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 Aug 11, 2026
2. DivitJain26 commented on Aug 18, 2026

   

   DivitJain26

   on Aug 18, 2026

   Member

   More actions

   • Guidance on feat: add strided/base/reinterpret-uint64 stdlib#14219

   Should the byte order of the returned Uint32Array view be:

   • Host-OS dependent: view[0] is whichever word (high or low) comes first in memory on the current platform (little-endian vs big-endian),
     or
   • Definitive: view[0] is always documented as HIGH (or always LOW)?

   This affects how examples in the documentation should be written.

   var Uint64Array = require( '@stdlib/array/uint64' );
   var reinterpret = require( './../lib' );
   
   // Define a 64-bit unsigned integer array:
   var x = new Uint64Array( [ 1, 2, 3, 4, 5, 6, 7, 8 ] );
   // returns <Uint64Array>
   
   // Reinterpret as a `uint32` array:
   var view = reinterpret( x, 0 );
   // returns <Uint32Array>
   
   // Set view elements:
   view[ 0 ] = 0;
   view[ 1 ] = 0;
   
   // Get the first element of the 64-bit unsigned integer array:
   var u = x.get( 0 );
   // returns <Uint64>[ 0n ]
   
   console.log( u.toString() );

   This particular example is endianness-safe; setting both words to 0 zeroes the element no matter which index is high vs. low. But it doesn't actually demonstrate the indexing convention, and any example that sets a specific non-zero value (e.g., "set only the high word") would need to know which index means what.
3. 0PrashantYadav0 commented on Aug 18, 2026

   

   0PrashantYadav0

   on Aug 18, 2026

   Member

   More actions

   Discussion on

   • @stdlib/blas/ext/base/nannsum/results/struct-factory - #14368 : dstruct schema: { sum, count }
   • @stdlib/blas/ext/base/nannsum/results/factory - #14368 : Results constructor factory (float64 | float32)
   • @stdlib/blas/ext/base/nannsum/results/float32 - #14348 : Float32Results JS constructor

   Initial inspiration :

   • ml/base/kmeans/results/struct-factory - #12356
   • @stdlib/stats/strided/dztest — stdlib_stats_ztest_one_sample_float64_results
   • @stdlib/stats/strided/dztest2 — stdlib_stats_ztest_two_sample_float64_results
   • @stdlib/stats/strided/sztest — stdlib_stats_ztest_one_sample_float32_results
   • @stdlib/stats/strided/sztest2 — stdlib_stats_ztest_two_sample_float32_results

   Project Idea:

   nannsum/results/
   ├── struct-factory/   - Layer 1: raw binary struct schema (dtype-parameterised)
   ├── factory/          - Layer 2: user-facing constructor factory (validates dtype, adds API)
   ├── float32/          - Layer 3a: pre-baked float32 constructor (shortcut for factory('float32'))
   ├── float64/          - Layer 3b: pre-baked float64 constructor (shortcut for factory('float64'))
   ├── to-json/          - Utility: serialize a results object to a plain JSON object
   └── to-string/        - Utility: serialize a results object to a human-readable string

   Example for struct-factory:

   var structFactory = require( '@stdlib/blas/ext/base/nannsum/results/struct-factory' );
   var BigInt = require( '@stdlib/bigint/ctor' );
   
   var Struct = structFactory( 'float64' );
   var results = new Struct({
       'sum': 11.7586,
       'count': BigInt( 5 )
   });
   
   var str = results.toString({
       'format': 'linear'
   });
   console.log( str );

   Example for factory:

   var BigInt = require( '@stdlib/bigint/ctor' );
   var resultsFactory = require( '@stdlib/blas/ext/base/nannsum/results/factory' );
   
   var Results = resultsFactory( 'float64' );
   var results = new Results({
       'sum': 11.7586,
       'count': BigInt( 5 )
   });
   
   var str = results.toString();
   console.log( str );

   Example for float32:

   var BigInt = require( '@stdlib/bigint/ctor' );
   var f32 = require( '@stdlib/number/float64/base/to-float32' );
   var Results = require( '@stdlib/blas/ext/base/nannsum/results/float32' );
   
   var results = new Results({
       'sum': f32( 11.7586 ),
       'count': BigInt( 5 )
   });
   
   var str = results.toString();
   console.log( str );
4. kgryte commented on Aug 18, 2026

