[Mesa-dev] [PATCH 1/2] radv/nir: call opt_remove_phis after trivial continues.

Timothy Arceri tarceri at itsqueeze.com
Wed Sep 13 03:52:46 UTC 2017



On 13/09/17 13:48, Dave Airlie wrote:
> On 13 September 2017 at 13:42, Timothy Arceri <tarceri at itsqueeze.com> wrote:
>>
>> On 13/09/17 12:57, Dave Airlie wrote:
>>>
>>> From: Dave Airlie <airlied at redhat.com>
>>>
>>> With the shaders in the ssao demo, the nir_opt_if wasn't
>>> working properly without this, after this the if gets optimised
>>> so that loop unrolling gets called.
>>>
>>> (loop unrolling fails due to instruction count, but at least
>>> it gets to do that.)
>>>
>>> Signed-off-by: Dave Airlie <airlied at redhat.com>
>>> ---
>>>    src/amd/vulkan/radv_shader.c | 1 +
>>>    1 file changed, 1 insertion(+)
>>>
>>> diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c
>>> index 1e25ea3..87deb7c 100644
>>> --- a/src/amd/vulkan/radv_shader.c
>>> +++ b/src/amd/vulkan/radv_shader.c
>>> @@ -129,6 +129,7 @@ radv_optimize_nir(struct nir_shader *shader)
>>>                    if (nir_opt_trivial_continues(shader)) {
>>>                            progress = true;
>>>                            NIR_PASS(progress, shader, nir_copy_prop);
>>> +                       NIR_PASS(progress, shader, nir_opt_remove_phis);
>>
>>
>> Any reason for not just putting this in the main nir opt loop rather than
>> inside this if?
> 
> It's already in there.
> 
> This is adding it after the second copy_prop.
> 

Right I just noticed that. I seems i965 just calls it once but later on, 
I wonder if radv should just do the same.

> Dave.
>>
>>
>>>                            NIR_PASS(progress, shader, nir_opt_dce);
>>>                    }
>>>                    NIR_PASS(progress, shader, nir_opt_if);
>>>
>>


More information about the mesa-dev mailing list