Skip to content

Advance the ndarray pin to master's current head - #7

Merged
AdaWorldAPI merged 1 commit into
mainfrom
claude/ndarray-simd-tract-o3jfrn
Sep 4, 2026
Merged

Advance the ndarray pin to master's current head#7
AdaWorldAPI merged 1 commit into
mainfrom
claude/ndarray-simd-tract-o3jfrn

Conversation

@AdaWorldAPI

Copy link
Copy Markdown
Owner

Lockfile-only: moves ndarray, p64 and fractal from e55af0e3 to f43338fb, so tract builds against ndarray master as it now stands rather than a pin predating ndarray sonos#297/sonos#298/sonos#299.

Why this one matters

ndarray sonos#299 repaired the AVX-512 test modules, which had been failing to compile whenever AVX-512 was enabled at build time — ndarray's CI does not pass -C target-cpu, so those modules were cfg'd out there and CI stayed green while the suite was broken under the target-cpu=x86-64-v4 configuration ndarray's own .cargo/config.toml ships. That suite covers the bf16 tile primitive this crate's AMX kernels call (bf16_tile_gemm_16x16_packed, PackedBf16B), so the old pin left tract's AMX path depending on ndarray code whose parity tests could not run on the configuration that matters here.

Verification

On the new pin, RUSTFLAGS="-C target-cpu=native":

  • cargo check -p tract-linalg — clean
  • cargo test -p tract-linalg --lib4463 passed, 0 failed, 3 ignored

Scope

No source changes; Cargo.lock only, three source = lines. No behaviour change intended and none observed.

Note for anyone picking up the AMX benchmarking thread: this container reports bf16_tile_gemm_tier() = "F32x16 FMA polyfill (bf16→f32 decode)" with amx_available() = false and no avx512bf16, so it cannot reproduce or extend PR #6's AMX measurements — that work needs the Sapphire-Rapids-class silicon the original numbers came from.

🍍

Co-Authored-By: Claude Opus 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_012wrzeZAdwGYTCKoxamwQht

🤖 Generated with Claude Code

https://claude.ai/code/session_012wrzeZAdwGYTCKoxamwQht


Generated by Claude Code

The lock still resolved ndarray, p64 and fractal at e55af0e3, which
predates ndarray sonos#297/sonos#298/sonos#299. The AVX-512 test modules that sonos#299
repaired are the ones covering the bf16 tile primitive this crate's
AMX kernels call, so building tract against the older pin exercised a
tree where that suite could not compile under target-cpu=x86-64-v4.

cargo check -p tract-linalg clean and cargo test -p tract-linalg --lib
4463 passed / 0 failed against the new pin.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012wrzeZAdwGYTCKoxamwQht
@cursor

cursor Bot commented Sep 4, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_7e711e6d-e8f0-4cb9-9edf-d9682d2e15ab)

@AdaWorldAPI
AdaWorldAPI marked this pull request as ready for review September 4, 2026 18:21
@AdaWorldAPI
AdaWorldAPI merged commit e9928da into main Sep 4, 2026
55 of 66 checks passed
@cursor

cursor Bot commented Sep 4, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_9e417cf8-e050-42af-b581-c0b9b3199ca6)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants