squiid/llvm
1
0
Fork 0
Code Issues Pull requests Activity

[DSE,MemorySSA] Add an early check for read clobbers to traversal.

Browse source 
Depending on the benchmark, this early exit can save a substantial
amount of compile-time:

http://llvm-compile-time-tracker.com/compare.php?from=505f2d817aa8e07ba98e5fd4a8f6ff0666f89df1&to=eb4e441147f9b4b7a5fcbbc57428cadbe9e01f10&stat=instructions
This commit is contained in:
Florian Hahn 2020-09-07 22:52:10 +01:00
parent 5f5a0bb087
commit efb8e156da
2 changed files with 70 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
llvm/lib/Transforms/Scalar/DeadStoreElimination.cpp
View file

@ -1901,6 +1901,18 @@ struct DSEState {
return None;
}
// Quick check if there are direct uses that are read-clobbers.
if (any_of(Current->uses(), [this, &DefLoc, StartAccess](Use &U) {
if (auto *UseOrDef = dyn_cast<MemoryUseOrDef>(U.getUser()))
return !MSSA.dominates(StartAccess, UseOrDef) &&
isReadClobber(DefLoc, UseOrDef->getMemoryInst());
return false;
})) {
Cache.KnownReads.insert(Current);
LLVM_DEBUG(dbgs() << " ... found a read clobber\n");
return None;
}
// If Current cannot be analyzed or is not removable, check the next
// candidate.
if (!hasAnalyzableMemoryWrite(CurrentI, TLI) || !isRemovable(CurrentI)) {

58
llvm/test/Transforms/DeadStoreElimination/MSSA/read-clobber-after-overwrite.ll Normal file
View file

@ -0,0 +1,58 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -dse -enable-dse-memoryssa -S %s | FileCheck %s
declare i1 @cond() readnone
define i32 @test() {
; CHECK-LABEL: @test(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[M0:%.*]] = alloca [4 x i32], align 16
; CHECK-NEXT: br label [[LOOP_1:%.*]]
; CHECK: loop.1:
; CHECK-NEXT: br label [[LOOP_2:%.*]]
; CHECK: loop.2:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[LOOP_1]] ], [ [[IV_NEXT:%.*]], [[LOOP_2]] ]
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr inbounds [4 x i32], [4 x i32]* [[M0]], i64 3, i64 [[IV]]
; CHECK-NEXT: [[PTR_2:%.*]] = getelementptr inbounds [4 x i32], [4 x i32]* [[M0]], i64 0, i64 [[IV]]
; CHECK-NEXT: store i32 20, i32* [[PTR_2]], align 4
; CHECK-NEXT: store i32 30, i32* [[PTR_1]], align 4
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
; CHECK-NEXT: [[C_3:%.*]] = call i1 @cond()
; CHECK-NEXT: br i1 [[C_3]], label [[LOOP_1_LATCH:%.*]], label [[LOOP_2]]
; CHECK: loop.1.latch:
; CHECK-NEXT: [[C_2:%.*]] = call i1 @cond()
; CHECK-NEXT: br i1 [[C_2]], label [[EXIT:%.*]], label [[LOOP_1]]
; CHECK: exit:
; CHECK-NEXT: [[PTR_3:%.*]] = getelementptr inbounds [4 x i32], [4 x i32]* [[M0]], i64 0, i64 1
; CHECK-NEXT: [[LV:%.*]] = load i32, i32* [[PTR_3]], align 16
; CHECK-NEXT: ret i32 [[LV]]
;
entry:
%M0 = alloca [4 x i32], align 16
br label %loop.1
loop.1:
br label %loop.2
loop.2:
%iv = phi i64 [ 0, %loop.1 ], [ %iv.next, %loop.2 ]
%ptr.1 = getelementptr inbounds [4 x i32], [4 x i32]* %M0, i64 3, i64 %iv
store i32 10, i32* %ptr.1, align 4
%ptr.2 = getelementptr inbounds [4 x i32], [4 x i32]* %M0, i64 0, i64 %iv
store i32 20, i32* %ptr.2, align 4
store i32 30, i32* %ptr.1, align 4
%iv.next = add nuw nsw i64 %iv, 1
%c.3 = call i1 @cond()
br i1 %c.3, label %loop.1.latch, label %loop.2
loop.1.latch:
%c.2 = call i1 @cond()
br i1 %c.2, label %exit, label %loop.1
exit:
%ptr.3 = getelementptr inbounds [4 x i32], [4 x i32]* %M0, i64 0, i64 1
%lv = load i32, i32* %ptr.3, align 16
ret i32 %lv
}