Re: Re: [f-cpu] GCC 3.1 for F-CPU port

[whygee@club-internet.fr]

hi !

>De: devik
>> >using sufix like .8/.16/.32 is okay but .64 is wrong
>>
>> why ?...
>> at least, using these numbers is not ambiguous at all.
>
>Probably because .8 is valid ONLY if SR_SIZE_0 == 1 ...

and what ?

>But if you will use .8, .128... then assembler should
>generate .note in ELF to describe wanted content of
>SR_SIZE_[0-3] and linker should check whether only
>4 values all used in all linked files.
>Maybe it is not bad idea at all !

heheh.

There is also a little convention : SR_SIZE_3 contains
the largest size. This way, algorithms that need the largest
size (for maximal performance) do not need to fiddle with
SRs. But this can be changed or enhanced.

>By the way, I read in some older post that SIMD bit
>in insns is used only to mask result.
well, that's how it's implemented in FC0.

> So that if some core would decide to ignore the flag
> will it still work correctly ?
more or less.

>IMHO it should.
yup but Cédric said the contrary (at least from the
compiler's point of view).

> So that if I do non-simd add.8
(that can be also named "scalar" :-D)

> I should not rely on bits 8 and higher, these can have
> any value, correct ?

now, the mask clears the MSB so you can rely on this to be zero.

If you have no idea whether SIMD flag is set, then you can't
rely on MSB. but at least the LSB are sure :-)

>devik
YG

*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/