This should help us identify exactly how we're building our releases.
PiperOrigin-RevId: 380829893
Change-Id: I279b789351504d6d6d9bafcac40d3c3a6174db5f
This should help us identify exactly how we're building our releases.
PiperOrigin-RevId: 380700058
Change-Id: I21cf9eef338dfdda468afc232d9ac2dbd673637f
Before, the build would fail with errors such as: "relocation truncated to fit: R_X86_64_32 against .debug_info'". The issue was the debug info was too large. I believe the issue was occurring because offsets into the .debug_info section are stored as 32-bit integers, and so that section cannot exceed 4GiB. To fix, debug info is only included for files under tensorflow/, excluding kernels. This brings the size of the .debug_info section down to about 1.4GiB, well under the 4GiB limit.
Unfortunately, TF kernels and TF dependencies do not have debugging info anymore, but I suspect these are rarely debugged. Debugging info for specific kernels/dependencies can still be explicitly included by the user, e.g. by passing the bazel flags: --config=dbg --per_file_copt=+tensorflow/core/kernels/identity_op.*@-g
See https://github.com/tensorflow/tensorflow/issues/48919 for more context.
PiperOrigin-RevId: 378910826
Change-Id: I4b94e3d53bb3ca00c30d5c83d2a57e4bd390c5a8
Currently we can disable entire test targets via the bazel tags `noasan`,
`nomsan`, `notsan`, `noubsan`. However, in some instances, we may expect
one particular edge-case to fail a sanitizer test (for known reason), but
still want the rest of the test suite to be analyzed. By disabling individual
test methods or parts thereof locally, we allow for this without the need to
separate the offending case into a different target.
There already was some functionality in tflite, which seems to currently
be unused (though has been previously to locally skip tests).
Refactored and moved to tf so can be more widely used.
Also added corresponding defines to the `.bazelrc` file.
PiperOrigin-RevId: 374231232
Change-Id: I62ca36ebe9edec18e0d6d358496b6d07efb43574
A roll-forward of cl/370462021, but only on Windows.
This avoids creating runfiles tree on the local machine, which is very slow.
PiperOrigin-RevId: 370679217
Change-Id: I11f331d236575e408ceaa3b46d3830010177b386
This avoids downloading remote build outputs to local machine, which can speed up build and test time, especially on Windows.
PiperOrigin-RevId: 370574647
Change-Id: Ied7b9c4df06165d0543b71bcb1e7c94219fd3d43
On Windows, py_binary & py_test are by default built as self-extracting archives of their runfiles tree. It is very time consuming, but we don't need that
for the RBE build because the runfiles tree is created in the remote worker anyway.
This significantly speeds up TF's python tests in the Windows RBE build.
Related doc: https://bazel.build/designs/2016/09/05/build-python-on-windows.html
PiperOrigin-RevId: 370464199
Change-Id: Ic91fea41f1fc1b0f5de3b6aac4d054cc43536ebf
This avoids downloading remote build outputs to local machine, which can speed up build and test time, especially on Windows.
PiperOrigin-RevId: 370462021
Change-Id: I5531d8a5e5e517373d74f8528e28472b199e6661
With --experimental_strict_action_env, CUDA DLLs are not in PATH, tests cannot load them, and GPU tests fall back to CPU.
--experimental_strict_action_env became the default for Windows builds in cl/283405622. This change restricts it to Windows RBE builds, which is CPU only.
PiperOrigin-RevId: 365530221
Change-Id: I09684440e7c64fc9151694da5758faa4ecf3dd7b
Implicit options are evaluated first no matter where they are. Following https://blog.bazel.build/2018/01/19/config-parsing-order.html, the .bazelrc should ideally be ordered by evaluation order.
PiperOrigin-RevId: 364409118
Change-Id: Iec71a709dd0c40b7de196b16bb980d247d21fdee
This avoid double-expansion of --config=release_base for --config=release_gpu_{linux,windows}.
PiperOrigin-RevId: 364358564
Change-Id: I20adf018cfaf7f02e9fdfe79dba252116f05840b
This is a preparation change to remove running ./configure in release build scripts.
PiperOrigin-RevId: 363624089
Change-Id: I88a67df4e784c489539d928641f9eda574ad1dff
This time, don't move `--enable_platform_specific_config` because it affects the expanded flag order.
PiperOrigin-RevId: 363584510
Change-Id: I2290305f280f71e469fea9c64307d59b5cc336a5
Previously, XLA was enabled by default with --config=rbe_linux and when running ./configure.
PiperOrigin-RevId: 363433913
Change-Id: I9afa53496a9da2ae29b71a9ef38fc63b90dfba02
Use --config=avx_{linux,windows} instead of --config=opt in release builds. This is an NFC because release build scripts set build:opt to enable AVX.
PiperOrigin-RevId: 363193986
Change-Id: Ifb0788e29485873e35469d0936c1d33488315870
This PR introduces the support of Compute Library for the Arm
Architecture (ACL) as part of Bazel build.
Signed-off-by: Aleksandr Nikolaev <aleksandr.nikolaev@arm.com>
