def alone

bend-lint-rules@0.2.0.0 · performance/gpu_fit.bend:199

import Base

199def alone(+callee: String) -> String:
200  callee
201    ++ "! runs on one GPU thread, alone: no parallel let runs under "
202    ++ callee
203    ++ ". GPU: measured 6.5 s for a list walk of BendSR, against 0.08 s without the `!`. CPU (`--gpu off`): 1.6x slower with the `!`. JS: no GPU, so no change. Confidence: high, as one GPU thread is much slower than a CPU core. Remove the `!`, or fork inside "
204    ++ callee
205    ++ " with a parallel let. If a timing shows the `!` faster here, expect this finding."