Thank you for the details, Deepti. I'm not an owner but I would recommend going for a 3-release Origin Trial and assess the situation then and decide whether extending makes sense. If there isn't enough usage by then, extending would probably be an easy call. It seems better to me to assess it then than preemptively set the experiment length to 5 releases assuming the usage and feedback will take some time to ramp up.
-- Mounir On Fri, 3 Apr 2020 at 13:21, Deepti Gandluri <gdee...@chromium.org> wrote: > The partners we have lined up, currently have a (mostly) full SIMD version > of their applications ready. So they would be able to enable this right > away. > > There are a couple of issues I was hoping to address with a longer trial, > for the partners that currently have an in progress implementation, as well > as for partners where to get feedback, there is a level of indirection > involved when it comes to how users can take advantage of this. To clarify > what I mean by the second use case - for partners like Halide > <https://github.com/halide/Halide>, and Tensorflow.js > <https://blog.tensorflow.org/2020/03/introducing-webassembly-backend-for-tensorflow-js.html>, > which have WebAssembly backends with experimental SIMD support, to get > feedback their users will have to sign up for an origin trial, as they > serve as backends/libraries and are not end users themselves. So some of > that buffer was to ensure we can reach a larger number of users - I'm > somewhat unsure how to account for that type of indirection as I haven't > found many other examples where this is the case, so feedback here is > welcome. > > With respect to "ongoing standardization effort", I should have been > clearer there. The proposal is currently at Phase 3 > <https://github.com/WebAssembly/meetings/blob/master/process/phases.md#3-implementation-phase-community--working-group>, > and the standardization effort here refers to moving to Phase 4 > <https://github.com/WebAssembly/meetings/blob/master/process/phases.md#4-standardize-the-feature-working-group>, > the requirement for which is that there needs to be one other Web VM that > implements this. At this point, the shape of the API is quite stable, we > have an implementation in both V8, as well as the WebAssembly LLVM > toolchain, and we are waiting on more engine implementations to move the > proposal forward. > > On Friday, April 3, 2020 at 11:05:32 AM UTC-7, Mounir Lamouri wrote: >> >> Deepti, what did you mean by "ongoing standardization effort"? Is the API >> shape still being discussed? >> >> On Thu, 2 Apr 2020 at 14:12, Yoav Weiss <y...@yoav.ws> wrote: >> >>> I'm wondering if there's something preventing partners from implementing >>> SIMD versions of their applications with the behind-the-flag implementation? >>> >>> I'm assuming you already have some partners lined up for the experiment. >>> Would they be able to develop using the flagged implementation? That would >>> enable us to start the origin trial when they're ready, and as a result, >>> have a shorter trial which purpose is to gather real life usage metrics. >>> >>> On Fri, Mar 27, 2020 at 11:02 PM Deepti Gandluri <gdee...@chromium.org> >>> wrote: >>> >>>> Hi Mounir, >>>> >>>> The reason for a longer trial here is that this proposal is somewhat >>>> large, and while enabling auto-vectorization can give some performance >>>> benefits, to really experiment with this applications may have to rewrite >>>> their existing code that uses native SIMD intrinsics to use portable Wasm >>>> intrinsics (example: Emscripten SIMD intrinsics header >>>> <https://github.com/emscripten-core/emscripten/blob/master/system/include/wasm_simd128.h>), >>>> or builtins that map to Wasm operations. >>>> >>>> The intent is to build in enough time for new applications to >>>> experiment with this in parallel to ongoing standardization effort in the >>>> WebAssembly CG. The ballpark estimate I've used here is from previous >>>> applications experimenting with SIMD, with some buffer to gather data. That >>>> said, I do understand that this is a longer trial than is usual, so would >>>> be open to adjusting the timeline based on concerns. >>>> >>>> Thanks, >>>> Deepti >>>> >>>> On Friday, March 27, 2020 at 1:29:09 PM UTC-7, Mounir Lamouri wrote: >>>>> >>>>> Hi Deepti, >>>>> >>>>> This Intent to Experiment looks good, though, I'm wondering why this >>>>> is expected to run for 5 releases, which is roughly 6 months. Is there >>>>> something specific we are concerned about and that this timeline aims to >>>>> solve? >>>>> >>>>> -- Mounir >>>>> >>>>> On Fri, 27 Mar 2020 at 07:22, Deepti Gandluri <gdee...@chromium.org> >>>>> wrote: >>>>> >>>>>> Contact emails >>>>>> >>>>>> gdee...