Skip to content

Commit 32c4db0

Browse files
rlavaeerlavaee
committed
[BasicBlockSections] Allow mixing of -basic-block-sections with MFS.
1 parent 4acf935 commit 32c4db0

File tree

3 files changed

+66
-9
lines changed

3 files changed

+66
-9
lines changed

‎llvm/lib/CodeGen/MachineFunctionSplitter.cpp‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -28,6 +28,7 @@
2828
#include "llvm/Analysis/EHUtils.h"
2929
#include "llvm/Analysis/ProfileSummaryInfo.h"
3030
#include "llvm/CodeGen/BasicBlockSectionUtils.h"
31+
#include "llvm/CodeGen/BasicBlockSectionsProfileReader.h"
3132
#include "llvm/CodeGen/MachineBasicBlock.h"
3233
#include "llvm/CodeGen/MachineBlockFrequencyInfo.h"
3334
#include "llvm/CodeGen/MachineFunction.h"
@@ -128,6 +129,9 @@ static bool isColdBlock(const MachineBasicBlock &MBB,
128129
}
129130

130131
bool MachineFunctionSplitter::runOnMachineFunction(MachineFunction &MF) {
132+
// Do not split functions when -basic-block-sections=all is specified.
133+
if (MF.getTarget().getBBSectionsType() == llvm::BasicBlockSection::All)
134+
return false;
131135
// We target functions with profile data. Static information in the form
132136
// of exception handling code may be split to cold if user passes the
133137
// mfs-split-ehcode flag.
@@ -139,6 +143,13 @@ bool MachineFunctionSplitter::runOnMachineFunction(MachineFunction &MF) {
139143
if (!TII.isFunctionSafeToSplit(MF))
140144
return false;
141145

146+
// Do not split functions with BasicBlockSections profiles as they will
147+
// be split according to that profile by the BasicBlockSections pass.
148+
auto BBSPRWP =
149+
getAnalysisIfAvailable<BasicBlockSectionsProfileReaderWrapperPass>();
150+
if (BBSPRWP != nullptr && BBSPRWP->getBBSPR().isFunctionHot(MF.getName()))
151+
return false;
152+
142153
// Renumbering blocks here preserves the order of the blocks as
143154
// sortBasicBlocksAndUpdateBranches uses the numeric identifier to sort
144155
// blocks. Preserving the order of blocks is essential to retaining decisions
@@ -201,6 +212,7 @@ void MachineFunctionSplitter::getAnalysisUsage(AnalysisUsage &AU) const {
201212
AU.addRequired<MachineModuleInfoWrapperPass>();
202213
AU.addRequired<MachineBlockFrequencyInfoWrapperPass>();
203214
AU.addRequired<ProfileSummaryInfoWrapperPass>();
215+
AU.addUsedIfAvailable<BasicBlockSectionsProfileReaderWrapperPass>();
204216
}
205217

206218
char MachineFunctionSplitter::ID = 0;

‎llvm/lib/CodeGen/TargetPassConfig.cpp‎

Lines changed: 9 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -1235,13 +1235,13 @@ void TargetPassConfig::addMachinePasses() {
12351235
addPass(createMIRAddFSDiscriminatorsPass(
12361236
sampleprof::FSDiscriminatorPass::PassLast));
12371237

1238-
bool NeedsBBSections =
1239-
TM->getBBSectionsType() != llvm::BasicBlockSection::None;
1240-
// Machine function splitter uses the basic block sections feature. Both
1241-
// cannot be enabled at the same time. We do not apply machine function
1242-
// splitter if -basic-block-sections is requested.
1243-
if (!NeedsBBSections && (TM->Options.EnableMachineFunctionSplitter ||
1244-
EnableMachineFunctionSplitter)) {
1238+
// Machine function splitter uses the basic block sections feature.
1239+
// When used along with `-basic-block-sections=`, the basic-block-sections
1240+
// feature takes precedence. This means functions eligible for
1241+
// basic-block-sections optimizations (`=all`, or `=list=` with function
1242+
// included in the list profile) will get that optimization instead.
1243+
if (TM->Options.EnableMachineFunctionSplitter ||
1244+
EnableMachineFunctionSplitter) {
12451245
const std::string ProfileFile = getFSProfileFile(TM);
12461246
if (!ProfileFile.empty()) {
12471247
if (EnableFSDiscriminator) {
@@ -1260,7 +1260,8 @@ void TargetPassConfig::addMachinePasses() {
12601260
}
12611261
// We run the BasicBlockSections pass if either we need BB sections or BB
12621262
// address map (or both).
1263-
if (NeedsBBSections || TM->Options.BBAddrMap) {
1263+
if (TM->getBBSectionsType() != llvm::BasicBlockSection::None ||
1264+
TM->Options.BBAddrMap) {
12641265
if (TM->getBBSectionsType() == llvm::BasicBlockSection::List) {
12651266
addPass(llvm::createBasicBlockSectionsProfileReaderWrapperPass(
12661267
TM->getBBSectionsFuncListBuf()));

‎llvm/test/CodeGen/Generic/machine-function-splitter.ll‎

Lines changed: 45 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,12 +2,21 @@
22
; REQUIRES: x86-registered-target
33

