Move fusion executor files to their own directory

CMakeLists.txt

@@ -123,13 +123,13 @@ list(APPEND NVFUSER_SRCS
   ${NVFUSER_SRCS_DIR}/dynamic_transform.cpp
   ${NVFUSER_SRCS_DIR}/evaluator_common.cpp
   ${NVFUSER_SRCS_DIR}/exceptions.cpp
-  ${NVFUSER_SRCS_DIR}/executor.cpp
-  ${NVFUSER_SRCS_DIR}/executor_kernel_arg.cpp
-  ${NVFUSER_SRCS_DIR}/executor_params.cpp
-  ${NVFUSER_SRCS_DIR}/executor_utils.cpp
   ${NVFUSER_SRCS_DIR}/expr_evaluator.cpp
   ${NVFUSER_SRCS_DIR}/expr_simplifier.cpp
   ${NVFUSER_SRCS_DIR}/fusion.cpp
+  ${NVFUSER_SRCS_DIR}/fusion_executor/executor.cpp
+  ${NVFUSER_SRCS_DIR}/fusion_executor/executor_kernel_arg.cpp
+  ${NVFUSER_SRCS_DIR}/fusion_executor/executor_params.cpp
+  ${NVFUSER_SRCS_DIR}/fusion_executor/executor_utils.cpp
   ${NVFUSER_SRCS_DIR}/fusion_segmenter.cpp
   ${NVFUSER_SRCS_DIR}/global_allocator.cpp
   ${NVFUSER_SRCS_DIR}/grouped_reduction.cpp

benchmarks/cpp/batch_norm_channels_first.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/batch_norm_channels_first_backward.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/batch_norm_channels_last.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/batch_norm_channels_last_backward.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/bert.cpp

@@ -6,8 +6,8 @@
  */
 // clang-format on
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/broadcast.cpp

@@ -6,8 +6,8 @@
  */
 // clang-format on
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/utils.h>
 #include <ops/arith.h>

benchmarks/cpp/gelu_backward.cpp

@@ -9,8 +9,8 @@
 // Based on NVFuserTest.FusionBiasGeluBwd_CUDA
 
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/builder.h>
 #include <ops/arith.h>
 #include <scheduler/all_schedulers.h>

benchmarks/cpp/gelu_backward_reduction.cpp

@@ -6,8 +6,8 @@
  */
 // clang-format on
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/utils.h>
 #include <ops/arith.h>

benchmarks/cpp/heuristic_cache.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/heuristic_lookup.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/indexselect.cpp

@@ -9,8 +9,8 @@
 // Based on NVFuserTest.FusionBiasGeluBwd_CUDA
 
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/builder.h>
 #include <ops/arith.h>
 #include <scheduler/all_schedulers.h>

benchmarks/cpp/instance_norm.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/builder.h>
 #include <ops/all_ops.h>
 #include <scheduler/all_schedulers.h>

benchmarks/cpp/layer_norm.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/layer_norm_backward.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/layer_norm_fused.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/lstm_cell.cpp

@@ -6,8 +6,8 @@
  */
 // clang-format on
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ops/all_ops.h>
 #include <scheduler/all_schedulers.h>
 

benchmarks/cpp/matmul.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/analysis/bank_conflict.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/utils.h>
 #include <ops/all_ops.h>

benchmarks/cpp/reduction.cpp

@@ -6,8 +6,8 @@
  */
 // clang-format on
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/utils.h>
 #include <ops/arith.h>

benchmarks/cpp/rms_norm.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/rms_norm_backward.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/scale_bias_relu.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ops/arith.h>
 #include <scheduler/all_schedulers.h>
 

benchmarks/cpp/shape_inference.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/softmax.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/utils.h>
 #include <ops/all_ops.h>

benchmarks/cpp/softmax_backward.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/utils.h>
 #include <ops/all_ops.h>

benchmarks/cpp/softmax_dropout.cpp

@@ -7,8 +7,8 @@
 // clang-format on
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 #include <ir/utils.h>

benchmarks/cpp/timm.cpp

@@ -5,8 +5,8 @@
  * SPDX-License-Identifier: BSD-3-Clause
  */
 // clang-format on
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/builder.h>
 

benchmarks/cpp/transpose.cpp

@@ -6,8 +6,8 @@
  */
 // clang-format on
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/utils.h>
 #include <ops/all_ops.h>

benchmarks/cpp/utils.h

@@ -9,8 +9,8 @@
 
 #include <csrc/exceptions.h>
 #include <device_lower/lower2device.h>
-#include <executor.h>
 #include <fusion.h>
+#include <fusion_executor/executor.h>
 #include <ir/all_nodes.h>
 #include <ir/utils.h>
 #include <kernel_cache.h>

csrc/device_lower/analysis/bank_conflict.h

@@ -8,7 +8,7 @@
 #pragma once
 
 #include <exceptions.h>
-#include <executor_params.h>
+#include <fusion_executor/executor_params.h>
 #include <ir/base_nodes.h>
 #include <kernel.h>
 #include <polymorphic_value.h>

csrc/device_lower/lower2device.h

@@ -23,8 +23,8 @@
 #include <device_lower/pass/scalar_hoist.h>
 #include <device_lower/pass/warp_reduce.h>
 #include <exceptions.h>
-#include <executor_params.h>
 #include <expr_simplifier.h>
+#include <fusion_executor/executor_params.h>
 #include <id_model/id_model.h>
 #include <id_model/indexing.h>
 #include <ir/all_nodes.h>

csrc/device_lower/pass/loop_rotation.h

@@ -8,7 +8,7 @@
 #pragma once
 
 #include <exceptions.h>
-#include <executor_params.h>
+#include <fusion_executor/executor_params.h>
 #include <ir/all_nodes.h>
 
 namespace nvfuser {

csrc/dynamic_transform.cpp

@@ -7,10 +7,10 @@
 // clang-format on
 #include <device_lower/utils.h>
 #include <dynamic_transform.h>
-#include <executor_kernel_arg.h>
-#include <executor_utils.h>
 #include <expr_evaluator.h>
 #include <fusion.h>
+#include <fusion_executor/executor_kernel_arg.h>
+#include <fusion_executor/executor_utils.h>
 #include <ir/cloner.h>
 #include <ir/utils.h>
 #include <logical_domain_map.h>

csrc/evaluator_common.cpp

@@ -9,8 +9,8 @@
 
 #include <debug.h>
 #include <device_lower/lower2device.h>
-#include <executor_kernel_arg.h>
 #include <expr_evaluator.h>
+#include <fusion_executor/executor_kernel_arg.h>
 #include <instrumentation.h>
 #include <ir/utils.h>
 #include <tensor_metadata.h>

csrc/evaluator_common.h

@@ -8,8 +8,8 @@
 #pragma once
 #include <device_lower/lower2device.h>
 #include <exceptions.h>
-#include <executor_params.h>
 #include <fusion.h>
+#include <fusion_executor/executor_params.h>
 #include <ir/all_nodes.h>
 #include <polymorphic_value.h>
 #include <utils.h>

csrc/fusion.cpp

@@ -10,8 +10,8 @@
 #include <device_lower/analysis/bank_conflict.h>
 #include <device_lower/lower2device.h>
 #include <disjoint_set.h>
-#include <executor_params.h>
 #include <fusion.h>
+#include <fusion_executor/executor_params.h>
 #include <fusion_segmenter.h>
 #include <host_ir/container.h>
 #include <instrumentation.h>

csrc/fusion.h

@@ -11,7 +11,7 @@
 #include <exceptions.h>
 
 #include <debug.h>
-#include <executor_params.h>
+#include <fusion_executor/executor_params.h>
 #include <ir/base_nodes.h>
 #include <ir/container.h>
 #include <iter_visitor.h>

csrc/executor.cpp → csrc/fusion_executor/executor.cpp

@@ -6,14 +6,14 @@
  */
 // clang-format on
 
-#include <executor.h>
+#include <fusion_executor/executor.h>
 
 #include <codegen.h>
 #include <debug.h>
 #include <device_lower/analysis/bank_conflict.h>
 #include <driver_api.h>
-#include <executor_kernel_arg.h>
-#include <executor_utils.h>
+#include <fusion_executor/executor_kernel_arg.h>
+#include <fusion_executor/executor_utils.h>
 #include <fusion_profiler.h>
 #include <global_allocator.h>
 #include <instrumentation.h>