【custom】split Custom depend header

paddle/cinn/hlir/dialect/operator/transforms/check_infer_symbolic_util.cc

@@ -19,6 +19,7 @@
 #include <functional>
 #include <memory>
 #include <optional>
+#include <random>
 #include "paddle/cinn/hlir/dialect/operator/transforms/check_infer_symbolic_pass.h"
 #include "paddle/cinn/hlir/dialect/operator/transforms/local_infer_symbolic_util.h"
 #include "paddle/cinn/hlir/dialect/operator/transforms/split_generate_shape_into_shape_ops_pass.h"

paddle/fluid/framework/new_executor/garbage_collector/garbage_collector.h

@@ -18,6 +18,7 @@
 #include "paddle/common/errors.h"
 #include "paddle/common/flags.h"
 #include "paddle/fluid/framework/new_executor/instruction/instruction_base.h"
+#include "paddle/fluid/framework/new_executor/new_executor_defs.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/phi/core/memory/allocation/spin_lock.h"
 #include "paddle/phi/core/platform/device_event.h"

paddle/fluid/framework/new_executor/instruction/cinn_jit_instruction.h

@@ -16,6 +16,7 @@
 
 #include <memory>
 #include "paddle/fluid/framework/new_executor/instruction/instruction_base.h"
+#include "paddle/phi/core/ddim.h"
 
 namespace pir {
 class Operation;

paddle/fluid/framework/new_executor/instruction/control_flow/if_instruction.h

@@ -16,6 +16,7 @@
 
 #include "paddle/fluid/framework/new_executor/instruction/instruction_base.h"
 #include "paddle/fluid/framework/new_executor/interpreter/execution_config.h"
+#include "paddle/fluid/framework/new_executor/new_executor_defs.h"
 
 namespace ir {
 class Operation;

paddle/fluid/framework/new_executor/instruction/control_flow/while_instruction.h

@@ -18,6 +18,7 @@
 
 #include "paddle/fluid/framework/new_executor/instruction/instruction_base.h"
 #include "paddle/fluid/framework/new_executor/interpreter/execution_config.h"
+#include "paddle/fluid/framework/new_executor/new_executor_defs.h"
 
 namespace ir {
 class Operation;

paddle/fluid/framework/new_executor/instruction/instruction_base.h

@@ -19,7 +19,7 @@
 #include <unordered_map>
 #include <vector>
 
-#include "paddle/fluid/framework/new_executor/new_executor_defs.h"
+#include "paddle/fluid/framework/new_executor/instruction/instruction_defs.h"
 #include "paddle/phi/api/profiler/event.h"
 
 namespace pir {
@@ -29,8 +29,8 @@ class Value;
 namespace paddle {
 namespace framework {
 class ValueExecutionInfo;
-
-using SchedulingPriority = int64_t;
+class Variable;
+class Scope;
 
 class InstructionBase {
  public:

paddle/fluid/framework/new_executor/instruction/instruction_defs.h

@@ -0,0 +1,52 @@
+// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+#pragma once
+
+#include <unordered_set>
+#include "paddle/phi/core/platform/device_event_base.h"
+#include "paddle/pir/include/core/operation.h"
+
+namespace paddle {
+namespace framework {
+
+using SchedulingPriority = int64_t;
+
+constexpr const char* kCoalesceTensor = "coalesce_tensor";
+
+// stream types
+constexpr const char* kCustomStream = "CustomStream";
+constexpr const char* kDefaultStream = "DefaultStream";
+constexpr const char* kD2HStream = "D2HStream";
+constexpr const char* kH2DStream = "H2DStream";
+
+constexpr int kEmptyVarIndex = 0;
+
+struct EventInter {
+  explicit EventInter(size_t instr_id,
+                      std::shared_ptr<platform::DeviceEvent> event,
+                      platform::DeviceType waiter_type)
+      : instr_id_(instr_id), event_(event), waiter_type_(waiter_type) {}
+  size_t instr_id_;
+  std::shared_ptr<platform::DeviceEvent> event_;
+  platform::DeviceType waiter_type_;
+};
+
+enum class OpFuncType {
+  kCpuSync,  // CPU kernel, block host
+  kGpuSync,  // GPU or other device kernel without asynchronous operation
+  kGpuAsync  // GPU or other device kernel with asynchronous operation
+};
+
+}  // namespace framework
+}  // namespace paddle

