Re: [RFC][PATCH] Preferred rename register in regrename pass

[Christophe Lyon <christophe.lyon@linaro.org>]

On 9 November 2015 at 18:01, Robert Suchanek <Robert.Suchanek@imgtec.com> wrote:
> Hi,
>
>> On 11/09/2015 02:32 PM, Robert Suchanek wrote:
>> > The results below were generated for CSiBE benchmark and the numbers in
>> > columns express bytes in format 'net (gain/loss)' to show the difference
>> > with and without the patch when -frename-registers switch is used.
>>
>> I'm not entirely sure what the numbers represent. I can see how you'd
>> measure at a net size change (I assume a negative net is the intended
>> goal), but how did you arrive at gain/loss numbers?
>>
>> In any case, assuming negative is good, the results seem pretty decent.
>
> The gain/loss was calculated based on per function analysis.
> Each flavour e.g. MIPS n64 -Os was ran with/without the patch and compared to
> the base i.e. without the patch. The patched version of each function may
> show either positive (larger code size), negative or no difference to
> the code size. The gain/loss in a cell is the sum of all positive/negative
> numbers for a test. The negatives, as you said, are the good ones.
>
>>
>> > +         gcc_assert
>> > +           (terminated_this_insn->regno == REGNO (recog_data.operand[1]));
>>
>> Maybe break the line before the == so that you can start the arguments
>> on the same line as the assert.
>>
>> > +  /* Nonzero if the chain is renamed.  */
>> > +  unsigned int renamed:1;
>>
>> I'd write "has already been renamed" since that is maybe slightly less
>> ambiguous.
>>
>> Ok with those changes.
>>
>>
>> Bernd
>
> Will do the changes and apply.
>

Hi,

Since you committed this (r230087 if I'm correct), I can see that GCC
fails to build
ligfortran for target arm-none-linuxgnueabi --with-cpu=cortex-a9.

The backtrace is:
/tmp/8079076_3.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/libgfortran/generated/matmul_i8.c:
In function
 'matmul_i8':
/tmp/8079076_3.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/libgfortran/generated/matmul_i8.c:374:1:
internal compiler error: in scan_rtx_reg, at regrename.c:1074
 }
 ^
0xa13940 scan_rtx_reg
        /tmp/8079076_3.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/regrename.c:1074
0xa1451d record_out_operands
        /tmp/8079076_3.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/regrename.c:1554
0xa14d12 build_def_use
        /tmp/8079076_3.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/regrename.c:1802
0xa1533e regrename_analyze(bitmap_head*)
        /tmp/8079076_3.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/regrename.c:726
0xa161f9 regrename_optimize
        /tmp/8079076_3.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/regrename.c:1871
0xa161f9 execute
        /tmp/8079076_3.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/regrename.c:1908
Please submit a full bug report,

Can you have a look?


> Regards,
> Robert
>