On 17/05/2018 12:24, Chris Wilson wrote:
Quoting Tvrtko Ursulin (2018-05-17 11:58:21)


On 17/05/2018 08:40, Chris Wilson wrote:
Store whether or not we need to kick the guc's execlists emulation on
the engine itself to avoid chasing the device info.

We do not chase device info but modparams in this case.

gen8_cs_irq_handler                          512     428     -84

I guess my point from before, (unfortunately I forgot to reply), was how
much of the saving remains if GEM_BUG_ON is compiled out?

Remember the motto of killing off checking globals and only checking
derived state? :) (I'm definitely not in favour of sprinkling more global
checking CAPS over the code.)

Out of the debug build

gen8_cs_irq_handler                          170     185     +15
gen8_cs_irq_handler                          170     128     -42 (later)
gen8_cs_irq_handler                          170     128     -42 (+ 
USES_GUC_SUBMISSION)

I don't get which is which.

[snip]

If nothing or almost nothing, I don't see a need to fiddle with this now.

Also please consider later patches which also need conditionals as
execlists is unfortunately gaining new tricks that are harder to pull
off with the current guc submission :|
(Might be time to stop mixing the two backends?)

I might say fine if the commit message contained truth. :)

Regards,

Tvrtko

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reply via email to