[naga wgsl-in] Experimental 64-bit floating-point literals. by jimblandy · Pull Request #4747 · gfx-rs/wgpu

jimblandy · 2023-11-22T04:25:23Z

In the WGSL front end, support an l suffix on floating-point literals to yield 64-bit integer literals.

Unless Capabilities::FLOAT64 is selected, these literals will always be validated out, so it doesn't affect standards compliance.

This PR is questionable. I guess it's nice to have f64 support to play with (edit: and the front and back ends already parse and emit an f64 WGSL type), but it's hardly urgent. Who knows whether the spec will ever actually have f64 support, and if it does, what the syntax will be?

But the truth is, I have ulterior motives for this patch. It's a little circuitous:

On the abstract type branch, all constructor built-in functions know how to consume abstract values. Calls to scalar constructor built-in functions (say, f32(5)) are handled by generating As expressions and letting the constant evaluator convert abstract values to concrete values. So f32(5) constant-evaluates the AbstractInt 5 to the f32 5.0.
It happens that the test case tests/out/wgsl/double-math-functions.frag.wgsl contains expressions like f64(1.0). It's translated from GLSL, which does have a double type.
Naga should either accept f64(1.0) or reject it.
- If Naga rejects it, then we cannot translate tests/in/glsl/double-math-functions.frag to WGSL. The GLSL snapshots aren't really set up to generate other output languages; I don't really want to change that. Fine, maybe we back this test out, it's hardly essential. But let's assume that we want to accept it.
- If Naga accepts it, then the only reasonable semantics for this expression should be that applying the f64 constructor built-in function to an AbstractFloat gives you an f64.
ConstantEvaluator::cast doesn't happen to support casting to f64. Since it returns an error, the WGSL front end assumes that the expression is not constant-evaluatable, and leaves it in the IR as anAbstractFloat literal, which the validator then rejects.
Expanding ConstantEvaluator::cast to support casting to f64 allows the out/wgsl file to validate.
However, it also means that Literal::F64 values start making it to the WGSL backend, which errors out because WGSL has no f64 literal syntax. Hence the present PR.

So it seems to me we have two options:

Give WGSL syntax for 64-bit literals, and have it generate them in tests/out/wgsl/double-math-functions.frag.wgsl and then read and validate them.
Remove tests/in/glsl/double-math-functions.frag and forget all this.

Checklist

Run cargo fmt.
Run cargo clippy. If applicable, add:
Run cargo xtask test to run tests.
Add change to CHANGELOG.md. See simple instructions inside file.

jimblandy · 2023-11-22T05:22:20Z

Point in favor of this PR: It seems like the front and back ends already support an f64 WGSL type. We just don't have a syntax for literals of that type.

If we do feel that adding experimental f64 support is okay, then it would probably make sense to pair it with backend support for literals as well. There should then be snapshot tests for the literals too.

cwfitzgerald · 2023-11-22T06:22:12Z

Speaking from community side, having support for this is one step closer to end-to-end f64 support, something that is asked for a decent amount, particularly from the scientifically minded folks.

teoxoy · 2023-11-22T11:46:15Z

Give WGSL syntax for 64-bit literals, it generate them in tests/out/wgsl/double-math-functions.frag.wgsl and then read and validate them.

I'm also in favor of supporting f64 👍

naga/src/front/wgsl/parse/number.rs

jimblandy · 2023-11-22T18:09:48Z

Okay, here's full front and back end support with an lf suffix.

In the WGSL front and back ends, support an `lf` suffix on floating-point literals to yield 64-bit integer literals.

