[Nouveau] [PATCH 1/2] drm/nv50: align size of buffer object to the right boundaries.

[Maarten Maathuis]

On Sun, Dec 27, 2009 at 2:43 PM, Christoph Bumiller
<e0425955 at student.tuwien.ac.at> wrote:
> On 12/27/2009 12:41 PM, Maarten Maathuis wrote:
>> - Depth and stencil buffers are supposed to be large enough in general.
>>
>> Signed-off-by: Maarten Maathuis <madman2003 at gmail.com>
>> ---
>>  drivers/gpu/drm/nouveau/nouveau_bo.c |    9 ++++-----
>>  1 files changed, 4 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index e342a41..9fc4bd6 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -65,8 +65,9 @@ nouveau_bo_fixup_align(struct drm_device *dev,
>>
>>       /*
>>        * Some of the tile_flags have a periodic structure of N*4096 bytes,
>> -      * align to to that as well as the page size. Overallocate memory to
>> -      * avoid corruption of other buffer objects.
>> +      * align to to that as well as the page size. Align the size to the
>> +      * appropriate boundaries. This does imply that sizes are rounded up
>> +      * 3-7 pages, so make sure your "special" buffer sizes are large enough.
>>        */
> No - 16x16 depth textures or whatever crazy idea some app might have
> won't be large enough.
> Taking care of size in userspace and of alignment in kernel ... not nice
> to split in my opinion.

Userspace doesn't know all constraints, so the kernel has to do
something. How to deal with these limitations is another question.
Either we over allocate (and hide the overallocation from everyone),
align the size, or we try to make the allocator smarter to group them
together based on their tile flags.

>>       if (dev_priv->card_type == NV_50) {
>>               uint32_t block_size = nouveau_mem_fb_amount(dev) >> 15;
>> @@ -77,22 +78,20 @@ nouveau_bo_fixup_align(struct drm_device *dev,
>>               case 0x2800:
>>               case 0x4800:
>>               case 0x7a00:
>> -                     *size = roundup(*size, block_size);
>>                       if (is_power_of_2(block_size)) {
>> -                             *size += 3 * block_size;
>>                               for (i = 1; i < 10; i++) {
>>                                       *align = 12 * i * block_size;
>>                                       if (!(*align % 65536))
>>                                               break;
>>                               }
>>                       } else {
>> -                             *size += 6 * block_size;
>>                               for (i = 1; i < 10; i++) {
>>                                       *align = 8 * i * block_size;
>>                                       if (!(*align % 65536))
>>                                               break;
>>                               }
>>                       }
>> +                     *size = roundup(*size, *align);
>>                       break;
>>               default:
>>                       break;
>
>