[llvm] acabad9 - [InstCombine] try to canonicalize icmp with trunc op into mask and cmp
Sanjay Patel via llvm-commits
llvm-commits at lists.llvm.org
Tue Oct 26 14:56:22 PDT 2021
Yes, that should be better. I knew there was a sibling 'ugt' fold, but I
hadn't considered signed preds. Will add tests and try to enhance in the
near future.
On Tue, Oct 26, 2021 at 5:47 PM Roman Lebedev <lebedev.ri at gmail.com> wrote:
> Should this simply use `decomposeBitTestICmp()`?
>
> Roman
>
> On Wed, Oct 27, 2021 at 12:45 AM Sanjay Patel via llvm-commits
> <llvm-commits at lists.llvm.org> wrote:
> >
> >
> > Author: Sanjay Patel
> > Date: 2021-10-26T17:43:28-04:00
> > New Revision: acabad9ff6bf13e00305d9d8621ee8eafc1f8b08
> >
> > URL:
> https://github.com/llvm/llvm-project/commit/acabad9ff6bf13e00305d9d8621ee8eafc1f8b08
> > DIFF:
> https://github.com/llvm/llvm-project/commit/acabad9ff6bf13e00305d9d8621ee8eafc1f8b08.diff
> >
> > LOG: [InstCombine] try to canonicalize icmp with trunc op into mask and
> cmp
> >
> > The motivating test is based on:
> > https://llvm.org/PR52260
> >
> > We have better analysis for X == 0, so try harder to form that.
> >
> > Added:
> >
> >
> > Modified:
> > llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
> > llvm/test/Transforms/InstCombine/icmp-trunc.ll
> > llvm/test/Transforms/InstCombine/signed-truncation-check.ll
> >
> > Removed:
> >
> >
> >
> >
> ################################################################################
> > diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
> b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
> > index 54d751e30ee3..a2db1bc83e76 100644
> > --- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
> > +++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
> > @@ -4589,6 +4589,31 @@ Instruction
> *InstCombinerImpl::foldICmpEquality(ICmpInst &I) {
> > return nullptr;
> > }
> >
> > +static Instruction *foldICmpWithTrunc(ICmpInst &ICmp,
> > + InstCombiner::BuilderTy &Builder)
> {
> > + const ICmpInst::Predicate Pred = ICmp.getPredicate();
> > + Value *Op0 = ICmp.getOperand(0), *Op1 = ICmp.getOperand(1);
> > +
> > + // Try to canonicalize trunc + compare-to-constant into a mask + cmp.
> > + // The trunc masks high bits while the compare may effectively mask
> low bits.
> > + Value *X;
> > + const APInt *C;
> > + if (match(Op0, m_OneUse(m_Trunc(m_Value(X)))) && match(Op1,
> m_Power2(C))) {
> > + if (Pred == ICmpInst::ICMP_ULT) {
> > + // (trunc X) u< Pow2C --> (X & MaskC) == 0
> > + unsigned SrcBits = X->getType()->getScalarSizeInBits();
> > + unsigned DstBits = Op0->getType()->getScalarSizeInBits();
> > + APInt MaskC = APInt::getOneBitSet(SrcBits, DstBits) -
> C->zext(SrcBits);
> > + Value *And = Builder.CreateAnd(X, MaskC);
> > + Constant *Zero = ConstantInt::getNullValue(X->getType());
> > + return new ICmpInst(ICmpInst::ICMP_EQ, And, Zero);
> > + }
> > + // TODO: Handle ugt.
> > + }
> > +
> > + return nullptr;
> > +}
> > +
> > static Instruction *foldICmpWithZextOrSext(ICmpInst &ICmp,
> > InstCombiner::BuilderTy
> &Builder) {
> > assert(isa<CastInst>(ICmp.getOperand(0)) && "Expected cast for
> operand 0");
> > @@ -4732,6 +4757,9 @@ Instruction
> *InstCombinerImpl::foldICmpWithCastOp(ICmpInst &ICmp) {
> > return new ICmpInst(ICmp.getPredicate(), Op0Src, NewOp1);
> > }
> >
> > + if (Instruction *R = foldICmpWithTrunc(ICmp, Builder))
> > + return R;
> > +
> > return foldICmpWithZextOrSext(ICmp, Builder);
> > }
> >
> >
> > diff --git a/llvm/test/Transforms/InstCombine/icmp-trunc.ll
> b/llvm/test/Transforms/InstCombine/icmp-trunc.ll
> > index 7f961f2ccfaf..39ca767a6d2a 100644
> > --- a/llvm/test/Transforms/InstCombine/icmp-trunc.ll
> > +++ b/llvm/test/Transforms/InstCombine/icmp-trunc.ll
> > @@ -5,8 +5,8 @@ declare void @use(i8)
> >
> > define i1 @ult_2(i32 %x) {
> > ; CHECK-LABEL: @ult_2(
