[XLA] Remove xla_cpu_*_eigen flags from CPU backends.

These flags are currently de-facto unused; parallelism should be controlled
through the cpu_parallel backend. For configuring Eigen, if needed, the options
should be piped more directly to the code.

PiperOrigin-RevId: 159746509

14 files changed, 4 insertions, 223 deletions

diff --git a/tensorflow/compiler/aot/BUILD b/tensorflow/compiler/aot/BUILD
index 31637358c3..3e1670dce3 100644
--- a/tensorflow/compiler/aot/BUILD
+++ b/tensorflow/compiler/aot/BUILD
@@ -128,7 +128,6 @@ cc_library(
         ":tfcompile_proto",
         "//tensorflow/compiler/xla/legacy_flags:buffer_assignment_flags",
         "//tensorflow/compiler/xla/legacy_flags:compiler_functor_flags",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/legacy_flags:debug_options_flags",
         "//tensorflow/compiler/xla/legacy_flags:hlo_graph_dumper_flags",
         "//tensorflow/compiler/xla/legacy_flags:service_flags",
diff --git a/tensorflow/compiler/aot/tfcompile_main.cc b/tensorflow/compiler/aot/tfcompile_main.cc
index e03d28cd96..26a1b203ac 100644
--- a/tensorflow/compiler/aot/tfcompile_main.cc
+++ b/tensorflow/compiler/aot/tfcompile_main.cc
@@ -25,7 +25,6 @@ limitations under the License.
 #include "tensorflow/compiler/aot/tfcompile_util.h"
 #include "tensorflow/compiler/xla/legacy_flags/buffer_assignment_flags.h"
 #include "tensorflow/compiler/xla/legacy_flags/compiler_functor_flags.h"
-#include "tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h"
 #include "tensorflow/compiler/xla/legacy_flags/debug_options_flags.h"
 #include "tensorflow/compiler/xla/legacy_flags/hlo_graph_dumper_flags.h"
 #include "tensorflow/compiler/xla/legacy_flags/service_flags.h"
@@ -135,7 +134,6 @@ int main(int argc, char** argv) {
   AppendMainFlags(&flag_list, &flags);
   xla::legacy_flags::AppendBufferAssignmentFlags(&flag_list);
   xla::legacy_flags::AppendCompilerFunctorFlags(&flag_list);
-  xla::legacy_flags::AppendCpuRuntimeFlags(&flag_list);
   xla::legacy_flags::AppendHloGraphDumperFlags(&flag_list);
   xla::legacy_flags::AppendDebugOptionsFlags(&flag_list);
   xla::legacy_flags::AppendServiceFlags(&flag_list);
diff --git a/tensorflow/compiler/tests/BUILD b/tensorflow/compiler/tests/BUILD
index 044857d422..45642c9aa2 100644
--- a/tensorflow/compiler/tests/BUILD
+++ b/tensorflow/compiler/tests/BUILD
@@ -535,7 +535,6 @@ tf_library(
     cpp_class = "LSTMLayerInference",
     graph = "lstm_layer_inference.pbtxt",
     tags = ["manual"],
-    tfcompile_flags = "--xla_cpu_multi_thread_eigen=false",
 )
 
 # -----------------------------------------------------------------------------
diff --git a/tensorflow/compiler/xla/legacy_flags/BUILD b/tensorflow/compiler/xla/legacy_flags/BUILD
index fafd5f591b..ac049c9b7a 100644
--- a/tensorflow/compiler/xla/legacy_flags/BUILD
+++ b/tensorflow/compiler/xla/legacy_flags/BUILD
@@ -80,18 +80,6 @@ cc_library(
 )
 
