[SimplifyCFG] Fold the contiguous wrapping cases into ICmp. #161000
Merged
+331
−47
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-coroutines @llvm/pr-subscribers-llvm-transforms Author: dianqk (dianqk) ChangesFixes #157113. Take the following IR as an example; we know the destination of the define i32 @<!-- -->src(i8 range(i8 0, 6) %arg) {
switch i8 %arg, label %else [
i8 0, label %if
i8 4, label %if
i8 5, label %if
]
if:
ret i32 0
else:
ret i32 1
}We can first try the non-wrapping range for both destinations, but I don't see how that would be any better. Proof: https://alive2.llvm.org/ce/z/acdWD4. Full diff: https://github.com/llvm/llvm-project/pull/161000.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
index 5e719c6c8cbb7..b02aed3f8f5ea 100644
--- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -5761,15 +5761,49 @@ bool SimplifyCFGOpt::simplifyUnreachable(UnreachableInst *UI) {
return Changed;
}
-static bool casesAreContiguous(SmallVectorImpl<ConstantInt *> &Cases) {
+static bool casesAreContiguous(Value *Condition,
+ SmallVectorImpl<ConstantInt *> &Cases,
+ ConstantInt *&ContiguousCasesMin,
+ ConstantInt *&ContiguousCasesMax,
+ bool &IsWrapping) {
assert(Cases.size() >= 1);
array_pod_sort(Cases.begin(), Cases.end(), constantIntSortPredicate);
- for (size_t I = 1, E = Cases.size(); I != E; ++I) {
- if (Cases[I - 1]->getValue() != Cases[I]->getValue() + 1)
+ auto Min = Cases.back()->getValue();
+ auto Max = Cases.front()->getValue();
+ auto Offset = Max - Min;
+ auto ContiguousOffset = Cases.size() - 1;
+ if (Offset == ContiguousOffset) {
+ ContiguousCasesMin = Cases.back();
+ ContiguousCasesMax = Cases.front();
+ return true;
+ }
+ ConstantRange CR = computeConstantRange(Condition, /*ForSigned=*/false);
+ // If this is a wrapping contiguous range, that is, [Min, OtherMin] +
+ // [OtherMax, Max] (also [OtherMax, OtherMin]), [OtherMin+1, OtherMax-1] is a
+ // contiguous range for the other destination. N.B. If CR is not a full range,
+ // Max+1 is not equal to Min. It's not continuous in arithmetic.
+ if (Max == CR.getUnsignedMax() && Min == CR.getUnsignedMin()) {
+ assert(Cases.size() >= 2);
+ auto *It =
+ std::adjacent_find(Cases.begin(), Cases.end(), [](auto L, auto R) {
+ return L->getValue() != R->getValue() + 1;
+ });
+ if (It == Cases.end())
return false;
+ auto *OtherMax = *It;
+ auto *OtherMin = *(It + 1);
+ if ((Max - OtherMax->getValue()) + (OtherMin->getValue() - Min) ==
+ Cases.size() - 2) {
+ ContiguousCasesMin = cast<ConstantInt>(
+ ConstantInt::get(OtherMin->getType(), OtherMin->getValue() + 1));
+ ContiguousCasesMax = cast<ConstantInt>(
+ ConstantInt::get(OtherMax->getType(), OtherMax->getValue() - 1));
+ IsWrapping = true;
+ return true;
+ }
}
- return true;
+ return false;
}
static void createUnreachableSwitchDefault(SwitchInst *Switch,
@@ -5840,25 +5874,34 @@ bool SimplifyCFGOpt::turnSwitchRangeIntoICmp(SwitchInst *SI,
assert(!CasesA.empty() || HasDefault);
// Figure out if one of the sets of cases form a contiguous range.
- SmallVectorImpl<ConstantInt *> *ContiguousCases = nullptr;
+ ConstantInt *ContiguousCasesMin = nullptr;
+ ConstantInt *ContiguousCasesMax = nullptr;
BasicBlock *ContiguousDest = nullptr;
BasicBlock *OtherDest = nullptr;
- if (!CasesA.empty() && casesAreContiguous(CasesA)) {
- ContiguousCases = &CasesA;
+ bool IsWrapping = false;
+ if (!CasesA.empty() &&
+ casesAreContiguous(SI->getCondition(), CasesA, ContiguousCasesMin,
+ ContiguousCasesMax, IsWrapping)) {
ContiguousDest = DestA;
OtherDest = DestB;
- } else if (casesAreContiguous(CasesB)) {
- ContiguousCases = &CasesB;
+ } else if (casesAreContiguous(SI->getCondition(), CasesB, ContiguousCasesMin,
+ ContiguousCasesMax, IsWrapping)) {
ContiguousDest = DestB;
OtherDest = DestA;
} else
return false;
+ if (IsWrapping)
+ std::swap(ContiguousDest, OtherDest);
+
// Start building the compare and branch.
