RPCS3/llvm
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm.git synced 2024-12-14 15:39:06 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[X86][SSE] Rename truncateVectorCompareWithPACKSS to truncateVectorWithPACKSS. NFC.

Browse Source 
We no longer rely on the vector source being a comparison result, just have sufficient sign bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@316834 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Simon Pilgrim 2017-10-28 17:59:56 +00:00
parent 3635128533
commit 2146c5ac40
1 changed files with 11 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
lib/Target/X86/X86ISelLowering.cpp
View File

@ -15990,14 +15990,13 @@ static SDValue LowerZERO_EXTEND(SDValue Op, const X86Subtarget &Subtarget,
}
/// Helper to recursively truncate vector elements in half with PACKSS.
/// It makes use of the fact that vector comparison results will be all-zeros
/// or all-ones to prevent the PACKSS from saturating the results.
/// It makes use of the fact that vectors with enough leading sign bits
/// prevent the PACKSS from saturating the results.
/// AVX2 (Int256) sub-targets require extra shuffling as the PACKSS operates
/// within each 128-bit lane.
static SDValue truncateVectorCompareWithPACKSS(EVT DstVT, SDValue In,
const SDLoc &DL,
SelectionDAG &DAG,
const X86Subtarget &Subtarget) {
static SDValue truncateVectorWithPACKSS(EVT DstVT, SDValue In, const SDLoc &DL,
SelectionDAG &DAG,
const X86Subtarget &Subtarget) {
// Requires SSE2 but AVX512 has fast truncate.
if (!Subtarget.hasSSE2() || Subtarget.hasAVX512())
return SDValue();
@ -16065,18 +16064,18 @@ static SDValue truncateVectorCompareWithPACKSS(EVT DstVT, SDValue In,
// If 512bit -> 128bit truncate another stage.
EVT PackedVT = EVT::getVectorVT(Ctx, PackedSVT, NumElems);
Res = DAG.getBitcast(PackedVT, Res);
return truncateVectorCompareWithPACKSS(DstVT, Res, DL, DAG, Subtarget);
return truncateVectorWithPACKSS(DstVT, Res, DL, DAG, Subtarget);
}
// Recursively pack lower/upper subvectors, concat result and pack again.
assert(SrcSizeInBits >= 512 && "Expected 512-bit vector or greater");
EVT PackedVT = EVT::getVectorVT(Ctx, PackedSVT, NumSubElts);
Lo = truncateVectorCompareWithPACKSS(PackedVT, Lo, DL, DAG, Subtarget);
Hi = truncateVectorCompareWithPACKSS(PackedVT, Hi, DL, DAG, Subtarget);
Lo = truncateVectorWithPACKSS(PackedVT, Lo, DL, DAG, Subtarget);
Hi = truncateVectorWithPACKSS(PackedVT, Hi, DL, DAG, Subtarget);
PackedVT = EVT::getVectorVT(Ctx, PackedSVT, NumElems);
SDValue Res = DAG.getNode(ISD::CONCAT_VECTORS, DL, PackedVT, Lo, Hi);
return truncateVectorCompareWithPACKSS(DstVT, Res, DL, DAG, Subtarget);
return truncateVectorWithPACKSS(DstVT, Res, DL, DAG, Subtarget);
}
static SDValue LowerTruncateVecI1(SDValue Op, SelectionDAG &DAG,
@ -16140,7 +16139,7 @@ SDValue X86TargetLowering::LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
// Truncate with PACKSS if we are truncating a vector zero/all-bits result.
if (InVT.getScalarSizeInBits() == DAG.ComputeNumSignBits(In))
if (SDValue V = truncateVectorCompareWithPACKSS(VT, In, DL, DAG, Subtarget))
if (SDValue V = truncateVectorWithPACKSS(VT, In, DL, DAG, Subtarget))
return V;
if ((VT == MVT::v4i32) && (InVT == MVT::v4i64)) {
@ -34328,7 +34327,7 @@ static SDValue combineVectorSignBitsTruncation(SDNode *N, SDLoc &DL,
if (InSVT != MVT::i16 && InSVT != MVT::i32 && InSVT != MVT::i64)
return SDValue();
return truncateVectorCompareWithPACKSS(VT, In, DL, DAG, Subtarget);
return truncateVectorWithPACKSS(VT, In, DL, DAG, Subtarget);
}
static SDValue combineTruncate(SDNode *N, SelectionDAG &DAG,