[llvm] r270019 - [GuardWidening] Use getEquivalentICmp to fold constant compares
Sanjoy Das via llvm-commits
llvm-commits at lists.llvm.org
Wed May 18 20:53:19 PDT 2016
Author: sanjoy
Date: Wed May 18 22:53:17 2016
New Revision: 270019
URL: http://llvm.org/viewvc/llvm-project?rev=270019&view=rev
Log:
[GuardWidening] Use getEquivalentICmp to fold constant compares
`ConstantRange::getEquivalentICmp` is more general, and better
factored.
Modified:
llvm/trunk/lib/Transforms/Scalar/GuardWidening.cpp
llvm/trunk/test/Transforms/GuardWidening/basic.ll
Modified: llvm/trunk/lib/Transforms/Scalar/GuardWidening.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/GuardWidening.cpp?rev=270019&r1=270018&r2=270019&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/GuardWidening.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/GuardWidening.cpp Wed May 18 22:53:17 2016
@@ -356,21 +356,31 @@ bool GuardWideningImpl::widenCondCommon(
if (match(Cond0, m_ICmp(Pred0, m_Value(LHS), m_ConstantInt(RHS0))) &&
match(Cond1, m_ICmp(Pred1, m_Specific(LHS), m_ConstantInt(RHS1)))) {
- // TODO: This logic should be generalized and refactored into a new
- // Constant::getEquivalentICmp helper.
- if (Pred0 == ICmpInst::ICMP_NE && RHS0->isZero())
- Pred0 = ICmpInst::ICMP_UGT;
- if (Pred1 == ICmpInst::ICMP_NE && RHS1->isZero())
- Pred1 = ICmpInst::ICMP_UGT;
+ ConstantRange CR0 =
+ ConstantRange::makeExactICmpRegion(Pred0, RHS0->getValue());
+ ConstantRange CR1 =
+ ConstantRange::makeExactICmpRegion(Pred1, RHS1->getValue());
- if (Pred0 == ICmpInst::ICMP_UGT && Pred1 == ICmpInst::ICMP_UGT) {
+ // SubsetIntersect is a subset of the actual mathematical intersection of
+ // CR0 and CR1, while SupersetIntersect is a superset of the the actual
+ // mathematical intersection. If these two ConstantRanges are equal, then
+ // we know we were able to represent the actual mathematical intersection
+ // of CR0 and CR1, and can use the same to generate an icmp instruction.
+ //
+ // Given what we're doing here and the semantics of guards, it would
+ // actually be correct to just use SubsetIntersect, but that may be too
+ // aggressive in cases we care about.
+ auto SubsetIntersect = CR0.inverse().unionWith(CR1.inverse()).inverse();
+ auto SupersetIntersect = CR0.intersectWith(CR1);
+
+ APInt NewRHSAP;
+ CmpInst::Predicate Pred;
+ if (SubsetIntersect == SupersetIntersect &&
+ SubsetIntersect.getEquivalentICmp(Pred, NewRHSAP)) {
if (InsertPt) {
- ConstantInt *NewRHS =
- RHS0->getValue().ugt(RHS1->getValue()) ? RHS0 : RHS1;
- Result = new ICmpInst(InsertPt, ICmpInst::ICMP_UGT, LHS, NewRHS,
- "wide.chk");
+ ConstantInt *NewRHS = ConstantInt::get(Cond0->getContext(), NewRHSAP);
+ Result = new ICmpInst(InsertPt, Pred, LHS, NewRHS, "wide.chk");
}
-
return true;
}
}
Modified: llvm/trunk/test/Transforms/GuardWidening/basic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/GuardWidening/basic.ll?rev=270019&r1=270018&r2=270019&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/GuardWidening/basic.ll (original)
+++ llvm/trunk/test/Transforms/GuardWidening/basic.ll Wed May 18 22:53:17 2016
@@ -126,7 +126,7 @@ leave:
define void @f_5(i32 %a) {
; CHECK-LABEL: @f_5(
entry:
-; CHECK: %wide.chk = icmp ugt i32 %a, 10
+; CHECK: %wide.chk = icmp uge i32 %a, 11
; CHECK: call void (i1, ...) @llvm.experimental.guard(i1 %wide.chk) [ "deopt"() ]
; CHECK: br i1 undef, label %left, label %right
@@ -335,3 +335,47 @@ entry:
call void(i1, ...) @llvm.experimental.guard(i1 %cond) [ "deopt"() ]
ret void
}
+
+define void @f_13(i32 %a) {
+; CHECK-LABEL: @f_13(
+entry:
+; CHECK: %wide.chk = icmp ult i32 %a, 10
+; CHECK: call void (i1, ...) @llvm.experimental.guard(i1 %wide.chk) [ "deopt"() ]
+; CHECK: br i1 undef, label %left, label %right
+
+ %cond_0 = icmp ult i32 %a, 14
+ call void(i1, ...) @llvm.experimental.guard(i1 %cond_0) [ "deopt"() ]
+ br i1 undef, label %left, label %right
+
+left:
+ %cond_1 = icmp slt i32 %a, 10
+ call void(i1, ...) @llvm.experimental.guard(i1 %cond_1) [ "deopt"() ]
+ ret void
+
+right:
+ ret void
+}
+
+define void @f_14(i32 %a) {
+; CHECK-LABEL: @f_14(
+entry:
+; CHECK: %cond_0 = icmp ult i32 %a, 14
+; CHECK: call void (i1, ...) @llvm.experimental.guard(i1 %cond_0) [ "deopt"() ]
+; CHECK: br i1 undef, label %left, label %right
+
+ %cond_0 = icmp ult i32 %a, 14
+ call void(i1, ...) @llvm.experimental.guard(i1 %cond_0) [ "deopt"() ]
+ br i1 undef, label %left, label %right
+
+left:
+; CHECK: left:
+; CHECK: %cond_1 = icmp sgt i32 %a, 10
+; CHECK: call void (i1, ...) @llvm.experimental.guard(i1 %cond_1) [ "deopt"() ]
+
+ %cond_1 = icmp sgt i32 %a, 10
+ call void(i1, ...) @llvm.experimental.guard(i1 %cond_1) [ "deopt"() ]
+ ret void
+
+right:
+ ret void
+}
More information about the llvm-commits
mailing list