On Sun, 21 Apr 2002, José Fonseca wrote: > On 2002.04.21 19:40 Leif Delgass wrote: > > On Sun, 21 Apr 2002, José Fonseca wrote: > > > > > > We just need to add the fifo check now. > > > > > > I've just add it but it makes gears drop from 222 to 185 fps in my > > system. > > > I don't know if this is caused by droped triangles when there is no > > FIFO > > > check, or if the FIFO check makes a big overload. So I #ifdef'd it out. > > > > > Note that I even avoided to use udelay or any other time function, but > > it > > > didn't help. > > > > I've always wondered why we never ran into lockups before. The previous > > branch had places where potentially more than 16 writes would happen > > without checking the fifo. Since the fifo depth is acutally deeper than > > 16 and the check just looks at the last 16 slots, maybe we've been lucky. > > > > In that case I can try to replace the check for a idle engine to a check > of 16 empty slots. Perhaps that may give a green light sooner.
Yes, I think waiting for idle here is overkill. According to the programmer's guide, you only need to wait for idle when reading a register or bit field updated by the draw engine, or when the draw engine and an apterture read/write are acting on the same memory region (e.g. Read/WritePixels). > > With DMA, we won't have this issue, but I think while were using > > pseudo-DMA as a debugging aid and if we leave it in as a fallback, we > > should keep the fifo check there just to rule out engine lockups caused > > by a fifo overflow. I don't expect this to be as fast as doing the > > register writes client side, so I'm not too worried about the performance > > hit. > > > > Ok. > > > -- > > Leif Delgass > > http://www.retinalburn.net > > > > José Fonseca > -- Leif Delgass http://www.retinalburn.net _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
