Daily snapshots of tree-ssa branch available

Daily snapshots of tree-ssa branch available

[Diego Novillo]

I received requests to make source snapshots of the tree-ssa
branch.  The snapshots are going to be available at

http://people.redhat.com/dnovillo/pub/snapshot/

The SPEC tester will upload a tarball named tree-ssa-YYYYMMDD.tar.bz
after a successful bootstrap.


Diego.

Re: Daily snapshots of tree-ssa branch available

[Toon Moene]

Diego Novillo wrote:

> I received requests to make source snapshots of the tree-ssa
> branch.  The snapshots are going to be available at
> 
> http://people.redhat.com/dnovillo/pub/snapshot/
> 
> The SPEC tester will upload a tarball named tree-ssa-YYYYMMDD.tar.bz
> after a successful bootstrap.

Good idea !

BTW, without http://gcc.gnu.org/ml/gcc-patches/2003-05/msg00575.html the 
tree-ssa branch won't build on Alpha.

Oh, and you need about a dozen no-warn entries added to 
gcc/gcc/Makefile.in :-(

For test results (c,f77,f95 only), see:

http://gcc.gnu.org/ml/gcc-testresults/2003-05/msg00630.html

Cheers,

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

Re: Daily snapshots of tree-ssa branch available

[Diego Novillo]

On Sat, 2003-05-10 at 04:06, Toon Moene wrote:
> Diego Novillo wrote:
> 
> > I received requests to make source snapshots of the tree-ssa
> > branch.  The snapshots are going to be available at
> > 
> > http://people.redhat.com/dnovillo/pub/snapshot/
> > 
> > The SPEC tester will upload a tarball named tree-ssa-YYYYMMDD.tar.bz
> > after a successful bootstrap.
> 
> Good idea !
> 
> BTW, without http://gcc.gnu.org/ml/gcc-patches/2003-05/msg00575.html the 
> tree-ssa branch won't build on Alpha.
> 
I merged this weekend, so hopefully this patch made it in.  New
snapshots should be available soon.

> Oh, and you need about a dozen no-warn entries added to 
> gcc/gcc/Makefile.in :-(
> 
Yes.  This is an issue we need to fix before merging into mainline. 
GIMPLE tends to emit conditionals that trigger warnings like 'used
before definition'.  I believe that Jason was working on a change that
would fix this.  In any case, for now they are OK to list as no-warn,
but we will need to remove those settings before merging.  Could you add
these entries near the existing FIXME note when you patch Makefile.in?


Thanks.  Diego.

Re: Daily snapshots of tree-ssa branch available

[Joe Buck]

On Mon, May 12, 2003 at 08:43:54AM -0400, Diego Novillo wrote:
> Yes.  This is an issue we need to fix before merging into mainline. 
> GIMPLE tends to emit conditionals that trigger warnings like 'used
> before definition'.  I believe that Jason was working on a change that
> would fix this.

Are these warnings valid, or is it the well-known problem that gcc can't
deal with

	pointer* p;
	bool flag = false;
	if (expensive()) {
		p = init_p();
		flag = true;
	}
	do_something_else();
	if (flag) {
		use(p);
	}
?

That is, gcc's uninitialized warning misses correlations between variables,
so this kind of thing gives a false warning at use(p).

Re: Daily snapshots of tree-ssa branch available

[Toon Moene]

Diego Novillo wrote:

> On Sat, 2003-05-10 at 04:06, Toon Moene wrote:
>>
>>BTW, without http://gcc.gnu.org/ml/gcc-patches/2003-05/msg00575.html the 
>>tree-ssa branch won't build on Alpha.
> 
> I merged this weekend, so hopefully this patch made it in.  New
> snapshots should be available soon.

No, it didn't make it - that's why I made a remark about it :-)

>>Oh, and you need about a dozen no-warn entries added to 
>>gcc/gcc/Makefile.in :-(
> 
> Yes.  This is an issue we need to fix before merging into mainline. 
> GIMPLE tends to emit conditionals that trigger warnings like 'used
> before definition'.  I believe that Jason was working on a change that
> would fix this.  In any case, for now they are OK to list as no-warn,
> but we will need to remove those settings before merging.  Could you add
> these entries near the existing FIXME note when you patch Makefile.in?

