Revert "[Flang-RT] Changes required for embedded GPU LLVM IR Flang runtime (#757)"

ronlieb · ronlieb · commit f301308971a3 · 2025-12-15T06:30:23.000-06:00
This reverts commit 4b92bcf.
diff --git a/flang-rt/CMakeLists.txt b/flang-rt/CMakeLists.txt
@@ -152,8 +152,6 @@ if (NOT "${FLANG_RT_LIBCXX_PROVIDER}" IN_LIST FLANG_RT_SUPPORTED_PROVIDERS)
 endif ()
 
 option(FLANG_RT_ENABLE_STATIC "Build Flang-RT as a static library." ON)
-option(FLANG_RT_EMBED_GPU_LLVM_IR "Build Flang-RT as GPU LLVM IR library" ON)
-
 if (WIN32)
   # Windows DLL currently not implemented.
   set(FLANG_RT_ENABLE_SHARED OFF)
@@ -329,9 +327,7 @@ endif ()
 
 if (FLANG_RT_INCLUDE_TESTS)
   add_subdirectory(test)
-  if (NOT "${LLVM_RUNTIMES_TARGET}" MATCHES "^amdgcn|^nvptx")
-    add_subdirectory(unittests)
-  endif()
+  add_subdirectory(unittests)
 else ()
   add_custom_target(check-flang-rt)
 endif()
diff --git a/flang-rt/cmake/modules/AddFlangRT.cmake b/flang-rt/cmake/modules/AddFlangRT.cmake
@@ -219,12 +219,6 @@ function (add_flangrt_library name)
     # Minimum required C++ version for Flang-RT, even if CMAKE_CXX_STANDARD is defined to something else.
     target_compile_features(${tgtname} PRIVATE cxx_std_17)
 
-    # Determine which version of GPU Flang-RT we want to build:
-    # flang_rt.hostdevice or the implicitly linked device flang_rt.runtime.
-    if (FLANG_RT_EMBED_GPU_LLVM_IR)
-      target_compile_definitions(${tgtname} PRIVATE EMBED_FLANG_RT_GPU_LLVM_IR)
-    endif ()
-
     # When building the flang runtime if LTO is enabled the archive file
     # contains LLVM IR rather than object code. Currently flang is not
     # LTO aware so cannot link this file to compiled Fortran code.
diff --git a/flang-rt/include/flang-rt/runtime/lock.h b/flang-rt/include/flang-rt/runtime/lock.h
@@ -23,9 +23,7 @@
 #endif
 
 #if USE_PTHREADS
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 #include <pthread.h>
-#endif
 #elif defined(_WIN32)
 #include "flang/Common/windows-include.h"
 #else
@@ -47,7 +45,6 @@ class Lock {
   RT_API_ATTRS void Drop() {}
   RT_API_ATTRS bool TakeIfNoDeadlock() { return true; }
 #elif USE_PTHREADS
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
   Lock() { pthread_mutex_init(&mutex_, nullptr); }
   ~Lock() { pthread_mutex_destroy(&mutex_); }
   void Take() {
@@ -71,14 +68,6 @@ class Lock {
     isBusy_ = false;
     pthread_mutex_unlock(&mutex_);
   }
-#else
-  RT_API_ATTRS void Take(){}
-  RT_API_ATTRS bool TakeIfNoDeadlock() {return true;}
-  RT_API_ATTRS bool Try() {return true;}
-  RT_API_ATTRS void Drop() {}
-  Lock() {}
-  ~Lock() {}
-#endif
 #elif defined(_WIN32)
   Lock() { InitializeCriticalSection(&cs_); }
   ~Lock() { DeleteCriticalSection(&cs_); }
@@ -102,11 +91,9 @@ class Lock {
 #if RT_USE_PSEUDO_FILE_UNIT
   // No state.
 #elif USE_PTHREADS
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
   pthread_mutex_t mutex_{};
   volatile bool isBusy_{false};
   volatile pthread_t holder_;
-#endif
 #elif defined(_WIN32)
   CRITICAL_SECTION cs_;
 #else
diff --git a/flang-rt/include/flang-rt/runtime/tools.h b/flang-rt/include/flang-rt/runtime/tools.h
@@ -42,13 +42,6 @@
 #define RT_USE_PSEUDO_FILE_UNIT 1
 #endif
 
