Add modern logging utils (#4486)

* add a modern logger

* replace the old logger

* lint

* Refactor logging statements to remove prefixes

* update

* Refactor logging

* check using logger

* fix

* fix lint

* fix msvc build

* fix MSVC build

* fix MSVC build

Author: lzhangzz

CMakeLists.txt

@@ -21,6 +21,12 @@ if (MSVC)
     add_compile_options($<$<COMPILE_LANGUAGE:CXX>:/Zc:preprocessor>)
     add_compile_options($<$<COMPILE_LANGUAGE:CXX>:/Zc:__cplusplus>)
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler=/Zc:preprocessor -Xcompiler=/Zc:__cplusplus")
+    # nvcc's front-end has a known bug (fixed in NVCC 13.1) where the fmt 11.x
+    # literal-encoding probe (fmt/base.h:is_utf8_enabled) always returns false,
+    # causing a static_assert failure.  Disable fmt's Unicode check project-wide;
+    # without /utf-8 the MSVC host compiler also sees use_utf8=false, so all TUs
+    # agree and there is no ODR inconsistency.
+    add_compile_definitions(FMT_UNICODE=0)
 endif ()
 
 find_package(CUDAToolkit REQUIRED)
@@ -39,6 +45,19 @@ option(BUILD_FAST_MATH "Build in fast math mode" ON)
 
 include(FetchContent)
 
+FetchContent_Declare(
+  fmt
+  GIT_REPOSITORY https://github.com/fmtlib/fmt.git
+  GIT_TAG        11.1.4
+  GIT_SHALLOW    ON
+  GIT_PROGRESS   TRUE
+  USES_TERMINAL_DOWNLOAD TRUE
+  EXCLUDE_FROM_ALL
+)
+FetchContent_MakeAvailable(fmt)
+set_target_properties(fmt PROPERTIES POSITION_INDEPENDENT_CODE ON)
+include_directories(${fmt_SOURCE_DIR}/include)
+
 if (BUILD_TEST)
   FetchContent_Declare(
     Catch2
@@ -269,6 +288,11 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-extended-lambda")
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --std=c++${CXX_STD}")
+# Suppress nvcc warning 128 (loop not reachable) in fmt headers - nvcc false positive
+# Suppress nvcc warning 27 (character value out of range) from fmt/format.h's
+# fractional_part_rounding_thresholds which uses U32 string literals with values
+# exceeding 0x10FFFF - harmless, pending fix in fmtlib (PR #4719)
+set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --diag-suppress=128 --diag-suppress=27")
 
 string(REPLACE "-O2" "" CMAKE_CXX_FLAGS_RELEASE         "${CMAKE_CXX_FLAGS_RELEASE}")
 string(REPLACE "-O2" "" CMAKE_CUDA_FLAGS_RELEASE        "${CMAKE_CUDA_FLAGS_RELEASE}")

src/turbomind/comm/CMakeLists.txt

@@ -5,11 +5,11 @@ cmake_minimum_required(VERSION 3.11)
 find_package(Threads)
 
 add_library(host_comm STATIC host_comm.cc thread_comm.cc)
-target_link_libraries(host_comm PRIVATE core logger Threads::Threads)
+target_link_libraries(host_comm PRIVATE core Threads::Threads)
 set_property(TARGET host_comm PROPERTY POSITION_INDEPENDENT_CODE ON)
 
 add_library(device_comm STATIC device_comm.cc)
-target_link_libraries(device_comm PRIVATE core logger)
+target_link_libraries(device_comm PRIVATE core)
 set_property(TARGET device_comm PROPERTY POSITION_INDEPENDENT_CODE ON)
 set_property(TARGET device_comm PROPERTY CUDA_RESOLVE_DEVICE_SYMBOLS ON)
 

src/turbomind/comm/cuda_ipc/CMakeLists.txt

@@ -16,7 +16,7 @@ target_link_libraries(cuda_ipc_comm PRIVATE
         core
         cuda_utils
         CUDA::cuda_driver
-        logger)
+       )
 
 set_property(TARGET cuda_ipc_comm PROPERTY POSITION_INDEPENDENT_CODE  ON)
 set_property(TARGET cuda_ipc_comm PROPERTY CUDA_RESOLVE_DEVICE_SYMBOLS  ON)

src/turbomind/comm/cuda_ipc/cuda_ipc_comm.cu

@@ -17,8 +17,8 @@
 
 #include "src/turbomind/comm/cuda_ipc/semaphore.h"
 
+#include "src/turbomind/core/logger.h"
 #include "src/turbomind/utils/cuda_utils.h"
-#include "src/turbomind/utils/logger.h"
 
 namespace turbomind::comm {
 
@@ -156,7 +156,7 @@ CudaIpcCommImpl::~CudaIpcCommImpl()
     }
 
     for (const auto& a : allocation_) {
-        TM_LOG_WARNING("[COMM][%d] Allocation (%p, %lu) is not freed", global_rank_, a.uc_beg, a.size);
+        TM_LOG_WARN("Rank {}: Allocation ({}, {}) is not freed", global_rank_, a.uc_beg, a.size);
     }
 
     cudaStreamSynchronize(0);
@@ -220,7 +220,7 @@ void CudaIpcCommImpl::Free(void* ptr)
         allocation_.erase(it);
     }
     else {
-        TM_LOG_WARNING("[TM][COMM][%d] Freeing %p which is not allocated by this module", global_rank_, ptr);
+        TM_LOG_WARN("Rank {}: Freeing {} which is not allocated by this module", global_rank_, ptr);
     }
 }
 
