Re: [PATCH v15 0/5] MR514: tests: Use the vulkan runner to run SM1 compilation tests.

[Giovanni Mascellani (＠giomasce)]

Giovanni Mascellani (@giomasce) commented about libs/vkd3d-shader/ir.c:
> - unsigned int components_read = 3 + (parser->shader_version.major >= 2);
> - unsigned int tmp_idx = ~0u;
> - unsigned int i, k;
> -
> - for (i = 0; i < instructions->count; ++i)
> - {
> -
>    texkill_ins = &instructions->elements[i];
> -
> -
>    if (texkill_ins->handler_idx != VKD3DSIH_TEXKILL)
> -
>        continue;
> -
> -
>    if (!shader_instruction_array_insert_at(instructions, i + 1, components_read + 1))
> -
>        return VKD3D_ERROR_OUT_OF_MEMORY;
> -
> -
>    if (tmp_idx == ~0u)
> -
>        tmp_idx = parser->shader_desc.temp_count++;

Right now you're also supposed to update the DCL_TEMPS instruction (and add it if it's not there). Since @cmccarthy is having the same problem in !564 too and I'm going to need it soon too, I think we could handle this problem in a uniform way:
* either we add a preliminary pass that adds `dcl_temps 0` if there is no DCL_TEMPS around (and then everybody increments it just like we do for `temp_count`);
* or we add a preliminary pass that removes any DCL_TEMPS around and then rely just on `temp_count`; this would require some minor changes to the validator.

I would vote for the second one.