PMCTrack v1.5

Author: jcsaezals

README.md

@@ -6,16 +6,20 @@ Despite being an OS-oriented tool, PMCTrack still allows gathering PMC values fr
 
 ## Project Contributors
 
-* Juan Carlos Saez Alcaide (<jcsaezal@ucm.es>) - Creator of PMCTrack and main maintainer. 
-* Jorge Casas Hernan - Maintainer of PMCTrack-GUI
-* Abel Serrano Juste (aka [@Akronix](https://github.com/Akronix))
-* Javier Setoain
+* Juan Carlos Saez Alcaide (<jcsaezal@ucm.es>) - Creator of PMCTrack and main maintainer 
+* Adrián García García (aka [@Mizadri](https://github.com/mizadri))
+* Jaime Sáez de Buruaga Brouns (aka [@jaimesaez97](https://github.com/jaimesaez97))
 
 ## Past Contributors
 
-* Guillermo Martinez Fernandez
-* Sergio Sanchez Gordo
-* Sofia Dronda Merino
+* Jorge Casas Hernan 
+* Abel Serrano Juste (aka [@Akronix](https://github.com/Akronix))
+* Germán Franco Dorca
+* Andrés Plaza Hernando
+* Javier Setoain
+* Guillermo Martinez Fernández
+* Sergio Sánchez Gordo
+* Sofía Dronda Merino
 
 ## Publications
 
@@ -27,11 +31,11 @@ Despite being an OS-oriented tool, PMCTrack still allows gathering PMC values fr
 To support PMCTrack, a patched Linux kernel must be installed on the machine. A number of kernel patches for various Linux versions can be found in the `src/kernel-patches` directory. The name of each patch file encodes the Linux kernel version where the patch must be applied to as well as the processor architecture supported. The format is as follows: 
 
 	pmctrack_linux-<kernel_version>_<architecture>.patch 
-		
+
 To build the kernel for PMCTrack, the following option must be enabled when configuring the kernel:
 
 	CONFIG_PMCTRACK=y
-		
+
 
 The kernel headers for the patched Linux version must be installed on the system as well. This is necessary for a successful out-of-tree build of PMCTrack's kernel module. An out-of-tree-ready Makefile can be found in the sources for the different flavors of the kernel module.
 
@@ -56,7 +60,7 @@ On Mac OS X, PMCTrack-GUI has been succesfully tested after installing the above
 	## Set up default configuration for matplotlib 
 	$ mkdir  ~/.matplotlib
 	$ cp /opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/matplotlib/mpl-data/matplotlibrc  ~/.matplotlib
-
+	
 	## Select MacPorts Python27 interpreter by default
 	$sudo port select --set python python27
 	$sudo port select --set ipython ipython27
@@ -67,7 +71,7 @@ On Mac OS X, PMCTrack-GUI has been succesfully tested after installing the above
 The `PMCTRACK_ROOT` environment variable must be defined for a successful execution of the various PMCTrack command-line tools. The `shrc` script found in the repository's root directory can be used to set the `PMCTRACK_ROOT` variable appropriately as well as to add command-line tools' directories to the PATH. To make this possible, run the following command in the root directory of the repository:
 
 	$ . shrc
-		
+
 Now kernel-level and user-level components can be easily built with the `pmctrack-manager` script as follows:
 
 	$ pmctrack-manager build
@@ -77,7 +81,7 @@ Now kernel-level and user-level components can be easily built with the `pmctrac
 	Processor_bitwidth=64
 	***********************************************
 	Press ENTER to start the build process...
-
+	
 	*************************************************
 	*** Building supported PMCTrack kernel modules **
 	*************************************************
@@ -174,15 +178,16 @@ After that, load any of the available flavors of the PMCTrack's kernel module co
 
 The following table summarizes the properties of the various flavors of the kernel module:
 
