Mika Kuoppala <mika.kuoppala@xxxxxxxxxxxxxxx> writes:
> Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> writes:
>
>> Quoting Mika Kuoppala (2017-11-02 10:38:30)
>>> Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> writes:
>>>
>>> > Quoting Mika Kuoppala (2017-10-31 15:27:33)
>>> >> +static inline struct execlist_port *
>>> >> +execlists_port_next(struct intel_engine_execlists * const execlists,
>>> >> + const struct execlist_port * const port)
>>> >> +{
>>> >> + const unsigned int n = __port_add(port_index(port, execlists),
>>> >> + 1,
>>> >> + execlists->port_mask);
>>> >
>>> > How does this compare to
>>> >
>>> > if (port++ == execlists->port + execlists->port_mask)
>>> > port = execlists->port;
>>> >
>>> > return port;
>>> > ?
>>>
>>> add/remove: 0/0 grow/shrink: 1/1 up/down: 29/-29 (0)
>>> function old new delta
>>> i915_guc_irq_handler 2584 2613 +29
>>> intel_lrc_irq_handler 2963 2934 -29
>>> Total: Before=1123627, After=1123627, chg +0.00%
>>
>> Hmm, your functions saw little difference and are twice as big as
>> mine... Weird.
>>
>> I used gcc version 6.3.0 20170516 (Debian 6.3.0-18) with defconfig.
>> Yourself?
>
> I had debugs on, sigh...
>
> Now with the defconfig and gcc (Ubuntu 6.3.0-12ubuntu2) 6.3.0 20170406:
>
> add/remove: 0/0 grow/shrink: 0/2 up/down: 0/-139 (-139)
> function old new delta
> i915_guc_irq_handler 1620 1617 -3
> intel_lrc_irq_handler 1926 1790 -136
>
> So we have a clear winner.
>
And to be precise on what diff lead to above:
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 387667fe50d3..9131d66fb628 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -599,12 +599,12 @@ execlists_port_tail(struct intel_engine_execlists * const execlists)
static inline struct execlist_port *
execlists_port_next(struct intel_engine_execlists * const execlists,
- const struct execlist_port * const port)
+ struct execlist_port *port)
{
- const unsigned int n = __port_add(port_index(port, execlists),
- 1,
- execlists->port_mask);
- return &execlists->port[n];
+ if (port++ == execlists->port + execlists->port_mask)
+ port = execlists->port;
+
+ return port;
}
> -Mika
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
