From: Jerome G. <gl...@fr...> - 2009-08-20 23:01:59
|
On Thu, 2009-08-20 at 19:13 +0300, Pauli Nieminen wrote: > I don't understand how this particular commit is causing performance > regression. > > There is now optimisation for dma buffer objects in kms case where > reusing old buffer object would cause wait if gpu haven't yet handled > the rendering operation associated buffer. > > This change requires kernel changes in KMS case because there was a > bug in the latest drm pull for linus. > I attached a diff which fixes the problem in kernel untill Dave sends > next pull request for Linus. Nevertheless we should not unmap buffer if we are going to remap it, map/unmap is way too costly with KMS. Pauli any good reason to unmap ? Cheers, Jerome |