ibsidorenko commented on code in PR #12398: URL: https://github.com/apache/tvm/pull/12398#discussion_r997408406
########## python/tvm/topi/hexagon/qnn.py: ########## @@ -0,0 +1,678 @@ +# 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. +"""Hexagon QNN operators""" +# pylint: disable=invalid-name + +import tvm +from tvm import te, topi +from ..utils import get_const_tuple +from ..nn.utils import get_pad_tuple +from ..nn.pad import pad +from .. import tag, nn +from ..x86.concat import concatenate + + +def clip_cast(val, dtype): + # clip + cast: + const_min = tvm.tir.min_value(dtype) + const_max = tvm.tir.max_value(dtype) + return te.max(tvm.te.min(val, const_max), const_min).astype(dtype) + + +def get_qnn_param(param, indices, axis): + # Account scalar and 1D quantization parameters: + if len(param.shape) == 0: + return param + + param_idx = tvm.tir.indexmod(indices[axis], topi.shape(param)[0]) + return param[param_idx] + + +def default_schedule(outs): + """Simple default schedule for QNN ops. + + Parameters + ---------- + outs: Array of Tensor + The computation graph description of dense in the format + of an array of tensors. + + Returns + ------- + sch: Schedule + The computation schedule for the op. + """ + outs = [outs] if isinstance(outs, tvm.te.tensor.Tensor) else outs + s = tvm.te.create_schedule([x.op for x in outs]) + tvm.te.schedule.AutoInlineInjective(s) + return s + + +def qnn_quantize(data, output_scale, output_zero_point, axis, out_dtype): + """Compute for qnn.quantize + + Note! This is POC code. There was no goal to implement high performance compute function. Review Comment: Removed. It was a reminder for QC people. -- 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]
