Skip to content

[ARM] Reduce loop unroll when low overhead branching is available #120065

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

Merged
merged 1 commit into from
Dec 18, 2024
Merged

[ARM] Reduce loop unroll when low overhead branching is available #120065

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
23 changes: 22 additions & 1 deletion llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2592,11 +2592,32 @@ void ARMTTIImpl::getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
return;
}

// For processors with low overhead branching (LOB), runtime unrolling the
// innermost loop is often detrimental to performance. In these cases the loop
// remainder gets unrolled into a series of compare-and-jump blocks, which in
// deeply nested loops get executed multiple times, negating the benefits of
// LOB. This is particularly noticable when the loop trip count of the
// innermost loop varies within the outer loop, such as in the case of
// triangular matrix decompositions. In these cases we will prefer to not
// unroll the innermost loop, with the intention for it to be executed as a
// low overhead loop.
bool Runtime = true;
if (ST->hasLOB()) {
if (SE.hasLoopInvariantBackedgeTakenCount(L)) {
const auto *BETC = SE.getBackedgeTakenCount(L);
auto *Outer = L->getOutermostLoop();
if ((L != Outer && Outer != L->getParentLoop()) ||
(L != Outer && BETC && !SE.isLoopInvariant(BETC, Outer))) {
Runtime = false;
}
}
}

LLVM_DEBUG(dbgs() << "Cost of loop: " << Cost << "\n");
LLVM_DEBUG(dbgs() << "Default Runtime Unroll Count: " << UnrollCount << "\n");

UP.Partial = true;
UP.Runtime = true;
UP.Runtime = Runtime;
UP.UnrollRemainder = true;
UP.DefaultUnrollRuntimeCount = UnrollCount;
UP.UnrollAndJam = true;
Expand Down
27 changes: 18 additions & 9 deletions llvm/test/Transforms/LoopUnroll/ARM/lob-unroll.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,17 +1,23 @@
; RUN: opt -mcpu=cortex-m7 -mtriple=thumbv8.1m.main -passes=loop-unroll -S %s -o - | FileCheck %s --check-prefix=NLOB
; RUN: opt -mcpu=cortex-m55 -mtriple=thumbv8.1m.main -passes=loop-unroll -S %s -o - | FileCheck %s --check-prefix=LOB

; This test checks behaviour of loop unrolling on processors with low overhead branching available

; LOB-CHECK-LABEL: for.body{{.*}}.prol
; LOB-COUNT-1: fmul fast float
; LOB-CHECK-LABEL: for.body{{.*}}.prol.1
; LOB-COUNT-1: fmul fast float
; LOB-CHECK-LABEL: for.body{{.*}}.prol.2
; LOB-COUNT-1: fmul fast float
; LOB-CHECK-LABEL: for.body{{.*}}
; LOB-COUNT-4: fmul fast float
; NLOB-LABEL: for.body{{.*}}.prol:
; NLOB-COUNT-1: fmul fast float
; NLOB-LABEL: for.body{{.*}}.prol.1:
; NLOB-COUNT-1: fmul fast float
; NLOB-LABEL: for.body{{.*}}.prol.2:
; NLOB-COUNT-1: fmul fast float
; NLOB-LABEL: for.body{{.*}}:
; NLOB-COUNT-4: fmul fast float
; NLOB-NOT: fmul fast float

; LOB-LABEL: for.body{{.*}}:
; LOB: fmul fast float
; LOB-NOT: fmul fast float


; Function Attrs: nofree norecurse nosync nounwind memory(argmem: readwrite)
define dso_local void @test(i32 noundef %n, ptr nocapture noundef %pA) local_unnamed_addr #0 {
entry:
Expand All @@ -20,7 +26,7 @@ entry:

for.cond.loopexit: ; preds = %for.cond6.for.cond.cleanup8_crit_edge.us, %for.body
%exitcond49.not = icmp eq i32 %add, %n
br i1 %exitcond49.not, label %for.cond.cleanup, label %for.body
br i1 %exitcond49.not, label %for.cond.cleanup, label %for.body, !llvm.loop !0

for.cond.cleanup: ; preds = %for.cond.loopexit, %entry
ret void
Expand Down Expand Up @@ -61,3 +67,6 @@ for.cond6.for.cond.cleanup8_crit_edge.us: ; preds = %for.body9.us
br i1 %exitcond48.not, label %for.cond.loopexit, label %for.cond6.preheader.us
}

!0 = distinct !{!0, !1, !2}
!1 = !{!"llvm.loop.mustprogress"}
!2 = !{!"llvm.loop.unroll.disable"}
Loading