pytc.xtc._estimate_delta_u_direct_tile_bytes¶
- pytc.xtc._estimate_delta_u_direct_tile_bytes(Np, Nq, Nr, Ns, N_rank, *, include_d=True)[source]¶
Estimate device memory for the balanced direct Delta U tile kernel.
Delegates to the canonical formula in
pytc.utils.tile_memory.isdf_tile_peak_bytes()so that the runtime memory guard and the build-phase estimators ingpu_memory.pyalways agree.include_dshould be False when D is already a persistent on-device resident (cached by the per-device cache); otherwise D gets counted both inin_use(via the free-bytes probe) and again in the total peak — effectively doubling its contribution and triggering spurious “tile exceeds device memory” refusals.