[Mesa-dev] [PATCH 8/8] intel/hsw: Enable hiz

[Eric Anholt]

Chad Versace <chad.versace at linux.intel.com> writes:

> On 04/09/2013 04:15 PM, Eric Anholt wrote:
>> Kenneth Graunke <kenneth at whitecape.org> writes:
>>
>>> From: Chad Versace <chad.versace at linux.intel.com>
>>>
>>> Enable hiz by setting intel_context::has_hiz.  However, to work around
>>> a hardware bug, we selectively enable hiz for only nicely aligned miptree
>>> slices.
>>>
>>> No Piglit regressions on Haswell 0x0d26 rev07 when based atop
>>> mesa-master-97e40a5.
>>>
>>> Improves the performance of GLB27_TRex_C24Z16_FixedTimeStep by 18.52%
>>> (hsw-0x0d26-rev07; kernel-3.9.0-rc1; GLBenchmark 2.7.0 Release a68901;
>>> samples=3).
>>
>>> diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
>>> index 6c27bab..654f0be 100644
>>> --- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
>>> +++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
>>> @@ -29,6 +29,7 @@
>>>   #include <GL/internal/dri_interface.h>
>>>
>>>   #include "intel_batchbuffer.h"
>>> +#include "intel_chipset.h"
>>>   #include "intel_context.h"
>>>   #include "intel_mipmap_tree.h"
>>>   #include "intel_regions.h"
>>> @@ -1036,7 +1037,38 @@ intel_miptree_slice_has_hiz(struct intel_mipmap_tree *mt,
>>>                               uint32_t layer)
>>>   {
>>>      intel_miptree_check_level_layer(mt, level, layer);
>>> -   return mt->hiz_mt != NULL;
>>> +
>>> +   if (!mt->hiz_mt)
>>> +      return false;
>>> +
>>> +   int devid = drm_intel_bufmgr_gem_get_devid(mt->region->bo->bufmgr);
>>> +   if (IS_HASWELL(devid)) {
>>
>> This conditional is checking for about 37 different PCI IDs, and
>> shouldn't appear in a hot path.  Please use intel->is_haswell, by
>> passing the intel_context into this function.
>
> Passing the intel_context into this function would require changing
> the signature of many functions. Rather than change all those 
> signatures, I'd prefer to store a boolean flag 'has_hiz' in
> intel_mipmap_slice. What do you think?

That would get even more code out of the hot paths if it's precomputed
at miptree creation, so I'd be in favor.  Slice structs pack nicely into
cachelines at the moment, but oh well.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20130410/9a33fbdd/attachment.pgp>