Fix PR16807

- Lower signed division by constant powers-of-2 to target-independent
  DAG operators instead of target-dependent ones to support them better
  on targets where vector types are legal but shift operators on that
  types are illegal. E.g., on AVX, PSRAW is only available on <8 x i16>
  though <16 x i16> is a legal type.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192721 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Michael Liao 2013-10-15 17:51:02 +00:00
parent d45b3c4653
commit dc8c044a9a
2 changed files with 34 additions and 6 deletions

View File

@ -12467,14 +12467,24 @@ static SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) {
(SplatValue.isPowerOf2() || (-SplatValue).isPowerOf2())) {
unsigned lg2 = SplatValue.countTrailingZeros();
// Splat the sign bit.
SDValue Sz = DAG.getConstant(EltTy.getSizeInBits()-1, MVT::i32);
SDValue SGN = getTargetVShiftNode(X86ISD::VSRAI, dl, VT, N0, Sz, DAG);
SmallVector<SDValue, 16> Sz(NumElts,
DAG.getConstant(EltTy.getSizeInBits() - 1,
EltTy));
SDValue SGN = DAG.getNode(ISD::SRA, dl, VT, N0,
DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Sz[0],
NumElts));
// Add (N0 < 0) ? abs2 - 1 : 0;
SDValue Amt = DAG.getConstant(EltTy.getSizeInBits() - lg2, MVT::i32);
SDValue SRL = getTargetVShiftNode(X86ISD::VSRLI, dl, VT, SGN, Amt, DAG);
SmallVector<SDValue, 16> Amt(NumElts,
DAG.getConstant(EltTy.getSizeInBits() - lg2,
EltTy));
SDValue SRL = DAG.getNode(ISD::SRL, dl, VT, SGN,
DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Amt[0],
NumElts));
SDValue ADD = DAG.getNode(ISD::ADD, dl, VT, N0, SRL);
SDValue Lg2Amt = DAG.getConstant(lg2, MVT::i32);
SDValue SRA = getTargetVShiftNode(X86ISD::VSRAI, dl, VT, ADD, Lg2Amt, DAG);
SmallVector<SDValue, 16> Lg2Amt(NumElts, DAG.getConstant(lg2, EltTy));
SDValue SRA = DAG.getNode(ISD::SRA, dl, VT, ADD,
DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Lg2Amt[0],
NumElts));
// If we're dividing by a positive value, we're done. Otherwise, we must
// negate the result.

View File

@ -0,0 +1,18 @@
; RUN: llc < %s -mtriple=x86_64-linux-gnu -mcpu=core-avx-i | FileCheck %s
define <16 x i16> @f_fu(<16 x i16> %bf) {
allocas:
%avg.i.i = sdiv <16 x i16> %bf, <i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4>
ret <16 x i16> %avg.i.i
}
; CHECK: f_fu
; CHECK: psraw
; CHECK: psrlw
; CHECK: paddw
; CHECK: psraw
; CHECK: psraw
; CHECK: psrlw
; CHECK: paddw
; CHECK: psraw
; CHECK: ret