-#if (defined(__AMDGPU__) ||  defined(__NVPTX__)) && defined(EMBED_FLANG_RT_GPU_LLVM_IR)
-// Use the pseudo lock and pseudo file unit implementations
-// for the device.
-#define RT_USE_PSEUDO_LOCK 1
-#define RT_USE_PSEUDO_FILE_UNIT 1
-#endif
-
 namespace Fortran::runtime {
 
 class Terminator;
diff --git a/flang-rt/lib/runtime/CMakeLists.txt b/flang-rt/lib/runtime/CMakeLists.txt
@@ -179,11 +179,7 @@ else ()
 endif ()
 
 if ("${LLVM_RUNTIMES_TARGET}" MATCHES "^amdgcn|^nvptx")
-  if (FLANG_RT_EMBED_GPU_LLVM_IR)
-    set(sources ${supported_sources} ${gpu_sources})
-  else ()
-    set(sources ${gpu_sources})
-  endif ()
+  set(sources ${gpu_sources})
 elseif(FLANG_RT_EXPERIMENTAL_OFFLOAD_SUPPORT STREQUAL "CUDA")
   set(sources ${supported_sources})
 else ()
diff --git a/flang-rt/lib/runtime/assign.cpp b/flang-rt/lib/runtime/assign.cpp
@@ -7,9 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "flang/Runtime/assign.h"
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 #include "flang/Runtime/stop.h"
-#endif
 #include "flang-rt/runtime/assign-impl.h"
 #include "flang-rt/runtime/derived.h"
 #include "flang-rt/runtime/descriptor.h"
@@ -864,6 +862,17 @@ void RTDEF(AssignPolymorphic)(Descriptor &to, const Descriptor &from,
           PolymorphicLHS);
 }
 
+#if defined(OMP_OFFLOAD_BUILD)
+// To support a recently added use of variant in the OpenMP offload build,
+// added an abort wrapper which calls the flang-rt FortranAAbort.
+// Avoids the following linker error:
+//   ld.lld: error: undefined symbol: abort
+//   >>> referenced by /tmp/device_aassign.amdgcn.gfx90a-34a7ed.img.lto.o:(std::__throw_bad_variant_access(char const*))
+extern "C" void abort(void) {
+  RTNAME(Abort)();
+}
+#endif
+
 RT_EXT_API_GROUP_END
 } // extern "C"
 } // namespace Fortran::runtime
diff --git a/flang-rt/lib/runtime/descriptor.cpp b/flang-rt/lib/runtime/descriptor.cpp
@@ -8,9 +8,7 @@
 
 #include "flang-rt/runtime/descriptor.h"
 #include "ISO_Fortran_util.h"
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 #include "memory.h"
-#endif
 #include "flang-rt/runtime/allocator-registry.h"
 #include "flang-rt/runtime/derived.h"
 #include "flang-rt/runtime/stat.h"
diff --git a/flang-rt/lib/runtime/edit-input.cpp b/flang-rt/lib/runtime/edit-input.cpp
@@ -569,11 +569,7 @@ static RT_API_ATTRS void RaiseFPExceptions(
 #ifdef feraisexcept // a macro in some environments; omit std::
 #define RAISE feraiseexcept
 #else
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined (EMBED_FLANG_RT_GPU_LLVM_IR)
 #define RAISE std::feraiseexcept
-#else
-#define RAISE
-#endif
 #endif
 #endif // !defined(RT_DEVICE_COMPILATION)
 
diff --git a/flang-rt/lib/runtime/environment.cpp b/flang-rt/lib/runtime/environment.cpp
@@ -6,7 +6,6 @@
 //
 //===----------------------------------------------------------------------===//
 
-#if (not defined (__AMDGPU__) && not defined(__NVPTX__)) || not defined (EMBED_FLANG_RT_GPU_LLVM_IR)
 #include "flang-rt/runtime/environment.h"
 #include "environment-default-list.h"
 #include "memory.h"
@@ -335,4 +334,3 @@ bool RTNAME(RegisterConfigureEnv)(
 } // extern "C"
 
 } // namespace Fortran::runtime
-#endif
diff --git a/flang-rt/lib/runtime/file.cpp b/flang-rt/lib/runtime/file.cpp
@@ -6,7 +6,6 @@
 //
 //===----------------------------------------------------------------------===//
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 #include "flang-rt/runtime/file.h"
 #include "flang-rt/runtime/memory.h"
 #include "flang-rt/runtime/tools.h"
@@ -487,4 +486,3 @@ RT_API_ATTRS std::int64_t SizeInBytes(const char *path) {
 #endif // defined(RT_DEVICE_COMPILATION)
 
