On 07/18/2013 12:31 AM, Kenneth Graunke wrote:
On 07/17/2013 04:46 PM, Chad Versace wrote:
Set MOCS to L3_CACHEABLE for the following packets, during the normal
draw upload path and during blorp:
3DSTATE_CONSTANT_PS
3DSTATE_CONSTANT_VS
3DSTATE_DEPTH_BUFFER
3DSTATE_HIER_DEPTH_BUFFER
3DSTATE_STENCIL_BUFFER
3DSTATE_VERTEX_BUFFERS
CMD_STATE_BASE_ADDRESS
SURFACE_STATE
It's not possible to set the MOCS of the MCS buffer, because the MCS
inherits MOCS from the parent surface.
Gives +5.68% on Xonotic 1920x1080 on Haswell GT2.
At this point, I'm quite happy to accept your choices of what to cache.
Two comments:
I'd really prefer to see each change split into a separate patch. You don't
need to benchmark them separately or
anything, but it'd be nice. I know Eric had hoped to see this happen as well.
I considered that, but hesitated. I agree, such a patch series will be nicer.
V3 coming soon.
I'd also prefer to see:
uint32_t mocs = brw->is_haswell ? GEN7_MOCS_L3 : 0;
rather than if-blocks...it's just more concise. Not a big deal either way,
though.
Sure. The if-blocks were an artifact of a more complex experiment I was doing.
But,
now that the MOCS settings are so simple, there's no longer a need for the
if-block.
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
