g++ (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 changes

.gitmodules

@@ -0,0 +1,18 @@
+[submodule "vendor/Catch2"]
+	path = vendor/Catch2
+	url = https://github.com/catchorg/Catch2
+[submodule "vendor/fmt"]
+	path = vendor/fmt
+	url = https://github.com/fmtlib/fmt
+[submodule "vendor/jsoncpp"]
+	path = vendor/jsoncpp
+	url = https://github.com/open-source-parsers/jsoncpp
+[submodule "vendor/libevent"]
+	path = vendor/libevent
+	url = https://github.com/libevent/libevent
+[submodule "vendor/plog"]
+	path = vendor/plog
+	url = https://github.com/SergiusTheBest/plog
+[submodule "vendor/yaml-cpp"]
+	path = vendor/yaml-cpp
+	url = https://github.com/jbeder/yaml-cpp

CMakeLists.txt

@@ -1,4 +1,14 @@
 # Copyright (c) 2023, NVIDIA CORPORATION.  All rights reserved.
+set (CMAKE_CUDA_ARCHITECTURES "native")
+set (CUDA_TOOLKIT_ROOT_DIR "/usr/local/cuda-12.3")
+set (CMAKE_CUDA_COMPILER "/usr/local/cuda-12.3/bin/nvcc")
+list(APPEND LIBRARIES ${CUDA_CUBLAS_LIBRARIES})
+enable_language(CUDA)
+
+set(CUDA12_INCLUDE_DIR "/usr/local/cuda-12.3/include")
+set(CUDA12_INCLUDE_DIR "/usr/local/cuda-12.3/include")
+include_directories("/usr/local/cuda-12.3/targets/x86_64-linux/include")
+include_directories(${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -312,6 +322,8 @@ add_custom_command(
 )
 add_custom_target(dcgm_decode_db ALL DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/dcgm_decode_db.txt)
 
+
+
 set(DCGM_PRIVATE_DIR "${CMAKE_SOURCE_DIR}/dcgm_private")
 if (EXISTS ${DCGM_PRIVATE_DIR})
     add_subdirectory(${DCGM_PRIVATE_DIR})
@@ -447,3 +459,4 @@ configure_file(
     @ONLY)
 
 include(CPack)
+cmake_minimum_required(VERSION 3.18)

cmake/FindCuda.cmake

@@ -16,8 +16,8 @@
 
 include(utils)
 
-set(Cuda10_prefix usr/local/cuda-10.0)
-set(Cuda11_prefix usr/local/cuda-11.8)
+#set(Cuda10_prefix usr/local/cuda-10.0)
+#set(Cuda11_prefix usr/local/cuda-11.8)
 set(Cuda12_prefix usr/local/cuda-12.0)
 
 macro (load_cuda cuda_version)
@@ -95,18 +95,18 @@ macro (load_cuda cuda_version)
 
 endmacro()
 
-if (NOT DEFINED CUDA10_INCLUDE_DIR AND NOT ${CMAKE_SYSTEM_PROCESSOR} STREQUAL "aarch64")
-    load_cuda(10)
-endif()
+#if (NOT DEFINED CUDA10_INCLUDE_DIR AND NOT ${CMAKE_SYSTEM_PROCESSOR} STREQUAL "aarch64")
+#    load_cuda(10)
+#endif()
 
-if (NOT DEFINED CUDA11_INCLUDE_DIR)
-    load_cuda(11)
-endif()
+#if (NOT DEFINED CUDA11_INCLUDE_DIR)
+#    load_cuda(11)
+#endif()
 
 if (NOT DEFINED CUDA12_INCLUDE_DIR)
     load_cuda(12)
 endif()
 
