[Mesa-dev] [PATCH 1/8] nir: add varying array splitting pass

[Nicolai Hähnle]

On 29.11.2017 12:20, Timothy Arceri wrote:
> On 29/11/17 21:47, Nicolai Hähnle wrote:
>> On 21.11.2017 04:28, Timothy Arceri wrote:
>>> +      nir_ssa_dest_init(&element_intr->instr, &element_intr->dest,
>>> +                        intr->num_components, 
>>> intr->dest.ssa.bit_size, NULL);
>>> +
>>> +      if (intr->intrinsic == nir_intrinsic_interp_var_at_offset ||
>>> +          intr->intrinsic == nir_intrinsic_interp_var_at_sample)
>>> +         nir_src_copy(element_intr->src, intr->src, 
>>> &element_intr->instr);
>>
>> &element_int->src[0], &intr->src[0] for clarity, I think.

?


>>> +
>>> +      nir_ssa_def_rewrite_uses(&intr->dest.ssa,
>>> + nir_src_for_ssa(&element_intr->dest.ssa));
>>> +   } else {
>>> +      nir_intrinsic_set_write_mask(element_intr, 
>>> nir_intrinsic_write_mask(intr));
>>> +      element_intr->src[0] = intr->src[0];
>>
>> I'm still not 100% up on my NIR, but doesn't this also need to be a 
>> nir_src_copy? I think it makes a difference if there's an indirect 
>> access involved.
> 
> We don't split indirects so it shouldn't matter here.

Maybe I'm reading this wrong, but AFAIU this would be for indirects in 
the source of a nir_intrinsic_store_var. So, storing to a non-indirect 
output variable, but taking a value from an indirect access to e.g. a 
temporary array. Those could still exist, right?

(And even if they can't it might be good to use nir_src_copy anyway for 
defensive programming.)

Cheers,
Nicolai
-- 
Lerne, wie die Welt wirklich ist,
Aber vergiss niemals, wie sie sein sollte.