/* module_analysis */
int module_analysis(void) {
OUTPUT(("%s", _YELLOW("Analysing measurements")));
if (PERFEXPERT_SUCCESS != output_analysis()) {
OUTPUT(("%s", _ERROR("printing analysis report")));
return PERFEXPERT_ERROR;
}
return PERFEXPERT_SUCCESS;
}
/* module_compile */
int module_compile(void) {
OUTPUT(("%s [%s]", _YELLOW("Compiling"), globals.program));
if (PERFEXPERT_SUCCESS != run_clean()) {
return PERFEXPERT_ERROR;
}
if (PERFEXPERT_SUCCESS != run_make()) {
return PERFEXPERT_ERROR;
}
return PERFEXPERT_SUCCESS;
}
/* module_measurements */
int module_measurements(void) {
OUTPUT(("%s", _YELLOW("Setting performance events")));
my_module_globals.hpctoolkit = (perfexpert_module_hpctoolkit_t *)
perfexpert_module_available("hpctoolkit");
if (PERFEXPERT_SUCCESS != my_module_globals.hpctoolkit->set_event(
"PAPI_TOT_CYC")) {
OUTPUT(("%s", _ERROR("unable to generate add PAPI_TOT_CYC")));
return PERFEXPERT_ERROR;
}
return PERFEXPERT_SUCCESS;
}
/* parse_cli_params */
int parse_module_args(int argc, char *argv[]) {
int i = 0;
/* If some environment variable is defined, use it! */
if (PERFEXPERT_SUCCESS != parse_env_vars()) {
OUTPUT(("%s", _ERROR("parsing environment variables")));
return PERFEXPERT_ERROR;
}
/* Parse arguments */
argp_parse(&argp, argc, argv, 0, 0, NULL);
/* Sanity check: threshold is mandatory, check limits */
if (((0 >= my_module_globals.threshold) ||
(1 < my_module_globals.threshold))) {
OUTPUT(("%s", _ERROR("invalid threshold")));
return PERFEXPERT_ERROR;
}
OUTPUT_VERBOSE((7, "%s", _BLUE("Summary of options")));
OUTPUT_VERBOSE((7, " Threshold: %f", my_module_globals.threshold));
OUTPUT_VERBOSE((7, " Verbose level: %d", my_module_globals.verbose));
/* Not using OUTPUT_VERBOSE because I want only one line */
if (8 <= my_module_globals.verbose) {
i = 0;
printf("%s %s", PROGRAM_PREFIX, _YELLOW("options:"));
for (i = 0; i < argc; i++) {
printf(" [%s]", argv[i]);
}
printf("\n");
fflush(stdout);
}
return PERFEXPERT_SUCCESS;
}
/* lcpi_compute */
int lcpi_compute(profile_t *profile) {
lcpi_t *hotspot_lcpi = NULL, *lcpi = NULL;
procedure_t *hotspot = NULL;
double *values = NULL;
char **names = NULL;
int count = 0;
OUTPUT_VERBOSE((4, " %s", _CYAN("Calculating LCPI metrics")));
/* For each hotspot in this profile... */
hotspot = (procedure_t *)perfexpert_list_get_first(&(profile->hotspots));
while ((perfexpert_list_item_t *)hotspot != &(profile->hotspots.sentinel)) {
OUTPUT_VERBOSE((8, " [%d] %s", hotspot->id, _YELLOW(hotspot->name)));
/* For each LCPI definition */
for (lcpi = globals.lcpi_by_name; lcpi != NULL;
lcpi = lcpi->hh_str.next) {
/* Copy LCPI definitions into this hotspot LCPI hash */
PERFEXPERT_ALLOC(lcpi_t, hotspot_lcpi, sizeof(lcpi_t));
strcpy(hotspot_lcpi->name_md5, lcpi->name_md5);
hotspot_lcpi->name = lcpi->name;
hotspot_lcpi->value = lcpi->value;
hotspot_lcpi->expression = lcpi->expression;
/* Get the list of variables and their values */
evaluator_get_variables(hotspot_lcpi->expression, &names, &count);
if (0 < count) {
int i = 0;
PERFEXPERT_ALLOC(double, values, (sizeof(double*) * count));
for (i = 0; i < count; i++) {
metric_t *metric = NULL;
char key_md5[33];