…7ab0a860eec. r=webgpu-reviewers,supply-chain-reviewers,ErichDonGubler # Changelog * #4708 Fix max_vertex_buffers validation By nical in gfx-rs/wgpu#4708 * #4729 Remove expected failure on AMD/DX12 from msaa example By teoxoy in gfx-rs/wgpu#4729 * #4711 [naga] Add `Literal::I64`, for signed 64-bit integer literals. By jimblandy in gfx-rs/wgpu#4711 * #4736 Bump serde from 1.0.192 to 1.0.193 By dependabot[bot] in gfx-rs/wgpu#4736 * #4735 [naga]: Let `TypeInner::Matrix` hold a `Scalar`, not just a width. By jimblandy in gfx-rs/wgpu#4735 * #4741 [naga] Fix type error in test. By jimblandy in gfx-rs/wgpu#4741 * #4742 [naga]: Make snapshot tests include paths in errors. By jimblandy in gfx-rs/wgpu#4742 * #4723 Test And Normalize Vertex Behavior on All Backends By cwfitzgerald in gfx-rs/wgpu#4723 * #4746 Skip Flaky Test By cwfitzgerald in gfx-rs/wgpu#4746 * #4744 Ensure DeviceLostClosureC callbacks have null-terminated message strings By bradwerth in gfx-rs/wgpu#4744 * #4745 [naga wgsl-in] Test hex float suffix handling corner case. By jimblandy in gfx-rs/wgpu#4745 * #4737 Make the command_encoder_clear_buffer's size an Option<BufferAddress> By nical in gfx-rs/wgpu#4737 * #4701 [naga msl-out] Emit and init `struct` member padding always By ErichDonGubler in gfx-rs/wgpu#4701 * #4701 [naga msl-out] Emit and init `struct` member padding always By ErichDonGubler in gfx-rs/wgpu#4701 * #4701 [naga msl-out] Emit and init `struct` member padding always By ErichDonGubler in gfx-rs/wgpu#4701 * #4701 [naga msl-out] Emit and init `struct` member padding always By ErichDonGubler in gfx-rs/wgpu#4701 * #4733 [naga wgsl-in] Preserve type names in `alias` declarations. By jimblandy in gfx-rs/wgpu#4733 * #4734 [naga] Make compaction preserve named types, even if unused. By jimblandy in gfx-rs/wgpu#4734 * #4752 Logging cleanups in device/global.rs By nical in gfx-rs/wgpu#4752 * #4753 Fix Mac Event Loop By cwfitzgerald in gfx-rs/wgpu#4753 * #4754 wgpu-hal(vk): Add WGPU_ALLOW_NONCOMPLIANT_ADAPTER By i509VCB in gfx-rs/wgpu#4754 * #4748 Allow Tests to Expect Certain Panic or Validation Messages By cwfitzgerald in gfx-rs/wgpu#4748 * #4756 Move to A Single Example Executable By cwfitzgerald in gfx-rs/wgpu#4756 * #4747 [naga wgsl-in] Experimental 64-bit floating-point literals. By jimblandy in gfx-rs/wgpu#4747 * #4747 [naga wgsl-in] Experimental 64-bit floating-point literals. By jimblandy in gfx-rs/wgpu#4747 * #4761 [naga] Make the `example_wgsl` test build without `wgsl-in` feature. By jimblandy in gfx-rs/wgpu#4761 * #4769 Conditionally lift API logging from trace to info level By nical in gfx-rs/wgpu#4769 * #4771 Downgrade some of wgpu_core's logging level from info to trace and debug By nical in gfx-rs/wgpu#4771 * #4760 Rename `ALLOW_NONCOMPLIANT_ADAPTER` to `ALLOW_UNDERLYING_NONCOMPLIANT_ADAPTER` By teoxoy in gfx-rs/wgpu#4760 * #4772 Downgrade resource lifetime management log level to trace. By nical in gfx-rs/wgpu#4772 * #4765 Revamp Examples to Match Website By cwfitzgerald in gfx-rs/wgpu#4765 * #4774 Update examples readme files By roffs in gfx-rs/wgpu#4774 * #4781 remove_abandoned fix By gents83 in gfx-rs/wgpu#4781 * #4777 Bump web-sys to 0.3.65 By torokati44 in gfx-rs/wgpu#4777 * #4780 Fixes for Skybox and All Examples By cwfitzgerald in gfx-rs/wgpu#4780 * #4782 Remove resources ONLY when needed inside wgpu and not in user land By gents83 in gfx-rs/wgpu#4782 * #4786 Don't keep a strong ref in storage for destroyed resources By gents83 in gfx-rs/wgpu#4786 * #4795 Fix Arcanization changelog By a1phyr in gfx-rs/wgpu#4795 * #4794 Delete/rename now-unused gpu.rs By exrook in gfx-rs/wgpu#4794 * #4794 Delete/rename now-unused gpu.rs By exrook in gfx-rs/wgpu#4794 * #4573 Support nv12 texture format By xiaopengli89 in gfx-rs/wgpu#4573 * #4789 Bump js-sys from 0.3.65 to 0.3.66 By dependabot[bot] in gfx-rs/wgpu#4789 * #4784 Bump winit from 0.29.3 to 0.29.4 By dependabot[bot] in gfx-rs/wgpu#4784 * #4798 Bump JamesIves/github-pages-deploy-action from 4.4.3 to 4.5.0 By dependabot[bot] in gfx-rs/wgpu#4798 * #4800 Bump web-sys from 0.3.65 to 0.3.66 By dependabot[bot] in gfx-rs/wgpu#4800 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4799 Bump wasm-bindgen-futures from 0.4.38 to 0.4.39 By dependabot[bot] in gfx-rs/wgpu#4799 * #4796 Remove surface extent validation (and thus fix the annoying `Requested size ... is outside of the supported range` warning) By Wumpf in gfx-rs/wgpu#4796 * #4804 Add space to a comment inside make_spirv_raw By ComfyFluffy in gfx-rs/wgpu#4804 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4805 [naga] Improve ConstantEvaluatorError::InvalidCastArg message. By jimblandy in gfx-rs/wgpu#4805 * #4801 update deno By crowlKats in gfx-rs/wgpu#4801 * #4806 Fixes and changes to the documentation for increasing clarity By Blatko1 in gfx-rs/wgpu#4806 * #4809 [naga wgsl-in] Use a better span for errors in constructors. By jimblandy in gfx-rs/wgpu#4809 * #4808 [naga wgsl-in] Drop spanless labels from front-end error messages. By jimblandy in gfx-rs/wgpu#4808 * #4822 Fix expected error message By teoxoy in gfx-rs/wgpu#4822 * #4759 Add feature float32-filterable By almarklein in gfx-rs/wgpu#4759 * #4813 Bump core-graphics-types from 0.1.2 to 0.1.3 By dependabot[bot] in gfx-rs/wgpu#4813 Differential Revision: https://phabricator.services.mozilla.com/D195438

