Allen Guo gglin001

Run RISC-V Debian GNU/Linux bullseye/sid via QEMU.

Run the latest version of Debian on regular x86_64 box (at least ver 10 Buster, better to run ver 11 Bullseye)
If opensbi and u-boot-qemu packages are not found add testing apt repository (aka bullseye). Or even unstable (aka sid)

sudo vi /etc/apt/sources.list

# Add testing repo (or unstable)
deb http://cdn-aws.deb.debian.org/debian testing main
deb-src http://cdn-aws.deb.debian.org/debian testing main

PyTorch Channels Last Memory Format Performance Optimization on CPU Path

("mkldnn" has been renamed to "oneDNN", but exsiting PyTorch APIs still use "mkldnn", future work will align PyTorch user level APIs to "oneDNN")

NB: Memory format refers to data representation that describes how multidimensional arrays (nD) are stored in linear (1D) memory address space. Memory format has the same semantic with layout in oneDNN. Layout in PyTorch has other semantic ofdescribing dense or sparse with the attributes: 'torch.strided', 'torch.sparse_coo'.

This is a technique for extracting all imported modules from a packaged Python application as .pyc files, then decompiling them. The target program needs to be run from scratch, but no debugging symbols are necessary (assuming an unmodified build of Python is being used).

This was originally performed on 64-bit Linux with a Python 3.6 target. The Python scripts have since been updated to handle pyc files for Python 2.7 - 3.9.

Theory

In Python we can leverage the fact that any module import involving a .py* file will eventually arrive as ready-to-execute Python code object at this function:

PyObject* PyEval_EvalCode(PyObject *co, PyObject *globals, PyObject *locals);

Docker Hub 镜像加速器

国内从 Docker Hub 拉取镜像有时会遇到困难，此时可以配置镜像加速器。

Dockerized 实践 https://github.com/y0ngb1n/dockerized

1️⃣ Docker daemon 配置代理（推荐）

参考 Docker daemon 配置代理

vi /etc/environment

add these lines...

LANG=en_US.utf-8
LC_ALL=en_US.utf-8

Alternatively,

	#!/bin/bash

	# Attempt to set up the Nvidia GeForce GT 710 on a Pi CM4.
	#
	# I have tried both armv7l and aarch64 versions of the proprietary driver, in
	# addition to the nouveau open source driver (which needs to be compiled into
	# a custom Raspberry Pi kernel).
	#
	# tl;dr - None of the drivers worked :P

	# This isn't supposed to run as a bash script, i named it with ".sh" for syntax highlighting.

	# https://developer.nvidia.com/nsight-systems
	# https://docs.nvidia.com/nsight-systems/profiling/index.html

	# My preferred nsys (command line executable used to create profiles) commands
	#
	# In your script, write
	# torch.cuda.nvtx.range_push("region name")
	# ...

	@echo off

	SET PhpStormPath=C:\Program Files\JetBrains\PhpStorm 2020.3.3\bin\phpstorm64.exe

	echo Adding file entries
	@reg add "HKEY_CLASSES_ROOT\*\shell\PhpStorm" /t REG_SZ /v "" /d "Open in PhpStorm" /f
	@reg add "HKEY_CLASSES_ROOT\*\shell\PhpStorm" /t REG_EXPAND_SZ /v "Icon" /d "%PhpStormPath%,0" /f
	@reg add "HKEY_CLASSES_ROOT\*\shell\PhpStorm\command" /t REG_SZ /v "" /d "%PhpStormPath% \"%%1\"" /f

	echo Adding within a folder entries

	import os, sys

	import google.protobuf
	from tensorflow.core.protobuf import saved_model_pb2
	import tensorflow as tf


	def convert_saved_model_to_pbtxt(path):
	saved_model = saved_model_pb2.SavedModel()
	with open(os.path.join(path, 'saved_model.pb'), 'rb') as f:

	import sys
	from ctypes import POINTER, py_object, Structure, c_ssize_t, c_void_p, sizeof
	from typing import Any, Iterator, Optional, Sequence, Union


	__all__ = ("OpStack", )


	class Frame(Structure):
	_fields_ = (

Allen Guo gglin001

PyTorch Channels Last Memory Format Performance Optimization on CPU Path

Table of Contents

Theory

Docker Hub 镜像加速器

1️⃣ Docker daemon 配置代理（推荐）