# Copyright 2022 The IREE Authors # # Licensed under the Apache License v2.0 with LLVM Exceptions. # See https://llvm.org/LICENSE.txt for license information. # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception # Run commands below on the workstation that the phone is attached to. # Prerequisites: # Manual installations of the Android NDK and ADB are needed. See https://iree.dev/building-from-source/android/#install-android-ndk-and-adb for instructions. # Manual installations of the Termux App and python are needed on the Android device. See README.md for instructions. #!/bin/bash set -euo pipefail GPU_TYPE="mali" #GPU_TYPE="andreno" # Create root dir. ROOT_DIR=/tmp/benchmarks rm -rf "${ROOT_DIR}" mkdir "${ROOT_DIR}" mkdir "${ROOT_DIR}/models" mkdir "${ROOT_DIR}/models/tflite" mkdir "${ROOT_DIR}/models/iree" mkdir "${ROOT_DIR}/setup" mkdir "${ROOT_DIR}/test_data" mkdir "${ROOT_DIR}/output" # Touch result file as adb doesn't push empty dirs. touch "${ROOT_DIR}/output/results.csv" wget https://storage.googleapis.com/iree-model-artifacts/tflite_squad_test_data.zip -O /tmp/tflite_squad_test_data.zip unzip /tmp/tflite_squad_test_data.zip -d "${ROOT_DIR}/test_data/" wget https://storage.googleapis.com/iree-model-artifacts/mobilebert-baseline-tf2-quant.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/mobilebert_float_384_gpu.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/mobilenet_v2_224_1.0_uint8.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/mobilenet_v2_1.0_224.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/deeplabv3.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/person_detect.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/ssd_mobilenet_v2_static_1.0_int8.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/resnet_v2_101_1_default_1.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/asr_conformer_int8.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/albert_lite_base_squadv1_1.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/ssd_mobilenet_v2_fpnlite_fp32.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/ssd_mobilenet_v2_fpnlite_uint8.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/inception_v4_299_fp32.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/inception_v4_299_uint8.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/efficientnet_lite0_fp32_2.tflite -P "${ROOT_DIR}/models/tflite/" wget https://storage.googleapis.com/iree-model-artifacts/efficientnet_lite0_int8_2.tflite -P "${ROOT_DIR}/models/tflite/" # Build IREE source. SOURCE_DIR=/tmp/github rm -rf "${SOURCE_DIR}" mkdir "${SOURCE_DIR}" cd "${SOURCE_DIR}" git clone https://github.com/openxla/iree.git cd iree cp "${SOURCE_DIR}/iree/build_tools/benchmarks/set_adreno_gpu_scaling_policy.sh" "${ROOT_DIR}/setup/" cp "${SOURCE_DIR}/iree/build_tools/benchmarks/set_android_scaling_governor.sh" "${ROOT_DIR}/setup/" cp "${SOURCE_DIR}/iree/build_tools/benchmarks/set_pixel6_gpu_scaling_policy.sh" "${ROOT_DIR}/setup/" git submodule update --init cmake -GNinja -B ../iree-build/ -S . -DCMAKE_BUILD_TYPE=RelWithDebInfo -DIREE_ENABLE_ASSERTIONS=ON -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DIREE_ENABLE_LLD=ON cmake --build ../iree-build/ export CC=clang export CXX=clang++ python3 configure_bazel.py # TODO(mariecwhite): Use Python-based importers (no Bazel build) cd integrations/tensorflow bazel build -c opt iree_tf_compiler:iree-import-tflite ./symlink_binaries.sh echo "Done building iree-import-tflite" echo IREE_IMPORT_TFLITE_PATH=${SOURCE_DIR}/iree/integrations/tensorflow/bazel-bin/iree_tf_compiler/iree-import-tflite