263 lines
8.0 KiB
Python
263 lines
8.0 KiB
Python
# Licensed to the Apache Software Foundation (ASF) under one
|
|
# or more contributor license agreements. See the NOTICE file
|
|
# distributed with this work for additional information
|
|
# regarding copyright ownership. The ASF licenses this file
|
|
# to you under the Apache License, Version 2.0 (the
|
|
# "License"); you may not use this file except in compliance
|
|
# with the License. You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing,
|
|
# software distributed under the License is distributed on an
|
|
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
|
|
# KIND, either express or implied. See the License for the
|
|
# specific language governing permissions and limitations
|
|
# under the License.
|
|
# ruff: noqa: E501, F821
|
|
"""A script to measure GPU memory bandwidth"""
|
|
|
|
import argparse
|
|
import itertools
|
|
|
|
import numpy as np
|
|
|
|
import tvm
|
|
from tvm import te
|
|
from tvm.s_tir.meta_schedule.runner import EvaluatorConfig, RPCConfig
|
|
from tvm.testing import local_run, rpc_run
|
|
|
|
|
|
def _parse_args() -> argparse.Namespace:
|
|
def _parse_list_int(source: str):
|
|
return [int(i) for i in source.split(",")]
|
|
|
|
parser = argparse.ArgumentParser(
|
|
prog="GPU memory bandwidth testing",
|
|
description="""Example for host GPU:
|
|
python -m tvm.exec.gpu_memory_bandwidth "nvidia/geforce-rtx-3090-ti" \
|
|
--dtype "float32"
|
|
--bx "8,16,32,64,128,256" \
|
|
--tx "32,64,128,256,512,1024" \
|
|
--vec "1,2,4" \
|
|
|
|
Example for Android GPU: \
|
|
python -m tvm.exec.gpu_memory_bandwidth "opencl" --target_host '{"kind": "llvm", "mtriple": "arm64-linux-android"}' \
|
|
--rpc_host "127.0.0.1" \
|
|
--rpc_port 9190 \
|
|
--rpc_key "android" \
|
|
--export_func "ndk" \
|
|
--dtype "float32" \
|
|
--bx "8,16,32,64,128,256" \
|
|
--tx "32,64,128,256,512,1024" \
|
|
--vec "1,2,4" \
|
|
""",
|
|
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
|
|
)
|
|
parser.add_argument(
|
|
"target",
|
|
type=str,
|
|
help="The target to be benchmarked",
|
|
)
|
|
parser.add_argument(
|
|
"--target_host",
|
|
type=str,
|
|
default=None,
|
|
help="The target host for build",
|
|
)
|
|
parser.add_argument(
|
|
"--xo",
|
|
type=int,
|
|
default=1024,
|
|
help="The value of `XO` in [XO, K, XI] => [XO, XI] reduction",
|
|
)
|
|
parser.add_argument(
|
|
"--k",
|
|
type=int,
|
|
default=64,
|
|
help="The value of `K` in [XO, K, XI] => [XO, XI] reduction",
|
|
)
|
|
parser.add_argument(
|
|
"--xi",
|
|
type=int,
|
|
default=4096,
|
|
help="The value of `XI` in [XO, K, XI] -> [XO, XI] reduction",
|
|
)
|
|
parser.add_argument(
|
|
"--dtype",
|
|
type=str,
|
|
default="float32",
|
|
help="The data type to be used in the workload",
|
|
)
|
|
parser.add_argument(
|
|
"--bx",
|
|
type=_parse_list_int,
|
|
default=[8, 16, 32, 64, 128, 256],
|
|
help="The value to be used to split `XO` into [BX, _]",
|
|
)
|
|
parser.add_argument(
|
|
"--tx",
|
|
type=_parse_list_int,
|
|
default=[32, 64, 128, 256, 512, 1024],
|
|
help="Number of threads to be used",
|
|
)
|
|
parser.add_argument(
|
|
"--vec",
|
|
type=_parse_list_int,
|
|
default=[1, 2, 4],
|
|
help="Vector length to be used in vectorized load",
|
|
)
|
|
parser.add_argument(
|
|
"--rpc_host",
|
|
type=str,
|
|
default=None,
|
|
help="The address of RPC host (default: None, that means that RPC is not used)",
|
|
)
|
|
parser.add_argument(
|
|
"--rpc_port",
|
|
type=int,
|
|
default=None,
|
|
help="The port of RPC connection (default: None, that means that RPC is not used)",
|
|
)
|
|
parser.add_argument(
|
|
"--rpc_key",
|
|
type=str,
|
|
default=None,
|
|
help="The device key in RPC tracker (default: None, that means that RPC is not used)",
|
|
)
