merrymercy commented on a change in pull request #5962: URL: https://github.com/apache/incubator-tvm/pull/5962#discussion_r449730892
########## File path: python/tvm/ansor/auto_schedule.py ########## @@ -0,0 +1,207 @@ +# 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. + +""" +User interface for Ansor auto-scheduler. + +The basic schedule search process for Ansor is designed to be: +`Program sampling` -> `Performance Tuning`. + +In `Program sampling`, we use some predefined precise or heuristic rules to generate several +initial schedules. Based on these initial starting points, we perform `Performance Tuning` which +uses cost model based evolutionary search to select schedules with the best performance. + +Candidate schedules are measured against the specific hardware target. +""" + +import tvm._ffi +from tvm.runtime import Object +from .compute_dag import ComputeDAG +from .measure import LocalBuilder, LocalRunner +from . import _ffi_api + + +@tvm._ffi.register_object("ansor.HardwareParams") +class HardwareParams(Object): + """ The parameters of target hardware used to guide the search process of SearchPolicy. + + TODO(jcf94): This is considering to merge with the new Target: + https://discuss.tvm.ai/t/rfc-tvm-target-specification/6844 + + Parameters + ---------- + num_cores : int + The number of device cores. + vector_unit_bytes : int + The width of vector units in bytes. + cache_line_bytes : int + The size of cache line in bytes. + max_unroll_vec : int + The max length of an axis to be unrolled or vectorized. + max_innermost_split_factor : int + The max split factor for the innermost tile. + """ + def __init__(self, num_cores, vector_unit_bytes, cache_line_bytes, + max_unroll_vec, max_innermost_split_factor): + self.__init_handle_by_constructor__(_ffi_api.HardwareParams, num_cores, + vector_unit_bytes, cache_line_bytes, + max_unroll_vec, max_innermost_split_factor) + + +@tvm._ffi.register_object("ansor.SearchTask") +class SearchTask(Object): + """ The computation information and hardware parameters for a specific schedule search task. + + Parameters + ---------- + dag : ComputeDAG + The ComputeDAG for the target compute declaration. + workload_key : str + The workload key for the target compute declaration. Review comment: ```suggestion The workload key for the compute declaration. ``` Do not overuse `target` because `target` already has its meaning (i.e. the hardware target) ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
