[Bug ipa/108130] New: [13 Regression] LTO compile time hog seen on bootstrap-lto config since r13-4684-g7450b25566b7a7

[Bug ipa/108130] New: [13 Regression] LTO compile time hog seen on bootstrap-lto config since r13-4684-g7450b25566b7a7

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108130

            Bug ID: 108130
           Summary: [13 Regression] LTO compile time hog seen on
                    bootstrap-lto config since r13-4684-g7450b25566b7a7
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Keywords: compile-time-hog
          Severity: normal
          Priority: P3
         Component: ipa
          Assignee: unassigned at gcc dot gnu.org
          Reporter: marxin at gcc dot gnu.org
                CC: jamborm at gcc dot gnu.org, marxin at gcc dot gnu.org
  Target Milestone: ---

I noticed linking of e.g. cc1plus gets much slower in stage2 with a WPA hog in:

    58.15%  lto1                [.] lto_symtab_encoder_lookup
    19.95%  lto1                [.] may_need_named_section_p
    18.77%  lto1                [.] rename_statics
     0.50%  lto1                [.] lto_symtab_encoder_lookup
     0.09%  lto1                [.] lto_symtab_encoder_encode_body_p

The WPA phase takes >5 minutes, while it used to take ~30 seconds on my
machine.

[Bug ipa/108130] [13 Regression] LTO compile time hog seen on bootstrap-lto config since r13-4684-g7450b25566b7a7

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108130

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2022-12-15
           Priority|P3                          |P1
             Status|UNCONFIRMED                 |NEW
   Target Milestone|---                         |13.0

[Bug ipa/108130] [13 Regression] LTO compile time hog seen on bootstrap-lto config since r13-4684-g7450b25566b7a7

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108130

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
We probably have all the "external" summaries multiple times now?

[Bug ipa/108130] [13 Regression] LTO compile time hog seen on bootstrap-lto config since r13-4684-g7450b25566b7a7

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108130

--- Comment #2 from Martin Liška <marxin at gcc dot gnu.org> ---
Apparently, Godot links for more than 2 hours now:
https://build.opensuse.org/package/live_build_log/home:marxin:home:marxin:gcc-periodic-testing-v2/godot/openSUSE_Tumbleweed/x86_64

[Bug ipa/108130] [13 Regression] LTO compile time hog seen on bootstrap-lto config since r13-4684-g7450b25566b7a7

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108130

--- Comment #3 from Martin Liška <marxin at gcc dot gnu.org> ---
I noticed one more issue starting with this revision and it's linker error when
building systemd-mini package:

https://build.opensuse.org/package/live_build_log/home:marxin:home:marxin:gcc-periodic-testing-v2/systemd-mini/openSUSE_Tumbleweed/x86_64

