[Mesa-dev] [PATCH 6/7] nir: fix missing increments of num_inputs/num_outputs

[Rob Clark]

On Sat, Nov 7, 2015 at 10:59 AM, Jason Ekstrand <jason at jlekstrand.net> wrote:
> On Sat, Nov 7, 2015 at 5:21 AM, Rob Clark <robdclark at gmail.com> wrote:
>> On Fri, Nov 6, 2015 at 10:20 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>>> On Fri, Nov 6, 2015 at 5:10 PM, Rob Clark <robdclark at gmail.com> wrote:
>>>> On Fri, Nov 6, 2015 at 6:39 PM, Rob Clark <robdclark at gmail.com> wrote:
>>>>> On Fri, Nov 6, 2015 at 6:23 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>>>>>> On Fri, Nov 6, 2015 at 8:35 AM, Rob Clark <robdclark at gmail.com> wrote:
>>>>>>> From: Rob Clark <robclark at freedesktop.org>
>>>>>>>
>>>>>>> Signed-off-by: Rob Clark <robclark at freedesktop.org>
>>>>>>> ---
>>>>>>>  src/glsl/nir/nir_lower_clip.c            | 2 ++
>>>>>>>  src/glsl/nir/nir_lower_two_sided_color.c | 2 ++
>>>>>>>  2 files changed, 4 insertions(+)
>>>>>>>
>>>>>>> diff --git a/src/glsl/nir/nir_lower_clip.c b/src/glsl/nir/nir_lower_clip.c
>>>>>>> index 31ccfb2..4a91527 100644
>>>>>>> --- a/src/glsl/nir/nir_lower_clip.c
>>>>>>> +++ b/src/glsl/nir/nir_lower_clip.c
>>>>>>> @@ -55,9 +55,11 @@ create_clipdist_var(nir_shader *shader, unsigned drvloc,
>>>>>>>
>>>>>>>     if (output) {
>>>>>>>        exec_list_push_tail(&shader->outputs, &var->node);
>>>>>>> +      shader->num_outputs++;
>>>>>>>     }
>>>>>>>     else {
>>>>>>>        exec_list_push_tail(&shader->inputs, &var->node);
>>>>>>> +      shader->num_inputs++;
>>>>>>
>>>>>> I'm not sure what I think about this.  Usually, num_inputs/outputs is
>>>>>> set by nir_lower_io or similar.  They don't have to be vec4's.  In the
>>>>>> i965 driver, FS inputs are in terms of floats.  Maybe tgsi_to_nir
>>>>>> provides you those guarantees but we haven't for i965.
>>>>>
>>>>> hmm, what do you recommend then?  There isn't really any
>>>>> straightforward way to run this *prior* to lower_io... tgsi->nir gives
>>>>> me something that is already i/o lowered, and what I'm doing so far w/
>>>>> gallium support for glsl->nir is doing lower_io (and a few other
>>>>> steps) in gallium so that the result the driver gets is equivalent to
>>>>> tgsi->nir..
>>>
>>> Hrm... That does make things a bit sticky.  If you'd like, you can go
>>> ahead and push it with
>>>
>>> Akced-by: Jason Ekstrand <jason.ekstrand at intel.com>
>>>
>>>> btw Jason, how do you feel about stuffing the type_size fxn ptr in
>>>> nir_shader_compiler_options?  It shouldn't ever change over the
>>>> lifetime of the shader, we could drop it as arg to
>>>> nir_assign_var_locations() (err, well, replace w/ nir_shader ptr), and
>>>> we could use it to dtrt here..
>>>
>>> That won't work.  The type_size function we use depends on shader
>>> stage, gen, and type of thing we're lowering.  Stuffing it into
>>> nir_shader_compiler_options won't work.
>>
>> hmm, but the type_size should be invariant over the lifetime of the
>> nir_shader object, shouldn't it?  (Well, maybe we need to be able to
>> specify different one's for uniform/input/output?)  I guess if you
>> don't already have different compiler_options's for scalar vs simd
>> modes, then they could be passed in to nir_shader_create() instead..
>
> We could restructure things to make nir_compiler_options something we
> create more on-the-fly but, as it currently stands, we create 1 or 2
> of them total for the life of the driver.
>
>> The other option is just to pass type_size into these other lowering
>> passes and use that instead of +=1..  but seems like type_size should
>> be constant for a given nir_shader, so not having to pass them around
>> all over the place seems nice.
>
> Like I said above, go ahead and push it.  We can sort out the issues
> once we have multiple drivers using it.  Trying to sort it out now is
> kind of theoretical at best.

ok.. I'll have one more pass, potentially inserting a uniform (this
time, ypos-transform), as part of my evolving gallium glsl_to_nir
patchset.. but for now I'll just add some notes in appropriate spots
in the code that the num_{uniforms,inputs,outputs} increment should
use type_size()..

BR,
-R