strcpy(key_md5, perfexpert_md5_string(names[i]));
perfexpert_hash_find_str(hotspot->metrics_by_name,
(char *)&key_md5, metric);
if (NULL == metric) {
perfexpert_hash_find_str(globals.machine_by_name,
(char *)&key_md5, metric);
if (NULL == metric) {
values[i] = 0.0;
} else {
values[i] = metric->value;
}
} else {
values[i] = metric->value;
}
}
/* Evaluate the LCPI expression */
hotspot_lcpi->value = evaluator_evaluate(
hotspot_lcpi->expression, count, names, values);
PERFEXPERT_DEALLOC(values);
}
/* Add the LCPI to the hotspot's list of LCPIs */
perfexpert_hash_add_str(hotspot->lcpi_by_name, name_md5,
hotspot_lcpi);
OUTPUT_VERBOSE((10, " %s=[%g]", hotspot_lcpi->name,
hotspot_lcpi->value));
}
/* Move on */
hotspot = (procedure_t *)perfexpert_list_get_next(hotspot);
}
/* parse_cli_params */
int parse_cli_params(int argc, char *argv[]) {
int parameter = 0; /* Temporary variable to hold parameter */
int option_index = 0; /* getopt_long() stores the option index here */
/* If some environment variable is defined, use it! */
if (PERFEXPERT_SUCCESS != parse_env_vars()) {
OUTPUT(("%s", _ERROR("Error: parsing environment variables")));
return PERFEXPERT_ERROR;
}
while (1) {
/* get parameter */
parameter = getopt_long(argc, argv, "achi:l:m:M:o:O:t:T:v:w:",
long_options, &option_index);
/* Detect the end of the options */
if (-1 == parameter) {
break;
}
switch (parameter) {
/* Aggregated metrics */
case 'a':
globals.aggregate = PERFEXPERT_TRUE;
OUTPUT_VERBOSE((1, "option 'a' set"));
break;
/* Activate colorful mode */
case 'c':
globals.colorful = PERFEXPERT_TRUE;
OUTPUT_VERBOSE((1, "option 'c' set"));
break;
/* Show help */
case 'h':
OUTPUT_VERBOSE((1, "option 'h' set"));
show_help();
return PERFEXPERT_ERROR;
/* Input file */
case 'i':
globals.inputfile = optarg;
OUTPUT_VERBOSE((1, "option 'i' set [%s]", globals.inputfile));
break;
/* LCPI file */
case 'l':
globals.lcpifile = optarg;
OUTPUT_VERBOSE((1, "option 'l' set [%s]", globals.lcpifile));
break;
/* Measurement tool */
case 'm':
globals.tool = optarg;
OUTPUT_VERBOSE((1, "option 'm' set [%s]", globals.tool));
break;
/* Machine file */
case 'M':
globals.machinefile = optarg;
OUTPUT_VERBOSE((1, "option 'M' set [%s]",
globals.machinefile));
break;
/* Output file */
case 'o':
globals.outputfile = optarg;
OUTPUT_VERBOSE((1, "option 'o' set [%s]", globals.outputfile));
break;
/* Sorting order */
case 'O':
globals.order = optarg;
OUTPUT_VERBOSE((1, "option 'O' set [%s]", globals.order));
break;
/* Threshold */
case 't':
globals.threshold = atof(optarg);
OUTPUT_VERBOSE((1, "option 't' set [%f]", globals.threshold));
break;
/* Thread ID */
case 'T':
globals.thread = atoi(optarg);
OUTPUT_VERBOSE((1, "option 'T' set [%d]", globals.thread));
break;
/* Verbose */
case 'v':
globals.verbose = atoi(optarg);
OUTPUT_VERBOSE((1, "option 'v' set [%d]", globals.verbose));
break;
/* Workdir */
case 'w':
globals.workdir = optarg;
OUTPUT_VERBOSE((1, "option 'w' set [%s]", globals.workdir));
break;
/* Unknown option */
case '?':
default:
show_help();
return PERFEXPERT_ERROR;
}
}
/* Print summary */
OUTPUT_VERBOSE((7, "%s", _BLUE("Summary of options")));
OUTPUT_VERBOSE((7, " Aggregate? %s", globals.aggregate ?
