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>
290 lines
9.3 KiB
Python
290 lines
9.3 KiB
Python
"""Tests for DataExecutor Phase 2 execution (ADR-0020 D6)."""
|
|
import numpy as np
|
|
|
|
from kernbench.sim_engine.data_executor import DataExecutor
|
|
from kernbench.sim_engine.memory_store import MemoryStore
|
|
from kernbench.sim_engine.op_log import OpRecord
|
|
|
|
|
|
def test_gemm_execution():
|
|
"""Phase 2 GEMM: out = a @ b with f32 accumulation."""
|
|
store = MemoryStore()
|
|
a = np.ones((4, 8), dtype=np.float16)
|
|
b = np.ones((8, 4), dtype=np.float16) * 2.0
|
|
store.write("tcm", 0x0, a)
|
|
store.write("tcm", 0x100, b)
|
|
|
|
op = OpRecord(
|
|
t_start=0.0, t_end=100.0,
|
|
component_id="pe_gemm",
|
|
op_kind="gemm", op_name="gemm_f16",
|
|
params={
|
|
"src_a_addr": 0x0, "src_b_addr": 0x100, "dst_addr": 0x200,
|
|
"shape_a": (4, 8), "shape_b": (8, 4), "shape_out": (4, 4),
|
|
"dtype_in": "f16", "dtype_acc": "f32", "dtype_out": "f16",
|
|
"addr_space": "tcm",
|
|
},
|
|
)
|
|
|
|
executor = DataExecutor([op], store)
|
|
executor.run()
|
|
|
|
result = store.read("tcm", 0x200)
|
|
expected = (a.astype(np.float32) @ b.astype(np.float32)).astype(np.float16)
|
|
assert np.allclose(result, expected)
|
|
|
|
|
|
def test_math_exp():
|
|
store = MemoryStore()
|
|
x = np.array([0.0, 1.0, 2.0], dtype=np.float32)
|
|
store.write("tcm", 0x0, x)
|
|
|
|
op = OpRecord(
|
|
t_start=0.0, t_end=10.0,
|
|
component_id="pe_math",
|
|
op_kind="math", op_name="exp",
|
|
params={
|
|
"op": "exp",
|
|
"input_addrs": [0x0], "input_shapes": [(3,)],
|
|
"dst_addr": 0x100, "shape_out": (3,),
|
|
"dtype": "f32", "axis": None, "addr_space": "tcm",
|
|
},
|
|
)
|
|
|
|
executor = DataExecutor([op], store)
|
|
executor.run()
|
|
|
|
result = store.read("tcm", 0x100)
|
|
assert np.allclose(result, np.exp(x))
|
|
|
|
|
|
def test_math_extra_ops():
|
|
"""Phase 2 replay of tl.maximum/minimum/fma/clamp/softmax."""
|
|
store = MemoryStore()
|
|
a = np.array([1.0, 5.0, 3.0], dtype=np.float32)
|
|
b = np.array([4.0, 2.0, 6.0], dtype=np.float32)
|
|
c = np.array([0.5, 0.5, 0.5], dtype=np.float32)
|
|
store.write("tcm", 0x0, a)
|
|
store.write("tcm", 0x100, b)
|
|
store.write("tcm", 0x200, c)
|
|
|
|
def _math(name, op, dst, inputs, axis=None):
|
|
return OpRecord(
|
|
t_start=float(dst), t_end=float(dst) + 1.0,
|
|
component_id="pe_math", op_kind="math", op_name=name,
|
|
params={
|
|
"op": op,
|
|
"input_addrs": [a for a, _ in inputs],
|
|
"input_shapes": [s for _, s in inputs],
|
|
"input_spaces": ["tcm"] * len(inputs),
|
|
"input_dtypes": ["f32"] * len(inputs),
|
|
"dst_addr": dst, "dst_space": "tcm",
|
|
"shape_out": (3,), "dtype": "f32", "axis": axis,
|
|
},
|
|
)
|
|
|
|
ops = [
|
|
_math("maximum", "maximum", 0x300, [(0x0, (3,)), (0x100, (3,))]),
|
|
_math("minimum", "minimum", 0x400, [(0x0, (3,)), (0x100, (3,))]),
|
|
_math("fma", "fma", 0x500, [(0x0, (3,)), (0x100, (3,)), (0x200, (3,))]),
|
|
_math("clamp", "clamp", 0x600, [(0x0, (3,)), (0x200, (3,)), (0x100, (3,))]),
|
|
]
|
|
DataExecutor(ops, store).run()
|
|
|
|
