[llvm] 0776c48 - Recommit "[LICM] Only create load in ph when promoting load or store doesn't exec."
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Sun May 29 14:00:29 PDT 2022
Author: Florian Hahn
Date: 2022-05-29T21:57:14+01:00
New Revision: 0776c48f9b7e69fa447bee57c7c0985caa856be9
URL: https://github.com/llvm/llvm-project/commit/0776c48f9b7e69fa447bee57c7c0985caa856be9
DIFF: https://github.com/llvm/llvm-project/commit/0776c48f9b7e69fa447bee57c7c0985caa856be9.diff
LOG: Recommit "[LICM] Only create load in ph when promoting load or store doesn't exec."
This reverts the revert commit ad95255b9215a.
The updated version also creates a load when the store may not execute.
In those cases, we still need to introduce a load in a function where
there may not have been one before, so this doesn't completely resolve
issue #51248.
Original message:
When only a store is sunk, there is no need to create a load in the
pre-header, as the result of the load will never get used.
The dead load can can introduce UB, if the function is marked as
writeonly.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D123473
Added:
Modified:
llvm/lib/Transforms/Scalar/LICM.cpp
llvm/test/Transforms/LICM/scalar-promote.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Scalar/LICM.cpp b/llvm/lib/Transforms/Scalar/LICM.cpp
index 74ae5977a3663..da079e42a8c49 100644
--- a/llvm/lib/Transforms/Scalar/LICM.cpp
+++ b/llvm/lib/Transforms/Scalar/LICM.cpp
@@ -1959,6 +1959,7 @@ bool llvm::promoteLoopAccessesToScalars(
bool DereferenceableInPH = false;
bool SafeToInsertStore = false;
+ bool StoreIsGuanteedToExecute = false;
bool FoundLoadToPromote = false;
SmallVector<Instruction *, 64> LoopUses;
@@ -2039,10 +2040,12 @@ bool llvm::promoteLoopAccessesToScalars(
// alignment than any other guaranteed stores, in which case we can
// raise the alignment on the promoted store.
Align InstAlignment = Store->getAlign();
-
+ bool GuaranteedToExecute =
+ SafetyInfo->isGuaranteedToExecute(*UI, DT, CurLoop);
+ StoreIsGuanteedToExecute |= GuaranteedToExecute;
if (!DereferenceableInPH || !SafeToInsertStore ||
(InstAlignment > Alignment)) {
- if (SafetyInfo->isGuaranteedToExecute(*UI, DT, CurLoop)) {
+ if (GuaranteedToExecute) {
DereferenceableInPH = true;
SafeToInsertStore = true;
Alignment = std::max(Alignment, InstAlignment);
@@ -2156,21 +2159,26 @@ bool llvm::promoteLoopAccessesToScalars(
// Set up the preheader to have a definition of the value. It is the live-out
// value from the preheader that uses in the loop will use.
- LoadInst *PreheaderLoad = new LoadInst(
- AccessTy, SomePtr, SomePtr->getName() + ".promoted",
- Preheader->getTerminator());
- if (SawUnorderedAtomic)
- PreheaderLoad->setOrdering(AtomicOrdering::Unordered);
- PreheaderLoad->setAlignment(Alignment);
- PreheaderLoad->setDebugLoc(DebugLoc());
- if (AATags)
- PreheaderLoad->setAAMetadata(AATags);
- SSA.AddAvailableValue(Preheader, PreheaderLoad);
-
- MemoryAccess *PreheaderLoadMemoryAccess = MSSAU.createMemoryAccessInBB(
- PreheaderLoad, nullptr, PreheaderLoad->getParent(), MemorySSA::End);
- MemoryUse *NewMemUse = cast<MemoryUse>(PreheaderLoadMemoryAccess);
- MSSAU.insertUse(NewMemUse, /*RenameUses=*/true);
+ LoadInst *PreheaderLoad = nullptr;
+ if (FoundLoadToPromote || !StoreIsGuanteedToExecute) {
+ PreheaderLoad =
+ new LoadInst(AccessTy, SomePtr, SomePtr->getName() + ".promoted",
+ Preheader->getTerminator());
+ if (SawUnorderedAtomic)
+ PreheaderLoad->setOrdering(AtomicOrdering::Unordered);
+ PreheaderLoad->setAlignment(Alignment);
+ PreheaderLoad->setDebugLoc(DebugLoc());
+ if (AATags)
+ PreheaderLoad->setAAMetadata(AATags);
+
+ MemoryAccess *PreheaderLoadMemoryAccess = MSSAU.createMemoryAccessInBB(
+ PreheaderLoad, nullptr, PreheaderLoad->getParent(), MemorySSA::End);
+ MemoryUse *NewMemUse = cast<MemoryUse>(PreheaderLoadMemoryAccess);
+ MSSAU.insertUse(NewMemUse, /*RenameUses=*/true);
+ SSA.AddAvailableValue(Preheader, PreheaderLoad);
+ } else {
+ SSA.AddAvailableValue(Preheader, PoisonValue::get(AccessTy));
+ }
if (VerifyMemorySSA)
MSSAU.getMemorySSA()->verifyMemorySSA();
@@ -2181,7 +2189,7 @@ bool llvm::promoteLoopAccessesToScalars(
if (VerifyMemorySSA)
MSSAU.getMemorySSA()->verifyMemorySSA();
// If the SSAUpdater didn't use the load in the preheader, just zap it now.
