[llvm] 4ab77d1 - [LICM] Allow promotion with non-load/store users

[Nikita Popov via llvm-commits]

Author: Nikita Popov
Date: 2022-09-09T13:09:59+02:00
New Revision: 4ab77d16776ac819dfa1de3b5dd56dc6da036108

URL: https://github.com/llvm/llvm-project/commit/4ab77d16776ac819dfa1de3b5dd56dc6da036108
DIFF: https://github.com/llvm/llvm-project/commit/4ab77d16776ac819dfa1de3b5dd56dc6da036108.diff

LOG: [LICM] Allow promotion with non-load/store users

If there are non-load/store users of the promoted pointer, we
currently abort promotion. However, having such users isn't really
relevant to the transform. We already separately check that a)
there are no instructions that modref the promoted pointer and
b) that a pointer capture disables store promotion.

In the affected @test_captured_in_loop test case we have a readnone
capture of the promoted pointer, which means that load promotion
can be performed (while store promotion cannot).

Differential Revision: https://reviews.llvm.org/D133485

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 29e14d8dfe0b4..48958dc55ed0d 100644
--- a/llvm/lib/Transforms/Scalar/LICM.cpp
+++ b/llvm/lib/Transforms/Scalar/LICM.cpp
@@ -2093,7 +2093,7 @@ bool llvm::promoteLoopAccessesToScalars(
               Store->getAlign(), MDL, Preheader->getTerminator(), DT, TLI);
         }
       } else
-        return false; // Not a load or store.
+        continue; // Not a load or store.
 
       if (!AccessTy)
         AccessTy = getLoadStoreType(UI);

diff  --git a/llvm/test/Transforms/LICM/promote-capture.ll b/llvm/test/Transforms/LICM/promote-capture.ll
index a409a492ee543..7d4e5868ea525 100644
--- a/llvm/test/Transforms/LICM/promote-capture.ll
+++ b/llvm/test/Transforms/LICM/promote-capture.ll
@@ -61,18 +61,20 @@ define void @test_captured_in_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:    [[C_INC:%.*]] = add i32 [[C_INC2]], 1
 ; CHECK-NEXT:    store i32 [[C_INC]], i32* [[COUNT]], align 4
 ; CHECK-NEXT:    call void @capture(i32* [[COUNT]])
 ; 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]]