⚠️ DISCONTINUATION OF PROJECT - This project will no longer be maintained by Intel. Intel has ceased development and contributions including, but not limited to, maintenance, bug fixes, new releases, or updates, to this project. Intel no longer accepts patches to this project. If you have an ongoing need to use this project, are interested in independently developing it, or would like to maintain patches for the open source software community, please create your own fork of this project.

Latte depends on Python 3. The preferred method of installation is using Miniconda, which provides a clean solution to manage Python distributions and dependencies. Otherwise, follow the instructions for your distribution to install Python 3.

wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

The following guides are for those working on Latte internals.

For developing Latte, it is convenient to tell Python to install the current working directoty as a "symbolic" package. This is done with:

cd ~/path/to/repo
pip install -r requirements.txt  # Install dependencies
pip install -e .

Now changes will be automatically imported without having to reinstall the package.

Configure the target compiler using the file ~/.ctree.cfg, for example, on a avx-2 platform:

[c]
CC = icpc
CFLAGS = -fPIC -O3 -std=c++11 -ansi-alias -xCORE-AVX2
LDFLAGS = -tbb

On a avx-512 platform (Xeon):

[c]
CC = icpc
CFLAGS = -fPIC -O3 -std=c++11 -ansi-alias -xCORE-AVX512
LDFLAGS = -tbb

On a avx-512 platform (Xeon Phi):

[c]
CC = icpc
CFLAGS = -fPIC -O3 -std=c++11 -ansi-alias -xMIC-AVX512
LDFLAGS = -tbb

Latte's parallelization strategy is configured with the environment variable LATTE_PARALLEL_STRATEGY. Supported values are:

• SIMPLE_LOOP - nested parallel_for (like basic TBB)
• FLOWGRAPH_LOOP - FlowGraph model for first level parallelism, parallel_for for nested
• OPENMP - pragma omp parallel for (supports collapse(2))
• OPENCL_SIMPLE_LOOP - converts parallel loops to NDRange kernel

**Note: ** When using the OpenCL or OpenMP based parallelization strategies, be sure to update your ~/.ctree.cfg to contain the correct flags so the latte compiler includes and links the proper files. For example, OpenMP strategies need the -openmp compiler flag.

Other command line flags:

• LATTE_VEC_CONFIG - either AVX-2 or AVX-512
• LATTE_NUM_THREADS - number of threads
• LATTE_CODEGEN_STRATEGY - set to AUTOVEC by default
• LATTE_MODE - either DEV or RELEASE (default should be RELEASE)
• LATTE_PREFETCH_MODE - either ON or OFF: enables software prefetching (default ON)
• LATTE_TIMER - either ON or OFF: prints layer by layer timing

Latte py.test package for testing purposes. The entire test suite can be run with with:

py.test test

or individual tests with

py.test test/<test_file>.py

Checkout the py.test documentation for more detailed usage documentation.