-| Name | Path of the .ko file | Supported processors |  
-| -----| ---------------------| ---------------------|  
-| intel-core | `src/modules/pmcs/intel-core/mchw_intel_core.ko` | Most Intel multi-core processors are compatible with this module, including recent processors based on the Intel "Broadwell" microarchitecture. | 
-| amd | `src/modules/pmcs/amd/mchw_amd.ko` | This module has been successfully tested on AMD opteron processors. Nevertheless, it should be compatible with all AMD multicore processors. | 
-| arm | `src/modules/pmcs/arm/mchw_arm.ko` | This module has been successfully tested on ARM systems featuring 32-bit big.LITTLE processors, which combine ARM Cortex A7 cores with and ARM Cortex A15 cores. Specifically, tests were performed on the ARM Coretile Express Development Board (TC2). | 
-| odroid-xu | `src/modules/pmcs/odroid-xu/mchw_odroid_xu.ko` | Specific module for Odroid XU3 and XU4 boards. More information on these boards can be found at [www.hardkernel.com](http://www.hardkernel.com) | 
-| arm64 | `src/modules/pmcs/arm64/mchw_arm64.ko` | This module has been successfully tested on ARM systems featuring 64-bit big.LITTLE processors, which combine ARM Cortex A57 cores with and ARM Cortex A53 cores. Specifically, tests were performed on the ARM Juno Development Board. | 
-| xeon-phi | `src/modules/pmcs/xeon-phi/mchw_phi.ko` | Intel Xeon Phi Coprocessor | 
-| core2 | `src/modules/pmcs/phi/mchw_core2.ko` | This module has been specifically designed for the Intel QuickIA prototype system. The Intel QuickIA is a dual-socket asymmetric multicore system that features a quad-core Intel Xeon E5450 processor and a dual-core Intel Atom N330. The module also works with Intel Atom processors as well as "old" Intel multicore processors, such as the Intel Core 2 Duo. Nevertheless, given the numerous existing hacks for the QuickIA in this module, users are advised to use the more general "intel-core" flavor.  | 
+| Name | Path of the .ko file | Supported processors |
+| -----| ---------------------| ---------------------|
+| intel-core | `src/modules/pmcs/intel-core/mchw_intel_core.ko` | Most Intel multi-core processors are compatible with this module, including recent processors based on the Intel "Broadwell" microarchitecture. |
+| amd | `src/modules/pmcs/amd/mchw_amd.ko` | This module has been successfully tested on AMD opteron processors. Nevertheless, it should be compatible with all AMD multicore processors. |
+| arm | `src/modules/pmcs/arm/mchw_arm.ko` | This module has been successfully tested on ARM systems featuring 32-bit big.LITTLE processors, which combine ARM Cortex A7 cores with and ARM Cortex A15 cores. Specifically, tests were performed on the ARM Coretile Express Development Board (TC2). |
+| odroid-xu | `src/modules/pmcs/odroid-xu/mchw_odroid_xu.ko` | Specific module for Odroid XU3 and XU4 boards. More information on these boards can be found at [www.hardkernel.com](http://www.hardkernel.com) |
+| arm64 | `src/modules/pmcs/arm64/mchw_arm64.ko` | This module has been successfully tested on ARM systems featuring 64-bit big.LITTLE processors, which combine ARM Cortex A57 cores with and ARM Cortex A53 cores. Specifically, tests were performed on the ARM Juno Development Board. |
+| xeon-phi | `src/modules/pmcs/xeon-phi/mchw_phi.ko` | Intel Xeon Phi Coprocessor |
+| core2 | `src/modules/pmcs/phi/mchw_core2.ko` | This module has been specifically designed for the Intel QuickIA prototype system. The Intel QuickIA is a dual-socket asymmetric multicore system that features a quad-core Intel Xeon E5450 processor and a dual-core Intel Atom N330. The module also works with Intel Atom processors as well as "old" Intel multicore processors, such as the Intel Core 2 Duo. Nevertheless, given the numerous existing hacks for the QuickIA in this module, users are advised to use the more general "intel-core" flavor.  |
+| perf | `src/modules/pmcs/perf/mchw_perf.ko` | Experimental backend that uses Perf Events's kernel API to access performance monitoring counters |
 
 
 Once the most suitable kernel model for the system has been identified, the module can be loaded in the running PMCTrack-enabled kernel as follows:
@@ -408,7 +413,7 @@ The `pmc-events` command can be used to list the virtual counters exported by th
 
