Test: CI build optimizations (CUDA compute capabilities and non-curses mode from cl/945255455)

.bazelrc

@@ -17,7 +17,8 @@ build:cuda_clang --host_copt=-Wno-error=unused-command-line-argument
 build:cuda_clang --repo_env TF_NEED_TENSORRT=0
 build:cuda_clang --action_env=TF_CUDA_CLANG="1"
 build:cuda_clang --@local_config_cuda//:cuda_compiler=clang
-build:cuda_clang --repo_env=TF_CUDA_COMPUTE_CAPABILITIES="sm_60,sm_70,sm_80,sm_89,compute_90"
+build:cuda_clang --repo_env=TF_CUDA_COMPUTE_CAPABILITIES="sm_80,sm_89,compute_90"
+build:cuda_clang --copt=-Dtypeof=__typeof__
 
 common:cuda_clang --repo_env=HERMETIC_CUDA_VERSION="12.5.1"
 common:cuda_clang --repo_env=HERMETIC_CUDNN_VERSION="9.3.0"

WORKSPACE

@@ -24,8 +24,8 @@ local_repository(
 load("//tensorflow_serving:repo.bzl", "tensorflow_http_archive")
 tensorflow_http_archive(
     name = "org_tensorflow",
-    sha256 = "54bfd56d61dfced2b82bdeee3262a24fb3d59bed4dbb292d19834af002d85b3b",
-    git_commit = "c1be979f3130b85f14e8890c05cbca147784f474",
+    sha256 = "582f5dc1c0887ca5b2eb364dc39e0c15fd31ee138b250e24163b6eca340f701f",
+    git_commit = "d875953b4fd431c15fedb1d408d5396e7f9d4d23",
     patch = "//third_party/tensorflow:tensorflow.patch",
     patch_cmds = [
         "sed -i '/cc_library = _cc_library/d' tensorflow/core/platform/rules_cc.bzl",
@@ -120,6 +120,9 @@ http_archive(
         "https://storage.googleapis.com/mirror.tensorflow.org/github.com/google-ml-infra/rules_ml_toolchain/archive/398d613aea7a4c294da49b79a6d6f3f8732bd84c.tar.gz",
         "https://github.com/google-ml-infra/rules_ml_toolchain/archive/398d613aea7a4c294da49b79a6d6f3f8732bd84c.tar.gz",
     ],
+    patch_cmds = [
+        "cat << 'EOF' >> cc/cuda/clang/clang_cuda_runtime_wrapper.h\n#if defined(__clang__) && defined(__CUDA__)\n#include <vector_types.h>\n#ifndef typeof\n#define typeof __typeof__\n#endif\n__device__ inline void __stwt(uint4* ptr, uint4 value) {\n  asm volatile(\"st.global.wt.v4.u32 [%0], {%1, %2, %3, %4};\" : : \"l\"(ptr), \"r\"(value.x), \"r\"(value.y), \"r\"(value.z), \"r\"(value.w) : \"memory\");\n}\n#endif\nEOF",
+    ],
 )
 
