This is an archive of the discontinued LLVM Phabricator instance.

Differential D141499

[AArch64] Set MaxInterleaveFactor for Apple A14, A15, A16.
ClosedPublic

Authored by fhahn on Jan 11 2023, 7:23 AM.

Details

Reviewers
t.p.northover
ab
jroelofs
Commits
rG830d0bc56b41: [AArch64] Set MaxInterleaveFactor for Apple A14, A15, A16.
Summary

Those CPUs can benefit from additional interleaving.

Diff Detail

Event Timeline

fhahn created this revision.Jan 11 2023, 7:23 AM
Herald added a project: Restricted Project. · View Herald TranscriptJan 11 2023, 7:23 AM
Herald added subscribers: StephenFan, hiraditya, kristof.beyls. · View Herald Transcript
fhahn requested review of this revision.Jan 11 2023, 7:23 AM
Herald added a project: Restricted Project. · View Herald TranscriptJan 11 2023, 7:23 AM
Herald added a subscriber: • pcwang-thead. · View Herald Transcript
Harbormaster completed remote builds in B207085: Diff 488209.Jan 11 2023, 8:30 AM
jroelofs accepted this revision.Jan 11 2023, 8:36 AM

LGTM

This revision is now accepted and ready to land.Jan 11 2023, 8:36 AM
This revision was landed with ongoing or failed builds.Jan 11 2023, 10:53 AM
Closed by commit rG830d0bc56b41: [AArch64] Set MaxInterleaveFactor for Apple A14, A15, A16. (authored by fhahn). · Explain Why
This revision was automatically updated to reflect the committed changes.
fhahn added a commit: rG830d0bc56b41: [AArch64] Set MaxInterleaveFactor for Apple A14, A15, A16..

Revision Contents

PathSize
llvm/
lib/
Target/
AArch64/
9 lines
test/
Transforms/
LoopVectorize/
AArch64/
7 lines
7 lines

Diff 488288

llvm/lib/Target/AArch64/AArch64Subtarget.cpp

Show First 20 Lines • Show All 166 Lines • ▼ Show 20 Linesvoid AArch64Subtarget::initializeProperties() {
case AppleA13: case AppleA13:
case AppleA14: case AppleA14:
case AppleA15: case AppleA15:
case AppleA16: case AppleA16:
CacheLineSize = 64; CacheLineSize = 64;
PrefetchDistance = 280; PrefetchDistance = 280;
MinPrefetchStride = 2048; MinPrefetchStride = 2048;
MaxPrefetchIterationsAhead = 3; MaxPrefetchIterationsAhead = 3;
switch (ARMProcFamily) {
case AppleA14:
case AppleA15:
case AppleA16:
MaxInterleaveFactor = 4;
break;
default:
break;
}
break; break;
case ExynosM3: case ExynosM3:
MaxInterleaveFactor = 4; MaxInterleaveFactor = 4;
MaxJumpTableSize = 20; MaxJumpTableSize = 20;
PrefFunctionLogAlignment = 5; PrefFunctionLogAlignment = 5;
PrefLoopLogAlignment = 4; PrefLoopLogAlignment = 4;
break; break;
case Falkor: case Falkor:
▲ Show 20 Lines • Show All 283 Lines • Show Last 20 Lines

llvm/test/Transforms/LoopVectorize/AArch64/interleaving-load-store.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a57 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a57 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a75 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a75 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-m1 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-m1 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a14 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a14 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a15 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a15 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a16 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; Tests for selecting interleave counts for loops with loads and stores. ; Tests for selecting interleave counts for loops with loads and stores.
define void @interleave_single_load_store(ptr %src, ptr %dst, i64 %N, i8 %a, i8 %b) { define void @interleave_single_load_store(ptr %src, ptr %dst, i64 %N, i8 %a, i8 %b) {
; INTERLEAVE-4-LABEL: @interleave_single_load_store( ; INTERLEAVE-4-LABEL: @interleave_single_load_store(
; INTERLEAVE-4-NEXT: iter.check: ; INTERLEAVE-4-NEXT: iter.check:
; INTERLEAVE-4-NEXT: [[SRC2:%.*]] = ptrtoint ptr [[SRC:%.*]] to i64 ; INTERLEAVE-4-NEXT: [[SRC2:%.*]] = ptrtoint ptr [[SRC:%.*]] to i64
; INTERLEAVE-4-NEXT: [[DST1:%.*]] = ptrtoint ptr [[DST:%.*]] to i64 ; INTERLEAVE-4-NEXT: [[DST1:%.*]] = ptrtoint ptr [[DST:%.*]] to i64
▲ Show 20 Lines • Show All 249 Lines • Show Last 20 Lines

llvm/test/Transforms/LoopVectorize/AArch64/interleaving-reduction.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a57 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a57 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a75 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=cortex-a75 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-m1 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-m1 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a14 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a14 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a15 -S %s | FileCheck --check-prefix=INTERLEAVE-2 %s ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a15 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-macos -mcpu=apple-a16 -S %s | FileCheck --check-prefix=INTERLEAVE-4 %s
; Tests for selecting the interleave count for loops with reductions. ; Tests for selecting the interleave count for loops with reductions.
define i32 @interleave_integer_reduction(ptr %src, i64 %N) { define i32 @interleave_integer_reduction(ptr %src, i64 %N) {
; INTERLEAVE-4-LABEL: @interleave_integer_reduction( ; INTERLEAVE-4-LABEL: @interleave_integer_reduction(
; INTERLEAVE-4-NEXT: entry: ; INTERLEAVE-4-NEXT: entry:
; INTERLEAVE-4-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 16 ; INTERLEAVE-4-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 16
; INTERLEAVE-4-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; INTERLEAVE-4-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
▲ Show 20 Lines • Show All 122 Lines • Show Last 20 Lines