 cc_library(
-    name = "cpu_runtime_flags",
-    srcs = ["cpu_runtime_flags.cc"],
-    hdrs = ["cpu_runtime_flags.h"],
-    deps =
-        [
-            ":parse_flags_from_env",
-            "//tensorflow/core:framework_internal",
-            "//tensorflow/core:lib",
-        ],
-)
-
-cc_library(
     name = "compiler_functor_flags",
     srcs = ["compiler_functor_flags.cc"],
     hdrs = ["compiler_functor_flags.h"],
diff --git a/tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.cc b/tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.cc
deleted file mode 100644
index d7817c5d54..0000000000
--- a/tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.cc
+++ /dev/null
@@ -1,71 +0,0 @@
-/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
-==============================================================================*/
-
-// Legacy flags for XLA's cpu_runtime module.
-
-#include <mutex>  // NOLINT(build/c++11): only using std::call_once, not mutex.
-#include <vector>
-
-#include "tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h"
-#include "tensorflow/compiler/xla/legacy_flags/parse_flags_from_env.h"
-#include "tensorflow/core/platform/types.h"
-#include "tensorflow/core/util/command_line_flags.h"
-
-namespace xla {
-namespace legacy_flags {
-
-// Pointers to the parsed value of the flags and flag descriptors, initialized
-// via flags_init.
-static CpuRuntimeFlags* flags;
-static std::vector<tensorflow::Flag>* flag_list;
-static std::once_flag flags_init;
-
-// Allocate *flags.  Called via call_once(&flags_init,...).
-static void AllocateFlags() {
-  flags = new CpuRuntimeFlags;
-  flags->xla_cpu_use_eigen = true;
-  flags->xla_cpu_multi_thread_eigen = true;
-  flag_list = new std::vector<tensorflow::Flag>({
-      tensorflow::Flag(
-          "xla_cpu_use_eigen", &flags->xla_cpu_use_eigen,
-          "Use Eigen for matrix multiply on the CPU platform. This "
-          "is a useful hack for performance comparisons against "
-          "XLA's implementation."),
-      tensorflow::Flag(
-          "xla_cpu_multi_thread_eigen", &flags->xla_cpu_multi_thread_eigen,
-          "When generating calls to Eigen for matmul and conv, should "
-          "single or multi-threaded eigen be used? "
-          "Only used when --xla_cpu_use_eigen is true."),
-  });
-  ParseFlagsFromEnv(*flag_list);
-}
-
-// Append to *append_to flag definitions associated with XLA's cpu_runtime
-// module.
-void AppendCpuRuntimeFlags(std::vector<tensorflow::Flag>* append_to) {
-  std::call_once(flags_init, &AllocateFlags);
-  append_to->insert(append_to->end(), flag_list->begin(), flag_list->end());
-}
-
-// Return a pointer to the CpuRuntimeFlags struct;
-// repeated calls return the same pointer.
-// This should be called only after Flags::Parse() has returned.
-CpuRuntimeFlags* GetCpuRuntimeFlags() {
-  std::call_once(flags_init, &AllocateFlags);
-  return flags;
-}
-
-}  // namespace legacy_flags
-}  // namespace xla
diff --git a/tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h b/tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h
deleted file mode 100644
index e3ff30da36..0000000000
--- a/tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h
+++ /dev/null
@@ -1,51 +0,0 @@
-/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
-==============================================================================*/
-
-#ifndef TENSORFLOW_COMPILER_XLA_LEGACY_FLAGS_CPU_RUNTIME_FLAGS_H_
-#define TENSORFLOW_COMPILER_XLA_LEGACY_FLAGS_CPU_RUNTIME_FLAGS_H_
-
-// Legacy flags for the XLA's cpu_runtime module.
-
-#include <vector>
-
-#include "tensorflow/core/platform/types.h"
-#include "tensorflow/core/util/command_line_flags.h"
-
-namespace xla {
-namespace legacy_flags {
-
-// Append to *flag_list flag definitions associated with XLA's cpu_runtime
-// module.
-void AppendCpuRuntimeFlags(std::vector<tensorflow::Flag>* flag_list);
-
-// The values of flags associated with XLA's cpu_runtime module.
-typedef struct {
-  // Use Eigen for matrix multiply on the CPU platform. This is a useful hack
-  // for performance comparisons against XLA's implementation.
-  bool xla_cpu_use_eigen;
-  // When generating calls to Eigen for matmul and conv, should single or
