Merge remote-tracking branch 'meinersbur/users/meinersbur/flang_runtime' into HEAD

Author: Meinersbur

flang-rt/CMakeLists.txt

@@ -73,13 +73,22 @@ if (LLVM_LIBRARY_OUTPUT_INTDIR AND LLVM_RUNTIME_OUTPUT_INTDIR AND PACKAGE_VERSIO
 endif()
 
 if (LLVM_TREE_AVAILABLE)
+  # In a bootstrap build emit the libraries into a default search path in the
+  # build directory of the just-built compiler. This allows using the
+  # just-built compiler without specifying paths to runtime libraries.
+  #
   # Despite Clang in the name, get_clang_resource_dir does not depend on Clang
   # being added to the build. Flang uses the same resource dir as clang.
   include(GetClangResourceDir)
   get_clang_resource_dir(FLANG_RT_BUILD_LIB_DIR PREFIX "${LLVM_LIBRARY_OUTPUT_INTDIR}/.." SUBDIR "lib${LLVM_LIBDIR_SUFFIX}")
   get_clang_resource_dir(FLANG_RT_INSTALL_LIB_DIR SUBDIR "lib${LLVM_LIBDIR_SUFFIX}") # No prefix, CMake's install command finds the install prefix itself
 else ()
-  set(FLANG_RT_BUILD_LIB_DIR "${LLVM_LIBRARY_OUTPUT_INTDIR}")
+  # In a runtimes build never write into LLVM's build dir. It might be reused
+  # for mutliple Flang-RT builds (e.g. Debug/Release). Instead create our own
+  # library directory.
+  #
+  # TODO: Support multi-config generators
+  set(FLANG_RT_BUILD_LIB_DIR "${CMAKE_BINARY_DIR}/lib${LLVM_LIBDIR_SUFFIX}")
   set(FLANG_RT_INSTALL_LIB_DIR "lib${LLVM_LIBDIR_SUFFIX}")
 endif ()
 
@@ -102,19 +111,23 @@ endif ()
 option(FLANG_RT_INCLUDE_TESTS "Generate build targets for the flang-rt unit and regression-tests." "${LLVM_INCLUDE_TESTS}")
 
 
-set(FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT "" CACHE STRING "Compile flang-rt with GPU support (CUDA or OpenMP)")
+set(FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT "" CACHE STRING "Compile Flang-RT with GPU support (CUDA or OpenMP)")
 set_property(CACHE FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT PROPERTY STRINGS
     ""
     CUDA
     OpenMP
   )
 if (NOT FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT)
+  # Support for GPUs disabled
 elseif (FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT STREQUAL "CUDA")
+  # Support for CUDA
   set(FLANG_RT_LIBCUDACXX_PATH "" CACHE PATH "Path to libcu++ package installation")
   option(FLANG_RT_CUDA_RUNTIME_PTX_WITHOUT_GLOBAL_VARS "Do not compile global variables' definitions when producing PTX library" OFF)
 elseif (FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT STREQUAL "OpenMP")
-set(FLANG_RT_DEVICE_ARCHITECTURES "all" CACHE STRING
-  "List of OpenMP device architectures to be used to compile the Fortran runtime (e.g. 'gfx1103;sm_90')")
+  # Support for OpenMP offloading
+  set(FLANG_RT_DEVICE_ARCHITECTURES "all" CACHE STRING
+      "List of OpenMP device architectures to be used to compile the Fortran runtime (e.g. 'gfx1103;sm_90')"
+    )
 else ()
   message(FATAL_ERROR "Invalid value '${FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT}' for FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT; must be empty, 'CUDA', or 'OpenMP'")
 endif ()

flang-rt/cmake/modules/AddFlangRT.cmake

@@ -65,7 +65,10 @@ function (add_flangrt_library name)
     set_target_properties(${name} PROPERTIES FOLDER "Fortran Runtime/Libraries")
   endif ()
 
