[Mesa-dev] [PATCH 10/19] i965/fs: Lower LOAD_PAYLOAD and clean up.

[Matt Turner]

On Wed, Apr 23, 2014 at 10:55 PM, Kenneth Graunke <kenneth at whitecape.org> wrote:
> On 04/18/2014 11:56 AM, Matt Turner wrote:
>> Clean up with with register_coalesce()/dead_code_eliminate().
>> ---
>>  src/mesa/drivers/dri/i965/brw_fs.cpp | 37 ++++++++++++++++++++++++++++++++++++
>>  src/mesa/drivers/dri/i965/brw_fs.h   |  1 +
>>  2 files changed, 38 insertions(+)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
>> index e963ee8..602fc4a 100644
>> --- a/src/mesa/drivers/dri/i965/brw_fs.cpp
>> +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
>> @@ -2635,6 +2635,38 @@ fs_visitor::lower_uniform_pull_constant_loads()
>>     }
>>  }
>>
>> +bool
>> +fs_visitor::lower_load_payload()
>> +{
>> +   bool progress = false;
>> +
>> +   foreach_list_safe(node, &instructions) {
>> +      fs_inst *inst = (fs_inst *)node;
>> +
>> +      if (inst->opcode == SHADER_OPCODE_LOAD_PAYLOAD) {
>> +         fs_reg dst = inst->dst;
>> +
>
> It would be great to have a comment here such as:
>
> /* src[0] represents the (optional) message header. */
>
> It might also be worth adding a comment above the opcode definition in
> the previous patch explaining that src[0] is reserved for an optional
> message header, and could be BAD_FILE, while the rest of the parameters
> follow in src[1..n].  Notably, this is the first opcode where you can
> have a BAD_FILE early, and real parameters later (AFAIK).

Yeah. Another problem this causes is dump_instructions() stops
printing the arguments following BAD_FILE.

I could special-case that code, but I wonder if it'd be more clear to
use a new enum type ("MSG_HEADER")?