Feat/issue 356 aicpu launch new interface

.gitignore

@@ -41,3 +41,7 @@ compile_commands.json
 # Built nanobind extensions
 python/_task_interface*.so
 python/_task_interface*.dylib
+
+# Log files
+*.log
+profiling_logs_*/

python/simpler/runtime_compiler.py

@@ -233,6 +233,12 @@ def _build(actual_build_dir: str) -> Union[bytes, Path]:
                 od.mkdir(parents=True, exist_ok=True)
                 dest = od / binary_name
                 shutil.copy2(binary_path, dest)
+                dispatcher_so = Path(actual_build_dir) / "libaicpu_dispatcher.so"
+                if dispatcher_so.is_file():
+                    dest_dispatcher = od / "libaicpu_dispatcher.so"
+                    shutil.copy2(dispatcher_so, dest_dispatcher)
+                    # Strip debug info to match CANN built-in SO format
+                    subprocess.run(["strip", "-s", str(dest_dispatcher)], check=True)
                 return dest
             else:
                 with open(binary_path, "rb") as f:

src/a2a3/platform/onboard/aicpu/CMakeLists.txt

@@ -10,6 +10,7 @@ project(aicpu_kernel LANGUAGES C CXX)
 set(CMAKE_CUSTOM_INCLUDE_DIRS "")
 list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${CMAKE_CURRENT_SOURCE_DIR}/../../include")
 list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${CMAKE_CURRENT_SOURCE_DIR}/../../../../common/task_interface")
+list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${CMAKE_CURRENT_SOURCE_DIR}/../../../../common/aicpu_dispatcher")
 if(DEFINED CUSTOM_INCLUDE_DIRS)
     foreach(INC_DIR ${CUSTOM_INCLUDE_DIRS})
         list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${INC_DIR}")
@@ -75,3 +76,38 @@ target_link_directories(aicpu_kernel
 # Output name
 set_target_properties(aicpu_kernel PROPERTIES OUTPUT_NAME aicpu_kernel)
 
+# Build dispatcher SO (two-layer architecture)
+set(AICPU_DISPATCHER_SOURCES
+    "${CMAKE_CURRENT_SOURCE_DIR}/../../../../common/aicpu_dispatcher/aicpu_dispatcher.cpp"
+)
+add_library(aicpu_dispatcher SHARED ${AICPU_DISPATCHER_SOURCES})
+
+# Compiler options for dispatcher (same as AICPU kernel)
+target_compile_options(aicpu_dispatcher
+    PRIVATE
+        -Wall
+        -Wextra
+        -rdynamic
+        -O3
+        -fPIC
+        -g
+        $<$<COMPILE_LANGUAGE:CXX>:-std=gnu++17>
+)
+
+# Include directories for dispatcher
+target_include_directories(aicpu_dispatcher
+    PRIVATE
+        ${CMAKE_CURRENT_SOURCE_DIR}
+        ${CMAKE_CUSTOM_INCLUDE_DIRS}
+        ${ASCEND_HOME_PATH}/include
+)
+
+# Link against dl for dlopen/dlsym
+target_link_libraries(aicpu_dispatcher PRIVATE dl)
+
+# Match CANN built-in SO properties: SYMBOLIC flag, build-id, stripped
+set_target_properties(aicpu_dispatcher PROPERTIES
+    LINK_FLAGS "-Wl,-Bsymbolic,--build-id"
+    OUTPUT_NAME "aicpu_dispatcher"
+)
+