assert np.array_equal(store.read("tcm", 0x300), np.maximum(a, b))
|
|
assert np.array_equal(store.read("tcm", 0x400), np.minimum(a, b))
|
|
assert np.array_equal(store.read("tcm", 0x500), a * b + c)
|
|
assert np.array_equal(
|
|
store.read("tcm", 0x600), np.minimum(np.maximum(a, c), b)
|
|
)
|
|
|
|
|
|
def test_math_softmax():
|
|
store = MemoryStore()
|
|
x = np.array([[1.0, 2.0, 3.0], [10.0, 20.0, 30.0]], dtype=np.float32)
|
|
store.write("tcm", 0x0, x)
|
|
|
|
op = OpRecord(
|
|
t_start=0.0, t_end=1.0,
|
|
component_id="pe_math", op_kind="math", op_name="softmax",
|
|
params={
|
|
"op": "softmax",
|
|
"input_addrs": [0x0], "input_shapes": [(2, 3)],
|
|
"input_spaces": ["tcm"], "input_dtypes": ["f32"],
|
|
"dst_addr": 0x100, "dst_space": "tcm",
|
|
"shape_out": (2, 3), "dtype": "f32", "axis": -1,
|
|
},
|
|
)
|
|
DataExecutor([op], store).run()
|
|
expected = np.exp(x - x.max(axis=-1, keepdims=True))
|
|
expected /= expected.sum(axis=-1, keepdims=True)
|
|
assert np.allclose(store.read("tcm", 0x100), expected)
|
|
|
|
|
|
def test_math_add():
|
|
store = MemoryStore()
|
|
a = np.array([1.0, 2.0], dtype=np.float32)
|
|
b = np.array([3.0, 4.0], dtype=np.float32)
|
|
store.write("tcm", 0x0, a)
|
|
store.write("tcm", 0x100, b)
|
|
|
|
op = OpRecord(
|
|
t_start=0.0, t_end=5.0,
|
|
component_id="pe_math",
|
|
op_kind="math", op_name="add",
|
|
params={
|
|
"op": "add",
|
|
"input_addrs": [0x0, 0x100], "input_shapes": [(2,), (2,)],
|
|
"dst_addr": 0x200, "shape_out": (2,),
|
|
"dtype": "f32", "axis": None, "addr_space": "tcm",
|
|
},
|
|
)
|
|
|
|
executor = DataExecutor([op], store)
|
|
executor.run()
|
|
|
|
result = store.read("tcm", 0x200)
|
|
assert np.array_equal(result, np.array([4.0, 6.0], dtype=np.float32))
|
|
|
|
|
|
def test_math_sum_reduction():
|
|
store = MemoryStore()
|
|
x = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
|
|
store.write("tcm", 0x0, x)
|
|
|
|
op = OpRecord(
|
|
t_start=0.0, t_end=5.0,
|
|
component_id="pe_math",
|
|
op_kind="math", op_name="sum",
|
|
params={
|
|
"op": "sum",
|
|
"input_addrs": [0x0], "input_shapes": [(2, 2)],
|
|
"dst_addr": 0x100, "shape_out": (1, 2),
|
|
"dtype": "f32", "axis": 0, "addr_space": "tcm",
|
|
},
|
|
)
|
|
|
|
executor = DataExecutor([op], store)
|
|
executor.run()
|
|
|
|
result = store.read("tcm", 0x100)
|
|
assert np.array_equal(result, np.array([[4.0, 6.0]], dtype=np.float32))
|
|
|
|
|
|
def test_verify_pass():
|
|
store = MemoryStore()
|
|
store.write("hbm", 0x0, np.array([1.0, 2.0], dtype=np.float32))
|
|
|
|
executor = DataExecutor([], store)
|
|
results = executor.verify({
|
|
("hbm", 0x0): np.array([1.0, 2.0], dtype=np.float32),
|
|
})
|
|
assert results["hbm:0x0"] is True
|
|
|
|
|
|
def test_verify_fail():
|
|
store = MemoryStore()
|
|
store.write("hbm", 0x0, np.array([1.0, 2.0], dtype=np.float32))
|
|
|
|
executor = DataExecutor([], store)
|
|
results = executor.verify({
|
|
("hbm", 0x0): np.array([9.0, 9.0], dtype=np.float32),
|
|
})
|
|
assert results["hbm:0x0"] is False
|
|
|
|
|
|
def test_memory_ops_skipped():
|
|
"""Memory ops in op_log should be skipped (handled in Phase 1)."""
