MemDepResult getCallSiteDependencyFrom(CallSite C, bool isReadOnlyCall,
BasicBlock::iterator ScanIt,
BasicBlock *BB);
- bool getNonLocalPointerDepFromBB(const PHITransAddr &Pointer,
+ bool getNonLocalPointerDepFromBB(Instruction *QueryInst,
+ const PHITransAddr &Pointer,
const AliasAnalysis::Location &Loc,
bool isLoad, BasicBlock *BB,
SmallVectorImpl<NonLocalDepResult> &Result,
DenseMap<BasicBlock*, Value*> &Visited,
bool SkipFirstBlock = false);
- MemDepResult GetNonLocalInfoForBlock(const AliasAnalysis::Location &Loc,
+ MemDepResult GetNonLocalInfoForBlock(Instruction *QueryInst,
+ const AliasAnalysis::Location &Loc,
bool isLoad, BasicBlock *BB,
NonLocalDepInfo *Cache,
unsigned NumSortedEntries);
// a block with multiple different pointers. This can happen during PHI
// translation.
DenseMap<BasicBlock*, Value*> Visited;
- if (!getNonLocalPointerDepFromBB(Address, Loc, isLoad, FromBB,
+ if (!getNonLocalPointerDepFromBB(QueryInst, Address, Loc, isLoad, FromBB,
Result, Visited, true))
return;
Result.clear();
/// lookup (which may use dirty cache info if available). If we do a lookup,
/// add the result to the cache.
MemDepResult MemoryDependenceAnalysis::
-GetNonLocalInfoForBlock(const AliasAnalysis::Location &Loc,
+GetNonLocalInfoForBlock(Instruction *QueryInst,
+ const AliasAnalysis::Location &Loc,
bool isLoad, BasicBlock *BB,
NonLocalDepInfo *Cache, unsigned NumSortedEntries) {
}
// Scan the block for the dependency.
- MemDepResult Dep = getPointerDependencyFrom(Loc, isLoad, ScanPos, BB);
+ MemDepResult Dep = getPointerDependencyFrom(Loc, isLoad, ScanPos, BB,
+ QueryInst);
// If we had a dirty entry for the block, update it. Otherwise, just add
// a new entry.
/// not compute dependence information for some reason. This should be treated
/// as a clobber dependence on the first instruction in the predecessor block.
bool MemoryDependenceAnalysis::
-getNonLocalPointerDepFromBB(const PHITransAddr &Pointer,
+getNonLocalPointerDepFromBB(Instruction *QueryInst,
+ const PHITransAddr &Pointer,
const AliasAnalysis::Location &Loc,
bool isLoad, BasicBlock *StartBB,
SmallVectorImpl<NonLocalDepResult> &Result,
} else if (CacheInfo->Size > Loc.Size) {
// This query's Size is less than the cached one. Conservatively restart
// the query using the greater size.
- return getNonLocalPointerDepFromBB(Pointer,
+ return getNonLocalPointerDepFromBB(QueryInst, Pointer,
Loc.getWithNewSize(CacheInfo->Size),
isLoad, StartBB, Result, Visited,
SkipFirstBlock);
CacheInfo->NonLocalDeps.clear();
}
if (Loc.AATags)
- return getNonLocalPointerDepFromBB(Pointer, Loc.getWithoutAATags(),
+ return getNonLocalPointerDepFromBB(QueryInst,
+ Pointer, Loc.getWithoutAATags(),
isLoad, StartBB, Result, Visited,
SkipFirstBlock);
}
// Get the dependency info for Pointer in BB. If we have cached
// information, we will use it, otherwise we compute it.
DEBUG(AssertSorted(*Cache, NumSortedEntries));
- MemDepResult Dep = GetNonLocalInfoForBlock(Loc, isLoad, BB, Cache,
+ MemDepResult Dep = GetNonLocalInfoForBlock(QueryInst,
+ Loc, isLoad, BB, Cache,
NumSortedEntries);
// If we got a Def or Clobber, add this to the list of results.
// result conflicted with the Visited list; we have to conservatively
// assume it is unknown, but this also does not block PRE of the load.
if (!CanTranslate ||
- getNonLocalPointerDepFromBB(PredPointer,
+ getNonLocalPointerDepFromBB(QueryInst, PredPointer,
Loc.getWithNewPtr(PredPtrVal),
isLoad, Pred,
Result, Visited)) {
ret i32 %add
}
+; With the invariant.load metadata, what would otherwise
+; be a case for PRE becomes a full redundancy.
+define i32 @test3(i1 %cnd, i32* %p, i32* %q) {
+; CHECK-LABEL: test3
+; CHECK-NOT: load
+entry:
+ %v1 = load i32* %p
+ br i1 %cnd, label %bb1, label %bb2
+
+bb1:
+ store i32 5, i32* %q
+ br label %bb2
+
+bb2:
+ %v2 = load i32* %p, !invariant.load !0
+ %res = sub i32 %v1, %v2
+ ret i32 %res
+}
+
+; This test is here to document a case which doesn't optimize
+; as well as it could.
+define i32 @test4(i1 %cnd, i32* %p, i32* %q) {
+; CHECK-LABEL: test4
+; %v2 is redundant, but GVN currently doesn't catch that
+entry:
+ %v1 = load i32* %p, !invariant.load !0
+ br i1 %cnd, label %bb1, label %bb2
+
+bb1:
+ store i32 5, i32* %q
+ br label %bb2
+
+bb2:
+ %v2 = load i32* %p
+ %res = sub i32 %v1, %v2
+ ret i32 %res
+}
+
!0 = !{ }
-; RUN: opt -basicaa -gvn -S < %s | FileCheck %s
+; RUN: opt -tbaa -basicaa -gvn -S < %s | FileCheck %s
define i32 @test1(i8* %p, i8* %q) {
; CHECK: @test1(i8* %p, i8* %q)
ret i32 %c
}
+
+
+define i32 @test8(i32* %p, i32* %q) {
+; CHECK-LABEL: test8
+; CHECK-NEXT: store i32 15, i32* %p
+; CHECK-NEXT: ret i32 0
+; Since we know the location is invariant, we can forward the
+; load across the potentially aliasing store.
+
+ %a = load i32* %q, !tbaa !10
+ store i32 15, i32* %p
+ %b = load i32* %q, !tbaa !10
+ %c = sub i32 %a, %b
+ ret i32 %c
+}
+define i32 @test9(i32* %p, i32* %q) {
+; CHECK-LABEL: test9
+; CHECK-NEXT: call void @clobber()
+; CHECK-NEXT: ret i32 0
+; Since we know the location is invariant, we can forward the
+; load across the potentially aliasing store (within the call).
+
+ %a = load i32* %q, !tbaa !10
+ call void @clobber()
+ %b = load i32* %q, !tbaa !10
+ %c = sub i32 %a, %b
+ ret i32 %c
+}
+
+
+declare void @clobber()
declare i32 @foo(i8*) readonly
; CHECK: [[TAGC]] = !{[[TYPEC:!.*]], [[TYPEC]], i64 0}
!6 = !{!"A", !2}
!7 = !{!"B", !6}
!8 = !{!"another root", null}
+
+
+;; A TBAA structure who's only point is to have a constant location
+!9 = !{!"yet another root"}
+!10 = !{!"node", !9, i64 1}
+