[lld] [lld][BP] Avoid ordering ICF'ed sections (PR #126327)
Ellis Hoag via llvm-commits
llvm-commits at lists.llvm.org
Mon Feb 10 09:22:17 PST 2025
https://github.com/ellishg updated https://github.com/llvm/llvm-project/pull/126327
>From ca93859385faf9cdce53d5cd2208a1259128e6c0 Mon Sep 17 00:00:00 2001
From: Ellis Hoag <ellishoag at meta.com>
Date: Fri, 7 Feb 2025 16:14:57 -0800
Subject: [PATCH 1/2] [lld][BP] Avoid ordering ICF'ed sections
---
lld/MachO/BPSectionOrderer.cpp | 2 +
lld/test/ELF/bp-section-orderer.s | 66 ++++++++++++++++++++++++-----
lld/test/MachO/bp-section-orderer.s | 12 +++++-
3 files changed, 68 insertions(+), 12 deletions(-)
diff --git a/lld/MachO/BPSectionOrderer.cpp b/lld/MachO/BPSectionOrderer.cpp
index 689afd67712a417..f693e5e59f8faea 100644
--- a/lld/MachO/BPSectionOrderer.cpp
+++ b/lld/MachO/BPSectionOrderer.cpp
@@ -117,6 +117,8 @@ DenseMap<const InputSection *, int> lld::macho::runBalancedPartitioning(
auto *isec = subsec.isec;
if (!isec || isec->data.empty())
continue;
+ if (isa<ConcatInputSection>(isec) && !isec->isLive(0))
+ continue;
size_t idx = sections.size();
sections.emplace_back(isec);
for (auto *sym : BPOrdererMachO::getSymbols(*isec)) {
diff --git a/lld/test/ELF/bp-section-orderer.s b/lld/test/ELF/bp-section-orderer.s
index 2e18107c02ca306..6a61e7e8cef2b8d 100644
--- a/lld/test/ELF/bp-section-orderer.s
+++ b/lld/test/ELF/bp-section-orderer.s
@@ -1,3 +1,4 @@
+# NOTE: Code has been autogenerated by utils/update_test_body.py
# REQUIRES: aarch64
# RUN: rm -rf %t && split-file %s %t && cd %t
@@ -25,30 +26,33 @@
# RUN: ld.lld -o out.s a.o --irpgo-profile=a.profdata --bp-startup-sort=function
# RUN: llvm-nm -jn out.s | tr '\n' , | FileCheck %s --check-prefix=STARTUP
-# STARTUP: s5,s4,s3,s2,s1,A,B,C,F,E,D,_start,d4,d3,d2,d1,{{$}}
+# STARTUP: s5,s4,s3,s2,s1,A,B,C,F,E,D,merged1,merged2,_start,d4,d3,d2,d1,{{$}}
# RUN: ld.lld -o out.os a.o --irpgo-profile=a.profdata --bp-startup-sort=function --symbol-ordering-file a.txt
# RUN: llvm-nm -jn out.os | tr '\n' , | FileCheck %s --check-prefix=ORDER-STARTUP
-# ORDER-STARTUP: s2,s1,s5,s4,s3,A,F,E,D,B,C,_start,d3,d2,d4,d1,{{$}}
+# ORDER-STARTUP: s2,s1,s5,s4,s3,A,F,E,D,B,C,merged1,merged2,_start,d3,d2,d4,d1,{{$}}
# RUN: ld.lld -o out.cf a.o --verbose-bp-section-orderer --bp-compression-sort=function 2>&1 | FileCheck %s --check-prefix=BP-COMPRESSION-FUNC
+# RUN: ld.lld -o out.cf.icf a.o --verbose-bp-section-orderer --bp-compression-sort=function --icf=all 2>&1 | FileCheck %s --check-prefix=BP-COMPRESSION-ICF-FUNC
# RUN: llvm-nm -jn out.cf | tr '\n' , | FileCheck %s --check-prefix=CFUNC
-# CFUNC: s5,s4,s3,s2,s1,F,C,E,D,B,A,_start,d4,d3,d2,d1,{{$}}
+# CFUNC: s5,s4,s3,s2,s1,A,F,merged1,merged2,C,E,D,B,_start,d4,d3,d2,d1,{{$}}
# RUN: ld.lld -o out.cd a.o --verbose-bp-section-orderer --bp-compression-sort=data 2>&1 | FileCheck %s --check-prefix=BP-COMPRESSION-DATA
# RUN: llvm-nm -jn out.cd | tr '\n' , | FileCheck %s --check-prefix=CDATA
-# CDATA: s5,s3,s4,s2,s1,F,C,E,D,B,A,_start,d4,d1,d3,d2,{{$}}
+# CDATA: s5,s3,s4,s2,s1,F,C,E,D,B,A,merged1,merged2,_start,d4,d1,d3,d2,{{$}}
# RUN: ld.lld -o out.cb a.o --verbose-bp-section-orderer --bp-compression-sort=both 2>&1 | FileCheck %s --check-prefix=BP-COMPRESSION-BOTH
-# RUN: llvm-nm -jn out.cb | tr '\n' , | FileCheck %s --check-prefix=CDATA
