Cambios cruzados

app_profiler.py

@@ -42,14 +42,23 @@ Function that calls the profiler based on the injection mode
 """
 
 
-def profiler_caller(gdb_exec, kernel, benchmark_binary, benchmark_args,device,section,kernel_end):
+def profiler_caller(gdb_exec, kernels, benchmark_binary, benchmark_args,device,section):
     acc_time = 0
     acc_time_profiler=0
     script = 'env CUDA_VISIBLE_DEVICES={} {} -ex \'py arg0 = {}\' -n -batch -x {}'
     benchmark_args_striped = benchmark_args.replace('\\n', '').replace('\\', '')
-    print ("KERNEL"+kernel)
+   # print ("KERNEL"+kernels)
     #init_string = '"file {}; set args {}"'.format(benchmark_binary, benchmark_args_striped)
-    init_string = '"{};{};file {}; set args {}; break {}"'.format(section,kernel_end,benchmark_binary, benchmark_args_striped,kernel)
+    print ("SECTION {}".format(section))
+    
+    init_string = '"{};{};{};file {}; set args {}; set cuda break_on_launch application"'.format(False,True,kernels,benchmark_binary, benchmark_args_striped)
+    profiler_cmd = script.format(device, gdb_exec, init_string, cp.PROFILER_SCRIPT)
+    max_registers=os.system(profiler_cmd) >>8
+    print(max_registers,max_registers>>8)
+    if  bool(section):
+      init_string = '"{};{};{};file {}; set args {}; break {}; break {}"'.format( bool(section),False,kernels,benchmark_binary,benchmark_args_striped,section['begin'],section['end'])
+    else:  
+      init_string = '"{};{};{};file {}; set args {}; break {}"'.format(False,False,kernels,benchmark_binary, benchmark_args_striped,kernels.split(",")[0])
     profiler_cmd = script.format(device, gdb_exec, init_string, cp.PROFILER_SCRIPT)
     print ("Profiler caller")
     if cp.DEBUG:
@@ -64,7 +73,7 @@ def profiler_caller(gdb_exec, kernel, benchmark_binary, benchmark_args,device,se
         acc_time += end - start
         cf.kill_all("killall -9 {}; killall -9 {}".format(
             os.path.basename(gdb_exec), os.path.basename(benchmark_binary)))
-    return acc_time_profiler / cp.MAX_TIMES_TO_PROFILE, acc_time / cp.MAX_TIMES_TO_PROFILE
+    return acc_time_profiler / cp.MAX_TIMES_TO_PROFILE, acc_time / cp.MAX_TIMES_TO_PROFILE, max_registers
 
 
 
@@ -81,7 +90,7 @@ def generate_gold(gdb_exec, benchmark_binary, benchmark_args,device):
     os.system("rm -rf " + tmp_path + "/*")
 
     script = 'env CUDA_VISIBLE_DEVICES={} {} -ex \'py arg0 = {}\' -n -batch -x {} > {} 2> {}'
-    init_string = '"file {}; set args {}"'.format(benchmark_binary, benchmark_args)
+    init_string = '"{};{};{};file  {}; set args {}"'.format(False,True,"",benchmark_binary, benchmark_args)
     profiler_cmd = script.format(device, gdb_exec, init_string, cp.PROFILER_SCRIPT, cp.GOLD_OUTPUT_PATH, cp.GOLD_ERR_PATH)
     if cp.DEBUG:
         print("PROFILER CMD: {}".format(profiler_cmd))
@@ -118,28 +127,29 @@ def main():
       benchmark_args = conf.get('DEFAULT', 'benchmarkArgs_noverificar')
     else:
       benchmark_args = conf.get('DEFAULT', 'benchmarkArgs')
-    section= 'kernel_end' in conf['DEFAULT']
-    kernel_end=''
-    if (section):
-       kernel_end=conf.get('DEFAULT','kernel_end')
+ 
+    section={}
+    if ('section_end' in conf['DEFAULT']):    
+       section['begin']=conf.get('DEFAULT','section_begin')
+       section['end']=conf.get('DEFAULT','section_end')
     gdb_exec = conf.get("DEFAULT", "gdbExecName")
-    kernel=conf.get('DEFAULT', 'kernel')
-    [max_time_kernel,max_time_app] = profiler_caller(gdb_exec=gdb_exec,kernel=kernel, benchmark_binary=benchmark_binary, benchmark_args=benchmark_args,device=args.device,section=section,kernel_end=kernel_end)
+    kernels=conf.get('DEFAULT', 'kernels')
+    [max_time_kernel,max_time_app,max_regs] = profiler_caller(gdb_exec=gdb_exec,kernels=kernels, benchmark_binary=benchmark_binary, benchmark_args=benchmark_args,device=args.device,section=section)
     print ("Time kernel= "+str(max_time_kernel)+ "Time app "+str(max_time_app))
     # saving gold
     print ("Saving gold");
-    #generate_gold_result = generate_gold(gdb_exec=gdb_exec,
-    #                                     benchmark_binary=benchmark_binary, benchmark_args=benchmark_args,device=args.device)
+    generate_gold_result = generate_gold(gdb_exec=gdb_exec,
+                                         benchmark_binary=benchmark_binary, benchmark_args=benchmark_args,device=args.device)
 
