On Fri, May 26, 2017 at 10:47:18AM -0700, Kenneth Graunke wrote: > On Friday, May 26, 2017 4:25:20 AM PDT Chris Wilson wrote: > > Passing the index of the target buffer via the reloc.target_handle is > > marginally more efficient for the kernel (it can avoid some allocations, > > and can use a direct lookup rather than a hash or search). It is also > > useful for ourselves as we can use the index into our exec_bos for other > > tasks. > > > > Signed-off-by: Chris Wilson <ch...@chris-wilson.co.uk> > > Cc: Kenneth Graunke <kenn...@whitecape.org> > > Cc: Jason Ekstrand <jason.ekstr...@intel.com> > > --- > > src/mesa/drivers/dri/i965/intel_batchbuffer.c | 30 > > ++++++++++++++++++--------- > > 1 file changed, 20 insertions(+), 10 deletions(-) > > I wrote this patch as well, and it turned out to hurt performance by a > small amount in every CPU-limited benchmark I tried.
If you are end up looking at many relocs, yeah. I've used yet another array just to store the batch relocs that needed fixing up, but with BATCH_FIRST it's not required. > I think the extra loop to rewrite -1 to the batch index was enough to > make it hurt. That would go away with I915_EXEC_BATCH_FIRST... With a very small change, it looks like we can apply patch 3 first and then only apply HANDLE_LUT when we have BATCH_FIRST. I'll play a bit more, but I was surprised at how close the code was for NO_RELOC and wanted to share! -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