> > -; CHECK-NEXT: [[T:%.*]] = trunc i32 [[X:%.*]] to i8
> > -; CHECK-NEXT: [[R:%.*]] = icmp ult i8 [[T]], 2
> > +; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[X:%.*]], 254
> > +; CHECK-NEXT: [[R:%.*]] = icmp eq i32 [[TMP1]], 0
> > ; CHECK-NEXT: ret i1 [[R]]
> > ;
> > %t = trunc i32 %x to i8
> > @@ -16,8 +16,8 @@ define i1 @ult_2(i32 %x) {
> >
> > define <2 x i1> @ult_16_splat(<2 x i16> %x) {
> > ; CHECK-LABEL: @ult_16_splat(
> > -; CHECK-NEXT: [[T:%.*]] = trunc <2 x i16> [[X:%.*]] to <2 x i11>
> > -; CHECK-NEXT: [[R:%.*]] = icmp ult <2 x i11> [[T]], <i11 16, i11 16>
> > +; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i16> [[X:%.*]], <i16 2032, i16
> 2032>
> > +; CHECK-NEXT: [[R:%.*]] = icmp eq <2 x i16> [[TMP1]], zeroinitializer
> > ; CHECK-NEXT: ret <2 x i1> [[R]]
> > ;
> > %t = trunc <2 x i16> %x to <2 x i11>
> > @@ -25,6 +25,8 @@ define <2 x i1> @ult_16_splat(<2 x i16> %x) {
> > ret <2 x i1> %r
> > }
> >
> > +; negative test - need power-of-2 constant
> > +
> > define i1 @ult_3(i32 %x) {
> > ; CHECK-LABEL: @ult_3(
> > ; CHECK-NEXT: [[T:%.*]] = trunc i32 [[X:%.*]] to i8
> > @@ -36,6 +38,8 @@ define i1 @ult_3(i32 %x) {
> > ret i1 %r
> > }
> >
> > +; negative test - no extra use allowed
> > +
> > define i1 @ult_2_use(i32 %x) {
> > ; CHECK-LABEL: @ult_2_use(
> > ; CHECK-NEXT: [[T:%.*]] = trunc i32 [[X:%.*]] to i8
> > @@ -53,12 +57,7 @@ define i1 @ult_2_use(i32 %x) {
> >
> > define i1 @PR52260(i32 %x) {
> > ; CHECK-LABEL: @PR52260(
> > -; CHECK-NEXT: [[IDXPROM:%.*]] = sext i32 [[X:%.*]] to i64
> > -; CHECK-NEXT: [[IDX:%.*]] = getelementptr inbounds [3 x i32], [3 x
> i32]* @a, i64 0, i64 [[IDXPROM]]
> > -; CHECK-NEXT: [[T1:%.*]] = load i32, i32* [[IDX]], align 4
> > -; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 [[T1]] to i8
> > -; CHECK-NEXT: [[TOBOOL:%.*]] = icmp ult i8 [[TMP1]], 2
> > -; CHECK-NEXT: ret i1 [[TOBOOL]]
> > +; CHECK-NEXT: ret i1 true
> > ;
> > %idxprom = sext i32 %x to i64
> > %idx = getelementptr inbounds [3 x i32], [3 x i32]* @a, i64 0, i64
> %idxprom
> > @@ -69,4 +68,3 @@ define i1 @PR52260(i32 %x) {
> > %tobool = icmp eq i8 %conv2, 0
> > ret i1 %tobool
> > }
> > -
> >
> > diff --git
> a/llvm/test/Transforms/InstCombine/signed-truncation-check.ll
> b/llvm/test/Transforms/InstCombine/signed-truncation-check.ll
> > index 105e77284b93..6681d5021472 100644
> > --- a/llvm/test/Transforms/InstCombine/signed-truncation-check.ll
> > +++ b/llvm/test/Transforms/InstCombine/signed-truncation-check.ll
> > @@ -397,8 +397,8 @@ define i1 @positive_trunc_signbit_logical(i32 %arg) {
> >
> > define i1 @positive_trunc_base(i32 %arg) {
> > ; CHECK-LABEL: @positive_trunc_base(
> > -; CHECK-NEXT: [[T1:%.*]] = trunc i32 [[ARG:%.*]] to i16
> > -; CHECK-NEXT: [[T5_SIMPLIFIED:%.*]] = icmp ult i16 [[T1]], 128
> > +; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[ARG:%.*]], 65408
> > +; CHECK-NEXT: [[T5_SIMPLIFIED:%.*]] = icmp eq i32 [[TMP1]], 0
> > ; CHECK-NEXT: ret i1 [[T5_SIMPLIFIED]]
> > ;
> > %t1 = trunc i32 %arg to i16
> > @@ -411,8 +411,8 @@ define i1 @positive_trunc_base(i32 %arg) {
> >
> > define i1 @positive_trunc_base_logical(i32 %arg) {
> > ; CHECK-LABEL: @positive_trunc_base_logical(
> > -; CHECK-NEXT: [[T1:%.*]] = trunc i32 [[ARG:%.*]] to i16
> > -; CHECK-NEXT: [[T5_SIMPLIFIED:%.*]] = icmp ult i16 [[T1]], 128
> > +; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[ARG:%.*]], 65408
> > +; CHECK-NEXT: [[T5_SIMPLIFIED:%.*]] = icmp eq i32 [[TMP1]], 0
> > ; CHECK-NEXT: ret i1 [[T5_SIMPLIFIED]]
> > ;
> > %t1 = trunc i32 %arg to i16
> >
> >
> >
> > _______________________________________________
> > llvm-commits mailing list
> > llvm-commits at lists.llvm.org
> > https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20211026/4f7bae00/attachment.html>
More information about the llvm-commits
mailing list