- Constant *Offset = ConstantExpr::getNeg(ContiguousCases->back());
- Constant *NumCases =
- ConstantInt::get(Offset->getType(), ContiguousCases->size());
+ auto ContiguousCasesSize =
+ (ContiguousCasesMax->getValue() - ContiguousCasesMin->getValue())
+ .getZExtValue() +
+ 1;
+ Constant *Offset = ConstantExpr::getNeg(ContiguousCasesMin);
+ Constant *NumCases = ConstantInt::get(Offset->getType(), ContiguousCasesSize);
Value *Sub = SI->getCondition();
if (!Offset->isNullValue())
@@ -5866,7 +5909,7 @@ bool SimplifyCFGOpt::turnSwitchRangeIntoICmp(SwitchInst *SI,
Value *Cmp;
// If NumCases overflowed, then all possible values jump to the successor.
- if (NumCases->isNullValue() && !ContiguousCases->empty())
+ if (NumCases->isNullValue() && ContiguousCasesSize != 0)
Cmp = ConstantInt::getTrue(SI->getContext());
else
Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch");
@@ -5895,14 +5938,14 @@ bool SimplifyCFGOpt::turnSwitchRangeIntoICmp(SwitchInst *SI,
// Prune obsolete incoming values off the successors' PHI nodes.
for (auto BBI = ContiguousDest->begin(); isa<PHINode>(BBI); ++BBI) {
- unsigned PreviousEdges = ContiguousCases->size();
+ unsigned PreviousEdges = ContiguousCasesSize;
if (ContiguousDest == SI->getDefaultDest())
++PreviousEdges;
for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I)
cast<PHINode>(BBI)->removeIncomingValue(SI->getParent());
}
for (auto BBI = OtherDest->begin(); isa<PHINode>(BBI); ++BBI) {
- unsigned PreviousEdges = SI->getNumCases() - ContiguousCases->size();
+ unsigned PreviousEdges = SI->getNumCases() - ContiguousCasesSize;
if (OtherDest == SI->getDefaultDest())
++PreviousEdges;
for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I)
diff --git a/llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll b/llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll
index 8f2ae2d054f1e..c6904fe273f09 100644
--- a/llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll
+++ b/llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll
@@ -188,4 +188,97 @@ exit:
ret void
}
+define i32 @wrapping_known_range(i8 range(i8 0, 6) %arg) {
+; CHECK-LABEL: @wrapping_known_range(
+; CHECK-NEXT: [[ARG_OFF:%.*]] = add i8 [[ARG:%.*]], -1
+; CHECK-NEXT: [[SWITCH:%.*]] = icmp ult i8 [[ARG_OFF]], 3
+; CHECK-NEXT: br i1 [[SWITCH]], label [[ELSE:%.*]], label [[IF:%.*]]
+; CHECK: common.ret:
+; CHECK-NEXT: [[COMMON_RET_OP:%.*]] = phi i32 [ [[I0:%.*]], [[IF]] ], [ [[I1:%.*]], [[ELSE]] ]
+; CHECK-NEXT: ret i32 [[COMMON_RET_OP]]
+; CHECK: if:
+; CHECK-NEXT: [[I0]] = call i32 @f(i32 0)
+; CHECK-NEXT: br label [[COMMON_RET:%.*]]
+; CHECK: else:
+; CHECK-NEXT: [[I1]] = call i32 @f(i32 1)
+; CHECK-NEXT: br label [[COMMON_RET]]
+;
+ switch i8 %arg, label %else [
+ i8 0, label %if
+ i8 4, label %if
+ i8 5, label %if
+ ]
+
+if:
+ %i0 = call i32 @f(i32 0)
+ ret i32 %i0
+
+else:
+ %i1 = call i32 @f(i32 1)
+ ret i32 %i1
+}
+
+define i32 @wrapping_range(i8 %arg) {
+; CHECK-LABEL: @wrapping_range(
+; CHECK-NEXT: [[ARG_OFF:%.*]] = add i8 [[ARG:%.*]], -1
+; CHECK-NEXT: [[SWITCH:%.*]] = icmp ult i8 [[ARG_OFF]], -4
