slorado

Building CUDA version of slorado on x84_64

To build for the NVIDIA GPU, you need to have the CUDA toolkit installed. We have tested with CUDA 10, 11 and 12.
A minimum g++ version of 5.4 (available on Ubuntu 16.04 or higher) is required as of slorado 0.2.0 due to libtorch v2.0.0.

Install zlib development files needed for slow5lib:

 On Debian/Ubuntu : sudo apt-get install zlib1g-dev
 On Fedora/CentOS : sudo dnf/yum install zlib-devel

Clone the slorado repository recursively

 git clone --recursive https://github.com/BonsonW/slorado
 cd slorado

Run the script that downloads and extracts torchlib.
```
 scripts/install-torch2.sh cuda
```
Invoke make
```
 make cuda=1 -j
```
See the slorado version
```
 ./slorado --version
```

By default it is assumed that you have CUDA on the standard system location (/usr/local/cuda/). There should be the nvcc compiler at /usr/local/cuda/bin/nvcc. The library files (.a files) should be present under /usr/local/cuda/lib64/. If the CUDA location is different, you can specify the path manually as:
```
 make cuda=1 CUDA_ROOT=/path/to/cuda/
```
Make sure you have nvcc at /path/to/cuda/bin/nvcc and library files at /path/to/cuda/lib64.
You can provide the CUDA architecture as make cuda=1 CUDA_ARCH=-arch=sm_xy
Custom libtorch path:
```
  make LIBTORCH_DIR=/path/to/libtorch
```
C++11 ABI (if you are using torch version with C++11 ABI):
```
  make cxx11_abi=1
```
You can optionally enable zstd support for built-in slow5lib when building slorado by invoking make zstd=1. This requires zstd 1.3 development libraries installed on your system (libzstd1-dev package for apt, libzstd-devel for yum/dnf).

This site is open source. Improve this page.