-unset(Cuda10_prefix)
-unset(Cuda11_prefix)
+#unset(Cuda10_prefix)
+#unset(Cuda11_prefix)
 unset(Cuda12_prefix)

cmake/FindJsoncpp.cmake

@@ -16,8 +16,8 @@
 
 if (NOT TARGET JsonCpp::JsonCpp)
     find_package(jsoncpp REQUIRED CONFIG)
-    set(JSONCPP_STATIC_LIBS jsoncpp_static)
-    set(JSONCPP_INCLUDE_PATH $<TARGET_PROPERTY:jsoncpp_static,INTERFACE_INCLUDE_DIRECTORIES>)
+    set(JSONCPP_STATIC_LIBS jsoncpp)
+    set(JSONCPP_INCLUDE_PATH $<TARGET_PROPERTY:jsoncpp,INTERFACE_INCLUDE_DIRECTORIES>)
 endif()
 # set(Jsoncpp_PATH_PREFIXES /usr/local "${Jsoncpp_ROOT}" "$ENV{HOME}")
 # foreach(prefix ${Jsoncpp_PATH_PREFIXES})

cmake/FindLibevent.cmake

@@ -1,61 +1,18 @@
-#
-# Copyright (c) 2023, NVIDIA CORPORATION.  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.
-#
-# - Find Libevent (a cross event library)
-# This module defines
-# LIBEVENT_INCLUDE_DIR, where to find Libevent headers
-# LIBEVENT_STATIC_LIBS, Libevent static libraries
-# Libevent_FOUND, If false, do not try to use libevent
+include(FindPackageHandleStandardArgs)
 
-set(Libevent_EXTRA_PREFIXES / /lib /lib64 /usr/local /opt/local "$ENV{HOME}" "${Libevent_ROOT}")
-foreach(prefix ${Libevent_EXTRA_PREFIXES})
-  list(APPEND Libevent_INCLUDE_PATHS "${prefix}/include")
-  list(APPEND Libevent_LIB_PATHS "${prefix}/lib" "${prefix}/lib64")
-endforeach()
+find_library(LIBEVENT_LIBRARY event
+  PATHS ${LIBEVENT_LIBRARYDIR})
 
-find_path(LIBEVENT_INCLUDE_DIR evhttp.h event.h PATHS ${Libevent_INCLUDE_PATHS})
-find_library(LIBEVENT_STATIC_LIB NAMES libevent.a libevent_core.a libevent_extra.a PATHS ${Libevent_LIB_PATHS})
-find_library(LIBEVENT_PTHREAD_STATIC_LIB NAMES libevent_pthreads.a PATHS ${Libevent_LIB_PATHS})
+find_path(LIBEVENT_INCLUDE_DIR event.h
+  PATHS ${LIBEVENT_INCLUDEDIR})
 
-if (LIBEVENT_INCLUDE_DIR AND LIBEVENT_STATIC_LIB AND LIBEVENT_PTHREAD_STATIC_LIB)
-    set(Libevent_FOUND TRUE)
-    add_library(libevent_event_static STATIC IMPORTED)
-    set_target_properties(libevent_event_static PROPERTIES IMPORTED_LOCATION ${LIBEVENT_STATIC_LIB})
-    add_library(libevent_event_pthread STATIC IMPORTED)
-    set_target_properties(libevent_event_pthread PROPERTIES IMPORTED_LOCATION ${LIBEVENT_PTHREAD_STATIC_LIB})
-    set(LIBEVENT_STATIC_LIBS libevent_event_static libevent_event_pthread)
-else ()
-    set(Libevent_FOUND FALSE)
-endif ()
-
-if (Libevent_FOUND)
-    if (NOT Libevent_FIND_QUIETLY)
-        message(STATUS "Found libevent: ${LIBEVENT_LIB}")
-    endif ()
-else ()
-    if (Libevent_FIND_REQUIRED)
-        message(FATAL_ERROR "Could NOT find libevent and libevent_pthread.")
-    endif ()
-    message(STATUS "libevent and libevent_pthread NOT found.")
-endif ()
-
-unset(Libevent_EXTRA_PREFIXES)
-unset(LIBEVENT_PTHREAD_STATIC_LIB)
-unset(LIBEVENT_STATIC_LIB)
+find_package_handle_standard_args(libevent DEFAULT_MSG
+  LIBEVENT_LIBRARY
+  LIBEVENT_INCLUDE_DIR)
 
