_x86(c, cpu_flags, flags);
  }