Skip to content

[TapirTaskInfo] Modify check for promotable allocas to prevent optimi… #354

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: dev/19.x
Choose a base branch
from
Open

[TapirTaskInfo] Modify check for promotable allocas to prevent optimi… #354

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 13 additions & 0 deletions llvm/lib/Analysis/TapirTaskInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1168,6 +1168,11 @@ static void computeLiveInBlocks(
if (SI->getOperand(1) != AI)
continue;

if (SI->isAtomic() && SI->getOrdering() != AtomicOrdering::Unordered)
// Treat this atomic store as a use that makes the alloca live-in
// here.
break;

// We found a store to the alloca before a load. The alloca is not
// actually live-in here.
LiveInBlockWorklist[i] = LiveInBlockWorklist.back();
Expand Down Expand Up @@ -1257,9 +1262,17 @@ bool TaskInfo::isAllocaParallelPromotable(const AllocaInst *AIP) const {
if (StoreInst *SI = dyn_cast<StoreInst>(User)) {
// Remember the basic blocks which define new values for the alloca
DefBlocks.insert(SI->getParent());
if (SI->isAtomic() && SI->getOrdering() != AtomicOrdering::Unordered)
// Treat this atomic store as a use, to avoid removing uses of atomics
// to synchronize tasks.
UsingBlocks.push_back(SI->getParent());
} else if (LoadInst *LI = dyn_cast<LoadInst>(User)) {
// Otherwise it must be a load instruction, keep track of variable reads.
UsingBlocks.push_back(LI->getParent());
if (LI->isAtomic() && LI->getOrdering() != AtomicOrdering::Unordered)
// Treat this atomic load as a def, to avoid removing uses of atomics
// to synchronize tasks.
DefBlocks.insert(LI->getParent());
} else continue;

if (OnlyUsedInOneSpindle)
Expand Down
327 changes: 327 additions & 0 deletions llvm/test/Transforms/Tapir/spawn-while-flag.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,327 @@
; Check that a spawned task can spin on an atomic flag that is set in the continuation.
;
; Optimizing away this loop and the continuation are technically allowed by the serial projection,
; but it's convenient to allow this use of atomics.
;
; RUN: opt < %s -passes="default<O1>" -S | FileCheck %s
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

%"struct.std::atomic" = type { %"struct.std::__atomic_base" }
%"struct.std::__atomic_base" = type { i8 }
%class.anon = type { i8 }

$_ZNSt6atomicIbE5storeEbSt12memory_order = comdat any

$_ZNKSt6atomicIbE4loadESt12memory_order = comdat any

$_ZNKSt13__atomic_baseIbE4loadESt12memory_order = comdat any

$_ZStanSt12memory_orderSt23__memory_order_modifier = comdat any

$_ZNSt13__atomic_baseIbE5storeEbSt12memory_order = comdat any

; Function Attrs: mustprogress uwtable
define dso_local void @_Z3foov() #0 {
entry:
%done_flag = alloca %"struct.std::atomic", align 1
%syncreg = call token @llvm.syncregion.start()
call void @llvm.lifetime.start.p0(i64 1, ptr %done_flag) #9
call void @llvm.memset.p0.i64(ptr align 1 %done_flag, i8 0, i64 1, i1 false)
%0 = call token @llvm.taskframe.create()
%ref.tmp = alloca %class.anon, align 1
call void @llvm.lifetime.start.p0(i64 1, ptr %ref.tmp) #9
detach within %syncreg, label %det.achd, label %det.cont

det.achd: ; preds = %entry
call void @llvm.taskframe.use(token %0)
call void @"_ZZ3foovENK3$_0clERSt6atomicIbE"(ptr noundef nonnull align 1 dereferenceable(1) %ref.tmp, ptr noundef nonnull align 1 dereferenceable(1) %done_flag)
call void @llvm.lifetime.end.p0(i64 1, ptr %ref.tmp) #9
reattach within %syncreg, label %det.cont

det.cont: ; preds = %det.achd, %entry
call void @_Z3bazv()
call void @_ZNSt6atomicIbE5storeEbSt12memory_order(ptr noundef nonnull align 1 dereferenceable(1) %done_flag, i1 noundef zeroext true, i32 noundef 3) #9
sync within %syncreg, label %sync.continue

sync.continue: ; preds = %det.cont
call void @llvm.sync.unwind(token %syncreg)
call void @llvm.lifetime.end.p0(i64 1, ptr %done_flag) #9
ret void
}