"yes" : "no"));
OUTPUT_VERBOSE((7, " Colorful verbose? %s", globals.colorful ?
"yes" : "no"));
OUTPUT_VERBOSE((7, " Input file: %s", globals.inputfile));
OUTPUT_VERBOSE((7, " LCPI file: %s", globals.lcpifile));
OUTPUT_VERBOSE((7, " Measurement tool: %s", globals.tool));
OUTPUT_VERBOSE((7, " Machine file: %s", globals.machinefile));
OUTPUT_VERBOSE((7, " Output file: %s", globals.outputfile));
OUTPUT_VERBOSE((7, " Sorting order: %s", globals.order));
OUTPUT_VERBOSE((7, " Threshold: %f", globals.threshold));
OUTPUT_VERBOSE((7, " Thread ID: %d", globals.thread));
OUTPUT_VERBOSE((7, " Verbose level: %d", globals.verbose));
OUTPUT_VERBOSE((7, " Output metrcis: %s", globals.outputmetrics));
OUTPUT_VERBOSE((7, " Workdir: %s", globals.workdir));
/* Not using OUTPUT_VERBOSE because I want only one line */
if (8 <= globals.verbose) {
int i;
printf("%s %s", PROGRAM_PREFIX, _YELLOW("command line:"));
for (i = 0; i < argc; i++) {
printf(" %s", argv[i]);
}
printf("\n");
}
/* Sanity check: verbose level should be between 1-10 */
if ((0 > globals.verbose) || (10 < globals.verbose)) {
OUTPUT(("%s", _ERROR("Error: invalid verbose level")));
show_help();
return PERFEXPERT_ERROR;
}
/* Sanity check: threshold is mandatory */
if ((0 >= globals.threshold) || (1 < globals.threshold)) {
OUTPUT(("%s", _ERROR("Error: undefined or invalid threshold")));
show_help();
return PERFEXPERT_ERROR;
}
/* Sanity check: inputfile is mandatory */
if (NULL == globals.inputfile) {
OUTPUT(("%s", _ERROR("Error: undefined input file")));
show_help();
return PERFEXPERT_ERROR;
}
/* Sanity check: thread ID must be <= 0, but -1 means NO_THREADS */
if (-1 > globals.thread) {
OUTPUT(("%s", _ERROR("Error: invalid thread ID")));
show_help();
return PERFEXPERT_ERROR;
}
return PERFEXPERT_SUCCESS;
}
/* parse_cli_params */
int parse_module_args(int argc, char *argv[]) {
int i = 0;
/* Set default values */
arg_options = (arg_options_t) {
.prefix = NULL,
.before = NULL,
.after = NULL,
.mainsrc = NULL,
};
if (((0 >= my_module_globals.threshold) ||
(1 < my_module_globals.threshold))) {
OUTPUT(("%s", _ERROR("invalid threshold")));
return PERFEXPERT_ERROR;
}
/* If some environment variable is defined, use it! */
if (PERFEXPERT_SUCCESS != parse_env_vars()) {
OUTPUT(("%s", _ERROR("parsing environment variables")));
return PERFEXPERT_ERROR;
}
/* Parse arguments */
argp_parse(&argp, argc, argv, 0, 0, NULL);
/* Expand AFTERs, BEFOREs, and PREFIXs arguments */
// TODO: right now we only use the before, after and
// prefix from globals. We should also consider values coming
// throught args_options.* (that have been set in the previous
// instruction
if (NULL != arg_options.after) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.after), my_module_globals.after, ' ');
}
if (NULL != arg_options.before) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.before), my_module_globals.before, ' ');
}
if (NULL != arg_options.prefix) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.prefix), my_module_globals.prefix, ' ');
}
if (NULL != arg_options.mainsrc) {
my_module_globals.mainsrc = arg_options.mainsrc;
}
OUTPUT_VERBOSE((7, "%s", _BLUE("Summary of options")));
if (7 <= globals.verbose) {
