Fix: Use __USE_NVTX macro to avoid NVTX linking errors in tests.

Clarify in docs that timer_enable_nvtx parameter only takes effect on CUDA platforms.
Signed-off-by：Tianxiang Wang<[email protected]>, Contributed under MetaX Integrated Circuits (Shanghai) Co., Ltd.

Author: wangtianxiang

CMakeLists.txt

@@ -287,7 +287,7 @@ if (USE_SW)
   set(SW ON)
   include_directories(${SW_MATH}/include)
   include_directories(${SW_FFT}/include)
-  
+
   target_link_libraries(${ABACUS_BIN_NAME} ${SW_FFT}/lib/libfftw3.a)
   target_link_libraries(${ABACUS_BIN_NAME} ${SW_MATH}/libswfft.a)
   target_link_libraries(${ABACUS_BIN_NAME} ${SW_MATH}/libswscalapack.a)
@@ -373,6 +373,7 @@ if(USE_CUDA)
   if(USE_CUDA)
     add_compile_definitions(__CUDA)
     add_compile_definitions(__UT_USE_CUDA)
+    target_compile_definitions(${ABACUS_BIN_NAME} PRIVATE __USE_NVTX)
     if (CMAKE_BUILD_TYPE STREQUAL "Debug")
       set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -g -G" CACHE STRING "CUDA flags for debug build" FORCE)
     endif()
@@ -520,7 +521,7 @@ if(ENABLE_MLALGO)
     include_directories(${libnpy_INCLUDE_DIR})
   endif()
   include_directories(${libnpy_SOURCE_DIR}/include)
-  
+
   add_compile_definitions(__MLALGO)
 endif()
 
@@ -560,7 +561,7 @@ if (ENABLE_CNPY)
     include_directories(${cnpy_INCLUDE_DIR})
   endif()
   include_directories(${cnpy_SOURCE_DIR})
-  
+
   # find ZLIB and link
   find_package(ZLIB REQUIRED)
   target_link_libraries(${ABACUS_BIN_NAME} cnpy ZLIB::ZLIB)

docs/advanced/input_files/input-main.md

@@ -710,7 +710,7 @@ If only one value is set (such as `kspacing 0.5`), then kspacing values of a/b/c
 ### timer_enable_nvtx
 
 - **Type**: Boolean
-- **Description**:
+- **Description**: Controls whether NVTX profiling labels are emitted by the timer. This feature is only effective on CUDA platforms.
 
   - True: Enable NVTX profiling labels in the timer.
   - False: Disable NVTX profiling labels in the timer.

source/source_base/timer.cpp

@@ -14,7 +14,7 @@
 #include "chrono"
 #include "source_base/formatter.h"
 
-#ifdef __CUDA
+#if defined(__CUDA) && defined(__USE_NVTX)
 #include <nvToolsExt.h>
 #include "source_io/module_parameter/parameter.h"
 #endif
@@ -98,7 +98,7 @@ void timer::tick(const std::string &class_name,const std::string &name)
 #endif
 			++timer_one.calls;
 			timer_one.start_flag = false;
-#ifdef __CUDA
+#if defined(__CUDA) && defined(__USE_NVTX)
             if (PARAM.inp.timer_enable_nvtx){
                 std::string label = class_name + ":" + name;
                 nvtxRangePushA(label.data());
@@ -118,7 +118,7 @@ void timer::tick(const std::string &class_name,const std::string &name)
 			timer_one.cpu_second += (cpu_time() - timer_one.cpu_start);
 #endif
 			timer_one.start_flag = true;
-#ifdef __CUDA
+#if defined(__CUDA) && defined(__USE_NVTX)
             if (PARAM.inp.timer_enable_nvtx){
                 nvtxRangePop();
             }