Ecma TC57 - HLSL

Minutes of the:16th meeting of Ecma TC57
held in:Teleconference
on:13 October 2026
Chairs:Chris Bieneman (Microsoft)
Attendees:Alexander Johnston (AMD), Daniel Brown (AMD), Greg Mitrano (Microsoft), João Saffran (Microsoft), Mark Herdeg (AMD), Hai Nguyen (Nvidia), Gregory Roth (Nvidia), Helena Kotas (Microsoft), Keith Stockdale (Microsoft), Damyan Pepper (Microsoft), Sarah Spall (Microsoft), Jesse Barker (Invited Expert)
Secretary:Aki Rose Braun (Ecma International)

1 Welcome, opening and meeting logistics

Chris Bieneman, chair of TC57, began the teleconference and welcomed attendees. Attendees were reminded of the committee’s code of conduct, Ecma’s IPR policy, and TC57’s royalty-free patent policy extension. All attendees were informed that the meeting was recorded and transcribed by Zoom for the purpose of note-taking, and directed to review the video recording and legal disclaimer if they were not familiar.

2 Approval of the minutes of previous meetings

There were no minutes available for approval at this meeting.

3 Review of the agenda

The meeting’s agenda pull request was proposed and approved at the start of the meeting.

4 Editors’ report

There was no editors’ report today.

5 Proposals

5.1 New proposals

There were no proposals today.

5.2 Stage advancement

Any proposals failing to advance will be addressed in the next meeting.

5.3 Future stage advancement

The following proposals and PRs will seek advancement/approval to merge at the 17th plenary on 27 October. The chair requests that all delegates review and engage in the conversation to resolve any potential concerns before then.

6 Discussions

6.1 Normative consensus PRs

The only open normative consensus PR was Math.GeneralFunctions.

Gregory Roth (Nvidia) expressed a lack of confidence in where the prose describes smooth step, bringing up that there are open questions regarding how it is defined. He suggested removing that section to prevent his concern from blocking the rest of the pull request, and adding the smooth step piece at a later date. Keith had left a series of comments, all primarily targeted at editorial changes such as typos; Gregory provided easy-to-apply suggestions on the PR for many of them. Gregory also brought up fused vs unfused operations, and contested the text that described fused operations as higher precision. Chris agreed and provided a tweak to the language. The next comment reviewed was one from Keith regarding the “precise” keyword, and its guarantees (or lack thereof). Chris explained that “precise” actually more accurately accomplished consistency over precision. The committee came up with replacement text to clarify.

The committee worked through the language and approved the PR to be merged.

6.2 Should structs with mutable fields be allowed in constant resources?

Chris proposes that HLSL error when a struct containing a mutable member is inside a resource that cannot be written to, such as constant buffers, texture buffers, or non-writable structure buffers. This can be a protective measure for both the user experience and compiler resilience. There is no C++ prior art to rely on, because C++ has no notion of memory that is truly read-only. Hai Nguyen (Nvidia) was interested in what the structure of that error message would be, Chris proposed something to the effect of “you cannot put a structure with a mutable field inside a read-only resource”.

Keith & Greg Mitrano (Microsoft) were trying to figure out how to do this without then resulting in a user making a copy of that struct to run some operation on it, leading to two copies of a struct in play.

The committee explored producing the error as a compilation error, but lacked confidence that the error would be statically knowable in the long-term development of the language, such as implementing references.

Ultimately, Hai expressed an appreciation for this suggestion from the perspective of language purity, but suggested that pragmatically it was not an ideal outcome for HLSL developers without HLSL gaining other planned features such as conversion functions.

Helena Kotas (Microsoft) initially suggested that this should at least be a warning, even if the committee wasn’t completely sold on throwing an error.

After some additional back and forth, Chris pointed out that due to the fact that const member functions is a new feature, existing code wouldn’t be broken by making this an error. The committee then quickly reached consensus on making it an error.

6.3 Initializers on groupshared values

DXC ignores initializers on group shared values, but only sometimes the compiler provides a warning—others it ignores them silently. Chris proposes introducing some consistency here, aligning ignored initializers on group shared memory with ignored initializers on cbuffer values (i.e. error).

There is a viable alternative Chris provided—making it work. The drawback here is that it could impact existing code unexpectedly. Gregory Mitrano shared that there is some hardware that can initialize to 0, so this may be more than a basic quality-of-life improvement. Chris suggested making a change for the purpose of that accommodation might be something Microsoft & Kronos may choose to standardize separately.

Overall, the committee is aligned on this being illegal. Chris will prep a PR for a future meeting.

6.4 Add 0/1 swizzle syntax

There’s currently a proposal (0011) to change initialization and remove the ability to append a literal value to the end of a vector in order to change its dimension or add a homogeneous coordinate. It’s not currently planned for 202x, but at any point it where to be added, Gregory Roth proposes 0/1 swizzle syntax, which could restore some developer ergonomics lost by 0011’s strictness.

There was support in the committee to pursue this in either 202x or 202y and independent of 0011, specifically allowing placement a 0/1 in any element for non-literals rather than the previously-suggested to only allow 0/1 after the first position

7 Any other business

.

8 Next meetings

The next meeting will be held on Tuesday, 00 MONTH 2026 at 09:00 PDT.

Meetings are held via Zoom call every other Tuesday at 09:00 PDT (12:00 EDT, 18:00 CEST). Meeting times during seasonal time changes will be scheduled based on US Pacific (America/Los_Angeles) time.

8.1 Future Agendas

Agendas are viewable in GitHub and remain available in perpetuity.

9 Conclusion

The Chairs wrapped up the meeting and the call concluded.