From: James Almer <jamrial@gmail.com>
Date: Wed, 9 Apr 2014 06:33:04 +0000 (-0300)
Subject: x86util: XOP optimized HADDD
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=9b77dffab04e3ea242598454282b40800e720353;p=libx264

x86util: XOP optimized HADDD
---

diff --git a/common/x86/x86util.asm b/common/x86/x86util.asm
index 79ff34f3..b18a1d10 100644
--- a/common/x86/x86util.asm
+++ b/common/x86/x86util.asm
@@ -298,11 +298,16 @@
     paddd   %1, %2
 %endif
 %if mmsize >= 16
+%if cpuflag(xop) && sizeof%1 == 16
+    vphadddq %1, %1
+%endif
     movhlps %2, %1
     paddd   %1, %2
 %endif
+%if notcpuflag(xop) || sizeof%1 != 16
     PSHUFLW %2, %1, q0032
     paddd   %1, %2
+%endif
 %undef %1
 %undef %2
 %endmacro