'num_threads' has no effect on benchmark_model target

[felix-johnny]

Hi!
I am doing the switch from using TFLite to LiteRT and these issues are related to that. If you see any incorrect usages or any tips, feel free to suggest!

OS: MacOS 15.5

Build Command: bazelisk build -c opt --config macos //litert/tools:benchmark_model --define xnn_enable_profiler=true --define tflite_with_xnnpack=true
--define xnn_enable_arm_i8mm=true
--define tflite_with_xnnpack_qs8=true
--define tflite_with_xnnpack_qu8=true --repo_env=HERMETIC_PYTHON_VERSION=3.12

Execution 1 : ./bazel-bin/litert/tools/benchmark_model --use_profiler=true --require_full_delegation=false --enable_op_profiling=true --num_threads=4 --use_profiler=true --require_full_delegation=false --enable_op_profiling=true --graph=my_model.tflite

Execution 2 : same_as_execution_1 num_threads=1

The inference times are the same wheras they should be different like in TFLite.

[felix-johnny]

Sha: af5583e

[pkgoogle]

Seems like a bug. Thanks.