Skip to contents

Warm up GPU backends to eliminate cold-start latency

Source: R/backend-warmup.R
amatrix_warm.Rd

Pre-compiles GPU kernels by running tiny dummy operations through each requested backend. Call once before timed work to pay JIT compilation costs upfront. Errors are silently swallowed; warming never alters numerical state.

Usage

amatrix_warm(
  backend = NULL,
  ops = c("matmul", "crossprod", "qr", "chol"),
  size = c(64L, 64L),
  quiet = FALSE
)

Arguments

backend

Character vector of backend names to warm, or NULL to warm all non-CPU backends currently registered.

ops

Character vector of operation names to trigger. Recognised values: "matmul", "crossprod", "tcrossprod", "qr", "chol", "svd", "solve".

size

Integer vector of length 2 giving the dimensions c(nrow, ncol) of the dummy matrices used during warming.

quiet

Logical; suppress progress messages when TRUE.

Value

An invisible named list, one entry per backend, each a list with elements warmed (logical) and elapsed_ms (numeric milliseconds, or NA when unavailable).

See also

amatrix_backend_names

Examples

# \donttest{
results <- amatrix_warm(quiet = TRUE)
# }