[Bug target/7198] [IA64] missing usual fnma patterns

[Bug target/7198] [IA64] missing usual fnma patterns

[dhazeghi@yahoo.com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198


dhazeghi@yahoo.com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
   Last reconfirmed|0000-00-00 00:00:00         |2003-06-02 03:46:40
               date|                            |


------- Additional Comments From dhazeghi@yahoo.com  2003-06-02 03:46 -------
Should have been marked NEW per Tim's comments.



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug target/7198] [IA64] missing usual fnma patterns

[dhazeghi at yahoo dot com]

------- Additional Comments From dhazeghi at yahoo dot com  2003-12-09 23:16 -------
Still present. I'll ping the list...

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198

[Bug target/7198] [IA64] missing usual fnma patterns

[wilson at gcc dot gnu dot org]

------- Additional Comments From wilson at gcc dot gnu dot org  2003-12-16 08:32 -------
The only copyright assignment I see for Tim Prince is for g77, and this is not a
g77 patch.  So technically I need an updated copyright assignment from him
before I can accept his patch.

The patch is pretty trivial though. I was able to write my own patch from the
description of the problem before looking at Tim's patch.  So another way to
solve this problem is to write my own patch.

This does need a bit of work to match current sources.  XFmode is used instead
of TFmode for the 80-bit FP format now.  We should have a minus equivalent for
every plus/neg pattern, of which there are 9.  We already have 1 of them, so we
need 8 new ones.  It is probably better to replace the existing plus/neg pattern
s rather than adding new ones, as I don't think the current ones can be
generated except from the divide expanders.  That requires verification.  So
that means deleting the SFmode plus/neg/mult pattern, and converting the other 8
patterns to use minus/mult.  By the time I am done with that, I don't think I'll
need a copyright assignment as I will have a completely new patch.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198

[Bug target/7198] [IA64] missing usual fnma patterns

[tprince at computer dot org]

------- Additional Comments From tprince at computer dot org  2003-12-16 13:36 -------
Subject: Re:  [IA64] missing usual fnma patterns

At 12:33 AM 12/16/2003, wilson at gcc dot gnu dot org wrote:


>------- Additional Comments From wilson at gcc dot gnu dot org  2003-12-16 
>08:32 -------
>The only copyright assignment I see for Tim Prince is for g77, and this is 
>not a
>g77 patch.  So technically I need an updated copyright assignment from him
>before I can accept his patch.
>
>The patch is pretty trivial though. I was able to write my own patch from the
>description of the problem before looking at Tim's patch.  So another way to
>solve this problem is to write my own patch.
>
>This does need a bit of work to match current sources.  XFmode is used instead
>of TFmode for the 80-bit FP format now.  We should have a minus equivalent for
>every plus/neg pattern, of which there are 9.  We already have 1 of them, 
>so we
>need 8 new ones.  It is probably better to replace the existing plus/neg 
>pattern
>s rather than adding new ones, as I don't think the current ones can be
>generated except from the divide expanders.  That requires verification.  So
>that means deleting the SFmode plus/neg/mult pattern, and converting the 
>other 8
>patterns to use minus/mult.  By the time I am done with that, I don't 
>think I'll
>need a copyright assignment as I will have a completely new patch.
>
>--
>
>
>http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198
>
>------- You are receiving this mail because: -------
>You reported the bug, or are watching the reporter.

Thanks.  I noticed the change from TFmode to XFmode when I built 
gfortran-ssa recently, but I hadn't noticed the additional patterns.  I 
didn't know about a distinction in the copyright assignment; I assumed that 
it was for all gcc events.


Tim Prince 



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198

[Bug target/7198] [IA64] missing usual fnma patterns

[wilson at specifixinc dot com]

------- Additional Comments From wilson at specifixinc dot com  2004-01-26 23:44 -------
Subject: Re: Candidate patch for PR 7198

On Mon, 2004-01-26 at 10:23, Zack Weinberg wrote:
> Here's a candidate patch for PR 7198.  I cannot bootstrap mainline due
> to what looks like an unrelated bug in cselib.  However, the trivial
> test case

There is a known problem with sched-ebb's use of cselib that results in
accesses to memory that has already been freed.  This causes bootstrap
compare problems that go away if your environment size changes. 
However, I'd call this a sched-ebb bug rather than a cselib bug, because
sched-ebb is using cselib in a way that it was never designed for.  I've
never seen this on my machine, and there have been so many other IA-64
problems to deal with that it has never made it to my high priority list
of problems.  This doesn't result in incorrect code, it only results in
code that is scheduled differently.

Thanks for fixing PR 7198.

I don't really see a need for the nmaddsf4_alts pattern.  This can only
be created via expanders, and we currently don't have any expanders that
need it.  It doesn't hurt to have it though, and it does make the
patterns more symmetrical, so I don't mind if you add it.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198

[Bug target/7198] [IA64] missing usual fnma patterns

[zack at codesourcery dot com]

------- Additional Comments From zack at codesourcery dot com  2004-01-27 18:04 -------
Subject: Re: Candidate patch for PR 7198

Jim Wilson <wilson@specifixinc.com> writes:

> I don't really see a need for the nmaddsf4_alts pattern.  This can only
> be created via expanders, and we currently don't have any expanders that
> need it.  It doesn't hurt to have it though, and it does make the
> patterns more symmetrical, so I don't mind if you add it.

I may be adding a mechanism for specifying that all floating-point
arithmetic operations are to use an alternative FPSR, in which case
we need _alts patterns for all basic arithmetic patterns; that's why I
added this.

zw



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198

[Bug target/7198] [IA64] missing usual fnma patterns

[wilson at specifixinc dot com]

------- Additional Comments From wilson at specifixinc dot com  2004-01-28 00:07 -------
Subject: Re:  [IA64] missing usual fnma patterns

zack at codesourcery dot com wrote:
> I may be adding a mechanism for specifying that all floating-point
> arithmetic operations are to use an alternative FPSR, in which case
> we need _alts patterns for all basic arithmetic patterns; that's why I
> added this.

OK.

If you do this, you might want to look at PR 7285 which is a vaguely 
related issue.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198

[Bug target/7198] [IA64] missing usual fnma patterns

[zack at gcc dot gnu dot org]

------- Additional Comments From zack at gcc dot gnu dot org  2004-01-28 21:40 -------
Fixed in 3.5.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198

[Bug target/7198] [IA64] missing usual fnma patterns

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7198