.bazelrc

############################################################################
# All default build options below.

# Make Bazel print out all options from rc files.
build --announce_rc

# TODO(goranpetrovic): figure out visibility of tensorflow libraries.
build --nocheck_visibility

build --enable_platform_specific_config

build --experimental_cc_shared_library

# Disable enabled-by-default TensorFlow features that we don't care about.
build --define=no_aws_support=true
build --define=no_hdfs_support=true
build --define=no_hdfs_support=true
build --define=no_kafka_support=true
build --define=no_ignite_support=true

build --define=grpc_no_ares=true

build -c opt

build --config=short_logs

# PyTorch/XLA uses exceptions to communicate with Python.
build --copt=-fexceptions

# Force GCC because clang/bazel has issues.
build --action_env=CC=gcc
build --action_env=CXX=g++
build --spawn_strategy=standalone

# TODO(https://github.com/pytorch/xla/issues/9336): Enable bzlmod.
build --noenable_bzlmod

###########################################################################

build:clang --action_env=CC=/usr/bin/clang-17
build:clang --action_env=CXX=/usr/bin/clang++-17

# clang requires the sandbox mode. Without this, `bazel build` generates an
# error like:
#
# ERROR: .../external/llvm-project/llvm/BUILD.bazel:2006:11: Compiling
# llvm/lib/CodeGenTypes/LowLevelType.cpp failed: undeclared inclusion(s) in
# rule '@llvm-project//llvm:CodeGenTypes':
# this rule is missing dependency declarations for the following files
# included by 'llvm/lib/CodeGenTypes/LowLevelType.cpp':
#   'bazel-out/k8-opt/bin/external/llvm-project/llvm/config.cppmap'
#   'bazel-out/k8-opt/bin/external/llvm-project/llvm/Demangle.cppmap'
#
# The sandbox mode requires all source files to be readable by others, as
# inside the docker we build PyTorch/XLA as root, not the original user who
# created the files. You can ensure this by either:
# 1. `umask 022` before running the `git clone` commands, or
# 2. after running the `git clone` commands, go to the workspace directory
#    and run `chmod -R o+rX .`. The X (capital X) is important: it gives
#    execute permission to directories, and to files only if they already have
#    execute permission for the owner (or group or others). This is generally
#    safer than o+rx.
build:clang --spawn_strategy=sandboxed

###########################################################################

# Make all symbols hidden, by default.
# Ref: https://github.com/pytorch/xla/pull/9693
#
# Without this flag, GCC was associating some variables as "global unique".
# Meaning that the linker would make sure that there was only one instance
# of that variable per process.
#
# Which Variables?
# ================
# - Function-local static variables
# - Inside an inline function (e.g. class member functions)
# - Inside an non-internal linkage context
#
# One example is: `protobuf` holds [a `ShutdownData`][1] which is function-local
# and static, living inside an inline function, which lives inside a class with
# external linkage.
#
# What Happens?
# =============
# [GCC marks those variables with STB_GNU_UNIQUE binding][2], making sure those
# variables are unique for the entire process.
#
# Alternatives
# ============
# There is also the `--no-gnu-unique` option, which disables exactly that.
# However, I believe this is a good opportunity to make everything hidden, unless
# otherwise specified.
#
# [1]: https://github.com/protocolbuffers/protobuf/blob/13fe37f25f187b7a2e79faa962df6a69bdb3d5b8/src/google/protobuf/message_lite.cc#L776
# [2]: https://gcc.gnu.org/onlinedocs/gcc/Code-Gen-Options.html?utm_source=chatgpt.com#index-fno-gnu-unique
build:posix --copt=-fvisibility=hidden

build:posix --copt=-Wno-sign-compare
build:posix --cxxopt=-std=c++17
build:posix --host_cxxopt=-std=c++17

build:avx_posix --copt=-mavx
build:avx_posix --host_copt=-mavx

build:avx_linux --copt=-mavx
build:avx_linux --host_copt=-mavx

build:native_arch_posix --copt=-march=native
build:native_arch_posix --host_copt=-march=native

build:mkl_open_source_only --define=tensorflow_mkldnn_contraction_kernel=1

build:acl --define==build_with_acl=true

build:nonccl --define=no_nccl_support=true

build:linux --config=posix

# Suppress all warning messages.
build:short_logs --output_filter=DONT_MATCH_ANYTHING

#build:tpu --@xla//xla/python:enable_tpu=true
build:tpu --define=with_tpu_support=true

# Run tests serially with TPU (only 1 device is available).
test:tpu --local_test_jobs=1

#########################################################################
# RBE config options below.
# Flag to enable remote config
common --experimental_repo_remote_exec

# Inherit environmental variables that are used in testing.
test --test_env=TPU_NUM_DEVICES --test_env=CPU_NUM_DEVICES --test_env=XRT_LOCAL_WORKER
test --test_env=XRT_TPU_CONFIG --test_env=XRT_DEVICE_MAP --test_env=XRT_WORKERS --test_env=XRT_MESH_SERVICE_ADDRESS
test --test_env=XRT_SHARD_WORLD_SIZE --test_env=XRT_MULTI_PROCESSING_DEVICE --test_env=XRT_HOST_ORDINAL --test_env=XRT_SHARD_ORDINAL
test --test_env=XRT_START_LOCAL_SERVER --test_env=TPUVM_MODE --test_env=PJRT_DEVICE --test_env=PJRT_TPU_MAX_INFLIGHT_COMPUTATIONS
test --test_env=PJRT_CPU_ASYNC_CLIENT --test_env=TPU_LIBRARY_PATH --test_env=PJRT_DIST_SERVICE_ADDR
test --test_env=PJRT_LOCAL_PROCESS_RANK