|
|
store = MemoryStore()
|
|
op = OpRecord(
|
|
t_start=0.0, t_end=5.0,
|
|
component_id="pe_dma",
|
|
op_kind="memory", op_name="dma_read",
|
|
params={"src_addr": 0x0, "nbytes": 64, "handle_id": "t0"},
|
|
)
|
|
# Should not raise
|
|
executor = DataExecutor([op], store)
|
|
executor.run()
|
|
|
|
|
|
def test_sequential_gemm_then_math():
|
|
"""GEMM output feeds into math op."""
|
|
store = MemoryStore()
|
|
a = np.eye(2, dtype=np.float16)
|
|
b = np.ones((2, 2), dtype=np.float16)
|
|
store.write("tcm", 0x0, a)
|
|
store.write("tcm", 0x100, b)
|
|
|
|
ops = [
|
|
OpRecord(
|
|
t_start=0.0, t_end=50.0,
|
|
component_id="pe_gemm",
|
|
op_kind="gemm", op_name="gemm_f16",
|
|
params={
|
|
"src_a_addr": 0x0, "src_b_addr": 0x100, "dst_addr": 0x200,
|
|
"shape_a": (2, 2), "shape_b": (2, 2), "shape_out": (2, 2),
|
|
"dtype_in": "f16", "dtype_acc": "f32", "dtype_out": "f32",
|
|
"addr_space": "tcm",
|
|
},
|
|
),
|
|
OpRecord(
|
|
t_start=50.0, t_end=55.0,
|
|
component_id="pe_math",
|
|
op_kind="math", op_name="exp",
|
|
params={
|
|
"op": "exp",
|
|
"input_addrs": [0x200], "input_shapes": [(2, 2)],
|
|
"dst_addr": 0x300, "shape_out": (2, 2),
|
|
"dtype": "f32", "axis": None, "addr_space": "tcm",
|
|
},
|
|
),
|
|
]
|
|
|
|
executor = DataExecutor(ops, store)
|
|
executor.run()
|
|
|
|
gemm_result = store.read("tcm", 0x200)
|
|
expected_gemm = (a.astype(np.float32) @ b.astype(np.float32)).astype(np.float32)
|
|
assert np.allclose(gemm_result, expected_gemm)
|
|
|
|
exp_result = store.read("tcm", 0x300)
|
|
assert np.allclose(exp_result, np.exp(expected_gemm))
|
|
|
|
|
|
def test_parallel_same_timestamp_ops():
|
|
"""Multiple independent ops at the same t_start produce correct results
|
|
when executed in parallel (ThreadPoolExecutor)."""
|
|
store = MemoryStore()
|
|
n_ops = 8
|
|
# Each op: independent GEMM writing to a different address
|
|
for i in range(n_ops):
|
|
a = np.full((4, 4), float(i + 1), dtype=np.float16)
|
|
b = np.eye(4, dtype=np.float16)
|
|
store.write("tcm", 0x1000 * i, a)
|
|
store.write("tcm", 0x1000 * i + 0x800, b)
|
|
|
|
ops = [
|
|
OpRecord(
|
|
t_start=0.0, t_end=100.0,
|
|
component_id=f"pe{i}.pe_gemm",
|
|
op_kind="gemm", op_name="gemm_f16",
|
|
params={
|
|
"src_a_addr": 0x1000 * i,
|
|
"src_b_addr": 0x1000 * i + 0x800,
|
|
"dst_addr": 0x80000 + 0x1000 * i,
|
|
"shape_a": (4, 4), "shape_b": (4, 4), "shape_out": (4, 4),
|
|
"dtype_in": "f16", "dtype_acc": "f32", "dtype_out": "f16",
|
|
"addr_space": "tcm",
|
|
},
|
|
)
|
|
for i in range(n_ops)
|
|
]
|
|
|
|
executor = DataExecutor(ops, store)
|
|
executor.run()
|
|
|
|
for i in range(n_ops):
|
|
result = store.read("tcm", 0x80000 + 0x1000 * i)
|
|
expected = np.full((4, 4), float(i + 1), dtype=np.float16)
|
|
assert np.allclose(result, expected), f"op {i}: expected {expected}, got {result}"
|