metal-linalg for PyTorch

QR, symmetric eigendecomposition and SVD for batches of matrices on Apple GPUs, on PyTorch tensors.

import torch
import metal_linalg_torch as mlt

a = torch.randn(1000, 64, 32, device="mps")   # a batch of 1000 matrices; "cpu" works too

Q, R = mlt.qr(a)                     # Q (1000, 64, 32), R (1000, 32, 32)
U, S, Vh = mlt.svd(a)                # thin: U (1000, 64, 32), S (1000, 32), Vh (1000, 32, 32)
L, V = mlt.eigh(a.mT @ a)            # L ascending (1000, 32), V (1000, 32, 32)

The functions take the arguments of their torch.linalg namesakes and return the same result types (Q, R = ..., result.eigenvalues, ...), on the device of their input. Each call is routed to the fastest Metal kernel for its shape and batch, or to LAPACK on the CPU, by a policy measured on the Mac it runs on:

mlt.device_name()                      # 'Apple M5 Pro'
mlt.eigh_policy_source()               # 'tuned:Apple M5 Pro'
mlt.eigh_backend(32, 4096)             # 'ql': 4096 matrices of 32x32 go to the GPU
mlt.svd_backend(4096, 4096)            # 'bidiag': GPU bidiagonalization, LAPACK's solver
mlt.set_eigh_policy(gpu_min_batch=1)   # override the measured policy

The kernels, the routing and the measurements behind them are described in the main README.

Installing

pip install metal-linalg-torch

On an Apple Silicon Mac with macOS 14 or later, Python 3.10 or later and PyTorch 2.4 or later. This is the PyTorch package; for MLX arrays install metal-linalg instead. The two are independent: this one does not install or load MLX.

One wheel serves every PyTorch and every Python: the package calls the library's C API (on plain float buffers) and is not compiled against either, so upgrading torch never breaks it.

To build from source instead (needs Xcode's command line tools):

pip install ./python-torch                                                      # in a clone
pip install "git+https://github.com/c0rmac/metal-linalg.git#subdirectory=python-torch"

Functions

A is [..., M, N] with any number of batch dimensions, on "cpu" or "mps". What differs from torch.linalg:

• float32. The kernels compute in float32. Other real dtypes (float64, float16, bfloat16, integers) are converted, and results are float32; complex input raises TypeError.
• Thin factors only. svd defaults to full_matrices=False (torch's default is True), and full_matrices=True or qr(mode="complete") on a non-square matrix raises NotImplementedError.
• No exceptions for bad matrices. A matrix with a NaN or an infinity gives NaN results for that matrix alone, leaving the rest of its batch intact, where torch raises.

Autograd and torch.compile

Underneath, the functions are custom operators, torch.ops.metal_linalg.{qr, eigh, eigvalsh, svd, svdvals}, with fake implementations and the backward formulas torch.linalg uses:

A = torch.randn(64, 32, 32, device="mps", requires_grad=True)
S = mlt.svdvals(A)
S.sum().backward()                       # A.grad: the gradient of the nuclear norm

f = torch.compile(lambda x: mlt.eigh(x @ x.mT).eigenvalues.sum())
f(torch.randn(8, 16, 16))

As in torch, the gradients of eigh and svd are defined only for distinct eigenvalues or singular values (they divide by their differences), and are not unique for a loss that depends on the signs of the vectors. Second derivatives are not supported. eigvalsh and svdvals compute the vectors as well when their input requires grad, since the gradient needs them.

Performance

Against torch.linalg on an M5 Pro with PyTorch 2.14 and metal-linalg 2.10 (best of five; the same tensors on MPS for torch's MPS path and for this package, copies included):

It is ahead on every row: 1.6-2.9x over PyTorch's MPS kernels for batches of small matrices (the SVD of 256 matrices of 128×64 runs on the library's CPU path, which spreads a batch over every core, so routing an MPS tensor to the CPU can still be the fast choice), and 2-50x in QR and in large matrices, where torch falls back to the CPU. Which backend a shape gets on your Mac: mlt.svd_backend(m, n, batch) and its siblings.

MPS tensors

The library runs its own Metal kernels on host memory. An MPS tensor is copied to the CPU (which waits for the work PyTorch has queued on the GPU) and the results are copied back: two copies at memory bandwidth around a decomposition. A CPU tensor is used in place, and its results stay on the CPU, though the work still runs on the GPU where that is faster.

Calls are thread-safe; they are serialised, and release the GIL.

Calibration

Each Mac's routing comes from measurements of that Mac. On one that has none (or older ones), the import raises a CalibrationWarning once per decomposition, and mlt.calibration_status() says where each stands. The library works either way, with a cautious default; measuring the Mac takes one command and improves it for everyone with that Mac: see which Macs are measured and how to contribute. METAL_LINALG_NO_CALIBRATION_NOTICE=1 silences the warning.