Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> writes:
> Quoting Mika Kuoppala (2017-10-19 15:39:41)
>> while (test_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted)) {
>> + struct execlist_port *port;
>> +
>> /* The HWSP contains a (cacheable) mirror of the CSB */
>> const u32 *buf =
>> &engine->status_page.page_addr[I915_HWS_CSB_BUF0_INDEX];
>> @@ -855,7 +865,7 @@ static void intel_lrc_irq_handler(unsigned long data)
>>
>> if (status & GEN8_CTX_STATUS_ACTIVE_IDLE &&
>> buf[2*head + 1] == PREEMPT_ID) {
>> - execlist_cancel_port_requests(execlists);
>> + execlists_cancel_port_requests(execlists);
>>
>> spin_lock_irq(&engine->timeline->lock);
>> unwind_incomplete_requests(engine);
>> @@ -870,6 +880,8 @@ static void intel_lrc_irq_handler(unsigned long data)
>> execlists->preempt)
>> continue;
>>
>> + port = execlists_port_head(execlists);
>> +
>> /* Check the context/desc id for this event matches */
>> GEM_DEBUG_BUG_ON(buf[2 * head + 1] != port->context_id);
>>
>> @@ -890,7 +902,7 @@ static void intel_lrc_irq_handler(unsigned long data)
>> }
>>
>> /* After the final element, the hw should be idle */
>> - GEM_BUG_ON(port_count(port) == 0 &&
>> + GEM_BUG_ON(port_count(execlists_port_head(execlists)) == 0 &&
>> !(status & GEN8_CTX_STATUS_ACTIVE_IDLE));
>> }
>
> Can you try reworking this such that port is kept local without having
> to go back to the struct on every loop? And then compare code
> generation.
New one is with the last_port introduce and set outside loop.
add/remove: 0/0 grow/shrink: 1/1 up/down: 10/-7 (3)
function old new delta
intel_lrc_irq_handler 1847 1857 +10
i915_guc_irq_handler 1327 1320 -7
Without looking at the assembler, I would go with last_port
as a net win.
-Mika
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
