feat: Add missing std functions - #2925
Open
reczkok wants to merge 1 commit into
Open
Conversation
|
pkg.pr.new packages benchmark commit |
Bundle size comparison (
|
| 🟢 Decreased | ➖ Unchanged | 🔴 Increased (max 17.74%) | ❔ Unknown |
|---|---|---|---|
| 0 | 276 | 49 | 23 |
import * as ... in PR vs import * as ... in target (did bundle size increase?):
Click to reveal the results table (72 entries).
| Test | tsdown |
|---|---|
| std_textureBarrier.ts | 20.86 kB ( |
| std_atomicLoad.ts | 20.85 kB ( |
| std_atomicStore.ts | 20.85 kB ( |
| std_atomicAdd.ts | 20.86 kB ( |
| std_atomicAnd.ts | 20.86 kB ( |
| std_atomicMax.ts | 20.86 kB ( |
| std_atomicMin.ts | 20.86 kB ( |
| std_atomicOr.ts | 20.86 kB ( |
| std_atomicSub.ts | 20.86 kB ( |
| std_atomicXor.ts | 20.86 kB ( |
| std_storageBarrier.ts | 20.86 kB ( |
| std_workgroupBarrier.ts | 20.85 kB ( |
| std_pack2x16float.ts | 40.86 kB ( |
| std_pack4x8unorm.ts | 40.86 kB ( |
| std_unpack2x16float.ts | 40.86 kB ( |
| std_unpack4x8unorm.ts | 40.86 kB ( |
| STATIC_std.ts | 103.68 kB ( |
| std_textureDimensions.ts | 28.99 kB ( |
| std_textureGather.ts | 28.99 kB ( |
| std_textureLoad.ts | 28.99 kB ( |
| std_textureSample.ts | 28.99 kB ( |
| std_textureSampleBaseClampToEdge.ts | 28.99 kB ( |
| std_textureSampleBias.ts | 28.99 kB ( |
| std_textureSampleCompare.ts | 28.99 kB ( |
| std_textureSampleCompareLevel.ts | 28.99 kB ( |
| std_textureSampleGrad.ts | 28.99 kB ( |
| std_textureSampleLevel.ts | 28.99 kB ( |
| std_textureStore.ts | 28.99 kB ( |
| STATIC_allImports.ts | 305.05 kB ( |
| STATIC_tgpu.ts | 275.56 kB ( |
| tgpu_fn.ts | 275.56 kB ( |
| tgpu_init.ts | 275.56 kB ( |
| tgpu_lazy.ts | 275.56 kB ( |
| tgpu_slot.ts | 275.56 kB ( |
| tgpu_const.ts | 275.56 kB ( |
| tgpu_unroll.ts | 275.57 kB ( |
| tgpu_resolve.ts | 275.57 kB ( |
| tgpu_accessor.ts | 275.57 kB ( |
| tgpu_comptime.ts | 275.57 kB ( |
| tgpu_vertexFn.ts | 275.57 kB ( |
| tgpu_computeFn.ts | 275.57 kB ( |
| tgpu_fragmentFn.ts | 275.57 kB ( |
| tgpu_privateVar.ts | 275.57 kB ( |
| tgpu_vertexLayout.ts | 275.57 kB ( |
| tgpu_workgroupVar.ts | 275.57 kB ( |
| tgpu_initFromDevice.ts | 275.57 kB ( |
| tgpu_bindGroupLayout.ts | 275.57 kB ( |
| tgpu_mutableAccessor.ts | 275.57 kB ( |
| tgpu_resolveWithContext.ts | 275.58 kB ( |
| std_atomicCompareExchangeWeak.ts | 20.85 kB |
| std_atomicExchange.ts | 20.85 kB |
| std_pack2x16snorm.ts | 40.86 kB |
| std_pack2x16unorm.ts | 40.86 kB |
| std_pack4x8snorm.ts | 40.86 kB |
| std_pack4xI8.ts | 40.86 kB |
| std_pack4xI8Clamp.ts | 40.86 kB |
| std_pack4xU8.ts | 40.86 kB |
| std_pack4xU8Clamp.ts | 40.86 kB |
| std_quadBroadcast.ts | 16.56 kB |
