| BasedOnStyle: LLVM | |
| IndentWidth: 8 | |
| UseTab: Always | |
| BreakBeforeBraces: Linux | |
| AllowShortIfStatementsOnASingleLine: false | |
| IndentCaseLabels: false | |
| ColumnLimit: 0 | |
| AlignTrailingComments: true | |
| SpacesBeforeTrailingComments: 2 |
This is a short post that explains how to write a high-performance matrix multiplication program on modern processors. In this tutorial I will use a single core of the Skylake-client CPU with AVX2, but the principles in this post also apply to other processors with different instruction sets (such as AVX512).
Matrix multiplication is a mathematical operation that defines the product of
I'm posting here a detailed summary of the problem and the situation with the Jetson TK1 board and compiling the final caffe release. The final caffe release requires cudnn v.5. If you attempt to compile on the Jetson TK1 board you will give an error from nvcc saying "Unsupported gpu architecture". I describe the
On the Jetson TK1 if you clone the final caffe release and compile:
make all -j 4
The nvcc compiler will show the following error:
| # -*- mode: dockerfile -*- | |
| # Dockerfile to build libmxnet.so for armv6 | |
| FROM dockcross/linux-armv6 | |
| ENV ARCH armv6l | |
| ENV BUILD_OPTS "USE_BLAS=openblas USE_SSE=0 USE_OPENCV=0" | |
| ENV CC /usr/bin/arm-linux-gnueabihf-gcc | |
| ENV CXX /usr/bin/arm-linux-gnueabihf-g++ | |
| ENV FC /usr/bin/arm-linux-gnueabihf-gfortran | |
| ENV HOSTCC gcc |
| """References: | |
| Simonyan, Karen, and Andrew Zisserman. "Very deep convolutional networks for | |
| large-scale image recognition." arXiv preprint arXiv:1409.1556 (2014). | |
| """ | |
| import mxnet as mx | |
| def depthwise_conv(data, kernel, pad, num_filter, name, num_group): | |
| conv = mx.symbol.Convolution(data=data, kernel=kernel, pad=pad, | |
| num_filter=num_group, name=name+'_depthwise', num_group=num_group) |
Pre-reqs:
| #!/bin/sh | |
| # | |
| # Modified from http://qiita.com/janus_wel/items/cfc6914d6b7b8bf185b6 | |
| # | |
| # An example hook script to verify what is about to be committed. | |
| # Called by "git commit" with no arguments. The hook should | |
| # exit with non-zero status after issuing an appropriate message if | |
| # it wants to stop the commit. | |
| # | |
| # To enable this hook, rename this file to "pre-commit". |
| # Make sure you grab the latest version | |
| curl -OL https://github.com/google/protobuf/releases/download/v3.2.0/protoc-3.2.0-linux-x86_64.zip | |
| # Unzip | |
| unzip protoc-3.2.0-linux-x86_64.zip -d protoc3 | |
| # Move protoc to /usr/local/bin/ | |
| sudo mv protoc3/bin/* /usr/local/bin/ | |
| # Move protoc3/include to /usr/local/include/ |
| // half - IEEE 754-based half-precision floating point library. | |
| // | |
| // Copyright (c) 2012-2013 Christian Rau <rauy@users.sourceforge.net> | |
| // | |
| // Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation | |
| // files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, | |
| // modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the | |
| // Software is furnished to do so, subject to the following conditions: | |
| // | |
| // The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. |
