From 542640616034266053f7bb92a34781a8bb73ec1f Mon Sep 17 00:00:00 2001 From: Solomon Bekele Date: Thu, 31 Oct 2024 23:51:39 +0000 Subject: [PATCH] separate sampling --- ze/Makefile.am | 5 +- ze/gen_ze.rb | 2 +- ze/tracer_ze_helpers.include.c | 485 +-------------------------------- 3 files changed, 10 insertions(+), 482 deletions(-) diff --git a/ze/Makefile.am b/ze/Makefile.am index 60f88f30..ee6174bf 100644 --- a/ze/Makefile.am +++ b/ze/Makefile.am @@ -148,12 +148,13 @@ BUILT_SOURCES = \ $(ZE_PROBES_INCL) \ $(ZE_STATIC_PROBES_INCL) -tracer_ze.c: $(srcdir)/gen_ze.rb $(srcdir)/tracer_ze_helpers.include.c $(srcdir)/ze.h.include $(ZE_MODEL) $(ZE_PROBES_INCL) $(ZE_STATIC_PROBES_INCL) +tracer_ze.c: $(srcdir)/gen_ze.rb $(srcdir)/tracer_ze_helpers.include.c $(srcdir)/sampling_daemon.c $(srcdir)/sampling_daemon.h $(srcdir)/ze.h.include $(ZE_MODEL) $(ZE_PROBES_INCL) $(ZE_STATIC_PROBES_INCL) SRC_DIR=$(srcdir) $(RUBY) $< > $@ EXTRA_DIST += \ gen_ze.rb \ - tracer_ze_helpers.include.c + tracer_ze_helpers.include.c \ + sampling_daemon.c CLEANFILES += tracer_ze.c diff --git a/ze/gen_ze.rb b/ze/gen_ze.rb index 4c821939..1a421b0f 100644 --- a/ze/gen_ze.rb +++ b/ze/gen_ze.rb @@ -139,7 +139,7 @@ def gen_struct_printer(namespace, types) EOF puts File::read(File.join(SRC_DIR,"tracer_ze_helpers.include.c")) - +puts File::read(File.join(SRC_DIR,"sampling_daemon.c")) common_block = lambda { |c, provider, types| params = c.parameters ? c.parameters.collect(&:name) : [] tp_params = if c.parameters diff --git a/ze/tracer_ze_helpers.include.c b/ze/tracer_ze_helpers.include.c index 8a18f161..bf403c72 100644 --- a/ze/tracer_ze_helpers.include.c +++ b/ze/tracer_ze_helpers.include.c @@ -1,4 +1,5 @@ #include "thapi_sampling.h" +#include "sampling_daemon.h" #ifdef THAPI_DEBUG #define TAHPI_LOG stderr @@ -789,483 +790,13 @@ static inline void _dump_memory_info(ze_command_list_handle_t hCommandList, cons {perror((MSG)); fprintf(stderr,"errno=%d at %d(%s)",errno,__LINE__,__FILE__);\ } while (0) -static int _sampling_freq_initialized = 0; -static int _sampling_fabricPorts_initialized = 0; -static int _sampling_memModules_initialized = 0; -static int _sampling_pwr_initialized = 0; -static int _sampling_engines_initialized = 0; -// Static handles to stay throughout the execution -static zes_driver_handle_t *_sampling_hDrivers = NULL; -static zes_device_handle_t **_sampling_hDevices = NULL; -static zes_freq_handle_t ***_sampling_hFrequencies = NULL; -static zes_pwr_handle_t ***_sampling_hPowers = NULL; -static zes_engine_handle_t ***_sampling_engineHandles = NULL; -static zes_fabric_port_handle_t ***_sampling_hFabricPort = NULL; -static zes_mem_handle_t ***_sampling_hMemModule = NULL; -static uint32_t _sampling_driverCount = 0; -static uint32_t *_sampling_deviceCount = NULL; -static uint32_t **_sampling_freqDomainCounts = NULL; -static uint32_t **_sampling_fabricPortCount = NULL; -static uint32_t **_sampling_memModuleCount = NULL; -static uint32_t **_sampling_powerDomainCounts = NULL; -static uint32_t **_sampling_engineCounts = NULL; - -static void intializeFrequency() { - ze_result_t res; - _sampling_hFrequencies = - (zes_freq_handle_t ***)calloc(_sampling_driverCount, sizeof(zes_freq_handle_t **)); - _sampling_freqDomainCounts = (uint32_t **)calloc(_sampling_driverCount, sizeof(uint32_t *)); - for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) { - _sampling_freqDomainCounts[driverIdx] = - (uint32_t *)calloc(_sampling_deviceCount[driverIdx], sizeof(uint32_t)); - _sampling_hFrequencies[driverIdx] = - (zes_freq_handle_t **)calloc(_sampling_deviceCount[driverIdx], sizeof(zes_freq_handle_t *)); - for (uint32_t deviceIdx = 0; deviceIdx < _sampling_deviceCount[driverIdx]; deviceIdx++) { - // Get frequency domains for each device - res = ZES_DEVICE_ENUM_FREQUENCY_DOMAINS_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_freqDomainCounts[driverIdx][deviceIdx], - NULL); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("1st ZES_DEVICE_ENUM_FREQUENCY_DOMAINS_PTR", res); - _sampling_freqDomainCounts[driverIdx][deviceIdx] = 0; - continue; - } - _sampling_hFrequencies[driverIdx][deviceIdx] = (zes_freq_handle_t *)calloc( - _sampling_freqDomainCounts[driverIdx][deviceIdx], sizeof(zes_freq_handle_t)); - res = ZES_DEVICE_ENUM_FREQUENCY_DOMAINS_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_freqDomainCounts[driverIdx][deviceIdx], - _sampling_hFrequencies[driverIdx][deviceIdx]); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("2nd ZES_DEVICE_ENUM_FREQUENCY_DOMAINS_PTR", res); - _sampling_freqDomainCounts[driverIdx][deviceIdx] = 0; - free(_sampling_hFrequencies[driverIdx][deviceIdx]); - } - for (uint32_t domainIdx = 0; domainIdx < _sampling_freqDomainCounts[driverIdx][deviceIdx]; - domainIdx++) { - zes_freq_properties_t freqProps = {0}; - freqProps.stype = ZES_STRUCTURE_TYPE_FREQ_PROPERTIES; - res = ZES_FREQUENCY_GET_PROPERTIES_PTR( - _sampling_hFrequencies[driverIdx][deviceIdx][domainIdx], &freqProps); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_FREQUENCY_GET_PROPERTIES_PTR", res); - free(_sampling_hFrequencies[driverIdx][deviceIdx][domainIdx]); - } - do_tracepoint(lttng_ust_ze_sampling, freqProperties, - (ze_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_freq_handle_t)_sampling_hFrequencies[driverIdx][deviceIdx][domainIdx], - &freqProps); - } - } - } - _sampling_freq_initialized = 1; -} - -static void intializePower() { - ze_result_t res; - _sampling_hPowers = - (zes_pwr_handle_t ***)calloc(_sampling_driverCount, sizeof(zes_pwr_handle_t **)); - _sampling_powerDomainCounts = (uint32_t **)calloc(_sampling_driverCount, sizeof(uint32_t *)); - for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) { - _sampling_hPowers[driverIdx] = - (zes_pwr_handle_t **)calloc(_sampling_deviceCount[driverIdx], sizeof(zes_pwr_handle_t *)); - _sampling_powerDomainCounts[driverIdx] = - (uint32_t *)calloc(_sampling_deviceCount[driverIdx], sizeof(uint32_t)); - for (uint32_t deviceIdx = 0; deviceIdx < _sampling_deviceCount[driverIdx]; deviceIdx++) { - // Get power domains for each device - res = ZES_DEVICE_ENUM_POWER_DOMAINS_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_powerDomainCounts[driverIdx][deviceIdx], - NULL); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("1st ZES_DEVICE_ENUM_POWER_DOMAINS_PTR", res); - _sampling_powerDomainCounts[driverIdx][deviceIdx] = 0; - continue; - } - _sampling_hPowers[driverIdx][deviceIdx] = (zes_pwr_handle_t *)calloc( - _sampling_powerDomainCounts[driverIdx][deviceIdx], sizeof(zes_pwr_handle_t)); - res = ZES_DEVICE_ENUM_POWER_DOMAINS_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_powerDomainCounts[driverIdx][deviceIdx], - _sampling_hPowers[driverIdx][deviceIdx]); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("2nd ZES_DEVICE_ENUM_POWER_DOMAINS_PTR", res); - _sampling_powerDomainCounts[driverIdx][deviceIdx] = 0; - free(_sampling_hPowers[driverIdx][deviceIdx]); - } - for (uint32_t domainIdx = 0; domainIdx < _sampling_powerDomainCounts[driverIdx][deviceIdx]; - domainIdx++) { - zes_power_properties_t powerProperties = {0}; - powerProperties.stype = ZES_STRUCTURE_TYPE_POWER_PROPERTIES; - res = ZES_POWER_GET_PROPERTIES_PTR(_sampling_hPowers[driverIdx][deviceIdx][domainIdx], - &powerProperties); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_POWER_GET_PROPERTIES_PTR", res); - free(_sampling_hPowers[driverIdx][deviceIdx][domainIdx]); - } - do_tracepoint(lttng_ust_ze_sampling, powerProperties, - (ze_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_pwr_handle_t)_sampling_hPowers[driverIdx][deviceIdx][domainIdx], - &powerProperties); - } - } - } - _sampling_pwr_initialized = 1; -} - -static void intializeEngines() { - ze_result_t res; - _sampling_engineHandles = - (zes_engine_handle_t ***)calloc(_sampling_driverCount, sizeof(zes_engine_handle_t **)); - _sampling_engineCounts = (uint32_t **)calloc(_sampling_driverCount, sizeof(uint32_t *)); - for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) { - _sampling_engineHandles[driverIdx] = (zes_engine_handle_t **)calloc( - _sampling_deviceCount[driverIdx], sizeof(zes_engine_handle_t *)); - _sampling_engineCounts[driverIdx] = - (uint32_t *)calloc(_sampling_deviceCount[driverIdx], sizeof(uint32_t)); - for (uint32_t deviceIdx = 0; deviceIdx < _sampling_deviceCount[driverIdx]; deviceIdx++) { - // Get engine counts for each device - res = ZES_DEVICE_ENUM_ENGINE_GROUPS_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_engineCounts[driverIdx][deviceIdx], NULL); - if (res != ZE_RESULT_SUCCESS || _sampling_engineCounts[driverIdx][deviceIdx] == 0) { - _ZE_ERROR_MSG("1st ZES_DEVICE_ENUM_ENGINE_GROUPS_PTR", res); - _sampling_engineCounts[driverIdx][deviceIdx] = 0; - continue; - } - _sampling_engineHandles[driverIdx][deviceIdx] = (zes_engine_handle_t *)calloc( - _sampling_engineCounts[driverIdx][deviceIdx], sizeof(zes_engine_handle_t)); - res = ZES_DEVICE_ENUM_ENGINE_GROUPS_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_engineCounts[driverIdx][deviceIdx], - _sampling_engineHandles[driverIdx][deviceIdx]); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("2nd ZES_DEVICE_ENUM_ENGINE_GROUPS_PTR", res); - _sampling_engineCounts[driverIdx][deviceIdx] = 0; - free(_sampling_engineHandles[driverIdx][deviceIdx]); - } - for (uint32_t engineIdx = 0; engineIdx < _sampling_engineCounts[driverIdx][deviceIdx]; - ++engineIdx) { - zes_engine_properties_t engineProps = {0}; - engineProps.stype = ZES_STRUCTURE_TYPE_ENGINE_PROPERTIES; - res = ZES_ENGINE_GET_PROPERTIES_PTR( - _sampling_engineHandles[driverIdx][deviceIdx][engineIdx], &engineProps); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_ENGINE_GET_PROPERTIES_PTR", res); - } - do_tracepoint(lttng_ust_ze_sampling, engineProperties, - (ze_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_engine_handle_t)_sampling_engineHandles[driverIdx][deviceIdx][engineIdx], - &engineProps); - } - } - } - _sampling_engines_initialized = 1; -} - -static void intializeFabricPorts() { - ze_result_t res; - _sampling_hFabricPort = (zes_fabric_port_handle_t ***)calloc(_sampling_driverCount, - sizeof(zes_fabric_port_handle_t **)); - _sampling_fabricPortCount = (uint32_t **)calloc(_sampling_driverCount, sizeof(uint32_t *)); - for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) { - _sampling_fabricPortCount[driverIdx] = - (uint32_t *)calloc(_sampling_deviceCount[driverIdx], sizeof(uint32_t)); - _sampling_hFabricPort[driverIdx] = (zes_fabric_port_handle_t **)calloc( - _sampling_deviceCount[driverIdx], sizeof(zes_fabric_port_handle_t *)); - for (uint32_t deviceIdx = 0; deviceIdx < _sampling_deviceCount[driverIdx]; deviceIdx++) { - // Get fabric ports for each device - res = - ZES_DEVICE_ENUM_FABRIC_PORTS_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_fabricPortCount[driverIdx][deviceIdx], NULL); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("1st ZES_DEVICE_ENUM_FABRIC_PORTS_PTR", res); - _sampling_fabricPortCount[driverIdx][deviceIdx] = 0; - continue; - } - _sampling_hFabricPort[driverIdx][deviceIdx] = (zes_fabric_port_handle_t *)calloc( - _sampling_fabricPortCount[driverIdx][deviceIdx], sizeof(zes_fabric_port_handle_t)); - res = ZES_DEVICE_ENUM_FABRIC_PORTS_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_fabricPortCount[driverIdx][deviceIdx], - _sampling_hFabricPort[driverIdx][deviceIdx]); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("2nd ZES_DEVICE_ENUM_FABRIC_PORTS_PTR", res); - _sampling_fabricPortCount[driverIdx][deviceIdx] = 0; - free(_sampling_hFabricPort[driverIdx][deviceIdx]); - } - for (uint32_t fabricPortIdx = 0; - fabricPortIdx < _sampling_fabricPortCount[driverIdx][deviceIdx]; ++fabricPortIdx) { - - zes_fabric_port_properties_t fabricPortProps = {0}; - res = ZES_FABRIC_PORT_GET_PROPERTIES_PTR( - _sampling_hFabricPort[driverIdx][deviceIdx][fabricPortIdx], &fabricPortProps); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_FABRIC_PORT_GET_PROPERTIES_PTR", res); - } - // Dump fabricPortProperties once - do_tracepoint( - lttng_ust_ze_sampling, fabricPortProperties, - (ze_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_fabric_port_handle_t)_sampling_hFabricPort[driverIdx][deviceIdx][fabricPortIdx], - &fabricPortProps); - } - } - } - _sampling_fabricPorts_initialized = 1; -} - -static void intializeMemModules() { - ze_result_t res; - _sampling_hMemModule = - (zes_mem_handle_t ***)calloc(_sampling_driverCount, sizeof(zes_mem_handle_t **)); - _sampling_memModuleCount = (uint32_t **)calloc(_sampling_driverCount, sizeof(uint32_t *)); - for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) { - _sampling_memModuleCount[driverIdx] = - (uint32_t *)calloc(_sampling_deviceCount[driverIdx], sizeof(uint32_t)); - _sampling_hMemModule[driverIdx] = - (zes_mem_handle_t **)calloc(_sampling_deviceCount[driverIdx], sizeof(zes_mem_handle_t *)); - for (uint32_t deviceIdx = 0; deviceIdx < _sampling_deviceCount[driverIdx]; deviceIdx++) { - // Get fabric ports for each device - res = - ZES_DEVICE_ENUM_MEMORY_MODULES_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_memModuleCount[driverIdx][deviceIdx], NULL); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("1st ZES_DEVICE_ENUM_MEMORY_MODULES_PTR", res); - _sampling_memModuleCount[driverIdx][deviceIdx] = 