Re: [sdcc-devel] bug 1839277 and size regression

[Borut R. <bor...@si...>]

Maarten,

I ran the pic16 regression tests with the un-patched and patched 
versions, and the results are the same. I also compiled sdcc-mullong.c 
for pic16 and the result was still the same for both versions, and for 
sdcc 2.7.0 too. It seems that pic16 was not affected by the regression 
at all.

Borut


Maarten Brock wrote:
> Borut,
>
> The changes in gen.c are for the bug fix, the changes in SDCCicode.c are
> for the code regression.
>
> Before I commit, can someone please have a look how this affects the PIC
> ports?
>
> Some explanation: What happens here in SDCCicode is that it uses an
> assignment instead of a cast for 'ptr = array' or 'dataptr = codeptr'
> icodes. For the mcs51 Harvard architecture this is dangerous. How this is
> for PIC I really do not know.
>
> Maarten
>
>
>   
>> Maarten,
>>
>> once more I leave the decision to you, but I'm favorable to apply it. We
>> still have one week of time for testing before the release.
>>
>>
>> I saw that you made changes in mcs51 and ds390 code generator, but
>> Philipp discovered the regression on z80 target. Does the fix solve the
>> z80 regression too?
>>
>>
>> Borut
>>
>>
>> Maarten Brock wrote:
>>     
>>> Borut,
>>>
>>> I have a fix for bug 1839277 and for the size regression. See the
>>> attached
>>> patch.
>>>
>>> After this patch there are still some size regressions in the form of
>>> extra spils. In some places they are compensated for by less pushes/pops
>>> of registers, in other places they are not. My guess is they result from
>>> a
>>> bugfix in the register allocator. I would not want to change this
>>> anymore
>>> before the release.
>>>
>>> But what to do with my patch? Apply it or not?
>>> Comments are welcome and if I need to explain things, just ask.
>>>
>>> Greets,
>>> Maarten
>>>