Fix rare overflow in 10-bit intra_satd_x3_16x16 asm

diff --git a/common/x86/pixel-a.asm b/common/x86/pixel-a.asm
index b0aa65956a720249f03beae744474868438b7f31..6f3076cfb8e37b40e2781a35acc319fc1bf5adea 100644 (file)
--- 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