ywkang
998cc85762
Major changes:
PE-level IPCQ infrastructure:
- New PE_IPCQ component: ring-buffer control plane with 4-direction
neighbor mapping, head/tail pointers, backpressure (poll/sleep).
- PE_DMA extended with vc_comm channel for IPCQ outbound/inbound DMA,
including in-flight data snapshot (D9) and op_log recording at
outbound time for Phase 2 replay correctness.
- IpcqDmaToken piggyback model: data + metadata travel together,
atomic visibility at receiver (invariant I6).
- Credit return fast path: bottleneck-BW latency, no fabric vc_comm.
Phase 2 data execution (ADR-0020 integration):
- op_log extended: DmaWriteCmd now captures src_space/src_addr for
Phase 2 dma_write copy; ipcq_copy ops recorded at outbound time.
- DataExecutor replays dma_write + ipcq_copy in t_start order.
- Engine._flush_data_phase: incremental cursor-based replay after
each engine.wait() so host reads see post-Phase-2 data.
- KernelRunner Phase 1 writes disabled when op_log is active to
prevent stale data from corrupting the MemoryStore snapshot.
TLContext / kernel API:
- tl.send(dir, src=TensorHandle), tl.recv(dir, shape, dtype),
tl.recv_async, tl.wait(RecvFuture), copy_to_dst mode.
- TensorHandle operator overloading (add/sub/mul/div) via thread-local
active TLContext → MathCmd dispatch through PE_MATH.
- PE-local scratch allocator for math output handles.
- tl.load returns space="hbm" handles for correct Phase 2 addressing.
- Additional math functions: maximum, minimum, fma, clamp, softmax, cdiv.
Unified ccl_allreduce bench (PyTorch-compat host code):
- Single benches/ccl_allreduce.py with run() + worker(rank, ws, torch)
split matching real PyTorch DDP worker pattern.
- torch.distributed facade: init_process_group, get_world_size,
get_rank, get_backend, all_reduce, barrier — only real PyTorch names.
- AhbmCCLBackend: eager install_ipcq at init, all_reduce dispatches
kernel via tensor shard metadata (n_elem from shards[0].nbytes).
- world_size derived from topology spec (sips × cubes × pes_per_cube)
with optional algorithm-level override in ccl.yaml.
Tensor API (PyTorch-compat surface):
- Tensor.numpy(): gather-aware (all shards via VA-based addressing).
- Tensor.copy_(source): scatter from host tensor into sharded target.
- RuntimeContext.from_numpy(arr): host-side staging tensor.
- Tensor.data property fixed to use numpy() (was shards[0]-only).
Algorithm modules moved to src/kernbench/ccl/algorithms/:
- ring_allreduce, mesh_allreduce, tree_allreduce, hello_send.
- Each module exports kernel_args(world_size, n_elem) helper.
- ccl.yaml module paths updated to kernbench.ccl.algorithms.*.
Dead code removed:
- 7 per-variant bench files (ccl_allreduce_{tcm,hbm,sram}, etc.).
- _run_ccl_bench greenlet-per-SIP scheduler.
- benches.loader.is_ccl_bench + run_rank detection.
- benches/ccl/ directory.
Tests:
- New test_ccl_allreduce_matrix.py: 7 parametrized cases
(ring×3 buffers, ring 8/16, mesh 4, tree 7).
- New test_runtime_api_tensor.py: copy_/numpy/from_numpy unit tests.
- Existing tests updated for new import paths + world_size_override.
Docs:
- Korean ccl-author-guide.md and ADR-0023 paths updated.
- New English versions: ccl-author-guide.en.md, ADR-0023.en.md.