- if (PreheaderLoad->use_empty())
+ if (PreheaderLoad && PreheaderLoad->use_empty())
eraseInstruction(*PreheaderLoad, *SafetyInfo, MSSAU);
return true;
diff --git a/llvm/test/Transforms/LICM/scalar-promote.ll b/llvm/test/Transforms/LICM/scalar-promote.ll
index ba274835d05f8..8304479f4f238 100644
--- a/llvm/test/Transforms/LICM/scalar-promote.ll
+++ b/llvm/test/Transforms/LICM/scalar-promote.ll
@@ -644,10 +644,9 @@ define void @test_sink_store_only() writeonly {
; CHECK: Function Attrs: writeonly
; CHECK-LABEL: @test_sink_store_only(
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[GLB_PROMOTED:%.*]] = load i8, i8* @glb, align 1
; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
; CHECK: loop.header:
-; CHECK-NEXT: [[DIV1:%.*]] = phi i8 [ [[GLB_PROMOTED]], [[ENTRY:%.*]] ], [ [[DIV:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT: [[DIV1:%.*]] = phi i8 [ poison, [[ENTRY:%.*]] ], [ [[DIV:%.*]], [[LOOP_LATCH:%.*]] ]
; CHECK-NEXT: [[I:%.*]] = phi i8 [ 0, [[ENTRY]] ], [ [[ADD:%.*]], [[LOOP_LATCH]] ]
; CHECK-NEXT: [[CMP:%.*]] = icmp ult i8 [[I]], 4
; CHECK-NEXT: br i1 [[CMP]], label [[LOOP_LATCH]], label [[EXIT:%.*]]
@@ -683,10 +682,9 @@ define void @test_sink_store_to_local_object_only_loop_must_execute() writeonly
; CHECK-LABEL: @test_sink_store_to_local_object_only_loop_must_execute(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[A:%.*]] = alloca i8, align 1
-; CHECK-NEXT: [[A_PROMOTED:%.*]] = load i8, i8* [[A]], align 1
; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
; CHECK: loop.header:
-; CHECK-NEXT: [[DIV1:%.*]] = phi i8 [ [[A_PROMOTED]], [[ENTRY:%.*]] ], [ [[DIV:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT: [[DIV1:%.*]] = phi i8 [ poison, [[ENTRY:%.*]] ], [ [[DIV:%.*]], [[LOOP_LATCH:%.*]] ]
; CHECK-NEXT: [[I:%.*]] = phi i8 [ 0, [[ENTRY]] ], [ [[ADD:%.*]], [[LOOP_LATCH]] ]
; CHECK-NEXT: [[CMP:%.*]] = icmp ult i8 [[I]], 4
; CHECK-NEXT: br i1 [[CMP]], label [[LOOP_LATCH]], label [[EXIT:%.*]]
@@ -842,17 +840,16 @@ define void @sink_store_lcssa_phis(i32* %ptr, i1 %c) {
; CHECK: loop.2.header:
; CHECK-NEXT: br i1 false, label [[LOOP_3_HEADER_PREHEADER:%.*]], label [[LOOP_1_LATCH:%.*]]
; CHECK: loop.3.header.preheader:
-; CHECK-NEXT: [[PTR_PROMOTED:%.*]] = load i32, i32* [[PTR:%.*]], align 4
; CHECK-NEXT: br label [[LOOP_3_HEADER:%.*]]
; CHECK: loop.3.header:
-; CHECK-NEXT: [[I_11:%.*]] = phi i32 [ [[I_1:%.*]], [[LOOP_3_LATCH:%.*]] ], [ [[PTR_PROMOTED]], [[LOOP_3_HEADER_PREHEADER]] ]
+; CHECK-NEXT: [[I_11:%.*]] = phi i32 [ [[I_1:%.*]], [[LOOP_3_LATCH:%.*]] ], [ poison, [[LOOP_3_HEADER_PREHEADER]] ]
; CHECK-NEXT: [[I_1]] = phi i32 [ 1, [[LOOP_3_LATCH]] ], [ 0, [[LOOP_3_HEADER_PREHEADER]] ]
; CHECK-NEXT: br i1 true, label [[LOOP_3_LATCH]], label [[LOOP_2_LATCH:%.*]]
; CHECK: loop.3.latch:
; CHECK-NEXT: br label [[LOOP_3_HEADER]]
; CHECK: loop.2.latch:
; CHECK-NEXT: [[I_11_LCSSA:%.*]] = phi i32 [ [[I_11]], [[LOOP_3_HEADER]] ]
-; CHECK-NEXT: store i32 [[I_11_LCSSA]], i32* [[PTR]], align 4
+; CHECK-NEXT: store i32 [[I_11_LCSSA]], i32* [[PTR:%.*]], align 4
; CHECK-NEXT: br label [[LOOP_2_HEADER]]
; CHECK: loop.1.latch:
; CHECK-NEXT: br i1 [[C:%.*]], label [[LOOP_1_HEADER]], label [[EXIT:%.*]]
More information about the llvm-commits
mailing list