44
; COM: Machine function splitting with FDO profiles
5-
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -split-machine-functions | FileCheck %s -check-prefixes=MFS-DEFAULTS,MFS-DEFAULTS-X86
5+
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -split-machine-functions | FileCheck %s -check-prefixes=MFS-DEFAULTS,MFS-DEFAULTS-X86,MFS-NOBBSECTIONS
66
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -split-machine-functions -mfs-psi-cutoff=0 -mfs-count-threshold=2000 | FileCheck %s --dump-input=always -check-prefixes=MFS-OPTS1,MFS-OPTS1-X86
77
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -split-machine-functions -mfs-psi-cutoff=950000 | FileCheck %s -check-prefixes=MFS-OPTS2,MFS-OPTS2-X86
88
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -split-machine-functions -mfs-split-ehcode | FileCheck %s -check-prefixes=MFS-EH-SPLIT,MFS-EH-SPLIT-X86
99
; RUN: llc < %s -mtriple=x86_64 -split-machine-functions -O0 -mfs-psi-cutoff=0 -mfs-count-threshold=10000 | FileCheck %s -check-prefixes=MFS-O0,MFS-O0-X86
1010

11+
; COM: Machine function splitting along with -basic-block-sections=list
12+
; RUN: echo 'v1' > %t
13+
; RUN: echo 'ffoo21' >> %t
14+
; RUN: echo 'c0' >> %t
15+
; RUN: echo 'ffoo1' >> %t
16+
; RUN: echo 'c0 1' >> %t
17+
; RUN: echo 'c2' >> %t
18+
; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -basic-block-sections=%t -split-machine-functions | FileCheck %s --check-prefixes=MFS-DEFAULT,MFS-BBSECTIONS
19+
1120
; RUN: llc < %s -mtriple=aarch64-unknown-linux-gnu -aarch64-min-jump-table-entries=4 -enable-split-machine-functions | FileCheck %s -check-prefixes=MFS-DEFAULTS,MFS-DEFAULTS-AARCH64
1221
; RUN: llc < %s -mtriple=aarch64-unknown-linux-gnu -aarch64-min-jump-table-entries=4 -enable-split-machine-functions -mfs-psi-cutoff=0 -mfs-count-threshold=2000 | FileCheck %s --dump-input=always -check-prefixes=MFS-OPTS1,MFS-OPTS1-AARCH64
1322
; RUN: llc < %s -mtriple=aarch64-unknown-linux-gnu -aarch64-min-jump-table-entries=4 -enable-split-machine-functions -mfs-psi-cutoff=950000 | FileCheck %s -check-prefixes=MFS-OPTS2,MFS-OPTS2-AARCH64
@@ -20,9 +29,16 @@
2029
; RUN: llc < %t.ll -mtriple=x86_64-unknown-linux-gnu -split-machine-functions | FileCheck %s --check-prefix=FSAFDO-MFS
2130
; RUN: llc < %t.ll -mtriple=x86_64-unknown-linux-gnu -split-machine-functions | FileCheck %s --check-prefix=FSAFDO-MFS2
2231

32+
2333
define void @foo1(i1 zeroext %0) nounwind !prof !14 !section_prefix !15 {
2434
;; Check that cold block is moved to .text.split.
2535
; MFS-DEFAULTS-LABEL: foo1
36+
;; Check that we create one extra section with -basic-block-sections.
37+
; MFS-BBSECTIONS: callq bar
38+
; MFS-BBSECTIONS: .section .text.hot.foo1
39+
; MFS-BBSECTIONS-NEXT: foo1.__part.1:
40+
; MFS-BBSECTIONS: callq baz
41+
; MFS_NOBBSECTIONS-NOT: .section .text.hot.foo1
2642
; MFS-DEFAULTS: .section .text.split.foo1
2743
; MFS-DEFAULTS-NEXT: foo1.cold:
2844
; MFS-DEFAULTS-X86-NOT: callq bar
@@ -610,6 +626,34 @@ cold_asm_target:
610626
ret void
611627
}
612628

629+
define void @foo21(i1 zeroext %0) {
630+
;; Check that a function with basic-block-sections profile is properly split
631+
;; when the profile is used along with mfs.
632+
; MFS-BBSECTIONS: .section .text.hot.foo21
633+
; MFS-NOBBSECTIONS-NOT: .section .text.hot.foo21
634+
; MFS-DEFAULT-DAG: foo21
635+
; MFS-NOBBSECTIONS-NOT: foo21.cold
636+
; MFS-BBSECTIONS: .section .text.split.foo21
637+
; MFS-BBSECTIONS-DAG: oo21.cold
638+
%2 = alloca i8, align 1
639+
%3 = zext i1 %0 to i8
640+
store i8 %3, ptr %2, align 1
641+
%4 = load i8, ptr %2, align 1
642+
%5 = trunc i8 %4 to i1
643+
br i1 %5, label %6, label %8
644+
645+
6: ; preds = %1
646+
%7 = call i32 @bar()
647+
br label %10
648+
649+
8: ; preds = %1
650+
%9 = call i32 @baz()
651+
br label %10
652+
653+
10: ; preds = %8, %6
654+
ret void
655+
}
656+
613657
declare i32 @bar()
614658
declare i32 @baz()
615659
declare i32 @bam()

0 commit comments

Comments
 (0)