printf("%s Threshold: %f", PROGRAM_PREFIX, my_module_globals.threshold);
printf("\n%s Prefix: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.prefix[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.prefix[i]) {
printf(" [%s]", (char *)my_module_globals.prefix[i]);
i++;
}
}
printf("\n%s Before each run: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.before[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.before[i]) {
printf(" [%s]", (char *)my_module_globals.before[i]);
i++;
}
}
printf("\n%s After each run: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.after[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.after[i]) {
printf(" [%s]", (char *)my_module_globals.after[i]);
i++;
}
}
printf("\n%s Main src file: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.mainsrc) {
printf(" (null)");
} else {
printf(" [%s]", (char *)my_module_globals.mainsrc);
}
printf("\n");
fflush(stdout);
}
/* Not using OUTPUT_VERBOSE because I want only one line */
if (8 <= globals.verbose) {
i = 0;
printf("%s %s", PROGRAM_PREFIX, _YELLOW("options:"));
for (i = 0; i < argc; i++) {
printf(" [%s]", argv[i]);
}
printf("\n");
fflush(stdout);
}
return PERFEXPERT_SUCCESS;
}
// This function takes the output of running VTune and creates a report CSV file
// The results of running VTune are in results_folder
// The output of this function is in parse_file
// The command it executes is like:
// amplxe-cl -R hw-events -r results_folder -group-by=function -format=csv -report-output=parse_file -csv-delimiter=','
// amplxe-cl -report hw-events -group-by=function -group-by=module -group-by=source-file -group-by=source-line -group-by=thread -r results_folder -format=csv -report-output=parse-file -csv-delimiter=','
int create_report(char* results_folder, const char* parse_file) {
struct timespec time_start, time_end, time_diff;
char *argv[MAX_ARGUMENTS_COUNT];
int argc = 0, i, rc;
char buffer[MAX_BUFFER_SIZE];
test_t test;
argv[argc] = VTUNE_CL_BIN;
argc++;
argv[argc] = VTUNE_ACT_REPORT;
argc++;
argv[argc] = VTUNE_RPT_HW;
argc++;
argv[argc] = "-group-by=function";
argc++;
argv[argc] = "-group-by=module";
argc++;
argv[argc] = "-group-by=source-file";
argc++;
argv[argc] = "-group-by=source-line";
argc++;
if (globals.output_mode == HYBRID_OUTPUT) {
argv[argc] = "-group-by=thread";
argc++;
}
argv[argc] = "-r";
argc++;
argv[argc] = my_module_globals.res_folder;
argc++;
argv[argc] = "-format=csv";
argc++;
strcpy(buffer, "-report-output=");
strcat(buffer, parse_file);
argv[argc] = buffer;
argc++;
argv[argc] = "-csv-delimiter=','";
argc++;
argv[argc] = NULL;
/* Not using OUTPUT_VERBOSEbecause I want only one line */
if (8 <= globals.verbose) {
printf("%s %s", PROGRAM_PREFIX, _YELLOW("command line:"));
for (i = 0; i < argc; i++) {
printf(" %s", argv[i]);
}
printf("\n");
}
/* The super-ninja test sctructure */
PERFEXPERT_ALLOC(char, test.output, (strlen(globals.moduledir) + 14));
sprintf(test.output, "%s/vtune.output", globals.moduledir);
test.input = my_module_globals.inputfile;
test.info = globals.program;
clock_gettime(CLOCK_MONOTONIC, &time_start);
/* fork_and_wait_and_pray */
rc = perfexpert_fork_and_wait(&test, (char **)argv);
/* Evaluate results if required to */
