Re: [PATCH 19/70] target/hexagon/idef-parser: Use gen_constant for gen_extend_tcg_width_op

[Richard Henderson]

On 2/27/23 11:55, Taylor Simpson wrote:
> > -        HexValue mask = gen_tmp_value(c, locp, mask_str,
> > -                                     dst_width, UNSIGNED);
> > +        HexValue mask = gen_constant(c, locp, "-1", dst_width,
> > + UNSIGNED);
> >           OUT(c, locp, "tcg_gen_shr_i", &dst_width, "(",
> > -            &mask, ", ", &mask, ", ", &shift, ");\n");
> > +            &res, ", ", &mask, ", ", &shift, ");\n");
> >           OUT(c, locp, "tcg_gen_and_i", &dst_width, "(",
> > -            &res, ", ", value, ", ", &mask, ");\n");
> > +            &res, ", ", &res, ", ", value, ");\n");
>
> What's the advantage of putting the result of the tcg_gen_shr into res instead 
> of mask?  Is there something in TCG code generation that takes advantage?

With this patch, mask is read-only, so a write to it is illegal.


r~