IREE_COMPILE_PATH="${SOURCE_DIR}/iree-build/tools/iree-compile" TFLITE_MODEL_DIR="${ROOT_DIR}/models/tflite" IREE_MODEL_DIR="${ROOT_DIR}/models/iree" mkdir -p "${IREE_MODEL_DIR}/vulkan" mkdir -p "${IREE_MODEL_DIR}/llvm-cpu" # Runs `iree-compile` on all TFLite files in directory. If compilation fails, we # keep going. for i in $(ls ${ROOT_DIR}/models/tflite/); do MODEL_NAME=$(basename $i .tflite) echo "Processing ${MODEL_NAME} ..." ${IREE_IMPORT_TFLITE_PATH} "${TFLITE_MODEL_DIR}/${MODEL_NAME}.tflite" -o "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" || true echo -e "\tCompiling ${MODEL_NAME}.vmfb for aarch64..." "${IREE_COMPILE_PATH}" \ --iree-input-type=tosa \ --iree-hal-target-backends=llvm-cpu \ --iree-llvmcpu-target-triple=aarch64-none-linux-android29 \ --iree-llvmcpu-debug-symbols=false \ --iree-vm-bytecode-module-strip-source-map=true \ --iree-vm-emit-polyglot-zip=false \ "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" \ --o "${IREE_MODEL_DIR}/llvm-cpu/${MODEL_NAME}.vmfb" || true echo -e "\tCompiling ${MODEL_NAME}_padfuse.vmfb for aarch64..." "${IREE_COMPILE_PATH}" \ --iree-input-type=tosa \ --iree-hal-target-backends=llvm-cpu \ --iree-llvmcpu-target-triple=aarch64-none-linux-android29 \ --iree-llvmcpu-debug-symbols=false \ --iree-vm-bytecode-module-strip-source-map=true \ --iree-vm-emit-polyglot-zip=false \ "--iree-flow-enable-fuse-padding-into-linalg-consumer-ops" \ "--iree-llvmcpu-enable-pad-consumer-fusion" \ "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" \ --o "${IREE_MODEL_DIR}/llvm-cpu/${MODEL_NAME}_padfuse.vmfb" || true echo -e "\tCompiling ${MODEL_NAME}_mmt4d.vmfb for aarch64..." "${IREE_COMPILE_PATH}" \ --iree-input-type=tosa \ --iree-hal-target-backends=llvm-cpu \ --iree-llvmcpu-target-triple=aarch64-none-linux-android29 \ --iree-opt-data-tiling \ --iree-llvmcpu-target-cpu-features=+dotprod \ "--iree-flow-enable-fuse-padding-into-linalg-consumer-ops" \ "--iree-llvmcpu-enable-pad-consumer-fusion" \ --iree-llvmcpu-debug-symbols=false \ --iree-vm-bytecode-module-strip-source-map=true \ --iree-vm-emit-polyglot-zip=false \ "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" \ --o "${IREE_MODEL_DIR}/llvm-cpu/${MODEL_NAME}_mmt4d.vmfb" || true echo -e "\tCompiling ${MODEL_NAME}_im2col_mmt4d.vmfb for aarch64..." "${IREE_COMPILE_PATH}" \ --iree-input-type=tosa \ --iree-hal-target-backends=llvm-cpu \ --iree-llvmcpu-target-triple=aarch64-none-linux-android29 \ --iree-opt-data-tiling \ --iree-llvmcpu-target-cpu-features=+dotprod \ "--iree-flow-enable-fuse-padding-into-linalg-consumer-ops" \ "--iree-llvmcpu-enable-pad-consumer-fusion" \ --iree-flow-enable-conv-img2col-transform \ --iree-llvmcpu-debug-symbols=false \ --iree-vm-bytecode-module-strip-source-map=true \ --iree-vm-emit-polyglot-zip=false \ "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" \ --o "${IREE_MODEL_DIR}/llvm-cpu/${MODEL_NAME}_im2col_mmt4d.vmfb" || true if [[ "${GPU_TYPE}" = "mali" ]]; then echo -e "\tCompiling ${MODEL_NAME}.vmfb for vulkan mali..." "${IREE_COMPILE_PATH}" \ --iree-input-type=tosa \ --iree-hal-target-backends=vulkan-spirv \ --iree-vulkan-target-triple=valhall-unknown-android31 \ --iree-llvmcpu-debug-symbols=false \ --iree-vm-bytecode-module-strip-source-map=true \ --iree-vm-emit-polyglot-zip=false \ "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" \ --o "${IREE_MODEL_DIR}/vulkan/${MODEL_NAME}.vmfb" || true echo -e "\tCompiling ${MODEL_NAME}_padfuse.vmfb for vulkan mali..." "${IREE_COMPILE_PATH}" \ --iree-input-type=tosa \ --iree-hal-target-backends=vulkan-spirv \ --iree-vulkan-target-triple=valhall-unknown-android31 \ --iree-llvmcpu-debug-symbols=false \ --iree-vm-bytecode-module-strip-source-map=true \ --iree-vm-emit-polyglot-zip=false \ --iree-flow-enable-fuse-padding-into-linalg-consumer-ops \ "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" \ --o "${IREE_MODEL_DIR}/vulkan/${MODEL_NAME}_padfuse.vmfb" || true echo -e "\tCompiling ${MODEL_NAME}_fp16.vmfb for vulkan mali..." "${IREE_COMPILE_PATH}" \ --iree-input-type=tosa \ --iree-hal-target-backends=vulkan-spirv \ --iree-vulkan-target-triple=valhall-unknown-android31 \ --iree-opt-demote-f32-to-f16 \ --iree-llvmcpu-debug-symbols=false \ --iree-vm-bytecode-module-strip-source-map=true \ --iree-vm-emit-polyglot-zip=false \ --iree-flow-enable-fuse-padding-into-linalg-consumer-ops \ "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" \ --o "${IREE_MODEL_DIR}/vulkan/${MODEL_NAME}_fp16.vmfb" || true else echo -e "\tCompiling ${MODEL_NAME}.vmfb for vulkan adreno..." "${IREE_COMPILE_PATH}" \ --iree-input-type=tosa \ --iree-hal-target-backends=vulkan-spirv \ --iree-vulkan-target-triple=adreno-unknown-android31 \ --iree-llvmcpu-debug-symbols=false \ --iree-vm-bytecode-module-strip-source-map=true \ --iree-vm-emit-polyglot-zip=false \ --iree-flow-enable-fuse-padding-into-linalg-consumer-ops \ "${IREE_MODEL_DIR}/${MODEL_NAME}.mlir" \ --o "${IREE_MODEL_DIR}/vulkan/${MODEL_NAME}.vmfb" || true fi done echo -e "\nCross-compile IREE benchmark binary.\n" cd "${SOURCE_DIR}/iree" cmake -GNinja -B ../iree-build/ \ -DCMAKE_INSTALL_PREFIX=../iree-build/install \ -DCMAKE_BUILD_TYPE=RelWithDebInfo \ . cmake --build ../iree-build/ --target install rm -rf ${SOURCE_DIR}/iree-build-android cmake -GNinja -B ../iree-build-android/ \ -DCMAKE_TOOLCHAIN_FILE="${ANDROID_NDK?}/build/cmake/android.toolchain.cmake" \ -DIREE_HOST_BIN_DIR="${PWD}/../iree-build/install/bin" \ -DANDROID_ABI="arm64-v8a" \ -DANDROID_PLATFORM="latest" \ -DIREE_BUILD_COMPILER=OFF \ . cmake --build ../iree-build-android/ cp "${SOURCE_DIR}/iree-build-android/tools/iree-benchmark-module" "${ROOT_DIR}/" # Cross-compile TFLite benchmark binary. sudo apt-get install libgles2-mesa-dev export CC=clang export CXX=clang++ cd "${SOURCE_DIR}" git clone https://github.com/tensorflow/tensorflow.git cd tensorflow echo echo Configuring TensorFlow echo Select defaults. Answer Yes to configuring ./WORKSPACE for Android builds. echo Use Version 21 for Android NDK, 29 for Android SDK. echo python configure.py bazel build -c opt --config=android_arm64 \ --copt="-Wno-error=implicit-function-declaration" \ tensorflow/lite/tools/benchmark:benchmark_model cp "${SOURCE_DIR}/tensorflow/bazel-bin/tensorflow/lite/tools/benchmark/benchmark_model" "${ROOT_DIR}/" echo "Pushing benchmarking artifacts to device." DEVICE_ROOT_DIR=/data/local/tmp/benchmarks adb shell rm -r "${DEVICE_ROOT_DIR}" || true adb push "${ROOT_DIR}" /data/local/tmp adb shell chmod +x "${DEVICE_ROOT_DIR}/benchmark_model" adb shell chmod +x "${DEVICE_ROOT_DIR}/iree-benchmark-module" echo Setup device. adb shell "su root sh ${DEVICE_ROOT_DIR}/setup/set_android_scaling_governor.sh performance" if [[ "${GPU_TYPE}" = "mali" ]]; then adb shell "su root sh ${DEVICE_ROOT_DIR}/setup/set_pixel6_gpu_scaling_policy.sh performance" else adb shell "su root sh ${DEVICE_ROOT_DIR}/setup/set_adreno_gpu_scaling_policy.sh performance" fi echo Running benchmark. adb push "${SOURCE_DIR}/iree/build_tools/benchmarks/comparisons" /data/local/tmp/ adb shell "su root /data/data/com.termux/files/usr/bin/python /data/local/tmp/comparisons/run_benchmarks.py --device_name=Pixel6 --mode=mobile --base_dir=${DEVICE_ROOT_DIR} --output_dir=${DEVICE_ROOT_DIR}/output" adb shell cat "${DEVICE_ROOT_DIR}/output/results.csv" | tee ${ROOT_DIR}/output/results.csv