+# RUN: llvm-nm -jn out.cb | tr '\n' , | FileCheck %s --check-prefix=CBOTH
+# CBOTH: s5,s3,s4,s2,s1,A,F,merged1,merged2,C,E,D,B,_start,d4,d1,d3,d2,{{$}}
# RUN: ld.lld -o out.cbs a.o --verbose-bp-section-orderer --bp-compression-sort=both --irpgo-profile=a.profdata --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=BP-COMPRESSION-BOTH
# RUN: llvm-nm -jn out.cbs | tr '\n' , | FileCheck %s --check-prefix=CBOTH-STARTUP
-# CBOTH-STARTUP: s5,s3,s4,s2,s1,A,B,C,F,E,D,_start,d4,d1,d3,d2,{{$}}
+# CBOTH-STARTUP: s5,s3,s4,s2,s1,A,B,C,F,E,D,merged1,merged2,_start,d4,d1,d3,d2,{{$}}
-# BP-COMPRESSION-FUNC: Ordered 7 sections using balanced partitioning
+# BP-COMPRESSION-FUNC: Ordered 9 sections using balanced partitioning
+# BP-COMPRESSION-ICF-FUNC: Ordered 8 sections using balanced partitioning
# BP-COMPRESSION-DATA: Ordered 9 sections using balanced partitioning
-# BP-COMPRESSION-BOTH: Ordered 16 sections using balanced partitioning
+# BP-COMPRESSION-BOTH: Ordered 18 sections using balanced partitioning
#--- a.proftext
:ir
@@ -125,6 +129,9 @@ int C(int a) { A(); return a + 2; }
int B(int a) { A(); return a + 1; }
void A() {}
+int merged1(int a) { return F(a + 101); }
+int merged2(int a) { return F(a + 101); }
+
int _start() { return 0; }
#--- gen
@@ -236,6 +243,44 @@ A: // @A
.Lfunc_end5:
.size A, .Lfunc_end5-A
// -- End function
+ .section .text.merged1,"ax", at progbits
+ .globl merged1 // -- Begin function merged1
+ .p2align 2
+ .type merged1, at function
+merged1: // @merged1
+// %bb.0: // %entry
+ sub sp, sp, #32
+ stp x29, x30, [sp, #16] // 16-byte Folded Spill
+ add x29, sp, #16
+ stur w0, [x29, #-4]
+ ldur w8, [x29, #-4]
+ add w0, w8, #101
+ bl F
+ ldp x29, x30, [sp, #16] // 16-byte Folded Reload
+ add sp, sp, #32
+ ret
+.Lfunc_end6:
+ .size merged1, .Lfunc_end6-merged1
+ // -- End function
+ .section .text.merged2,"ax", at progbits
+ .globl merged2 // -- Begin function merged2
+ .p2align 2
+ .type merged2, at function
+merged2: // @merged2
+// %bb.0: // %entry
+ sub sp, sp, #32
+ stp x29, x30, [sp, #16] // 16-byte Folded Spill
+ add x29, sp, #16
+ stur w0, [x29, #-4]
+ ldur w8, [x29, #-4]
+ add w0, w8, #101
+ bl F
+ ldp x29, x30, [sp, #16] // 16-byte Folded Reload
+ add sp, sp, #32
+ ret
+.Lfunc_end7:
+ .size merged2, .Lfunc_end7-merged2
+ // -- End function
.section .text._start,"ax", at progbits
.globl _start // -- Begin function _start
.p2align 2
@@ -244,8 +289,8 @@ _start: // @_start
// %bb.0: // %entry
mov w0, wzr
ret
-.Lfunc_end6:
- .size _start, .Lfunc_end6-_start
+.Lfunc_end8:
+ .size _start, .Lfunc_end8-_start
// -- End function
.type s5, at object // @s5
.section .rodata.s5,"a", at progbits
@@ -330,6 +375,7 @@ d1:
.section ".note.GNU-stack","", at progbits
.addrsig
+ .addrsig_sym F
.addrsig_sym C
.addrsig_sym B
.addrsig_sym A
diff --git a/lld/test/MachO/bp-section-orderer.s b/lld/test/MachO/bp-section-orderer.s
index 2eaff04bdc04700..f0ac1bd99f7227a 100644
--- a/lld/test/MachO/bp-section-orderer.s
+++ b/lld/test/MachO/bp-section-orderer.s
@@ -42,13 +42,15 @@
