All of lore.kernel.org
 help / color / mirror / Atom feed
* [amdkfd] where are AQL packets processed after writing the userspace doorbell
@ 2020-12-22  5:12 Smith John
  2020-12-22  6:17 ` Bridgman, John
  0 siblings, 1 reply; 2+ messages in thread
From: Smith John @ 2020-12-22  5:12 UTC (permalink / raw)
  To: amd-gfx


[-- Attachment #1.1: Type: text/plain, Size: 414 bytes --]

Hi! I noticed that the AQL packets are more concise compared with PM4
packets. It seemed that AQL packets need more post-processing than PM4
packets.
I was wondering where the AQL packets are processed, such like calculating
the code address using code_entry_offset, resetting packets' headers into
INVALID, and writing values to the completion signal when finished.
Are all these operations done by the firmware?

[-- Attachment #1.2: Type: text/html, Size: 472 bytes --]

[-- Attachment #2: Type: text/plain, Size: 154 bytes --]

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

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [amdkfd] where are AQL packets processed after writing the userspace doorbell
  2020-12-22  5:12 [amdkfd] where are AQL packets processed after writing the userspace doorbell Smith John
@ 2020-12-22  6:17 ` Bridgman, John
  0 siblings, 0 replies; 2+ messages in thread
From: Bridgman, John @ 2020-12-22  6:17 UTC (permalink / raw)
  To: amd-gfx


[-- Attachment #1.1: Type: text/plain, Size: 835 bytes --]

[AMD Official Use Only - Internal Distribution Only]

>Hi! I noticed that the AQL packets are more concise compared with PM4 packets. It seemed that AQL packets need more post-processing than PM4 packets.
>I was wondering where the AQL packets are processed, such like calculating the code address using code_entry_offset, resetting packets' headers into INVALID, and writing values to the completion signal when finished.
>Are all these operations done by the firmware?

Yes, these operations are performed entirely by MEC firmware.

There were special cases (eg certain debug scenarios) where we used a "soft AQL" layer in the ROC runtime which interpreted AQL packets and translated them into a series of PM4 packets, but I don't believe that mechanism is used any more. It was never used during normal processing anyways.

[-- Attachment #1.2: Type: text/html, Size: 1405 bytes --]

[-- Attachment #2: Type: text/plain, Size: 154 bytes --]

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

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2020-12-22  6:17 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-12-22  5:12 [amdkfd] where are AQL packets processed after writing the userspace doorbell Smith John
2020-12-22  6:17 ` Bridgman, John

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.