[PATCH vkd3d 06/17] vkd3d-shader/hlsl: Lower scalar-to-vector casts to swizzles.

[Zebediah Figura]

On 12/2/21 09:26, Giovanni Mascellani wrote:
> Signed-off-by: Giovanni Mascellani <gmascellani at codeweavers.com>
> ---
> Though with a few comments...
> 
> On 01/12/21 17:14, Matteo Bruni wrote:
>> +    if (instr->type != HLSL_IR_EXPR)
>> +        return false;
>> +    cast = hlsl_ir_expr(instr);
>> +    src_type = cast->operands[0].node->data_type;
>> +    dst_type = cast->node.data_type;
> 
> Here operands[0].node is assumed to be valid before we know the 
> operation is checked to be OP1_CAST. I know that we don't have nullary 
> operators and I can't see why we might eventually have one, but I still 
> don't like the idea.

Yes, I suppose I would agree.

> 
>> +    if (cast->op == HLSL_OP1_CAST
>> +            && src_type->type <= HLSL_CLASS_VECTOR && dst_type->type 
>> <= HLSL_CLASS_VECTOR
>> +            && src_type->dimx == 1)
> 
> I don't like how this is formatted. I would either put each conjoint on 
> a dedicated line, or put two in one line and two in another. This 1-2-1 
> structure makes it more difficult to parse to the eye, at least to my eye.

The mental logic is that it compares "op", then "type" (for both 
sources), then "dimx". I don't feel strongly about it, though.