RISC-V で TensorFlow Lite を動かす。
November RISC-V Bay Area Meetup hosted by Antmicroにて、
引用 "TensorFlow Lite on RISC-V" (Pete Warden - Google) Many emerging new microcontrollers use RISC-V, and the TensorFlow Lite team has been exploring how to best support these new platforms. In this talk, Pete Warden, Lead of the TensorFlow Mobile/Embedded team at Google will discuss how they've used Renode, gcc, and other open-source technologies to target RISC-V for machine learning.
おー、聴いてみたい。。。遠いけどね。
TensorFlow Liteには特別な仕組みは無くて、Makefileを作るだけでできるんだよね。
実は、r1.11 から入っているんだよね。ここ
riscv_makefile.inc
# Settings for RiscV platforms. ifeq ($(TARGET), riscv) TARGET_ARCH := riscv TARGET_TOOLCHAIN_PREFIX := riscv32-unknown-elf- #CXXFLAGS += -march=gap8 CXXFLAGS += -DTFLITE_MCU LIBS += -ldl BUILD_TYPE := micro endif じゃー、どうやって、Build するかというと、https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/lite/g3doc/rpi.md RasPiと同じ。 https://github.com/tensorflow/tensorflow/tree/r1.11/tensorflow/contrib/lite/tools/make ここにあるbuild_rpi_lib.shをベースに以下のように変更するだけよ。 {{{ set -e SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" cd "$SCRIPT_DIR/../../../../.." CC_PREFIX=arm-linux-gnueabihf- make -j 3 -f tensorflow/contrib/lite/tools/make/Makefile TARGET=riscv TARGET_ARCH=riscvl