Repository navigation

CI run: Highway CPU kernels on Linux arm64 (not for merge) - #2

Draft

rcfa wants to merge 38 commits into

ci-base-t167p2from

t167p2-arm64

Draft

CI run: Highway CPU kernels on Linux arm64 (not for merge)#2
rcfa wants to merge 38 commits into
ci-base-t167p2from
t167p2-arm64

Commits on Sep 27, 2026
- Add Google Highway 1.4.0 as a submodule, not yet in any build
  rcfa
  committed
  ee8d4c1
- Add a check that SwiftPM and Xcode compile the same Cmlx code on macOS
  rcfa
  committed
  b56f6be
- Build the CPU backend's Highway kernels on Linux x86-64
  rcfa
  committed
  285a36b
- Add CmlxCPUShim: the CPU backend's Highway, int8 and thread controls
  rcfa
  committed
  0d7985e
- Add VMLXCPUKernelTests and its runner
  rcfa
  committed
  3e1781b
- Show the values of failed expectations in the kernel-test runner
  rcfa
  committed
  53f8dd3
- Measure MLX's polynomial accuracy on the scalar CPU code
  rcfa
  committed
  75759e0
- Test the CPU quantized matmul against exact data and a float32 bound
  rcfa
  committed
  c5f80c9
- Test the CPU fp32 and half-precision GEMMs around their thresholds
  rcfa
  committed
  1c95117
Commits on Sep 28, 2026
- Test CPU norms, RoPE and attention against float64 with derived bounds
  rcfa
  committed
  8420fbd
- Test the CPU elementwise functions, reductions and softmax
  rcfa
  committed
  b8e5806
- Run the CPU kernel tests in CI, and move Linux SwiftPM to Swift 6.4
  rcfa
  committed
  31a02da
- Expect OpenBLAS pinned at every CPU pool size
  rcfa
  committed
  cb79fae
- Test that the int8 switch starts off
  rcfa
  committed
  c00f8a4
- Compute the GEMM bound's reference on the host, in Double
  rcfa
  committed
  f262336
- Give the kernel-test runner distinct exit statuses, and a 1200 s stall
  rcfa
  committed
  df9796e
- Skip the tests that need Highway kernels, instead of passing them
  rcfa
  committed
  42c470e
- Make a measuring run of the polynomial tests fail
  rcfa
  committed
  b5e46a5
- Bound erf in half precision with the floor of its baseline
  rcfa
  committed
  0f7a013
- Judge the matmul bounds on the host, and name the worst element
  rcfa
  committed
  682fb90
- Check the shim's families and targets, and read back the int8 switch
  rcfa
  committed
  fd09f40
- Refuse Highway kernels off x86-64, and check the highway submodule
  rcfa
  committed
  fd2c2d5
- Correct the kernel tests' comments, and pin them to the CPU device
  rcfa
  committed
  fb96493
- Compare the kernel tests' exact results bit for bit
  rcfa
  committed
  9c74ec9
- Test the CPU gather quantized matmul, batched by expert and pooled
  rcfa
  committed
  db9276f
- Test untransposed and 2-bit quantized weights on the CPU
  rcfa
  committed
  a880c09
- Test the CPU's threaded gathers and general copies
  rcfa
  committed
  e3fd836
- Test float16 infinities and NaN through the norm kernels' loads
  rcfa
  committed
  a391d4f
- Replace private plan references with their reasons
  rcfa
  committed
  7800ffe
Commits on Sep 29, 2026
- Take the fork's thread-pool fix
  rcfa
  committed
  847adfc
Commits on Sep 30, 2026
- Take the fork's fix for Simd<bool> of a vector
  rcfa
  committed
  473b2e2
- Take the fork's fix for counting cores by the CPUs they hold
  rcfa
  committed
  68fe776
Commits on Oct 1, 2026
- Suppress two upstream MLX races for ThreadSanitizer runs
  rcfa
  committed
  2df59e8
- Build the Highway kernels on Linux arm64, with the SVE family disabled
  rcfa
  committed
  edb191a
- Build without Highway on request, and name the CPU on arm64
  rcfa
  committed
  65c404c
- Test fromFP8, dequantize and max/min against oracles of their own
  rcfa
  committed
  a03dafa
- Run arm64 CI with Highway, add a leg without it, and a TSan job
  rcfa
  committed
  b1609cb
- Take the fork's Highway build for Linux arm64
  rcfa
  committed
  b6481b8

Commits

Commits on Sep 27, 2026

Commits on Sep 28, 2026

Commits on Sep 29, 2026

Commits on Sep 30, 2026

Commits on Oct 1, 2026