[lld] [ELF] Making cdsort default for function reordering (PR #68638)

via llvm-commits llvm-commits at lists.llvm.org
Mon Oct 9 19:25:58 PDT 2023


llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-lld

Author: None (spupyrev)

<details>
<summary>Changes</summary>

CDSort function reordering outperforms the existing default heuristic (
hfsort/C^3) in terms of the performance of generated binaries while
being (almost) as fast. Thus, the suggestion is to change the default.
The speedup is up to 1.5% perf for large front-end binaries, and can be
moderate/neutral for "small" benchmarks.

High-level **perf impact** on two selected binaries:
clang-10 binary (built with LTO+AutoFDO/CSSPGO): wins on top of C^3 in [0.3%..0.8%]
rocksDB-8 binary (built with LTO+CSSPGO): wins on top of C^3 in [0.8%..1.5%]

More detailed measurements on the clang binary is at [here](https://reviews.llvm.org/D152834#<!-- -->4445042)

---
Full diff: https://github.com/llvm/llvm-project/pull/68638.diff


8 Files Affected:

- (modified) lld/ELF/Driver.cpp (+1-1) 
- (modified) lld/docs/ld.lld.1 (+2-2) 
- (modified) lld/test/ELF/cgprofile-bad-clusters.s (+1-1) 
- (modified) lld/test/ELF/cgprofile-icf.s (+2-2) 
- (modified) lld/test/ELF/cgprofile-print.s (+1-4) 
- (modified) lld/test/ELF/cgprofile-rela.test (+1-1) 
- (modified) lld/test/ELF/cgprofile-reproduce.s (+1-4) 
- (modified) lld/test/ELF/cgprofile-txt.s (+3-3) 


``````````diff
diff --git a/lld/ELF/Driver.cpp b/lld/ELF/Driver.cpp
index 6272276e94b2d35..e2100d00d54ede6 100644
--- a/lld/ELF/Driver.cpp
+++ b/lld/ELF/Driver.cpp
@@ -1095,7 +1095,7 @@ static void ltoValidateAllVtablesHaveTypeInfos(opt::InputArgList &args) {
 }
 
 static CGProfileSortKind getCGProfileSortKind(opt::InputArgList &args) {
-  StringRef s = args.getLastArgValue(OPT_call_graph_profile_sort, "hfsort");
+  StringRef s = args.getLastArgValue(OPT_call_graph_profile_sort, "cdsort");
   if (s == "hfsort")
     return CGProfileSortKind::Hfsort;
   if (s == "cdsort")
diff --git a/lld/docs/ld.lld.1 b/lld/docs/ld.lld.1
index 2e46fc18132f3e0..12b17dd37796d13 100644
--- a/lld/docs/ld.lld.1
+++ b/lld/docs/ld.lld.1
@@ -128,9 +128,9 @@ may be:
 .It Cm none
 Ignore call graph profile.
 .It Cm hfsort
-Use hfsort (default).
+Use hfsort.
 .It Cm cdsort
-Use cdsort.
+Use cdsort (default).
 .El
 .Pp
 .It Fl -color-diagnostics Ns = Ns Ar value
diff --git a/lld/test/ELF/cgprofile-bad-clusters.s b/lld/test/ELF/cgprofile-bad-clusters.s
index c162e981acdd633..88e68bfb7b2c0a1 100644
--- a/lld/test/ELF/cgprofile-bad-clusters.s
+++ b/lld/test/ELF/cgprofile-bad-clusters.s
@@ -10,7 +10,7 @@
 # RUN: echo "F G 6" >> %t.call_graph
 # RUN: echo "G H 5" >> %t.call_graph
 # RUN: echo "H I 4" >> %t.call_graph
-# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t2
+# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph --call-graph-profile-sort=hfsort  -o %t2
 # RUN: llvm-readobj --symbols %t2 | FileCheck %s
 
     .section    .text.A,"ax", at progbits
diff --git a/lld/test/ELF/cgprofile-icf.s b/lld/test/ELF/cgprofile-icf.s
index a9de5613917cbf1..e28630d0eb30bf0 100644
--- a/lld/test/ELF/cgprofile-icf.s
+++ b/lld/test/ELF/cgprofile-icf.s
@@ -5,9 +5,9 @@
 # RUN: echo "A B 100" > %t.call_graph
 # RUN: echo "A C 40" >> %t.call_graph
 # RUN: echo "C D 61" >> %t.call_graph
-# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t.out -icf=all
+# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph --call-graph-profile-sort=hfsort -o %t.out -icf=all
 # RUN: llvm-readobj --symbols %t.out | FileCheck %s
-# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t2.out
+# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph --call-graph-profile-sort=hfsort -o %t2.out
 # RUN: llvm-readobj --symbols %t2.out | FileCheck %s --check-prefix=NOICF
 
     .section    .text.D,"ax", at progbits
diff --git a/lld/test/ELF/cgprofile-print.s b/lld/test/ELF/cgprofile-print.s
index b103ef5109effbb..d6f15f2926b5729 100644
--- a/lld/test/ELF/cgprofile-print.s
+++ b/lld/test/ELF/cgprofile-print.s
@@ -5,7 +5,7 @@
 # RUN: echo "B C 50" >> %t.call_graph
 # RUN: echo "C D 40" >> %t.call_graph
 # RUN: echo "D B 10" >> %t.call_graph
-# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t2 --print-symbol-order=%t3
+# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t2 --call-graph-profile-sort=hfsort --print-symbol-order=%t3
 # RUN: FileCheck %s --input-file %t3
 
 # CHECK: B
@@ -32,6 +32,3 @@ C:
 .globl  D
 D:
  nop
-
-
-
diff --git a/lld/test/ELF/cgprofile-rela.test b/lld/test/ELF/cgprofile-rela.test
index 189f169e65481ee..141dfd4c65b1ea1 100644
--- a/lld/test/ELF/cgprofile-rela.test
+++ b/lld/test/ELF/cgprofile-rela.test
@@ -3,7 +3,7 @@
 # REQUIRES: x86
 
 # RUN: yaml2obj %s -o %t.o
-# RUN: ld.lld %t.o -o %t
+# RUN: ld.lld --call-graph-profile-sort=hfsort %t.o -o %t
 # RUN: llvm-nm --no-sort %t | FileCheck %s
 # RUN: ld.lld --no-call-graph-profile-sort %t.o -o %t
 # RUN: llvm-nm --no-sort %t | FileCheck %s --check-prefix=NO-CG
diff --git a/lld/test/ELF/cgprofile-reproduce.s b/lld/test/ELF/cgprofile-reproduce.s
index b9cb269e4580d78..1b1b36151da99d1 100644
--- a/lld/test/ELF/cgprofile-reproduce.s
+++ b/lld/test/ELF/cgprofile-reproduce.s
@@ -5,7 +5,7 @@
 # RUN: echo "B C 50" >> %t.call_graph
 # RUN: echo "C D 40" >> %t.call_graph
 # RUN: echo "D B 10" >> %t.call_graph
-# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t2 --print-symbol-order=%t3
+# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t2 --call-graph-profile-sort=hfsort --print-symbol-order=%t3
 # RUN: ld.lld -e A %t --symbol-ordering-file %t3 -o %t2
 # RUN: llvm-readobj --symbols %t2 | FileCheck %s
 
@@ -37,6 +37,3 @@ C:
 .globl  D
 D:
  nop
-
-
-
diff --git a/lld/test/ELF/cgprofile-txt.s b/lld/test/ELF/cgprofile-txt.s
index c9194bbbc43cbe0..cf5b17627cfb63c 100644
--- a/lld/test/ELF/cgprofile-txt.s
+++ b/lld/test/ELF/cgprofile-txt.s
@@ -26,12 +26,12 @@
 # RUN: echo "TooManyPreds10 TooManyPreds 11" >> %t.call_graph
 # RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph --call-graph-profile-sort=hfsort -o %t2
 # RUN: llvm-readobj --symbols %t2 | FileCheck %s
-## --call-graph-profile-sort=hfsort is the default.
-# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t2b
-# RUN: cmp %t2 %t2b
 
 # RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph --call-graph-profile-sort=cdsort -o %t2
 # RUN: llvm-readobj --symbols %t2 | FileCheck %s --check-prefix=CDSORT
+## --call-graph-profile-sort=cdsort is the default.
+# RUN: ld.lld -e A %t --call-graph-ordering-file %t.call_graph -o %t2b
+# RUN: cmp %t2 %t2b
 
 # RUN: not ld.lld -e A %t --call-graph-ordering-file %t.call_graph --call-graph-profile-sort=sort \
 # RUN:   -o /dev/null 2>&1 | FileCheck %s --check-prefix=UNKNOWN

``````````

</details>


https://github.com/llvm/llvm-project/pull/68638


More information about the llvm-commits mailing list