Re: tree-ssa branch: Call for preliminary testing

[qiongc at cse dot unsw dot edu dot au (Qiong Cai)]

bootstrap on a 933Mhz PIII Linux 2.4.20-workstation with 256MB.

> ../gcc/configure --prefix=/import/qin/1/qiongc/gcc-ssa --enable-languages=c --disable-libgcj : (reconfigured)
Thread model: posix
gcc version 3.5-tree-ssa 20030504 (experimental)

15 C programs from SPEC2000 are compiled and run with test input.

Compiling times (user time) in secs are shown  as follows for those 
successfully compiled. In (x/y), x is compiling time
by gcc 3.3 prerelease and y compiling time by gcc ssa.

Compiling and Running Passed:
gzip(4.5/6.55), gcc(113.86/176.88), mcf(2.92/3.94), crafty(22.10/33.64), 
parser(9.26/12.93), 
gap(40.89/62.00), bzip2(2.66/3.65), twolf(26.32/42.56), mesa(46.36/69.02), 
art(1.49/1.96), 
equake(1.84/2.68), ammp(13.21/20.39)

Compiling Passed:
vpr(12.87/18.81), vortex(46.68/68.43)

Compiling Error:
1) perlbmk
universal.c: In function `XS_UNIVERSAL_isa':
universal.c:123: internal compiler error: in simplify_stmt_expr, at c-simplify.c:991

Running Error:
1) vpr
unning 175.vpr test base x86_linux default
*** Miscompare of route_log.out, see /import/qin/1/qiongc/cpu2000/benchspec/CINT2000/175.vpr/run/00000002/route_log.out.mis
Invalid run; unable to continue.  If you wish to ignore errors please use '-I' or ignore_errors

2) vortex
Running 255.vortex test base x86_linux default
*** Miscompare of vortex.out, see /import/qin/1/qiongc/cpu2000/benchspec/CINT2000/255.vortex/run/00000002/vortex.out.mis

And the compiling base flag is -O2 -march=i686 -fomit-frame-pointer.

BTW, I haven't tried SSAPRE. Is the implemenation complete now?


Thanks.

Qiong

On Wed, May 07, 2003 at 10:25:03AM -0400, Diego Novillo wrote:
> I'm looking for volunteers to try out the tree-ssa branch with
> their favourite C/C++ code.  Be warned that at this time you
> shouldn't expect the branch compiler to be any better than
> mainline.  In fact, it will probably be worse.
> 
> This call for testing is mostly to gather data that will help
> finding the most obvious problems with the branch.  Besides the
> obvious correctness issues, I'm mostly concerned with compile and
> run time performance.
> 
> If you're willing to help, I'd like to know:
> 
> - Compile time comparisons vs mainline (-ftime-report or gprof).
> 
> 
> - Run time comparisons vs mainline (oprofile or gprof or other
>   profilers).
>   
>   Sometimes it's useful to compare the pretty-printed dumps given
>   by -fdump-tree-all.  Particularly, comparing .t05.simple vs
>   .t14.optimized shows the difference before and after
>   optimization.
> 
>   Also, comparing the .00.rtl between the branch and mainline may
>   give hints.  Direct comparison on large codes will probably not
>   be useful because the branch generates RTL from GIMPLE trees,
>   while mainline generates RTL directly from FE trees.
> 
> 
> - Correctness.  Does the branch compiler fail or generate
>   incorrect code?  Does it fail to bootstrap on your platform?
> 
> 
> Thanks.  Diego.