MacBook Pro (Retina, Mid 2012) - Catalina 10.15.7

[pilipicus]

OS Platform and Distribution

Mac Pro Catalina 10.15.7

Compiler version

No response

Programming Language and version

C++

Installed using virtualenv? pip? Conda?(if python)

No response

MediaPipe version

0.10.19

Bazel version

No response

XCode and Tulsi versions (if iOS)

Version 12.4

Android SDK and NDK versions (if android)

No response

Android AAR (if android)

None

OpenCV version (if running on desktop)

4.10.0

Describe the problem

The provided examples do not run

Complete Logs

My configuration: 
Bazelisk version: 1.24.0
Mediapipe version: 0.10.19
opencv_version: 4.10.0
Xcode version: 1.24.0
clang version: Apple clang version 12.0.0 (clang-1200.0.32.29)
Target: x86_64-apple-darwin19.6.0
Thread model: posix

I've followed the guide for MacOS: https://ai.google.dev/edge/mediapipe/framework/getting_started/install.md#installing_on_macos  and downloaded from git: https://github.com/google/mediapipe.git the examples.


The "HelloWorld" example works. Instead all the others don't! 

To run "HelloWorld" I do this:
$ export GLOG_logtostderr=1

# if you are running on Linux desktop with CPU only
$ bazel run --define MEDIAPIPE_DISABLE_GPU=1 \
    mediapipe/examples/desktop/hello_world:hello_world

I0000 00:00:1732381130.229285 8588331 hello_world.cc:58] Hello World!
I0000 00:00:1732381130.229467 8588331 hello_world.cc:58] Hello World!
...


To run "hand_tracking", that doesn't work at the I do this:

bazel build -c opt --define MEDIAPIPE_DISABLE_GPU=1 mediapipe/examples/desktop/hand_tracking:hand_tracking_cpu --verbose_failures --sandbox_debug

Th error that I receive is: 
clang: error: unknown argument: '-mavxvnni'
Error in child process '/usr/bin/xcrun'. 1
Target //mediapipe/examples/desktop/hand_tracking:hand_tracking_cpu failed to build
INFO: Elapsed time: 36.360s, Critical Path: 35.94s
INFO: 212 processes: 128 internal, 84 darwin-sandbox.
FAILED: Build did NOT complete successfully

[kalyan2789g]

Hi @pilipicus,

Could you please confirm the latest version of the OpenCv in your system? Currently Mediapipe supports OpenCv v4.

Thanks,
@kalyan2789g

[pilipicus]

Hi @kalyan2789g,

Before it was the OpenCV version 4.10.0. Now, following the guide online (link), but also looking at what was written in the WORSPACE file in the mediapipe folder, I've installed the version of OpenCV 3.4.11.

The error that I've receive is:
ERROR: /private/var/tmp/_bazel_Filippo/a66de5a5c3e5be3ced057a7a2c8bed2c/external/XNNPACK/BUILD.bazel:661:36: Compiling src/qs8-rsum/gen/qs8-rsum-avxvnni-u128-acc2.c failed: (Exit 1): wrapped_clang failed: error executing command (from target @XNNPACK//:avxvnni_prod_microkernels) external/local_config_cc/wrapped_clang '-D_FORTIFY_SOURCE=1' -fstack-protector -fcolor-diagnostics -Wall -Wthread-safety -Wself-assign -fno-omit-frame-pointer -g0 -O2 -DNDEBUG '-DNS_BLOCK_ASSERTIONS=1' ... (remaining 108 arguments skipped)

Use --sandbox_debug to see verbose messages from the sandbox and retain the sandbox build root for debugging
clang: error: unknown argument: '-mavxvnni'
Error in child process '/usr/bin/xcrun'. 1
Target //mediapipe/examples/desktop/hand_tracking:hand_tracking_cpu failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 37.144s, Critical Path: 36.79s
INFO: 207 processes: 129 internal, 78 darwin-sandbox.
FAILED: Build did NOT complete successfully

I've also tried to disable AVX-512 modyifing the file .bazelrc , in the root of the MediaPipe project:
build --copt=-DXNN_ENABLE_AVX512FP16=0
build --copt=-DXNN_ENABLE_AVXVNNI=0
build --copt=-DXNN_ENABLE_AVXVNNIINT8=0
build --copt=-DXNN_ENABLE_AVX512AMX=0
build --copt=-mno-avx512f
build --copt=-mno-avx512cd
build --copt=-mno-avx512bw
build --copt=-mno-avx512dq
build --copt=-mno-avx512vl
build --copt=-mno-avx512vnni
build --copt=-mno-gfni
build --copt=-mno-avx512fp16

[kuaashish]

Hi @pilipicus,

Could you please execute the following command to disable "avxvnni" as below and update Clang to version 18.0?

echo "build --define=xnn_enable_avxvnni=false" >> .bazelrc

Thank you!!

[pilipicus]

Hi @kalyan2789g,

I've installed the version 18.1 because the 18.0 seems not being available or I couldn't find it:

clang --version

clang version 18.1.0rc
Target: x86_64-apple-darwin19.6.0
Thread model: posix
InstalledDir: /usr/local/llvm-18.1.0/bin

After, I moved to the folder mediapipe, I run the command echo "build --define=xnn_enable_avxvnni=false" >> .bazelrc and I run again the command bazel build -c opt --define MEDIAPIPE_DISABLE_GPU=1 mediapipe/examples/desktop/hand_tracking:hand_tracking_cpu --verbose_failures --sandbox_debug

The error that I receive Is still the same:
error: unknown argument: '-mavx512fp16'
Error in child process '/usr/bin/xcrun'. 1
Target //mediapipe/examples/desktop/hand_tracking:hand_tracking_cpu failed to build

Here the screenshot of the error:

[kuaashish]

Hi @pilipicus,

This appears to be a different error. Could you try applying a similar approach(Clang 18.1 fine) for that flag using the following command?

echo "build --define=xnn_enable_avx512fp16=false" >> .bazelrc

[pilipicus]

Hi @kalyan2789g,

I've tried your command line and the error that I get is the following one:

[kuaashish]

@pilipicus,

Please follow the instructions below to disable all flags. Unfortunately, we do not have an older macOS version to reproduce the issue and help with disabling the flags.

clang: error: the clang compiler does not support '-mavxvnniint8’'

• Step 1: Search for the flag in the XNNPack repo:
  https://github.com/search?q=repo%3Agoogle%2FXNNPACK++mavxvnniint8&type=code

• Step 2: Go to result and ding condition name:

  https://github.com/google/XNNPACK/blob/fc3c00d97d135650b6b9e268c9ddd141f053dcb5/build_params.bzl#L568
  Name is avxvnniint8_enabled

• Step 3: Search for avxvnniint8_enabled

  https://github.com/search?q=repo%3Agoogle%2FXNNPACK%20avxvnniint8_enabled&type=code

• Go to result in BUILD.bazel:

  https://github.com/google/XNNPACK/blob/fc3c00d97d135650b6b9e268c9ddd141f053dcb5/BUILD.bazel#L1797

  Find condition name and search in this file: xnn_enable_avxvnniint8_explicit_true

• Go to config name:

  https://github.com/google/XNNPACK/blob/fc3c00d97d135650b6b9e268c9ddd141f053dcb5/BUILD.bazel#L1313

  config_setting(
  name = "xnn_enable_avxvnniint8_explicit_true",
  define_values = {"xnn_enable_avxvnniint8": "true"},
  )

This is the flag. So you can disable this by passing –define=xnn_enable_avxvnniint8=false by same command.