502 tests pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
107 lines
3.3 KiB
Python
107 lines
3.3 KiB
Python
"""Tests for tl.recv_async + tl.wait (ADR-0023 D4)."""
|
|
from __future__ import annotations
|
|
|
|
import numpy as np
|
|
|
|
from kernbench.ccl.testing import run_kernel_in_mock
|
|
|
|
|
|
def kernel_async_recv(t_ptr, n_elem, tl):
|
|
"""Each PE issues recv_async first, then send, then wait — this exercises
|
|
the non-blocking path. Uses TensorHandle math (PE_MATH) for accumulation
|
|
so Phase 2 produces correct final HBM contents."""
|
|
rank = tl.program_id(axis=0)
|
|
world_size = tl.num_programs(axis=0)
|
|
nbytes = n_elem * 2
|
|
|
|
pe_addr = t_ptr + rank * nbytes
|
|
acc = tl.load(pe_addr, shape=(n_elem,), dtype="f16")
|
|
current = acc
|
|
|
|
for _step in range(world_size - 1):
|
|
future = tl.recv_async(dir="W", shape=(n_elem,), dtype="f16")
|
|
tl.send(dir="E", src=current)
|
|
recv = tl.wait(future)
|
|
acc = acc + recv
|
|
current = recv # forward W's tile to E next round
|
|
|
|
tl.store(pe_addr, acc)
|
|
|
|
|
|
def test_recv_async_mock_runtime():
|
|
n_elem = 8
|
|
inputs = [
|
|
np.full((n_elem,), float(r + 1), dtype=np.float16)
|
|
for r in range(4)
|
|
]
|
|
expected = sum(inputs)
|
|
|
|
outputs = run_kernel_in_mock(
|
|
kernel_fn=kernel_async_recv,
|
|
world_size=4,
|
|
topology="ring_1d",
|
|
inputs=inputs,
|
|
kernel_args=(n_elem,),
|
|
)
|
|
for r in range(4):
|
|
assert np.allclose(outputs[r], expected)
|
|
|
|
|
|
def test_recv_async_simpy_runner():
|
|
"""Run the async kernel through the real SimPy stack via the
|
|
install_ipcq + launch path.
|
|
"""
|
|
import importlib
|
|
|
|
from kernbench.runtime_api.bench_runner import run_bench
|
|
from kernbench.runtime_api.types import resolve_device
|
|
from kernbench.sim_engine.engine import GraphEngine
|
|
from kernbench.topology.builder import resolve_topology
|
|
|
|
# Re-use the standard 8-PE bench skeleton but swap in the async kernel.
|
|
topo = resolve_topology("topology.yaml")
|
|
|
|
# Build a tiny inline bench module
|
|
import types
|
|
mod = types.ModuleType("inline_bench_async")
|
|
|
|
from kernbench.policy.placement.dp import DPPolicy
|
|
|
|
def run(torch):
|
|
plan = torch.install_ipcq(
|
|
algorithm="ring_allreduce_tcm", world_size_override=8,
|
|
)
|
|
a = torch.zeros(
|
|
(1, 8 * 8),
|
|
dtype="f16",
|
|
dp=DPPolicy(
|
|
sip="replicate", cube="replicate", pe="column_wise",
|
|
num_sips=1, num_cubes=1,
|
|
),
|
|
name="async_in",
|
|
)
|
|
store = torch.engine.memory_store
|
|
base = a._handle.va_base or a._handle.shards[0].pa
|
|
nbytes = 8 * 2
|
|
for r in range(8):
|
|
store.write("hbm", base + r * nbytes,
|
|
np.full((8,), float(r + 1), dtype=np.float16))
|
|
|
|
torch.launch("ring_allreduce_tcm", kernel_async_recv, a, 8)
|
|
|
|
for r in range(8):
|
|
result = store.read("hbm", base + r * nbytes, shape=(8,), dtype="f16")
|
|
expected = float(sum(range(1, 9))) # 36
|
|
assert np.allclose(result, expected, rtol=1e-2, atol=1e-2), \
|
|
f"rank {r}: got {result}, expected {expected}"
|
|
|
|
mod.run = run
|
|
result = run_bench(
|
|
topology=topo, bench_fn=mod.run,
|
|
device=resolve_device("all"),
|
|
engine_factory=lambda t, d: GraphEngine(
|
|
getattr(t, "topology_obj", t), enable_data=True
|
|
),
|
|
)
|
|
assert result.completion.ok
|