diff --git a/tensorflow/compiler/mlir/xla/BUILD b/tensorflow/compiler/mlir/xla/BUILD
index ec98d9d29e5..32a2ed1c272 100644
--- a/tensorflow/compiler/mlir/xla/BUILD
+++ b/tensorflow/compiler/mlir/xla/BUILD
@@ -238,6 +238,7 @@ cc_library(
     deps = [
         ":type_to_shape",
         "//tensorflow/compiler/mlir/hlo",
+        "//tensorflow/compiler/mlir/hlo:hlo_dialect_force_registration",
         "//tensorflow/compiler/mlir/tensorflow:convert_type",
         "//tensorflow/compiler/mlir/tensorflow:error_util",
         "//tensorflow/compiler/tf2xla:common",
@@ -388,6 +389,7 @@ cc_library(
         ":xla_legalize_tf_with_tf2xla",
         "//tensorflow/compiler/mlir/hlo",
         "//tensorflow/compiler/mlir/hlo:chlo_legalize_to_hlo",
+        "//tensorflow/compiler/mlir/hlo:hlo_dialect_force_registration",
         "//tensorflow/compiler/mlir/hlo:hlo_legalize_to_lhlo",
         "//tensorflow/compiler/mlir/hlo:legalize_control_flow",
         "//tensorflow/compiler/mlir/hlo:legalize_tanh_to_approximation",
diff --git a/tensorflow/compiler/xla/service/gpu/BUILD b/tensorflow/compiler/xla/service/gpu/BUILD
index ce761d8e0ae..d1d0827981e 100644
--- a/tensorflow/compiler/xla/service/gpu/BUILD
+++ b/tensorflow/compiler/xla/service/gpu/BUILD
@@ -254,7 +254,6 @@ cc_library(
         ":target_util",
         ":thunk",
         ":thunk_emitter",
-        "//tensorflow/compiler/mlir/hlo",
         "//tensorflow/compiler/mlir/hlo:lhlo",
         "//tensorflow/compiler/mlir/xla:hlo_utils",
         "//tensorflow/compiler/mlir/xla:mhlo_to_lhlo_with_xla",
diff --git a/tensorflow/compiler/xla/service/gpu/ir_emitter_context.h b/tensorflow/compiler/xla/service/gpu/ir_emitter_context.h
index 2a493fe4494..7d5a8d032e6 100644
--- a/tensorflow/compiler/xla/service/gpu/ir_emitter_context.h
+++ b/tensorflow/compiler/xla/service/gpu/ir_emitter_context.h
@@ -17,10 +17,7 @@ limitations under the License.
 #define TENSORFLOW_COMPILER_XLA_SERVICE_GPU_IR_EMITTER_CONTEXT_H_
 
 #include "llvm/IR/Module.h"
-#include "mlir/Dialect/StandardOps/IR/Ops.h"  // from @llvm-project
 #include "mlir/IR/MLIRContext.h"  // from @llvm-project
-#include "tensorflow/compiler/mlir/hlo/include/mlir-hlo/Dialect/mhlo/IR/hlo_ops.h"
-#include "tensorflow/compiler/mlir/hlo/include/mlir-hlo/Dialect/mhlo/IR/lhlo_ops.h"
 #include "tensorflow/compiler/xla/service/buffer_assignment.h"
 #include "tensorflow/compiler/xla/service/gpu/launch_dimensions.h"
 #include "tensorflow/compiler/xla/service/hlo_execution_profile.h"
@@ -47,11 +44,7 @@ class IrEmitterContext {
         cuda_compute_capability_(cuda_compute_capability),
         profile_index_map_(profile_index_map),
         mlir_context_(mlir_context),
-        llvm_module_(llvm_module) {
-    mlir_context_
-        ->loadDialect<mlir::lmhlo::LmhloDialect, mlir::mhlo::MhloDialect,
-                      mlir::StandardOpsDialect>();
-  }
+        llvm_module_(llvm_module) {}
   // Disallow copy and assign.
   IrEmitterContext(const IrEmitterContext&) = delete;
   IrEmitterContext& operator=(const IrEmitterContext&) = delete;
diff --git a/tensorflow/compiler/xla/service/mlir_gpu/BUILD b/tensorflow/compiler/xla/service/mlir_gpu/BUILD
index af670eb059f..786f28c7705 100644
--- a/tensorflow/compiler/xla/service/mlir_gpu/BUILD
+++ b/tensorflow/compiler/xla/service/mlir_gpu/BUILD
@@ -41,12 +41,9 @@ cc_library(
     srcs = ["emission_context.cc"],
     hdrs = ["emission_context.h"],
     deps = [
-        "//tensorflow/compiler/mlir/hlo",
-        "//tensorflow/compiler/mlir/hlo:lhlo",
         "//tensorflow/compiler/xla/service:hlo",
         "@com_google_absl//absl/strings",
         "@llvm-project//mlir:IR",
-        "@llvm-project//mlir:StandardOps",
     ],
 )
 
diff --git a/tensorflow/compiler/xla/service/mlir_gpu/emission_context.cc b/tensorflow/compiler/xla/service/mlir_gpu/emission_context.cc
index 06c7ebd1099..cb5ea946c1b 100644
--- a/tensorflow/compiler/xla/service/mlir_gpu/emission_context.cc
+++ b/tensorflow/compiler/xla/service/mlir_gpu/emission_context.cc
@@ -16,11 +16,8 @@ limitations under the License.
 #include "tensorflow/compiler/xla/service/mlir_gpu/emission_context.h"
 
 #include "absl/strings/substitute.h"
-#include "mlir/Dialect/StandardOps/IR/Ops.h"  // from @llvm-project
 #include "mlir/IR/Location.h"  // from @llvm-project
 #include "mlir/IR/MLIRContext.h"  // from @llvm-project
-#include "tensorflow/compiler/mlir/hlo/include/mlir-hlo/Dialect/mhlo/IR/hlo_ops.h"
-#include "tensorflow/compiler/mlir/hlo/include/mlir-hlo/Dialect/mhlo/IR/lhlo_ops.h"
 #include "tensorflow/compiler/xla/service/hlo_instruction.h"
 
 namespace xla {
@@ -28,8 +25,7 @@ namespace mlir_gpu {
 
 EmissionContext::EmissionContext(std::unique_ptr<HloModule> module)
     : module_(std::move(module)), context_() {
-  context_.loadDialect<mlir::mhlo::MhloDialect, mlir::lmhlo::LmhloDialect,
-                       mlir::StandardOpsDialect>();
+  context_.loadAllGloballyRegisteredDialects();
   error_handler_ = [](const ErrorMap& instructions_with_error,
                       HloModule* module) {
     std::set<const HloComputation*> computations_with_error;