add fastmath flag and bug fix

Author: wilfonba

CMakeLists.txt

@@ -20,6 +20,7 @@ option(MFC_MPI           "Build with MPI"                                     ON
 option(MFC_OpenACC       "Build with OpenACC"                                OFF)
 option(MFC_GCov          "Build with GCov"                                   OFF)
 option(MFC_Unified       "Build with unified CPU & GPU memory (GH-200 only)" OFF)
+option(MFC_Fastmath      "Build with -fastmath on NV GPUs"                   OFF)
 option(MFC_PRE_PROCESS   "Build pre_process"                                 OFF)
 option(MFC_SIMULATION    "Build simulation"                                  OFF)
 option(MFC_POST_PROCESS  "Build post_process"                                OFF)
@@ -487,6 +488,9 @@ function(MFC_SETUP_TARGET)
                     "-foffload=amdgcn-amdhsa='-march=gfx90a'"
                     "-foffload-options=-lgfortran\ -lm"
                     "-fno-exceptions")
+                if (MFC_Fastmath)
+                    message(WARNING "--fastmath has no effect with the GNU compiler")
+                endif()
             elseif(CMAKE_Fortran_COMPILER_ID STREQUAL "NVHPC" OR CMAKE_Fortran_COMPILER_ID STREQUAL "PGI")
                 foreach (cc ${MFC_CUDA_CC})
                     target_compile_options(${a_target}
@@ -498,6 +502,12 @@ function(MFC_SETUP_TARGET)
                     PRIVATE -gpu=keep,ptxinfo,lineinfo
                 )
 
+                if (MFC_Fastmath)
+                    target_compile_options(${a_target}
+                        PRIVATE -gpu=fastmath
+                    )
+                endif()
+
                 # GH-200 Unified Memory Support
                 if (MFC_Unified)
                     target_compile_options(${ARGS_TARGET}
@@ -521,11 +531,18 @@ function(MFC_SETUP_TARGET)
                         PRIVATE -DFRONTIER_UNIFIED)
                 endif()
 
+                if (MFC_Fastmath)
+                    message(WARNING "--fastmath has no effect with the CCE")
+                endif()
+
                 find_package(hipfort COMPONENTS hip CONFIG REQUIRED)
                 target_link_libraries(${a_target} PRIVATE hipfort::hip hipfort::hipfort-amdgcn)
             endif()
         elseif (CMAKE_Fortran_COMPILER_ID STREQUAL "Cray")
             target_compile_options(${a_target} PRIVATE "SHELL:-h noacc" "SHELL:-x acc")
+            if (MFC_Fastmath)
+                message(WARNING "--fastmath has no effect with the CCE")
+            endif()
         endif()
 
         if (CMAKE_Fortran_COMPILER_ID STREQUAL "NVHPC" OR CMAKE_Fortran_COMPILER_ID STREQUAL "PGI")

toolchain/mfc/build.py

@@ -147,6 +147,7 @@ def configure(self, case: Case):
             flags.append(f"-DMFC_OpenACC={'ON' if ARG('gpu') else 'OFF'}")
             flags.append(f"-DMFC_GCov={   'ON' if ARG('gcov') else 'OFF'}")
             flags.append(f"-DMFC_Unified={'ON' if ARG('unified') else 'OFF'}")
+            flags.append(f"-DMFC_Fastmath={'ON' if ARG('fastmath') else 'OFF'}")
 
         command = ["cmake"] + flags + ["-S", cmake_dirpath, "-B", build_dirpath]
 

toolchain/mfc/lock.py

@@ -5,7 +5,7 @@
 from .printer import cons
 
 
-MFC_LOCK_CURRENT_VERSION: int = 5
+MFC_LOCK_CURRENT_VERSION: int = 6
 
 
 @dataclasses.dataclass

toolchain/mfc/state.py

@@ -3,12 +3,13 @@
 
 @dataclasses.dataclass
 class MFCConfig:
-    mpi:     bool = True
-    gpu:     bool = False
-    debug:   bool = False
-    gcov:    bool = False
-    unified: bool = False
-    single: bool = False
+    mpi:       bool = True
+    gpu:       bool = False
+    debug:     bool = False
+    gcov:      bool = False
+    unified:   bool = False
+    single:    bool = False
+    fastmath : bool = False
 
     @staticmethod
     def from_dict(d: dict):

toolchain/templates/santis.mako

@@ -3,9 +3,8 @@
 <%namespace name="helpers" file="helpers.mako"/>
 
 % if engine == 'batch':
-#SBATCH --uenv=icon/25.2:v1
+#SBATCH --uenv=icon/25.2:v1@santis
 #SBATCH --nodes=${nodes}
-#SBATCH --reservation=g183
 #SBATCH --ntasks-per-node=${tasks_per_node}
 #SBATCH --job-name="${name}"
 #SBATCH --output="${name}.out"
@@ -78,9 +77,9 @@ echo
                 --gpus-per-task 1                                    \
             % endif
                 --wait 200 --bcast=/tmp/${target.name}               \
-                "${target.get_home_dirpath(case)}/misc/nvidia_uvm/bind.sh" \
+                "${target.get_home_dirpath()}/misc/nvidia_uvm/bind.sh" \
             % if target.name == 'simulation':
-                "${target.get_home_dirpath(case)}/misc/nvidia_uvm/nsys.sh" \
+                "${target.get_home_dirpath()}/misc/nvidia_uvm/nsys.sh" \
             % endif
                 "${target.get_install_binpath(case)}")
     % endif