Raspberry Pie 4

[shijunti19]

Raspberry Pie 4

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device:
• TensorFlow installed from (source or binary):
• TensorFlow version:
• Python version:
• Installed using virtualenv? pip? conda?:
• Bazel version (if compiling from source):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version:
• GPU model and memory:

Describe the problem
pi@raspberrypi:/xiaolu/u/soft/tensorflow $ sudo ./tensorflow/lite/tools/make/build_rpi_lib.sh

• set -e
  +++ dirname ./tensorflow/lite/tools/make/build_rpi_lib.sh
  ++ cd ./tensorflow/lite/tools/make
  ++ pwd
• SCRIPT_DIR=/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make
• TENSORFLOW_DIR=/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/../../../..
• make -j 4 TARGET=rpi -C /xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/../../../.. -f tensorflow/lite/tools/make/Makefile -latomic
  make: 进入目录“/xiaolu/u/soft/tensorflow”
  arm-linux-gnueabihf-g++ -O3 -DNDEBUG -fPIC --std=c++11 -march=armv7-a -mfpu=neon-vfpv4 -funsafe-math-optimizations -ftree-vectorize -fPIC -I. -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/../../../../../ -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/../../../../../../ -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/downloads/ -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/downloads/eigen -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/downloads/absl -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/downloads/gemmlowp -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/downloads/neon_2_sse -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/downloads/farmhash/src -I/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/downloads/flatbuffers/include -I -I/usr/local/include -c tensorflow/lite/experimental/ruy/kernel_arm32.cc -o /xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/gen/rpi_armv7l/obj/tensorflow/lite/experimental/ruy/kernel_arm32.o
  /tmp/ccIPb9CY.s: Assembler messages:
  /tmp/ccIPb9CY.s:451: 错误：Neon quad precision register expected -- `vld1.32 q11,[r2]'
  make: *** [tensorflow/lite/tools/make/Makefile:244：/xiaolu/u/soft/tensorflow/tensorflow/lite/tools/make/gen/rpi_armv7l/obj/tensorflow/lite/experimental/ruy/kernel_arm32.o] 错误 1
  make: 离开目录“/xiaolu/u/soft/tensorflow”

Provide the exact sequence of commands / steps that you executed before running into the problem

Any other info / logs
Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.

[shijunti19]

I have the same problem with cross-compiling with Ubuntu on Windows

/tmp/ccuFx1oC.s: Assembler messages:
/tmp/ccuFx1oC.s:455: Error: Neon quad precision register expected -- `vld1.32 q11,[r2]'
tensorflow/lite/tools/make/Makefile:244: recipe for target '/mnt/z/linux/tensorflow/tensorflow/lite/tools/make/gen/rpi_armv7l/obj/tensorflow/lite/experimental/ruy/kernel_arm32.o' failed
make: *** [/mnt/z/linux/tensorflow/tensorflow/lite/tools/make/gen/rpi_armv7l/obj/tensorflow/lite/experimental/ruy/kernel_arm32.o] Error 1

[paulbauriegel]

Having the same problem under Ubuntu 19.04 when cross-compiling with Docker:

Dockerfile

FROM tensorflow/tensorflow:nightly-devel

USER root

RUN apt-get update && \
    apt-get install -y --no-install-recommends \
    build-essential \
    crossbuild-essential-armhf \
    git

RUN git clone https://github.com/tensorflow/tensorflow

RUN ./tensorflow/tensorflow/lite/tools/make/download_dependencies.sh

WORKDIR ./tensorflow
RUN ./tensorflow/lite/tools/make/build_rpi_lib.sh

[wangtz]

Looks like Ruy was using something unsupported on RasberryPi.

[bjacob]

It's not a hardware issue, it's an assembler syntax issue. One line of asm is sloppy and had been accidentally accepted by the android toolchain, but the RPi toolchain is being more rigorous here. The fix should be below. We're submitting this internally so you can expect this to be working soon, but always interesting to hear a confirmation.

diff --git a/tensorflow/lite/experimental/ruy/kernel_arm32.cc b/tensorflow/lite/experimental/ruy/kernel_arm32.cc
--- a/tensorflow/lite/experimental/ruy/kernel_arm32.cc
+++ b/tensorflow/lite/experimental/ruy/kernel_arm32.cc
@@ -947,7 +947,7 @@ void Kernel8bitNeonOutOfOrder(const Kern
         "ldr r5, [%[params], #" RUY_STR(RUY_OFFSET_RHS_ZERO_POINT) "]\n"
 
         // Load 4 lhs_sums values.
-        "vld1.32 q11, [r2]\n"
+        "vld1.32 {d22, d23}, [r2]\n"
         "vdup.32 d13, r5\n" // rhs_zero_point
 
         // Compute lhs_sums * rhs_zero_point.

[paulbauriegel]

@bjacob Thanks for the fix. I can confirm that the patch works for my setup.