@chromium.org, nattes...@chromium.org, z...@chromium.org >>>>>> >>>>>> >>>>>> Spec >>>>>> >>>>>> Overview document: >>>>>> https://github.com/WebAssembly/simd/blob/master/proposals/simd/SIMD.md >>>>>> >>>>>> Proposal directory: >>>>>> https://github.com/WebAssembly/simd/tree/master/proposals/simd >>>>>> >>>>>> Tag review: https://github.com/w3ctag/design-reviews/issues/487 >>>>>> >>>>>> >>>>>> Summary >>>>>> >>>>>> The WebAssembly SIMD proposal defines a portable, performant subset >>>>>> of SIMD operations that map to vector instructions that are available >>>>>> across most modern hardware platforms. >>>>>> >>>>>> >>>>>> This is purely a WebAssembly performance feature that adds >>>>>> WebAssembly operations and does not affect web API behavior, but is still >>>>>> useful for developers to be aware of as it can change performance >>>>>> characteristics of applications using WebAssembly. >>>>>> >>>>>> >>>>>> Link to “Intent to Prototype” blink-dev discussion: >>>>>> >>>>>> This links to the Intent to Implement discussion for Simd.js >>>>>> <https://groups.google.com/a/chromium.org/forum/#!searchin/blink-dev/Simd.js%7Csort:date/blink-dev/rhPPSQVodq0/zr4HRxXjBgAJ>, >>>>>> which was the basis for the WebAssembly SIMD. The Spec section above has >>>>>> updated links for WebAssembly SIMD, as well as the Tag review. >>>>>> >>>>>> >>>>>> Goals for experimentation >>>>>> >>>>>> Performance: >>>>>> >>>>>> The goal for this experiment is to determine the performance benefit >>>>>> of having WebAssembly SIMD enabled applications in production code, and >>>>>> determine real world performance numbers. We currently have data from >>>>>> isolated benchmarks, and demos, but these do not always reflect real >>>>>> world >>>>>> usage. >>>>>> >>>>>> >>>>>> API Shape: >>>>>> >>>>>> The SIMD proposal currently consists of a set of operations that are >>>>>> well supported across most modern hardware platforms. The set of >>>>>> instructions available on hardware is large and varied, and we are >>>>>> looking >>>>>> for feedback on whether this portable subset is sufficient for a cross >>>>>> section of applications. >>>>>> >>>>>> >>>>>> Experimental timeline >>>>>> >>>>>> M84-M89 >>>>>> >>>>>> >>>>>> Any risks when the experiment finishes? >>>>>> >>>>>> The risk is a drop in performance for sites that were enhancing >>>>>> performance using WebAssembly SIMD operations. >>>>>> >>>>>> >>>>>> Ongoing technical constraints >>>>>> >>>>>> None. >>>>>> >>>>>> >>>>>> Debuggability >>>>>> >>>>>> WebAssembly SIMD needs debugging support in devtools for 128-bit >>>>>> values, as well as baseline compiler support in V8. Both of these efforts >>>>>> are ongoing. (Tracking bugs: crbug.com/v8/10347, crbug.com/v8/9909) >>>>>> >>>>>> >>>>>> Will this feature be supported on all five Blink platforms supported >>>>>> by Origin Trials (Windows, Mac, Linux, Chrome OS, and Android)? >>>>>> >>>>>> Yes. >>>>>> >>>>>> >>>>>> Link to entry on the feature dashboard >>>>>> <https://www.chromestatus.com/> >>>>>> >>>>>> https://www.chromestatus.com/feature/6533147810332672 >>>>>> >>>>>> -- >>>>>> 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 on the web visit >>>>>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/e4010247-bb4c-47a4-9b82-ea8188c3b232%40chromium.org >>>>>> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/e4010247-bb4c-47a4-9b82-ea8188c3b232%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 on the web visit >>>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/e9f7ebf3-495e-46b7-82bd-a0f963f1625d%40chromium.org >>>> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/e9f7ebf3-495e-46b7-82bd-a0f963f1625d%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 on the web visit >>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CACj%3DBEh8BicQHAygqbS0icM4R_%2B9qgzWmOS%2BmqtAicWiqZmRkA%40mail.gmail.com >>> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CACj%3DBEh8BicQHAygqbS0icM4R_%2B9qgzWmOS%2BmqtAicWiqZmRkA%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 on the web visit > https://groups.google.com/a/chromium.org/d/msgid/blink-dev/722401d6-1424-47c5-913f-33092ae4b6d0%40chromium.org > <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/722401d6-1424-47c5-913f-33092ae4b6d0%40chromium.org?utm_medium=email&utm_source=footer> > . > -- -- v8-dev mailing list v8-dev@googlegroups.com http://groups.google.com/group/v8-dev --- You received this message because you are subscribed to the Google Groups "v8-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to v8-dev+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/v8-dev/CA%2B-LeH8KR2jk3g6zCJxtd7pBqoEL%2B2fp32BUQvpDgnMzwGTjrA%40mail.gmail.com.
