[llvm] d440f9a - [LICM] Make capture check more precise

[Nikita Popov via llvm-commits]

Author: Nikita Popov
Date: 2021-04-19T20:34:23+02:00
New Revision: d440f9a326f06df067b26adf48df3d52e9a07b7b

URL: https://github.com/llvm/llvm-project/commit/d440f9a326f06df067b26adf48df3d52e9a07b7b
DIFF: https://github.com/llvm/llvm-project/commit/d440f9a326f06df067b26adf48df3d52e9a07b7b.diff

LOG: [LICM] Make capture check more precise

During store promotion, we check whether the pointer was captured
to exclude potential reads from other threads. However, we're only
interested in captures before or inside the loop. Check this using
PointerMayBeCapturedBefore against the loop header.

Differential Revision: https://reviews.llvm.org/D100706

Added: 
    

Modified: 
    llvm/lib/Transforms/Scalar/LICM.cpp
    llvm/test/Transforms/LICM/promote-capture.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Scalar/LICM.cpp b/llvm/lib/Transforms/Scalar/LICM.cpp
index a63318bc9535f..c2bc93c1187b8 100644
--- a/llvm/lib/Transforms/Scalar/LICM.cpp
+++ b/llvm/lib/Transforms/Scalar/LICM.cpp
@@ -1954,10 +1954,21 @@ class LoopPromoter : public LoadAndStorePromoter {
   }
 };
 
+bool isNotCapturedBeforeOrInLoop(const Value *V, const Loop *L,
+                                 DominatorTree *DT) {
+  // We can perform the captured-before check against any instruction in the
+  // loop header, as the loop header is reachable from any instruction inside
+  // the loop.
+  // TODO: ReturnCaptures=true shouldn't be necessary here.
+  return !PointerMayBeCapturedBefore(V, /* ReturnCaptures */ true,
+                                     /* StoreCaptures */ true,
+                                     L->getHeader()->getTerminator(), DT);
+}
 
 /// Return true iff we can prove that a caller of this function can not inspect
 /// the contents of the provided object in a well defined program.
-bool isKnownNonEscaping(Value *Object, const TargetLibraryInfo *TLI) {
+bool isKnownNonEscaping(Value *Object, const Loop *L,
+                        const TargetLibraryInfo *TLI, DominatorTree *DT) {
   if (isa<AllocaInst>(Object))
     // Since the alloca goes out of scope, we know the caller can't retain a
     // reference to it and be well defined.  Thus, we don't need to check for
@@ -1974,7 +1985,7 @@ bool isKnownNonEscaping(Value *Object, const TargetLibraryInfo *TLI) {
   //      weaker condition and handle only AllocLikeFunctions (which are
   //      known to be noalias).  TODO
   return isAllocLikeFn(Object, TLI) &&
-    !PointerMayBeCaptured(Object, true, true);
+         isNotCapturedBeforeOrInLoop(Object, L, DT);
 }
 
 } // namespace
@@ -2060,7 +2071,7 @@ bool llvm::promoteLoopAccessesToScalars(
     // this by proving that the caller can't have a reference to the object
     // after return and thus can't possibly load from the object.
     Value *Object = getUnderlyingObject(SomePtr);
-    if (!isKnownNonEscaping(Object, TLI))
+    if (!isKnownNonEscaping(Object, CurLoop, TLI, DT))
       return false;
     // Subtlety: Alloca's aren't visible to callers, but *are* potentially
     // visible to other threads if captured and used during their lifetimes.
@@ -2195,7 +2206,7 @@ bool llvm::promoteLoopAccessesToScalars(
       Value *Object = getUnderlyingObject(SomePtr);
       SafeToInsertStore =
           (isAllocLikeFn(Object, TLI) || isa<AllocaInst>(Object)) &&
-          !PointerMayBeCaptured(Object, true, true);
+          isNotCapturedBeforeOrInLoop(Object, CurLoop, DT);
     }
   }
 

diff  --git a/llvm/test/Transforms/LICM/promote-capture.ll b/llvm/test/Transforms/LICM/promote-capture.ll
index 9ece3c01edddb..1a2603d1c9866 100644
--- a/llvm/test/Transforms/LICM/promote-capture.ll
+++ b/llvm/test/Transforms/LICM/promote-capture.ll
@@ -9,21 +9,24 @@ define void @test_captured_after_loop(i32 %len) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COUNT:%.*]] = alloca i32, align 4
 ; CHECK-NEXT:    store i32 0, i32* [[COUNT]], align 4
+; CHECK-NEXT:    [[COUNT_PROMOTED:%.*]] = load i32, i32* [[COUNT]], align 4
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LATCH:%.*]] ]
+; CHECK-NEXT:    [[C_INC2:%.*]] = phi i32 [ [[COUNT_PROMOTED]], [[ENTRY:%.*]] ], [ [[C_INC1:%.*]], [[LATCH:%.*]] ]
+; CHECK-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[I_NEXT:%.*]], [[LATCH]] ]
 ; CHECK-NEXT:    [[COND:%.*]] = call i1 @cond(i32 [[I]])
 ; CHECK-NEXT:    br i1 [[COND]], label [[IF:%.*]], label [[LATCH]]
 ; CHECK:       if:
-; CHECK-NEXT:    [[C:%.*]] = load i32, i32* [[COUNT]], align 4
-; CHECK-NEXT:    [[C_INC:%.*]] = add i32 [[C]], 1
-; CHECK-NEXT:    store i32 [[C_INC]], i32* [[COUNT]], align 4
+; CHECK-NEXT:    [[C_INC:%.*]] = add i32 [[C_INC2]], 1
 ; CHECK-NEXT:    br label [[LATCH]]
 ; CHECK:       latch:
+; CHECK-NEXT:    [[C_INC1]] = phi i32 [ [[C_INC]], [[IF]] ], [ [[C_INC2]], [[LOOP]] ]
 ; CHECK-NEXT:    [[I_NEXT]] = add nuw i32 [[I]], 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[I_NEXT]], [[LEN:%.*]]
 ; CHECK-NEXT:    br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
 ; CHECK:       exit:
+; CHECK-NEXT:    [[C_INC1_LCSSA:%.*]] = phi i32 [ [[C_INC1]], [[LATCH]] ]
+; CHECK-NEXT:    store i32 [[C_INC1_LCSSA]], i32* [[COUNT]], align 4
 ; CHECK-NEXT:    call void @capture(i32* [[COUNT]])
 ; CHECK-NEXT:    ret void
 ;