[OpenMP][libomp] Introduce oneAPI compiler support

Introduce KMP_COMPILER_ICX macro to represent compilation with oneAPI
compiler.

Fixup flag detection and compiler ID detection in CMake. Older CMake's
detect IntelLLVM as Clang.

Fix compiler warnings.

Fixup many of the tests to have non-empty parallel regions as they are
elided by oneAPI compiler.

Author: jpeyton52

openmp/cmake/DetectTestCompiler/CMakeLists.txt

@@ -10,6 +10,7 @@ function(write_compiler_information lang)
   set(information "${information}\\;${CMAKE_${lang}_COMPILER_VERSION}")
   set(information "${information}\\;${${lang}_FLAGS}")
   set(information "${information}\\;${${lang}_HAS_TSAN_FLAG}")
+  set(information "${information}\\;${${lang}_HAS_OMIT_FRAME_POINTER}")
   file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/${lang}CompilerInformation.txt ${information})
 endfunction(write_compiler_information)
 
@@ -40,6 +41,9 @@ if (CMAKE_C_COMPILER_ID STREQUAL "Clang")
   add_experimental_isel_flag(CXX)
 endif()
 
+check_c_compiler_flag("-fno-omit-frame-pointer" C_HAS_OMIT_FRAME_POINTER)
+check_cxx_compiler_flag("-fno-omit-frame-pointer" CXX_HAS_OMIT_FRAME_POINTER)
+
 SET(CMAKE_REQUIRED_FLAGS "-fsanitize=thread")
 check_c_compiler_flag("" C_HAS_TSAN_FLAG)
 check_cxx_compiler_flag("" CXX_HAS_TSAN_FLAG)

openmp/cmake/OpenMPTesting.cmake

@@ -76,12 +76,14 @@ macro(extract_test_compiler_information lang file)
   list(GET information 2 version)
   list(GET information 3 openmp_flags)
   list(GET information 4 has_tsan_flags)
+  list(GET information 5 has_omit_frame_pointer_flags)
 
   set(OPENMP_TEST_${lang}_COMPILER_PATH ${path})
   set(OPENMP_TEST_${lang}_COMPILER_ID ${id})
   set(OPENMP_TEST_${lang}_COMPILER_VERSION ${version})
   set(OPENMP_TEST_${lang}_COMPILER_OPENMP_FLAGS ${openmp_flags})
   set(OPENMP_TEST_${lang}_COMPILER_HAS_TSAN_FLAGS ${has_tsan_flags})
+  set(OPENMP_TEST_${lang}_COMPILER_HAS_OMIT_FRAME_POINTER_FLAGS ${has_omit_frame_pointer_flags})
 endmacro()
 
 # Function to set variables with information about the test compiler.
@@ -98,6 +100,7 @@ function(set_test_compiler_information dir)
     set(OPENMP_TEST_COMPILER_VERSION "${OPENMP_TEST_C_COMPILER_VERSION}" PARENT_SCOPE)
     set(OPENMP_TEST_COMPILER_OPENMP_FLAGS "${OPENMP_TEST_C_COMPILER_OPENMP_FLAGS}" PARENT_SCOPE)
     set(OPENMP_TEST_COMPILER_HAS_TSAN_FLAGS "${OPENMP_TEST_C_COMPILER_HAS_TSAN_FLAGS}" PARENT_SCOPE)
+    set(OPENMP_TEST_COMPILER_HAS_OMIT_FRAME_POINTER_FLAGS "${OPENMP_TEST_C_COMPILER_HAS_OMIT_FRAME_POINTER_FLAGS}" PARENT_SCOPE)
 
     # Determine major version.
     string(REGEX MATCH "[0-9]+" major "${OPENMP_TEST_C_COMPILER_VERSION}")
@@ -149,6 +152,7 @@ else()
   endif()
   # TODO: Implement blockaddress in GlobalISel and remove this flag!
   set(OPENMP_TEST_COMPILER_OPENMP_FLAGS "-fopenmp ${OPENMP_TEST_COMPILER_THREAD_FLAGS} -fno-experimental-isel")
+  set(OPENMP_TEST_COMPILER_HAS_OMIT_FRAME_POINTER_FLAGS 1)
 endif()
 
 # Function to set compiler features for use in lit.

openmp/cmake/config-ix.cmake

@@ -1,4 +1,20 @@
 include(CheckCXXCompilerFlag)
+include(CheckCXXSourceCompiles)
+
+# Check for oneAPI compiler (some older CMake versions detect as Clang)
+if (CMAKE_C_COMPILER_ID STREQUAL "Clang")
+  check_cxx_source_compiles("#if (defined(__INTEL_CLANG_COMPILER) || defined(__INTEL_LLVM_COMPILER))
+                             int main() { return 0; }
+                             #else
+                             not oneAPI
+                             #endif" OPENMP_HAVE_ONEAPI_COMPILER)
+  if (OPENMP_HAVE_ONEAPI_COMPILER)
+    # According to CMake documentation, the compiler id should
+    # be IntelLLVM when detected oneAPI
+    set(CMAKE_C_COMPILER_ID "IntelLLVM")
+    set(CMAKE_CXX_COMPILER_ID "IntelLLVM")
+  endif()
+endif()
 
 check_cxx_compiler_flag(-Wall OPENMP_HAVE_WALL_FLAG)
 check_cxx_compiler_flag(-Werror OPENMP_HAVE_WERROR_FLAG)

openmp/runtime/cmake/LibompCheckLinkerFlag.cmake

@@ -25,7 +25,7 @@ function(libomp_check_linker_flag flag boolean)
      add_library(foo SHARED src_to_link.c)")
   # Compiling as a part of runtimes introduces ARCH-unknown-linux-gnu as a part
   # of a working directory.  So adding a guard for unknown.
-  set(failed_regexes "[Ee]rror;[Uu]nknown[^-];[Ss]kipping;LINK : warning")
+  set(failed_regexes "[Ee]rror;[Uu]nknown[^-];[Ss]kipping;LINK : warning;Unsupported command line")
   set(base_dir ${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/link_flag_check_${boolean})
   file(MAKE_DIRECTORY ${base_dir})
   file(MAKE_DIRECTORY ${base_dir}/build)

openmp/runtime/cmake/config-ix.cmake

@@ -139,7 +139,7 @@ elseif(NOT APPLE)
 endif()
 
 # Check Intel(R) C Compiler specific flags
-if(CMAKE_C_COMPILER_ID STREQUAL "Intel")
+if(CMAKE_C_COMPILER_ID STREQUAL "Intel" OR CMAKE_C_COMPILER_ID STREQUAL "IntelLLVM")
   check_cxx_compiler_flag(/Qlong_double LIBOMP_HAVE_LONG_DOUBLE_FLAG)
   check_cxx_compiler_flag(/Qdiag-disable:177 LIBOMP_HAVE_DIAG_DISABLE_177_FLAG)
   check_cxx_compiler_flag(/Qinline-min-size=1 LIBOMP_HAVE_INLINE_MIN_SIZE_FLAG)
@@ -247,7 +247,7 @@ libomp_check_version_symbols(LIBOMP_HAVE_VERSION_SYMBOLS)
 # Check if quad precision types are available
 if(CMAKE_C_COMPILER_ID STREQUAL "GNU")
   set(LIBOMP_HAVE_QUAD_PRECISION TRUE)
-elseif(CMAKE_C_COMPILER_ID STREQUAL "Intel")
+elseif(CMAKE_C_COMPILER_ID STREQUAL "Intel" OR CMAKE_C_COMPILER_ID STREQUAL "IntelLLVM")
   if(LIBOMP_HAVE_EXTENDED_FLOAT_TYPES_FLAG)
     set(LIBOMP_HAVE_QUAD_PRECISION TRUE)
   else()

openmp/runtime/src/kmp.h

@@ -1124,7 +1124,7 @@ extern void __kmp_init_target_mem();
 #if KMP_OS_UNIX && (KMP_ARCH_X86 || KMP_ARCH_X86_64)
 // HW TSC is used to reduce overhead (clock tick instead of nanosecond).
 extern kmp_uint64 __kmp_ticks_per_msec;
-#if KMP_COMPILER_ICC
+#if KMP_COMPILER_ICC || KMP_COMPILER_ICX
 #define KMP_NOW() ((kmp_uint64)_rdtsc())
 #else
 #define KMP_NOW() __kmp_hardware_timestamp()

openmp/runtime/src/kmp_affinity.cpp

@@ -1765,7 +1765,7 @@ static bool __kmp_affinity_create_hwloc_map(kmp_i18n_id_t *const msg_id) {
 
   hw_thread_index = 0;
   pu = NULL;
-  while (pu = hwloc_get_next_obj_by_type(tp, HWLOC_OBJ_PU, pu)) {
+  while ((pu = hwloc_get_next_obj_by_type(tp, HWLOC_OBJ_PU, pu))) {
     int index = depth - 1;
     bool included = KMP_CPU_ISSET(pu->os_index, __kmp_affin_fullMask);
     kmp_hw_thread_t &hw_thread = __kmp_topology->at(hw_thread_index);

openmp/runtime/src/kmp_atomic.cpp

@@ -2452,6 +2452,7 @@ ATOMIC_CMPXCHG_CPT(float8, mul_cpt, kmp_real64, 64, *,
                                RTYPE, LCK_ID, MASK, GOMP_FLAG)                 \
   ATOMIC_BEGIN_CPT_MIX(TYPE_ID, OP_ID, TYPE, RTYPE_ID, RTYPE)                  \
   TYPE new_value;                                                              \
+  (void)new_value;                                                             \
   OP_GOMP_CRITICAL_CPT(TYPE, OP, GOMP_FLAG)                                    \
   OP_CMPXCHG_CPT(TYPE, BITS, OP)                                               \
   }
@@ -2461,6 +2462,7 @@ ATOMIC_CMPXCHG_CPT(float8, mul_cpt, kmp_real64, 64, *,
                                 LCK_ID, GOMP_FLAG)                             \
   ATOMIC_BEGIN_CPT_MIX(TYPE_ID, OP_ID, TYPE, RTYPE_ID, RTYPE)                  \
   TYPE new_value;                                                              \
+  (void)new_value;                                                             \
   OP_GOMP_CRITICAL_CPT(TYPE, OP, GOMP_FLAG) /* send assignment */              \
   OP_UPDATE_CRITICAL_CPT(TYPE, OP, LCK_ID) /* send assignment */               \
   }
@@ -3162,6 +3164,7 @@ ATOMIC_CRITICAL_CPT_REV(cmplx16, div_a16_cpt_rev, kmp_cmplx128_a16_t, /, 32c,
                                    RTYPE, LCK_ID, MASK, GOMP_FLAG)             \
   ATOMIC_BEGIN_CPT_MIX(TYPE_ID, OP_ID, TYPE, RTYPE_ID, RTYPE)                  \
   TYPE new_value;                                                              \
+  (void)new_value;                                                             \
   OP_GOMP_CRITICAL_CPT_REV(TYPE, OP, GOMP_FLAG)                                \
   OP_CMPXCHG_CPT_REV(TYPE, BITS, OP)                                           \
   }
@@ -3171,6 +3174,7 @@ ATOMIC_CRITICAL_CPT_REV(cmplx16, div_a16_cpt_rev, kmp_cmplx128_a16_t, /, 32c,
                                     LCK_ID, GOMP_FLAG)                         \
   ATOMIC_BEGIN_CPT_MIX(TYPE_ID, OP_ID, TYPE, RTYPE_ID, RTYPE)                  \
   TYPE new_value;                                                              \
+  (void)new_value;                                                             \
   OP_GOMP_CRITICAL_CPT_REV(TYPE, OP, GOMP_FLAG) /* send assignment */          \
   OP_CRITICAL_CPT_REV(TYPE, OP, LCK_ID) /* send assignment */                  \
   }

openmp/runtime/src/kmp_atomic.h

@@ -251,6 +251,9 @@ struct KMP_DO_ALIGN(4) kmp_cmplx128_a4_t {
 
   kmp_cmplx128_a4_t() : q() {}
 
+#if defined(__cplusplus) && (KMP_OS_WINDOWS)
+  kmp_cmplx128_a4_t(const std::complex<_Quad> &c128) : q(c128) {}
+#endif
   kmp_cmplx128_a4_t(const kmp_cmplx128 &c128) : q(c128) {}
 
   kmp_cmplx128_a4_t operator+(const kmp_cmplx128_a4_t &b) {
@@ -314,6 +317,9 @@ struct KMP_DO_ALIGN(16) kmp_cmplx128_a16_t {
 
   kmp_cmplx128_a16_t() : q() {}
 
+#if defined(__cplusplus) && (KMP_OS_WINDOWS)
+  kmp_cmplx128_a16_t(const std::complex<_Quad> &c128) : q(c128) {}
+#endif
   kmp_cmplx128_a16_t(const kmp_cmplx128 &c128) : q(c128) {}
 
   kmp_cmplx128_a16_t operator+(const kmp_cmplx128_a16_t &b) {

openmp/runtime/src/kmp_barrier.cpp

@@ -2163,7 +2163,6 @@ void __kmp_join_barrier(int gtid) {
 
   kmp_info_t *this_thr = __kmp_threads[gtid];
   kmp_team_t *team;
-  kmp_uint nproc;
   int tid;
 #ifdef KMP_DEBUG
   int team_id;
@@ -2176,12 +2175,14 @@ void __kmp_join_barrier(int gtid) {
     itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
 #endif
 #endif /* USE_ITT_BUILD */
+#if ((USE_ITT_BUILD && USE_ITT_NOTIFY) || defined KMP_DEBUG)
+  int nproc = this_thr->th.th_team_nproc;
+#endif
   KMP_MB();
 
   // Get current info
   team = this_thr->th.th_team;
-  nproc = this_thr->th.th_team_nproc;
-  KMP_DEBUG_ASSERT((int)nproc == team->t.t_nproc);
+  KMP_DEBUG_ASSERT(nproc == team->t.t_nproc);
   tid = __kmp_tid_from_gtid(gtid);
 #ifdef KMP_DEBUG
   team_id = team->t.t_id;
@@ -2354,7 +2355,7 @@ void __kmp_join_barrier(int gtid) {
           // Set arrive time to zero to be able to check it in
           // __kmp_invoke_task(); the same is done inside the loop below
           this_thr->th.th_bar_arrive_time = 0;
-          for (kmp_uint i = 1; i < nproc; ++i) {
+          for (int i = 1; i < nproc; ++i) {
             delta += (cur_time - other_threads[i]->th.th_bar_arrive_time);
             other_threads[i]->th.th_bar_arrive_time = 0;
           }