src/a2a3/platform/onboard/host/CMakeLists.txt

@@ -19,6 +19,8 @@ set(CMAKE_CUSTOM_INCLUDE_DIRS "")
 list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${CMAKE_CURRENT_SOURCE_DIR}/../../include")
 list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${CMAKE_CURRENT_SOURCE_DIR}/../../../../common/task_interface")
 list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${CMAKE_CURRENT_SOURCE_DIR}/../../../../common/worker")
+list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${CMAKE_CURRENT_SOURCE_DIR}/../../../../common/host")
+list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${CMAKE_CURRENT_SOURCE_DIR}/../../../../common/aicpu_dispatcher")
 if(DEFINED CUSTOM_INCLUDE_DIRS)
     foreach(INC_DIR ${CUSTOM_INCLUDE_DIRS})
         list(APPEND CMAKE_CUSTOM_INCLUDE_DIRS "${INC_DIR}")
@@ -35,10 +37,15 @@ list(APPEND HOST_RUNTIME_SOURCES
     "${CMAKE_CURRENT_SOURCE_DIR}/pto_runtime_c_api.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/platform_compile_info.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/host_regs.cpp"
+    "${CMAKE_CURRENT_SOURCE_DIR}/aicpu_loader.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/../../src/host/host_log.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/../../src/host/unified_log_host.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/../../src/host/performance_collector.cpp"
 )
+# Add common/host sources (LoadAicpuOp)
+list(APPEND HOST_RUNTIME_SOURCES
+    "${CMAKE_CURRENT_SOURCE_DIR}/../../../../common/host/load_aicpu_op.cpp"
+)
 if(DEFINED CUSTOM_SOURCE_DIRS)
     foreach(SRC_DIR ${CUSTOM_SOURCE_DIRS})
         file(GLOB DIR_SOURCES "${SRC_DIR}/*.cpp" "${SRC_DIR}/*.c")
@@ -84,11 +91,16 @@ target_include_directories(host_runtime
         ${ASCEND_HOME_PATH}/${CMAKE_SYSTEM_PROCESSOR}-linux/include/driver
 )
 
-target_link_directories(host_runtime
-    PRIVATE
-        ${ASCEND_HOME_PATH}/lib64
-        ${ASCEND_HOME_PATH}/runtime/lib64
-)
+# Conditional compilation for new CANN interface
+option(BUILD_WITH_NEW_CANN "Use new rtsLaunchCpuKernel interface (CANN 7.0+)" ON)
+if(BUILD_WITH_NEW_CANN)
+    target_compile_definitions(host_runtime PRIVATE BUILD_WITH_NEW_CANN)
+    # Add additional include path for new RTS headers (CANN 7.0+)
+    target_include_directories(host_runtime PRIVATE
+        ${ASCEND_HOME_PATH}/pkg_inc/runtime/runtime
+    )
+    message(STATUS "Building with new CANN rtsLaunchCpuKernel interface")
+endif()
 
 # Link against CANN runtime libraries
 # ascend_hal is dynamically loaded at runtime via dlopen in device_runner
@@ -100,4 +112,10 @@ target_link_libraries(host_runtime
         dl
 )
 
+target_link_directories(host_runtime
+    PRIVATE
+        ${ASCEND_HOME_PATH}/lib64
+        ${ASCEND_HOME_PATH}/runtime/lib64
+)
+
 set_target_properties(host_runtime PROPERTIES OUTPUT_NAME "host_runtime")

src/a2a3/platform/onboard/host/aicpu_loader.cpp

@@ -0,0 +1,71 @@
+/*
+ * Copyright (c) PyPTO Contributors.
+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
+ * CANN Open Software License Agreement Version 2.0 (the "License").
+ * Please refer to the License for details. You may not use this file except in compliance with the License.
+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See LICENSE in the root of the software repository for the full text of the License.
+ * -----------------------------------------------------------------------------------------------------------
+ */
+/**
+ * AICPU Loader Implementation (Legacy Interface)
+ *
+ * Provides AICPU kernel launching via the legacy rtAicpuKernelLaunchExWithArgs API.
+ * Used when BUILD_WITH_NEW_CANN is OFF. When BUILD_WITH_NEW_CANN is ON,
+ * device_runner uses LoadAicpuOp (src/common/host/load_aicpu_op.h) instead.
+ */
+
+#include "aicpu_loader.h"
+
+#include <cstring>
+
+#include "common/unified_log.h"
+#include "common/kernel_args.h"
+
+int AicpuLoader::init_with_binary(
+    const std::vector<uint8_t> &aicpu_binary, const std::vector<std::string> &kernel_names
+) {
+    // Legacy interface: No pre-loading needed
+    (void)aicpu_binary;
+    (void)kernel_names;
+    LOG_INFO("AicpuLoader: Using legacy rtAicpuKernelLaunchExWithArgs interface");
+    return 0;
+}
+
+int AicpuLoader::init(const std::string &so_path, const std::vector<std::string> &kernel_names) {
+    // Legacy interface: No pre-loading needed
+    (void)so_path;
+    (void)kernel_names;
+    LOG_INFO("AicpuLoader: Using legacy rtAicpuKernelLaunchExWithArgs interface");
+    return 0;
+}
+
+int AicpuLoader::launch(rtStream_t stream, KernelArgs *k_args, const char *kernel_name, int aicpu_num) {
+    // Legacy interface: rtAicpuKernelLaunchExWithArgs
+    struct Args {
+        KernelArgs k_args;
+        char kernel_name[32];
+        const char so_name[32] = {"libaicpu_extend_kernels.so"};
+        const char op_name[32] = {""};
+    } args;
+
+    args.k_args = *k_args;
+    std::strncpy(args.kernel_name, kernel_name, sizeof(args.kernel_name) - 1);
+    args.kernel_name[sizeof(args.kernel_name) - 1] = '\0';
+
+    rtAicpuArgsEx_t rt_args;
+    std::memset(&rt_args, 0, sizeof(rt_args));
+    rt_args.args = &args;
+    rt_args.argsSize = sizeof(args);
+    rt_args.kernelNameAddrOffset = offsetof(struct Args, kernel_name);
+    rt_args.soNameAddrOffset = offsetof(struct Args, so_name);
+
+    return rtAicpuKernelLaunchExWithArgs(
+        rtKernelType_t::KERNEL_TYPE_AICPU_KFC, "AST_DYN_AICPU", aicpu_num, &rt_args, nullptr, stream, 0
+    );
+}
+
+void AicpuLoader::finalize() {
+    // Legacy interface: No-op
+}