 load(

tensorflow_serving/opensource_only/.bazelrc

@@ -0,0 +1,128 @@
+# Optimizations used for TF Serving release builds.
+build:release --copt=-mavx
+build:release --copt=-msse4.2
+
+# Options used to build with CUDA.
+build:cuda --repo_env TF_NEED_CUDA=1
+build:cuda --crosstool_top=@local_config_cuda//crosstool:toolchain
+build:cuda --@local_config_cuda//:enable_cuda
+
+# Options used to build with CUDA clang
+build:cuda_clang --config=cuda
+build:cuda_clang --copt=-Wno-gnu-offsetof-extensions
+build:cuda_clang --copt=-Wno-error=unused-command-line-argument
+build --linkopt=-Wno-unused-command-line-argument
+build --host_linkopt=-Wno-unused-command-line-argument
+build:cuda_clang --host_copt=-Wno-error=unused-command-line-argument
+build:cuda_clang --repo_env TF_NEED_TENSORRT=0
+build:cuda_clang --action_env=TF_CUDA_CLANG="1"
+build:cuda_clang --@local_config_cuda//:cuda_compiler=clang
+build:cuda_clang --repo_env=TF_CUDA_COMPUTE_CAPABILITIES="sm_80,sm_89,compute_90"
+build:cuda_clang --copt=-Dtypeof=__typeof__
+
+common:cuda_clang --repo_env=HERMETIC_CUDA_VERSION="12.5.1"
+common:cuda_clang --repo_env=HERMETIC_CUDNN_VERSION="9.3.0"
+common:cuda_clang --repo_env=HERMETIC_NVSHMEM_VERSION="3.2.5"
+common:cuda_clang --repo_env=HERMETIC_NCCL_VERSION="2.29.7"
+build:cuda_clang --action_env=GCC_HOST_COMPILER_PATH="/usr/bin/gcc-10"
+build:cuda_clang --action_env=CLANG_CUDA_COMPILER_PATH="/usr/lib/llvm-17/bin/clang"
+
+# Options used to build with TPU support.
+build:tpu --define=with_tpu_support=true --define=framework_shared_object=false
+build:tpu --copt=-DLIBTPU_ON_GCE
+
+# Please note that MKL on MacOS or windows is still not supported.
+# If you would like to use a local MKL instead of downloading, please set the
+# environment variable "TF_MKL_ROOT" every time before build.
+build:mkl --define=build_with_mkl=true --define=enable_mkl=true --define=build_with_openmp=false
+build:mkl --define=tensorflow_mkldnn_contraction_kernel=0
+
+# This config option is used to enable MKL-DNN open source library only,
+# without depending on MKL binary version.
+build:mkl_open_source_only --define=build_with_mkl_dnn_only=true
+build:mkl_open_source_only --define=build_with_mkl=true --define=enable_mkl=true
+build:mkl_open_source_only --define=tensorflow_mkldnn_contraction_kernel=0
+
+# Config setting to build oneDNN with Compute Library for the Arm Architecture (ACL).
+# This build is for the inference regime only.
+build:mkl_aarch64 --define=build_with_mkl_aarch64=true --define=enable_mkl=true
+build:mkl_aarch64 --define=tensorflow_mkldnn_contraction_kernel=0
+build:mkl_aarch64 --define=build_with_mkl_opensource=true
+build:mkl_aarch64 --define=build_with_openmp=true
+build:mkl_aarch64 --copt=-march=armv8.2-a
+build:mkl_aarch64 --copt=-O3
+
+build --define=build_with_onednn_v2=true
+build --define=xnn_enable_avxvnni=false
+build --define=xnn_enable_avxvnniint8=false
+build --define=xnn_enable_avx256vnnigfni=false
+build --define=xnn_enable_avx512amx=false
+build --define=xnn_enable_avx512fp16=false
+
+# Processor native optimizations (depends on build host capabilities).
+build:nativeopt --copt=-march=native
+build:nativeopt --host_copt=-march=native
+build:nativeopt --copt=-O3
+
+build --keep_going
+build --verbose_failures=true
+build --spawn_strategy=standalone
+build --genrule_strategy=standalone
+
+build --define=grpc_no_ares=true
+
+# Sets the default Apple platform to macOS.
+build --apple_platform_type=macos
+
+build -c opt
+
+# LLVM, MLIR and TF requires C++17.
+build --cxxopt=-std=c++17
+build --host_cxxopt=-std=c++17
+
+# Adding "--cxxopt=-D_GLIBCXX_USE_CXX11_ABI=0" creates parity with TF
+# compilation options. It also addresses memory use due to
+# copy-on-write semantics of std::strings of the older ABI.
+build --cxxopt=-D_GLIBCXX_USE_CXX11_ABI=0
+
+build --workspace_status_command=/proc/self/cwd/tools/gen_status_stamp.sh
+
+build --experimental_repo_remote_exec
+
+# TF now has `cc_shared_library` targets, so it needs the experimental flag
+build --experimental_cc_shared_library
+
+# Yggdrasil Decision Forests (internal library of TensorFlow Decision Forests)
+# uses TensorFlow for all IO operations.
+build --define=use_tensorflow_io=1
+
+# TensorFlow Decision Forests does not use Absl concurrency primitives on MacOs.
+# Reason: TensorFlow/ABSL ODR violation (b/214189609) # copybara:strip
+build:macos --define std_synchronization_primitives=1
+
+# Taken from https://github.com/openxla/xla/blob/99559d7a4f7c55490f46385ad29a3cbf9c3911af/warnings.bazelrc#L6
+# We silence warnings for code in `external`.
+build --per_file_copt=external/.*@-w
+build --host_per_file_copt=external/.*@-w
+build --copt=-Wno-macro-redefined # absl vs tsl logging clash
+build --copt=-Wno-sign-compare # int as loop variable
+build --copt=-Wno-deprecated-declarations
+build --copt=-Wno-unused-but-set-variable # due to `ifdefs` in ml_dtypes
+
+# Revert to the legacy WORKSPACE dependency resolution system
+# and disable Bzlmod.
+common --enable_bzlmod=false
+common --incompatible_enable_cc_toolchain_resolution
+common --repo_env USE_HERMETIC_CC_TOOLCHAIN=1
+common --repo_env=HERMETIC_PYTHON_VERSION="3.10"
+common --repo_env=USE_PYWRAP_RULES=1
+build --features=-layering_check
+build --define=framework_shared_object=false
+build --dynamic_mode=off
+
+
+common:clang_local --noincompatible_enable_cc_toolchain_resolution
+common:clang_local --noincompatible_enable_android_toolchain_resolution
+common:clang_local --@rules_ml_toolchain//cc_toolchain/config:enable_hermetic_cc=False
+common:clang_local --repo_env USE_HERMETIC_CC_TOOLCHAIN=0
+