Will work on that when I find some time ...

Cheers,

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

Re: Daily snapshots of tree-ssa branch available

[Diego Novillo]

On Mon, 2003-05-12 at 18:17, Toon Moene wrote:

> No, it didn't make it - that's why I made a remark about it :-)
> 
I neglected to mention that I still have not checked in the merge.  So,
you couldn't have possibly seen it yet :)

With any luck, the merge should be committed tomorrow or Wed.  I still
have some testsuite failures to go through.


Diego.

Re: Daily snapshots of tree-ssa branch available

[Diego Novillo]

On Mon, 2003-05-12 at 13:01, Joe Buck wrote:
> On Mon, May 12, 2003 at 08:43:54AM -0400, Diego Novillo wrote:
> > Yes.  This is an issue we need to fix before merging into mainline. 
> > GIMPLE tends to emit conditionals that trigger warnings like 'used
> > before definition'.  I believe that Jason was working on a change that
> > would fix this.
> 
> Are these warnings valid.
>
Kind of.  The more common false positive is due to the way the
gimplifier restructures complex conditional expressions like 'if (v &&
(x = foo()))'.  This sometimes leaves code similar to what you quote in
your message.

To properly fix this we need to restructure the code emitted by the
gimplifier and, probably, move some of the other warnings into the
tree-ssa passes.


Diego.

Re: Daily snapshots of tree-ssa branch available

[Toon Moene]

[-- Attachment #1: Type: text/plain, Size: 962 bytes --]

Toon Moene wrote:

> Diego Novillo wrote:
> 
>> On Sat, 2003-05-10 at 04:06, Toon Moene wrote:

>>> Oh, and you need about a dozen no-warn entries added to 
>>> gcc/gcc/Makefile.in :-(
>>
>> Yes.  This is an issue we need to fix before merging into mainline. 
>> GIMPLE tends to emit conditionals that trigger warnings like 'used
>> before definition'.  I believe that Jason was working on a change that
>> would fix this.  In any case, for now they are OK to list as no-warn,
>> but we will need to remove those settings before merging.  Could you add
>> these entries near the existing FIXME note when you patch Makefile.in?
> 
> Will work on that when I find some time ...

Here ya go:

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

[-- Attachment #2: make.patch --]
[-- Type: text/plain, Size: 1513 bytes --]

2003-05-13  Toon Moene  <toon@moene.indiv.nluug.nl>

	* Makefile.in: Add additional -Wno-error targets for Alpha.

*** Makefile.in.orig	Tue May 13 17:54:16 2003
--- Makefile.in	Tue May 13 19:29:14 2003
*************** gengtype-lex.o-warn = -Wno-error
*** 172,177 ****
--- 172,182 ----
  # the gimplifier re-writes va_start with a temporary instead of using the
  # last named argument of the function.
+ alias.o-warn = -Wno-error
  bitmap.o-warn = -Wno-error
+ builtins.o-warn = -Wno-error
+ caller-save.o-warn = -Wno-error
  combine.o-warn = -Wno-error
+ cfgrtl.o-warn = -Wno-error
+ cse.o-warn = -Wno-error
  c-semantics.o-warn = -Wno-error
  c-typeck.o-warn = -Wno-error
*************** emit-rtl.o-warn = -Wno-error
*** 180,184 ****
--- 185,196 ----
  expr.o-warn = -Wno-error
  fold-const.o-warn = -Wno-error
+ ifcvt.o-warn = -Wno-error
  genattrtab.o-warn = -Wno-error
+ jump.o-warn = -Wno-error
+ local-alloc.o-warn = -Wno-error
+ ra-build.o-warn = -Wno-error
+ simplify-rtx.o-warn = -Wno-error
+ sched-vis.o-warn = -Wno-error
+ stmt.o-warn = -Wno-error
  reload1.o-warn = -Wno-error
  reload.o-warn = -Wno-error
*************** rtlanal.o-warn = -Wno-error
*** 186,190 ****
--- 198,206 ----
  tree.o-warn = -Wno-error
  varasm.o-warn = -Wno-error
+ f/com.o-warn = -Wno-error
+ f/equiv.o-warn = -Wno-error
  f/expr.o-warn = -Wno-error
+ f/intrin.o-warn = -Wno-error
+ f/where.o-warn = -Wno-error
  # The warning for 'prev_file_name' in profile.c is odd.  The variable *is*
  # initialized before being used.