-    #if generate_gold_result != 0:
-    #   raise EnvironmentError("Gold generation did not finish well, the fault injection will not work")
+    if generate_gold_result != 0:
+       raise EnvironmentError("Gold generation did not finish well, the fault injection will not work")
 
     # Remove trash GDB info from the std output and the err output
     cf.remove_useless_information_from_output(cp.GOLD_OUTPUT_PATH)
     cf.remove_useless_information_from_output(cp.GOLD_ERR_PATH)
 
     # Save the kernel configuration txt file
-    cf.save_file(file_path=cp.KERNEL_INFO_DIR, data={'max_time': max_time_app,'max_time_kernel': max_time_kernel})
+    cf.save_file(file_path=cp.KERNEL_INFO_DIR, data={'max_time': max_time_app,'max_time_kernel': max_time_kernel,'max_regs':max_regs})
 
     print("1 - Profile finished\n###################################################")
 

codes/mmElem/matrixmul.conf

@@ -17,7 +17,7 @@ faultModel = 0
 # RF -> Register File
 # INST_OUT -> Instruction Output (NOT IMPLEMENTED YET)
 # INST_composed -> Instruction Adress (NOT IMPLEMENTED YET)
-injectionSite = RF
+injectionSite = INST_OUT
 # Max time factor to finish the app, this will be multiplied by the application running time
 # For example if your app spend 2s, and the maxWaitTimes is 5, the max running time before it is
 # Considered as a crash is 10s
@@ -42,6 +42,6 @@ seqSignals = 20
 # Initial sleep time in seconds before start sending signals
 # Generally the memory setup time
 initSleep = 2.1
-#kernel = matrixMulCUDA
-kernel = matrixMul.cu:205
-kernel_end = matrixMul.cu:217
+kernels = matrixMulCUDA
+section_begin = matrixMul.cu:205
+section_end = matrixMul.cu:217

fault_injector.py

@@ -300,6 +300,7 @@ def check_injection_outcome(host_thread, logging, injection_site):
             # if fault was injected ASSM_LINE MUST be in the logfile
             assm_line = logging.search("ASSM_LINE")
             instruction = re.match(r".*:\t(\S+) .*", assm_line).group(1)
+            instruction = assm_line
 
     except TypeError as te:
         instruction = new_value = old_value = None
@@ -701,7 +702,7 @@ def main():
             'current_path': current_path,
             'seq_signals': int(conf.get('DEFAULT', 'seqSignals')),
             'init_sleep': float(conf.get('DEFAULT', 'initSleep')),
-            'kernel':conf.get('DEFAULT', 'kernel'),
+            'kernel':conf.get('DEFAULT', 'section_begin'),
             'gold_check_script': "{}/{}".format(current_path, conf.get('DEFAULT', 'goldenCheckScript')),
             'summary_file': summary_file,
             'header': fieldnames
@@ -718,7 +719,7 @@ def main():
 
     for thread in gpus_threads:
         thread.join()
-        ret += thread.exit_code
+        #ret += thread.exit_code
 
