On Sun, Apr 11, 2010 at 8:12 PM, Jerome Glisse glisse@freedesktop.orgwrote:
On Sun, Apr 11, 2010 at 06:39:05AM +0200, Marek Olšák wrote:
Hi devs,
The first attached patch fixes the calculation of mipmapped 3D texture
sizes
in the CS checker, the 3rd dimension (depth) should be minified too. This should probably go to 2.6.34.
The second patch adds 2 new regs:
- VAP_ALT_NUM_VERTICES, along with an update to the CS checker.
- VAP_INDEX_OFFSET, I don't think this one needs to be tracked, because
we
have min/max vertex index clamping, but I might be wrong.
I need to recheck but iirc VAP_INDEX_OFFSET doesn't count in the min max clipping ie we need to track it.
GL places index clamping prior to adding the index offset and if hw follows this rule, there is nothing sane to be done. VAP_INDEX_OFFSET is completely orthogonal to VAP_VF_MAX_VTX_INDX. The latter contains a minimum allowed buffer-object size divided by a stride, having nothing to do with actual indices. To support index offsets, we would have to check content of every index buffer, which I am not a fan of. Opinions?
-Marek
On Wed, Apr 14, 2010 at 03:42:57AM +0200, Marek Olšák wrote:
On Sun, Apr 11, 2010 at 8:12 PM, Jerome Glisse glisse@freedesktop.orgwrote:
On Sun, Apr 11, 2010 at 06:39:05AM +0200, Marek Olšák wrote:
Hi devs,
The first attached patch fixes the calculation of mipmapped 3D texture
sizes
in the CS checker, the 3rd dimension (depth) should be minified too. This should probably go to 2.6.34.
The second patch adds 2 new regs:
- VAP_ALT_NUM_VERTICES, along with an update to the CS checker.
- VAP_INDEX_OFFSET, I don't think this one needs to be tracked, because
we
have min/max vertex index clamping, but I might be wrong.
I need to recheck but iirc VAP_INDEX_OFFSET doesn't count in the min max clipping ie we need to track it.
GL places index clamping prior to adding the index offset and if hw follows this rule, there is nothing sane to be done. VAP_INDEX_OFFSET is completely orthogonal to VAP_VF_MAX_VTX_INDX. The latter contains a minimum allowed buffer-object size divided by a stride, having nothing to do with actual indices. To support index offsets, we would have to check content of every index buffer, which I am not a fan of. Opinions?
-Marek
Yeah we should problably drop vertex/index tracking stuff there it would be very hard to use it to read specific memory and vertex shader are too limited to be abuse to do anythings that could lead to forwarding data to app.
Cheers, Jerome
dri-devel@lists.freedesktop.org
-
Jerome Glisse -
Marek Olšák