Halide init time very large

halide / Halide

a language for fast, portable data-parallel computation

https://halide-lang.org

Other

5.91k stars 1.07k forks source link

Halide init time very large #7954

Open hbkaye opened 1 year ago

hbkaye commented 1 year ago

With debug flag enabled, halide_initialize_kernels show to be 5 times larger than rest of the execution. Is there a way of caching for Halide GPU so that init time is reduced. If need more details let me know what should I provide.

mcourteaux commented 1 year ago

The required extra info would include:

which GPU backend are you talking about?
what are the timings you are experiencing right now with and without debug flag?
the debug flag you are talking about is the Halide (generator) target-feature? or the C++ compiler debug flag, such as -g for GCC/Clang?
are you running generators (ahead of time) or JIT-compiled pipelines?