declare(parallel()) can generate incorrect results for scalar accumulation loops

[t-kalinowski]

Summary

declare(parallel()) can emit an OpenMP loop with a shared scalar accumulator update (total <- total + x[i]), which is a data race and produces incorrect/nondeterministic results.

Original report and discussion: #88 (comment)

Verified on

• Branch: main
• Local reproduction date: 2026-02-17

Minimal reproducible example

library(devtools)
load_all(quiet = TRUE)

loop_sum_unsafe <- function(x) {
  declare(type(x = double(n)))
  total <- 0.0
  declare(parallel())
  for (i in seq_along(x)) {
    total <- total + x[i]
  }
  total
}

Sys.setenv(OMP_NUM_THREADS = "4", OMP_THREAD_LIMIT = "4", OMP_DYNAMIC = "false")
q_unsafe <- quick(loop_sum_unsafe)

x <- rep(1, 100000L)
vapply(1:6, function(i) q_unsafe(x), numeric(1))

Expected

Each run should return 100000.

Observed

Repeated calls return incorrect and varying results (example from local run):
30961, 31102, 25000, 31544, 30804, 31507

Generated code shape

r2f(loop_sum_unsafe) emits:

!$omp parallel do
do i = 1, size(x)
  total = (total + x(i))
end do
!$omp end parallel do

This loop updates shared total without a reduction/private strategy.

Notes

declare(parallel(private = total)) is not currently accepted on main (parallel() does not accept arguments yet.), so users cannot express a reduction/private fix in the current interface.