From 8a028a206e13fab14cdeb4c1eb2cd35cfc32f751 Mon Sep 17 00:00:00 2001
From: Hans Wennborg <hans@hanshq.net>
Date: Fri, 27 Jan 2017 16:37:00 +0000
Subject: [PATCH] Merging r292712 and r292713:
 ------------------------------------------------------------------------
 r292712 | ctopper | 2017-01-20 22:59:35 -0800 (Fri, 20 Jan 2017) | 1 line

[X86] Add test cases that show bad commuting being allowed to create a phsub operation.
------------------------------------------------------------------------

------------------------------------------------------------------------
r292713 | ctopper | 2017-01-20 22:59:38 -0800 (Fri, 20 Jan 2017) | 3 lines

[X86] Don't allow commuting to form phsub operations.

Fixes PR31714.
------------------------------------------------------------------------


git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_40@293299 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp |  4 +--
 test/CodeGen/X86/phaddsub.ll       | 58 ++++++++++++++++++++++++++++++
 2 files changed, 60 insertions(+), 2 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index d9a267b2a36..08fe2bad281 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -33693,11 +33693,11 @@ static SDValue combineSub(SDNode *N, SelectionDAG &DAG,
     }
   }
 
-  // Try to synthesize horizontal adds from adds of shuffles.
+  // Try to synthesize horizontal subs from subs of shuffles.
   EVT VT = N->getValueType(0);
   if (((Subtarget.hasSSSE3() && (VT == MVT::v8i16 || VT == MVT::v4i32)) ||
        (Subtarget.hasInt256() && (VT == MVT::v16i16 || VT == MVT::v8i32))) &&
-      isHorizontalBinOp(Op0, Op1, true))
+      isHorizontalBinOp(Op0, Op1, false))
     return DAG.getNode(X86ISD::HSUB, SDLoc(N), VT, Op0, Op1);
 
   return OptimizeConditionalInDecrement(N, DAG);
diff --git a/test/CodeGen/X86/phaddsub.ll b/test/CodeGen/X86/phaddsub.ll
index 44ad05ec6ed..08015258867 100644
--- a/test/CodeGen/X86/phaddsub.ll
+++ b/test/CodeGen/X86/phaddsub.ll
@@ -225,3 +225,61 @@ define <4 x i32> @phsubd4(<4 x i32> %x) {
   %r = sub <4 x i32> %a, %b
   ret <4 x i32> %r
 }
+
+define <8 x i16> @phsubw1_reverse(<8 x i16> %x, <8 x i16> %y) {
+; SSSE3-LABEL: phsubw1_reverse:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
+; SSSE3-NEXT:    movdqa %xmm1, %xmm4
+; SSSE3-NEXT:    pshufb %xmm3, %xmm4
+; SSSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSSE3-NEXT:    pshufb %xmm3, %xmm2
+; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm4[0]
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; SSSE3-NEXT:    pshufb %xmm3, %xmm1
+; SSSE3-NEXT:    pshufb %xmm3, %xmm0
+; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSSE3-NEXT:    psubw %xmm0, %xmm2
+; SSSE3-NEXT:    movdqa %xmm2, %xmm0
+; SSSE3-NEXT:    retq
+;
+; AVX-LABEL: phsubw1_reverse:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
+; AVX-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
+; AVX-NEXT:    vpshufb %xmm2, %xmm0, %xmm2
+; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
+; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX-NEXT:    vpsubw %xmm0, %xmm2, %xmm0
+; AVX-NEXT:    retq
+  %a = shufflevector <8 x i16> %x, <8 x i16> %y, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+  %b = shufflevector <8 x i16> %x, <8 x i16> %y, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+  %r = sub <8 x i16> %a, %b
+  ret <8 x i16> %r
+}
+
+define <4 x i32> @phsubd1_reverse(<4 x i32> %x, <4 x i32> %y) {
+; SSSE3-LABEL: phsubd1_reverse:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movaps %xmm0, %xmm2
+; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,3],xmm1[1,3]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; SSSE3-NEXT:    psubd %xmm0, %xmm2
+; SSSE3-NEXT:    movdqa %xmm2, %xmm0
+; SSSE3-NEXT:    retq
+;
+; AVX-LABEL: phsubd1_reverse:
+; AVX:       # BB#0:
+; AVX-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[1,3],xmm1[1,3]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX-NEXT:    vpsubd %xmm0, %xmm2, %xmm0
+; AVX-NEXT:    retq
+  %a = shufflevector <4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+  %b = shufflevector <4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+  %r = sub <4 x i32> %a, %b
+  ret <4 x i32> %r
+}
+
-- 
2.40.0