[Mesa-dev] [PATCH v4 106/129] nir: Rework opt_copy_prop_vars to use deref instructions

[Jason Ekstrand]

On June 22, 2018 11:52:54 Kenneth Graunke <kenneth at whitecape.org> wrote:

> On Friday, June 22, 2018 8:28:47 AM PDT Jason Ekstrand wrote:
>> On Thu, Jun 21, 2018 at 1:03 AM, Kenneth Graunke <kenneth at whitecape.org>
>> wrote:
>>
>>> On Thursday, May 31, 2018 10:06:47 PM PDT Jason Ekstrand wrote:
>>> [snip]
>>>> @@ -529,57 +509,55 @@ static bool
>>>> load_from_deref_entry_value(struct copy_prop_var_state *state,
>>>>                         struct copy_entry *entry,
>>>>                         nir_builder *b, nir_intrinsic_instr *intrin,
>>>> -                            nir_deref_var *src, struct value *value)
>>>> +                            nir_deref_instr *src, struct value *value)
>>>> {
>>>> *value = entry->src;
>>>>
>>>> -   /* Walk the deref to get the two tails and also figure out if we
>>> need to
>>>> -    * specialize any wildcards.
>>>> -    */
>>>> -   bool need_to_specialize_wildcards = false;
>>>> -   nir_deref *entry_tail = &entry->dst->deref;
>>>> -   nir_deref *src_tail = &src->deref;
>>>> -   while (entry_tail->child && src_tail->child) {
>>>> -      assert(src_tail->child->deref_type ==
>>> entry_tail->child->deref_type);
>>>> -      if (src_tail->child->deref_type == nir_deref_type_array) {
>>>> -         nir_deref_array *entry_arr = nir_deref_as_array(entry_tail-
>>>> child);
>>>> -         nir_deref_array *src_arr = nir_deref_as_array(src_tail->
>>> child);
>>>> -
>>>
>>> I think there might be a bug here...note this condition...
>>>
>>>> -         if (src_arr->deref_array_type != nir_deref_array_type_wildcard
>>> &&
>>>> -             entry_arr->deref_array_type ==
>>> nir_deref_array_type_wildcard)
>>>
>>> Old: Source NOT wildcard, dest is wildcard.
>>>
>>>> -            need_to_specialize_wildcards = true;
>>>> -      }
>>>> +   b->cursor = nir_instr_remove(&intrin->instr);
>>>>
>>>> -      entry_tail = entry_tail->child;
>>>> -      src_tail = src_tail->child;
>>>> +   nir_deref_path entry_dst_path, src_path;
>>>> +   nir_deref_path_init(&entry_dst_path, entry->dst, state->mem_ctx);
>>>> +   nir_deref_path_init(&src_path, src, state->mem_ctx);
>>>> +
>>>> +   bool need_to_specialize_wildcards = false;
>>>> +   nir_deref_instr **entry_p = &entry_dst_path.path[1];
>>>> +   nir_deref_instr **src_p = &src_path.path[1];
>>>> +   while (*entry_p && *src_p) {
>>>> +      nir_deref_instr *entry_tail = *entry_p++;
>>>> +      nir_deref_instr *src_tail = *src_p++;
>>>> +
>>>> +      if (src_tail->deref_type == nir_deref_type_array &&
>>>> +          entry_tail->deref_type == nir_deref_type_array_wildcard)
>>>
>>> New: Source IS wildcard, dest is wildcard.  I think you want != on the
>>> source condition to match the old behavior.
>>
>> I don't think there's a bug here.  With the old mechanism, we had
>> deref_type_array and then within that deref_array_type_direct, _indirect,
>> and _wildcard.  In the new scheme, we have deref_type_array and
>> deref_type_array_wildcard.  So deref_type == nir_deref_type_array in the
>> new translates to deref_type == nir_deref_type_array && deref_array_type !=
>> nir_deref_array_type_wildcard.  Does that make sense?
>
> I missed that it also changed from != wildcard to == array. :(

No worries. Thanks for double checking!

--Jason