[PATCH vkd3d 09/17] vkd3d-shader/hlsl: Replace register offsets with index paths in load initializations.

[Zebediah Figura]

On 7/15/22 18:31, Francisco Casas wrote:
>> But, that said, I don't think that either of these are prohibitive, or
>> make things worse than the alternative.
>>
> 
> So, just to confirm, I should replace compute_component_path() with a
> hlsl_deref_from_component_index() in v3. Right?

Without having seen it, I think I'd be in favor of it. In a sense it's 
the same function anyway, mostly, just with a few differences.

> 
>>> +struct hlsl_type *hlsl_type_get_component_type(struct hlsl_ctx *ctx,
>>> struct hlsl_type *type,
>>> +        unsigned int index)
>>> +{
>>> +    while (!type_is_single_component(type))
>>> +        subtype_index_from_component_index(ctx, &type, &index);
>>> +
>>> +    return type;
>>> +}
>>> +
>>> +/* Returns the path of a given component within a type, given its index.
>>> + * *path_len will be set to the lenght of the path.
>>> + * Memory should be free afterwards.
>>> + */
>>> +static unsigned int *compute_component_path(struct hlsl_ctx *ctx,
>>> struct hlsl_type *type,
>>> +        unsigned int index, unsigned int *path_len)
>>> +{
>>> +    struct hlsl_type *path_type;
>>> +    unsigned int *path, path_index;
>>> +
>>> +    *path_len = 0;
>>> +    path_type = type;
>>> +    path_index = index;
>>> +    while (!type_is_single_component(path_type))
>>> +    {
>>> +        subtype_index_from_component_index(ctx, &path_type,
>>> &path_index);
>>> +        ++*path_len;
>>> +    }
>>> +    if (!(path = hlsl_alloc(ctx, *path_len * sizeof(unsigned int) + 1)))
>>> +        return NULL;
>>> +
>>> +    *path_len = 0;
>>> +    path_type = type;
>>> +    path_index = index;
>>> +    while (!type_is_single_component(path_type))
>>> +    {
>>> +        path[*path_len] = subtype_index_from_component_index(ctx,
>>> &path_type, &path_index);
>>> +        ++*path_len;
>>> +    }
>>> +
>>> +    return path;
>>> +}
>>
>> Could we return path_type from this function, since we've already
>> calculated it, and throw out hlsl_type_get_component_type() entirely?
>>
> 
> We would be giving the caller the responsibility of freeing the returned
> path even if it is not used (which I think was a concern in v1). It also
> happens often that the path is required but not the component type.
> 
> So, given that the implementation of both functions is small and doesn't
> introduce too much overhead, I think that keeping them separate is
> better, but, as always, if you insist I will change it.

Oh, right, I forgot that add_cast() needed 
hlsl_type_get_component_type() but not this function. Never mind, then.