-  // multi-threaded eigen be used?  Only used when --xla_cpu_use_eigen is true.
-  bool xla_cpu_multi_thread_eigen;
-} CpuRuntimeFlags;
-
-// Return a pointer to the CpuRuntimeFlags struct;
-// repeated calls return the same pointer.
-// This should be called only after Flags::Parse() has returned.
-CpuRuntimeFlags* GetCpuRuntimeFlags();
-
-}  // namespace legacy_flags
-}  // namespace xla
-
-#endif  // TENSORFLOW_COMPILER_XLA_LEGACY_FLAGS_CPU_RUNTIME_FLAGS_H_
diff --git a/tensorflow/compiler/xla/service/cpu/BUILD b/tensorflow/compiler/xla/service/cpu/BUILD
index 68cd545695..93b9281874 100644
--- a/tensorflow/compiler/xla/service/cpu/BUILD
+++ b/tensorflow/compiler/xla/service/cpu/BUILD
@@ -197,7 +197,6 @@ cc_library(
         "//tensorflow/compiler/xla:util",
         "//tensorflow/compiler/xla:window_util",
         "//tensorflow/compiler/xla:xla_data_proto",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/service:buffer_assignment",
         "//tensorflow/compiler/xla/service:elemental_ir_emitter",
         "//tensorflow/compiler/xla/service:hlo",
@@ -228,7 +227,6 @@ cc_library(
         "//tensorflow/compiler/xla:types",
         "//tensorflow/compiler/xla:util",
         "//tensorflow/compiler/xla:xla_data_proto",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/service:hlo",
         "//tensorflow/compiler/xla/service/llvm_ir:ir_array",
         "//tensorflow/compiler/xla/service/llvm_ir:llvm_loop",
@@ -290,7 +288,6 @@ cc_library(
         "//tensorflow/compiler/xla:types",
         "//tensorflow/compiler/xla:util",
         "//tensorflow/compiler/xla/legacy_flags:compiler_functor_flags",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/service/llvm_ir:llvm_util",
         "//tensorflow/core:lib",
         "@llvm//:analysis",
@@ -485,7 +482,6 @@ cc_library(
         ":cpu_runtime",
         "//tensorflow/compiler/xla:shape_util",
         "//tensorflow/compiler/xla:window_util",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/service:hlo",
     ],
 )
@@ -512,7 +508,6 @@ cc_library(
         "//tensorflow/compiler/xla:shape_util",
         "//tensorflow/compiler/xla:util",
         "//tensorflow/compiler/xla:xla_data_proto",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/service:hlo",
         "//tensorflow/compiler/xla/service:hlo_pass",
         "//tensorflow/core:lib",
diff --git a/tensorflow/compiler/xla/service/cpu/compiler_functor.cc b/tensorflow/compiler/xla/service/cpu/compiler_functor.cc
index 8ebf9ab110..93895c3482 100644
--- a/tensorflow/compiler/xla/service/cpu/compiler_functor.cc
+++ b/tensorflow/compiler/xla/service/cpu/compiler_functor.cc
@@ -36,7 +36,6 @@ limitations under the License.
 #include "external/llvm/include/llvm/Transforms/IPO/AlwaysInliner.h"
 #include "external/llvm/include/llvm/Transforms/IPO/PassManagerBuilder.h"
 #include "tensorflow/compiler/xla/legacy_flags/compiler_functor_flags.h"
-#include "tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h"
 #include "tensorflow/compiler/xla/ptr_util.h"
 #include "tensorflow/compiler/xla/service/cpu/cpu_runtime.h"
 #include "tensorflow/compiler/xla/service/cpu/cpu_runtime_avx.h"
@@ -159,9 +158,7 @@ std::vector<llvm::VecDesc> VectorFunctionsForTargetLibraryInfoImpl(
   // Our vectorized library calls are currently implement by calling into Eigen.
   // As such, only emit calls to these routines if --xla_cpu_use_eigen is
   // enabled.
-  legacy_flags::CpuRuntimeFlags* flags = legacy_flags::GetCpuRuntimeFlags();
-  if (flags->xla_cpu_use_eigen &&
-      (arch == llvm::Triple::x86 || llvm::Triple::x86_64)) {
+  if (arch == llvm::Triple::x86 || llvm::Triple::x86_64) {
     llvm::SmallVector<llvm::StringRef, 32> features;
     feature_string.split(features, ',', -1, /*KeepEmpty=*/false);
     if (std::find(features.begin(), features.end(), "+sse4.1") !=
diff --git a/tensorflow/compiler/xla/service/cpu/conv_canonicalization.cc b/tensorflow/compiler/xla/service/cpu/conv_canonicalization.cc
index cdf43587b6..069979c661 100644
--- a/tensorflow/compiler/xla/service/cpu/conv_canonicalization.cc
+++ b/tensorflow/compiler/xla/service/cpu/conv_canonicalization.cc
@@ -15,7 +15,6 @@ limitations under the License.
 