| std_quadSwapDiagonal.ts | 16.57 kB |
| std_quadSwapX.ts | 16.57 kB |
| std_quadSwapY.ts | 16.57 kB |
| std_textureGatherCompare.ts | 28.99 kB |
| std_textureNumLayers.ts | 28.99 kB |
| std_textureNumLevels.ts | 28.99 kB |
| std_textureNumSamples.ts | 28.99 kB |
| std_unpack2x16snorm.ts | 40.86 kB |
| std_unpack2x16unorm.ts | 40.86 kB |
| std_unpack4x8snorm.ts | 40.86 kB |
| std_unpack4xI8.ts | 40.86 kB |
| std_unpack4xU8.ts | 40.86 kB |
| std_workgroupUniformLoad.ts | 20.85 kB |
import { ... } in PR vs import * as ... in PR (is the library tree-Shakeable?):
| Test | tsdown |
|---|---|
| tgpu_init.ts | 266.03 kB ( |
| tgpu_initFromDevice.ts | 265.49 kB ( |
| tgpu_resolve.ts | 161.75 kB ( |
| tgpu_resolveWithContext.ts | 161.68 kB ( |
| tgpu_bindGroupLayout.ts | 62.32 kB ( |
| tgpu_mutableAccessor.ts | 57.04 kB ( |
| tgpu_accessor.ts | 57.04 kB ( |
| tgpu_privateVar.ts | 55.73 kB ( |
| tgpu_workgroupVar.ts | 55.73 kB ( |
| tgpu_const.ts | 55.15 kB ( |
| tgpu_lazy.ts | 54.95 kB ( |
| tgpu_fragmentFn.ts | 39.68 kB ( |
| tgpu_fn.ts | 39.62 kB ( |
| tgpu_vertexFn.ts | 39.50 kB ( |
| tgpu_computeFn.ts | 39.20 kB ( |
| tgpu_vertexLayout.ts | 28.33 kB ( |
| tgpu_comptime.ts | 15.93 kB ( |
| tgpu_unroll.ts | 1.75 kB ( |
| tgpu_slot.ts | 1.70 kB ( |
If you wish to run a comparison for other, slower bundlers, run the 'Tree-shake test' from the GitHub Actions menu.
Resolution Time Benchmark---
config:
themeVariables:
xyChart:
plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
title "Random Branching (🔴 PR | 🔵 main | 🟢 release)"
x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
y-axis "time (ms)"
line [0.88, 1.77, 4.04, 5.27, 6.33, 11.13, 21.27, 21.82]
line [0.85, 1.79, 3.75, 5.59, 6.72, 11.76, 20.51, 21.20]
line [0.82, 1.65, 3.26, 5.94, 6.33, 9.31, 19.55, 22.05]
---
config:
themeVariables:
xyChart:
plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
title "Linear Recursion (🔴 PR | 🔵 main | 🟢 release)"
x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
y-axis "time (ms)"
line [0.32, 0.51, 0.64, 0.72, 1.02, 1.06, 1.26, 1.45]
line [0.26, 0.47, 0.61, 0.76, 1.18, 1.04, 1.27, 1.42]
line [0.31, 0.46, 0.56, 0.71, 1.00, 1.02, 1.16, 1.34]
---
config:
themeVariables:
xyChart:
plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
title "Full Tree (🔴 PR | 🔵 main | 🟢 release)"
x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
y-axis "time (ms)"
line [0.81, 2.08, 3.49, 6.23, 10.92, 23.36, 49.02, 100.29]
line [0.78, 1.84, 3.16, 6.58, 10.52, 23.09, 48.96, 99.39]
line [0.82, 1.83, 2.98, 5.52, 10.46, 23.09, 48.13, 96.34]
|
Contributor
There was a problem hiding this comment.
