diff --git a/tensorflow/compiler/xla/service/cpu/cpu_compiler.cc b/tensorflow/compiler/xla/service/cpu/cpu_compiler.cc
index d852d0aae5d..a3e224824ba 100644
--- a/tensorflow/compiler/xla/service/cpu/cpu_compiler.cc
+++ b/tensorflow/compiler/xla/service/cpu/cpu_compiler.cc
@@ -297,6 +297,7 @@ Status CpuCompiler::RunHloPassesThroughLayoutAssn(
     pass.AddInvariantChecker<HloVerifier>(/*layout_sensitive=*/false,
                                           /*allow_mixed_precision=*/false);
 
+    pass.AddPass<ScatterExpander>();
     pass.AddPass<BatchNormExpander>(
         /*rewrite_training_op=*/true,
         /*rewrite_inference_op=*/true,
@@ -340,8 +341,6 @@ Status CpuCompiler::RunHloPassesThroughLayoutAssn(
 
   pipeline.AddPass<CpuInstructionFusion>();
 
-  pipeline.AddPass<ScatterExpander>();
-
   ReducePrecisionInsertion::AddPasses(
       &pipeline, module->config().debug_options(),
       ReducePrecisionInsertion::PassTiming::AFTER_FUSION);