paddle/fluid/framework/new_executor/instruction/legacy_kernel_instruction.h

@@ -15,6 +15,7 @@
 #pragma once
 
 #include "paddle/fluid/framework/new_executor/instruction/instruction_base.h"
+#include "paddle/fluid/framework/new_executor/new_executor_defs.h"
 
 namespace pir {
 class Operation;

paddle/fluid/framework/new_executor/instruction/onednn/onednn_instruction.h

@@ -15,6 +15,7 @@
 #pragma once
 
 #include "paddle/fluid/framework/new_executor/instruction/instruction_base.h"
+#include "paddle/fluid/framework/new_executor/new_executor_defs.h"
 
 namespace pir {
 class Operation;

paddle/fluid/framework/new_executor/instruction/onednn/onednn_legacy_instruction.h

@@ -15,6 +15,7 @@
 #pragma once
 
 #include "paddle/fluid/framework/new_executor/instruction/instruction_base.h"
+#include "paddle/fluid/framework/new_executor/new_executor_defs.h"
 
 namespace pir {
 class Operation;

paddle/fluid/framework/new_executor/instruction/phi_kernel_instruction.h

@@ -15,6 +15,7 @@
 #pragma once
 
 #include "paddle/fluid/framework/new_executor/instruction/instruction_base.h"
+#include "paddle/fluid/framework/new_executor/new_executor_defs.h"
 
 namespace pir {
 class Operation;

paddle/fluid/framework/new_executor/new_executor_defs.h

@@ -19,6 +19,7 @@
 #include <vector>
 
 #include "paddle/fluid/framework/infershape_utils.h"
+#include "paddle/fluid/framework/new_executor/instruction/instruction_defs.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/variable_helper.h"
 #include "paddle/fluid/pir/dialect/operator/interface/infermeta.h"
@@ -48,18 +49,6 @@ using HookFunc = std::function<void(OperatorBase*, Scope*)>;
 using PirHookFunc =
     std::function<void(InstructionBase*, ValueExecutionInfo*, Scope*)>;
 
-using SchedulingPriority = int64_t;
-
-constexpr const char* kCoalesceTensor = "coalesce_tensor";
-
-// stream types
-constexpr const char* kCustomStream = "CustomStream";
-constexpr const char* kDefaultStream = "DefaultStream";
-constexpr const char* kD2HStream = "D2HStream";
-constexpr const char* kH2DStream = "H2DStream";
-
-constexpr int kEmptyVarIndex = 0;
-
 struct OpKernelFunc {
   OpKernelComputeFunc compute_func_;
 };
@@ -152,22 +141,6 @@ class VariableScope {
   std::vector<std::pair<std::string, int>> data_transfer_added_vars_;
 };
 
-struct EventInter {
-  explicit EventInter(size_t instr_id,
-                      std::shared_ptr<platform::DeviceEvent> event,
-                      platform::DeviceType waiter_type)
-      : instr_id_(instr_id), event_(event), waiter_type_(waiter_type) {}
-  size_t instr_id_;
-  std::shared_ptr<platform::DeviceEvent> event_;
-  platform::DeviceType waiter_type_;
-};
-
-enum class OpFuncType {
-  kCpuSync,  // CPU kernel, block host
-  kGpuSync,  // GPU or other device kernel without asynchronous operation
-  kGpuAsync  // GPU or other device kernel with asynchronous operation
-};
-
 struct OpFuncNode {
   int stream_priority_{0};  // lower value, higher priority
   // fit for phi kernel

