diff --git a/mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorRewriting.cpp b/mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorRewriting.cpp
--- a/mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorRewriting.cpp
+++ b/mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorRewriting.cpp
@@ -647,9 +647,9 @@
       src = rewriter.create<LoadOp>(loc, foreachOp.getResult(0), true);
     }
 
-    SparseTensorEncodingAttr encSrc = getSparseTensorEncoding(srcTp);
     // Sort the COO tensor so that its elements are ordered via increasing
     // indices for the storage ordering of the dst tensor.
+    SparseTensorEncodingAttr encSrc = getSparseTensorEncoding(srcTp);
     auto dynShape = {ShapedType::kDynamicSize};
     auto indTp =
         MemRefType::get(dynShape, getIndexOverheadType(rewriter, encSrc));
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.mlir
@@ -4,7 +4,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler=enable-runtime-library=false | \
 // RUN: TENSOR0="%mlir_src_dir/test/Integration/data/test.mtx" \
 // RUN: mlir-cpu-runner \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_bf16.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_bf16.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_bf16.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_bf16.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler=enable-runtime-library=false | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_f16.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_f16.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_f16.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_f16.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler=enable-runtime-library=false | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_codegen_foreach.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_codegen_foreach.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_codegen_foreach.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_codegen_foreach.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler=enable-runtime-library=false | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_dyn.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_dyn.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_dyn.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_dyn.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler="enable-runtime-library=false enable-buffer-initialization=true" | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_ptr.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_ptr.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_ptr.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_conversion_ptr.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler="enable-runtime-library=false enable-buffer-initialization=true" | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_matrix_ops.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_matrix_ops.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_matrix_ops.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_matrix_ops.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler=enable-runtime-library=false | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_mult_elt.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_mult_elt.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_mult_elt.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_mult_elt.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler="enable-runtime-library=false enable-buffer-initialization=true"| \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_reduction.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_reduction.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_reduction.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_out_reduction.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler="enable-runtime-library=false enable-buffer-initialization=true"| \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_sorted_coo.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_sorted_coo.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_sorted_coo.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_sorted_coo.mlir
@@ -5,7 +5,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler="enable-runtime-library=false enable-buffer-initialization=true" | \
 // RUN: TENSOR0="%mlir_src_dir/test/Integration/data/wide.mtx" \
 // RUN: TENSOR1="%mlir_src_dir/test/Integration/data/mttkrp_b.tns" \
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_storage.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_storage.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_storage.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_storage.mlir
@@ -3,7 +3,8 @@
 // RUN:  -e entry -entry-point-result=void  \
 // RUN:  -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
-
+//
+// Do the same run, but now with direct IR generation.
 // RUN: mlir-opt %s --sparse-compiler=enable-runtime-library=false | \
 // RUN: mlir-cpu-runner \
 // RUN:  -e entry -entry-point-result=void  \