RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 03:02:36 +01:00
Code Issues Projects Releases Wiki Activity

[VectorCombine][SVE] Do not fold bitcast shuffle for scalable type.

Browse Source 
First, shuffle cost for scalable type is not known for scalable type;
Second, we cannot reason if the narrowed shuffle mask for scalable type
is a splat or not.

E.g., Bitcast splat vector from type <vscale x 4 x i32> to <vscale x 8 x i16>
will involve narrowing shuffle mask <vscale x 4 x i32> zeroinitializer to
<vscale x 8 x i32> with element sequence of <0, 1, 0, 1, ...>, which cannot be
reasoned if it's a valid splat or not.

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D86995
This commit is contained in:
Huihui Zhang 2020-09-02 15:02:04 -07:00
parent 60cb217dc7
commit dc1e85f7b4
3 changed files with 32 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
lib/Transforms/Vectorize/VectorCombine.cpp
View File

@ -434,11 +434,14 @@ bool VectorCombine::foldBitcastShuf(Instruction &I) {
m_OneUse(m_Shuffle(m_Value(V), m_Undef(), m_Mask(Mask))))))
return false;
// Disallow non-vector casts and length-changing shuffles.
// 1) Do not fold bitcast shuffle for scalable type. First, shuffle cost for
// scalable type is unknown; Second, we cannot reason if the narrowed shuffle
// mask for scalable type is a splat or not.
// 2) Disallow non-vector casts and length-changing shuffles.
// TODO: We could allow any shuffle.
auto *DestTy = dyn_cast<VectorType>(I.getType());
auto *SrcTy = cast<VectorType>(V->getType());
if (!DestTy || I.getOperand(0)->getType() != SrcTy)
auto *DestTy = dyn_cast<FixedVectorType>(I.getType());
auto *SrcTy = dyn_cast<FixedVectorType>(V->getType());
if (!SrcTy || !DestTy || I.getOperand(0)->getType() != SrcTy)
return false;
// The new shuffle must not cost more than the old shuffle. The bitcast is
@ -447,10 +450,8 @@ bool VectorCombine::foldBitcastShuf(Instruction &I) {
TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, SrcTy))
return false;
// FIXME: it should be possible to implement the computation of the widened
// shuffle mask in terms of ElementCount to work with scalable shuffles.
unsigned DestNumElts = cast<FixedVectorType>(DestTy)->getNumElements();
unsigned SrcNumElts = cast<FixedVectorType>(SrcTy)->getNumElements();
unsigned DestNumElts = DestTy->getNumElements();
unsigned SrcNumElts = SrcTy->getNumElements();
SmallVector<int, 16> NewMask;
if (SrcNumElts <= DestNumElts) {
// The bitcast is from wide to narrow/equal elements. The shuffle mask can

2
test/Transforms/VectorCombine/AArch64/lit.local.cfg Normal file
View File

@ -0,0 +1,2 @@
if not 'AArch64' in config.root.targets:
config.unsupported = True

21
test/Transforms/VectorCombine/AArch64/vscale-bitcast-shuffle.ll Normal file
View File

@ -0,0 +1,21 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -vector-combine -S -mtriple=aarch64-- | FileCheck %s --check-prefixes=CHECK
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-unknown-linux-gnu"
; This test checks we are not crashing with TTI when trying to get shuffle cost.
; This test also check that shuffle mask <vscale x 4 x i32> zeroinitializer is
; not narrowed into <0, 1, 0, 1, ...>, which we cannot reason if it's a valid
; splat or not.
define <vscale x 8 x i16> @bitcast_shuffle(<vscale x 4 x i32> %a) {
; CHECK-LABEL: @bitcast_shuffle(
; CHECK-NEXT: [[I:%.*]] = shufflevector <vscale x 4 x i32> [[A:%.*]], <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
; CHECK-NEXT: [[R:%.*]] = bitcast <vscale x 4 x i32> [[I]] to <vscale x 8 x i16>
; CHECK-NEXT: ret <vscale x 8 x i16> [[R]]
;
%i = shufflevector <vscale x 4 x i32> %a, <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
%r = bitcast <vscale x 4 x i32> %i to <vscale x 8 x i16>
ret <vscale x 8 x i16> %r
}