quantlab.factor.kunquant_compile

Compiled KunQuant libraries, kept in the process and on disk: compile_graphs.

KunQuant turns a graph into C++ and compiles it, which takes a second or more each time. Every compilation in quantlab goes through compile_graphs, which compiles a graph once and then reuses the library: the same process gets the library it already loaded, and another process (a ChunkedBuilder worker, the next daily run) loads it from the cache directory.

A library is keyed by everything that changes the binary: each graph’s text (str(Function), deterministic across processes, unlike the generated C++), its KunCompilerConfig, the source of every quantlab module that defines one of its operators (a custom operator’s C++ is written by its Python class), the CppCompilerConfig and the compiler’s version, the host CPU (KunQuant compiles with -march=native) and the KunQuant installation.

The cache directory is $QUANTLAB_KUNQUANT_CACHE, else $XDG_CACHE_HOME/quantlab/kunquant, else ~/.cache/quantlab/kunquant; an empty QUANTLAB_KUNQUANT_CACHE keeps libraries in the process only. Each entry is a directory named by the key holding the library and a complete marker with the library’s digest; an entry without the marker, or whose library does not match it, is compiled again. Deleting the directory clears the cache.

Functions

cache_dir	Return the directory compiled libraries are cached in, or None for none.
compile_graphs	Return the compiled library of graphs, compiling only when no cache has it.
graph_key	Return the hex digest a library of graphs is cached under.