paddle/fluid/ir_adaptor/translator/pd_op_sig.h

@@ -0,0 +1,45 @@
+// Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <string>
+#include <vector>
+
+namespace paddle {
+namespace dialect {
+struct PdOpSig {
+  std::string name;
+  std::vector<std::string> inputs;
+  std::vector<std::string> outputs;
+  PdOpSig() = default;
+  PdOpSig(const PdOpSig& input_info) = default;
+
+  PdOpSig(const std::string& name,
+          const std::vector<std::string>& inputs,
+          const std::vector<std::string>& outputs)
+      : name(name), inputs(inputs), outputs(outputs) {}
+};
+
+bool HaveOpToMultiKernelsMap(std::string op_name);
+
+const std::vector<PdOpSig>& LegacyOpToPdOpsMapping(std::string op_name);
+const std::vector<PdOpSig>& SparseOpToPdOpsMapping(std::string op_name);
+
+#ifdef PADDLE_WITH_DNNL
+bool IsOneDNNOnlyOp(std::string op_name);
+#endif
+
+}  // namespace dialect
+}  // namespace paddle

paddle/fluid/ir_adaptor/translator/utils.h

@@ -19,38 +19,12 @@
 
 #include "paddle/fluid/framework/data_type.h"
 #include "paddle/fluid/framework/program_desc.h"
+#include "paddle/fluid/ir_adaptor/translator/pd_op_sig.h"
 #include "paddle/fluid/ir_adaptor/translator/program_translator.h"
 #include "paddle/pir/include/core/ir_context.h"
 #include "paddle/pir/include/core/operation.h"
 #include "paddle/pir/include/core/program.h"
 
-namespace paddle {
-namespace dialect {
-struct PdOpSig {
-  std::string name;
-  std::vector<std::string> inputs;
-  std::vector<std::string> outputs;
-  PdOpSig() = default;
-  PdOpSig(const PdOpSig& input_info) = default;
-
-  PdOpSig(const std::string& name,
-          const std::vector<std::string>& inputs,
-          const std::vector<std::string>& outputs)
-      : name(name), inputs(inputs), outputs(outputs) {}
-};
-
-bool HaveOpToMultiKernelsMap(std::string op_name);
-
-const std::vector<PdOpSig>& LegacyOpToPdOpsMapping(std::string op_name);
-const std::vector<PdOpSig>& SparseOpToPdOpsMapping(std::string op_name);
-
-#ifdef PADDLE_WITH_DNNL
-bool IsOneDNNOnlyOp(std::string op_name);
-#endif
-
-}  // namespace dialect
-}  // namespace paddle
-
 namespace paddle {
 namespace translator {
 

paddle/fluid/pir/dialect/op_generator/op_gen.py

@@ -113,8 +113,7 @@
 #ifdef PADDLE_WITH_DNNL
 #include "paddle/fluid/pir/dialect/operator/trait/onednn.h"
 #endif
-#include "paddle/fluid/framework/infershape_utils.h"
-#include "paddle/fluid/ir_adaptor/translator/utils.h"
+#include "paddle/fluid/ir_adaptor/translator/pd_op_sig.h"
 #include "paddle/fluid/pir/dialect/operator/ir/manual_op.h"
 #include "paddle/fluid/pir/dialect/operator/trait/custom_vjp.h"
 #include "paddle/phi/core/infermeta_utils.h"

paddle/fluid/pir/dialect/op_generator/op_kerneltype_gen.py

@@ -44,7 +44,7 @@
 
 OP_COMPLEX_PROMOTE_CHECK_TEMPLATE = """
   // deal complex_promote
-  if (framework::IsComplexType(expected_kernel_dtype)) {{
+  if (phi::IsComplexType(expected_kernel_dtype)) {{
     // only promote inputs’s types when contains complex input
     return tensor_dtype;
   }}

paddle/fluid/pir/dialect/operator/ir/manual_op.h

@@ -15,7 +15,6 @@
 #pragma once
 #include <vector>
 
-#include "paddle/fluid/framework/infershape_utils.h"
 #include "paddle/fluid/pir/dialect/operator/interface/decomp.h"
 #include "paddle/fluid/pir/dialect/operator/interface/get_kernel_type_for_var.h"
 #include "paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/infer_symbolic_shape.h"

paddle/fluid/pir/transforms/gpu/conv2d_add_fuse_pass.cc

@@ -16,6 +16,9 @@
 #include "paddle/fluid/pir/dialect/operator/ir/op_attribute.h"
 #include "paddle/fluid/pir/dialect/operator/ir/pd_op.h"
 #include "paddle/fluid/pir/drr/include/drr_pattern_base.h"
+#ifdef PADDLE_WITH_CUDA
+#include "paddle/phi/core/platform/device/gpu/gpu_info.h"
+#endif
 
