答复: [COMMIT] romcc 0.63 and some fixes...

[YhLu]

I'd like to wait it to produce smaller code. Otherwise if I enable superio +
ati support, It will be greater than 64K in debug mode.

When will that be ready?

Regards

YH

-----邮件原件-----
发件人: [e-mail] [mailto:[e-mail]] 
发送时间: 2004年5月28日 7:26
收件人: LinuxBIOS
主题: [COMMIT] romcc 0.63 and some fixes...

I finally have romcc to the point where I can seriously look
at size optimizations.  I am finally there after a lot of bug fixing
and experimentation.

Because of added correctness in handling of small types I have
measured about a 1K growth on the hdama port or about 3%.  But things
should start shrinking soon.

More optimization opportunities have been exposed so which breaks
some code that was not right but lucky.

Have fun with this.

Eric

- Upgrade to romcc version 0.63
  This includes more test cases
  Lots of small bug fixes
  A complete C preprocessor is now built in
  Initial support for not inlining everything
  __attribute__((noinline)) works
  Better command line options and help
  Constants arrays can be read at compile time
  Asm statements that are not volatile will now be removed when their
outputs go unused
  Loads and stores that are not volatile will be removed when their values
go unused
  The number of FIXMES in the code is finally starting to go down.

Tags romcc_0_38 and romcc_0_63 are present.  In case you need to regress the
version of romcc.

- Added volatile to asm statements in auto.c and failover.c
- Updated the romcc version in Config.lb
- Fixed type sizes in romcc_io.h and io.h inl() returning a byte was nasty

Modified Files:
    src/arch/i386/include/arch/romcc_io.h
    src/arch/i386/include/arch/io.h
    src/mainboard/arima/hdama/auto.c
    src/mainboard/arima/hdama/failover.c src/config/Config.lb

_______________________________________________
Linuxbios mailing list
[e-mail]
http://www.clustermatic.org/mailman/listinfo/linuxbios

YhLu  writes:
> I'd like to wait it to produce smaller code. Otherwise if I enable superio +
> ati support, It will be greater than 64K in debug mode.
>
> When will that be ready?

The code is feature complete so it will correctly not inline functions
(producing smaller code) if it does not run out of registers.

My simple test cases have worked well.

My complicated test case that uses an old version of production code
ran out of registers after consuming just the 8 general purpose
registers.  So mostly it is a matter of playing with the code, and
figuring out why weird things are happening.

Eric