mukesh
b610cb0d9a
Convert the multidevice allreduce correctness + latency/buffer-kind sweeps to run through the real PyTorch-distributed path (init_process_group(backend="ahbm") -> mp.spawn -> dist.all_reduce) instead of direct ctx.launch, and reorganize the CCL/allreduce tests into a tests/sccl/ package split one test per file. Production change (required for the distributed path on non-square SIP grids): - AhbmCCLBackend now reads explicit system.sips.w/h from the spec, with a square-only sqrt fallback that raises on ambiguity, instead of silently guessing round(sqrt(count)). This fixes the 2x3 / 3x2 torus + mesh cases, which previously resolved to a wrong 2x2 grid. Mirrors the test helper's _sip_topo_dims precedence (explicit w/h > square fallback > raise). Test reorganization (tests/sccl/): - _allreduce_helpers.py: shared plumbing (distributed driver, config writers, direct-launch run_allreduce parity reference, sweep/buffer-kind constants, plot aggregators, topology-diagram + FSIM-comparison emitters). - test_allreduce_ring_torus_mesh.py: correctness across ring/torus/mesh. - test_distributed_default_topology.py: full distributed path on topology.yaml. - test_plot_latency_sweep.py / test_plot_buffer_kind_sweep.py: sweep rows. - test_plot_topology_diagram.py / test_plot_comparison_fsim.py: plot emitters. - test_intercube_root_center.py: moved in (ADR-0032 center-root latency guard). Also: - Move the FSIM comparison plot generator out of scripts/ into the sccl suite. - Delete superseded test files (test_allreduce_multidevice, test_distributed_lrab_hierarchical_allreduce, test_allreduce_buffer_kind_sweep) and repoint conftest aggregators + the ipcq buffer-kind importers. - Regenerate the allreduce_latency_plots derived artifacts from the full sweep. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
36 lines
1.0 KiB
Python
36 lines
1.0 KiB
Python
"""Correctness of intercube allreduce across SIP topologies (distributed path).
|
||
|
||
Routes through init_process_group → mp.spawn → dist.all_reduce for ring_1d,
|
||
torus_2d (2×3), and mesh_2d_no_wrap (2×3). Per-rank correctness is asserted
|
||
inside the worker; spawn raises on failure.
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import pytest
|
||
|
||
from tests.sccl._allreduce_helpers import (
|
||
CONFIGS,
|
||
DEFAULT_N_ELEM,
|
||
_crit_ns,
|
||
_run_distributed,
|
||
_write_temp_configs,
|
||
)
|
||
|
||
|
||
@pytest.mark.parametrize(
|
||
"algorithm,sip_topology,n_sips,sip_w,sip_h", CONFIGS,
|
||
)
|
||
def test_allreduce(
|
||
tmp_path, monkeypatch, algorithm, sip_topology, n_sips, sip_w, sip_h,
|
||
):
|
||
topo_path, _ = _write_temp_configs(
|
||
tmp_path, sip_topology, n_sips, algorithm,
|
||
sip_w=sip_w, sip_h=sip_h,
|
||
)
|
||
engine, _n_cubes = _run_distributed(
|
||
tmp_path, monkeypatch, topo_path,
|
||
f"test_{algorithm}_{sip_topology}", DEFAULT_N_ELEM,
|
||
)
|
||
# A positive critical path confirms the kernel actually ran.
|
||
assert _crit_ns(engine) > 0.0
|