Re: Daily snapshots of tree-ssa branch available

[Diego Novillo]

On Tue, 2003-05-13 at 16:29, Toon Moene wrote:

> Here ya go:
>
Thanks.  This is OK.


Diego.

Re: Daily snapshots of tree-ssa branch available

[Jason Merrill]

On Tue, 13 May 2003 22:29:16 +0200, Toon Moene <toon@moene.indiv.nluug.nl> wrote:

> 2003-05-13  Toon Moene  <toon@moene.indiv.nluug.nl>
>
> 	* Makefile.in: Add additional -Wno-error targets for Alpha.

Could you check to see which of these are still needed after my recent changes?

Thanks,
Jason

Re: Daily snapshots of tree-ssa branch available

[Toon Moene]

[-- Attachment #1: Type: text/plain, Size: 653 bytes --]

Jason Merrill wrote:

> On Tue, 13 May 2003 22:29:16 +0200, Toon Moene <toon@moene.indiv.nluug.nl> wrote:
> 
> 
>>2003-05-13  Toon Moene  <toon@moene.indiv.nluug.nl>
>>
>>	* Makefile.in: Add additional -Wno-error targets for Alpha.
> 
> 
> Could you check to see which of these are still needed after my recent changes?

With the attached patch I'm getting through the bootstrap.

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

[-- Attachment #2: make2.patch --]
[-- Type: text/plain, Size: 1399 bytes --]

*** Makefile.in.orig	Thu May 15 20:32:03 2003
--- Makefile.in	Wed May 21 23:03:31 2003
*************** gengtype-lex.o-warn = -Wno-error
*** 172,182 ****
  # the gimplifier re-writes va_start with a temporary instead of using the
  # last named argument of the function.
- alias.o-warn = -Wno-error
  bitmap.o-warn = -Wno-error
- builtins.o-warn = -Wno-error
  caller-save.o-warn = -Wno-error
  combine.o-warn = -Wno-error
- cfgrtl.o-warn = -Wno-error
- cse.o-warn = -Wno-error
  c-semantics.o-warn = -Wno-error
  c-typeck.o-warn = -Wno-error
--- 172,178 ----
*************** emit-rtl.o-warn = -Wno-error
*** 185,196 ****
  expr.o-warn = -Wno-error
  fold-const.o-warn = -Wno-error
- ifcvt.o-warn = -Wno-error
  genattrtab.o-warn = -Wno-error
- jump.o-warn = -Wno-error
- local-alloc.o-warn = -Wno-error
- ra-build.o-warn = -Wno-error
- simplify-rtx.o-warn = -Wno-error
- sched-vis.o-warn = -Wno-error
- stmt.o-warn = -Wno-error
  regmove.o-warn = -Wno-error
  reload1.o-warn = -Wno-error
--- 181,185 ----
*************** rtlanal.o-warn = -Wno-error
*** 199,207 ****
  tree.o-warn = -Wno-error
  varasm.o-warn = -Wno-error
- f/com.o-warn = -Wno-error
- f/equiv.o-warn = -Wno-error
  f/expr.o-warn = -Wno-error
- f/intrin.o-warn = -Wno-error
- f/where.o-warn = -Wno-error
  # The warning for 'prev_file_name' in profile.c is odd.  The variable *is*
  # initialized before being used.
--- 188,192 ----