Pull request overview
This PR expands TypeGPU’s std surface area by adding several missing WGSL builtins (texture queries, gather-compare, quad ops, additional packing/unpacking, and atomic helpers) and introduces snapshot/unit tests to validate codegen and typing.
Changes:
- Add texture query builtins (
textureNumLayers/Levels/Samples) andtextureGatherCompare, with coverage tests. - Add quad subgroup builtins (
quadBroadcast,quadSwapDiagonal/X/Y) and atomic builtins (atomicExchange,atomicCompareExchangeWeak,workgroupUniformLoad) with tests. - Extend packing std with additional WGSL pack/unpack builtins and broaden packing test coverage.
Reviewed changes
Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| packages/typegpu/tests/std/texture/textureQueries.test.ts | Adds snapshot/type tests for texture query and gather-compare builtins. |
| packages/typegpu/tests/std/quad.test.ts | Adds snapshot/type tests for quad subgroup builtins (requires subgroups). |
| packages/typegpu/tests/std/packing.test.ts | Adds tests for new pack/unpack functions and rounding/clamping expectations. |
| packages/typegpu/tests/std/atomic.test.ts | Adds snapshot/type tests for new atomic and workgroupUniformLoad builtins. |
| packages/typegpu/src/std/texture.ts | Implements missing texture query builtins and textureGatherCompare codegen/signatures. |
| packages/typegpu/src/std/quad.ts | Implements quad subgroup builtins via dualImpl. |
| packages/typegpu/src/std/packing.ts | Refactors packing internals and adds multiple missing WGSL pack/unpack builtins. |
| packages/typegpu/src/std/index.ts | Re-exports the newly added std functions. |
| packages/typegpu/src/std/atomic.ts | Adds atomicExchange, atomicCompareExchangeWeak, and workgroupUniformLoad. |
Suppressed comments (3)
packages/typegpu/src/std/packing.ts:89
pack2x16unormusesMath.floor(0.5 + x)rounding, which disagrees with the project’s WGSLround(ties-to-even) semantics (seecpuRoundinstd/numeric.ts). This can produce different packed results for values that land exactly on.5after scaling.
normalImpl: (e: v2f): number => {
const x = Math.floor(0.5 + 65535 * clampScalar(e.x, 0, 1));
const y = Math.floor(0.5 + 65535 * clampScalar(e.y, 0, 1));
return writePacked((writer) => {
writer.writeUint16(x);
packages/typegpu/src/std/packing.ts:231
pack4x8snormusesMath.floor(0.5 + x)rounding, which doesn’t implement WGSLround(ties-to-even) semantics used elsewhere in the codebase (cpuRoundinstd/numeric.ts). This is particularly visible for negative.5cases (e.g. -63.5), where the current logic rounds toward zero instead of to the even integer.
export const pack4x8snorm = dualImpl({
name: 'pack4x8snorm',
normalImpl: (e: v4f): number =>
writePacked((writer) => {
writer.writeInt8(Math.floor(0.5 + 127 * clampScalar(e.x, -1, 1)));
writer.writeInt8(Math.floor(0.5 + 127 * clampScalar(e.y, -1, 1)));
writer.writeInt8(Math.floor(0.5 + 127 * clampScalar(e.z, -1, 1)));
writer.writeInt8(Math.floor(0.5 + 127 * clampScalar(e.w, -1, 1)));
}),
packages/typegpu/src/std/packing.ts:212
pack4x8unormcurrently rounds withMath.floor(0.5 + x), which isn’t ties-to-even and can disagree with WGSLroundsemantics used elsewhere in the codebase (cpuRoundinstd/numeric.ts). This can yield different results for values that scale toN + 0.5whereNis even/odd (e.g. 2.5).
