levelzero: disable ZES_ENABLE_SYSMAN=1 when zesInit() is available

When zesInit() is available (starting in specs 1.5), we don't need
all the stuff to set ZES_ENABLE_SYSMAN=1 in the environment anymore.

This assumes that zesInit() works for real, which was not the case
in first compute-runtime releases (it returned "unsupported").
Those releases won't be supported anymore because they'd need
a runtime-check at configure time to find out that the ZES_ENABLE_SYSMAN=1
stuff must be enabled.

Only use a compile-check to switch between zesInit() and
ZES_ENABLE_SYSMAN=1, and tell users to avoid those L0 runtime
versions where zesInit() fails.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

hwloc/topology-levelzero.c

@@ -604,8 +604,7 @@ hwloc_levelzero_discover(struct hwloc_backend *backend, struct hwloc_disc_status
   uint32_t nbdrivers, i, k, zeidx;
   struct hwloc_osdev_array oarray;
   struct hwloc_levelzero_ports hports;
-  int sysman_maybe_missing = 0; /* 1 if ZES_ENABLE_SYSMAN=1 was NOT set early, 2 if ZES_ENABLE_SYSMAN=0 */
-  char *env;
+  int sysman_maybe_missing = 0; /* 1 if ZES_ENABLE_SYSMAN=1 was NOT set early and zesInit() isn't available, 2 if ZES_ENABLE_SYSMAN=0 */
 
   assert(dstatus->phase == HWLOC_DISC_PHASE_IO);
 
@@ -620,18 +619,22 @@ hwloc_levelzero_discover(struct hwloc_backend *backend, struct hwloc_disc_status
 #ifdef HWLOC_HAVE_ZESINIT
   res = zesInit(0);
   if (res != ZE_RESULT_SUCCESS) {
-    hwloc_debug("hwloc/levelzero: Failed to initialize LevelZero Sysman in zesInit(): 0x%x\n", (unsigned)res);
-    hwloc_debug("hwloc/levelzero: Continuing. Hopefully ZES_ENABLE_SYSMAN=1\n");
+    if (res == ZE_RESULT_ERROR_UNSUPPORTED_FEATURE) {
+      fprintf(stderr, "hwloc/levelzero: Unsupported zesInit(), please either an older or a more recent level-zero runtime.\n");
+    } else if (HWLOC_SHOW_ALL_ERRORS()) {
+      fprintf(stderr, "hwloc/levelzero: Failed to initialize LevelZero in zesInit(): 0x%x\n", (unsigned)res);
+    }
+    return 0;
   }
-#endif /* HWLOC_HAVE_ZESINIT */
-
+#else /* !HWLOC_HAVE_ZESINIT */
   /* Tell L0 to create sysman devices.
    * If somebody already initialized L0 without Sysman,
    * zesDeviceGetProperties() will fail and warn in hwloc__levelzero_properties_get().
    * The lib constructor and Windows DllMain tried to set ZES_ENABLE_SYSMAN=1 early (see topology.c),
    * we try again in case they didn't.
    */
-  env = getenv("ZES_ENABLE_SYSMAN");
+ {
+  char *env = getenv("ZES_ENABLE_SYSMAN");
   if (!env) {
     /* setenv() is safer than putenv() but not available on Windows */
 #ifdef HWLOC_WIN_SYS
@@ -644,6 +647,8 @@ hwloc_levelzero_discover(struct hwloc_backend *backend, struct hwloc_disc_status
   } else if (!atoi(env)) {
     sysman_maybe_missing = 2;
   }
+ }
+#endif /* !HWLOC_HAVE_ZESINIT */
 
   res = zeInit(0);
   if (res != ZE_RESULT_SUCCESS) {

hwloc/topology.c

@@ -54,9 +54,10 @@
 #endif
 
 
-#ifdef HWLOC_HAVE_LEVELZERO
+#if (defined HWLOC_HAVE_LEVELZERO) && !(defined HWLOC_HAVE_ZESINIT)
 /*
  * Define ZES_ENABLE_SYSMAN=1 early so that the LevelZero backend gets Sysman enabled.
+ * This is only for old releases (<1.5) without zesInit().
  *
  * Only if the levelzero was enabled in this build so that we don't enable sysman
  * for external levelzero users when hwloc doesn't need it. If somebody ever loads
@@ -101,7 +102,7 @@ BOOL WINAPI DllMain(HINSTANCE hinstDLL, DWORD fdwReason, LPVOID lpReserved)
   return TRUE;
 }
 #endif
-#endif /* HWLOC_HAVE_LEVELZERO */
+#endif /* HWLOC_HAVE_LEVELZERO && !HWLOC_HAVE_ZESINIT */
 
 
 unsigned hwloc_get_api_version(void)

tests/hwloc/levelzero.c

@@ -26,19 +26,19 @@ int main(void)
 #ifdef HWLOC_HAVE_ZESINIT
   res = zesInit(0);
   if (res != ZE_RESULT_SUCCESS) {
-    fprintf(stderr, "Failed to initialize LevelZero Sysman in zesInit(): %d\n", (int)res);
-    /* continuing, assuming ZES_ENABLE_SYSMAN=1 will be enough */
+    fprintf(stderr, "Failed to initialize LevelZero in zesInit(): %d\n", (int)res);
+    return 0;
   }
-#endif
-
+#else
   putenv((char *) "ZES_ENABLE_SYSMAN=1");
+#endif
 
   res = zeInit(0);
   if (res != ZE_RESULT_SUCCESS) {
     fprintf(stderr, "Failed to initialize LevelZero in zeInit(): %d\n", (int)res);
     return 0;
   }
-  
+
   hwloc_topology_init(&topology);
   hwloc_topology_set_io_types_filter(topology, HWLOC_TYPE_FILTER_KEEP_IMPORTANT);
   hwloc_topology_load(topology);