W3C home > Mailing lists > Public > public-gpu@w3.org > November 2017

Re: NXT design for memory barriers and buffer mapping.

From: Corentin Wallez <cwallez@google.com>
Date: Fri, 17 Nov 2017 14:58:20 -0800
Message-ID: <CAGdfWNOfS3ztHcJviwp5GpDx1HMQ_pNzWDgukQ-WpmvtBYutmg@mail.gmail.com>
To: Richard Schreyer <rschreyer@apple.com>
Cc: Markus Siglreithmaier <m.siglreith@gmail.com>, public-gpu <public-gpu@w3.org>
Buffered transitions update the state tracking synchronously when the
command buffer is submitted to a queue. This means that for the use-case
you described the user doesn't need to do anything to ensure ordering.

On Fri, Nov 17, 2017 at 2:18 PM, Richard Schreyer <rschreyer@apple.com>
wrote:

> For buffered transitions, do those conceptually happen synchronously upon
> submission to a queue, or does the user have to wait for some event to know
> that they have taken effect?
>
> ie, if a user encodes and submits a buffered transition, and then does an
> immediate transition, does anything need to happen between those two steps
> to ensure ordering?
>
> On Nov 17, 2017, at 11:20 AM, Corentin Wallez <cwallez@google.com> wrote:
>
> Hey Markus,
>
> This is correct, NXT defers building the native API command buffer until
> queue submit. There are pros and cons to it. Examples pros are being able
> to work around more bugs, and have more flexibility in the implementation,
> cons that it could increase latency (for VR).
>
> Cheers,
>
> Corentin
>
> On Fri, Nov 17, 2017 at 11:15 AM, Markus Siglreithmaier <
> m.siglreith@gmail.com> wrote:
>
>> Hi,
>>
>> I'm having a small question regarding the implementation of memory
>> barriers. In the proposal the previous resources states (D3D12) or access
>> masks (Vulkan)  are unknown at command buffer recording time and only known
>> on actual queue submission if I understand it correctly. Would this require
>> then to cache all commands internally and defer the actual command buffer
>> calls until queue submission?
>>
>> Cheers,
>> Markus
>>
>> Am 15.11.2017 um 05:51 schrieb Corentin Wallez:
>>
>> Hey all,
>>
>> We wrote some document to help everyone reason about NXT's proposals for
>> memory barriers and resource upload /download. Unfortunately we still don't
>> have a fleshed out proposal that minimizes the number of copies on UMA.
>> Instead the docs focus on explaining our current design for resource
>> upload/download and for memory barriers since they are very tied.
>> Eventually we'll have these docs in MarkDown in some repo, either WebGPU's
>> or NXT's.
>>
>>    - NXT "memory barriers"
>>    <https://docs.google.com/document/d/1k7lPmxP7M7MMQR4g210lNC5TPwmXCMLgKOQWNiuJxzA>
>>    <- Please read this first as buffer mapping depends on it.
>>    - NXT buffer mapping
>>    <https://docs.google.com/document/d/1HFzMMvDGHFtTgjNT0j-0SQ1fNU9R7woZ4JuNJdAXBjg>
>>
>> Cheers,
>>
>> Corentin
>>
>>
>>
>
>
Received on Friday, 17 November 2017 22:59:04 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 19:52:22 UTC