R API reference

library(autozyme)

The functions you call to use AutoZyme’s accelerators in your own pipeline. The R API mirrors the Python API one-to-one — same names, same semantics; the only differences are R idioms (list instead of dict, character vector instead of list[str]).

The shape of the API — almost everything is one of three moves:

On / off — activate(name) turns patches on; deactivate(name) / deactivate_all() turn them off; with_disabled({ ... }) turns them off for a single block.
Look — status() (what’s on), list_patches() (what’s available), inspect(name) (detail on one), dashboard() (one-screen view).
Extras — set_threads() (CPU threads), env_snapshot() (reproducibility record), speedups() (published numbers).

`activate(name)`

activate(name) -> logical | named logical

Activate one or more patches.

Parameters

name — patch name (character), bundle name, or character vector.

Returns

TRUE for a single named patch that activated.
Named logical for multiple (c(seurat = TRUE, cellchat = FALSE, ...)).

Example

activate("seurat")                              # TRUE
activate(c("seurat", "cellchat"))               # c(seurat=TRUE, cellchat=TRUE)
activate("scrna_signaling")                     # bundle: activates 4 patches

`deactivate(name)`

deactivate(name) -> invisible(NULL)

Inverse of activate — rebinds upstream to its captured original. Patch stays registered for cheap re-activation.

`deactivate_all()`

deactivate_all() -> invisible(NULL)

Deactivate every active patch.

`with_disabled(expr)`

with_disabled({ ...expr... })

Evaluate expr with all patches temporarily disabled. Equivalent to Python’s with autozyme.disabled():. Works for namespace patches AND S4/class-method patches.

Example

with_disabled({
  Seurat::FindAllMarkers(obj)           # uses upstream original
  Seurat::FindNeighbors(obj)            # also original
})

`inspect(name)`

inspect(name) -> list

Returns a list with name, status, tested_against, installed_version, upstream, targets (list of per-target info). (An uninstalled patch returns name, status, error, targets instead.)

`status()`

status() -> named character

c(seurat = "active", cellchat = "inactive", ...).

`list_patches(installed = FALSE)`

list_patches(installed = FALSE) -> character

All shipped patches. With installed=TRUE, only those whose upstream is installed via requireNamespace(..., quietly=TRUE).

`list_subsets()`

list_subsets() -> character

Bundle names: c("scrna_signaling", "scrna_spatial", "scrna_trajectory").

`subset_patches(name)`

subset_patches(name) -> character

Member patch names of a bundle. R uses subset_patches (not subset) because base R has subset.default.

`env_snapshot()`

env_snapshot() -> list

Structured environment snapshot: autozyme version, R version, platform, and per-patch state (status, installed upstream versions, tested-against versions). Does not include BLAS or thread counts.

`speedups(name, history = FALSE)`

speedups(name, history = FALSE) -> data.frame

Published rows from the shipped finalized speedup data. With history=TRUE, returns every historical attest run.

`set_threads(n)`

set_threads(n) -> integer

Sets the BLAS/OpenMP env vars (OMP_NUM_THREADS, OPENBLAS_NUM_THREADS, MKL_NUM_THREADS, VECLIB_MAXIMUM_THREADS, NUMEXPR_NUM_THREADS), options(autozyme.threads = n), and — when installed — RhpcBLASctl / RcppParallel thread counts. Returns the value actually set. Rejects n <= 0.

`dashboard()`

dashboard() -> invisible(<env_snapshot list>)

Prints a one-screen status table to console: every patch, upstream availability, activation state, headline speedup. R equivalent of python -m autozyme.

Differences from the Python API

Python	R	Why
`subset(name)`	`subset_patches(name)`	Base R reserves `subset()` as an S3 generic on data.frame
`with autozyme.disabled():`	`autozyme::with_disabled({ ... })`	R doesn’t have context managers — `with_disabled` takes a code block expression
`status() -> dict[str, str]`	`status() -> named character`	R’s idiomatic equivalent

Where the Python API has dict[str, X], R returns a named list or named vector — same data shape.