<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Thu, Apr 5, 2018 at 1:31 PM, Caio Marcelo de Oliveira Filho <span dir="ltr"><<a href="mailto:caio.oliveira@intel.com" target="_blank">caio.oliveira@intel.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="">> @@ -268,8 +288,30 @@ split_var_copies_impl(nir_<wbr>function_impl *impl)<br>
>     state.dead_ctx = ralloc_context(NULL);<br>
>     state.progress = false;<br>
><br>
> +   nir_builder b;<br>
> +   nir_builder_init(&b, impl);<br>
> +<br>
>     nir_foreach_block(block, impl) {<br>
>        split_var_copies_block(block, &state);<br>
> +<br>
> +      nir_foreach_instr_safe(instr, block) {<br>
> +         if (instr->type != nir_instr_type_intrinsic)<br>
> +            continue;<br>
> +<br>
> +         nir_intrinsic_instr *copy = nir_instr_as_intrinsic(instr);<br>
> +         if (copy->intrinsic != nir_intrinsic_copy_deref)<br>
> +            continue;<br>
<br>
</span>Could you 'continue' here if glsl_type_is_vector_or_scalar(<wbr>) returns<br>
true for one of the srcs? If I understood correctly, it would avoid<br>
cluttering the 'progress'.<br></blockquote><div><br></div><div>Yes, we could.  That sounds like a good follow-on.  I'd like to keep this patch as close in behavior to the original as possible.<br></div></div></div></div>