Activity

1. github-actions

   added

   Office HoursIssue announcing office hours and soliciting associated agenda items.Issue announcing office hours and soliciting associated agenda items.

   on Jun 2, 2026
2. manit2004 commented on Jun 9, 2026

   

   manit2004

   on Jun 9, 2026

   More actions

   Request to review Wald MGF #12399
3. DivitJain26 commented on Jun 9, 2026

   

   DivitJain26

   on Jun 9, 2026

   Last edited by DivitJain26

   Member

   More actions

   1. Why do we not have symmetric complex-valued routines like cspmv in our BLAS implementations?
   2. Please provide some guidance on feat: add array/uint64 stdlib#12562.
   3. How to convert Unit8Array, Uint8ClampedArray, Unit16Array buffer to Unit32Array buffer?
4. batpigandme commented on Jun 9, 2026

   

   batpigandme

   on Jun 9, 2026

   Member

   More actions

   AI-Generated Summary

   Caution

   These notes are likely imperfect, especially when it comes to spelling/function/package names

   Meeting Purpose

   Review project progress and guide new development.

   Key Takeaways

   • Uint64 Array Guidance: The Uint64Array constructor should use the Uint64 utility to handle numbers/BigInts, splitting them into high/low words for a Uint32Array buffer. The sole exception is reinterpreting an existing Uint64Array by directly copying its buffer.
   • WebGPU PoC: A proof-of-concept for ASCAL (scalar multiplication) wraps WebGPU's verbose API with CUDA-like helpers. While compute-only benchmarks are comparable to CUDA, the current design's CPU-GPU transfers are a major bottleneck.
   • YouTube Channel Launch: A new YouTube channel will launch with screencasts of feature development and agentic coding to onboard new contributors and showcase maintainer workflows.
   • BLAS Routines: Symmetric complex-valued BLAS routines (e.g., CSPMV) are missing from the master tracking issue. Divit will add them to ensure they are implemented.

   Topics

   BLAS Routines

   • Problem: Symmetric complex-valued BLAS routines are missing from the master tracking issue.
   • Rationale: No technical reason; they were likely overlooked.
   • Action: Divit will add the missing Level 2 routines to the issue for future implementation.

   Uint64 Array Implementation

   • Guidance: The Uint64Array constructor should follow the pattern of ComplexNumberArray and BoolArray.
   • Core Logic:
     • Input Handling: Treat all inputs (numbers, BigInts, typed arrays) as generic collections to be cast.
     • Casting: Use the Uint64 utility to convert each collection element. This handles BigInts and correctly splits values into high and low words.
     • Storage: Store the high and low words in a Uint32Array buffer.
     • Byte Order: Use the indices utility (which leverages assertIsLittleEndian) to determine the correct storage order for the host platform.
   • Sole Exception:
     • Scenario: Input is an existing Uint64Array.
     • Action: Reinterpret its buffer as a Uint32Array and copy it directly. This avoids the casting overhead because the words are already in the correct platform byte order.

   WebGPU Proof-of-Concept

   • Goal: Simplify WebGPU's verbose API with CUDA-like helper functions.
   • Architecture:
     • Shaders: WGSL code for GPU execution.
       • Limitation: No import support → requires duplicating logic (e.g., for custom Float64 arithmetic).
     • Utils: Helper functions abstracting WebGPU boilerplate (e.g., loadShader, uploadBuffer, readbackBuffer).
     • Routines: Simple wrappers for specific operations (e.g., ASCAL).
   • Benchmarking (ASCAL):
     • Setup: Compares WebGPU (via node-webgpu package, which uses DAWN) vs. CUDA.
     • Finding: Compute-only times are comparable.
     • Bottleneck: WebGPU's CPU-GPU transfer overhead is the primary performance limiter.
   • Feedback:
     • API Inspiration: Review three.js and TensorFlow.js for their WebGPU abstractions.
     • Performance Strategy: Adopt a "stay on GPU" model (upload once, chain operations, download results only when needed) to minimize transfer overhead.

   YouTube Channel

   • Goal: Onboard new contributors and showcase maintainer workflows.
   • Content:
     • Screencasts of feature development and agentic coding.
     • Potential future content: live PR reviews.
   • Status: Raw video material is being polished for publication.

   Full transcript

   Full office hours transcript
5. batpigandme

   closed this as completedon Jun 9, 2026