[  114s] /usr/bin/cc -o src/boot/efi/linuxx64.elf.stub -DGNU_EFI_USE_MS_ABI
-DSD_BOOT -ffreestanding -fshort-wchar -fvisibility=hidden -I
/home/abuild/rpmbuild/BUILD/systemd-v252.3+suse.40.gbf3fef9988/src/fundamental
-I /home/abuild/rpmbuild/BUILD/systemd-v252.3+suse.40.gbf3fef9988/src/boot/efi
-include src/boot/efi/efi_config.h -include version.h -isystem
/usr/include/efi/x86_64 -isystem /usr/include/efi -std=gnu11 -Wall -Wextra
-Wno-missing-field-initializers -Wno-unused-parameter -Wdate-time
-Wendif-labels -Werror=format=2 -Werror=format-signedness
-Werror=implicit-function-declaration -Werror=incompatible-pointer-types
-Werror=int-conversion -Werror=overflow -Werror=override-init
-Werror=return-type -Werror=shift-count-overflow -Werror=shift-overflow=2
-Werror=undef -Wfloat-equal -Wimplicit-fallthrough=5 -Winit-self -Wlogical-op
-Wmissing-include-dirs -Wmissing-noreturn -Wnested-externs
-Wold-style-definition -Wpointer-arith -Wredundant-decls -Wshadow
-Wstrict-aliasing=2 -Wstrict-prototypes -Wsuggest-attribute=noreturn
-Wunused-function -Wwrite-strings -Wno-maybe-uninitialized -Wno-unused-result
-ftrivial-auto-var-init=zero -fno-stack-protector -fno-strict-aliasing -fpic
-fwide-exec-charset=UCS2 -O1 -mno-red-zone -mno-sse -mno-mmx -O2 -flto=auto
-fuse-ld=bfd -L /usr/lib64 -nostdlib -T /usr/lib64/elf_x86_64_efi.lds
-Wl,--build-id=sha1 -Wl,--fatal-warnings -Wl,--no-undefined -Wl,--warn-common
-Wl,-Bsymbolic -z nocombreloc /usr/lib64/crt0-efi-x86_64.o
-Wl,--no-warn-execstack -Wl,--no-warn-rwx-segments -pie -Wl,--no-dynamic-linker
src/boot/efi/bootspec-fundamental.c.o src/boot/efi/efivars-fundamental.c.o
src/boot/efi/sha256.c.o src/boot/efi/string-util-fundamental.c.o
src/boot/efi/tpm-pcr.c.o src/boot/efi/assert.c.o src/boot/efi/console.c.o
src/boot/efi/devicetree.c.o src/boot/efi/disk.c.o src/boot/efi/efi-string.c.o
src/boot/efi/graphics.c.o src/boot/efi/initrd.c.o src/boot/efi/measure.c.o
src/boot/efi/pe.c.o src/boot/efi/secure-boot.c.o src/boot/efi/ticks.c.o
src/boot/efi/util.c.o src/boot/efi/cpio.c.o src/boot/efi/linux.c.o
src/boot/efi/splash.c.o src/boot/efi/stub.c.o src/boot/efi/linux_x86.c.o -lefi
-lgnuefi -lgcc
[  114s] /usr/bin/ld.bfd: /tmp/ccfbAVRm.ltrans0.ltrans.o: in function
`locate_sections.constprop.0':
[  114s] <artificial>:(.text+0x360c): undefined reference to `memcmp'
[  114s] collect2: error: ld returned 1 exit status

note it's a freestanding environment and the symbol memcmp is defined by
systemd. Note -flto-partition=one does not help us here.

memcmp.part.0/2514 (memcmp.part.0)
  Type: function definition analyzed
  Visibility: semantic_interposition artificial
  References: 
  Referring: 
  Read from file: /tmp/ccnpwegs.ltrans0.o
  Function memcmp.part.0/2514 is inline copy in efi_main/1352
  Clone of memcmp.part.0/2216
  Unit id: 10
  Function flags: count:29068074 (estimated locally) local split_part
nonfreeing_fn
  Called by: memcmp.lto_priv.0/2513 (inlined) (29068074 (estimated
locally),0.10 per call) 
  Calls: 
memcmp.lto_priv.0/2513 (memcmp)
  Type: function definition analyzed
  Visibility: semantic_interposition public visibility:hidden
  References: 
  Referring: 
  Read from file: /tmp/ccnpwegs.ltrans0.o
  Function memcmp/2513 is inline copy in efi_main/1352
  Clone of memcmp.lto_priv.0/503
  Unit id: 10
  Function flags: count:44042536 (estimated locally) local nonfreeing_fn
  Called by: is_direct_boot/1927 (inlined) (44042536 (estimated locally),0.15
per call) 
  Calls: memcmp.part.0/2514 (inlined) (29068074 (estimated locally),0.10 per
call)

Unfortunately, I can't easily reduce a self-contained test-case. Please let me
know on IRC and can debug it.

[Bug ipa/108130] [13 Regression] LTO compile time hog seen on bootstrap-lto config since r13-4684-g7450b25566b7a7

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108130

Martin Jambor <jamborm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|NEW                         |RESOLVED

--- Comment #4 from Martin Jambor <jamborm at gcc dot gnu.org> ---
This has been fixed by Honza's r13-4998-gedc038863d88ef

posted in https://gcc.gnu.org/pipermail/gcc-patches/2023-January/609406.html

If the undefined reference problem persists, it is a different issue.