if (PERFEXPERT_FALSE == my_module_globals.ignore_return_code) {
switch (rc) {
case PERFEXPERT_FAILURE:
case PERFEXPERT_ERROR:
OUTPUT(("%s (return code: %d) Usually, this means that an error"
" happened during the program execution. To see the program"
"'s output, check the content of this file: [%s]. If you "
"want to PerfExpert ignore the return code next time you "
"run this program, set the 'return-code' option for the "
"Vtune module. See 'perfexpert -H vtune' for details.",
_ERROR("the target program returned non-zero"), rc,
test.output));
return PERFEXPERT_ERROR;
case PERFEXPERT_SUCCESS:
OUTPUT_VERBOSE((7, "[ %s ]", _BOLDGREEN("OK")));
break;
default:
break;
}
}
/* Calculate and display runtime */
clock_gettime(CLOCK_MONOTONIC, &time_end);
perfexpert_time_diff(&time_diff, &time_start, &time_end);
OUTPUT((" [1/1] %lld.%.9ld seconds (includes measurement overhead)",
(long long)time_diff.tv_sec, time_diff.tv_nsec));
PERFEXPERT_DEALLOC(test.output);
/* Run the AFTER program */
if (NULL != my_module_globals.after[0]) {
PERFEXPERT_ALLOC(char, test.output, (strlen(globals.moduledir) + 20));
sprintf(test.output, "%s/after.output", globals.moduledir);
test.input = NULL;
test.info = my_module_globals.after[0];
if (0 != perfexpert_fork_and_wait(&test,
(char **)my_module_globals.after)) {
OUTPUT(("%s", _RED("'after' command return non-zero")));
}
PERFEXPERT_DEALLOC(test.output);
}
/* parse_cli_params */
int parse_module_args(int argc, char *argv[]) {
int i = 0;
/* Set default values */
arg_options = (arg_options_t) {
.prefix = NULL,
.before = NULL,
.after = NULL,
.knc_prefix = NULL,
.knc_before = NULL,
.knc_after = NULL
};
/* If some environment variable is defined, use it! */
if (PERFEXPERT_SUCCESS != parse_env_vars()) {
OUTPUT(("%s", _ERROR("parsing environment variables")));
return PERFEXPERT_ERROR;
}
/* Parse arguments */
argp_parse(&argp, argc, argv, 0, 0, NULL);
/* Expand AFTERs, BEFOREs, and PREFIXs arguments */
if (NULL != arg_options.after) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.after), my_module_globals.after, ' ');
}
if (NULL != arg_options.before) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.before), my_module_globals.before, ' ');
}
if (NULL != arg_options.prefix) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.prefix), my_module_globals.prefix, ' ');
}
if (NULL != arg_options.knc_after) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.knc_after), my_module_globals.knc_after, ' ');
}
if (NULL != arg_options.knc_before) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.knc_before), my_module_globals.knc_before, ' ');
}
if (NULL != arg_options.knc_prefix) {
perfexpert_string_split(perfexpert_string_remove_spaces(
arg_options.knc_prefix), my_module_globals.knc_prefix, ' ');
}
/* Sanity check: MIC options without MIC */
if ((NULL != my_module_globals.knc_prefix[0]) &&
(NULL == my_module_globals.knc)) {
OUTPUT(("%s option -P selected but no MIC card was specified, ignoring",
_BOLDRED("WARNING:")));
}
/* Sanity check: MIC options without MIC */
if ((NULL != my_module_globals.knc_before[0]) &&
(NULL == my_module_globals.knc)) {
OUTPUT(("%s option -B selected but no MIC card was specified, ignoring",
_BOLDRED("WARNING:")));
}
/* Sanity check: MIC options without MIC */
