This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Problems with the way we calculate costs
- To: law at cygnus dot com
- Subject: Re: Problems with the way we calculate costs
- From: Aldy Hernandez <aldyh at redhat dot com>
- Date: 09 Sep 2000 20:07:03 -0400
- Cc: Geoff Keating <geoffk at cygnus dot com>, Bernd Schmidt <bernds at redhat dot co dot uk>, gcc-patches at gcc dot gnu dot org
- References: <11452.968428535@upchuck>
>>>>> "Jeffrey" == Jeffrey A Law <law@cygnus.com> writes:
> In message <jm1yyv9137.fsf@envy.cygnus.com>you write:
> [ Big Snip ]
>> (insn 1033 1032 1034 (set (mem/s:QI (plus:SI (reg:SI 396)
>> (const_int 2 [0x2])) 0)
>> (reg:QI 399)) -1 (nil)
>> (nil))
>>
>> ; so CSE substitutes (reg:SI 3) in here, and the insn is unrecognizable
>> ; because the modes don't match.
>>
>>
>> However, I don't like the way that CSE stretched the lifespan of r3
>> over so many insns in the first place, and would much rather this
>> didn't happen. Explicit hard register uses like this are bad and
>> are likely to cause problems later in the compiler.
> Not unlike the problem Aldy & I ran into on an ARM variant we're working
> on. Bernd posted a patch this morning, but I haven't had a chance to
> look at it.
Indeed, quite similar.
I have tried Bernd's patch and my Arm builds are back to normal.
Aldy