[llvm] r308430 - [X86][XOP] Use default AVX2 lowering for v4i64 ashr by splat constants

[Simon Pilgrim via llvm-commits]

Author: rksimon
Date: Wed Jul 19 03:29:31 2017
New Revision: 308430

URL: http://llvm.org/viewvc/llvm-project?rev=308430&view=rev
Log:
[X86][XOP] Use default AVX2 lowering for v4i64 ashr by splat constants

XOP shifts only support 128-bit vectors, so we were ending up with less optimal codegen requiring constants

Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
    llvm/trunk/test/CodeGen/X86/vector-shift-ashr-256.ll

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=308430&r1=308429&r2=308430&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Wed Jul 19 03:29:31 2017
@@ -22022,8 +22022,9 @@ static SDValue LowerScalarImmediateShift
         return getTargetVShiftByConstNode(X86Opc, dl, VT, R, ShiftAmt, DAG);
 
       // i64 SRA needs to be performed as partial shifts.
-      if ((VT == MVT::v2i64 || (Subtarget.hasInt256() && VT == MVT::v4i64)) &&
-          Op.getOpcode() == ISD::SRA && !Subtarget.hasXOP())
+      if (((!Subtarget.hasXOP() && VT == MVT::v2i64) ||
+           (Subtarget.hasInt256() && VT == MVT::v4i64)) &&
+          Op.getOpcode() == ISD::SRA)
         return ArithmeticShiftRight64(ShiftAmt);
 
       if (VT == MVT::v16i8 ||

Modified: llvm/trunk/test/CodeGen/X86/vector-shift-ashr-256.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-shift-ashr-256.ll?rev=308430&r1=308429&r2=308430&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-shift-ashr-256.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-shift-ashr-256.ll Wed Jul 19 03:29:31 2017
@@ -1699,10 +1699,9 @@ define <4 x i64> @splatconstant_shift_v4
 ;
 ; XOPAVX2-LABEL: splatconstant_shift_v4i64:
 ; XOPAVX2:       # BB#0:
+; XOPAVX2-NEXT:    vpsrad $7, %ymm0, %ymm1
 ; XOPAVX2-NEXT:    vpsrlq $7, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [72057594037927936,72057594037927936,72057594037927936,72057594037927936]
-; XOPAVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpsubq %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatconstant_shift_v4i64: