diff --git a/tensorflow/compiler/xla/service/mlir_gpu/tests/BUILD b/tensorflow/compiler/xla/service/mlir_gpu/tests/BUILD
index ddf5a481c51..014b26c5c78 100644
--- a/tensorflow/compiler/xla/service/mlir_gpu/tests/BUILD
+++ b/tensorflow/compiler/xla/service/mlir_gpu/tests/BUILD
@@ -25,8 +25,6 @@ glob_lit_tests(
     ],
     driver = "@llvm-project//mlir:run_lit.sh",
     exclude = [
-        # TODO(b/149302060) Reenable once fusion is fixed.
-        "iota_add_multiply.hlo",
         # TODO(b/137624192): Reenable once we can fuse reductions.
         "fused_reduce.hlo",
     ],
diff --git a/tensorflow/compiler/xla/service/mlir_gpu/tests/iota_add_multiply.hlo b/tensorflow/compiler/xla/service/mlir_gpu/tests/iota_add_subtract.hlo
similarity index 56%
rename from tensorflow/compiler/xla/service/mlir_gpu/tests/iota_add_multiply.hlo
rename to tensorflow/compiler/xla/service/mlir_gpu/tests/iota_add_subtract.hlo
index 1c52d435ba3..f42a7cf7ca6 100644
--- a/tensorflow/compiler/xla/service/mlir_gpu/tests/iota_add_multiply.hlo
+++ b/tensorflow/compiler/xla/service/mlir_gpu/tests/iota_add_subtract.hlo
@@ -1,16 +1,16 @@
 // RUN: xla-gpu-opt -lowering-stage=GPU %s | FileCheck %s -dump-input-on-failure
-HloModule AddMultiply
+HloModule AddSubtract
 
-ENTRY %AddMultiply (x: s32[2,2], y: s32[2,2]) -> s32[2,2] {
+ENTRY %AddSubtract (x: s32[2,2], y: s32[2,2]) -> s32[2,2] {
   %x = s32[2,2]{1,0} parameter(0)
   %y = s32[2,2]{1,0} parameter(1)
 
   %add = s32[2,2]{1,0} add(s32[2,2]{1,0} %x, s32[2,2]{1,0} %y)
   %iota = s32[2, 2]{1,0} iota(), iota_dimension=0
 
-  ROOT %mul = s32[2,2]{1,0} multiply(s32[2,2]{1,0} %add, s32[2,2]{1,0} %iota)
+  ROOT %sub = s32[2,2]{1,0} subtract(s32[2,2]{1,0} %add, s32[2,2]{1,0} %iota)
 }
 
 //  CHECK-NOT:  store
-//  CHECK:      %[[RESULT:.*]] = muli %{{.*}}, %{{.*}}
-//  CHECK:      store %[[RESULT]]
+//  CHECK:      [[RESULT:%.*]] = subi %{{.*}}, %{{.*}}
+//  CHECK:      store [[RESULT]]