mbox series

[meta-tensorflow,master,0/14] Upgrade tensorflow from 2.19.0 to 2.21.0

Message ID 20260325021719.1526525-1-hongxu.jia@windriver.com
Headers show
Series Upgrade tensorflow from 2.19.0 to 2.21.0 | expand

Message

Hongxu Jia March 25, 2026, 2:17 a.m. UTC 
Summary: Upgrade meta-tensorflow from 2.19.0 to 2.21.0

* Git logs
[meta-tensorflow]
commit a4878fbdbaf0b0519b74e33b39c50763c7f4cdc0
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Tue Mar 24 17:54:08 2026 +0800

    tensorflow: remove obsolete patches
    
    They are not required any more
    - 0001-fix-compile-XNNPACK-failed-for-aarch64.patch
    - 0001-fix-compile-failed.patch
    - 0001-tensorflow-BUILD-fix-build-failure-for-aarch64.patch
    - 0001-third_party-ducc-fix-ambiguous-failure.patch
    - 0002-third_party-eigen_archive-workaround-ice-failure-whi.patch
    - 0004-drop-v2-config.patch
    - 0006-fix-build-failure.patch
    - 0007-support-hermetic-Python.patch
    - 0009-Fix-build-with-gcc-12.patch
    - 0012-disable-avxvnni-and-avx512fp16-for-x86.patch
    - 0013-fix-build-tensorflow-lite-examples-label_image-label.patch
    - 0016-CheckFeatureOrDie-use-warning-to-avoid-die.patch
    - 0017-support-32-bit-x64-and-arm-for-yocto.patch
    - 0019-fix-build-failure-for-2.19.patch
    - 0020-tensorflow-compiler-mlir-lite-fix-tensorflow_lite_qu.patch
    - 0001-Add-hermetic-PYTHON-3.13-requirements-lock-file-in-T.patch
    - 0001-support-gcc-15.patch
    - 0001-support-python-3.33.patch
    - 0010-fix-patchelf-not-found.patch
    - 0011-fix-llvm-project-timeout-occasionally.patch
    - 0021-build_pip_package_with_bazel.sh-correct-version.patch
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit b6d6729635d9d41695e85a468e4bc3c986b4a202
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Tue Mar 24 17:58:36 2026 +0800

    tensorflow: refresh patches for 2.21.0
    
    - 0001-third_party-tf_runtime-fix-compile-failure.patch
    - 0008-make-hermetic-python3-support-offline-build.patch
    - 0012-add-yocto-toolchain-to-support-cross-compiling.patch
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit 4c1a1a2e3dbf488e7f766bf0a3352d020008b274
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Wed Mar 11 15:28:02 2026 +0800

    tensorflow/tensorflow-native/tensorflow-lite: upgrade 2.19.0 -> 2.21.0
    
    1. Support python 3.14
    - 0001-Add-support-for-Python-3.14-to-TensorFlow-build-conf.patch
    - 0001-Update-grpcio-to-1.78.0-for-Python-3.14.patch
    - 0001-correct-h5py-version-for-python-3.14.patch
    - 0001-fix-pypi_keras-is-not-defined.patch
    
    2. Swith toolchain to clang
    - BUILD.yocto_compiler
    - cc_config.bzl.tpl
    - Clean up gcc specific CFALGS and patches
    
    3. Migrating to Platforms [1]
    - 0001-set-x83_64-to-arch-of-whl.patch
    - 0001-support-cross-compile-for-target-aarch64.patch
    - BUILD.in
    
    [1] https://bazel.build/concepts/platforms
    
    4. Remove TF_ENABLE_XLA=0 otherwise `import tensorflow' failed
    [snip]
    $ python3 -c "import tensorflow"
    ...
    |    from tensorflow.python.platform import _pywrap_cpu_feature_guard
    |ImportError: .venv/lib/python3.13/site-packages/tensorflow/python/
    platform/../lib_pywrap_tensorflow_common.so: undefined symbol: _ZN4mlir3sdy18TensorShardingAttr13getClosedLikeES1_
    [snip]
    
    5. According to [2], revert to use external Yocto toolchains
    - 0001-use-external-toolchains.patch
    
    [2] https://github.com/tensorflow/tensorflow/commit/ac56b0e840919cea88ed847f364468c8332b5333
    
