The origin trial resulted in the following findings: 1. This functionality can achieve the performance improvements we expected. Meet integrated the functionality into some of its ML shaders. Benchmarking subgroups vs integer dot products (previous best) for matrix-vector multiply shaders resulted in speed ups of 2.3 - 2.9x depending on the device. 2. Limits were not the right design choice for subgroup size reporting in the API. They have since been moved into GPUAdapterInfo as they are really more static properties of a device. 3. The initial design contained an extra diagnostic that defaulted to an error to increase portability, but was found to hamper the user experience more than the value it added. It prevented very common code patterns unrelated to the subgroup functionality from compiling cleanly. It was decided to remove it from the design.
On Wed, Jan 8, 2025 at 4:16 PM Chris Harrelson <chris...@chromium.org> wrote: > Ok thanks for clarifying that this is shipping and not further > experimentation. > > Could you share the findings from the origin trials? > > On Wed, Jan 8, 2025 at 11:31 AM 'Alan Baker' via blink-dev < > blink-dev@chromium.org> wrote: > >> I've filed the standard's position issues (and added to the feature). >> Here they are for reference: >> >> - Mozilla: https://github.com/mozilla/standards-positions/issues/1160 >> - WebKit: https://github.com/WebKit/standards-positions/issues/446 >> >> The name was a bad choice. I wasn't familiar with this process initially >> and only thought of the name with the Origin trial in mind. We don't >> consider this an experiment anymore. >> >> On Wed, Jan 8, 2025 at 11:31 AM Chris Harrelson <chris...@chromium.org> >> wrote: >> >>> Also: >>> >>> >>> * The spec PR needs to land >>> * What do you mean by "experimentation" in the title of this Intent? >>> >>> On Wed, Jan 8, 2025 at 8:30 AM Alex Russell <slightly...@chromium.org> >>> wrote: >>> >>>> Hey Alan, >>>> >>>> Just for the avoidance of future friction, we try to use the explicit >>>> signals systems from Apple and Mozilla because they've asked us not to use >>>> their rep's statements in WGs as assent. Is it possible for you to file >>>> those? >>>> >>>> Best, >>>> >>>> Alex >>>> >>>> On Monday, January 6, 2025 at 10:48:36 AM UTC-8 Alan Baker wrote: >>>> >>>>> Contact emailsalanba...@google.com >>>>> >>>>> Explainer >>>>> https://github.com/gpuweb/gpuweb/blob/main/proposals/subgroups.md >>>>> >>>>> Specificationhttps://github.com/gpuweb/gpuweb/pull/4963 >>>>> >>>>> Approved by Apple rep, editorial comment by Mozilla rep. >>>>> >>>>> Summary >>>>> >>>>> Adds subgroup functionality to WebGPU. Subgroup operations perform >>>>> SIMT operations to provide efficient communication and data sharing among >>>>> groups of invocations. These operations can be used to accelerate >>>>> applications by reducing memory overheads incurred by inter-invocation >>>>> communication. >>>>> >>>>> >>>>> Blink componentBlink>WebGPU >>>>> <https://issues.chromium.org/issues?q=customfield1222907:%22Blink%3EWebGPU%22> >>>>> >>>>> TAG reviewNone >>>>> >>>>> TAG review statusNot applicable >>>>> >>>>> Origin Trial NameWebGPU Subgroups Features >>>>> >>>>> Chromium Trial NameWebGPUSubgroupsFeatures >>>>> >>>>> Origin Trial documentation link >>>>> https://github.com/gpuweb/gpuweb/blob/main/proposals/subgroups.md >>>>> >>>>> WebFeature UseCounter namekWebGPUSubgroupsFeatures >>>>> >>>>> Risks >>>>> >>>>> >>>>> Interoperability and Compatibility >>>>> >>>>> None >>>>> >>>>> >>>>> *Gecko*: No signal >>>>> >>>>> *WebKit*: No signal >>>>> >>>>> *Web developers*: No signals >>>>> >>>>> *Other signals*: >>>>> >>>>> Security >>>>> >>>>> None. This feature exposes functionality for operations only within a >>>>> single draw/dispatch. Data is contained within a single users run. >>>>> >>>>> >>>>> WebView application risks >>>>> >>>>> Does this intent deprecate or change behavior of existing APIs, such >>>>> that it has potentially high risk for Android WebView-based applications? >>>>> >>>>> None >>>>> >>>>> >>>>> Debuggability >>>>> >>>>> No special debugging is available for this