Where is the definition or addition location of GGML_USE_HYBRID_THREADING?

[wfloveiu]

Where is the definition or addition location of GGML_USE_HYBRID_THREADING?

In the int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) function in the ggml.c file

#ifdef GGML_USE_HYBRID_THREADING
    int compute_status = (size_t) ggml_graph_compute_thread_hybrid(&workers[0]);
#else
    int compute_status = (size_t) ggml_graph_compute_thread(&workers[0]);
#endif

I see that the ggml_graph_compute_thread_hybrid function is added by you (I think this may be an importance function for PowerInfer), but my global search did not find the addition of GGML_USE_HYBRID_THREADING.

[hodlen]

The GGML_USE_HYBRID_THREADING flag acts as a switch to decide if all GPU computation nodes should be exclusively assigned to a single specific CPU worker. By default, this feature is not active, hence you won't find its definition within the source code. If you wish to activate this feature, it can be enabled through the definition in CMake's CXXFLAGS and CFLAGS as follows:

cmake -S . -B build -DCMAKE_CXXFLAGS="-DGGML_USE_HYBRID_THREADING" -DCMAKE_CFLAGS="-DGGML_USE_HYBRID_THREADING"

[wfloveiu]

Thank you for your answer