; CHECK: define {{.*}}void @_Z3foov()
; CHECK: detach within %[[SYNCREG:.+]], label %[[DETACHED:.+]], label %[[CONTINUE:.+]]
; CHECK: [[DETACHED]]:
; CHECK: call void @_Z3barv()
; CHECK: reattach within %[[SYNCREG]], label %[[CONTINUE]]
; CHECK: [[CONTINUE]]:
; CHECK: call void @_Z3bazv()
; CHECK: sync within %[[SYNCREG]],

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #1

; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: write)
declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg) #2

; Function Attrs: nounwind willreturn memory(argmem: readwrite)
declare token @llvm.syncregion.start() #3

; Function Attrs: nounwind willreturn memory(argmem: readwrite)
declare token @llvm.taskframe.create() #3

; Function Attrs: inlinehint mustprogress uwtable
define internal void @"_ZZ3foovENK3$_0clERSt6atomicIbE"(ptr noundef nonnull align 1 dereferenceable(1) %this, ptr noundef nonnull align 1 dereferenceable(1) %flag) #4 align 2 {
entry:
%this.addr = alloca ptr, align 8
%flag.addr = alloca ptr, align 8
store ptr %this, ptr %this.addr, align 8, !tbaa !5
store ptr %flag, ptr %flag.addr, align 8, !tbaa !5
%this1 = load ptr, ptr %this.addr, align 8
br label %while.cond

while.cond: ; preds = %while.body, %entry
%0 = load ptr, ptr %flag.addr, align 8, !tbaa !5
%call = call noundef zeroext i1 @_ZNKSt6atomicIbE4loadESt12memory_order(ptr noundef nonnull align 1 dereferenceable(1) %0, i32 noundef 2) #9
%lnot = xor i1 %call, true
br i1 %lnot, label %while.body, label %while.end

while.body: ; preds = %while.cond
call void @_Z3barv()
br label %while.cond, !llvm.loop !9

while.end: ; preds = %while.cond
ret void
}

; Function Attrs: nounwind willreturn memory(argmem: readwrite)
declare void @llvm.taskframe.use(token) #3

; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #1

declare void @_Z3bazv() #5

; Function Attrs: mustprogress nounwind uwtable
define linkonce_odr dso_local void @_ZNSt6atomicIbE5storeEbSt12memory_order(ptr noundef nonnull align 1 dereferenceable(1) %this, i1 noundef zeroext %__i, i32 noundef %__m) #6 comdat align 2 {
entry:
%this.addr = alloca ptr, align 8
%__i.addr = alloca i8, align 1
%__m.addr = alloca i32, align 4
store ptr %this, ptr %this.addr, align 8, !tbaa !5
%storedv = zext i1 %__i to i8
store i8 %storedv, ptr %__i.addr, align 1, !tbaa !12
store i32 %__m, ptr %__m.addr, align 4, !tbaa !14
%this1 = load ptr, ptr %this.addr, align 8
%_M_base = getelementptr inbounds %"struct.std::atomic", ptr %this1, i32 0, i32 0
%0 = load i8, ptr %__i.addr, align 1, !tbaa !12, !range !16, !noundef !17
%loadedv = trunc i8 %0 to i1
%1 = load i32, ptr %__m.addr, align 4, !tbaa !14
call void @_ZNSt13__atomic_baseIbE5storeEbSt12memory_order(ptr noundef nonnull align 1 dereferenceable(1) %_M_base, i1 noundef zeroext %loadedv, i32 noundef %1) #9
ret void
}

; Function Attrs: willreturn memory(argmem: readwrite)
declare void @llvm.sync.unwind(token) #7

