Using silt

Fundamentally, the goal of silt is to enable the building of modular tensor accelerated libraries without heavy infrastructure or modifications of your build system. Therefore, silt has been designed to be trivially includable in your CMake + CUDA project.

Building a CUDA Library

  1. Import git submodule:

silt is most easily imported into your project as a git submodule:

git submodule add git@github.com:erosiv/silt.git ext/silt
git submodule update --init --recursive

  1. Add to CMake:

silt can then be built by adding it as a subdirectory to your CMakeLists.txt.

add_subdirectory(${CMAKE_SOURCE_DIR}/ext/silt silt)

This will build and expose two targets, silt_lib which is a shared C++ library, and silt which is a python module.

  1. Link to your C++ Library:

Finally, you can make the headers available and link silt to your own C++ library in CMake:

target_link_libraries(${TARGET_NAME} PUBLIC silt_lib)

Full CMake Example

# my_lib CMakeLists.txt Example

set(TARGET_NAME my_lib)
add_library(${TARGET_NAME} STATIC)
set_target_properties(${TARGET_NAME} PROPERTIES
  LINKER_LANGUAGE CUDA
)

target_include_directories(${TARGET_NAME} PUBLIC
$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/source>
  .
)

target_sources(${TARGET_NAME} PUBLIC
  # ... .cpp source files ...
)

# ... other dependencies ...

add_subdirectory(${CMAKE_SOURCE_DIR}/ext/silt silt)
target_link_libraries(${TARGET_NAME} PUBLIC silt_lib)

Building Python Bindings

When building python bindings that expose functions which accept silt types, you can either link silt explicitly in CMake, or you can link your C++ library.

The silt repository itself generates the python bindings for silt, you can inspect CMakeLists.txt to see an example linking structure.

Note that when you use silt as a dependency in a separate project’s python bindings, it is interoperable with other libraries that use silt. This is possible because of dllexport directives on the primary exposed type: the tensor.