masahi commented on a change in pull request #10461: URL: https://github.com/apache/tvm/pull/10461#discussion_r818978034
########## File path: src/meta_schedule/schedule_rule/winograd.cc ########## @@ -0,0 +1,96 @@ +/* + * 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. + */ +#include "../utils.h" + +namespace tvm { +namespace tir { + +TVM_REGISTER_GLOBAL("meta_schedule.compute_inline") + .set_body_typed([](Schedule sch, BlockRV block) -> Array<Schedule> { + sch->ComputeInline(block); + return {sch}; + }); + +inline BlockRV GetOnlyProducer(Schedule sch, BlockRV block) { + Array<BlockRV> producers = sch->GetProducers(block); + ICHECK_EQ(producers.size(), 1); + return producers[0]; +} + +inline LoopRV ScheduleDataPack(Schedule sch, BlockRV block) { + Array<ExprRV> factors{nullptr}; + Array<LoopRV> loops = sch->GetLoops(block); + ICHECK_EQ(loops.size(), 6); + + factors = sch->SamplePerfectTile(loops[2], /*n=*/2, /*max_innermost_factor=*/64); + Array<LoopRV> t0 = sch->Split(loops[2], {factors.begin(), factors.end()}); + ICHECK_EQ(t0.size(), 2); + + factors = sch->SamplePerfectTile(loops[3], /*n=*/2, /*max_innermost_factor=*/64); + Array<LoopRV> t1 = sch->Split(loops[3], {factors.begin(), factors.end()}); + ICHECK_EQ(t1.size(), 2); + + sch->Unroll(loops[0]); + sch->Unroll(loops[1]); + sch->Unroll(loops[4]); + sch->Unroll(loops[5]); + sch->Reorder({ + t0[0], + t1[0], + t0[1], + t1[1], + loops[0], + loops[1], + loops[4], + loops[5], + }); + return t1[1]; +} + +TVM_REGISTER_GLOBAL("meta_schedule.winograd_inverse") + .set_body_typed([](Schedule sch, BlockRV block) -> Array<Schedule> { + ScheduleDataPack(sch, block); + return {sch}; + }); + +TVM_REGISTER_GLOBAL("meta_schedule.winograd_data_pack.cpu") + .set_body_typed([](Schedule sch, BlockRV data_pack) -> Array<Schedule> { + BlockRV input_tile = GetOnlyProducer(sch, data_pack); + BlockRV data_pad = GetOnlyProducer(sch, input_tile); + ScheduleDataPack(sch, data_pack); + sch->ComputeAt(input_tile, /*loop_rv=*/sch->SampleComputeLocation(input_tile), + /*preserve_unit_loops=*/true); + sch->ComputeAt(data_pad, /*loop_rv=*/sch->SampleComputeLocation(data_pad), + /*preserve_unit_loops=*/true); + return {sch}; + }); + +TVM_REGISTER_GLOBAL("meta_schedule.winograd_data_pack.cuda") + .set_body_typed([](Schedule sch, BlockRV data_pack) -> Array<Schedule> { + BlockRV input_tile = GetOnlyProducer(sch, data_pack); + BlockRV data_pad = GetOnlyProducer(sch, input_tile); + LoopRV loop = ScheduleDataPack(sch, data_pack); + sch->ComputeAt(input_tile, /*loop_rv=*/loop, /*preserve_unit_loops=*/true); + sch->SetScope(input_tile, /*buffer_index=*/0, /*storage_scope=*/"local"); + sch->ComputeInline(data_pad); + return {sch}; Review comment: Now that we have TIR constants, I wonder if we can remove the need for * the giant `Select` tree in the test file `conv2d_winograd_cpu.py` etc * the mandatory unroll + inline schedule rules in this file cc @manupa-arm @d-smirnov -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
