Skip to content

[SCEVExpander] Clear flags when reusing GEP #109293

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Oct 1, 2024
Merged

[SCEVExpander] Clear flags when reusing GEP #109293

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
054a3c5
[SCEVExpander] Clear flags when reusing GEP
nikic Sep 19, 2024
4c37dc3
add unit test
nikic Oct 1, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,7 @@ struct PoisonFlags {
unsigned Exact : 1;
unsigned Disjoint : 1;
unsigned NNeg : 1;
GEPNoWrapFlags GEPNW;

PoisonFlags(const Instruction *I);
void apply(Instruction *I);
Expand Down
19 changes: 14 additions & 5 deletions llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,7 @@ PoisonFlags::PoisonFlags(const Instruction *I) {
Exact = false;
Disjoint = false;
NNeg = false;
GEPNW = GEPNoWrapFlags::none();
if (auto *OBO = dyn_cast<OverflowingBinaryOperator>(I)) {
NUW = OBO->hasNoUnsignedWrap();
NSW = OBO->hasNoSignedWrap();
Expand All @@ -63,6 +64,8 @@ PoisonFlags::PoisonFlags(const Instruction *I) {
NUW = TI->hasNoUnsignedWrap();
NSW = TI->hasNoSignedWrap();
}
if (auto *GEP = dyn_cast<GetElementPtrInst>(I))
GEPNW = GEP->getNoWrapFlags();
}

void PoisonFlags::apply(Instruction *I) {
Expand All @@ -80,6 +83,8 @@ void PoisonFlags::apply(Instruction *I) {
I->setHasNoUnsignedWrap(NUW);
I->setHasNoSignedWrap(NSW);
}
if (auto *GEP = dyn_cast<GetElementPtrInst>(I))
GEP->setNoWrapFlags(GEPNW);
}

/// ReuseOrCreateCast - Arrange for there to be a cast of V to Ty at IP,
Expand Down Expand Up @@ -370,11 +375,15 @@ Value *SCEVExpander::expandAddToGEP(const SCEV *Offset, Value *V) {
// generated code.
if (isa<DbgInfoIntrinsic>(IP))
ScanLimit++;
if (IP->getOpcode() == Instruction::GetElementPtr &&
IP->getOperand(0) == V && IP->getOperand(1) == Idx &&
cast<GEPOperator>(&*IP)->getSourceElementType() ==
Builder.getInt8Ty())
return &*IP;
if (auto *GEP = dyn_cast<GetElementPtrInst>(IP)) {
if (GEP->getPointerOperand() == V &&
GEP->getSourceElementType() == Builder.getInt8Ty() &&
GEP->getOperand(1) == Idx) {
rememberFlags(GEP);
GEP->setNoWrapFlags(GEPNoWrapFlags::none());
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In #102133 this will become an intersect with SCEV nuw flag instead.

return &*IP;
}
}
if (IP == BlockBegin) break;
}
}
Expand Down
20 changes: 12 additions & 8 deletions llvm/test/CodeGen/WebAssembly/simd-shift-in-loop.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,19 +9,21 @@ target triple = "wasm32-unknown-unknown"
define void @shl_loop(ptr %a, i8 %shift, i32 %count) {
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A bit unfortunate that this regresses, but I think we'll recover this in the future (after #102133 + inference for gep nuw in instcombine, at which point we should be able to preserve the nuw flag here, which is all wasm cares about).

