http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50165
Bug #: 50165
Summary: [4.7 regression] Huge build time regression (Firefox
lto build)
Classification: Unclassified
Product: gcc
Version: 4.7.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: lto
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: mar...@trippelsdorf.de
CC: dnovi...@google.com
Since Diegos patches (rev. 177571, 177661, 177704) went in
lto1 seems to run forever (I've killed it after 40 minutes) when libxul
is linked during a Firefox LTO build.
It used to finish in 4-5 minutes before.
This is a profile captured during the lto1 run:
("perf record -ag sleep 5" and then "perf report -v -g --stdio")
# Events: 5K cycles
#
# Overhead Command Shared Object
Symbol
# ........ ............... ...............................................
..........................................................
#
93.93% lto1-wpa /usr/libexec/gcc/x86_64-pc-linux-gnu/4.7.0/lto1
0x74a633 d [.] htab_expand
|
--- htab_expand
1.05% swapper [kernel.kallsyms]
0xffffffff810460c2 k [k] default_idle
...
and sometimes (much more infrequent) this profile:
40.25% lto1-wpa /usr/libexec/gcc/x86_64-pc-linux-gnu/4.7.0/lto1
0x74ad05 d [.] htab_find_slot_with_hash
|
--- htab_find_slot_with_hash
30.74% lto1-wpa /usr/libexec/gcc/x86_64-pc-linux-gnu/4.7.0/lto1
0x225b10 d [.] eq_string_slot_node
|
--- eq_string_slot_node
2.06% lto1-wpa /lib/libz.so.1.2.5
...
