feat: bitcastF32toU32#2517
Conversation
|
pkg.pr.new packages benchmark commit |
Bundle size comparison (
|
| 🟢 Decreased | ➖ Unchanged | 🔴 Increased | ❔ Unknown |
|---|---|---|---|
| 0 | 207 | 113 | 1 |
import * as ... in PR vs import * as ... in target (did bundle size increase?):
| Test | tsdown |
|---|---|
| std_bitcastU32toF32.ts | 45.09 kB ( |
| std_bitcastU32toI32.ts | 45.09 kB ( |
| STATIC_std.ts | 106.77 kB ( |
| std_bitcastF32toU32.ts | 45.09 kB |
import { ... } in PR vs import * as ... in PR (is the library tree-Shakeable?):
Click to reveal the results table (298 entries).
| Test | tsdown |
|---|---|
| d_Void.ts | 87.49 kB ( |
| d_sampler.ts | 87.49 kB ( |
| d_isPtr.ts | 87.49 kB ( |
| d_comparisonSampler.ts | 87.50 kB ( |
| d_isSizeAttrib.ts | 87.50 kB ( |
| d_isWgslArray.ts | 87.49 kB ( |
| d_isAlignAttrib.ts | 87.50 kB ( |
| d_isWgslStruct.ts | 87.50 kB ( |
| d_isAtomic.ts | 87.49 kB ( |
| d_isBuiltinAttrib.ts | 87.50 kB ( |
| d_isLocationAttrib.ts | 87.50 kB ( |
| d_isDecorated.ts | 87.49 kB ( |
| d_isInterpolateAttrib.ts | 87.50 kB ( |
| d_atomic.ts | 87.49 kB ( |
| d_ptrHandle.ts | 87.49 kB ( |
| d_ptrUniform.ts | 87.49 kB ( |
| d_ptrStorage.ts | 87.49 kB ( |
| d_ptrPrivate.ts | 87.49 kB ( |
| d_ptrFn.ts | 87.49 kB ( |
| d_ptrWorkgroup.ts | 87.50 kB ( |
| d_textureExternal.ts | 87.50 kB ( |
| d_textureStorage1d.ts | 87.50 kB ( |
| d_textureStorage2d.ts | 87.50 kB ( |
| d_textureStorage3d.ts | 87.50 kB ( |
| d_textureStorage2dArray.ts | 87.50 kB ( |
| d_isDisarray.ts | 87.49 kB ( |
| d_isUnstruct.ts | 87.49 kB ( |
| d_isLooseDecorated.ts | 87.50 kB ( |
| d_isLooseData.ts | 87.49 kB ( |
| d_isWgslData.ts | 87.49 kB ( |
| d_unstruct.ts | 87.49 kB ( |
| d_isData.ts | 87.49 kB ( |
| d_deepEqual.ts | 87.49 kB ( |
| d_struct.ts | 87.49 kB ( |
| d_ref.ts | 87.49 kB ( |
| std_discard.ts | 103.92 kB ( |
| std_copy.ts | 103.92 kB ( |
| std_arrayLength.ts | 103.92 kB ( |
| std_isBeingTranspiled.ts | 103.93 kB ( |
| std_getTargetShaderLanguage.ts | 103.94 kB ( |
| std_extensionEnabled.ts | 103.93 kB ( |
| std_range.ts | 103.92 kB ( |
| std_dpdx.ts | 103.92 kB ( |
| std_fwidthCoarse.ts | 103.92 kB ( |
| std_dpdxCoarse.ts | 103.92 kB ( |
| std_dpdyCoarse.ts | 103.92 kB ( |
| std_fwidthFine.ts | 103.92 kB ( |
| std_dpdxFine.ts | 103.92 kB ( |
| std_dpdyFine.ts | 103.92 kB ( |
| std_fwidth.ts | 103.92 kB ( |
| std_dpdy.ts | 103.92 kB ( |
| d_bool.ts | 87.49 kB ( |
| d_f16.ts | 87.49 kB ( |
| d_f32.ts | 87.49 kB ( |
| d_i32.ts | 87.49 kB ( |
| d_u32.ts | 87.49 kB ( |
| d_u16.ts | 87.49 kB ( |
| std_atomicLoad.ts | 103.92 kB ( |
| std_textureBarrier.ts | 103.93 kB ( |
| std_atomicStore.ts | 103.92 kB ( |
| std_workgroupBarrier.ts | 103.93 kB ( |
| std_storageBarrier.ts | 103.93 kB ( |
| std_atomicAdd.ts | 103.92 kB ( |
| std_atomicAnd.ts | 103.92 kB ( |
| std_atomicMax.ts | 103.92 kB ( |
| std_atomicMin.ts | 103.92 kB ( |