@@ -230,7 +230,7 @@ void CudaIpcCommImpl::Register(void* ptr, size_t size)
     auto& symm = groups_.at(0).symmetric;
 
     if (symm.find(ptr) != symm.end()) {
-        TM_LOG_WARNING("[TM][COMM][%d] Duplicated registration on (%p, %lu)", global_rank_, ptr, size);
+        TM_LOG_WARN("Rank {}: Duplicated registration on ({}, {})", global_rank_, ptr, size);
         return;
     }
 
@@ -317,7 +317,7 @@ void CudaIpcCommImpl::Deregister(void* ptr)
             Deregister(s.extract(it).value());
         }
         else {
-            TM_LOG_WARNING("[TM][COMM][%d] Deregistering non-registered address %p", global_rank_, ptr);
+            TM_LOG_WARN("Rank {}: Deregistering non-registered address {}", global_rank_, ptr);
         }
     }
 }

src/turbomind/comm/env.h

@@ -7,7 +7,7 @@
 #include <string>
 #include <type_traits>
 
-#include "src/turbomind/utils/logger.h"
+#include "src/turbomind/core/logger.h"
 
 namespace turbomind {
 
@@ -40,7 +40,7 @@ auto GetEnv()
         if (is_set) {
             std::stringstream ss;
             ss << x;
-            TM_LOG_INFO("[%s] %s=%s", E::prefix, E::name, ss.str().c_str());
+            TM_LOG_INFO("{} {}={}", E::prefix, E::name, ss.str());
         }
         return x;
     }();

src/turbomind/comm/gloo/CMakeLists.txt

@@ -33,7 +33,7 @@ add_library(gloo_comm STATIC
     tcp_store.cc
 )
 set_property(TARGET gloo_comm PROPERTY POSITION_INDEPENDENT_CODE ON)
-target_link_libraries(gloo_comm PUBLIC gloo host_comm logger xgrammar)
+target_link_libraries(gloo_comm PUBLIC gloo host_comm xgrammar)
 
 add_executable(test_ipc_comm test_ipc_comm.cc)
 target_link_libraries(test_ipc_comm PRIVATE gloo_comm Threads::Threads)

src/turbomind/comm/gloo/gloo_comm.cc

@@ -23,7 +23,7 @@
 
 #include "src/turbomind/comm/gloo/tcp_store.h"
 #include "src/turbomind/comm/host_comm.h"
-#include "src/turbomind/utils/logger.h"
+#include "src/turbomind/core/logger.h"
 
 namespace turbomind::comm {
 
@@ -346,7 +346,7 @@ class GlooGroupId: public HostGroupId {
     void Initialize() override
     {
         info_ = GlobalStoreFactory::Instance().New();
-        TM_LOG_INFO("[TM][COMM] GlooGroupId=%s", info_.c_str());
+        TM_LOG_INFO("GlooGroupId={}", info_);
     }
 
     void Export(std::ostream& os) override

src/turbomind/comm/gloo/tcp_store.cc

@@ -9,7 +9,7 @@
 #include <gloo/transport/tcp/socket.h>
 
 #include "src/turbomind/comm/gloo/tcp_store.h"
-#include "src/turbomind/utils/logger.h"
+#include "src/turbomind/core/logger.h"
 
 namespace turbomind::comm {
 
@@ -148,7 +148,7 @@ TCPStore::TCPStore(const std::string& host, int port)
             }
         }
         catch (const std::exception& e) {
-            TM_LOG_WARNING("[TM][COMM] Failed to connect to store after %d retries: %s", retry, e.what());
+            TM_LOG_WARN("Failed to connect to store after {} retries: {}", retry, e.what());
             std::this_thread::sleep_for(std::chrono::seconds(1));
             retry += 1;
         }
@@ -209,7 +209,7 @@ void TCPStore::wait(const std::vector<std::string>& keys, const std::chrono::mil
                 ss << key << " ";
             }
             ss << "]";
-            TM_LOG_ERROR("[TM][COMM] %s, elapsed %lld s", ss.str().c_str(), elapsed.count());
+            TM_LOG_ERROR("{}, elapsed {} s", ss.str(), elapsed.count());
             throw std::runtime_error("Wait timeout for key(s): " + ss.str());
         }
         std::this_thread::sleep_for(std::chrono::milliseconds(1000));

src/turbomind/comm/host_comm.h

@@ -11,8 +11,8 @@
 #include <vector>
 
 #include "src/turbomind/core/data_type.h"
+#include "src/turbomind/core/logger.h"
 #include "src/turbomind/core/serdes.h"
-#include "src/turbomind/utils/logger.h"
 
 namespace turbomind::comm {
 

src/turbomind/comm/nccl/CMakeLists.txt

@@ -3,7 +3,7 @@
 cmake_minimum_required(VERSION 3.11)
 
 add_library(nccl_comm STATIC nccl.cu)
-target_link_libraries(nccl_comm PRIVATE rms_norm core ${NCCL_LIBRARIES} logger)
+target_link_libraries(nccl_comm PRIVATE rms_norm core ${NCCL_LIBRARIES})
 target_include_directories(nccl_comm PRIVATE ${NCCL_INCLUDE_DIRS})
 
 set_property(TARGET nccl_comm PROPERTY POSITION_INDEPENDENT_CODE  ON)