Remove redundant <fstream> includes (#563)

* Remove redundant <fstream> includes

* Fix fstream in examples/

* Fix <fstream> in test/

* Use consistent order for <fstream> (always after <iostream>)

* Remove an unneeded include in a file where std::ofstream usage is commented out

Co-authored-by: Ivan Komarov <dfyz@yandex-team.ru>

test/unit/conv/device/conv2d_testbed.h

@@ -33,6 +33,8 @@
 */
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 #include "cutlass/cutlass.h"
 

test/unit/conv/device/conv2d_testbed_interleaved.h

@@ -33,6 +33,8 @@
 */
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 #include "cutlass/cutlass.h"
 

test/unit/conv/device/conv2d_with_broadcast_testbed.h

@@ -37,6 +37,8 @@
 */
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 #include "cutlass/cutlass.h"
 

test/unit/conv/device/conv2d_with_reduction_testbed.h

@@ -33,6 +33,8 @@
 */
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 #include "cutlass/cutlass.h"
 

test/unit/conv/device/conv3d_testbed.h

@@ -33,6 +33,8 @@
 */
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 #include "cutlass/cutlass.h"
 

test/unit/epilogue/threadblock/epilogue_planar_complex.cu

@@ -32,8 +32,6 @@
     \brief Unit tests for thread-level GEMM
 */
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/epilogue/threadblock/epilogue_simt.cu

@@ -32,8 +32,6 @@
     \brief Unit tests for thread-level GEMM
 */
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/epilogue/threadblock/epilogue_simt_sm60.cu

@@ -32,8 +32,6 @@
     \brief Unit tests for thread-level GEMM
 */
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/epilogue/threadblock/epilogue_simt_sm61.cu

@@ -32,8 +32,6 @@
     \brief Unit tests for thread-level GEMM
 */
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/epilogue/threadblock/epilogue_tensor_op.cu

@@ -32,8 +32,6 @@
     \brief Unit tests for thread-level GEMM
 */
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/epilogue/threadblock/epilogue_volta_tensor_op.cu

@@ -32,8 +32,6 @@
     \brief Unit tests for thread-level GEMM
 */
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/epilogue/threadblock/epilogue_with_reduction_tensor_op.cu

@@ -33,8 +33,6 @@
   \brief Unit tests for thread-level GEMM
 */
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/epilogue/threadblock/epilogue_wmma_tensor_op_sm70.cu

@@ -35,8 +35,6 @@
 
 #ifdef CUTLASS_ARCH_WMMA_SM70_ENABLED
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/epilogue/threadblock/output_tile_threadmap.cu

@@ -32,8 +32,6 @@
     \brief Unit tests for thread-level GEMM
 */
 
-#include <fstream>
-
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/gemm/device/testbed_complex.h

@@ -35,7 +35,6 @@
 #pragma once
 
 #include <iostream>
-#include <fstream>
 #include <sstream>
 #include <stdexcept>
 

test/unit/gemm/device/testbed_grouped.h

@@ -36,6 +36,7 @@
 #pragma once
 
 #include <iostream>
+#include <fstream>
 
 #include "../../common/cutlass_unit_test.h"
 #include "cutlass/cutlass.h"

test/unit/gemm/device/testbed_sanity.h

@@ -33,7 +33,6 @@
 */
 
 #include <iostream>
-#include <fstream>
 #include <sstream>
 
 #include "../../common/cutlass_unit_test.h"

test/unit/gemm/device/testbed_splitk.h

@@ -35,7 +35,6 @@
 #pragma once
 
 #include <iostream>
-#include <fstream>
 #include <sstream>
 
 #include "../../common/cutlass_unit_test.h"

test/unit/gemm/threadblock/mma_multistage_testbed_slicedk.h

@@ -35,6 +35,8 @@
 
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/gemm/threadblock/mma_pipelined_testbed.h

@@ -34,6 +34,8 @@
 
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/gemm/threadblock/mma_pipelined_testbed_slicedk.h

@@ -35,6 +35,8 @@
 
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/aligned_buffer.h"

test/unit/gemm/threadblock/mma_planar_complex_testbed.h

@@ -34,6 +34,8 @@
 
 #pragma once
 
+#include <fstream>
+
 #include "../../common/cutlass_unit_test.h"
 
 #include "cutlass/cutlass.h"