Version en funcionamiento

app_profiler.py

@@ -128,11 +128,11 @@ def main():
     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)

classes/BitFlip.py

@@ -200,16 +200,17 @@ class BitFlip:
 
     def __select_register(self):
         # Start on 1 to consider R0
-        max_num_register = 1
+        max_num_register = 0
         # registers_list.popleft()
         registers_list = cf.execute_command(gdb=gdb, to_execute="info registers")
         for line in registers_list:
             m = re.match(r".*R.*0x([0-9a-fA-F]+).*", line)
             if m and m.group(1) != '0':
+                #self.__logging.info ("LIne salida:{}".format(m.group(1))
                 max_num_register += 1
-
-        self.__register = "R{}".format(random.randint(0, max_num_register))
-
+                #self.__logging.info("LIne entrada {}--max{}".format(line,max_num_register))
+        self.__register = "R{}".format(random.randint(0, max_num_register-1))
+        self.__logging.info("MAX_NUM_REGISTER:{}".format(max_num_register))
     """
     Instruction injector    
     """

codes/mmElem/matrixmul.conf

@@ -17,8 +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
@@ -26,10 +25,10 @@ maxWaitTimes = 5
 
 # binary file of the application
 # Must be full path
-benchmarkBinary = /home/badia/carol-fi-carol-fi_cuda-parallel/codes/mmElem/matrixMul
+benchmarkBinary = /home/badia/mycarol-fi/codes/mmElem/matrixMul
 
 # Commands to set the session inside GDB environment
-benchmarkArgs =  -device=1 -wA=16384 -hA=16384 -hB=16384 -wB=16384
+benchmarkArgs =   -wA=8192 -hA=8192 -hB=8192 -wB=8192
 
 # CSV output file. It will be overwrite at each injection
 csvFile = codes/mmElem/fi_matrix_mul_single_bit.csv
@@ -42,5 +41,7 @@ seqSignals = 20
 
 # Initial sleep time in seconds before start sending signals
 # Generally the memory setup time
-initSleep = 2.3
-
+initSleep = 2.1
+#kernel = matrixMulCUDA
+kernel = matrixMul.cu:205
+kernel_end = matrixMul.cu:217

codes/mmElem/matrixmul_16K.conf

@@ -17,8 +17,7 @@ faultModel = 0
 # RF -> Register File
 # INST_OUT -> Instruction Output (NOT IMPLEMENTED YET)
 # INST_composed -> Instruction Adress (NOT IMPLEMENTED YET)
-injectionSite = RF, INST_OUT
-
+injectionSite = RF
 # 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
@@ -32,10 +31,10 @@ benchmarkBinary = /home/badia/mycarol-fi/codes/mmElem/matrixMul
 benchmarkArgs =   -wA=16384 -hA=16384 -hB=16384 -wB=16384
 
 # CSV output file. It will be overwrite at each injection
-csvFile = codes/matrixMul/fi_matrix_mul_single_bit.csv
+csvFile = codes/mmElem/fi_matrix_mul_single_bit.csv
 
 # You should create a script on the benchmark source folder to verify GOLD_OUTPUT x INJ_OUTPUT
-goldenCheckScript = codes/matrixMul/sdc_check.sh
+goldenCheckScript = codes/mmElem/sdc_check.sh
 
 # Number of signals that will be sent to the application
 seqSignals = 20
@@ -44,5 +43,5 @@ seqSignals = 20
 # Generally the memory setup time
 initSleep = 2.1
 #kernel = matrixMulCUDA
-kernel = matrixMul.cu:208
-kernel_end = matrixMul.cu:216
+kernel = matrixMul.cu:205
+kernel_end = matrixMul.cu:217

fault_injector.py

@@ -61,10 +61,10 @@ def signal_handler(sig, frame):
     for cmd in kill_strings.split(";"):
         os.system(cmd + " > /dev/null 2>&1")
 
-  #  os.system("rm -f {}/bin/*".format(current_path))
-  # for th in gpus_threads:
-  #     th.join()
-  #  sys.exit(0)
+    os.system("rm -f {}/bin/*".format(current_path))
+    for th in gpus_threads:
+      th.join()
+    sys.exit(0)
 
 
 """
@@ -593,7 +593,6 @@ def fault_injection_by_signal(**kwargs):
 Main function
 """
 
-
 def main():
     global kill_strings, current_path, gpus_threads, lock, syncro, wait_finish
  

flip_value.py

@@ -30,7 +30,7 @@ signal
 
 def set_event(event):
     # Accessing global vars
-    global global_logging, was_hit, bit_lip,bp,t
+    global global_logging, was_hit, bit_flip,bp,t
     if (isinstance(event, gdb.BreakpointEvent)):
       global_logging.info("Before breakpoint"+ str(time.clock()-t))
       global_logging.info ("Enviado senal a "+ str(pid))
@@ -56,7 +56,7 @@ Main function
 
 
 def main():
-    global global_logging, register, injection_site, bits_to_flip, fault_model, was_hit, bit_flip, arg0
+    global global_logging, register, injection_site, bits_to_flip, fault_model, was_hit, bit_flip, arg0,t,kernel,pid,bp
 
     was_hit = False
 

profiler_new.py

@@ -25,13 +25,16 @@ def set_event(event):
       print ("Es mi primera vez"+ str(primera)+"  "+str(ocurrencias))
       if (isinstance(event, gdb.BreakpointEvent)):
         if (primera):
-          t=time.clock()
+          
+          t=time.time()
+          print ("Tomo tiempo " + str (t))
           ocurrencias=ocurrencias+1
         else:  
-          trun=(time.clock()-t)
+          print ("Para tiempo " + str (time.time()))
+          trun=(time.time()-t)
         primera=not primera  
-      else:  
-          trun=(time.clock()-t)  
+    #  else:  
+    #      trun=(time.clock()-t)  
 """
 Main function
 """

test_mult.sh

@@ -8,7 +8,7 @@ set -e
 FAULTS=1000
 #CONFFILE=codes/matrixMul/matrixmul_16K.conf
 
-CONFFILE=codes/mmElem/matrixmul_16K.conf
+CONFFILE=codes/mmElem/matrixmul.conf
 #CONFFILE=codes/lavaMD/lavaMD.conf
 echo "Step 1 - Profiling the application for fault injection"
 ./app_profiler.py -c ${CONFFILE} $*

test_mult_mBlock.sh

@@ -6,7 +6,7 @@ set -e
 #set -x
 
 FAULTS=1000
-CONFFILE=codes/matrixMul/matrixmul_16K.conf
+CONFFILE=codes/matrixMul/matrixmul.conf
 
 #CONFFILE=codes/mmElem/matrixmul_16K.conf
 #CONFFILE=codes/lavaMD/lavaMD.conf