elvin-n commented on code in PR #11161: URL: https://github.com/apache/tvm/pull/11161#discussion_r865391074
########## python/tvm/topi/adreno/utils.py: ########## @@ -0,0 +1,545 @@ +# 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. +# pylint: disable=invalid-name,unused-variable,unused-argument,no-else-return +"""util functions to be reused in different compute/schedule on Qualcomm Adreno GPU""" + +import tvm +import numpy +from tvm import te +from tvm.topi.utils import simplify +from tvm.topi import nn +from ..utils import get_const_tuple + + +def getDiv(value, start): + """Returns the maximum divider for `value` starting from `start` value""" + div = 1 + for d in range(start, 0, -1): + if (value % d) == 0: + div = d + break + return div + + +def split_to_chunks(trip_count, block): + """ + Splits the trip count value to chunks and block, returns the remainder as well + the chunks and blocks covers or overlaps the origin trip_count + + If trip_count can be divisible by block: + trip_count = chunks * block + else + trip_count = (chunks - 1) * block + tail + + Parameters + ---------- + trip_count: int + tripcount for original compute + + block: int + size of the block + + Returns + ---------- + out: tuple of the (chunks, block, tail) + """ + tail = trip_count % 4 + chunks = trip_count // 4 + if tail == 0: + tail = 4 + else: + chunks += 1 + return chunks, block, tail + + +def pack_input( + Input, layout, batch, in_channel_chunks, in_channel_block, in_channel_tail, in_height, in_width +): + """ + Adds compute stages for packing of the data in runtime. Extends channel dimensions + to be dividable by factor 4 + + Parameters + ---------- + Input: tvm.te.Tensor + Input tensor to be repacked in runtime + + layout: string + Layout of origin 4d tensor + NCHW or NHWC are acceptable + + batch: int + Batch size + + in_channel_chunks: int + Number of channel chunks been in the final tensor + + in_channel_block: int + Number of channel blocks been in the final tensor + + in_channel_tail: int + Tail in the latest chunk diffing original number of channels vs blocked one + If in_channel_tail != in_channel_block: + original_channels = in_channel_chunks * in_channel_block - in_channel_tail Review Comment: tried to do my best -- 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]