if ((NULL != my_module_globals.knc_after[0]) &&
(NULL == my_module_globals.knc)) {
OUTPUT(("%s option -A selected but no MIC card was specified, ignoring",
_BOLDRED("WARNING:")));
}
OUTPUT_VERBOSE((7, "%s", _BLUE("Summary of options")));
OUTPUT_VERBOSE((7, " Program input file: %s",
my_module_globals.inputfile));
if (7 <= globals.verbose) {
printf("%s Prefix: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.prefix[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.prefix[i]) {
printf(" [%s]", (char *)my_module_globals.prefix[i]);
i++;
}
}
printf("\n%s Before each run: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.before[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.before[i]) {
printf(" [%s]", (char *)my_module_globals.before[i]);
i++;
}
}
printf("\n%s After each run: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.after[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.after[i]) {
printf(" [%s]", (char *)my_module_globals.after[i]);
i++;
}
}
printf("\n");
fflush(stdout);
}
#if HAVE_KNC_SUPPORT
OUTPUT_VERBOSE((7, " MIC card: %s", my_module_globals.knc));
if (7 <= globals.verbose) {
printf("%s MIC prefix: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.knc_prefix[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.knc_prefix[i]) {
printf(" [%s]", (char *)my_module_globals.knc_prefix[i]);
i++;
}
}
printf("\n%s MIC before each run:", PROGRAM_PREFIX);
if (NULL == my_module_globals.knc_before[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.knc_before[i]) {
printf(" [%s]", (char *)my_module_globals.knc_before[i]);
i++;
}
}
printf("\n%s MIC after each run: ", PROGRAM_PREFIX);
if (NULL == my_module_globals.knc_after[0]) {
printf(" (null)");
} else {
i = 0;
while (NULL != my_module_globals.knc_after[i]) {
printf(" [%s]", (char *)my_module_globals.knc_after[i]);
i++;
}
}
printf("\n");
fflush(stdout);
}
#endif
/* Not using OUTPUT_VERBOSE because I want only one line */
if (8 <= globals.verbose) {
i = 0;
printf("%s %s", PROGRAM_PREFIX, _YELLOW("options:"));
for (i = 0; i < argc; i++) {
printf(" [%s]", argv[i]);
}
printf("\n");
fflush(stdout);
}
return PERFEXPERT_SUCCESS;
}
/* Set default values */
if (NULL == globals.machinefile) {
PERFEXPERT_ALLOC(char, globals.machinefile,
(strlen(PERFEXPERT_ETCDIR) + strlen(MACHINE_FILE) + 2));
sprintf(globals.machinefile, "%s/%s", PERFEXPERT_ETCDIR, MACHINE_FILE);
}
if (NULL == globals.lcpifile) {
PERFEXPERT_ALLOC(char, globals.lcpifile,
(strlen(PERFEXPERT_ETCDIR) + strlen(LCPI_FILE) + 2));
sprintf(globals.lcpifile, "%s/%s", PERFEXPERT_ETCDIR, LCPI_FILE);
}
/* Print to a file or STDOUT? */
if (NULL != globals.outputfile) {
OUTPUT_VERBOSE((7, " %s (%s)",
_YELLOW("printing recommendations to file"), globals.outputfile));
if (NULL == (globals.outputfile_FP = fopen(globals.outputfile, "w+"))) {
OUTPUT(("%s (%s)", _ERROR("Error: unable to open output file"),
globals.outputfile));
return PERFEXPERT_ERROR;
}
} else {
OUTPUT_VERBOSE((7, " printing recommendations to STDOUT"));
}
/* Parse LCPI metrics */
if (PERFEXPERT_SUCCESS != lcpi_parse_file(globals.lcpifile)) {
OUTPUT(("%s (%s)", _ERROR("Error: LCPI file is not valid"),
globals.lcpifile));
return PERFEXPERT_ERROR;
}