 #include "tensorflow/compiler/xla/service/cpu/conv_canonicalization.h"
 
-#include "tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h"
 #include "tensorflow/compiler/xla/service/cpu/cpu_runtime.h"
 #include "tensorflow/compiler/xla/service/cpu/ir_emission_utils.h"
 #include "tensorflow/compiler/xla/service/hlo_computation.h"
@@ -30,11 +29,6 @@ namespace xla {
 namespace cpu {
 
 StatusOr<bool> ConvCanonicalization::Run(HloModule* module) {
-  legacy_flags::CpuRuntimeFlags* flags = legacy_flags::GetCpuRuntimeFlags();
-  if (!flags->xla_cpu_use_eigen) {
-    return false;
-  }
-
   bool changed = false;
   for (HloInstruction* hlo :
        module->entry_computation()->MakeInstructionPostOrder()) {
diff --git a/tensorflow/compiler/xla/service/cpu/dot_op_emitter.cc b/tensorflow/compiler/xla/service/cpu/dot_op_emitter.cc
index 420f9cebc5..a8d2565b64 100644
--- a/tensorflow/compiler/xla/service/cpu/dot_op_emitter.cc
+++ b/tensorflow/compiler/xla/service/cpu/dot_op_emitter.cc
@@ -22,7 +22,6 @@ limitations under the License.
 #include "external/llvm/include/llvm/IR/Instructions.h"
 #include "external/llvm/include/llvm/IR/Module.h"
 #include "external/llvm/include/llvm/IR/Value.h"
-#include "tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h"
 #include "tensorflow/compiler/xla/service/cpu/cpu_runtime.h"
 #include "tensorflow/compiler/xla/service/cpu/ir_emission_utils.h"
 #include "tensorflow/compiler/xla/service/llvm_ir/llvm_util.h"
@@ -233,22 +232,16 @@ tensorflow::Status DotOpEmitter::EmitCallToRuntime() {
   // The two transpose_... parameters are actually booleans, but we use int32
   // to avoid target-dependent calling convention details.
 
-  legacy_flags::CpuRuntimeFlags* flags = legacy_flags::GetCpuRuntimeFlags();
-  bool multi_threaded = flags->xla_cpu_multi_thread_eigen;
   PrimitiveType type = target_array_.GetShape().element_type();
   llvm::Type* float_type;
   const char* fn_name;
   switch (type) {
     case F32:
-      fn_name = multi_threaded
-                    ? runtime::kEigenMatmulF32SymbolName
-                    : runtime::kEigenSingleThreadedMatmulF32SymbolName;
+      fn_name = runtime::kEigenMatmulF32SymbolName;
       float_type = ir_builder_->getFloatTy();
       break;
     case F64:
-      fn_name = multi_threaded
-                    ? runtime::kEigenMatmulF64SymbolName
-                    : runtime::kEigenSingleThreadedMatmulF64SymbolName;
+      fn_name = runtime::kEigenMatmulF64SymbolName;
       float_type = ir_builder_->getDoubleTy();
       break;
     default:
diff --git a/tensorflow/compiler/xla/service/cpu/ir_emission_utils.cc b/tensorflow/compiler/xla/service/cpu/ir_emission_utils.cc
index 2d855d0eb1..859329e2c1 100644
--- a/tensorflow/compiler/xla/service/cpu/ir_emission_utils.cc
+++ b/tensorflow/compiler/xla/service/cpu/ir_emission_utils.cc
@@ -16,7 +16,6 @@ limitations under the License.
 #include "tensorflow/compiler/xla/service/cpu/ir_emission_utils.h"
 
