This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [Patch, AVR]: Fix PR46779
Denis Chertykov schrieb:
> 2011/6/9 Georg-Johann Lay <avr@gjlay.de>:
>> Denis Chertykov schrieb:
>>> 2011/6/9 Georg-Johann Lay <avr@gjlay.de>:
>> I agree with you. However, I think that the risk of spill failure
>> should be minimized. I have no idea how to fix a splill failure like
>> the following that occurs in testsuite (with -Os, no matter if the
>> patch is applied or not):
>>
>> ./gcc/testsuite/gcc.c-torture/execute/pr38051.c:189:1: error: unable
>> to find a register to spill in class 'POINTER_REGS'
>> ./gcc/testsuite/gcc.c-torture/execute/pr38051.c:189:1: error: this is
>> the insn:
>> (insn 61 60 62 10 (set (reg/v:SI 12 r12 [orig:73 b0 ] [73])
>> (mem:SI (subreg:HI (reg/v:SI 70 [ srcp2 ]) 0) [2 *D.2218_56+0
>> S4 A8])) ./gcc/testsuite/gcc.c-torture/execute/pr38051.c:64 12 {*movsi}
>> (nil))
>> ./gcc/testsuite/gcc.c-torture/execute/pr38051.c:189:1: internal
>> compiler error: in spill_failure, at reload1.c:2113
>>
>> Actually I have no idea if the snip in avr_hard_regno_mode_ok actually
>> would reduce the risk of spill failure :-/
>
> I think, no.
> I will try to debug reload for pr38051.c (It will be a long process 1-2 weeks)
>
> Denis.
Thanks for doing that! Debugging reload is really no fun...
I see spill fails for these test cases with -mmcu=atmega128
* gcc.c-torture/execute/pr38051.c (-Os)
* gcc.dg/pr43670.c (-O -ftree-vrp -fcompare-debug)
* gcc.dg/20030324-1.c (-O -fstrict-aliasing -fgcse)
all complainins contain RTX like
(mem:SI/SF (subreg:HI (reg:SI xxx) 0/2))
Then I have a question on spill failures:
There is PR46278, an optimization flaw that goes as follows:
The avr BE defines fake addressing mode X+const that has to be written
down in asm as
X += const
a = *X
X -= const
The comment says that this is just to cover a corner case, however the
new register allocator uses this case more often or even greedily.
There is no way to describe the cost for such an access, and as X has
lower prologue/epilogue cost than Y, X is preferred over Y often.
In 4.7, I see that flaw less often than in 4.5. However, I think the
best way is not to lie at the register allocator and not to supply a
fake instruction like that.
So I started to work on a fix. The changes in avr.h are:
* config/avr/avr.h (BASE_REG_CLASS): Remove.
(REG_OK_FOR_BASE_NOSTRICT_P): Remove.
(REG_OK_FOR_BASE_STRICT_P): Remove.
(MODE_CODE_BASE_REG_CLASS): New Define.
(REGNO_MODE_CODE_OK_FOR_BASE_P): New Define.
The macros REGNO_MODE_CODE_OK_FOR_BASE_P and MODE_CODE_BASE_REG_CLASS
allow a better, fine grained control over addressing modes for each
hard register and allow to support X without fake instructions. The
code quality actually improves, but I see a new spill failure for the
test case
* gcc.c-torture/compile/950612-1.c
On the one hand, that test case has heavy long long use and is no
"real world code" to run on AVR. On the other hand, I don't think
trading code quality here against ICE there is a good idea.
What do you think about that? As I have no idea how to fix a spill
failure in the backend, I stopped working on a patch.
Then I observed trouble with DI patterns during libgcc build and had
to remove
* "zero_extendqidi2"
* "zero_extendhidi2"
* "zero_extendsidi2"
These are "orphan" insns: they deal with DI without having movdi
support so I removed them.
Johann