[x264-devel] x86: Correctly use v-prefix for instructions with opmasks

[Henrik Gramner]

x264 | branch: master | Henrik Gramner <henrik at gramner.com> | Sat Mar 31 01:31:57 2018 +0200| [5f7f950c80e330728ecb07bc133e17456870121a] | committer: Henrik Gramner

x86: Correctly use v-prefix for instructions with opmasks

This was always required, but accidentally happened to work correctly
in a few cases.

> http://git.videolan.org/gitweb.cgi/x264.git/?a=commit;h=5f7f950c80e330728ecb07bc133e17456870121a
---

 common/x86/dct-a.asm   | 10 +++++-----
 common/x86/mc-a2.asm   |  8 ++++----
 common/x86/pixel-a.asm |  2 +-
 3 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/common/x86/dct-a.asm b/common/x86/dct-a.asm
index ca842ee9..01311c49 100644
--- a/common/x86/dct-a.asm
+++ b/common/x86/dct-a.asm
@@ -621,8 +621,8 @@ cglobal sub16x16_dct, 3,3,6
     SBUTTERFLY wd, 1, 0, 2
     paddw      m2, m1, m0
     psubw      m3, m1, m0
-    paddw      m2 {k1}, m1       ; 0+1+2+3 0<<1+1-2-3<<1
-    psubw      m3 {k1}, m0       ; 0-1-2+3 0-1<<1+2<<1-3
+    vpaddw     m2 {k1}, m1       ; 0+1+2+3 0<<1+1-2-3<<1
+    vpsubw     m3 {k1}, m0       ; 0-1-2+3 0-1<<1+2<<1-3
     shufps     m1, m2, m3, q2323 ; a3 b3 a2 b2 c3 d3 c2 d2
     punpcklqdq m2, m3            ; a0 b0 a1 b1 c0 d0 c1 d1
     SUMSUB_BA   w, 1, 2, 3
@@ -630,8 +630,8 @@ cglobal sub16x16_dct, 3,3,6
     shufps     m1, m2, q2020     ; a0+a3 b0+b3 c0+c3 d0+d3 a0-a3 b0-b3 c0-c3 d0-d3
     paddw      m2, m1, m3
     psubw      m0, m1, m3
-    paddw      m2 {k2}, m1       ; 0'+1'+2'+3' 0'<<1+1'-2'-3'<<1
-    psubw      m0 {k2}, m3       ; 0'-1'-2'+3' 0'-1'<<1+2'<<1-3'
+    vpaddw     m2 {k2}, m1       ; 0'+1'+2'+3' 0'<<1+1'-2'-3'<<1
+    vpsubw     m0 {k2}, m3       ; 0'-1'-2'+3' 0'-1'<<1+2'<<1-3'
 %endmacro
 
 INIT_XMM avx512
@@ -743,7 +743,7 @@ cglobal sub8x8_dct_dc, 3,3
     paddw      xmm0, xmm2       ; 0+1 0+1 2+3 2+3
     punpckldq  xmm0, xmm1       ; 0+1 0+1 0-1 0-1 2+3 2+3 2-3 2-3
     punpcklqdq xmm1, xmm0, xmm0
-    psubw      xmm0 {k1}, xm3, xmm0
+    vpsubw     xmm0 {k1}, xm3, xmm0
     paddw      xmm0, xmm1       ; 0+1+2+3 0+1-2-3 0-1+2-3 0-1-2+3
     movhps     [r0], xmm0
     RET
diff --git a/common/x86/mc-a2.asm b/common/x86/mc-a2.asm
index e3658f23..3c031313 100644
--- a/common/x86/mc-a2.asm
+++ b/common/x86/mc-a2.asm
@@ -2518,8 +2518,8 @@ cglobal mbtree_propagate_list_internal, 5,7,21
     paddd           m6, m7             ; i_mb_x += 8
     pand            m3, m8             ; {x, y}
     vprold          m1, m3, 20         ; {y, x} << 4
-    psubw           m3 {k4}, m9, m3    ; {32-x, 32-y}, {32-x, y}
-    psubw           m1 {k5}, m10, m1   ; ({32-y, x}, {y, x}) << 4
+    vpsubw          m3 {k4}, m9, m3    ; {32-x, 32-y}, {32-x, y}
+    vpsubw          m1 {k5}, m10, m1   ; ({32-y, x}, {y, x}) << 4
     pmullw          m3, m1
     paddsw          m3, m3             ; prevent signed overflow in idx0 (32*32<<5 == 0x8000)
     pmulhrsw        m2, m3, m4         ; idx01weight idx23weightp
@@ -2530,11 +2530,11 @@ cglobal mbtree_propagate_list_internal, 5,7,21
     vpcmpuw         k2, ym1, ym20, 1    ; {mbx, mbx+1} < width
     kunpckwd        k2, k2, k2
     psrad           m1, m0, 16
-    paddd           m1 {k6}, m11
+    vpaddd          m1 {k6}, m11
     vpcmpud         k1 {k1}, m1, m13, 1 ; mby < height | mby+1 < height
 
     pmaddwd         m0, m15
-    paddd           m0 {k6}, m14        ; idx0 | idx2
+    vpaddd          m0 {k6}, m14        ; idx0 | idx2
     vmovdqu16       m2 {k2}{z}, m2      ; idx01weight | idx23weight
     vptestmd        k1 {k1}, m2, m2     ; mask out offsets with no changes
 
diff --git a/common/x86/pixel-a.asm b/common/x86/pixel-a.asm
index 4271c921..04fe7099 100644
--- a/common/x86/pixel-a.asm
+++ b/common/x86/pixel-a.asm
@@ -4744,7 +4744,7 @@ cglobal intra_sad_x9_8x8, 5,7,8
 %endmacro
 
 %macro SATD_AVX512_END 0-1 0 ; sa8d
-    paddw          m0 {k1}{z}, m1 ; zero-extend to dwords
+    vpaddw         m0 {k1}{z}, m1 ; zero-extend to dwords
 %if ARCH_X86_64
 %if mmsize == 64
     vextracti32x8 ym1, m0, 1