 mark_as_advanced(
-    LIBEVENT_STATIC_LIBS
-    LIBEVENT_INCLUDE_DIR
-    )
+  LIBEVENT_LIBRARY
+  LIBEVENT_INCLUDE_DIR)
+
+set(LIBEVENT_LIBRARIES ${LIBEVENT_LIBRARY})
+set(LIBEVENT_INCLUDE_DIRS ${LIBEVENT_INCLUDE_DIR})

common/CMakeLists.txt

@@ -86,7 +86,11 @@ target_link_libraries(dcgm_common PUBLIC
     sdk_nvml_essentials_objects
     fmt::fmt
     dl
+
 )
+target_link_libraries(dcgm_common  PRIVATE "/usr/local/lib/libevent.so")
+target_link_libraries(dcgm_common  PRIVATE "/usr/local/lib/libevent_pthreads.so")
+
 
 add_library(common_watch_objects STATIC)
 target_sources(common_watch_objects PRIVATE
@@ -97,3 +101,5 @@ target_sources(common_watch_objects PRIVATE
     DcgmWatchTable.h
 )
 target_link_libraries(common_watch_objects PUBLIC common_interface)
+
+

common/CudaWorker/CMakeLists.txt

@@ -31,21 +31,21 @@ macro(define_dcgm_cuda_worker cuda_version)
         CudaWorkerThread.hpp
     )
 
-    if (${cuda_version} GREATER 10)
+    #    if (${cuda_version} GREATER 10)
         target_sources(${dcgm_cuda_worker_lib} PRIVATE DcgmDgemm.cpp)
-    else()
-        target_compile_options(${dcgm_cuda_worker_lib} PRIVATE -Wno-volatile)
-    endif()
+    #else()
+    #    target_compile_options(${dcgm_cuda_worker_lib} PRIVATE -Wno-volatile)
+    #endif()
 
     set(LOCAL_DCGM_CUDA_WORKER "${LOCAL_DCGM_CUDA_WORKER};${dcgm_cuda_worker_lib}" PARENT_SCOPE)
 endmacro()
 
 set(LOCAL_DCGM_CUDA_WORKER "")
 
-if (NOT ${CMAKE_SYSTEM_PROCESSOR} STREQUAL "aarch64")
-    define_dcgm_cuda_worker(10)
-endif()
-define_dcgm_cuda_worker(11)
+#if (NOT ${CMAKE_SYSTEM_PROCESSOR} STREQUAL "aarch64")
+#    define_dcgm_cuda_worker(10)
+#endif()
+#define_dcgm_cuda_worker(11)
 define_dcgm_cuda_worker(12)
 
-set(DCGM_CUDA_WORKER ${LOCAL_DCGM_CUDA_WORKER} PARENT_SCOPE)
+set(DCGM_CUDA_WORKER ${LOCAL_DCGM_CUDA_WORKER} PARENT_SCOPE)

common/CudaWorker/CudaWorkerThread.hpp

@@ -18,6 +18,13 @@
 
 #include <chrono>
 
+
+inline int format_as(cudaError_enum type) {
+  return static_cast<int>(type);
+}
+
+
+
 #include "FieldWorkers.hpp"
 #include <DcgmTaskRunner.h>
 

common/CudaWorker/FieldWorkers.hpp

@@ -15,16 +15,21 @@
  */
 #pragma once
 
-#include <cublas_proxy.hpp>
+//#include <cublas_proxy.hpp>
+#include "../cublas_proxy/cublas_proxy.hpp"
 #include <cuda.h>
 
