[WIP] Fix gpumetrics sampler

ovis-hpc · Sep 5, 2023 · 920e281 · 920e281
1 parent 4209fbc
commit 920e281
Show file tree

Hide file tree

Showing 4 changed files with 6 additions and 5 deletions.
diff --git a/ldms/src/contrib/sampler/gpu_metrics_sampler/Makefile.am b/ldms/src/contrib/sampler/gpu_metrics_sampler/Makefile.am
@@ -66,7 +66,7 @@ AM_CPPFLAGS = @OVIS_INCLUDE_ABS@ $(SDLCFLAGS)
 AM_LDFLAGS = @OVIS_LIB_ABS@ $(SDLLDFLAGS)
 COMMON_LIBADD = -lsampler_base -lldms -lovis_util -lcoll -lovis_log\
 		@LDFLAGS_GETTIME@
-GMGLIBS = -L. -lgmg -L/usr/lib/x86_64-linux-gnu/ -lze_loader
+GMGLIBS = -L. ./libgmg.a -L/usr/lib/x86_64-linux-gnu/ -lze_loader
 
 if ENABLE_GPU_METRICS
 
@@ -80,7 +80,7 @@ pkglib_LTLIBRARIES += libgpumetrics.la
 
 bin_PROGRAMS = bin/gpu_metrics_gatherer
 bin_gpu_metrics_gatherer_SOURCES = gather_gpu_metrics.cpp
-bin_gpu_metrics_gatherer_LDADD = $(GMGLIBS)
+bin_gpu_metrics_gatherer_LDADD = $(GMGLIBS) -lovis_log
 
 
 endif
diff --git a/ldms/src/contrib/sampler/gpu_metrics_sampler/gather_gpu_metrics_from_one_api.c b/ldms/src/contrib/sampler/gpu_metrics_sampler/gather_gpu_metrics_from_one_api.c
@@ -48,6 +48,7 @@
 #include "gmg_log.h"
 #include "gather_gpu_metrics_from_one_api.h"
 #include "gmg_ldms_util.h"
+#include "ovis_log/ovis_log.h"
 #include <string.h>
 #include <time.h>
 #include <stdlib.h>
@@ -694,7 +695,7 @@ uint32_t readRasErrorPropAndState(ze_device_handle_t hDevice, zes_ras_properties
     ze_bool_t clear = 0;
     res = zesRasGetState(pHandle[0], clear, &state);
     if (res != ZE_RESULT_SUCCESS) {
-        GMGLOG(OVIS_LERROR, "!!!zesRasGetState(pHandle[0]=%p,clear=%p,&state=%p) => 0x%x\n",
+        GMGLOG(OVIS_LERROR, "!!!zesRasGetState(pHandle[0]=%p,clear=%d,&state=%p) => 0x%x\n",
                pHandle[0], clear, &state, res);
         return -9999;
     }

diff --git a/ldms/src/contrib/sampler/gpu_metrics_sampler/gmg_ldms_util.c b/ldms/src/contrib/sampler/gpu_metrics_sampler/gmg_ldms_util.c
@@ -112,7 +112,7 @@ int populateMetricSchema(ldms_schema_t schema, uint32_t numDevices) {
         for (size_t i = 0; i < c_numMetrics; i++) {
             char szMetricName[MAX_METRIC_NAME_LENGTH + 1] = {};
             constructMetricName(metricsDefinitions[i].name, deviceId, szMetricName);
-            GMGLOG(OVIS_LDEBUG, "metricsDefinitions[i=%d].name = %s\n", i, metricsDefinitions[i].name);
+            GMGLOG(OVIS_LDEBUG, "metricsDefinitions[i=%ld].name = %s\n", i, metricsDefinitions[i].name);
             GMGLOG(OVIS_LDEBUG, "szMetricName = %s\n", szMetricName);
             if (ldms_type_is_array(metricsDefinitions[i].type)) {
                 rc = ldms_schema_metric_array_add(schema, szMetricName,

diff --git a/ldms/src/contrib/sampler/gpu_metrics_sampler/gpu_metrics_ldms_sampler.c b/ldms/src/contrib/sampler/gpu_metrics_sampler/gpu_metrics_ldms_sampler.c
@@ -184,7 +184,7 @@ static int create_metric_set_schema_and_set(base_data_t base) {
 static void printValList(const char *szListName, struct attr_value_list *av_list) {
     size_t listSize = MIN(av_list->count, av_list->size);
     for (size_t i = 0; i < listSize; i++) {
-        ovis_log(__gpu_metrics_log, OVIS_LDEBUG, "%s[%d] = %s:%s\n",
+        ovis_log(__gpu_metrics_log, OVIS_LDEBUG, "%s[%ld] = %s:%s\n",
                szListName, i, av_name(av_list, i), av_value_at_idx(av_list, i));
     }
 }