feature as it resides >>>>> solely on the GPU. >>>>> >>>>> >>>>> Will this feature be supported on all six Blink platforms (Windows, >>>>> Mac, Linux, ChromeOS, Android, and Android WebView)?No >>>>> >>>>> Is this feature fully tested by web-platform-tests >>>>> <https://chromium.googlesource.com/chromium/src/+/main/docs/testing/web_platform_tests.md> >>>>> ?No >>>>> >>>>> Flag name on about://flagsNone >>>>> >>>>> Finch feature nameNone >>>>> >>>>> Non-finch justificationNone >>>>> >>>>> Requires code in //chrome?False >>>>> >>>>> Availability expectationBoth Apple and Mozilla have indicated they >>>>> plan to support this feature in the future. Those implementations are >>>>> catching up to Chrome in terms of core functionality currently. WebGPU is >>>>> part of Safari technical previews and Mozilla nightly builds. >>>>> >>>>> Adoption expectationThis feature should have wide adoption for >>>>> specific use cases. It is not expected that all WebGPU shaders would >>>>> utilize subgroups, but many applications (e.g. ML) could. >>>>> >>>>> Non-OSS dependencies >>>>> >>>>> Does the feature depend on any code or APIs outside the Chromium open >>>>> source repository and its open-source dependencies to function? >>>>> No >>>>> >>>>> Estimated milestones >>>>> Shipping on desktop 134 >>>>> Origin trial desktop first 128 >>>>> Origin trial desktop last 131 >>>>> Origin trial extension 1 end milestone 133 >>>>> Shipping on Android 134 >>>>> Origin trial Android first 128 >>>>> Origin trial Android last 131 >>>>> Shipping on WebView 134 >>>>> Origin trial WebView first 128 >>>>> Origin trial WebView last 131 >>>>> >>>>> Anticipated spec changes >>>>> >>>>> Open questions about a feature may be a source of future web compat or >>>>> interop issues. Please list open issues (e.g. links to known github issues >>>>> in the project for the feature specification) whose resolution may >>>>> introduce web compat/interop risk (e.g., changing to naming or structure >>>>> of >>>>> the API in a non-backward-compatible way). >>>>> Editorial changes only. Future features are likely to build on top of >>>>> this extension. >>>>> >>>>> Link to entry on the Chrome Platform Status >>>>> https://chromestatus.com/feature/5126409856221184?gate=5073549680443392 >>>>> >>>>> Links to previous Intent discussionsIntent to Experiment: >>>>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABg%2BrB67gB5gUeDVPaMZ7fZwmeARTUsizs1BTXFXKGpJ%3DpF1TQ%40mail.gmail.com >>>>> Intent to Extend Experiment 1: >>>>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABg%2BrB5jfO13yZGBxMAHRgZX_P0qbixCKSK10SujYkBxftujnQ%40mail.gmail.com >>>>> >>>>> >>>>> This intent message was generated by Chrome Platform Status >>>>> <https://chromestatus.com/>. >>>>> >>>> -- >>>> You received this message because you are subscribed to the Google >>>> Groups "blink-dev" group. >>>> To unsubscribe from this group and stop receiving emails from it, send >>>> an email to blink-dev+unsubscr...@chromium.org. >>>> To view this discussion visit >>>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/1969ee25-f88c-482d-adb1-88846d1d4e64n%40chromium.org >>>> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/1969ee25-f88c-482d-adb1-88846d1d4e64n%40chromium.org?utm_medium=email&utm_source=footer> >>>> . >>>> >>> -- >> You received this message because you are subscribed to the Google Groups >> "blink-dev" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to blink-dev+unsubscr...@chromium.org. >> To view this discussion visit >> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABg%2BrB7yinOiFXX1ud4hLU9Y2F-9w8O1EL4KZ-KTWDOGCt-B2A%40mail.gmail.com >> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABg%2BrB7yinOiFXX1ud4hLU9Y2F-9w8O1EL4KZ-KTWDOGCt-B2A%40mail.gmail.com?utm_medium=email&utm_source=footer> >> . >> > -- You received this message because you are subscribed to the Google Groups "blink-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+unsubscr...@chromium.org. To view this discussion visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABg%2BrB5PA7QJku0dWkS2_jXOVC4V6dWAttbF850jEZmKu684Fg%40mail.gmail.com.