 #include "tensorflow/compiler/xla/layout_util.h"
-#include "tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h"
 #include "tensorflow/compiler/xla/service/cpu/cpu_runtime.h"
 #include "tensorflow/compiler/xla/shape_util.h"
 #include "tensorflow/compiler/xla/window_util.h"
@@ -26,11 +25,6 @@ namespace cpu {
 
 bool PotentiallyImplementedAsEigenConvolution(
     const HloInstruction& convolution) {
-  legacy_flags::CpuRuntimeFlags* flags = legacy_flags::GetCpuRuntimeFlags();
-  if (!flags->xla_cpu_use_eigen) {
-    return false;
-  }
-
   // The following conditions are necessary (but not sufficient) for
   // implementing `convolution` with Eigen convolution:
   // - the input and kernel have a non-zero number of elements.
@@ -82,11 +76,6 @@ bool AreValidGemmShapes(const Shape& lhs_shape, const Shape& rhs_shape,
 }  // namespace
 
 bool PotentiallyImplementedAsEigenDot(const HloInstruction& hlo) {
-  legacy_flags::CpuRuntimeFlags* flags = legacy_flags::GetCpuRuntimeFlags();
-  if (!flags->xla_cpu_use_eigen) {
-    return false;
-  }
-
   // For certain types of Dot, we can call Eigen
   if (hlo.opcode() == HloOpcode::kDot) {
     const Shape& lhs_shape = hlo.operand(0)->shape();
diff --git a/tensorflow/compiler/xla/service/cpu/ir_emitter.cc b/tensorflow/compiler/xla/service/cpu/ir_emitter.cc
index fee5fd8830..695c8069c4 100644
--- a/tensorflow/compiler/xla/service/cpu/ir_emitter.cc
+++ b/tensorflow/compiler/xla/service/cpu/ir_emitter.cc
@@ -33,7 +33,6 @@ limitations under the License.
 #include "external/llvm/include/llvm/IR/Intrinsics.h"
 #include "external/llvm/include/llvm/IR/LLVMContext.h"
 #include "tensorflow/compiler/xla/layout_util.h"
-#include "tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h"
 #include "tensorflow/compiler/xla/map_util.h"
 #include "tensorflow/compiler/xla/service/buffer_assignment.h"
 #include "tensorflow/compiler/xla/service/cpu/cpu_runtime.h"
@@ -862,11 +861,7 @@ Status IrEmitter::HandleConvolution(HloInstruction* convolution,
            int64_type,    int64_type,     int64_type,     int64_type,
            int64_type,    int64_type,     int64_type,     int64_type},
           /*isVarArg=*/false);
-      legacy_flags::CpuRuntimeFlags* flags = legacy_flags::GetCpuRuntimeFlags();
-      const char* fn_name =
-          (flags->xla_cpu_multi_thread_eigen
-               ? runtime::kEigenConvF32SymbolName
-               : runtime::kEigenSingleThreadedConvF32SymbolName);
+      const char* fn_name = runtime::kEigenConvF32SymbolName;
       llvm::Function* conv_func = llvm::cast<llvm::Function>(
           module_->getOrInsertFunction(fn_name, conv_type));
       conv_func->setCallingConv(llvm::CallingConv::C);
diff --git a/tensorflow/compiler/xla/tests/BUILD b/tensorflow/compiler/xla/tests/BUILD
index a11ac0bec6..184b6b684e 100644
--- a/tensorflow/compiler/xla/tests/BUILD
+++ b/tensorflow/compiler/xla/tests/BUILD
@@ -496,7 +496,6 @@ xla_test(
         "//tensorflow/compiler/xla:shape_util",
         "//tensorflow/compiler/xla/client:computation_builder",
         "//tensorflow/compiler/xla/client:local_client",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/legacy_flags:debug_options_flags",
         "//tensorflow/compiler/xla/legacy_flags:layout_util_flags",
         "//tensorflow/compiler/xla/tests:client_library_test_base",
@@ -513,43 +512,6 @@ xla_test(
 xla_test(
     name = "dot_operation_runtime_test",
     srcs = ["dot_operation_test.cc"],
-    backend_args = {
-        "cpu": ["--xla_cpu_use_eigen"],
-        "cpu_parallel": ["--xla_cpu_use_eigen"],
-    },
-    deps = [
-        "//tensorflow/compiler/xla:array2d",
-        "//tensorflow/compiler/xla:array3d",