 } // namespace Fortran::runtime::io
-#endif
diff --git a/flang-rt/lib/runtime/io-api-minimal.cpp b/flang-rt/lib/runtime/io-api-minimal.cpp
@@ -23,7 +23,6 @@ namespace Fortran::runtime::io {
 RT_EXT_API_GROUP_BEGIN
 #endif
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 Cookie IODEF(BeginExternalListOutput)(
     ExternalUnit unitNumber, const char *sourceFile, int sourceLine) {
   return BeginExternalListIO<Direction::Output, ExternalListIoStatementState>(
@@ -34,7 +33,6 @@ enum Iostat IODEF(EndIoStatement)(Cookie cookie) {
   IoStatementState &io{*cookie};
   return static_cast<enum Iostat>(io.EndIoStatement());
 }
-#endif
 
 template <int KIND, typename INT = CppTypeFor<TypeCategory::Integer, KIND>>
 inline RT_API_ATTRS bool FormattedScalarIntegerOutput(
@@ -47,7 +45,6 @@ inline RT_API_ATTRS bool FormattedScalarIntegerOutput(
   }
 }
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 bool IODEF(OutputInteger8)(Cookie cookie, std::int8_t n) {
   return FormattedScalarIntegerOutput<1>(*cookie, n, "OutputInteger8");
 }
@@ -63,7 +60,6 @@ bool IODEF(OutputInteger32)(Cookie cookie, std::int32_t n) {
 bool IODEF(OutputInteger64)(Cookie cookie, std::int64_t n) {
   return FormattedScalarIntegerOutput<8>(*cookie, n, "OutputInteger64");
 }
-#endif
 
 #ifdef __SIZEOF_INT128__
 bool IODEF(OutputInteger128)(Cookie cookie, common::int128_t n) {
@@ -83,15 +79,13 @@ inline RT_API_ATTRS bool FormattedScalarRealOutput(
   }
 }
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 bool IODEF(OutputReal32)(Cookie cookie, float x) {
   return FormattedScalarRealOutput<4>(*cookie, x, "OutputReal32");
 }
 
 bool IODEF(OutputReal64)(Cookie cookie, double x) {
   return FormattedScalarRealOutput<8>(*cookie, x, "OutputReal64");
 }
-#endif
 
 template <int KIND,
     typename REAL = typename RealOutputEditing<KIND>::BinaryFloatingPoint>
@@ -116,7 +110,6 @@ inline RT_API_ATTRS bool FormattedScalarComplexOutput(
   return false;
 }
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 bool IODEF(OutputComplex32)(Cookie cookie, float re, float im) {
   return FormattedScalarComplexOutput<4>(*cookie, re, im, "OutputComplex32");
 }
@@ -152,7 +145,6 @@ bool IODEF(OutputLogical)(Cookie cookie, bool truth) {
     return false;
   }
 }
-#endif
 
 } // namespace Fortran::runtime::io
 
diff --git a/flang-rt/lib/runtime/io-api.cpp b/flang-rt/lib/runtime/io-api.cpp
@@ -199,14 +199,12 @@ RT_API_ATTRS Cookie BeginExternalFormattedIO(const char *format,
   }
 }
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 Cookie IODEF(BeginExternalFormattedOutput)(const char *format,
     std::size_t formatLength, const Descriptor *formatDescriptor,
     ExternalUnit unitNumber, const char *sourceFile, int sourceLine) {
   return BeginExternalFormattedIO<Direction::Output>(format, formatLength,
       formatDescriptor, unitNumber, sourceFile, sourceLine);
 }
-#endif
 
 Cookie IODEF(BeginExternalFormattedInput)(const char *format,
     std::size_t formatLength, const Descriptor *formatDescriptor,
diff --git a/flang-rt/lib/runtime/stop.cpp b/flang-rt/lib/runtime/stop.cpp
@@ -13,15 +13,11 @@
 #include "flang-rt/runtime/file.h"
 #include "flang-rt/runtime/io-error.h"
 #include "flang-rt/runtime/terminator.h"
-#if not defined(__AMDGPU__) && not defined(__NVPTX__)
 #include "flang/Runtime/main.h"
-#endif
 #include <cfenv>
 #include <cstdio>
 #include <cstdlib>
-#if not defined(__AMDGPU__) && not defined(__NVPTX__)
 #include <thread>
-#endif
 
 #ifdef HAVE_BACKTRACE
 #include BACKTRACE_HEADER
@@ -30,7 +26,6 @@
 extern "C" {
 
 [[maybe_unused]] static void DescribeIEEESignaledExceptions() {
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 #ifdef fetestexcept // a macro in some environments; omit std::
   auto excepts{fetestexcept(FE_ALL_EXCEPT)};
 #else
@@ -65,15 +60,13 @@ extern "C" {
 #endif
     std::fputc('\n', stderr);
   }
-#endif
 }
 
 static void CloseAllExternalUnits(const char *why) {
   Fortran::runtime::io::IoErrorHandler handler{why};
   Fortran::runtime::io::ExternalFileUnit::CloseAll(handler);
 }
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 [[noreturn]] RT_API_ATTRS void RTNAME(StopStatement)(
     int code, bool isErrorStop, bool quiet) {
 #if defined(RT_DEVICE_COMPILATION)
@@ -110,9 +103,7 @@ static void CloseAllExternalUnits(const char *why) {
   std::exit(code);
 #endif
 }
-#endif
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 [[noreturn]] RT_API_ATTRS void RTNAME(StopStatementText)(
     const char *code, std::size_t length, bool isErrorStop, bool quiet) {
 #if defined(RT_DEVICE_COMPILATION)
@@ -145,7 +136,6 @@ static void CloseAllExternalUnits(const char *why) {
   }
 #endif
 }
-#endif
 
 static bool StartPause() {
   if (Fortran::runtime::io::IsATerminal(0)) {
@@ -228,14 +218,13 @@ static RT_NOINLINE_ATTR void PrintBacktrace() {
 
 #endif
 }
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
+
 [[noreturn]] RT_OPTNONE_ATTR void RTNAME(Abort)() {
 #ifdef HAVE_BACKTRACE
   PrintBacktrace();
 #endif
   std::abort();
 }
-#endif
 
 RT_OPTNONE_ATTR void FORTRAN_PROCEDURE_NAME(backtrace)() { PrintBacktrace(); }
 
diff --git a/flang/include/flang/Runtime/extensions.h b/flang/include/flang/Runtime/extensions.h
@@ -18,13 +18,10 @@
 
 #define FORTRAN_PROCEDURE_NAME(name) name##_
 
-#if defined (_WIN32)
+#ifdef _WIN32
 // UID and GID don't exist on Windows, these exist to avoid errors.
 typedef std::uint32_t uid_t;
 typedef std::uint32_t gid_t;
-#elif (defined(__AMDGPU__) || defined(__NVPTX__)) && defined (EMBED_FLANG_RT_GPU_LLVM_IR)
-typedef std::uint32_t uid_t;
-typedef std::uint32_t gid_t;
 #else
 #include "sys/types.h" //pid_t
 #endif
diff --git a/flang/include/flang/Runtime/main.h b/flang/include/flang/Runtime/main.h
@@ -11,15 +11,11 @@
 
 #include "flang/Runtime/c-or-cpp.h"
 #include "flang/Runtime/entry-names.h"
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 #include <thread>
-#endif
 
 struct EnvironmentDefaultList;
 
-#if (not defined(__AMDGPU__) && not defined(__NVPTX__)) || not defined(EMBED_FLANG_RT_GPU_LLVM_IR)
 std::thread::id RTNAME(GetMainThreadId)();
-#endif
 
 FORTRAN_EXTERN_C_BEGIN
 void RTNAME(ProgramStart)(
diff --git a/libc/CMakeLists.txt b/libc/CMakeLists.txt
@@ -67,7 +67,6 @@ if(NOT LIBC_NAMESPACE MATCHES "^__llvm_libc")
   message(FATAL_ERROR "Invalid LIBC_NAMESPACE. Must start with '__llvm_libc' was '${LIBC_NAMESPACE}'")
 endif()
 
-string(REPLACE "." "_" LIBC_NAMESPACE "${LIBC_NAMESPACE}")
 message(STATUS "Setting LIBC_NAMESPACE namespace to '${LIBC_NAMESPACE}'")
 add_compile_definitions(LIBC_NAMESPACE=${LIBC_NAMESPACE})
 
diff --git a/libcxx/src/string.cpp b/libcxx/src/string.cpp
@@ -360,11 +360,9 @@ wstring to_wstring(unsigned long val) { return i_to_string<wstring>(val); }
 wstring to_wstring(unsigned long long val) { return i_to_string<wstring>(val); }
 #endif
 
-#if not defined(__AMDGPU__) && not defined(__NVPTX__)
 string to_string(float val) { return as_string(snprintf, initial_string< string>()(), "%f", val); }
 string to_string(double val) { return as_string(snprintf, initial_string< string>()(), "%f", val); }
 string to_string(long double val) { return as_string(snprintf, initial_string< string>()(), "%Lf", val); }
-#endif
 
 #if _LIBCPP_HAS_WIDE_CHARACTERS
 wstring to_wstring(float val) { return as_string(get_swprintf(), initial_string<wstring>()(), L"%f", val); }