+  # Minimum required C++ version for Flang-RT, even if CMAKE_CXX_STANDARD is defined to something else.
   target_compile_features(${name} PRIVATE cxx_std_17)
+
+  # Use compiler-specific options to disable exceptions and RTTI.
   if (LLVM_COMPILER_IS_GCC_COMPATIBLE)
     target_compile_options(${name} PRIVATE
         $<$<COMPILE_LANGUAGE:CXX>:-fno-exceptions -fno-rtti -fno-unwind-tables -fno-asynchronous-unwind-tables>
@@ -74,6 +77,23 @@ function (add_flangrt_library name)
     target_compile_options(${name} PRIVATE
         $<$<COMPILE_LANGUAGE:CXX>:/EHs-c- /GR->
       )
+  elseif (CMAKE_CXX_COMPILER_ID MATCHES "XL")
+    target_compile_options(${name} PRIVATE
+        $<$<COMPILE_LANGUAGE:CXX>:-qnoeh -qnortti>
+      )
+  endif ()
+
+  # Also for CUDA source when compiling with FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT=CUDA
+  if (CMAKE_CUDA_COMPILER_ID MATCHES "NVIDIA")
+    # Assuming gcc as host compiler.
+    target_compile_options(${name} PRIVATE
+        $<$<COMPILE_LANGUAGE:CUDA>:--no-exceptions -Xcompiler -fno-rtti -Xcompiler -fno-unwind-tables -Xcompiler -fno-asynchronous-unwind-tables>
+      )
+  else ()
+    # Assuming a clang-compatible CUDA compiler.
+    target_compile_options(${name} PRIVATE
+        $<$<COMPILE_LANGUAGE:CUDA>:-fno-exceptions -fno-rtti -fno-unwind-tables -fno-asynchronous-unwind-tables>
+      )
   endif ()
 
   # Flang-rt's public headers

flang-rt/cmake/modules/AddFlangRTOffload.cmake

@@ -42,21 +42,27 @@ macro(enable_cuda_compilation name files)
       "${CUDA_COMPILE_OPTIONS}"
       )
 
-    if (EXISTS "${FLANG_RT_LIBCUDACXX_PATH}/include")
-      # When using libcudacxx headers files, we have to use them
-      # for all files of F18 runtime.
-      include_directories(AFTER ${FLANG_RT_LIBCUDACXX_PATH}/include)
-      add_compile_definitions(RT_USE_LIBCUDACXX=1)
-    endif()
+    # Create a .a library consisting of CUDA PTX.
+    # This is different from a regular static library. The CUDA_PTX_COMPILATION
+    # property can only be applied to object libraries and create *.ptx files
+    # instead of *.o files. The .a will consist of those *.ptx files only.
+    add_flangrt_library(obj.${name}PTX OBJECT ${files})
+    set_property(TARGET obj.${name}PTX PROPERTY CUDA_PTX_COMPILATION ON)
+    add_flangrt_library(${name}PTX STATIC "$<TARGET_OBJECTS:obj.${name}PTX>")
 
-    # Add an OBJECT library consisting of CUDA PTX.
-    add_flangrt_library(${name}PTX OBJECT ${files})
-    set_property(TARGET ${name}PTX PROPERTY CUDA_PTX_COMPILATION ON)
+    # Apply configuration options
     if (FLANG_RT_CUDA_RUNTIME_PTX_WITHOUT_GLOBAL_VARS)
-      target_compile_definitions(${name}PTX
+      target_compile_definitions(obj.${name}PTX
         PRIVATE FLANG_RT_NO_GLOBAL_VAR_DEFS
         )
     endif()
+
+    # When using libcudacxx headers files, we have to use them
+    # for all files of Flang-RT.
+    if (EXISTS "${FLANG_RT_LIBCUDACXX_PATH}/include")
+      target_include_directories(obj.${name}PTX AFTER PRIVATE "${FLANG_RT_LIBCUDACXX_PATH}/include")
+      target_compile_definitions(obj.${name}PTX PRIVATE RT_USE_LIBCUDACXX=1)
+    endif ()
   endif()
 endmacro()
 

flang-rt/lib/CufRuntime/CMakeLists.txt

@@ -6,6 +6,9 @@
 #
 #===------------------------------------------------------------------------===#
 
+# Phony build target that does not include the CUDA version.
+add_custom_target(CufRuntime)
+
 # libCufRuntime depends on a certain version of CUDA. To be able to have
 # multiple build of this library with different CUDA version, the version is
 # added to the library name.
@@ -20,6 +23,7 @@ add_flangrt_library(${CUFRT_LIBNAME}
   memory.cpp
   registration.cpp
 )
+add_dependencies(CufRuntime ${CUFRT_LIBNAME})
 target_include_directories(${CUFRT_LIBNAME} PRIVATE ${CUDAToolkit_INCLUDE_DIRS})
 
 if (BUILD_SHARED_LIBS)

flang-rt/lib/CufRuntime/kernel.cpp

@@ -7,7 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "flang/Runtime/CUDA/kernel.h"
-#include "../terminator.h"
+#include "../flang_rt/terminator.h"
 #include "flang/Runtime/CUDA/common.h"
 
