[Bug target/97521] [11 Regression] wrong code with -mno-sse2 since r11-3394

[ams at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97521

--- Comment #22 from Andrew Stubbs <ams at gcc dot gnu.org> ---
(In reply to Andrew Stubbs from comment #21)
> (In reply to Richard Biener from comment #19)
> > GCN also uses MODE_INT for the mask mode and thus may be similarly affected.
> > Andrew - are the bits in the mask dense?  Thus for a V4SImode compare
> > would the mask occupy only the lowest 4 bits of the DImode mask?
> 
> Yes, that's correct.

Or rather, I should say that that *will* be the case when I add partial vector
support; right now it can only be done via masking V64SImode.

A have a patch set, but the last problem is that while_ult doesn't operate on
partial integer masks, leading to wrong code. AArch64 doesn't have a problem
with this because it uses VBI masks of the right size. I have a patch that adds
the vector size as an operand to while_ult; this seems to fix the problems on
GCN, but I need to make corresponding changes for AArch64 also before I can
submit those patches, and time is tight.