| std_atomicSub.ts | 103.92 kB ( |
| std_atomicXor.ts | 103.92 kB ( |
| std_atomicOr.ts | 103.92 kB ( |
| d_textureDepth2d.ts | 87.50 kB ( |
| d_textureDepthCube.ts | 87.50 kB ( |
| d_texture1d.ts | 87.49 kB ( |
| d_texture2d.ts | 87.49 kB ( |
| d_texture3d.ts | 87.49 kB ( |
| d_textureDepth2dArray.ts | 87.50 kB ( |
| d_textureCube.ts | 87.49 kB ( |
| d_textureDepthCubeArray.ts | 87.50 kB ( |
| d_textureDepthMultisampled2d.ts | 87.51 kB ( |
| d_texture2dArray.ts | 87.50 kB ( |
| d_textureCubeArray.ts | 87.50 kB ( |
| d_textureMultisampled2d.ts | 87.50 kB ( |
| d_disarrayOf.ts | 87.49 kB ( |
| d_vec2b.ts | 87.49 kB ( |
| d_vec2f.ts | 87.49 kB ( |
| d_vec2h.ts | 87.49 kB ( |
| d_vec2i.ts | 87.49 kB ( |
| d_vec2u.ts | 87.49 kB ( |
| d_vec3b.ts | 87.49 kB ( |
| d_vec3f.ts | 87.49 kB ( |
| d_vec3h.ts | 87.49 kB ( |
| d_vec3i.ts | 87.49 kB ( |
| d_vec3u.ts | 87.49 kB ( |
| d_vec4b.ts | 87.49 kB ( |
| d_vec4f.ts | 87.49 kB ( |
| d_vec4h.ts | 87.49 kB ( |
| d_vec4i.ts | 87.49 kB ( |
| d_vec4u.ts | 87.49 kB ( |
| std_subgroupAdd.ts | 103.92 kB ( |
| std_subgroupBroadcastFirst.ts | 103.93 kB ( |
| std_subgroupExclusiveAdd.ts | 103.93 kB ( |
| std_subgroupExclusiveMul.ts | 103.93 kB ( |
| std_subgroupInclusiveAdd.ts | 103.93 kB ( |
| std_subgroupInclusiveMul.ts | 103.93 kB ( |
| std_subgroupShuffleDown.ts | 103.93 kB ( |
| std_subgroupShuffleXor.ts | 103.93 kB ( |
| std_subgroupBroadcast.ts | 103.93 kB ( |
| std_subgroupShuffleUp.ts | 103.93 kB ( |
| std_subgroupShuffle.ts | 103.93 kB ( |
| std_subgroupBallot.ts | 103.93 kB ( |
| std_subgroupElect.ts | 103.93 kB ( |
| std_subgroupAll.ts | 103.92 kB ( |
| std_subgroupAnd.ts | 103.92 kB ( |
| std_subgroupAny.ts | 103.92 kB ( |
| std_subgroupMax.ts | 103.92 kB ( |
| std_subgroupMin.ts | 103.92 kB ( |
| std_subgroupMul.ts | 103.92 kB ( |
| std_subgroupXor.ts | 103.92 kB ( |
| std_subgroupOr.ts | 103.92 kB ( |
| d_formatToWGSLType.ts | 87.50 kB ( |
| d_uint8.ts | 87.49 kB ( |
| d_unorm10_10_10_2.ts | 87.50 kB ( |
| d_unorm8x4_bgra.ts | 87.50 kB ( |
| d_float16x2.ts | 87.49 kB ( |
| d_float16x4.ts | 87.49 kB ( |
| d_float32x2.ts | 87.49 kB ( |
| d_float32x3.ts | 87.49 kB ( |
| d_float32x4.ts | 87.49 kB ( |
| d_snorm16x2.ts | 87.49 kB ( |
| d_snorm16x4.ts | 87.49 kB ( |
| d_unorm16x2.ts | 87.49 kB ( |
| d_unorm16x4.ts | 87.49 kB ( |
| d_sint16x2.ts | 87.49 kB ( |
| d_sint16x4.ts | 87.49 kB ( |
| d_sint32x2.ts | 87.49 kB ( |
| d_sint32x3.ts | 87.49 kB ( |
| d_sint32x4.ts | 87.49 kB ( |
| d_snorm8x2.ts | 87.49 kB ( |
| d_snorm8x4.ts | 87.49 kB ( |
| d_uint16x2.ts | 87.49 kB ( |
| d_uint16x4.ts | 87.49 kB ( |
| d_uint32x2.ts | 87.49 kB ( |
| d_uint32x3.ts | 87.49 kB ( |
| d_uint32x4.ts | 87.49 kB ( |
| d_unorm8x2.ts | 87.49 kB ( |
| d_unorm8x4.ts | 87.49 kB ( |
| d_float16.ts | 87.49 kB ( |
| d_float32.ts | 87.49 kB ( |
| d_sint8x2.ts | 87.49 kB ( |