|
|
parser.add_argument(
|
|
"--export_func",
|
|
type=str,
|
|
default="tar",
|
|
help="Export function, actual only for RPC",
|
|
choices=["tar", "ndk"],
|
|
)
|
|
return parser.parse_args()
|
|
|
|
|
|
def _workload(
|
|
len_xo: int,
|
|
len_k: int,
|
|
len_xi: int,
|
|
dtype: str,
|
|
):
|
|
# pylint: disable=invalid-name
|
|
A = te.placeholder((len_xo, len_k, len_xi), dtype=dtype, name="A")
|
|
k = te.reduce_axis((0, len_k), "k")
|
|
B = te.compute(
|
|
(len_xo, len_xi),
|
|
lambda i, j: te.sum(A[i, k, j], axis=k),
|
|
name="B",
|
|
)
|
|
# pylint: enable=invalid-name
|
|
return te.create_prim_func([A, B])
|
|
|
|
|
|
def _schedule(
|
|
sch: s_tir.Schedule,
|
|
len_bx: int,
|
|
len_tx: int,
|
|
len_vec: int,
|
|
):
|
|
# pylint: disable=invalid-name
|
|
block = sch.get_sblock("B")
|
|
xo, xi, k = sch.get_loops(block)
|
|
bx, xo = sch.split(xo, factors=[len_bx, None])
|
|
xi, tx, vec = sch.split(xi, factors=[None, len_tx, len_vec])
|
|
sch.reorder(bx, xi, tx, xo, k, vec)
|
|
bx = sch.fuse(bx, xi)
|
|
sch.bind(bx, "blockIdx.x")
|
|
sch.bind(tx, "threadIdx.x")
|
|
ldg = sch.cache_read(block, 0, "local")
|
|
sch.compute_at(ldg, k, preserve_unit_loops=True)
|
|
sch.vectorize(sch.get_loops(ldg)[-1])
|
|
sch.decompose_reduction(block, k)
|
|
# pylint: enable=invalid-name
|
|
|
|
|
|
def main(): # pylint: disable=too-many-locals
|
|
"""Entry point"""
|
|
args = _parse_args()
|
|
# pylint: disable=invalid-name
|
|
target = tvm.target.Target(args.target)
|
|
if args.target_host is not None:
|
|
target = tvm.target.Target(args.target, host=args.target_host)
|
|
dtype = args.dtype
|
|
rpcConfig = None
|
|
if args.rpc_host is not None and args.rpc_port is not None and args.rpc_key is not None:
|
|
rpcConfig = RPCConfig(
|
|
tracker_host=args.rpc_host,
|
|
tracker_port=args.rpc_port,
|
|
tracker_key=args.rpc_key,
|
|
session_priority=1,
|
|
session_timeout_sec=600,
|
|
)
|
|
|
|
a = np.random.uniform(-1, 1, (args.xo, args.k, args.xi)).astype(dtype)
|
|
b = np.zeros((args.xo, args.xi), dtype=dtype)
|
|
num_bytes = a.size * a.itemsize + b.size * b.itemsize
|
|
print("###### Bandwidth Test ######")
|
|
print(
|
|
f"Workload [XO, K, XI] => [XO, XI]. "
|
|
f"[{args.xo}, {args.k}, {args.xi}] => [{args.xo}, {args.xi}]"
|
|
)
|
|
print(f"Input size: {num_bytes / 1048576} MB")
|
|
print(f"Target: {target}")
|
|
|
|
# pylint: enable=invalid-name
|
|
best_bandwidth = -1
|
|
for len_bx, len_tx, len_vec in itertools.product(
|
|
args.bx,
|
|
args.tx,
|
|
args.vec,
|
|
):
|
|
func = _workload(
|
|
len_xo=args.xo,
|
|
len_k=args.k,
|
|
len_xi=args.xi,
|
|
dtype=dtype,
|
|
)
|
|
sch = tvm.s_tir.Schedule(func)
|
|
_schedule(sch, len_bx, len_tx, len_vec)
|
|
|
|
if rpcConfig is None:
|
|
_, profile_result = local_run(
|
|
tvm.compile(sch.mod, target=target),
|
|
target.kind.name,
|
|
[a, b],
|
|
evaluator_config=EvaluatorConfig(
|
|
number=10,
|
|
repeat=1,
|
|
min_repeat_ms=100,
|
|
enable_cpu_cache_flush=False,
|
|
),
|
|
)
|
|
else:
|
|
_, profile_result = rpc_run(
|
|
tvm.compile(sch.mod, target=target),
|
|
target.kind.name,
|
|
[a, b],
|
|
evaluator_config=EvaluatorConfig(
|
|
number=10,
|
|
repeat=1,
|
|
min_repeat_ms=100,
|
|
enable_cpu_cache_flush=False,
|
|
),
|
|
rpc_config=rpcConfig,
|
|
export_func=args.export_func,
|
|
)
|
|
bandwidth = num_bytes / profile_result.mean / (1024**3)
|
|
bx = len_bx * args.xi // (len_tx * len_vec) # pylint: disable=invalid-name
|
|
mbs = num_bytes / 1024 / 1024
|
|
print(
|
|
f"bandwidth = {bandwidth:.3f} GB/s, bx = {bx}, tx = {len_tx}, "
|
|
f"len_vec = {len_vec}, bytes = {mbs} MB"
|
|
)
|
|
if bandwidth > best_bandwidth:
|
|
best_bandwidth = bandwidth
|
|
print(f"peak bandwidth: {best_bandwidth:.3f} GB/s")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|