shard_apply_matrix: Apply a Function Over Matrix Columns with Shared Inputs

In shard: Deterministic, Zero-Copy Parallel Execution for R

Apply a Function Over Matrix Columns with Shared Inputs

Description

A convenience wrapper for the common "per-column apply" pattern. The matrix is shared once and each worker receives a zero-copy column view when possible.

Usage

shard_apply_matrix(
  X,
  MARGIN = 2,
  FUN,
  VARS = NULL,
  workers = NULL,
  ...,
  policy = shard_apply_policy()
)

Arguments

X	A numeric/integer/logical matrix (or a shared matrix created by share()).
MARGIN	Must be 2 (columns).
FUN	Function of the form ⁠function(v, ...)⁠ returning a scalar atomic.
VARS	Optional named list of extra variables. Large atomic VARS are auto-shared based on policy$auto_share_min_bytes.
workers	Number of workers (passed to shard_map()).
...	Additional arguments forwarded to FUN.
policy	A shard_apply_policy() object.

Details

Current limitation: MARGIN must be 2 (columns). Row-wise apply would require strided/gather slicing and is intentionally explicit in shard via views/kernels.

Value

An atomic vector of length ncol(X) with the results.

Examples

X <- matrix(rnorm(400), 20, 20)
shard_apply_matrix(X, MARGIN = 2, FUN = mean)
pool_stop()

shard documentation built on April 3, 2026, 9:08 a.m.