| d_sint8x4.ts | 87.49 kB ( |
| d_snorm16.ts | 87.49 kB ( |
| d_uint8x2.ts | 87.49 kB ( |
| d_uint8x4.ts | 87.49 kB ( |
| d_unorm16.ts | 87.49 kB ( |
| d_sint16.ts | 87.49 kB ( |
| d_sint32.ts | 87.49 kB ( |
| d_snorm8.ts | 87.49 kB ( |
| d_uint16.ts | 87.49 kB ( |
| d_uint32.ts | 87.49 kB ( |
| d_unorm8.ts | 87.49 kB ( |
| d_sint8.ts | 87.49 kB ( |
| d_packedFormats.ts | 87.50 kB ( |
| d_isPackedData.ts | 87.50 kB ( |
| std_textureDimensions.ts | 103.93 kB ( |
| std_textureGather.ts | 103.93 kB ( |
| std_textureSampleBaseClampToEdge.ts | 103.94 kB ( |
| std_textureSampleCompareLevel.ts | 103.94 kB ( |
| std_textureSampleCompare.ts | 103.93 kB ( |
| std_textureSampleLevel.ts | 103.93 kB ( |
| std_textureSampleBias.ts | 103.93 kB ( |
| std_textureSampleGrad.ts | 103.93 kB ( |
| std_textureSample.ts | 103.93 kB ( |
| std_textureStore.ts | 103.92 kB ( |
| std_textureLoad.ts | 103.92 kB ( |
| d_alignmentOf.ts | 87.49 kB ( |
| std_translation4.ts | 103.92 kB ( |
| std_rotationX4.ts | 103.92 kB ( |
| std_rotationY4.ts | 103.92 kB ( |
| std_rotationZ4.ts | 103.92 kB ( |
| std_identity2.ts | 103.92 kB ( |
| std_identity3.ts | 103.92 kB ( |
| std_identity4.ts | 103.92 kB ( |
| std_scaling4.ts | 103.92 kB ( |
| d_isBuiltin.ts | 87.49 kB ( |
| d_sizeOf.ts | 87.49 kB ( |
| d_isContiguous.ts | 87.50 kB ( |
| d_getLongestContiguousPrefix.ts | 87.51 kB ( |
| d_arrayOf.ts | 87.49 kB ( |
| d_size.ts | 87.49 kB ( |
| d_align.ts | 87.49 kB ( |
| d_location.ts | 87.49 kB ( |
| d_interpolate.ts | 87.49 kB ( |
| d_invariant.ts | 87.49 kB ( |
| d_mat4x4f.ts | 87.49 kB ( |
| d_mat2x2f.ts | 87.49 kB ( |
| d_mat3x3f.ts | 87.49 kB ( |
| d_matToArray.ts | 87.49 kB ( |
| d_builtin.ts | 87.49 kB ( |
| std_unpack2x16float.ts | 103.93 kB ( |
| std_unpack4x8unorm.ts | 103.93 kB ( |
| std_pack2x16float.ts | 103.93 kB ( |
| std_pack4x8unorm.ts | 103.92 kB ( |
| std_bitcastF32toU32.ts | 103.93 kB ( |
| std_bitcastU32toF32.ts | 103.93 kB ( |
| std_bitcastU32toI32.ts | 103.93 kB ( |
| std_neg.ts | 103.92 kB ( |
| std_bitShiftRight.ts | 103.93 kB ( |
| std_bitShiftLeft.ts | 103.92 kB ( |
| std_add.ts | 103.92 kB ( |
| std_div.ts | 103.92 kB ( |
| std_mod.ts | 103.92 kB ( |
| std_mul.ts | 103.92 kB ( |
| std_sub.ts | 103.92 kB ( |
| std_translate4.ts | 103.92 kB ( |
| std_rotateX4.ts | 103.92 kB ( |
| std_scale4.ts | 103.92 kB ( |
| std_rotateY4.ts | 103.92 kB ( |
| std_rotateZ4.ts | 103.92 kB ( |
| d_memoryLayoutOf.ts | 87.50 kB ( |
| std_and.ts | 103.92 kB ( |
| std_or.ts | 103.92 kB ( |
| std_select.ts | 103.92 kB ( |
| std_allEq.ts | 103.92 kB ( |
| std_all.ts | 103.92 kB ( |
| std_not.ts | 103.92 kB ( |
| std_eq.ts | 103.92 kB ( |
| std_lt.ts | 103.92 kB ( |
| std_ne.ts | 103.92 kB ( |
| std_isCloseTo.ts | 103.92 kB ( |
| std_any.ts | 103.92 kB ( |
| std_ge.ts | 103.92 kB ( |
| std_gt.ts | 103.92 kB ( |
| std_le.ts | 103.92 kB ( |