 #include "cuda_runtime.h"

flang-rt/lib/CufRuntime/registration.cpp

@@ -7,7 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "flang/Runtime/CUDA/registration.h"
-#include "../terminator.h"
+#include "../flang_rt/terminator.h"
 #include "flang/Runtime/CUDA/common.h"
 
 #include "cuda_runtime.h"

flang-rt/test/Driver/ctofortran.f90

@@ -1,7 +1,9 @@
 ! UNSUPPORTED: system-windows
+
 ! RUN: split-file %s %t
-! RUN: chmod +x %t/runtest.sh
-! RUN: %t/runtest.sh %t %t/ffile.f90 %t/cfile.c %flang | FileCheck %s
+! RUN: %clang -c %t/cfile.c -o %t/cfile.o
+! RUN: %flang -L"%libdir" %deplibs %t/ffile.f90 %t/cfile.o -o %t/ctofortran
+! RUN: env LD_LIBRARY_PATH="$LD_LIBRARY_PATH:%libdir" %t/ctofortran | FileCheck %s
 
 !--- ffile.f90
 program fmain
@@ -65,24 +67,3 @@ end subroutine foo
   foo(desc);
   return;
 }
-!--- runtest.sh
-#!/bin/bash
-TMPDIR=$1
-FFILE=$2
-CFILE=$3
-FLANG=$4
-shift 4
-FLAGS="$*"
-BINDIR=`dirname $FLANG`
-LIBDIR=$BINDIR/../lib
-CCOMP=$BINDIR/clang
-if [ -x $CCOMP ]
-then
-  export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$LIBDIR
-  $CCOMP $FLAGS -c $CFILE -o $TMPDIR/cfile.o
-  $FLANG $FLAGS $FFILE $TMPDIR/cfile.o -o $TMPDIR/ctofortran
-  $TMPDIR/ctofortran # should print "PASS"
-else
-  # No clang compiler, just pass by default
-  echo "PASS"
-fi

flang-rt/test/Driver/exec.f90

@@ -1,7 +1,7 @@
 ! Verify that flang can correctly build executables.
 
-! RUN: %flang %s -o %t
-! RUN: env LD_LIBRARY_PATH="$LD_LIBRARY_PATH:%llvmshlibdir" %t | FileCheck %s
+! RUN: %flang -L"%libdir" %s %deplibs -o %t
+! RUN: env LD_LIBRARY_PATH="$LD_LIBRARY_PATH:%libdir" %t | FileCheck %s
 ! RUN: rm -f %t
 
 ! CHECK: Hello, World!

flang-rt/test/NonGtestUnit/lit.cfg.py

@@ -16,7 +16,7 @@
 
 # test_exec_root: The root path where tests should be run.
 # lit writes a '.lit_test_times.txt' file into this directory.
-config.test_exec_root = config.flangrt_binary_test_dir
+config.test_exec_root = config.flang_rt_binary_test_dir
 
 # testFormat: The test format to use to interpret tests.
 config.test_format = lit.formats.ExecutableTest()

flang-rt/test/NonGtestUnit/lit.site.cfg.py.in

@@ -3,12 +3,12 @@
 import os
 
 config.llvm_tools_dir = "@LLVM_TOOLS_DIR@"
-config.flangrt_source_dir = "@FLANG_RT_SOURCE_DIR@"
+config.flang_rt_source_dir = "@FLANG_RT_SOURCE_DIR@"
 config.flangrt_binary_dir = "@FLANG_RT_BINARY_DIR@"
-config.flangrt_binary_test_dir = os.path.dirname(__file__)
+config.flang_rt_binary_test_dir = os.path.dirname(__file__)
 
 import lit.llvm
 lit.llvm.initialize(lit_config, config)
 
 # Let the main config do the real work.
-lit_config.load_config(config, os.path.join(config.flangrt_source_dir, 'test', 'NonGtestUnit', 'lit.cfg.py'))
+lit_config.load_config(config, os.path.join(config.flang_rt_source_dir, 'test', 'NonGtestUnit', 'lit.cfg.py'))