 	$ pmc-events -V
 	[Virtual counters]
-
+	
 	energy_pkg
 	energy_dram
 

bin/pmc-metric

@@ -226,7 +226,8 @@ def parse_pmctrack_output(infile,outfile,metrics,bypass,verbose,acum_mode,csv_mo
 				try:
 					metric_val=eval(expr,counts)
 					if len(expr.co_names)==1: ## Only one variable
-						metric_val=metric_val/nr_samples
+						if not name.startswith("energy"):
+							metric_val=metric_val/nr_samples
 				except ZeroDivisionError:
 					metric_val=0.0
 				counts[name]=metric_val ## Evaluate recursive metrics

bin/pmctrack-manager

@@ -107,8 +107,8 @@ function update_compatible_modules(){
 	local vendor_str=$1
 	local bitw=$2
 	if [ "$vendor_str" == "Intel" ]; then
-		compatible_modules=(intel-core core2)
-		module_names=(mchw_intel_core mchw_core2)
+		compatible_modules=(intel-core core2 perf)
+		module_names=(mchw_intel_core mchw_core2 mchw_perf)
 	elif [ "$vendor_str" == "AMD" ]; then
 		compatible_modules=(amd)
 		module_names=(mchw_amd.ko)

etc/events/armv8.cortex_a73.csv

@@ -0,0 +1,35 @@
+event_name,subevent_name,event_code,description,flags
+cycles_allmodes,-,-,-,type=fixed;pmc=0
+cycles,-,0x11,-,-
+instr,-,0x8,-,-
+instr_arch_executed,-,0x8,-,-
+l2_cache_accesses,-,0x16,-,-
+l2_cache_refills,-,0x17,-,-
+l2_cache_writebacks,-,0x18,-,-
+l1_data_refills,-,0x03,-,-
+l1_data_accesses,-,0x04,-,-
+llc_accesses,-,0x16,-,-
+llc_references,-,0x16,-,-
+llc_misses,-,0x17,-,-
+data_read_retired,-,0x66,-,-
+data_write_retired,-,0x67,-,-
+l1_instr_refills,-,0x1,-,-
+itlb_misses,-,0x2,-,-
+dtlb_misses,-,0x5,-,-
+mispred_branches,-,0x10,-,-
+inmediate_branches_retired,-,0x0d,-,-
+potentially_predicted_branches,-,0x12,-,-
+bus_accesses,-,0x19,-,-
+data_mem_access,-,0x13,-,-
+instr_cache_access,-,0x14,-,-
+data_cache_eviction,-,0x15,-,-
+instr_spec_executed,-,0x1b,-,-
+bus_read_accesses,-,0x60,-,-
+bus_write_accesses,-,0x61,-,-
+ld_spec,-,0x70,Instruction speculatively executed. load,-
+st_spec,-,0x71,Instruction speculatively executed. store,-
+ldst_spec,-,0x72,Instruction speculatively executed. load/store,-
+dp_spec,-,0x73,Instruction speculatively executed. integer/data processing,-
+ase_spec,-,0x74,Instruction speculatively executed. advanced SIMD extension,-
+vfp_spec,-,0x75,Instruction speculatively executed. floating point extension,-
+br_indirect_spec,-,0x7a,Branch speculatively executed. indirect branch,-

etc/version

@@ -1 +1 @@
-1.0
+1.5

src/cmdtools/pmctrack/pmctrack.c

@@ -89,6 +89,7 @@ struct options {
 	int timeout_secs;
 	int msecs;
 	int max_samples;
+	int max_ebs_samples;
 	int kernel_buffer_size;
 	unsigned long cpumask;
 	int optind;
@@ -303,9 +304,8 @@ static void print_process_statistics(FILE* fout,
 	if (opt->flags & CMD_FLAG_SHOW_TIME_SECS) {
 		fprintf (fout, "real\t%ld.%02ld\n",
 		         end->tv_sec,
-		         end->tv_usec / 10000);		
-	} 
-	else {
+		         end->tv_usec / 10000);
+	} else {
 		if (end->tv_sec >= 3600)
 			fprintf (fout, "real\t%ld:%02ld:%02ld\n", /* Format: H:M:S  */
 			         end->tv_sec / 3600,
@@ -316,7 +316,7 @@ static void print_process_statistics(FILE* fout,
 			         end->tv_sec / 60,
 			         end->tv_sec % 60,
 			         end->tv_usec / 10000);
-	}	
+	}
 
