Re: [PATCH] rs6000: Fix incorrect RTL for Power LE when removing the UNSPECS [PR106069]

[Segher Boessenkool <segher@kernel.crashing.org>]

Hi!

On Tue, Aug 09, 2022 at 11:01:05AM +0800, Kewen.Lin wrote:
> on 2022/8/8 11:42, Xionghu Luo wrote:
> > Regression tested pass for Power8{LE,BE}{32,64} and Power{9,10}LE{32,64}
> 
> Sorry, no -m32 for LE testing.

You can use -m32 on powerpc64le-*, but the default configuration
disallows it.  There also is powerpcle-*, which in the distant past
actually was used (string insns (like lswi) and multiple insns (like
lmw) do not work, and unaligned accesses are more problematic as well,
but :-) )

It isn't something we support with ELFv2 at all, indeed.

> I have some concern on those changed "altivec_*_direct", IMHO the suffix
> "_direct" is normally to indicate the define_insn is mapped to the
> corresponding hw insn directly.

Exactly.  Let's please keep this intact.

> With this change, for example,
> altivec_vmrghb_direct can be mapped into vmrghb or vmrglb, this looks
> misleading.  Maybe we can add the corresponding _direct_le and _direct_be
> versions, both are mapped into the same insn but have different RTL
> patterns.

If that is the best we can do, that is the best we can do.  It would be
lovely if there was something nicer we can do though :-)


Segher