…7ab0a860eec. r=webgpu-reviewers,supply-chain-reviewers,ErichDonGubler # Changelog * #4708 Fix max_vertex_buffers validation By nical in gfx-rs/wgpu#4708 * #4729 Remove expected failure on AMD/DX12 from msaa example By teoxoy in gfx-rs/wgpu#4729 * #4711 [naga] Add `Literal::I64`, for signed 64-bit integer literals. By jimblandy in gfx-rs/wgpu#4711 * #4736 Bump serde from 1.0.192 to 1.0.193 By dependabot[bot] in gfx-rs/wgpu#4736 * #4735 [naga]: Let `TypeInner::Matrix` hold a `Scalar`, not just a width. By jimblandy in gfx-rs/wgpu#4735 * #4741 [naga] Fix type error in test. By jimblandy in gfx-rs/wgpu#4741 * #4742 [naga]: Make snapshot tests include paths in errors. By jimblandy in gfx-rs/wgpu#4742 * #4723 Test And Normalize Vertex Behavior on All Backends By cwfitzgerald in gfx-rs/wgpu#4723 * #4746 Skip Flaky Test By cwfitzgerald in gfx-rs/wgpu#4746 * #4744 Ensure DeviceLostClosureC callbacks have null-terminated message strings By bradwerth in gfx-rs/wgpu#4744 * #4745 [naga wgsl-in] Test hex float suffix handling corner case. By jimblandy in gfx-rs/wgpu#4745 * #4737 Make the command_encoder_clear_buffer's size an Option<BufferAddress> By nical in gfx-rs/wgpu#4737 * #4701 [naga msl-out] Emit and init `struct` member padding always By ErichDonGubler in gfx-rs/wgpu#4701 * #4701 [naga msl-out] Emit and init `struct` member padding always By ErichDonGubler in gfx-rs/wgpu#4701 * #4701 [naga msl-out] Emit and init `struct` member padding always By ErichDonGubler in gfx-rs/wgpu#4701 * #4701 [naga msl-out] Emit and init `struct` member padding always By ErichDonGubler in gfx-rs/wgpu#4701 * #4733 [naga wgsl-in] Preserve type names in `alias` declarations. By jimblandy in gfx-rs/wgpu#4733 * #4734 [naga] Make compaction preserve named types, even if unused. By jimblandy in gfx-rs/wgpu#4734 * #4752 Logging cleanups in device/global.rs By nical in gfx-rs/wgpu#4752 * #4753 Fix Mac Event Loop By cwfitzgerald in gfx-rs/wgpu#4753 * #4754 wgpu-hal(vk): Add WGPU_ALLOW_NONCOMPLIANT_ADAPTER By i509VCB in gfx-rs/wgpu#4754 * #4748 Allow Tests to Expect Certain Panic or Validation Messages By cwfitzgerald in gfx-rs/wgpu#4748 * #4756 Move to A Single Example Executable By cwfitzgerald in gfx-rs/wgpu#4756 * #4747 [naga wgsl-in] Experimental 64-bit floating-point literals. By jimblandy in gfx-rs/wgpu#4747 * #4747 [naga wgsl-in] Experimental 64-bit floating-point literals. By jimblandy in gfx-rs/wgpu#4747 * #4761 [naga] Make the `example_wgsl` test build without `wgsl-in` feature. By jimblandy in gfx-rs/wgpu#4761 * #4769 Conditionally lift API logging from trace to info level By nical in gfx-rs/wgpu#4769 * #4771 Downgrade some of wgpu_core's logging level from info to trace and debug By nical in gfx-rs/wgpu#4771 * #4760 Rename `ALLOW_NONCOMPLIANT_ADAPTER` to `ALLOW_UNDERLYING_NONCOMPLIANT_ADAPTER` By teoxoy in gfx-rs/wgpu#4760 * #4772 Downgrade resource lifetime management log level to trace. By nical in gfx-rs/wgpu#4772 * #4765 Revamp Examples to Match Website By cwfitzgerald in gfx-rs/wgpu#4765 * #4774 Update examples readme files By roffs in gfx-rs/wgpu#4774 * #4781 remove_abandoned fix By gents83 in gfx-rs/wgpu#4781 * #4777 Bump web-sys to 0.3.65 By torokati44 in gfx-rs/wgpu#4777 * #4780 Fixes for Skybox and All Examples By cwfitzgerald in gfx-rs/wgpu#4780 * #4782 Remove resources ONLY when needed inside wgpu and not in user land By gents83 in gfx-rs/wgpu#4782 * #4786 Don't keep a strong ref in storage for destroyed resources By gents83 in gfx-rs/wgpu#4786 * #4795 Fix Arcanization changelog By a1phyr in gfx-rs/wgpu#4795 * #4794 Delete/rename now-unused gpu.rs By exrook in gfx-rs/wgpu#4794 * #4794 Delete/rename now-unused gpu.rs By exrook in gfx-rs/wgpu#4794 * #4573 Support nv12 texture format By xiaopengli89 in gfx-rs/wgpu#4573 * #4789 Bump js-sys from 0.3.65 to 0.3.66 By dependabot[bot] in gfx-rs/wgpu#4789 * #4784 Bump winit from 0.29.3 to 0.29.4 By dependabot[bot] in gfx-rs/wgpu#4784 * #4798 Bump JamesIves/github-pages-deploy-action from 4.4.3 to 4.5.0 By dependabot[bot] in gfx-rs/wgpu#4798 * #4800 Bump web-sys from 0.3.65 to 0.3.66 By dependabot[bot] in gfx-rs/wgpu#4800 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4743 Implement WGSL abstract types for global `const` declarations and constructor calls. By jimblandy in gfx-rs/wgpu#4743 * #4799 Bump wasm-bindgen-futures from 0.4.38 to 0.4.39 By dependabot[bot] in gfx-rs/wgpu#4799 * #4796 Remove surface extent validation (and thus fix the annoying `Requested size ... is outside of the supported range` warning) By Wumpf in gfx-rs/wgpu#4796 * #4804 Add space to a comment inside make_spirv_raw By ComfyFluffy in gfx-rs/wgpu#4804 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4803 `features`/`limits` refactors By teoxoy in gfx-rs/wgpu#4803 * #4805 [naga] Improve ConstantEvaluatorError::InvalidCastArg message. By jimblandy in gfx-rs/wgpu#4805 * #4801 update deno By crowlKats in gfx-rs/wgpu#4801 * #4806 Fixes and changes to the documentation for increasing clarity By Blatko1 in gfx-rs/wgpu#4806 * #4809 [naga wgsl-in] Use a better span for errors in constructors. By jimblandy in gfx-rs/wgpu#4809 * #4808 [naga wgsl-in] Drop spanless labels from front-end error messages. By jimblandy in gfx-rs/wgpu#4808 * #4822 Fix expected error message By teoxoy in gfx-rs/wgpu#4822 * #4759 Add feature float32-filterable By almarklein in gfx-rs/wgpu#4759 * #4813 Bump core-graphics-types from 0.1.2 to 0.1.3 By dependabot[bot] in gfx-rs/wgpu#4813 Differential Revision: https://phabricator.services.mozilla.com/D195438 UltraBlame original commit: 985aabb5bf419ffcc9248010353ff190c225ba61

