W3C home > Mailing lists > Public > public-gpu@w3.org > November 2018

Re: draft execution environment for WebGPU, including WebGPU env spec for SPIR-V

From: Corentin Wallez <cwallez@google.com>
Date: Thu, 22 Nov 2018 20:47:14 +0100
Message-ID: <CAGdfWNO9w=_wRwXoDhV1rOjAbHkLbr_vMgh6sS28Z5em7p-WtQ@mail.gmail.com>
To: fpizlo@apple.com
Cc: Dan Sinclair <dsinclair@google.com>, Maciej Stachowiak <mjs@apple.com>, David Neto <dneto@google.com>, public-gpu <public-gpu@w3.org>
All, we updated the execution environment
<https://github.com/gpuweb/spirv-execution-env/blob/master/execution-env.md>
with more details on OOB behavior (in this pull request
<https://github.com/gpuweb/spirv-execution-env/pull/3>). Hopefully it
addresses your concerns.

Cheers,

Corentin

On Wed, Nov 14, 2018 at 10:12 PM Corentin Wallez <cwallez@google.com> wrote:

> After only 10 emails it's already hard to follow what are the separate
> concerns and what has already been stated.. Can you please open one issue
> per individual concern on https://github.com/gpuweb/spirv-execution-env?
> It would help keep each discussion focused on a single concern and get to a
> resolution/fix faster.
>
> On Wed, Nov 14, 2018 at 8:36 PM Filip Pizlo <fpizlo@apple.com> wrote:
>
>> It’s not great that to figure these things out, one must read multiple
>> specs instead of one.
>>
>> -Filip
>>
>> On Nov 14, 2018, at 11:29 AM, Dan Sinclair <dsinclair@google.com> wrote:
>>
>> On Wed, Nov 14, 2018 at 1:52 PM Maciej Stachowiak <mjs@apple.com> wrote:
>>
>>> Consider OpUndef. The whole purpose of this instruction is to invoke an
>>> undefined behavior for the sake of performance (since initializing may  be
>>> costly when it’s not necessary). Does any shader program containing OpUndef
>>> get rejected? Does it trap and halt? Does it return an arbitrary value that
>>> is uninitialized and therefore may lead to reading stale data ? Does it
>>> initialize to a specific known value (e.g. zero)?
>>>
>>>
>> The env spec states:
>>    OpUndef is not allowed.
>>
>> So, the shader would be rejected as invalid.
>>
>> dan
>>
>>
Received on Thursday, 22 November 2018 19:47:49 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 19:52:25 UTC