Re: [avr-gcc-list] What GCC bug currently effect the AVR port of GCC?

[Björn Haase]

Hi Eric,

thank you for preparing an overview of open avr-bugs.
> Finally, *this is NOT intended to start a discussion about the internal
> details of every little bug on this list*. Please start different
> threads for the bug in question, or better still, keep the discussion
> limited to being on the bug reports themselves.
I found that the most critical bugs that you mentioned have been fixed already 
although they are not marked as fixed in bugzilla. Also the honor for fixing 
some of the bugs is not mine but Roger Sayle's (ice due to reload problems) 
and Marek's (shifting issues).
If I see it correctly, there presently is no real known "wrong-code" bug.
 
One should probably add a reminder for the 16-Bit peripheral registers to 
update the specification and warn users when using pointers to peripheral 
registers. Possibly in the libc manual?

Yours,

Björn
>
> AVR specific bugs as of 2005.02.28:
> --------------------------------------
> 16563:  Looks to be a fairly major bug for the AVR. This needs testing
> to see if it still exists in the latest release (3.4.3) and in CVS 4.0
> branch.
There are reports, that this bug is no longer existent in head 4.0.0 but one 
should possibly confirm this.
> 18251: Major AVR bug (ICE on valid code). Patches are available for
> 4.0.0 and recently, 3.4.x. This needs to be finished up and get the
> patches into CVS.
There exists a patch by Andy Hutchinson for this bug for head 4.0.0. I can 
confirm that this bug now is removed.
> 18551: Wrong code bug for C++. This is assigned to Denis Chertykov but
> there has been no work on it.
Possibly this is not any bug.?
> 18887: ICE on valid code bug. Björn Haase posted a patch, but it needs
> to be tested and gone through the process of getting it into CVS.
This bug now longer provoques an ICE. It is merely a reminder for a strange 
way libgcc2 is presently implemented. (there has been work under way two 
weeks ago that have fixed another bug.)
> 18989: Comment #1 of this bug narrows what is a bug and what isn't with
> this report. This needs to be tested with a recent HEAD snapshot. Björn
> Haase's fixes for bit shifts may have fixed this, perhaps.
Better say Marek's (already commited) shifts have fixed this bug.
> 19087: Overflowed address in dwarf debug line information. Listed as
> UNCONFIRMED. Reported by Torleif of Atmel and affects AVR Studio DWARF2.
> Potential fix in comments. Needs more testing.
> 19154: missed optimization. Comments imply this is due to rtx_cost,
> which IIRC was recently rewritten and comitted. So this needs testing
> with recent HEAD to see if it still exists.
> 19329: This looks like it's fixed in 4.0 and 3.4.x and all it needs is
> to be is formally closed. Marek, you did the commits, can you close this?
> 19331: Patch posted by Roger Sayle. I think this bug might be fixed, but
> needs to be verified and closed.
> 19636: ICE on valid code. Listed as UNCONFIRMED. Somebody with
> sufficient permissions needs to verify that this is a bug.
> 19676: Minor bug: Loop optimizer fails to reverse simple loop. Andy
> Hutchinson is supposed to be looking into this I think.
> 19684: Wrong size in asm comment. Listed as UNCONFIRMED. Needs double
> checking that this is a bug.
> 19686: missed optimization. Minor bug. Loop performance decrease, not
> comparing against 0.
> 19815: Documentation change. Listed as UNCOFIRMED. Needs double checking
> by someone with sufficient permissions to be confirmed.
> 19835: missed optimization: Loop variable gets widened to long instead
> of int.
> 19885: avr dwarf-2 support is broken for head 4.0/4.1. This is a
> regression and needs to get fixed before 4.0 is released or AVR Studio
> can't be used for debugging a GCC project.
> 19959: Can't compile gnattools for the cross targets. This is a
> regression for the Ada language for the AVR. This is an issue in getting
> AVR-Ada into WinAVR for GCC 4.0.
> 20016: Enhancement bug. Easy to fix, and needs testing.
> 20082: ICE: unrecognizable insn. Listed as UNCONFIRMED. Needs verification.
> 20243: UCONFIRMED. This is difficult to tell what the bug is, or whether
> it's a design issue.
>
> There you have it.
>
> HTH
> Eric Weddington