10835 – [3.3/3.4 regression] combinatory explosion in scheduler on HyperSPARC (original) (raw)

When I try making bootstrap, the stage 2 compilation hangs when compiling c-typeck.c. I aborted the process after 12 hours CPU time. When compiling the file with the same flags but -mv8 or -msupersparc, the compilation completes in 1.5 minutes.

I configured gcc with the following options:

--with-cpu=hypersparc --disable-multilib --enable-shared=libstdc++,libffi,zlib,boehm-gc,libjava

Release: 3.3

Environment: SunOS continuity 5.8 Generic_108528-20 sun4m sparc SUNW,SPARCstation-10

Dual Ross RT625 processors @ 100MHz

How-To-Repeat: Bootstrap 3.3 on SPARC with "--with-cpu=hypersparc"

Comment 1 marcus 2003-05-17 13:16:00 UTC

Fix: Workaround: Use "--with-cpu=v8" instead.

Comment 2 Eric Botcazou 2003-05-20 16:40:21 UTC

From: Eric Botcazou <ebotcazou@libertysurf.fr> To: marcus@mc.pp.se Cc: gcc-gnats@gcc.gnu.org, gcc-bugs@gcc.gnu.org Subject: Re: bootstrap/10835: bootstraps hangs on hypersparc Date: Tue, 20 May 2003 16:40:21 +0200

When I try making bootstrap, the stage 2 compilation hangs when compiling c-typeck.c. I aborted the process after 12 hours CPU time. When compiling the file with the same flags but -mv8 or -msupersparc, the compilation completes in 1.5 minutes.

I could reproduce with an already bootstrapped compiler: compile c-typeck.c at -O2 -mcpu=hypersparc. The compiler seems to be hung in the haïfa scheduler, more specifically max_issue(), while compiling really_start_incremental_init().

Eric Botcazou

Comment 3 Eric Botcazou 2003-05-30 15:47:05 UTC

This is a regression from GCC 3.2.3 due to the new DFA scheduler.

Comment 4 Eric Botcazou 2003-06-03 10:38:01 UTC

Created attachment 4152 [details] Reduced testcase

The multiple independent assignments to the structures drive the DFA scheduler crazy:-)

Comment 7 Eric Botcazou 2003-06-24 12:06:19 UTC

Thanks Vlad for fixing this one.