     os.system("rm -f {}/bin/*".format(current_path))
     if exit_injector:

profiler_new.py

 import gdb
-import time
 
+import time
+import re
+import sys
+import common_functions as cf 
 
 def exit_handler(event):
    global nosalir
@@ -16,9 +19,40 @@ def exit_handler(event):
 Handler that will put a breakpoint on the kernel after
 signal
 """
+def numreg (kernel):
+  try:
+    disassemble_array = cf.execute_command(gdb=gdb, to_execute="disassemble {}".format(kernel))
+  except:
+    disassemble_array = cf.execute_command(gdb=gdb, to_execute="disassemble")
+  listareg=set()
+  listaregdst=set()
+  listaregcond=set()
+  for i in range(0, len(disassemble_array) - 1):
+     line = disassemble_array[i]
+     m=re.match(r".*:\t(\S+) .*",line)
+#      print("ASSM_LINE:{}".format(l))
+     todacadena="ASSM_LINE:{}".format(line)
+     print(todacadena)
+     lista=re.findall(r"R(\d+)", line)
+     if (len(lista) > 0):   	
+         listareg.update(lista)
+         print (listareg)
+         listaregdst.add (lista[0])
+         print (listaregdst)
+     lista=re.findall(r" P(\d+)", line)
+     if (len(lista) > 0): 
+         print (listaregcond)
+         listaregcond.update(lista)
+  #       if m and m.group(1) != '0':
+  #       Print ("Encontrado"
 
-
-
+  print ("Registros Visibles ({})".format(len(listareg)))
+  print (listareg)
+  print ("Registros destino ({})".format(len(listaregdst)))
+  print (listaregdst)
+  print ("Registros condicionales ({})".format(len(listaregcond)))
+  print (listaregcond)
+  return (len(listareg))
 def set_event(event):
       global trun,ocurrencias,t,primera
       
@@ -58,37 +92,29 @@ def main():
 
   # gdb_init_strings = str(os.environ["CAROL_FI_INFO"])
   gdb_init_strings = arg0
-  cadena=gdb_init_strings.split(";",2)
-  
-  #print (cadena,"-",cadena[0],'-',cadena[1],'-', cadena[2])
-  section =cadena[0]=="True"
-  kernel_end=cadena[1]
-  
+  cadena=gdb_init_strings.split(";",3)
+ 
+  print >>sys.stderr, (cadena[0]) #,"-",cadena[0],"-",cadena[1],"-", cadena[2],"#". cadena[3])
+  section =cadena[0]=="True" 
+  onlycount=cadena[1]=="True" 
   #print ("B "+section+"ke "+kernel_end+" ....")
   #print (cadena[2].split(";"))
-  for init_str in  cadena[2].split(";"):
+  for init_str in  cadena[3].split(";"):
      gdb.execute(init_str)
-  if (section):
-      gdb.execute ("break "+kernel_end)
+  max=0.  
   gdb.execute("r")
-
-  #nosalir=True
-  #while nosalir:
-    
-  if (section):
-      #print ("Point 1")
-      gdb.execute("c")
+  if (onlycount):     
+    ks=cadena[2].split(",",1)
+    max=numreg(ks[0])
   else:
-      gdb.execute("finish")
-  #print ("Punto 2")
-  #print (" Ocurrencias "+str(ocurrencias)+" Tiempo acumulado de ejecucciones "+ str(trun)+ "\n")
-  gdb.execute("c")
+     if (section):
+       gdb.execute("c")
+     else:
+       gdb.execute("finish")
   
-  #print (" Ocurrencias "+str(ocurrencias)+" Tiempo acumulado de ejecucciones "+ str(trun))
-  f=open("tmpxxx_return_profiler.conf","w")
-  f.write("[DEFAULT] \nOcurrencias = "+str(ocurrencias)+"\nTiempo = "+str(trun)+"\n")
-  f.close()	
-  #print ("End write file \n")	
-  #sys.stdout.flush()
+     f=open("tmpxxx_return_profiler.conf","w")
+     f.write("[DEFAULT] \nOcurrencias = "+str(ocurrencias)+"\nTiempo = "+str(trun)+"\n")
+     f.close()	
+  sys.exit(max)  
 
 main()