[Diffusion] rL196996: Move mips' datalayout computation out of line and add comments.

Rafael EspĂ­ndola rafael.espindola at gmail.com
Wed Dec 11 08:59:37 PST 2013


On 11 December 2013 06:42, Daniel Sanders <daniel.sanders at imgtec.com> wrote:
> Hi Rafael,
>
> Was the removal of the "-f128:128:128" part of the isABI_N64() datalayouts intentional? If so, what was the reason?

No, sorry about that. The intension was for this to produce an identical string.

I was curious how could this not have causes a test failure, then I
noticed that DataLayout::init looks like:

...
 // Default alignments
...
setAlignment(FLOAT_ALIGN,    16, 16, 128);
...
parseSpecifier(Desc);

So f:128:128:128 is one of the defaults.

I think it is actually a bit better to use the simpler string.
Otherwise we get the impression that ABI_N64 is different when it is
actually just being explicit while the other abi is implicit, but this
is your call. Do you want me to add this back?

I will send a patch for review cleaning up the defaults on other
targets. Thank you so much for noticing!

Cheers,
Rafael



More information about the llvm-commits mailing list