[Mesa-dev] [v2 3/6] glsl: ir_serializer class for the shader cache

[Tapani]

On 11/06/2013 10:55 PM, Paul Berry wrote:
> On 5 November 2013 23:45, Tapani Pälli <tapani.palli at intel.com 
> <mailto:tapani.palli at intel.com>> wrote:
>
>     On 11/05/2013 07:36 PM, Paul Berry wrote:
>>     On 1 November 2013 02:16, Tapani Pälli <tapani.palli at intel.com
>>     <mailto:tapani.palli at intel.com>> wrote:
>>
>>
>>         +
>>         +/**
>>         + * Function to create an unique string for a ir_variable.
>>         This is
>>         + * used by variable dereferences to indicate the exact
>>         ir_variable
>>         + * when deserialization happens.
>>
>>
>>     I still don't understand why we don't just use the pointer for
>>     this purpose.  It's unique, and it takes up much less storage
>>     than <name>_<decimal address>.
>
>     We need to construct a unique name to refer to when reading so
>     this is constructed here already. It could be also just a counter
>     like is used somewhere else (to have assignment_tmp at 1,
>     assignment_tmp at 2) but I wanted to keep this code minimal as it's
>     just unique naming.
>
>
> I'm sorry, there's still something I'm not understanding.  Before 
> serialization, the names are not unique.  So I don't see why we need 
> to create unique names when reading.  The goal should be to reproduce 
> the IR data structure that was saved as closely as possible.
>
> It seems like storing the address of the ir_variable (or some other 
> suitably determined unique numeric id) should be sufficient for this.  
> When we serialize the ir_variable, we write out its unique numeric 
> id.  When we serialize an ir_dereference_variable, we write out the 
> unique numeric id of the ir_variable it refers to.  When deserializing 
> an ir_variable, we store the mapping from its unique numeric id to its 
> new address in memory in a hash table.  When deserializing an 
> ir_dereference_variable, we look up the unique numeric id to find the 
> address of the ir_variable that's already been deserialized.
>
> It doesn't seem to me that uniqueness of names is necessary anywhere 
> in that process.  What am I missing?

I think we are somewhat on the same page but the method you describe 
here is another way to solve the same problem. Problem is to get 
references to variables pointing to the exact correct variable. This 
cannot be detected by variable name, which I wanted to use for it to be 
very clear which variable we are referring to and I did not see any harm 
in using such 'unique name', ir_print_visitor does similar renaming in 
its output. I can change to use your method if renaming is no good 
(haven't seen any issues with this though), I guess using the hash 
should not affect the parsing performance, correct variable is anyway 
searched from a list when creating a reference.

>
>>         +
>>         +   for (unsigned i = 0; i < ir->num_state_slots; i++) {
>>         +  blob.write_int32(&ir->state_slots[i].swizzle);
>>
>>
>>     Swizzles are unsigned 8-bit values.  This should be write_uint8.
>
>     OK, maybe the struct could also be changed to have only 8 bits
>     instead of a int then.
>
>
> I could get behind that change.
>
>
>
>>     +   CACHE_DEBUG("write %d prototypes\n", total);
>>
>>         +
>>         +   foreach_list_const(node, shader->ir) {
>>         +      ir_instruction *const inst = (ir_instruction *) node;
>>         +      if (inst->as_variable())
>>         +         if (save(inst))
>>         +            goto write_errors;
>>         +   }
>>         +
>>         +   foreach_list_const(node, shader->ir) {
>>         +      ir_instruction *const inst = (ir_instruction *) node;
>>         +      if (inst->as_function())
>>         +         if (save(inst))
>>         +            goto write_errors;
>>         +   }
>>
>>
>>     Why is it necessary to save the variables and instructions first?
>
>     This is because during parsing we might encounter a call to a
>     function that does not exist yet, same goes for variable
>     references. Another way would be to modify the reading side so
>     that it makes 2 passes over the data but I took this way as
>     originally reader did not use mmap so it was just simpler.
>
>
> Ok, I think you are correct about the functions.  But I believe for 
> variables, the ir_variable always appears in the IR before any 
> references to it.  Can someone confirm this? (Ken or Ian perhaps?)

This would be good to confirm. I'm pretty sure I did this to tackle a 
problem during parsing but I can't recall now  .. I will check this.

// Tapani
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20131107/46b30ecb/attachment-0001.html>