extended pickling for KNL

TACC · Nov 1, 2016 · e8c9c79 · e8c9c79
1 parent 4b793a0
commit e8c9c79
Show file tree

Hide file tree

Showing 8 changed files with 15 additions and 9 deletions.
diff --git a/tacc_stats.ini b/tacc_stats.ini
@@ -3,7 +3,7 @@
 # server	= database and rmq server hostname
 # data_dir	= where data is stored
 [DEFAULT]
-machine		= maverick
+machine		= stampede-knl-15
 data_dir 	= /tacc_stats_site/%(machine)s
 server 		= tacc-stats03.tacc.utexas.edu
 
@@ -16,8 +16,8 @@ rmq_queue 	= %(machine)s
 
 ## Configuration for Web Portal Support
 [PORTAL]
-acct_path       = %(data_dir)s/accounting
-host_list_dir   = %(data_dir)s/hostfile_logs
+acct_path       = /scratch/projects/accounting/tacc_jobs_completed_knl
+host_list_dir   = /scratch/projects/tacc_stats/hostfile_logs
 pickles_dir     = %(data_dir)s/pickles
 archive_dir	= %(data_dir)s/archive
 host_name_ext   = %(machine)s.tacc.utexas.edu

diff --git a/tacc_stats/analysis/exam/lowflops.py b/tacc_stats/analysis/exam/lowflops.py
@@ -23,7 +23,7 @@ def compute_metric(self):
     if self.ts.pmc_type == 'amd64' :
       gfloprate += self.arc(self.ts.data[0])
 
-    if self.ts.pmc_type == 'intel_hsw':
+    if self.ts.pmc_type == 'intel_hsw' or self.ts.pmc_type == 'intel_knl':
       # print "Haswell chips do not have FLOP counters"
       return
 

diff --git a/tacc_stats/analysis/exam/vecpercent.py b/tacc_stats/analysis/exam/vecpercent.py
@@ -24,7 +24,7 @@ def compute_metric(self):
     if self.ts.pmc_type == 'amd64' :
       gvecrate += self.arc(self.ts.data[0])
 
-    if self.ts.pmc_type == 'intel_hsw':
+    if self.ts.pmc_type == 'intel_hsw' or self.ts.pmc_type == 'intel_knl' :
       #print "Haswell does not support FLOP counters"
       return
     if self.ts.pmc_type == 'intel_snb':

diff --git a/tacc_stats/analysis/gen/tspl.py b/tacc_stats/analysis/gen/tspl.py
@@ -80,6 +80,8 @@ def __init__(self,file,k1,k2,job_data = None):
       self.pmc_type='intel_ivb'
     elif 'intel_hsw' in self.j.hosts.values()[0].stats:
       self.pmc_type='intel_hsw'
+    elif 'intel_knl' in self.j.hosts.values()[0].stats:
+      self.pmc_type='intel_knl'
 
     default_wayness = len(self.j.hosts.values()[0].stats[self.pmc_type].keys())
 

diff --git a/tacc_stats/analysis/job_printer.py b/tacc_stats/analysis/job_printer.py
@@ -8,8 +8,7 @@
 def main(**args):
 
     acct = batch_acct.factory(cfg.batch_system,
-                              cfg.acct_path,
-                              cfg.host_name_ext)
+                              cfg.acct_path)
     if args['jobid']:
         reader = acct.find_jobids(args['jobid']).next()    
         date_dir = os.path.join(cfg.pickles_dir,

diff --git a/tacc_stats/pickler/intel_process.py b/tacc_stats/pickler/intel_process.py
@@ -13,6 +13,8 @@
 ## Processor events
 def CORE_PERF_EVENT(event_select, unit_mask):
     return event_select | (unit_mask << 8) | (1L << 16) | (1L << 17) | (0L << 21) | (1L << 22)
+
+# bit 21 is any thread (every hardware thread on a core increments on event)
 def CORE_PERF_EVENT1(event_select, unit_mask):
     return event_select | (unit_mask << 8) | (1L << 16) | (1L << 17) | (1L << 21) | (1L << 22)
 ## Processor event map
@@ -49,6 +51,8 @@ def CORE_PERF_EVENT1(event_select, unit_mask):
     CORE_PERF_EVENT1(0x10,0x80) : 'SSE_DOUBLE_SCALAR,E',
     CORE_PERF_EVENT1(0x10,0x10) : 'SSE_DOUBLE_PACKED,E',
     CORE_PERF_EVENT1(0xF1,0x07) : 'L2_LINES_IN_ALL,E',
+    CORE_PERF_EVENT(0x04, 0x40) : 'MEM_UOPS_RETIRED_ALL_LOADS',
+    CORE_PERF_EVENT(0x04, 0x02) : 'MEM_UOPS_RETIRED_L2_HIT_LOADS',
     'FIXED0'                   : 'INSTRUCTIONS_RETIRED,E',
     'FIXED1'                   : 'CLOCKS_UNHALTED_CORE,E',
     'FIXED2'                   : 'CLOCKS_UNHALTED_REF,E',
@@ -262,6 +266,7 @@ def register(self,host):
               'intel_hsw' : cpu_event_map, 'intel_hsw_cbo' : cbo_event_map, 'intel_hsw_hau' : hau_event_map, 
               'intel_hsw_imc' : imc_event_map,  'intel_hsw_qpi' : qpi_event_map, 'intel_hsw_pcu' : pcu_event_map, 'intel_hsw_r2pci' : r2pci_event_map,
               'intel_hsw_ht' : cpu_event_map, 'intel_hsw_cbo_ht' : cbo_event_map,
+              'intel_knl' : cpu_event_map,
 }
 
 

diff --git a/tacc_stats/site/machine/update_db.py b/tacc_stats/site/machine/update_db.py
@@ -323,5 +323,5 @@ def update_metric_fields(date,rerun=False):
 
     for date in daterange(start, end):
         directory = date.strftime("%Y-%m-%d")
-        update_acct(directory, rerun = False)         
+        update(directory, rerun = False)         
         update_metric_fields(directory, rerun = False)
diff --git a/tacc_stats/site/tacc_stats_site/settings.py b/tacc_stats/site/tacc_stats_site/settings.py
@@ -19,7 +19,7 @@
         'NAME'  : cfg.machine + '_db',
         'USER': 'taccstats',
         'PASSWORD': 'taccstats',
-        'HOST': 'localhost',         
+        'HOST': cfg.server,         
         'PORT': '5432',               
         },
     # Uncomment this portion if an xalt database exists