   

   kgryte

   on Aug 18, 2026

   Member

   More actions

   var out = new Float64Array( 1 );
   var count = new Int64Array( 1 );
   
   dnannsum( N, x, strideX, offsetX, out, offsetOut, count, offsetCount );
5. kgryte commented on Aug 18, 2026

   

   kgryte

   on Aug 18, 2026

   Member

   More actions

   var Uint64Array = require( '@stdlib/array/uint64' );
   var reinterpret = require( './../lib' );
   
   // Define a 64-bit unsigned integer array:
   var x = new Uint64Array( [ 1, 2, 3, 4, 5, 6, 7, 8 ] );
   // returns <Uint64Array>
   
   // Reinterpret as a `uint32` array:
   var view = reinterpret( x, 0 );
   // returns <Uint32Array>
   
   // Set view elements:
   view[ 0 ] = 0;
   view[ 1 ] = 0;
   
   // Get the first element of the 64-bit unsigned integer array:
   var u = x.get( 0 );
   // e.g., returns <Uint64>[ 0n ]
6. batpigandme commented on Aug 18, 2026

   

   batpigandme

   on Aug 18, 2026

   Member

   More actions

   AI-Generated Summary

   Caution

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

   Meeting Purpose

   To align on technical direction and unblock contributors.

   Key Takeaways

   • NAN APIs will use separate out and count arrays, not structs, for better performance. This is blocked by the in-progress Int64Array PR and Node.js binding updates.
   • New contributors should focus on C ND-array wrappers for packages like blast-extended-base and stats, as this work is unblocked.
   • Athan will prioritize reviewing pending Level 3 BLAS PRs (e.g., strided-gemm-tr) after GSOC ends this week.
   • Follow-up PRs are the standard process for applying new best practices (e.g., benchmark structure, Int64Array views) to previously merged code.

   Topics

   NAN API Design

   • Problem: stats-strided APIs need to return heterogeneous types (e.g., float64 sum, int64 count).
   • Rejected Solution: Structs
     1. Rationale: Creates non-contiguous data, hindering performance for subsequent operations.
     2. Precedent: Structs are reserved for complex results objects (e.g., Z-test) that match user expectations.
   • Chosen Solution: Separate Output Arrays
     1. API: d_nan_sum(..., out, offset_out, count, offset_count)
     2. Rationale: Mimics C APIs and provides contiguous arrays for efficient downstream processing.
   • Blockers:
     1. Int64Array Support: Requires merging the in-progress PR by Nicole.
     2. Node.js Bindings: Must be updated to parse both Int32Array and Int64Array arguments.
     3. CBLAS int Macro: The JS runtime needs a way to link to the C CBLAS_INT macro.
   • Status: All NAN-related refactoring is on hold pending these blockers.

   BLAS & SIMD Performance

   • Pending Level 3 BLAS PRs:
     • Issue: Many Level 3 BLAS routines (e.g., strided-gemm-tr) are not on NPM.
     • Cause: Reviewer bandwidth.
     • Action: Athan will prioritize reviews after GSOC.
   • New Routine: gemm-tr
     • Action: Manit will implement gemm-tr, using the existing gemm module as a reference.
   • SIMD/WebAssembly Optimization:
     • Status: Exploratory.
     • Challenge: Requires a major architectural shift to C++ (e.g., Google Highway) and a new build process.
     • Timeline: A clearer path is expected in the next few months.

   Contributor Workflow & Best Practices

   • Unblocked Work: C ND-array wrappers for packages like blast-extended-base and stats.
     • Action: Prashant will create a tracking issue to guide new contributors.
   • Follow-up PRs: The standard process for applying new best practices to previously merged code.
     • Example 1: Add Int64Array views to blast package (Siddharam).
     • Example 2: Align benchmark.native.js with benchmark.js (Ujjwal).
   • Masked Array Implementations:
     • Action: Ujjwal will create a demo PR for a masked array routine (e.g., dmask_mean).
     • Rationale: This establishes a pattern for the full family of APIs (e.g., dmask_mean, dnanmask_mean).
   • NaN Constant in C:
     • Guidance: Use the standard library macro instead of 0.0 / 0.0.
     • Rationale: Ensures consistency and avoids potential issues with signaling vs. quiet NaNs.
     • Reference: Issue #11743.

   Full transcript

   Full office hours transcript
7. batpigandme

   closed this as completedon Aug 25, 2026