Re: [patch i386]: PR/38900 ICE for fastcall functions with -O2

[Kai Tietz <ktietz70 at googlemail dot com>]

2009/7/6 Richard Henderson <rth@redhat.com>:
> On 07/06/2009 10:11 AM, Kai Tietz wrote:
>>
>> Hello,
>>
>> the first patch I posted just cured the ICE but produces wrong code.
>> So I modified patch and add an additional testcase for dllimport'ed
>> fastcall method with more then one argument to my patch. The problem
>> is that ecx and edx are used by fastcall functions for register
>> passing, This patch makes sure that for the case that ecx and edx are
>> used for argument passing to sibcalled function, eax is used as call
>> register.
>
> This doesn't fix the problem either, it merely moves it around.
> You'll now see the same problem for regparm functions instead.
>
> You need to figure out why ecx and edx weren't already live,
> which should have forced the register allocator to chose eax
> without these changed annotations.
>
>
> r~
>

So, I tried to solve this by adding to FUNCTION_ARG_REGNO_P an
additional function type argument to make better choices about call
used registers (which is in fact an issue for df-scan.c and rtlanal.c,
but this had no effect on this spilling issue.
So for the first, I would like to avoid this ICE and afterward to
investigate in more detail why register allocator fails to see that
ecx and edx are in use.

ChangeLog for gcc

2009-07-06  Kai Tietz  <kai.tietz@onevision.com>

      * config/i386/i386.c (ix86_function_ok_for_sibcall): Disable
.     sibcalling for fastcall methods, which are dllimported.


ChangeLog for gcc/testsuite

2009-07-07  Kai Tietz  <kai.tietz@onevision.com>

      *gcc.dg/pr38900.c: New testcase.
      *gcc.dg/pr38900-1.c: New testcase.


Tested bootstrap for i686-pc-mingw32. Ok to apply to 4.5, and to 4.3
and 4.4 branches?

Cheers,
Kai

-- 
|  (\_/) This is Bunny. Copy and paste
| (='.'=) Bunny into your signature to help
| (")_(") him gain world domination

Attachment: disable_fastcall_dllimport_sibcall.diff
Description: Binary data