AMD GPUs: Call rsmi_init() once

LLNL · Jul 7, 2023 · 5150443 · 5150443
1 parent e9b6bcc
commit 5150443
Show file tree

Hide file tree

Showing 2 changed files with 15 additions and 60 deletions.
diff --git a/src/variorum/AMD_GPU/amd_gpu_power_features.c b/src/variorum/AMD_GPU/amd_gpu_power_features.c
@@ -25,16 +25,6 @@ void get_power_data(int chipid, int total_sockets, int verbose, FILE *output)
 
     gethostname(hostname, 1024);
 
-    ret = rsmi_init(0);
-    if (ret != RSMI_STATUS_SUCCESS)
-    {
-        variorum_error_handler("Could not initialize RSMI",
-                               VARIORUM_ERROR_PLATFORM_ENV,
-                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
-                               __LINE__);
-        exit(-1);
-    }
-
     ret = rsmi_num_monitor_devices(&num_devices);
     if (ret != RSMI_STATUS_SUCCESS)
     {
@@ -115,16 +105,6 @@ void get_power_limit_data(int chipid, int total_sockets, int verbose,
 
     gethostname(hostname, 1024);
 
-    ret = rsmi_init(0);
-    if (ret != RSMI_STATUS_SUCCESS)
-    {
-        variorum_error_handler("Could not initialize RSMI",
-                               VARIORUM_ERROR_PLATFORM_ENV,
-                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
-                               __LINE__);
-        exit(-1);
-    }
-
     ret = rsmi_num_monitor_devices(&num_devices);
     if (ret != RSMI_STATUS_SUCCESS)
     {
@@ -217,16 +197,6 @@ void get_thermals_data(int chipid, int total_sockets, int verbose, FILE *output)
 
     gethostname(hostname, 1024);
 
-    ret = rsmi_init(0);
-    if (ret != RSMI_STATUS_SUCCESS)
-    {
-        variorum_error_handler("Could not initialize RSMI",
-                               VARIORUM_ERROR_PLATFORM_ENV,
-                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
-                               __LINE__);
-        exit(-1);
-    }
-
     ret = rsmi_num_monitor_devices(&num_devices);
     if (ret != RSMI_STATUS_SUCCESS)
     {
@@ -307,16 +277,6 @@ void get_clocks_data(int chipid, int total_sockets, int verbose, FILE *output)
 
     gethostname(hostname, 1024);
 
-    ret = rsmi_init(0);
-    if (ret != RSMI_STATUS_SUCCESS)
-    {
-        variorum_error_handler("Could not initialize RSMI",
-                               VARIORUM_ERROR_PLATFORM_ENV,
-                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
-                               __LINE__);
-        exit(-1);
-    }
-
     ret = rsmi_num_monitor_devices(&num_devices);
     if (ret != RSMI_STATUS_SUCCESS)
     {
@@ -408,16 +368,6 @@ void get_gpu_utilization_data(int chipid, int total_sockets, int verbose,
 
     gethostname(hostname, 1024);
 
-    ret = rsmi_init(0);
-    if (ret != RSMI_STATUS_SUCCESS)
-    {
-        variorum_error_handler("Could not initialize RSMI",
-                               VARIORUM_ERROR_PLATFORM_ENV,
-                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
-                               __LINE__);
-        exit(-1);
-    }
-
     ret = rsmi_num_monitor_devices(&num_devices);
     if (ret != RSMI_STATUS_SUCCESS)
     {
@@ -493,16 +443,6 @@ void cap_each_gpu_power_limit(int chipid, int total_sockets,
 
     gethostname(hostname, 1024);
 
-    ret = rsmi_init(0);
-    if (ret != RSMI_STATUS_SUCCESS)
-    {
-        variorum_error_handler("Could not initialize RSMI",
-                               VARIORUM_ERROR_PLATFORM_ENV,
-                               getenv("HOSTNAME"), __FILE__, __FUNCTION__,
-                               __LINE__);
-        exit(-1);
-    }
-
     ret = rsmi_num_monitor_devices(&num_devices);
     if (ret != RSMI_STATUS_SUCCESS)
     {

diff --git a/src/variorum/AMD_GPU/config_amd_gpu.c b/src/variorum/AMD_GPU/config_amd_gpu.c
@@ -22,6 +22,7 @@ uint64_t *detect_amd_gpu_arch(void)
 
 int set_amd_gpu_func_ptrs(int idx)
 {
+    static int init_complete = 0;
     int err = 0;
 
     if (*g_platform[idx].arch_id == AMD_INSTINCT)
@@ -43,5 +44,19 @@ int set_amd_gpu_func_ptrs(int idx)
         err = VARIORUM_ERROR_UNSUPPORTED_PLATFORM;
     }
 
+
+    if (init_complete == 0)
+    {
+        ret = rsmi_init(0);
+        if (ret != RSMI_STATUS_SUCCESS)
+        {
+            variorum_error_handler("Could not initialize RSMI",
+                                   VARIORUM_ERROR_PLATFORM_ENV,
+                                   getenv("HOSTNAME"), __FILE__, __FUNCTION__,
+                                   __LINE__);
+            exit(-1);
+        }
+        init_complete = 1;
+    }
     return err;
 }