tqchen commented on a change in pull request #9315: URL: https://github.com/apache/tvm/pull/9315#discussion_r734574998
########## File path: gallery/tutorial/tensor_ir_blitz_course.py ########## @@ -0,0 +1,183 @@ +# 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. +""" +.. _tir_blitz: + +Blitz Course to TensorIR +======================== +**Author**: `Siyuan Feng <https://github.com/Hzfengsy>`_ + +TensorIR is a domain specific languages for deep learning programs serving two broad purposes: + +- An implementation for transforming and optimizing programs on various hardware backends. + +- An abstraction for automatic tensorized program optimization. + +""" + +import tvm +from tvm.ir.module import IRModule +from tvm.script import tir as T +import numpy as np + +################################################################################################ +# IRModule +# -------- +# An IRModule is the central data structure in TVM, which contains deep learning programs. +# It is the basic object of interest of IR transformation and model building. +# +# .. image:: https://raw.githubusercontent.com/Hzfengsy/web-data/main/images/design/tvm_life_of_irmodule.png +# :align: center +# :width: 85% +# +# This is the life cycle of an IRModule, which can be created from TVM Script. TensorIR schedule +# primitives and passes are two major ways to transform an IRModule. Also, a sequence of +# transformations on an IRModule is acceptable. Note that we can print an IRModule at **ANY** stage +# to TVMScript. After all transformations and optimizations are complete, we can build the IRModule +# to a runnable module to deploy on target devices. +# +# Based on the design of TensorIR and IRModule, we are able to create a new programming method: +# +# 1. Write a program by TVMScript (just like write python codes) +# +# 2. Transform and optimize a program with python api (by schedule primitives and passes) +# +# 3. Interactively inspect and try the performance (print or build at any stage of IRModule) + + +################################################################################################ +# Create an IRModule +# ------------------ +# IRModule can be created by writing TVMScript, which is a round-trippable syntax for TVM IR. +# +# Different than creating an computational expression by Tensor Expression Review comment: avoid mentioning TE tutorial for now since that can change in the future. We can just directly say TensorIR allow users to write programs through tvm script, a language embedded in python AST -- 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]
