Quoting Matthew Auld (2019-02-26 18:53:17) > On Thu, 14 Feb 2019 at 15:48, Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> wrote: > > > > Quoting Matthew Auld (2019-02-14 14:57:16) > > > We need to support doing relocations from the CPU when dealing with LMEM > > > objects. > > > > Why not just use the GPU reloc? Please do explain the relative merits. > > Are you suggesting just default to the GPU reloc path for LMEM, and > don't bother with CPU relocs? I guess that would work, I'm just > wondering if keeping the CPU reloc path around for debug might be > useful, it's at least been useful in the past, although maybe it's not > that relevant for upstream. Debug is a good argument; I just want to hear about the relative merits of each path. At the moment, we only use the GPU path if already busy, but maybe we need to reconsider? There's even more fun to be had where reservation is async and we need to defer the relocation processing until after we've acquired the memory and vma :) -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
