[x264-devel] Fix rare overflow in 10-bit intra_satd_x3_16x16 asm

[Anton Mitrofanov]

x264 | branch: master | Anton Mitrofanov <Bugmaster at narod.ru> | Tue Feb 21 12:58:22 2012 -0800| [b17c247178a24c218843639c3f46bcfde0edab0a] | committer: Jason Garrett-Glaser

Fix rare overflow in 10-bit intra_satd_x3_16x16 asm

> http://git.videolan.org/gitweb.cgi/x264.git/?a=commit;h=b17c247178a24c218843639c3f46bcfde0edab0a
---

 common/x86/pixel-a.asm |   11 ++++++++---
 1 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/common/x86/pixel-a.asm b/common/x86/pixel-a.asm
index 8d935ef..adeeff0 100644
--- a/common/x86/pixel-a.asm
+++ b/common/x86/pixel-a.asm
@@ -2007,9 +2007,14 @@ cglobal intra_satd_x3_16x16, 0,5
     inc         r4
     jl  .loop_x
 %if HIGH_BIT_DEPTH
-    mova        m7, [pw_1]
-    pmaddwd     m4, m7
-    pmaddwd     m0, m7
+    psrld       m7, m4, 16
+    pslld       m4, 16
+    psrld       m4, 16
+    paddd       m4, m7
+    psrld       m7, m0, 16
+    pslld       m0, 16
+    psrld       m0, 16
+    paddd       m0, m7
     paddd       m4, [sums+32]
     paddd       m0, [sums+24]
     mova [sums+32], m4