; Function Attrs: mustprogress nounwind uwtable
define linkonce_odr dso_local noundef zeroext i1 @_ZNKSt6atomicIbE4loadESt12memory_order(ptr noundef nonnull align 1 dereferenceable(1) %this, i32 noundef %__m) #6 comdat align 2 {
entry:
%this.addr = alloca ptr, align 8
%__m.addr = alloca i32, align 4
store ptr %this, ptr %this.addr, align 8, !tbaa !5
store i32 %__m, ptr %__m.addr, align 4, !tbaa !14
%this1 = load ptr, ptr %this.addr, align 8
%_M_base = getelementptr inbounds %"struct.std::atomic", ptr %this1, i32 0, i32 0
%0 = load i32, ptr %__m.addr, align 4, !tbaa !14
%call = call noundef zeroext i1 @_ZNKSt13__atomic_baseIbE4loadESt12memory_order(ptr noundef nonnull align 1 dereferenceable(1) %_M_base, i32 noundef %0) #9
ret i1 %call
}

declare void @_Z3barv() #5

; Function Attrs: alwaysinline mustprogress nounwind uwtable
define linkonce_odr dso_local noundef zeroext i1 @_ZNKSt13__atomic_baseIbE4loadESt12memory_order(ptr noundef nonnull align 1 dereferenceable(1) %this, i32 noundef %__m) #8 comdat align 2 {
entry:
%this.addr = alloca ptr, align 8
%__m.addr = alloca i32, align 4
%__b = alloca i32, align 4
%atomic-temp = alloca i8, align 1
store ptr %this, ptr %this.addr, align 8, !tbaa !5
store i32 %__m, ptr %__m.addr, align 4, !tbaa !14
%this1 = load ptr, ptr %this.addr, align 8
call void @llvm.lifetime.start.p0(i64 4, ptr %__b) #9
%0 = load i32, ptr %__m.addr, align 4, !tbaa !14
%call = call noundef i32 @_ZStanSt12memory_orderSt23__memory_order_modifier(i32 noundef %0, i32 noundef 65535) #9
store i32 %call, ptr %__b, align 4, !tbaa !14
br label %do.body

do.body: ; preds = %entry
br label %do.cond

do.cond: ; preds = %do.body
br label %do.end

do.end: ; preds = %do.cond
br label %do.body2

do.body2: ; preds = %do.end
br label %do.cond3

do.cond3: ; preds = %do.body2
br label %do.end4

do.end4: ; preds = %do.cond3
%_M_i = getelementptr inbounds %"struct.std::__atomic_base", ptr %this1, i32 0, i32 0
%1 = load i32, ptr %__m.addr, align 4, !tbaa !14
switch i32 %1, label %monotonic [
i32 1, label %acquire
i32 2, label %acquire
i32 5, label %seqcst
]

monotonic: ; preds = %do.end4
%2 = load atomic i8, ptr %_M_i monotonic, align 1
store i8 %2, ptr %atomic-temp, align 1
br label %atomic.continue

acquire: ; preds = %do.end4, %do.end4
%3 = load atomic i8, ptr %_M_i acquire, align 1
store i8 %3, ptr %atomic-temp, align 1
br label %atomic.continue

seqcst: ; preds = %do.end4
%4 = load atomic i8, ptr %_M_i seq_cst, align 1
store i8 %4, ptr %atomic-temp, align 1
br label %atomic.continue

atomic.continue: ; preds = %seqcst, %acquire, %monotonic
%5 = load i8, ptr %atomic-temp, align 1, !tbaa !12, !range !16, !noundef !17
%loadedv = trunc i8 %5 to i1
call void @llvm.lifetime.end.p0(i64 4, ptr %__b) #9
ret i1 %loadedv
}

; Function Attrs: mustprogress nounwind uwtable
define linkonce_odr dso_local noundef i32 @_ZStanSt12memory_orderSt23__memory_order_modifier(i32 noundef %__m, i32 noundef %__mod) #6 comdat {
entry:
%__m.addr = alloca i32, align 4
%__mod.addr = alloca i32, align 4
store i32 %__m, ptr %__m.addr, align 4, !tbaa !14
store i32 %__mod, ptr %__mod.addr, align 4, !tbaa !18
%0 = load i32, ptr %__m.addr, align 4, !tbaa !14
%1 = load i32, ptr %__mod.addr, align 4, !tbaa !18
%and = and i32 %0, %1
ret i32 %and
}