export const pack4x8unorm = dualImpl({
name: 'pack4x8unorm',
normalImpl: (e: v4f): number =>
writePacked((writer) => {
writer.writeUint8(Math.floor(0.5 + 255 * clampScalar(e.x, 0, 1)));
writer.writeUint8(Math.floor(0.5 + 255 * clampScalar(e.y, 0, 1)));
writer.writeUint8(Math.floor(0.5 + 255 * clampScalar(e.z, 0, 1)));
writer.writeUint8(Math.floor(0.5 + 255 * clampScalar(e.w, 0, 1)));
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Comment on lines
66
to
+70
| normalImpl: (e: v2f): number => { | ||
| const buffer = new ArrayBuffer(4); | ||
| const writer = new TB.BufferWriter(buffer); | ||
| writeFloat16(writer, e.x); | ||
| writeFloat16(writer, e.y); | ||
| const reader = new TB.BufferReader(buffer); | ||
| return u32(reader.readUint32()); | ||
| const x = Math.floor(0.5 + 32767 * clampScalar(e.x, -1, 1)); | ||
| const y = Math.floor(0.5 + 32767 * clampScalar(e.y, -1, 1)); | ||
| return writePacked((writer) => { | ||
| writer.writeInt16(x); |
![pullfrog[bot]](https://avatars.githubusercontent.com/in/1768019?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
- Atomics (
std/atomic.ts) — addsworkgroupUniformLoad,atomicExchange, andatomicCompareExchangeWeak(returning anabstructresult withold_value/exchanged), all consistent with the existingatomicOpSignature/isAtomic/atomicNormalErrorhelpers. - Packing (
std/packing.ts) — replaces per-callArrayBufferchurn with sharedpackedBuffer/packedWriter/packedReaderhelpers (explicitlittleendianness) and addspack/unpack2x16(snorm|unorm),pack/unpack4x8(snorm|unorm),pack/unpack4xI8,pack/unpack4xU8, and the twoClampvariants. - Quad (
std/quad.ts, new) —quadBroadcast(2nd arg validated asi32/u32viaunify),quadSwapDiagonal,quadSwapX,quadSwapY. - Texture (
std/texture.ts) —textureNumLayers/textureNumLevels/textureNumSampleswith typed texture-category unions, andtextureGatherCompare(array layer index coerced to[u32, i32], mirroring the existingtextureGathersignature). - Exports (
std/index.ts) — all new builtins exported. - Tests — inline WGSL snapshots plus
if (false)expectTypeOfassertions for atomics, quad, packing, and texture queries; all 9 pass.
I hand-verified the pack arithmetic (little-endian byte order, snorm -1 → -max symmetric mapping via coefficient 32767/127, clamp semantics, and writeInt8/writeUint8 wraparound for the non-clamp variants) against the WGSL spec and the expected values in the tests — all consistent, and the GPU codegenImpl emits the corresponding WGSL builtins.
ℹ️ Piggy-back notes for the author (nothing blocking)
- Forcing
littleendianness in the shared helpers is actually a correctness fix: the pre-existingpack2x16float/unpack*usednew TB.BufferWriter(buffer)with the system default endianness, which diverges from WGSL's little-endian packing on a big-endian host. No-op on the (effectively universal) little-endian case. - The CPU
pack*rounding usesMath.round/floor(0.5 + x)(round-half-up), which can differ from WGSL's round-to-nearest-ties-to-even at exact.5boundary values — so CPU and GPU results can disagree on that narrow input set. This matches an existing repo-wide known convention (jsPolyfills.tsalready notes the JS/WGSLrounddivergence), so it's fine to leave as-is.
ℹ️ Nitpicks
std/quad.tsredeclaresIdentityNumOrVec, which already exists instd/subgroup.ts— harmless duplication, but worth hoisting to a shared location if more quad-mirroring subgroup APIs land later.
DeepSeek Flash (free via Pullfrog for OSS) ｜ 𝕏
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.