jimblandy requested a review from a team November 22, 2023 04:25

jimblandy mentioned this pull request Nov 22, 2023

Implement WGSL abstract types for global const declarations and constructor calls. #4743

Merged

4 tasks

teoxoy requested changes Nov 22, 2023

View reviewed changes

naga/src/front/wgsl/parse/number.rs Outdated Show resolved Hide resolved

jimblandy force-pushed the naga-wgsl-f64-literals branch from f0d92aa to daf7dee Compare November 22, 2023 18:09

jimblandy requested a review from teoxoy November 22, 2023 18:10

jimblandy force-pushed the naga-wgsl-f64-literals branch 3 times, most recently from d3bb991 to 98b0b71 Compare November 22, 2023 18:48

jimblandy added 2 commits November 22, 2023 13:24

[naga wgsl] Experimental 64-bit floating-point literals.

ea23916

In the WGSL front and back ends, support an `lf` suffix on floating-point literals to yield 64-bit integer literals.

[naga] Support casting to f64 in the constant evaluator.

e8e57d4

jimblandy force-pushed the naga-wgsl-f64-literals branch from 98b0b71 to e8e57d4 Compare November 22, 2023 21:25

teoxoy approved these changes Nov 23, 2023

View reviewed changes

teoxoy merged commit 86562e6 into gfx-rs:trunk Nov 23, 2023

jimblandy deleted the naga-wgsl-f64-literals branch December 13, 2023 16:31

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[naga wgsl-in] Experimental 64-bit floating-point literals.#4747

[naga wgsl-in] Experimental 64-bit floating-point literals.#4747
teoxoy merged 2 commits intogfx-rs:trunkfrom
jimblandy:naga-wgsl-f64-literals

jimblandy commented Nov 22, 2023 •

edited

Loading

Uh oh!

jimblandy commented Nov 22, 2023

Uh oh!

cwfitzgerald commented Nov 22, 2023

Uh oh!

teoxoy commented Nov 22, 2023

Uh oh!

Uh oh!

jimblandy commented Nov 22, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

jimblandy commented Nov 22, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jimblandy commented Nov 22, 2023

Uh oh!

cwfitzgerald commented Nov 22, 2023

Uh oh!

teoxoy commented Nov 22, 2023

Uh oh!

Uh oh!

jimblandy commented Nov 22, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

jimblandy commented Nov 22, 2023 •

edited

Loading