* Problems in pr26180.c with own backend
@ 2012-11-14 14:03 Andreas Geißler
2012-11-14 23:15 ` Georg-Johann Lay
0 siblings, 1 reply; 3+ messages in thread
From: Andreas Geißler @ 2012-11-14 14:03 UTC (permalink / raw)
To: gcc-help
Hi. I need your help again. I wrote my own backend for the gcc-4.6.2
and with your help it works nearly perfectly.
But i have a problem with the test case pr26180.c. The compiler does
not load the register, which
is used for the comparison. And i don't get why.
C-Code:
...
x2 = (x > 0 ? x : (-x));
...
Wrong translated assembler code:
...
;;x > 0 ?
cmpi.l r8,-2147483648 <- r8 is not used in the entire function
and so it is not initialized with x.
jmp C,_L2
;;x2 = x
xori.l r8,-1
addi.l r8,1
_L2:
;;x2 = -x
...
If i only change the comparison integer from 0 to 1 the translation
works correct:
C-Code:
...
x2 = (x > 1 ? x : (-x));
...
Correct translated assembler code:
...
;;load r8
mov.l r8,fp
addi.l r8,4
mov.l r9,r8
ldr.l r8,r9
;;x > 1 ?
cmpi.l r8,1
sbs Z
jmpn N,_L2
;;x2 = -x
mov.l r8,fp
addi.l r8,4
mov.l r9,r8
ldr.l r8,r9
xori.l r8,-1
addi.l r8,1
jmp T,_L3
_L2:
;;x2 = x
mov.l r8,fp
addi.l r8,4
mov.l r9,r8
ldr.l r8,r9
_L3:
...
The translation works also correctly if i use "if else" instead of "?:".
So i you would like to ask you, which pattern is related to "x2 = (x >
0 ? x : (-x));"
or where i should search for this problem.
Thanks in advance.
Best regards,
Andreas
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Problems in pr26180.c with own backend
2012-11-14 14:03 Problems in pr26180.c with own backend Andreas Geißler
@ 2012-11-14 23:15 ` Georg-Johann Lay
2012-11-19 12:47 ` Andreas Geißler
0 siblings, 1 reply; 3+ messages in thread
From: Georg-Johann Lay @ 2012-11-14 23:15 UTC (permalink / raw)
To: Andreas Geißler; +Cc: gcc-help
Andreas GeiÃler schrieb:
> Hi. I need your help again. I wrote my own backend for the gcc-4.6.2
> and with your help it works nearly perfectly.
> But i have a problem with the test case pr26180.c. The compiler does
> not load the register, which
> is used for the comparison. And i don't get why.
>
> C-Code:
> ...
> x2 = (x > 0 ? x : (-x));
> ...
>
> Wrong translated assembler code:
> ...
Assembler don't tell anything; look at TL dump with -da, -fdump-rtl-all
or -fdump-rtl-all-details.
You might also want so set -dp and / or -dP and / or -fverbose-asm to
see insns along with assembly.
> The translation works also correctly if i use "if else" instead of "?:".
> So i you would like to ask you, which pattern is related to "x2 = (x >
> 0 ? x : (-x));"
Who knows if not you? That depends on your backend: Maybe absXX2, maybe
movXXcc, maybe jumping around and movXX and negXX2, ...
> or where i should search for this problem.
Look at .expand dump and track until where it gets wrong. Presumably a
wrong expander, wrong constraints (missing early clobber), wrong
peephole, whatever.
Johann
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Problems in pr26180.c with own backend
2012-11-14 23:15 ` Georg-Johann Lay
@ 2012-11-19 12:47 ` Andreas Geißler
0 siblings, 0 replies; 3+ messages in thread
From: Andreas Geißler @ 2012-11-19 12:47 UTC (permalink / raw)
To: Georg-Johann Lay; +Cc: gcc-help
Thanks to your help, i could find the problem. And you're absolutely right,
> Who knows if not you? That depends on your backend: Maybe absXX2, maybe
> movXXcc, maybe jumping around and movXX and negXX2, ...
It was absXX2.
> Look at .expand dump and track until where it gets wrong. Presumably a
> wrong expander, wrong constraints (missing early clobber), wrong peephole,
> whatever.
wrong expander.
Many thanks again.
Andreas
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2012-11-19 12:47 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-11-14 14:03 Problems in pr26180.c with own backend Andreas Geißler
2012-11-14 23:15 ` Georg-Johann Lay
2012-11-19 12:47 ` Andreas Geißler
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).