| std_dot4U8Packed.ts | 103.92 kB ( |
| std_length.ts | 103.92 kB ( |
| std_countTrailingZeros.ts | 103.93 kB ( |
| std_distance.ts | 103.92 kB ( |
| std_dot.ts | 103.92 kB ( |
| std_countLeadingZeros.ts | 103.93 kB ( |
| std_firstTrailingBit.ts | 103.93 kB ( |
| std_firstLeadingBit.ts | 103.93 kB ( |
| std_fract.ts | 103.92 kB ( |
| std_frexp.ts | 103.92 kB ( |
| std_modf.ts | 103.92 kB ( |
| std_quantizeToF16.ts | 103.93 kB ( |
| std_mix.ts | 103.92 kB ( |
| std_countOneBits.ts | 103.92 kB ( |
| std_dot4I8Packed.ts | 103.92 kB ( |
| std_determinant.ts | 103.92 kB ( |
| std_extractBits.ts | 103.92 kB ( |
| std_faceForward.ts | 103.92 kB ( |
| std_inverseSqrt.ts | 103.92 kB ( |
| std_reverseBits.ts | 103.92 kB ( |
| std_insertBits.ts | 103.92 kB ( |
| std_smoothstep.ts | 103.92 kB ( |
| std_normalize.ts | 103.92 kB ( |
| std_transpose.ts | 103.92 kB ( |
| std_saturate.ts | 103.92 kB ( |
| std_degrees.ts | 103.92 kB ( |
| std_radians.ts | 103.92 kB ( |
| std_reflect.ts | 103.92 kB ( |
| std_refract.ts | 103.92 kB ( |
| std_acosh.ts | 103.92 kB ( |
| std_asinh.ts | 103.92 kB ( |
| std_atan2.ts | 103.92 kB ( |
| std_atanh.ts | 103.92 kB ( |
| std_clamp.ts | 103.92 kB ( |
| std_cross.ts | 103.92 kB ( |
| std_floor.ts | 103.92 kB ( |
| std_ldexp.ts | 103.92 kB ( |
| std_round.ts | 103.92 kB ( |
| std_trunc.ts | 103.92 kB ( |
| std_acos.ts | 103.92 kB ( |
| std_asin.ts | 103.92 kB ( |
| std_atan.ts | 103.92 kB ( |
| std_ceil.ts | 103.92 kB ( |
| std_cosh.ts | 103.92 kB ( |
| std_exp2.ts | 103.92 kB ( |
| std_log2.ts | 103.92 kB ( |
| std_sign.ts | 103.92 kB ( |
| std_sinh.ts | 103.92 kB ( |
| std_sqrt.ts | 103.92 kB ( |
| std_step.ts | 103.92 kB ( |
| std_tanh.ts | 103.92 kB ( |
| std_abs.ts | 103.92 kB ( |
| std_cos.ts | 103.92 kB ( |
| std_exp.ts | 103.92 kB ( |
| std_fma.ts | 103.92 kB ( |
| std_log.ts | 103.92 kB ( |
| std_max.ts | 103.92 kB ( |
| std_min.ts | 103.92 kB ( |
| std_pow.ts | 103.92 kB ( |
| std_sin.ts | 103.92 kB ( |
| std_tan.ts | 103.92 kB ( |
| common_fullScreenTriangle.ts | 56.66 kB ( |
| common_writeSoA.ts | 56.65 kB ( |
| STATIC_allImports.ts | 287.20 kB ( |
| STATIC_d.ts | 87.48 kB ( |
| STATIC_std.ts | 103.91 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 [1.00, 1.98, 4.47, 6.60, 7.30, 11.64, 22.60, 24.74]
line [0.95, 1.99, 4.34, 6.32, 8.32, 11.65, 22.86, 25.14]
line [0.92, 1.93, 4.40, 5.90, 7.22, 10.44, 21.05, 23.99]
---
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 [0.29, 0.50, 0.69, 0.89, 1.16, 1.28, 1.44, 1.69]
line [0.27, 0.52, 0.70, 0.84, 1.16, 1.23, 1.47, 1.58]
line [0.27, 0.52, 0.73, 0.84, 1.13, 1.23, 1.49, 1.59]
---
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 [0.86, 2.19, 4.44, 6.50, 12.91, 26.10, 57.02, 114.79]
line [0.86, 2.10, 4.22, 6.73, 13.10, 27.44, 55.22, 115.99]
line [0.91, 2.06, 3.74, 6.33, 11.92, 25.10, 54.73, 111.06]
|
There was a problem hiding this comment.