# RUN: %no-fatal-warnings-lld -arch arm64 -lSystem -e _main -o %t/a.out %t/a.o --verbose-bp-section-orderer --compression-sort=both --irpgo-profile-sort=%t/a.profdata 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
# RUN: %lld -arch arm64 -lSystem -e _main -o %t/a.out %t/a.o --verbose-bp-section-orderer --bp-compression-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-FUNC
+# RUN: %lld -arch arm64 -lSystem -e _main -o %t/a.out %t/a.o --verbose-bp-section-orderer --bp-compression-sort=function --icf=all 2>&1 | FileCheck %s --check-prefix=COMPRESSION-ICF-FUNC
# RUN: %lld -arch arm64 -lSystem -e _main -o %t/a.out %t/a.o --verbose-bp-section-orderer --bp-compression-sort=data 2>&1 | FileCheck %s --check-prefix=COMPRESSION-DATA
# RUN: %lld -arch arm64 -lSystem -e _main -o %t/a.out %t/a.o --verbose-bp-section-orderer --bp-compression-sort=both 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
# RUN: %lld -arch arm64 -lSystem -e _main -o %t/a.out %t/a.o --verbose-bp-section-orderer --bp-compression-sort=both --irpgo-profile=%t/a.profdata --bp-startup-sort=function 2>&1 | FileCheck %s --check-prefix=COMPRESSION-BOTH
-# COMPRESSION-FUNC: Ordered 7 sections using balanced partitioning
+# COMPRESSION-FUNC: Ordered 9 sections using balanced partitioning
+# COMPRESSION-ICF-FUNC: Ordered 7 sections using balanced partitioning
# COMPRESSION-DATA: Ordered 7 sections using balanced partitioning
-# COMPRESSION-BOTH: Ordered 14 sections using balanced partitioning
+# COMPRESSION-BOTH: Ordered 16 sections using balanced partitioning
#--- a.s
.text
@@ -78,6 +80,12 @@ F:
add w0, w0, #3
bl l_C.__uniq.111111111111111111111111111111111111111.llvm.2222222222222222222
ret
+merged1:
+ add w0, w0, #101
+ ret
+merged2:
+ add w0, w0, #101
+ ret
.data
s1:
>From b65e7a132fb4416e22962ac41c887160ece7cb2f Mon Sep 17 00:00:00 2001
From: Ellis Hoag <ellishoag at meta.com>
Date: Mon, 10 Feb 2025 09:22:02 -0800
Subject: [PATCH 2/2] Skip live sections for ELF
---
lld/ELF/BPSectionOrderer.cpp | 7 +++++--
lld/MachO/BPSectionOrderer.cpp | 2 ++
lld/test/ELF/bp-section-orderer.s | 23 +++++++++++++++--------
3 files changed, 22 insertions(+), 10 deletions(-)
diff --git a/lld/ELF/BPSectionOrderer.cpp b/lld/ELF/BPSectionOrderer.cpp
index 01f77b33926f755..4adb42ef4ff9365 100644
--- a/lld/ELF/BPSectionOrderer.cpp
+++ b/lld/ELF/BPSectionOrderer.cpp
@@ -75,8 +75,11 @@ DenseMap<const InputSectionBase *, int> elf::runBalancedPartitioning(
auto *d = dyn_cast<Defined>(&sym);
if (!d)
return;
- auto *sec = dyn_cast_or_null<InputSectionBase>(d->section);
- if (!sec || sec->size == 0 || !orderer.secToSym.try_emplace(sec, d).second)
+ auto *sec = dyn_cast_or_null<InputSection>(d->section);
+ // Skip empty, discarded, ICF folded sections. Skipping ICF folded sections
+ // reduces duplicate detection work in BPSectionOrderer.
+ if (!sec || sec->size == 0 || !sec->isLive() || sec->repl != sec ||
+ !orderer.secToSym.try_emplace(sec, d).second)
return;
rootSymbolToSectionIdxs[CachedHashStringRef(getRootSymbol(sym.getName()))]
.insert(sections.size());
diff --git a/lld/MachO/BPSectionOrderer.cpp b/lld/MachO/BPSectionOrderer.cpp
index f693e5e59f8faea..950afd0421f0620 100644
--- a/lld/MachO/BPSectionOrderer.cpp
+++ b/lld/MachO/BPSectionOrderer.cpp
@@ -117,6 +117,8 @@ DenseMap<const InputSection *, int> lld::macho::runBalancedPartitioning(
auto *isec = subsec.isec;
if (!isec || isec->data.empty())
continue;
+ // ConcatInputSections are entirely live or dead, so the offset is
+ // irrelevant.