 #include "paddle/fluid/pir/utils/general_functions.h"
 #include "paddle/pir/include/pass/pass.h"

paddle/fluid/pir/transforms/gpu/fused_flash_attn_pass.cc

@@ -17,6 +17,9 @@
 #include "paddle/fluid/pir/dialect/operator/ir/pd_op.h"
 #include "paddle/fluid/pir/drr/include/drr_pattern_base.h"
 #include "paddle/fluid/pir/utils/general_functions.h"
+#ifdef PADDLE_WITH_CUDA
+#include "paddle/phi/core/platform/device/gpu/gpu_info.h"
+#endif
 
 #include "paddle/pir/include/pass/pass.h"
 #include "paddle/pir/include/pass/pass_registry.h"

paddle/fluid/pir/transforms/sub_graph_detector.cc

@@ -16,6 +16,7 @@
 
 #include <memory>
 
+#include <climits>
 #include <iterator>
 #include <queue>
 #include <regex>

paddle/fluid/pir/transforms/xpu/conv2d_bn_xpu_fuse_pass.cc

@@ -13,6 +13,8 @@
 // limitations under the License.
 
 #include "paddle/fluid/pir/transforms/xpu/conv2d_bn_xpu_fuse_pass.h"
+#include "paddle/fluid/framework/infershape_utils.h"
+#include "paddle/fluid/ir_adaptor/translator/utils.h"
 #include "paddle/fluid/pir/dialect/operator/ir/pd_op.h"
 #include "paddle/fluid/pir/drr/include/drr_pattern_base.h"
 #include "paddle/fluid/pir/utils/general_functions.h"

python/setup.py.in

@@ -1174,7 +1174,7 @@ headers = (
     list(find_files('token.h','@PADDLE_SOURCE_DIR@/paddle/pir/src/core/parser/'))+
     #pir ops and dependency
     list(find_files('pd_op.h','@PADDLE_BINARY_DIR@/paddle/fluid/pir/dialect/operator/ir/'))+
-    list(find_files('utils.h','@PADDLE_SOURCE_DIR@/paddle/fluid/ir_adaptor/translator/'))+
+    list(find_files('utils_defs.h','@PADDLE_SOURCE_DIR@/paddle/fluid/ir_adaptor/translator/'))+
     list(find_files('*.h','@PADDLE_SOURCE_DIR@/paddle/fluid/pir/dialect/operator/interface/'))+
     list(find_files('*.hpp','@PADDLE_SOURCE_DIR@/paddle/fluid/pir/dialect/operator/interface/'))+
     list(find_files('*.h','@PADDLE_SOURCE_DIR@/paddle/fluid/pir/dialect/operator/trait/'))+
@@ -1187,7 +1187,7 @@ headers = (
     #new_executor headers
     list(find_files('pir_adaptor_util.h','@PADDLE_SOURCE_DIR@/paddle/fluid/framework/new_executor/pir_adaptor/'))+
     list(find_files('custom_engine_instruction.h','@PADDLE_SOURCE_DIR@/paddle/fluid/framework/new_executor/instruction/'))+
-    list(find_files('new_executor_defs.h','@PADDLE_SOURCE_DIR@/paddle/fluid/framework/new_executor/'))+
+    list(find_files('instruction_defs.h','@PADDLE_SOURCE_DIR@/paddle/fluid/framework/new_executor/'))+
     list(find_files('instruction_base.h','@PADDLE_SOURCE_DIR@/paddle/fluid/framework/new_executor/instruction/')))
 jit_layer_headers = ['layer.h', 'serializer.h', 'serializer_utils.h', 'all.h', 'function.h']
 for f in jit_layer_headers: