[OpenMP] Fix preprocessor mismatches between include and usages of hwloc

openmp/runtime/src/kmp.h

@@ -97,12 +97,15 @@ class kmp_stats_list;
 // OMPD_SKIP_HWLOC used in libompd/omp-icv.cpp to avoid OMPD depending on hwloc
 #if KMP_USE_HWLOC && KMP_AFFINITY_SUPPORTED && !defined(OMPD_SKIP_HWLOC)
 #include "hwloc.h"
+#define KMP_HWLOC_ENABLED 1
 #ifndef HWLOC_OBJ_NUMANODE
 #define HWLOC_OBJ_NUMANODE HWLOC_OBJ_NODE
 #endif
 #ifndef HWLOC_OBJ_PACKAGE
 #define HWLOC_OBJ_PACKAGE HWLOC_OBJ_SOCKET
 #endif
+#else
+#define KMP_HWLOC_ENABLED 0
 #endif
 
 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
@@ -672,10 +675,10 @@ typedef BOOL (*kmp_SetThreadGroupAffinity_t)(HANDLE, const GROUP_AFFINITY *,
 extern kmp_SetThreadGroupAffinity_t __kmp_SetThreadGroupAffinity;
 #endif /* KMP_OS_WINDOWS */
 
-#if KMP_USE_HWLOC && !defined(OMPD_SKIP_HWLOC)
+#if KMP_HWLOC_ENABLED
 extern hwloc_topology_t __kmp_hwloc_topology;
 extern int __kmp_hwloc_error;
-#endif
+#endif // KMP_HWLOC_ENABLED
 
 extern size_t __kmp_affin_mask_size;
 #define KMP_AFFINITY_CAPABLE() (__kmp_affin_mask_size > 0)
@@ -784,10 +787,10 @@ class KMPAffinity {
   static void destroy_api();
   enum api_type {
     NATIVE_OS
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
     ,
     HWLOC
-#endif
+#endif // KMP_HWLOC_ENABLED
   };
   virtual api_type get_api_type() const {
     KMP_ASSERT(0);
@@ -856,9 +859,9 @@ enum affinity_top_method {
   affinity_top_method_group,
 #endif /* KMP_GROUP_AFFINITY */
   affinity_top_method_flat,
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   affinity_top_method_hwloc,
-#endif
+#endif // KMP_HWLOC_ENABLED
   affinity_top_method_default
 };
 
@@ -1125,9 +1128,9 @@ typedef struct kmp_allocator_t {
   omp_alloctrait_value_t target_access;
   omp_alloctrait_value_t atomic_scope;
   size_t part_size;
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   omp_alloctrait_value_t membind;
-#endif
+#endif // KMP_HWLOC_ENABLED
 } kmp_allocator_t;
 
 extern omp_allocator_handle_t __kmpc_init_allocator(int gtid,
@@ -2087,12 +2090,12 @@ typedef struct dispatch_shared_info {
 #if KMP_USE_HIER_SCHED
   void *hier;
 #endif
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   // When linking with libhwloc, the ORDERED EPCC test slows down on big
   // machines (> 48 cores). Performance analysis showed that a cache thrash
   // was occurring and this padding helps alleviate the problem.
   char padding[64];
-#endif
+#endif // KMP_HWLOC_ENABLED
 } dispatch_shared_info_t;
 
 typedef struct kmp_disp {

openmp/runtime/src/kmp_affinity.cpp

@@ -19,13 +19,13 @@
 #if KMP_USE_HIER_SCHED
 #include "kmp_dispatch_hier.h"
 #endif
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
 // Copied from hwloc
 #define HWLOC_GROUP_KIND_INTEL_MODULE 102
 #define HWLOC_GROUP_KIND_INTEL_TILE 103
 #define HWLOC_GROUP_KIND_INTEL_DIE 104
 #define HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP 220
-#endif
+#endif // KMP_HWLOC_ENABLED
 #include <ctype.h>
 
 // The machine topology
@@ -1440,15 +1440,15 @@ void KMPAffinity::pick_api() {
   KMPAffinity *affinity_dispatch;
   if (picked_api)
     return;
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   // Only use Hwloc if affinity isn't explicitly disabled and
   // user requests Hwloc topology method
   if (__kmp_affinity_top_method == affinity_top_method_hwloc &&
       __kmp_affinity.type != affinity_disabled) {
     affinity_dispatch = new KMPHwlocAffinity();
     __kmp_hwloc_available = true;
   } else
-#endif
+#endif // KMP_HWLOC_ENABLED
   {
     affinity_dispatch = new KMPNativeAffinity();
   }
@@ -1699,7 +1699,7 @@ kmp_affin_mask_t *__kmp_affin_fullMask = NULL;
 // Original mask is a subset of full mask in multiple processor groups topology
 kmp_affin_mask_t *__kmp_affin_origMask = NULL;
 
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
 static inline bool __kmp_hwloc_is_cache_type(hwloc_obj_t obj) {
 #if HWLOC_API_VERSION >= 0x00020000
   return hwloc_obj_type_is_cache(obj->type);
@@ -2007,7 +2007,7 @@ static bool __kmp_affinity_create_hwloc_map(kmp_i18n_id_t *const msg_id) {
   __kmp_topology->sort_ids();
   return true;
 }
-#endif // KMP_USE_HWLOC
+#endif // KMP_HWLOC_ENABLED
 
 // If we don't know how to retrieve the machine's processor topology, or
 // encounter an error in doing so, this routine is called to form a "flat"
@@ -4854,7 +4854,7 @@ static bool __kmp_aux_affinity_initialize_topology(kmp_affinity_t &affinity) {
 // In the default code path, errors are not fatal - we just try using
 // another method. We only emit a warning message if affinity is on, or the
 // verbose flag is set, an the nowarnings flag was not set.
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
     if (!success &&
         __kmp_affinity_dispatch->get_api_type() == KMPAffinity::HWLOC) {
       if (!__kmp_hwloc_error) {
@@ -4866,7 +4866,7 @@ static bool __kmp_aux_affinity_initialize_topology(kmp_affinity_t &affinity) {
         KMP_INFORM(AffIgnoringHwloc, env_var);
       }
     }
-#endif
+#endif // KMP_HWLOC_ENABLED
 
 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
     if (!success) {
@@ -4914,7 +4914,7 @@ static bool __kmp_aux_affinity_initialize_topology(kmp_affinity_t &affinity) {
 // If the user has specified that a paricular topology discovery method is to be
 // used, then we abort if that method fails. The exception is group affinity,
 // which might have been implicitly set.
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   else if (__kmp_affinity_top_method == affinity_top_method_hwloc) {
     KMP_ASSERT(__kmp_affinity_dispatch->get_api_type() == KMPAffinity::HWLOC);
     success = __kmp_affinity_create_hwloc_map(&msg_id);
@@ -4923,7 +4923,7 @@ static bool __kmp_aux_affinity_initialize_topology(kmp_affinity_t &affinity) {
       KMP_FATAL(MsgExiting, __kmp_i18n_catgets(msg_id));
     }
   }
-#endif // KMP_USE_HWLOC
+#endif // KMP_HWLOC_ENABLED
 
 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
   else if (__kmp_affinity_top_method == affinity_top_method_x2apicid ||
@@ -5322,12 +5322,12 @@ void __kmp_affinity_uninitialize(void) {
     __kmp_free(__kmp_osid_to_hwthread_map);
     __kmp_osid_to_hwthread_map = NULL;
   }
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   if (__kmp_hwloc_topology != NULL) {
     hwloc_topology_destroy(__kmp_hwloc_topology);
     __kmp_hwloc_topology = NULL;
   }
-#endif
+#endif // KMP_HWLOC_ENABLED
   if (__kmp_hw_subset) {
     kmp_hw_subset_t::deallocate(__kmp_hw_subset);
     __kmp_hw_subset = nullptr;

openmp/runtime/src/kmp_affinity.h

@@ -18,7 +18,7 @@
 #include <limits>
 
 #if KMP_AFFINITY_SUPPORTED
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
 class KMPHwlocAffinity : public KMPAffinity {
 public:
   class Mask : public KMPAffinity::Mask {
@@ -109,7 +109,7 @@ class KMPHwlocAffinity : public KMPAffinity {
       }
       return error;
     }
-#endif
+#endif // KMP_OS_WINDOWS
     int get_proc_group() const override {
       int group = -1;
 #if KMP_OS_WINDOWS
@@ -191,7 +191,7 @@ class KMPHwlocAffinity : public KMPAffinity {
   }
   api_type get_api_type() const override { return HWLOC; }
 };
-#endif /* KMP_USE_HWLOC */
+#endif /* KMP_HWLOC_ENABLED */
 
 #if KMP_OS_LINUX || KMP_OS_FREEBSD || KMP_OS_NETBSD || KMP_OS_DRAGONFLY ||     \
     KMP_OS_AIX

openmp/runtime/src/kmp_alloc.cpp

@@ -14,7 +14,7 @@
 #include "kmp_io.h"
 #include "kmp_wrapper_malloc.h"
 
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
 #if HWLOC_API_VERSION > 0x00020300
 #define KMP_HWLOC_LOCATION_TYPE_CPUSET HWLOC_LOCATION_TYPE_CPUSET
 #elif HWLOC_API_VERSION == 0x00020300
@@ -26,7 +26,7 @@ enum hwloc_memattr_id_e {
   HWLOC_MEMATTR_ID_CAPACITY
 };
 #endif
-#endif // KMP_USE_HWLOC
+#endif // KMP_HWLOC_ENABLED
 
 // Disable bget when it is not used
 #if KMP_USE_BGET
@@ -1545,7 +1545,7 @@ void __kmp_fini_memkind() {
 #endif
 }
 
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
 static bool __kmp_is_hwloc_membind_supported(hwloc_membind_policy_t policy) {
 #if HWLOC_API_VERSION >= 0x00020300
   const hwloc_topology_support *support;
@@ -1561,7 +1561,7 @@ static bool __kmp_is_hwloc_membind_supported(hwloc_membind_policy_t policy) {
   return false;
 #else
   return false;
-#endif
+#endif // KMP_HWLOC_ENABLED
 }
 
 void *__kmp_hwloc_alloc_membind(hwloc_memattr_id_e attr, size_t size,
@@ -1611,7 +1611,7 @@ void *__kmp_hwloc_membind_policy(omp_memspace_handle_t ms, size_t size,
   return NULL;
 #endif
 }
-#endif // KMP_USE_HWLOC
+#endif // KMP_HWLOC_ENABLED
 
 void __kmp_init_target_mem() {
   *(void **)(&kmp_target_alloc_host) = KMP_DLSYM("llvm_omp_target_alloc_host");
@@ -1680,13 +1680,13 @@ omp_allocator_handle_t __kmpc_init_allocator(int gtid, omp_memspace_handle_t ms,
       al->fb_data = RCAST(kmp_allocator_t *, traits[i].value);
       break;
     case omp_atk_partition:
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
       al->membind = (omp_alloctrait_value_t)traits[i].value;
       KMP_DEBUG_ASSERT(al->membind == omp_atv_environment ||
                        al->membind == omp_atv_nearest ||
                        al->membind == omp_atv_blocked ||
                        al->membind == omp_atv_interleaved);
-#endif
+#endif // KMP_HWLOC_ENABLED
       al->memkind = RCAST(void **, traits[i].value);
       break;
     case omp_atk_pin_device:
@@ -1980,7 +1980,7 @@ void *__kmp_alloc(int gtid, size_t algn, size_t size,
     }
   }
 
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   if (__kmp_hwloc_available) {
     if (__kmp_is_hwloc_membind_supported(HWLOC_MEMBIND_BIND)) {
       if (allocator < kmp_max_mem_alloc) {
@@ -2074,7 +2074,7 @@ void *__kmp_alloc(int gtid, size_t algn, size_t size,
       ptr = hwloc_alloc(__kmp_hwloc_topology, desc.size_a);
     }
   } else {
-#endif
+#endif // KMP_HWLOC_ENABLED
     if (__kmp_memkind_available) {
       if (allocator < kmp_max_mem_alloc) {
         // pre-defined allocator
@@ -2201,9 +2201,9 @@ void *__kmp_alloc(int gtid, size_t algn, size_t size,
         KMP_ASSERT(0); // abort fallback requested
       } // no sense to look for another fallback because of same internal alloc
     }
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   }
-#endif
+#endif // KMP_HWLOC_ENABLED
   KE_TRACE(10, ("__kmp_alloc: T#%d %p=alloc(%d)\n", gtid, ptr, desc.size_a));
   if (ptr == NULL)
     return NULL;
@@ -2339,7 +2339,7 @@ void ___kmpc_free(int gtid, void *ptr, omp_allocator_handle_t allocator) {
     kmp_target_unlock_mem(desc.ptr_alloc, device);
   }
 
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   if (__kmp_hwloc_available) {
     if (oal > kmp_max_mem_alloc && al->pool_size > 0) {
       kmp_uint64 used =
@@ -2349,7 +2349,7 @@ void ___kmpc_free(int gtid, void *ptr, omp_allocator_handle_t allocator) {
     }
     hwloc_free(__kmp_hwloc_topology, desc.ptr_alloc, desc.size_a);
   } else {
-#endif
+#endif // KMP_HWLOC_ENABLED
     if (__kmp_memkind_available) {
       if (oal < kmp_max_mem_alloc) {
         // pre-defined allocator
@@ -2378,9 +2378,9 @@ void ___kmpc_free(int gtid, void *ptr, omp_allocator_handle_t allocator) {
       }
       __kmp_thread_free(__kmp_thread_from_gtid(gtid), desc.ptr_alloc);
     }
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   }
-#endif
+#endif // KMP_HWLOC_ENABLED
 }
 
 /* If LEAK_MEMORY is defined, __kmp_free() will *not* free memory. It causes

openmp/runtime/src/kmp_dispatch.h

@@ -182,12 +182,12 @@ template <typename T> struct dispatch_shared_info_template {
 #if KMP_USE_HIER_SCHED
   kmp_hier_t<T> *hier;
 #endif
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
   // When linking with libhwloc, the ORDERED EPCC test slowsdown on big
   // machines (> 48 cores). Performance analysis showed that a cache thrash
   // was occurring and this padding helps alleviate the problem.
   char padding[64];
-#endif
+#endif // KMP_HWLOC_ENABLED
 };
 
 /* ------------------------------------------------------------------------ */

openmp/runtime/src/kmp_global.cpp

@@ -248,10 +248,10 @@ enum mic_type __kmp_mic_type = non_mic;
 
 KMPAffinity *__kmp_affinity_dispatch = NULL;
 
-#if KMP_USE_HWLOC
+#if KMP_HWLOC_ENABLED
 int __kmp_hwloc_error = FALSE;
 hwloc_topology_t __kmp_hwloc_topology = NULL;
-#endif
+#endif // KMP_HWLOC_ENABLED
 
 #if KMP_OS_WINDOWS
 #if KMP_GROUP_AFFINITY