 #if (CUDA_VERSION_USED >= 11)
 #include "DcgmDgemm.hpp"
 #endif
 
 #include <DcgmLogging.h>
+
 #include <fmt/format.h>
 #include <timelib.h>
+inline int format_as(cublasStatus_t type) {
+  return static_cast<int>(type);
+  }
 
 using namespace Dcgm;
 

common/DcgmLogging.h

@@ -31,7 +31,7 @@
 #include <mutex>
 #include <string>
 #include <syslog.h>
-
+#include "format.hpp" // now define formatting for classes in this module
 #define DCGM_LOGGING_SEVERITY_OPTIONS "NONE, FATAL, ERROR, WARN, INFO, DEBUG, VERB"
 
 #define DCGM_LOGGING_SEVERITY_STRING_VERBOSE "VERB"

common/DcgmMutex.cpp

@@ -21,6 +21,8 @@
 #include <chrono>
 #include <ratio>
 
+inline int format_as(dcgmMutexSt type) {  return static_cast<int>(type);}
+
 /*****************************************************************************/
 DcgmMutex::DcgmMutex(int timeoutMs)
     // Cast to long long to avoid overflowing before widening to a long long

common/DcgmWatchTable.cpp

@@ -18,7 +18,7 @@
 #include "DcgmLogging.h"
 #include "DcgmUtilities.h"
 #include "DcgmWatchTable.h"
-
+inline int format_as(DcgmWatcherType_t type) {  return static_cast<int>(type);}
 /*****************************************************************************/
 DcgmWatchTable::DcgmWatchTable()
     : m_entityWatchHashTable()

common/tests/TaskRunnerTests.cpp

@@ -288,7 +288,7 @@ TEST_CASE("TaskRunner: Limited Queue")
 
     tr.Stop();
     fmt::print("Wait iterations elapsed: {}\n", cWaitIterations - waitIterations);
-    fmt::print("Iterations: {}\nExecutions: {}\nFailed to add: {}\n", iterations, executed, failedToAdd);
+    fmt::print("Iterations: {}\nExecutions: {}\nFailed to add: {}\n", (int)iterations, (int)executed, (int)failedToAdd);
     REQUIRE(executed >= cTaskRunnerCapacity);
     REQUIRE((failedToAdd + executed) == iterations);
 }
@@ -310,4 +310,4 @@ TEST_CASE("TaskRunner: Task with attempts")
     REQUIRE(fut.has_value());
     REQUIRE_THROWS_AS((*fut).get(), std::future_error);
     tr.Stop();
-}
+}

cublas_proxy/CMakeLists.txt

@@ -28,24 +28,30 @@ macro(define_proxy_lib CUDA_VER)
     set(cublas_proxy_include_dir "${CMAKE_CURRENT_SOURCE_DIR}/..")
     get_absolute_path(${cublas_proxy_include_dir} cublas_proxy_include_dir)
     target_include_directories(${lib_name} PUBLIC $<BUILD_INTERFACE:${cublas_proxy_include_dir}>)
+
     target_include_directories(${lib_name} PUBLIC $<BUILD_INTERFACE:${CUDA${CUDA_VER}_INCLUDE_DIR}>)
     target_link_libraries(${lib_name} PRIVATE dl rt)
     target_link_libraries(${lib_name} PRIVATE dcgm dcgm_common dcgm_logging dcgm_mutex)
-    target_link_libraries(${lib_name} PRIVATE ${CUDA${CUDA_VER}_STATIC_CUBLAS_LIBS})
-    target_link_libraries(${lib_name} PRIVATE ${CUDA${CUDA_VER}_STATIC_LIBS})
-    target_link_libraries(${lib_name} PRIVATE ${CUDA${CUDA_VER}_LIBS})
+#    target_link_libraries(${lib_name} PRIVATE ${CUDA${CUDA_VER}_STATIC_CUBLAS_LIBS})
+#    target_link_libraries(${lib_name} PRIVATE ${CUDA${CUDA_VER}_STATIC_LIBS})
+    #    target_link_libraries(${lib_name} PRIVATE ${CUDA_CUBLAS_LIBRARIES}))
+    #set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} 
+    target_link_libraries(${lib_name} PRIVATE  "/usr/local/cuda-12.3/lib64/libcublas.so" )
+    target_link_libraries(${lib_name} PRIVATE  "/usr/local/cuda-12.3/targets/x86_64-linux/lib/libcublasLt.so" )
+
 