if (isa<ConcatInputSection>(isec) && !isec->isLive(0))
continue;
size_t idx = sections.size();
diff --git a/lld/test/ELF/bp-section-orderer.s b/lld/test/ELF/bp-section-orderer.s
index 6a61e7e8cef2b8d..4889db63cd4d18d 100644
--- a/lld/test/ELF/bp-section-orderer.s
+++ b/lld/test/ELF/bp-section-orderer.s
@@ -19,7 +19,7 @@
# RUN: llvm-mc -filetype=obj -triple=aarch64 a.s -o a.o
# RUN: llvm-profdata merge a.proftext -o a.profdata
-# RUN: ld.lld a.o --irpgo-profile=a.profdata --bp-startup-sort=function --verbose-bp-section-orderer --icf=all 2>&1 | FileCheck %s --check-prefix=STARTUP-FUNC-ORDER
+# RUN: ld.lld a.o --irpgo-profile=a.profdata --bp-startup-sort=function --verbose-bp-section-orderer --icf=all --gc-sections 2>&1 | FileCheck %s --check-prefix=STARTUP-FUNC-ORDER
# STARTUP-FUNC-ORDER: Ordered 3 sections using balanced partitioning
# STARTUP-FUNC-ORDER: Total area under the page fault curve: 3.
@@ -33,7 +33,7 @@
# ORDER-STARTUP: s2,s1,s5,s4,s3,A,F,E,D,B,C,merged1,merged2,_start,d3,d2,d4,d1,{{$}}
# RUN: ld.lld -o out.cf a.o --verbose-bp-section-orderer --bp-compression-sort=function 2>&1 | FileCheck %s --check-prefix=BP-COMPRESSION-FUNC
-# RUN: ld.lld -o out.cf.icf a.o --verbose-bp-section-orderer --bp-compression-sort=function --icf=all 2>&1 | FileCheck %s --check-prefix=BP-COMPRESSION-ICF-FUNC
+# RUN: ld.lld -o out.cf.icf a.o --verbose-bp-section-orderer --bp-compression-sort=function --icf=all --gc-sections 2>&1 | FileCheck %s --check-prefix=BP-COMPRESSION-ICF-FUNC
# RUN: llvm-nm -jn out.cf | tr '\n' , | FileCheck %s --check-prefix=CFUNC
# CFUNC: s5,s4,s3,s2,s1,A,F,merged1,merged2,C,E,D,B,_start,d4,d3,d2,d1,{{$}}
@@ -118,18 +118,22 @@ int d3[] = {5,6,7,8};
int d2[] = {7,8,9,10};
int d1[] = {3,4,5,6};
+// used is to suppress compiler garbage collection in ELF; retain is to suppress linker garbage collection; used is not needed for non-internal linkage symbols
+// used is for both compiler/linker GC in Mach-O; retain is ignored for Mach-O
+#define RETAIN [[gnu::used,gnu::retain]]
+
int C(int a);
int B(int a);
void A();
int F(int a) { return C(a + 3); }
-int E(int a) { return C(a + 2); }
-int D(int a) { return B(a + 2); }
+RETAIN int E(int a) { return C(a + 2); }
+RETAIN int D(int a) { return B(a + 2); }
int C(int a) { A(); return a + 2; }
int B(int a) { A(); return a + 1; }
void A() {}
-int merged1(int a) { return F(a + 101); }
+RETAIN int merged1(int a) { return F(a + 101); }
int merged2(int a) { return F(a + 101); }
int _start() { return 0; }
@@ -176,7 +180,7 @@ C: // @C
.Lfunc_end1:
.size C, .Lfunc_end1-C
// -- End function
- .section .text.E,"ax", at progbits
+ .section .text.E,"axR", at progbits
.globl E // -- Begin function E
.p2align 2
.type E, at function
@@ -195,7 +199,7 @@ E: // @E
.Lfunc_end2:
.size E, .Lfunc_end2-E
// -- End function
- .section .text.D,"ax", at progbits
+ .section .text.D,"axR", at progbits
.globl D // -- Begin function D
.p2align 2
.type D, at function
@@ -243,7 +247,7 @@ A: // @A
.Lfunc_end5:
.size A, .Lfunc_end5-A
// -- End function
- .section .text.merged1,"ax", at progbits
+ .section .text.merged1,"axR", at progbits
.globl merged1 // -- Begin function merged1
.p2align 2
.type merged1, at function
@@ -377,5 +381,8 @@ d1:
.addrsig
.addrsig_sym F
.addrsig_sym C
+ .addrsig_sym E
+ .addrsig_sym D
.addrsig_sym B
.addrsig_sym A
+ .addrsig_sym merged1
More information about the llvm-commits
mailing list