-        "//tensorflow/compiler/xla:reference_util",
-        "//tensorflow/compiler/xla:shape_util",
-        "//tensorflow/compiler/xla/client:computation_builder",
-        "//tensorflow/compiler/xla/client:local_client",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
-        "//tensorflow/compiler/xla/legacy_flags:debug_options_flags",
-        "//tensorflow/compiler/xla/legacy_flags:layout_util_flags",
-        "//tensorflow/compiler/xla/tests:client_library_test_base",
-        "//tensorflow/compiler/xla/tests:literal_test_util",
-        "//tensorflow/compiler/xla/tests:test_utils",
-        "//tensorflow/core:framework_internal",
-        "//tensorflow/core:lib",
-        "//tensorflow/core:test",
-    ],
-)
-
-# Repeat dot_operation_runtime_test with single-threded eigen.
-xla_test(
-    name = "dot_operation_single_threaded_runtime_test",
-    srcs = ["dot_operation_test.cc"],
-    backend_args = {
-        "cpu": [
-            "--xla_cpu_use_eigen",
-            "--xla_cpu_multi_thread_eigen=false",
-        ],
-        "cpu_parallel": [
-            "--xla_cpu_use_eigen",
-            "--xla_cpu_multi_thread_eigen=false",
-        ],
-    },
     deps = [
         "//tensorflow/compiler/xla:array2d",
         "//tensorflow/compiler/xla:array3d",
@@ -557,7 +519,6 @@ xla_test(
         "//tensorflow/compiler/xla:shape_util",
         "//tensorflow/compiler/xla/client:computation_builder",
         "//tensorflow/compiler/xla/client:local_client",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/legacy_flags:debug_options_flags",
         "//tensorflow/compiler/xla/legacy_flags:layout_util_flags",
         "//tensorflow/compiler/xla/tests:client_library_test_base",
@@ -574,11 +535,9 @@ xla_test(
     srcs = ["dot_operation_test.cc"],
     backend_args = {
         "cpu": [
-            "--xla_cpu_use_eigen",
             "--xla_default_layout=major2minor",
         ],
         "cpu_parallel": [
-            "--xla_cpu_use_eigen",
             "--xla_default_layout=major2minor",
         ],
     },
@@ -589,7 +548,6 @@ xla_test(
         "//tensorflow/compiler/xla:shape_util",
         "//tensorflow/compiler/xla/client:computation_builder",
         "//tensorflow/compiler/xla/client:local_client",
-        "//tensorflow/compiler/xla/legacy_flags:cpu_runtime_flags",
         "//tensorflow/compiler/xla/legacy_flags:debug_options_flags",
         "//tensorflow/compiler/xla/legacy_flags:layout_util_flags",
         "//tensorflow/compiler/xla/tests:client_library_test_base",
diff --git a/tensorflow/compiler/xla/tests/dot_operation_test.cc b/tensorflow/compiler/xla/tests/dot_operation_test.cc
index b06b5c5f47..7abef6a27b 100644
--- a/tensorflow/compiler/xla/tests/dot_operation_test.cc
+++ b/tensorflow/compiler/xla/tests/dot_operation_test.cc
@@ -20,7 +20,6 @@ limitations under the License.
 #include "tensorflow/compiler/xla/array3d.h"
 #include "tensorflow/compiler/xla/client/computation_builder.h"
 #include "tensorflow/compiler/xla/client/local_client.h"
-#include "tensorflow/compiler/xla/legacy_flags/cpu_runtime_flags.h"
 #include "tensorflow/compiler/xla/legacy_flags/debug_options_flags.h"
 #include "tensorflow/compiler/xla/legacy_flags/layout_util_flags.h"
 #include "tensorflow/compiler/xla/primitive_util.h"
@@ -461,7 +460,6 @@ int main(int argc, char** argv) {
   std::vector<tensorflow::Flag> flag_list;
   xla::legacy_flags::AppendLayoutUtilFlags(&flag_list);
   xla::legacy_flags::AppendDebugOptionsFlags(&flag_list);
-  xla::legacy_flags::AppendCpuRuntimeFlags(&flag_list);
   xla::string usage = tensorflow::Flags::Usage(argv[0], flag_list);
   const bool parse_result = tensorflow::Flags::Parse(&argc, argv, flag_list);
   if (!parse_result) {