[PATCH] D83715: AMDGPU/GlobalISel: Use clamp modifier for uaddsat/usubsat

Tue Jul 14 05:18:26 PDT 2020

foad added a comment.

In D83715#2150014 <https://reviews.llvm.org/D83715#2150014>, @arsenm wrote:

> In D83715#2149633 <https://reviews.llvm.org/D83715#2149633>, @foad wrote:
>
> > > There's an annoying asymmetry where we have signed saturating for
> > >  packed operations, but not for scalar 16-bit.
> >
> > My Vega Instruction Set Architecture Reference Guide says:
> >  "V_ADD_I16
> >  D.i16 = S0.i16 + S1.i16.
> >  Supports saturation (signed 16-bit integer domain)."
>
>
> Seems missing from the gfx8 and 10 manuals? I'm guessing it's really there though

The same text is in the GFX10 "RDNA 1.0 Instruction Set Architecture Reference Guide" too, but with the instructions renamed to v_add_nc_i16 and v_add_nc_u16.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D83715/new/

https://reviews.llvm.org/D83715