+; CHECK-NEXT: br i1 [[SWITCH]], label [[ELSE:%.*]], label [[IF:%.*]]
+; CHECK: common.ret:
+; CHECK-NEXT: [[COMMON_RET_OP:%.*]] = phi i32 [ [[I0:%.*]], [[IF]] ], [ [[I1:%.*]], [[ELSE]] ]
+; CHECK-NEXT: ret i32 [[COMMON_RET_OP]]
+; CHECK: if:
+; CHECK-NEXT: [[I0]] = call i32 @f(i32 0)
+; CHECK-NEXT: br label [[COMMON_RET:%.*]]
+; CHECK: else:
+; CHECK-NEXT: [[I1]] = call i32 @f(i32 1)
+; CHECK-NEXT: br label [[COMMON_RET]]
+;
+ switch i8 %arg, label %else [
+ i8 0, label %if
+ i8 -3, label %if
+ i8 -2, label %if
+ i8 -1, label %if
+ ]
+
+if:
+ %i0 = call i32 @f(i32 0)
+ ret i32 %i0
+
+else:
+ %i1 = call i32 @f(i32 1)
+ ret i32 %i1
+}
+
+define i32 @no_continuous_wrapping_range(i8 %arg) {
+; CHECK-LABEL: @no_continuous_wrapping_range(
+; CHECK-NEXT: switch i8 [[ARG:%.*]], label [[ELSE:%.*]] [
+; CHECK-NEXT: i8 0, label [[IF:%.*]]
+; CHECK-NEXT: i8 -3, label [[IF]]
+; CHECK-NEXT: i8 -1, label [[IF]]
+; CHECK-NEXT: ]
+; CHECK: common.ret:
+; CHECK-NEXT: [[COMMON_RET_OP:%.*]] = phi i32 [ [[I0:%.*]], [[IF]] ], [ [[I1:%.*]], [[ELSE]] ]
+; CHECK-NEXT: ret i32 [[COMMON_RET_OP]]
+; CHECK: if:
+; CHECK-NEXT: [[I0]] = call i32 @f(i32 0)
+; CHECK-NEXT: br label [[COMMON_RET:%.*]]
+; CHECK: else:
+; CHECK-NEXT: [[I1]] = call i32 @f(i32 1)
+; CHECK-NEXT: br label [[COMMON_RET]]
+;
+ switch i8 %arg, label %else [
+ i8 0, label %if
+ i8 -3, label %if
+ i8 -1, label %if
+ ]
+
+if:
+ %i0 = call i32 @f(i32 0)
+ ret i32 %i0
+
+else:
+ %i1 = call i32 @f(i32 1)
+ ret i32 %i1
+}
+
declare void @bar(ptr nonnull dereferenceable(4))
|
This was referenced Sep 27, 2025
|
@zyw-bot csmith-quick-fuzz |
|
Maybe limit this to >2 cases to start to avoid regressions? |
dianqk
force-pushed
the
switch-wrapping-range
branch
from
bb9720c to
9864e17
Compare
This is still needed, some 2 cases have not been transformed: https://llvm.godbolt.org/z/Y71EovY4M. I think it's fine because it runs later in the passes pipeline. |
|
I improved this with only one case also. |
| // Max+1 is not equal to Min. It's not continuous in arithmetic. | ||
| if (Max == CR.getUnsignedMax() && Min == CR.getUnsignedMin()) { | ||
| assert(Cases.size() >= 2); | ||
| auto *It = |
There was a problem hiding this comment.
How about something as follows before the logic here?
// Find the first non-consecutive pair, and ensure this pair
// happens to be unique.
There was a problem hiding this comment.
(I meant this before the std::adjacent_find, which is not appearing)
There was a problem hiding this comment.
I think this is what you want. std::adjacent_find finds one non-consecutive pair and checks their distance to check if this pair is unique.
Comment on lines
5729
to
5732
| auto Min = Cases.back()->getValue(); | ||
| auto Max = Cases.front()->getValue(); | ||
| auto Offset = Max - Min; | ||
| auto ContiguousOffset = Cases.size() - 1; |
There was a problem hiding this comment.
No need for auto here, having the type explicit (const APInt&, unsigned) would make the code more readable.
Comment on lines
5751
to
5752
| auto *OtherMax = *It; | ||
| auto *OtherMin = *(It + 1); |
There was a problem hiding this comment.
Suggested change
| auto *OtherMax = *It; | |
| auto *OtherMin = *(It + 1); | |
| auto [OtherMax, OtherMin] = std::make_pair(*It, *std::next(It)); |
|
|
||
| static bool casesAreContiguous(SmallVectorImpl<ConstantInt *> &Cases) { | ||
| static bool casesAreContiguous(Value *Condition, | ||
| SmallVectorImpl<ConstantInt *> &Cases, |
There was a problem hiding this comment.
I think the function name would deserve an update, e.g., findContiguousCases (and maybe return a std::optional struct instead of the three values as out parameters)?
dtcxzyw
reviewed
Oct 4, 2025
Comment on lines
5874
to
5876
| if (auto Result = findContiguousCases(SI->getCondition(), CasesA, CasesB, | ||
| DestA, DestB)) | ||
| ContiguousCases = *Result; |
There was a problem hiding this comment.
Suggested change
| if (auto Result = findContiguousCases(SI->getCondition(), CasesA, CasesB, | |
| DestA, DestB)) | |
| ContiguousCases = *Result; | |
| ContiguousCases = findContiguousCases(SI->getCondition(), CasesA, CasesB, | |
| DestA, DestB); |
Comment on lines
5879
to
5881
| if (auto Result = findContiguousCases(SI->getCondition(), CasesB, CasesA, | ||
| DestB, DestA)) | ||
| ContiguousCases = *Result; |
There was a problem hiding this comment.
Suggested change
| if (auto Result = findContiguousCases(SI->getCondition(), CasesB, CasesA, | |
| DestB, DestA)) | |
| ContiguousCases = *Result; | |
| ContiguousCases = findContiguousCases(SI->getCondition(), CasesB, CasesA, | |
| DestB, DestA); |
|
|
||
| // Start building the compare and branch. | ||
| // Correctness: Cases to the default destination cannot be contiguous cases. | ||
| if (!ContiguousCases && !HasDefault && !CasesA.empty()) |
There was a problem hiding this comment.
Suggested change
| if (!ContiguousCases && !HasDefault && !CasesA.empty()) | |
| else if (!HasDefault) |
!HasDefault implies !CasesA.empty().
dtcxzyw
approved these changes
Oct 5, 2025
|
@zyw-bot csmith-fuzz |
arsenm
reviewed
Oct 6, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/169/builds/15667 Here is the relevant piece of the build log for the reference |
dtcxzyw
added a commit
that referenced
this pull request
Oct 6, 2025
This patch implements the todo discussed in #158495 (comment). It also fixes a regression introduced by #161000. See also dtcxzyw/llvm-opt-benchmark#2890 (comment). IR diff: dtcxzyw/llvm-opt-benchmark#2892
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Oct 6, 2025
This patch implements the todo discussed in llvm/llvm-project#158495 (comment). It also fixes a regression introduced by llvm/llvm-project#161000. See also dtcxzyw/llvm-opt-benchmark#2890 (comment). IR diff: dtcxzyw/llvm-opt-benchmark#2892
aokblast
pushed a commit
to aokblast/llvm-project
that referenced
this pull request
Oct 6, 2025
…1000) Fixes llvm#157113. Take the following IR as an example; we know the destination of the `[1, 3]` cases is `%else`. ```llvm define i32 @src(i8 range(i8 0, 6) %arg) { switch i8 %arg, label %else [ i8 0, label %if i8 4, label %if i8 5, label %if ] if: ret i32 0 else: ret i32 1 } ``` We can first try the non-wrapping range for both destinations, but I don't see how that would be any better. Proof: https://alive2.llvm.org/ce/z/acdWD4.
aokblast
pushed a commit
to aokblast/llvm-project
that referenced
this pull request
Oct 6, 2025
This patch implements the todo discussed in llvm#158495 (comment). It also fixes a regression introduced by llvm#161000. See also dtcxzyw/llvm-opt-benchmark#2890 (comment). IR diff: dtcxzyw/llvm-opt-benchmark#2892
|
I am seeing a "Broken module" error when building the Linux kernel with full LTO (it shows up for AArch64 / I was able to find two If there is any other information I can provide, I am more than happy to do so. |
Reproduced. I will provide a smaller IR reproducer. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #157113.
Take the following IR as an example; we know the destination of the
[1, 3]cases is%else.We can first try the non-wrapping range for both destinations, but I don't see how that would be any better.
Proof: https://alive2.llvm.org/ce/z/acdWD4.