    6. Fix tensorflow-lite compile failed
    - 0001-do-not-use-pywrap-rules.patch
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit 5da554a2858b5d8d98456cb3ba2dc3876f846696
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Tue Mar 24 12:14:27 2026 +0800

    classes/unsupportarch.bbclass: update CPUs
    
    Do not support 32bit arm and x86
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit ab69230eec15d948f9de502286ba1e0cd12db78c
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Thu Mar 12 19:15:12 2026 +0800

    Switch tf-keras to python3-keras 3.13.2
    
    Add LICENSE from https://github.com/keras-team/keras/blob/master/LICENSE
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit 98bd429302d29c0b489ded70666a84f4ce3d02e9
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Thu Mar 12 12:54:48 2026 +0800

    bazel-native: fix compile failure
    
    According to [1][2], fix compile bazel failed
    
    [1] https://github.com/bazelbuild/bazel/issues/27349
    [2] https://github.com/bazelbuild/bazel/issues/26162
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit bc22a52f34d5e72d8cf3ea52ce75c7db39ac55dd
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Thu Mar 12 12:11:56 2026 +0800

    Upgrade openjdk-11 to openjdk-21
    
    The bazel 7.7.0 requires JDK 21 [1], upgrade openjdk-11 to openjdk-21
    
    [1] https://github.com/bazelbuild/bazel/commit/7a5c1fafac4971f9dc8e9a14dd3f5fee545e3877
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit eae659814fec9b43b62a1a1284f316ee702ad1fc
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Wed Mar 11 16:58:20 2026 +0800

    bazel-native: upgrade 6.0.0 -> 7.7.0
    
    Tensorflow 2.21 requires bazel 7.7.0
    
    Drop obsolete patches:
    - 0001-fix-compile-failure-on-gcc-13.patch
    - 1f2b3ed4ac717b814d02a0d125f160ddabe78003.patch
    - include_stdint.patch
    
    Refresh patches to 7.7.0
    - 0001-add-Yocto-native-sysroot-dir-to-the-default-Bazel-to.patch
    - 0001-cutsomize-native-toolchains-for-Yocto.patch
    - 0001-fix-unzip-command-not-found.patch
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit b6215d27018de635930bb890bab581ac1032df16
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Tue Mar 24 15:53:04 2026 +0800

    python3-opt-einsum: 3.3.0 -> 3.4.0
    
    Swtich to python_hatchling
    
    Remove obsolete patch
    - 0001-Use-ConfigParser-instead-of-SafeConfigParser.patch
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit a57d71ff251297e4549cb91ac7ff3a36d9ffd5c3
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Tue Mar 24 15:40:55 2026 +0800

    python3-mldtypes: 0.5.1 -> 0.5.4
    
    Remove obsolete patch
    - 0001-fix-setuptools-missing.patch
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit de859d2fabcdc5e68f60a313c5335282f80f8935
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Tue Mar 24 13:13:06 2026 +0800

    python3-optree: add recipe 0.19.0
    
    It is required by tensorflow and keras
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit 5a8e06ae0ac9233c79d274fc28ad04852b2e17e8
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Wed Mar 18 10:30:12 2026 +0800

    classes/bazel.bbclass: multiple fixes
    
    - Make run `rm -rf ${BAZEL_DIR}' could work
    
    - Remove znoexecstack
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit a6bb8260c0cbea533492bb45ba51b35dd29ef7db
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Tue Mar 17 16:09:32 2026 +0800

    tensorflow-estimator: remove recipe
    
    Due to commit [1], remove Estimator from Tensorflow.
    
    [1] https://github.com/tensorflow/tensorflow/commit/aa35dc27617676322972f2fd482ed43a7a7e1d16
    
    Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>

commit 32c443afe8f81273b89d0df97ce5f3a4d78a9b69
Author: Hongxu Jia <hongxu.jia@windriver.com>
Date:   Wed Mar 11 15:44:05 2026 +0800

    Revert "Add python3 3.13.11 back"
    
    This reverts commit 3fa809419857160ffa23762adbee2f20f9f9fbce.

Comments

Martin Jansa March 25, 2026, 7:34 a.m. UTC | #1 
On Wed, Mar 25, 2026 at 3:17 AM hongxu via lists.yoctoproject.org
<hongxu.jia=windriver.com@lists.yoctoproject.org> wrote:
>
> Summary: Upgrade meta-tensorflow from 2.19.0 to 2.21.0

Thanks you, can you please push it to master-next or somewhere else to
make it easier to test these changes in our CI as well? Or what's ETA
for merging this to master (as it cannot be used with latest oe-core
without wrynose in compatibility list)?