; CHECK-LABEL: shl_loop:
; CHECK: .functype shl_loop (i32, i32, i32) -> ()
; CHECK-NEXT: .local i32
; CHECK-NEXT: # %bb.0: # %entry
; CHECK-NEXT: .LBB0_1: # %body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
; CHECK-NEXT: loop # label0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: i32.add
; CHECK-NEXT: local.tee 3
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.load 0:p2align=0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.shl
; CHECK-NEXT: v128.store 16
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: i32.add
; CHECK-NEXT: v128.store 0
; CHECK-NEXT: local.get 3
; CHECK-NEXT: local.set 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i32.const -1
Expand Down Expand Up @@ -56,23 +58,25 @@ exit:
define void @shl_phi_loop(ptr %a, i8 %shift, i32 %count) {
; CHECK-LABEL: shl_phi_loop:
; CHECK: .functype shl_phi_loop (i32, i32, i32) -> ()
; CHECK-NEXT: .local i32
; CHECK-NEXT: # %bb.0: # %entry
; CHECK-NEXT: .LBB1_1: # %body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
; CHECK-NEXT: loop # label1:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: i32.add
; CHECK-NEXT: local.tee 3
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.load 0:p2align=0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.shl
; CHECK-NEXT: v128.store 16
; CHECK-NEXT: v128.store 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i32.const 1
; CHECK-NEXT: i32.and
; CHECK-NEXT: local.set 1
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: i32.add
; CHECK-NEXT: local.get 3
; CHECK-NEXT: local.set 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i32.const -1
Expand Down
8 changes: 4 additions & 4 deletions llvm/test/Transforms/LoopStrengthReduce/AMDGPU/preserve-addrspace-assert.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,11 +79,11 @@ define void @lsr_crash_preserve_addrspace_unknown_type2(ptr addrspace(5) %array,
; CHECK-NEXT: br label %[[FOR_BODY:.*]]
; CHECK: [[FOR_BODY]]:
; CHECK-NEXT: [[J:%.*]] = phi i32 [ [[ADD:%.*]], %[[FOR_INC:.*]] ], [ 0, %[[ENTRY]] ]
; CHECK-NEXT: [[IDX:%.*]] = getelementptr inbounds i8, ptr addrspace(5) [[ARRAY]], i32 [[J]]
; CHECK-NEXT: [[IDX1:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[ARRAY2]], i32 [[J]]
; CHECK-NEXT: [[T:%.*]] = getelementptr inbounds i8, ptr addrspace(5) [[ARRAY]], i32 [[J]]
; CHECK-NEXT: [[IDX:%.*]] = getelementptr i8, ptr addrspace(5) [[ARRAY]], i32 [[J]]
; CHECK-NEXT: [[IDX1:%.*]] = getelementptr i8, ptr addrspace(3) [[ARRAY2]], i32 [[J]]
; CHECK-NEXT: [[T:%.*]] = getelementptr i8, ptr addrspace(5) [[ARRAY]], i32 [[J]]
; CHECK-NEXT: [[N8:%.*]] = load i8, ptr addrspace(5) [[T]], align 4
; CHECK-NEXT: [[N7:%.*]] = getelementptr inbounds i8, ptr addrspace(5) [[T]], i32 42
; CHECK-NEXT: [[N7:%.*]] = getelementptr i8, ptr addrspace(5) [[T]], i32 42
; CHECK-NEXT: [[N9:%.*]] = load i8, ptr addrspace(5) [[N7]], align 4
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[J]], 42
; CHECK-NEXT: br i1 [[CMP]], label %[[IF_THEN17:.*]], label %[[FOR_INC]]
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/Transforms/LoopStrengthReduce/ARM/illegal-addr-modes.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ define ptr @negativeOneCase(ptr returned %a, ptr nocapture readonly %b, i32 %n)
; CHECK-NEXT: br label [[WHILE_COND:%.*]]
; CHECK: while.cond:
; CHECK-NEXT: [[P_0:%.*]] = phi ptr [ [[ADD_PTR]], [[ENTRY:%.*]] ], [ [[INCDEC_PTR:%.*]], [[WHILE_COND]] ]
; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i8, ptr [[P_0]], i32 1
; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr i8, ptr [[P_0]], i32 1
; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[INCDEC_PTR]], align 1
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i8 [[TMP0]], 0
; CHECK-NEXT: br i1 [[CMP]], label [[WHILE_COND2_PREHEADER:%.*]], label [[WHILE_COND]]
Expand Down
32 changes: 32 additions & 0 deletions llvm/unittests/Transforms/Utils/ScalarEvolutionExpanderTest.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -954,4 +954,36 @@ TEST_F(ScalarEvolutionExpanderTest, ExpandNonIntegralPtrWithNullBase) {
});
}

TEST_F(ScalarEvolutionExpanderTest, GEPFlags) {
LLVMContext C;
SMDiagnostic Err;
StringRef ModStr = R"(
define void @f(ptr %p, i64 %x) {
%gep_inbounds = getelementptr inbounds i8, ptr %p, i64 %x
ret void
})";
std::unique_ptr<Module> M = parseAssemblyString(ModStr, Err, C);

assert(M && "Could not parse module?");
assert(!verifyModule(*M) && "Must have been well formed!");

Function *F = M->getFunction("f");
ASSERT_NE(F, nullptr) << "Could not find function 'f'";
BasicBlock &Entry = F->getEntryBlock();
auto *GEP = cast<GetElementPtrInst>(&Entry.front());

ScalarEvolution SE = buildSE(*F);
const SCEV *Ptr = SE.getSCEV(F->getArg(0));
const SCEV *X = SE.getSCEV(F->getArg(1));
const SCEV *PtrX = SE.getAddExpr(Ptr, X);

SCEVExpander Exp(SE, M->getDataLayout(), "expander");
auto *I = cast<Instruction>(
Exp.expandCodeFor(PtrX, nullptr, Entry.getTerminator()));
// Check that the GEP is reused, but the inbounds flag cleared. We don't
// know that the newly introduced use is inbounds.
EXPECT_EQ(I, GEP);
EXPECT_EQ(GEP->getNoWrapFlags(), GEPNoWrapFlags::none());
}

} // end namespace llvm
Loading