src/a2a3/platform/onboard/host/aicpu_loader.h

@@ -0,0 +1,69 @@
+/*
+ * Copyright (c) PyPTO Contributors.
+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
+ * CANN Open Software License Agreement Version 2.0 (the "License").
+ * Please refer to the License for details. You may not use this file except in compliance with the License.
+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See LICENSE in the root of the software repository for the full text of the License.
+ * -----------------------------------------------------------------------------------------------------------
+ */
+/**
+ * AICPU Loader Abstraction (Legacy Interface)
+ *
+ * Provides AICPU kernel launching via the legacy rtAicpuKernelLaunchExWithArgs API.
+ * Used when BUILD_WITH_NEW_CANN is OFF. When BUILD_WITH_NEW_CANN is ON,
+ * device_runner uses LoadAicpuOp (src/common/host/load_aicpu_op.h) instead.
+ */
+
+#ifndef A2A3_PLATFORM_ONBOARD_HOST_AICPU_LOADER_H_
+#define A2A3_PLATFORM_ONBOARD_HOST_AICPU_LOADER_H_
+
+#include <cstdint>
+#include <string>
+#include <vector>
+
+#include <runtime/rt.h>
+
+// Forward declarations
+struct KernelArgs;
+
+/**
+ * @brief AICPU kernel loader (legacy interface)
+ *
+ * Launches AICPU kernels via the legacy rtAicpuKernelLaunchExWithArgs API.
+ * Used as the fallback when BUILD_WITH_NEW_CANN is OFF.
+ */
+class AicpuLoader {
+public:
+    AicpuLoader() = default;
+    ~AicpuLoader() = default;
+
+    /**
+     * @brief Initialize the AICPU loader with binary data (no-op for legacy interface)
+     */
+    int init_with_binary(const std::vector<uint8_t> &aicpu_binary, const std::vector<std::string> &kernel_names);
+
+    /**
+     * @brief Initialize the AICPU loader (no-op for legacy interface)
+     */
+    int init(const std::string &so_path, const std::vector<std::string> &kernel_names);
+
+    /**
+     * @brief Launch an AICPU kernel via legacy rtAicpuKernelLaunchExWithArgs
+     */
+    int launch(rtStream_t stream, KernelArgs *k_args, const char *kernel_name, int aicpu_num);
+
+    /**
+     * @brief Cleanup resources (no-op for legacy interface)
+     */
+    void finalize();
+
+    // Disable copy and move
+    AicpuLoader(const AicpuLoader &) = delete;
+    AicpuLoader &operator=(const AicpuLoader &) = delete;
+    AicpuLoader(AicpuLoader &&) = delete;
+    AicpuLoader &operator=(AicpuLoader &&) = delete;
+};
+
+#endif  // A2A3_PLATFORM_ONBOARD_HOST_AICPU_LOADER_H_