; Function Attrs: alwaysinline mustprogress nounwind uwtable
define linkonce_odr dso_local void @_ZNSt13__atomic_baseIbE5storeEbSt12memory_order(ptr noundef nonnull align 1 dereferenceable(1) %this, i1 noundef zeroext %__i, i32 noundef %__m) #8 comdat align 2 {
entry:
%this.addr = alloca ptr, align 8
%__i.addr = alloca i8, align 1
%__m.addr = alloca i32, align 4
%__b = alloca i32, align 4
%.atomictmp = alloca i8, align 1
store ptr %this, ptr %this.addr, align 8, !tbaa !5
%storedv = zext i1 %__i to i8
store i8 %storedv, ptr %__i.addr, align 1, !tbaa !12
store i32 %__m, ptr %__m.addr, align 4, !tbaa !14
%this1 = load ptr, ptr %this.addr, align 8
call void @llvm.lifetime.start.p0(i64 4, ptr %__b) #9
%0 = load i32, ptr %__m.addr, align 4, !tbaa !14
%call = call noundef i32 @_ZStanSt12memory_orderSt23__memory_order_modifier(i32 noundef %0, i32 noundef 65535) #9
store i32 %call, ptr %__b, align 4, !tbaa !14
br label %do.body

do.body: ; preds = %entry
br label %do.cond

do.cond: ; preds = %do.body
br label %do.end

do.end: ; preds = %do.cond
br label %do.body2

do.body2: ; preds = %do.end
br label %do.cond3

do.cond3: ; preds = %do.body2
br label %do.end4

do.end4: ; preds = %do.cond3
br label %do.body5

do.body5: ; preds = %do.end4
br label %do.cond6

do.cond6: ; preds = %do.body5
br label %do.end7

do.end7: ; preds = %do.cond6
%_M_i = getelementptr inbounds %"struct.std::__atomic_base", ptr %this1, i32 0, i32 0
%1 = load i32, ptr %__m.addr, align 4, !tbaa !14
%2 = load i8, ptr %__i.addr, align 1, !tbaa !12, !range !16, !noundef !17
%loadedv = trunc i8 %2 to i1
%storedv8 = zext i1 %loadedv to i8
store i8 %storedv8, ptr %.atomictmp, align 1, !tbaa !12
switch i32 %1, label %monotonic [
i32 3, label %release
i32 5, label %seqcst
]

monotonic: ; preds = %do.end7
%3 = load i8, ptr %.atomictmp, align 1
store atomic i8 %3, ptr %_M_i monotonic, align 1
br label %atomic.continue

release: ; preds = %do.end7
%4 = load i8, ptr %.atomictmp, align 1
store atomic i8 %4, ptr %_M_i release, align 1
br label %atomic.continue

seqcst: ; preds = %do.end7
%5 = load i8, ptr %.atomictmp, align 1
store atomic i8 %5, ptr %_M_i seq_cst, align 1
br label %atomic.continue

atomic.continue: ; preds = %seqcst, %release, %monotonic
call void @llvm.lifetime.end.p0(i64 4, ptr %__b) #9
ret void
}

attributes #0 = { mustprogress uwtable "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #1 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
attributes #2 = { nocallback nofree nounwind willreturn memory(argmem: write) }
attributes #3 = { nounwind willreturn memory(argmem: readwrite) }
attributes #4 = { inlinehint mustprogress uwtable "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #5 = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #6 = { mustprogress nounwind uwtable "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #7 = { willreturn memory(argmem: readwrite) }
attributes #8 = { alwaysinline mustprogress nounwind uwtable "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
attributes #9 = { nounwind }

!llvm.module.flags = !{!0, !1, !2, !3}
!llvm.ident = !{!4}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{i32 8, !"PIC Level", i32 2}
!2 = !{i32 7, !"PIE Level", i32 2}
!3 = !{i32 7, !"uwtable", i32 2}
!4 = !{!"clang version 19.1.7 ([email protected]:neboat/opencilk-project.git 5190d883dcdd73d55187685cb6013a7c01f3a7a8)"}
!5 = !{!6, !6, i64 0}
!6 = !{!"any pointer", !7, i64 0}
!7 = !{!"omnipotent char", !8, i64 0}
!8 = !{!"Simple C++ TBAA"}
!9 = distinct !{!9, !10, !11}
!10 = !{!"llvm.loop.mustprogress"}
!11 = !{!"llvm.loop.unroll.disable"}
!12 = !{!13, !13, i64 0}
!13 = !{!"bool", !7, i64 0}
!14 = !{!15, !15, i64 0}
!15 = !{!"_ZTSSt12memory_order", !7, i64 0}
!16 = !{i8 0, i8 2}
!17 = !{}
!18 = !{!19, !19, i64 0}
!19 = !{!"_ZTSSt23__memory_order_modifier", !7, i64 0}
Loading