Re: Results for egcs-2.92.22 19981122 (gcc2 ss-980609 experimental) testsuite on i586-pc-linux-gnu

[Nix <nix at esperi dot demon dot co dot uk>]

Jeffrey A Law writes:
>   > -O9 ICEs in stage1 with
> Err, I see man more options than just -O9.  If you want us to look into these
> problems in the future, you'll have to supply accurate information.

Er, yes. I know. What I meant was `the build which differs from all my
other builds solely by supplying an -O9 flag rather than, for
instance, an -O2'. I rearranged the mail during writing and somehow
the full options list ended up below this bit when it started out
above :( sorry.

>                                                                     When you
> do not provide this information

But I did! Look below.

>   > ./xgcc -B./ -B/usr/packages.bin/egcs/snapshot/i586-pc-linux-gnu/bin/ -O2   
[snip]
> This must be stage2, the C++ compiler is not built in stage1.

*arghwhoops*

My brain must have been *melted* when I wrote stage1. It's stage2, of
course. (*Using* the stage1 compiler, which was what I was
thinking. My fingers ran away and typed the wrong thing :( pardon me.)

> ANd I still don't know how the stage1 compiler was built.  I've got better

The BOOT_CFLAGS were

 -W -Wall -O[19] -funroll-all-loops -g -march=pentium -mcpu=pentium -fomit-frame-pointer

(depending on the optimisation level, of course).

Unless I'm much mistaken I listed them in the mail you've just responded
to. And in every test result, viz

configure flags: --with-gnu-ld --with-gnu-as --prefix=/usr/packages.bin/egcs/snapshot --enable-shared --disable-haifa
BOOT_CC:         egcs-2.91.57
BOOT_CFLAGS:     -W -Wall -O1 -funroll-all-loops -g -march=pentium -mcpu=pentium -fomit-frame-pointer
CFLAGS:          -W -Wall -O1 -march=pentium -mcpu=pentium -fomit-frame-pointer -g -D__NO_MATH_INLINES
CXXFLAGS:        -W -Wall -O1 -march=pentium -mcpu=pentium -fomit-frame-pointer -g -D__NO_MATH_INLINES

> you just started saying "-O1 and -O9" fail, without mentioning that you
> were still using -funroll-loops -fomit-frame-pointer and several more options.

Er, I *was* listing all the other options as well (just not in that
comment). If I wasn't being clear enough, I apologise; I'll try to
improve the comment clarity in future. I suppose it wouldn't have been
obvious to one who saw the results trickle in one by one that they
were all using similar options - this is a disadvantage of the
perspective I've got, forwarding all the mails out at once :(

My test cycle uses a constant base set of options, testing all
combinations of -O[0129s] and --(enable|disable)-haifa with those
options. I wonder if I should revise my test reports to make that
clear somehow, somewhere?  *muses*

-- 
`Anyone who says you can have a lot of widely dispersed people hack
 away on a complicated piece of code and avoid total anarchy has never
 managed a software project.' - Andy Tanenbaum in 1992 on comp.os.minix