caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[InstCombine] match usub.sat from umax intrinsic

Browse Source 
umax(X, Op1) - Op1 --> usub.sat(X, Op1)

https://alive2.llvm.org/ce/z/HpcGiJ

This happens in 2 or more steps with an icmp-select idiom
instead of an intrinsic. This is another step towards
canonicalization of the min/max intrinsics. See:
D98152
This commit is contained in:
Sanjay Patel
2021-11-05 17:32:24 -04:00
parent 025a2f73a3
commit 83c2fb9f66
2 changed files with 9 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
View File

@@ -2168,6 +2168,13 @@ Instruction *InstCombinerImpl::visitSub(BinaryOperator &I) {
return replaceInstUsesWith(
I, Builder.CreateIntrinsic(Intrinsic::umin, {I.getType()}, {Op0, Y}));
// umax(X, Op1) - Op1 --> usub.sat(X, Op1)
// TODO: The one-use restriction is not strictly necessary, but it may
// require improving other pattern matching and/or codegen.
if (match(Op0, m_OneUse(m_UMax(m_Value(X), m_Specific(Op1)))))
return replaceInstUsesWith(
I, Builder.CreateIntrinsic(Intrinsic::usub_sat, {Ty}, {X, Op1}));
// C - ctpop(X) => ctpop(~X) if C is bitwidth
if (match(Op0, m_SpecificInt(Ty->getScalarSizeInBits())) &&
match(Op1, m_OneUse(m_Intrinsic<Intrinsic::ctpop>(m_Value(X)))))

5
llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
View File

@@ -2132,9 +2132,8 @@ define <3 x i8> @umax_vector_splat_undef(<3 x i8> %x) {
define <3 x i8> @umax_sub_vec(<3 x i8> %x, <3 x i8> %y) {
; CHECK-LABEL: @umax_sub_vec(
; CHECK-NEXT: [[U:%.*]] = call <3 x i8> @llvm.umax.v3i8(<3 x i8> [[X:%.*]], <3 x i8> [[Y:%.*]])
; CHECK-NEXT: [[R:%.*]] = sub <3 x i8> [[U]], [[Y]]
; CHECK-NEXT: ret <3 x i8> [[R]]
; CHECK-NEXT: [[TMP1:%.*]] = call <3 x i8> @llvm.usub.sat.v3i8(<3 x i8> [[X:%.*]], <3 x i8> [[Y:%.*]])
; CHECK-NEXT: ret <3 x i8> [[TMP1]]
;
%u = call <3 x i8> @llvm.umax.v3i8(<3 x i8> %x, <3 x i8> %y)
%r = sub <3 x i8> %u, %y