TVM Mobile Benchmark

this folder contain the patch of benchmark needed, ios and android tuning and inference python script of tvm.

Build TVM

First get tvm code from github:

git clone https://github.com/apache/tvm.git

Checkout to the benchmark version:

cd tvm && git checkout df6ccaec3bff66f8879c97df279cf096a6e73079

Apply the patch:

git apply benchmark.patch

Then build tvm from source by ref.

Android Benchmark

Build an Android RPC apk at tvm/apps/android_rpc by ref, and run:

# Specify the RPC tracker
export TVM_TRACKER_HOST=0.0.0.0
export TVM_TRACKER_PORT=[PORT]
# Specify the standalone Android C++ compiler
export TVM_NDK_CC=/opt/android-toolchain-arm64/bin/aarch64-linux-android-g++

# start RPC tracker
python -m tvm.exec.rpc_tracker --port 9090 --host 0.0.0.0
# start tuning and inference
python android_tuning.py <test_model_file>

iOS Benchmark

Build an iOS RPC apk at tvm/apps/ios_rpc by ref, and run:

export TVM_IOS_RPC_PROXY_HOST=0.0.0.0
export TVM_IOS_RPC_ROOT=${TVM_HOME}/your/testscript/path
export TVM_IOS_CODESIGN='Apple Development: xxxxx(xxxxx)'
export TVM_IOS_RPC_DESTINATION='platform=iOS,id=xxxxx-xxxxxxx'

# start RPC tracker
python -m tvm.exec.rpc_tracker --host=192.168.101.49  --port=9190 --no-fork
# start RPC proxy
python -m tvm.exec.rpc_proxy --no-fork --host 0.0.0.0 --port 9090 --tracker 0.0.0.0:9190
# tuning and inference
# change the `network` in ios_tuning.py to specify test model
python ios_tuning.py

CUDA Benchmark

After building tvm from source, run the script: python tvmc_cuda_test.py

MNN Semi-Search Time

The time of MNN resizeSession contains all semi-search time using. Computing the time of resizeSession in MNN as below:

git apply mnn_semisearch_time.patch
# then run the benckmark script of MNN 
# get the output info: ### MNN Semi-Search Time is : xxx ms