Pull request overview
Adds a new std helper bitcastF32toU32 (CPU + WGSL codegen) to reinterpret f32 bit patterns as u32, including vector support, and extends the test suite to cover the new behavior.
Changes:
- Implement
std.bitcastF32toU32withdualImplsupport for scalars andvec2f/vec3f/vec4f. - Add CPU bitcast implementation (
bitcastF32toU32Impl) and vector-component CPU helpers. - Extend
bitcasttests to cover scalar/vector behavior and shader codegen snapshots.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| packages/typegpu/tests/std/bitcast.test.ts | Adds scalar/vector tests and shader snapshots for bitcastF32toU32 (currently introduces a Node Buffer typing issue). |
| packages/typegpu/src/std/index.ts | Exports bitcastF32toU32 from the std barrel. |
| packages/typegpu/src/std/bitcast.ts | Adds the new dualImpl wrapper for bitcastF32toU32 (signature currently doesn’t reject unsupported input types). |
| packages/typegpu/src/data/vectorOps.ts | Adds CPU vector implementations for bitcastF32toU32. |
| packages/typegpu/src/data/numberOps.ts | Adds the scalar CPU bitcast implementation for f32 -> u32. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
iwoplaza
left a comment
There was a problem hiding this comment.
I would add a few more dataType checks like the AI said, but other than that, you have my stamp of approval ![]()
|
@pullfrog review this PR |
There was a problem hiding this comment.
ℹ️ Minor suggestions inline.
Reviewed changes — adds bitcastF32toU32, refactors the existing bitcast functions to use stricter type unification, and optimizes the shared ArrayBuffer in the CPU implementations.
- Add
bitcastF32toU32— newdualImplfunction mirroringbitcastU32toF32andbitcastU32toI32, with runtime impl, shader codegen, andVectorOpssupport for all float→uint combinations. - Add
unifyStricttoconversion.ts— stricterunifythat rejects unknown types (viafindBestTypewithrecast: false) instead of silently passing them through. - Refactor existing bitcast functions —
bitcastU32toF32andbitcastU32toI32now useunifyStrict+SignatureNotSupportedErrorinstead of the lenientunifyfallback pattern. Overload types converted from intersection types to conditional generics. - Optimize CPU bitcast impls — shared module-level
ArrayBufferwith typed array views replaces per-callnew DataView(new ArrayBuffer(4)). - Tests — scalar, vector, edge-case (NaN/infinities/subnormals/±0), shader codegen, and error-path coverage for the new function.
DeepSeek Pro (free via Pullfrog for OSS) | 𝕏
There was a problem hiding this comment.
ℹ️ No critical issues — minor suggestion inline.
Reviewed changes — addresses the prior review's codegen concern and adds union-type tests.
- Fix
bitcastU32toF32andbitcastU32toI32shader codegen —codegenImplnow emitsbitcast<vecNf>/bitcast<vecNi>for vector arguments, matching the pattern already used inbitcastF32toU32. - Add union-type and type-error tests —
expectTypeOfverifies correct return-type narrowing for union arguments; error-path tests coverSignatureNotSupportedErrorrejection of invalid types.
DeepSeek Pro (free via Pullfrog for OSS) | 𝕏

Let's wait for #2615.
TODO:
sideEffect: false