# This environmental variable is important for properly integrating with XLA.
test --test_env=XLA_EXPERIMENTAL

# To find `libpython` that is required to run tests (they run using installed wheels).
test --test_env=LD_LIBRARY_PATH

# This fixes an issue where targets are configured differently because of `test_filter`.
# https://github.com/bazelbuild/bazel/issues/6842
test --notrim_test_configuration

# Stabilize the environmental variables used to minimize cache misses (src and env affects cache keys).
build --incompatible_strict_action_env

# By default in local builds, do not upload local results to cache.
build --noremote_upload_local_results

# Remote caching with local builds.
build:remote_cache --remote_cache=grpcs://remotebuildexecution.googleapis.com
build:remote_cache --remote_instance_name=projects/tpu-pytorch/instances/default_instance
build:remote_cache --google_default_credentials
build:remote_cache --remote_upload_local_results
build:remote_cache --bes_backend=buildeventservice.googleapis.com
build:remote_cache --bes_upload_mode=fully_async
build:remote_cache --bes_results_url="https://source.cloud.google.com/results/invocations"
build:remote_cache --bes_instance_name="tpu-pytorch"
build:remote_cache --bes_timeout=600s  # On longer builds, BES can cause a non-zero exit from bazel.

# Attempt to minimize the amount of data transfer between bazel and the remote
# workers:
build:remote_cache --remote_download_toplevel
#########################################################################

# Load rc file with user-specific options.
try-import %workspace%/.bazelrc.user

# Compile database generation config.
build:compdb --features=-layering_check

# Compiling tests requires Java.
build --java_runtime_version=remotejdk_11

# Coverage setup.
coverage --build_tests_only
coverage --config=coverage
coverage --instrumentation_filter="//torch_xla[/:],//third_party[/:],-//test[/:]"
coverage --combined_report=lcov
coverage --nocache_test_results
build:coverage --strategy=CoverageReport=sandboxed,local

build:coverage --test_tag_filters=-nocoverage

############################################################################
############## TensorFlow .bazelrc greatest hits ###########################
############################################################################

# Modular TF build options
build:dynamic_kernels --define=dynamic_loaded_kernels=true
build:dynamic_kernels --copt=-DAUTOLOAD_DYNAMIC_KERNELS
build --define=tf_api_version=2 --action_env=TF2_BEHAVIOR=1

# Default paths for TF_SYSTEM_LIBS
build:linux --define=PREFIX=/usr
build:linux --define=LIBDIR=$(PREFIX)/lib
build:linux --define=INCLUDEDIR=$(PREFIX)/include
build:linux --define=PROTOBUF_INCLUDE_PATH=$(PREFIX)/include

build:linux --define=build_with_onednn_v2=true

# On linux, we dynamically link small amount of kernels
build:linux --config=dynamic_kernels

# For projects which use TensorFlow as part of a Bazel build process, putting
# nothing in a bazelrc will default to a monolithic build. Here we force
# the monolitih build because otherwise there are missing dependencies and
# linking fails.
build --define framework_shared_object=false
build --define tsl_protobuf_header_only=false

build --define=use_fast_cpp_protos=true
build --define=allow_oversize_protos=true

# Enable XLA support by default.
build --define=with_xla_support=true

# Disable some xnnpack compilation flags which are not supported before gcc-13.
build:linux --define=xnn_enable_avxvnni=false
build:linux --define=xnn_enable_avx256vnni=false
build:linux --define=xnn_enable_avxvnniint8=false
build:linux --define=xnn_enable_avx512amx=false
build:linux --define=xnn_enable_avx512fp16=false

# See https://github.com/bazelbuild/bazel/issues/7362 for information on what
# --incompatible_remove_legacy_whole_archive flag does.
# This flag is set to true in Bazel 1.0 and newer versions. We tried to migrate
# Tensorflow to the default, however test coverage wasn't enough to catch the
# errors.
# There is ongoing work on Bazel team's side to provide support for transitive
# shared libraries. As part of migrating to transitive shared libraries, we
# hope to provide a better mechanism for control over symbol exporting, and
# then tackle this issue again.
#
# TODO: Remove this line once TF doesn't depend on Bazel wrapping all library
# archives in -whole_archive -no_whole_archive.
build --noincompatible_remove_legacy_whole_archive

# cc_shared_library ensures no library is linked statically more than once.
build --experimental_link_static_libraries_once=false

# Do not risk cache corruption. See:
# https://github.com/bazelbuild/bazel/issues/3360
build:linux --experimental_guard_against_concurrent_changes

# Prevent regressions on those two incompatible changes
# TODO: remove those flags when they are flipped in the default Bazel version TF uses.
build --incompatible_enforce_config_setting_visibility

# Suppress most C++ complier warnings to reduce log size but allow
# for specific warnings to still be present.
build:linux --copt="-Wno-all"
build:linux --copt="-Wno-extra"
build:linux --copt="-Wno-deprecated"
build:linux --copt="-Wno-deprecated-declarations"
build:linux --copt="-Wno-ignored-attributes"
build:linux --copt="-Wno-array-bounds"
# Add unused-result as an error on Linux.
build:linux --copt="-Wunused-result"
build:linux --copt="-Werror=unused-result"
# Add switch as an error on Linux.
build:linux --copt="-Wswitch"
build:linux --copt="-Werror=switch"
# Required for building with clang
build:linux --copt="-Wno-error=unused-but-set-variable"

# Only include debug info for files not under XLA.
build:dbg -c dbg
build:dbg --per_file_copt=external/xla/.*@-g0,-DNDEBUG