     set(LOCAL_CUBLAS_PROXY "${LOCAL_CUBLAS_PROXY};${lib_name}" PARENT_SCOPE)
 
+
 endmacro()
 
 set(LOCAL_CUBLAS_PROXY "")
 
 add_subdirectory(Cuda12)
-add_subdirectory(Cuda11)
-if (NOT ${CMAKE_SYSTEM_PROCESSOR} STREQUAL "aarch64")
-    add_subdirectory(Cuda10)
-endif()
+#add_subdirectory(Cuda11)
+#if (NOT ${CMAKE_SYSTEM_PROCESSOR} STREQUAL "aarch64")
+#    add_subdirectory(Cuda10)
+#endif()
 
 set(DCGM_CUBLAS_PROXY ${LOCAL_CUBLAS_PROXY} PARENT_SCOPE)
 

cublas_proxy/Cuda10/CMakeLists.txt

@@ -11,5 +11,5 @@
 # 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.
-define_proxy_lib(10)
-target_compile_options(${lib_name} PRIVATE -Wno-volatile)
+#define_proxy_lib(10)
+#target_compile_options(${lib_name} PRIVATE -Wno-volatile)

cublas_proxy/Cuda11/CMakeLists.txt

@@ -11,4 +11,4 @@
 # 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.
-define_proxy_lib(11)
+#define_proxy_lib(11)

dcgmlib/format.hpp

@@ -0,0 +1,30 @@
+#ifndef DCGM_FORMAT_H
+#define DCGM_FORMAT_H
+#ifdef DcgmWatcherType_t
+inline int format_as(DcgmWatcherType_t type) {  return static_cast<int>(type);}
+#endif
+inline int format_as(dcgmChipArchitecture_enum type) {  return static_cast<int>(type);}
+inline int format_as(dcgmConfigType_enum type) {  return static_cast<int>(type);}
+inline int format_as(dcgmDiagnosticLevel_t type) {  return static_cast<int>(type);}
+
+inline int format_as(dcgmGroupType_enum type) {  return static_cast<int>(type);}
+inline int format_as(dcgmHealthSystems_enum type) {  return static_cast<int>(type);}
+inline int format_as(dcgmModuleId_t type) {  return static_cast<int>(type);}
+inline int format_as(dcgmModuleStatus_t type) {  return static_cast<int>(type);}
+//inline int format_as(dcgmMutexSt type) {  return static_cast<int>(type);}
+
+#ifdef dcgmNvLinkLinkState
+inline int format_as(dcgmNvLinkLinkState type) {  return static_cast<int>(type);}
+#endif
+inline int format_as(dcgmOrder_enum type) {  return static_cast<int>(type);}
+inline int format_as(dcgmPolicyCondition_enum type) {  return static_cast<int>(type);}
+inline int format_as(dcgmPolicyValidation_enum type) {  return static_cast<int>(type);}
+inline int format_as(dcgmPolicyAction_enum type) {  return static_cast<int>(type);}
+
+inline int format_as(dcgmReturn_enum type) {  return static_cast<int>(type);}
+inline int format_as(dcgm_field_entity_group_t type) {  return static_cast<int>(type);}
+
+#ifdef nvmlReturn_enum
+inline int format_as(nvmlReturn_enum type) {  return static_cast<int>(type);}
+#endif
+#endif