 	/* User time.  */
 	fprintf (fout, "user\t%ld.%02ld\n",
@@ -498,7 +498,7 @@ static void monitoring_counters(struct options* opts,int optind,char** argv)
 	if(pid == -1) {
 		err(1,"Error forking process.");
 	} else if(pid == 0) {
-		// CHILD CODE:		
+		// CHILD CODE:
 		if (restore_signal_handlers())
 			exit(1);
 
@@ -509,6 +509,9 @@ static void monitoring_counters(struct options* opts,int optind,char** argv)
 		if (opts->kernel_buffer_size!=-1 && pmct_set_kernel_buffer_size(opts->kernel_buffer_size))
 			pmctrack_exit(1);
 
+		if (opts->max_ebs_samples>0)
+			pmct_config_max_ebs_samples(opts->max_ebs_samples);
+
 		/* Configure counters if there is something to configure */
 		if (opts->strcfg[0] && pmct_config_counters((const char**)opts->strcfg,0))
 			pmctrack_exit(1);
@@ -644,7 +647,7 @@ void monitoring_counters_syswide(struct options* opts,int optind,char** argv)
 	} else if(pid == 0) {
 		// CHILD CODE:
 		if (restore_signal_handlers())
-			exit(1);		
+			exit(1);
 
 		/* Bind first */
 		bind_process_cpumask(getpid(),opts->cpumask);
@@ -1122,6 +1125,7 @@ void init_options( struct options* opts)
 
 	opts->cpumask = NO_CPU_BINDING;
 	opts->max_samples = -1;
+	opts->max_ebs_samples= -1;
 	opts->flags=0;
 	opts->target_pid=-1;
 	opts->kernel_buffer_size = -1;
@@ -1238,7 +1242,7 @@ static void usage(const char* program_name,int status)
 		printf ("\n\t-n\t<max-samples>\n\t\tRun command until a given number of samples are collected");
 		printf ("\n\t-N\t<secs>\n\t\tRun command for secs seconds only");
 		printf ("\n\t-e\n\t\tEnable extended output");
-		printf ("\n\t-E\n\t\tShow additional column with elapsed time between samples");	
+		printf ("\n\t-E\n\t\tShow additional column with elapsed time between samples");
 		printf ("\n\t-A\n\t\tEnable aggregate count mode");
 		printf ("\n\t-k\t<kernel_buffer_size>\n\t\tSpecify the size of the kernel buffer used for the PMC samples");
 		printf ("\n\t-b\t<cpu or mask>\n\t\tbind monitor program to the specified cpu o cpumask.");
@@ -1247,8 +1251,9 @@ static void usage(const char* program_name,int status)
 		printf ("\n\t-P\t<pmu>\n\t\tSpecify the PMU id to use for the event configuration");
 		printf ("\n\t-L\n\t\tLegacy-mode: do not show counter-to-event mapping");
 		printf ("\n\t-t\n\t\tShow real, user and sys time of child process");
-		printf ("\n\t-st\n\t\tDisplay real time in seconds (when -t option is enabled)");		
+		printf ("\n\t-st\n\t\tDisplay real time in seconds (when -t option is enabled)");
 		printf ("\n\t-p\t<pid>\n\t\tAttach to existing process with given pid");
+		printf ("\n\t-K\t<nsamples>\n\t\tSetup maximum number of samples (in EBS mode) that the application will actually execute");
 		printf ("\nPROG + ARGS:\n\t\tCommand line for the program to be monitored.\n");
 		break;
 	case -2:
@@ -1278,7 +1283,7 @@ int main(int argc, char *argv[])
 		usage(argv[0],0);
 
 	/* Process command-line options ... */
-	while ((optc = getopt(argc, argv, "+hc:T:o:b:n:V:B:eAk:SrP:LtN:p:sE")) != (char)-1) {
+	while ((optc = getopt(argc, argv, "+hc:T:o:b:n:V:B:eAk:SrP:LtN:p:sEK:")) != (char)-1) {
 		switch (optc) {
 		case 'o':
 			if((fo = fopen(optarg, "w")) == NULL)
@@ -1343,7 +1348,10 @@ int main(int argc, char *argv[])
 			break;
 		case 'E':
 			opts.flags|=CMD_FLAG_SHOW_ELAPSED_TIME;
-			break;			
+			break;
+		case 'K':
+			opts.max_ebs_samples=atoi(optarg);
+			break;
 		default:
 			fprintf(stderr, "Wrong option: %c\n", optc);
 			exit(1);