On Sun, Jun 30, 2002 at 08:53:40PM -0400, Leif Delgass wrote: [...] > >For vertex data passing between the Mesa client driver and the drm, we >really don't even need to use AGP/DMA-able mem. We could probably get >away with a much smaller number of buffers for this since they can be >recycled as soon as the drm is finished copying the data to DMA buffers. >Actually, the buffer should be reusable as soon as the vertex dispatch >ioctl returns. In fact, couldn't each context just use keep reusing a >single vertex buffer for it's entire lifetime? That would eliminate the >need for the drmDMA ioctl on every vertex buffer flush.
These are details that we will surely face once we've determined that copying the client buffers is effective as we hope so. At that time I think the solution will be have the client submit is own buffers and copy from the client memory space. This way we wouldn't need to prealocate memory for clients on the DRM, but just the DMA'ble memory. I'm not sure whether at this moment we make one ord two IOCTLs per buffer, but note that we'd still need to call an IOCTL for the buffer submission. Jos� Fonseca ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
