* RE: Why reload make bad code?
@ 2002-12-24 9:12 Naveen Sharma, Noida
0 siblings, 0 replies; 2+ messages in thread
From: Naveen Sharma, Noida @ 2002-12-24 9:12 UTC (permalink / raw)
To: Wang Xiaowu; +Cc: gcc
> BUT after reload/global_alloc, It generate bad code.
>
> ...
> (insn:HI 16 15 387 0 0x4001d370 (set (reg:SI 28 %r12)
> (mem/s:SI (plus:SI (mem/f:SI (plus:SI (reg/f:SI 7 %sp)
> (const_int 76 [0x54])) [2 S+0 S4 A32])
> (const_int 16 [0x10])) [4 <variable>._wds+0
> S4 A32])) 2 {*zkrisc.md:141} (insn_list 4 (nil))
> (nil))
> ...
>
> below is some information:
> %ap is arg pointer register, it will be eliminated by %fp or %sp.
> In this function: (%ap + 4) == (%sp + 76)
> FIRST_PSEUDO_REGISTER is 32.
> all outgoing args store in stack
> (plus (REG) (CONST_INT)) is legitimate addres,
> (set (REG) (MEM (plus (REG) (CONST_INT)))) is legitimated also.
Have you defined LEGITIMIZE_RELOAD_ADDRESS macro? I think
reload uses that to check the legitimacy of an address.
Also check the documentation of this macro at
http://gcc.gnu.org/onlinedocs/gccint/Addressing-Modes.html
Best Regards,
Naveen Sharma.
> gcc version is 3.3(2002/12/16)
>
> I tried to find the reason, but I failed. I need your help,
> please tell me what's the matter?
> (Maybe the information is not enough, tell me if it is)
^ permalink raw reply [flat|nested] 2+ messages in thread
* Why reload make bad code?
@ 2002-12-23 9:43 Wang Xiaowu
0 siblings, 0 replies; 2+ messages in thread
From: Wang Xiaowu @ 2002-12-23 9:43 UTC (permalink / raw)
To: gcc
Hello,
I am writing gcc backend for a RISC arch, and I have some troubles.
The question will happend when compile with optimization and function have many local variables.
below is insns before reload/global_alloc call
...
(insn:HI 4 3 15 0 (nil) (set (reg/v/f:SI 39)
(mem/f:SI (plus:SI (reg/f:SI 2 %ap)
(const_int 4 [0x4])) [2 S+0 S4 A32])) 2 {*zkrisc.md:141} (nil)
(expr_list:REG_EQUIV (mem/f:SI (plus:SI (reg/f:SI 2 %ap)
(const_int 4 [0x4])) [2 S+0 S4 A32])
(nil)))
...
(insn:HI 16 15 17 0 0x4001d370 (set (reg/v:SI 40)
(mem/s:SI (plus:SI (reg/v/f:SI 39)
(const_int 16 [0x10])) [4 <variable>._wds+0 S4 A32])) 2 {*zkrisc.md:141} (insn_list 4 (nil))
(nil))
...
BUT after reload/global_alloc, It generate bad code.
...
(insn:HI 16 15 387 0 0x4001d370 (set (reg:SI 28 %r12)
(mem/s:SI (plus:SI (mem/f:SI (plus:SI (reg/f:SI 7 %sp)
(const_int 76 [0x54])) [2 S+0 S4 A32])
(const_int 16 [0x10])) [4 <variable>._wds+0 S4 A32])) 2 {*zkrisc.md:141} (insn_list 4 (nil))
(nil))
...
below is some information:
%ap is arg pointer register, it will be eliminated by %fp or %sp.
In this function: (%ap + 4) == (%sp + 76)
FIRST_PSEUDO_REGISTER is 32.
all outgoing args store in stack
(plus (REG) (CONST_INT)) is legitimate addres,
(set (REG) (MEM (plus (REG) (CONST_INT)))) is legitimated also.
gcc version is 3.3(2002/12/16)
I tried to find the reason, but I failed. I need your help, please tell me what's the matter?
(Maybe the information is not enough, tell me if it is)
--
Your's Sincerely,
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2002-12-24 7:34 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2002-12-24 9:12 Why reload make bad code? Naveen Sharma, Noida
-- strict thread matches above, loose matches on Subject: below --
2002-12-23 9:43 Wang Xiaowu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).