0; - continue; - } - _sampling_hMemModule[driverIdx][deviceIdx] = (zes_mem_handle_t *)calloc( - _sampling_memModuleCount[driverIdx][deviceIdx], sizeof(zes_mem_handle_t)); - res = ZES_DEVICE_ENUM_MEMORY_MODULES_PTR(_sampling_hDevices[driverIdx][deviceIdx], - &_sampling_memModuleCount[driverIdx][deviceIdx], - _sampling_hMemModule[driverIdx][deviceIdx]); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("2nd ZES_DEVICE_ENUM_MEMORY_MODULES_PTR", res); - _sampling_memModuleCount[driverIdx][deviceIdx] = 0; - free(_sampling_hMemModule[driverIdx][deviceIdx]); - } - for (uint32_t memModuleIdx = 0; memModuleIdx < _sampling_memModuleCount[driverIdx][deviceIdx]; - ++memModuleIdx) { - zes_mem_properties_t memProps = {0}; - memProps.stype = ZES_STRUCTURE_TYPE_MEM_PROPERTIES; - res = ZES_MEMORY_GET_PROPERTIES_PTR( - _sampling_hMemModule[driverIdx][deviceIdx][memModuleIdx], &memProps); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_MEMORY_GET_PROPERTIES_PTR", res); - } - // Dump fabricPortProperties once - do_tracepoint(lttng_ust_ze_sampling, memoryProperties, - (ze_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_mem_handle_t)_sampling_hMemModule[driverIdx][deviceIdx][memModuleIdx], - &memProps); - } - } - } - _sampling_memModules_initialized = 1; -} - -static int initializeHandles() { - ze_result_t res; - res = ZES_INIT_PTR(0); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_INIT_PTR", res); - return -1; - } - - // Query driver - _sampling_driverCount = 0; - res = ZES_DRIVER_GET_PTR(&_sampling_driverCount, NULL); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("1st ZES_DRIVER_GET_PTR", res); - return -1; - } - _sampling_hDrivers = - (zes_driver_handle_t *)calloc(_sampling_driverCount, sizeof(zes_driver_handle_t)); - res = ZES_DRIVER_GET_PTR(&_sampling_driverCount, _sampling_hDrivers); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("2nd ZES_DRIVER_GET_PTR", res); - return -1; - } - _sampling_deviceCount = (uint32_t *)calloc(_sampling_driverCount, sizeof(uint32_t)); - _sampling_hDevices = - (zes_device_handle_t **)calloc(_sampling_driverCount, sizeof(zes_device_handle_t *)); - for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) { - res = - ZES_DEVICE_GET_PTR(_sampling_hDrivers[driverIdx], &_sampling_deviceCount[driverIdx], NULL); - if (res != ZE_RESULT_SUCCESS || _sampling_deviceCount[driverIdx] == 0) { - fprintf(stderr, "ERROR: No device found!\n"); - _ZE_ERROR_MSG("1st ZES_DEVICE_GET_PTR", res); - return -1; - } - _sampling_hDevices[driverIdx] = (zes_device_handle_t *)calloc(_sampling_deviceCount[driverIdx], - sizeof(zes_device_handle_t)); - res = ZES_DEVICE_GET_PTR(_sampling_hDrivers[driverIdx], &_sampling_deviceCount[driverIdx], - _sampling_hDevices[driverIdx]); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("2nd ZES_DEVICE_GET_PTR", res); - free(_sampling_hDevices[driverIdx]); - return -1; - } - for (uint32_t deviceIdx = 0; deviceIdx < _sampling_deviceCount[driverIdx]; deviceIdx++) { - - zes_device_properties_t deviceProps = {0}; - deviceProps.stype = ZES_STRUCTURE_TYPE_DEVICE_PROPERTIES; - deviceProps.pNext = NULL; - res = ZES_DEVICE_GET_PROPERTIES_PTR(_sampling_hDevices[driverIdx][deviceIdx], &deviceProps); - if (res != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_DEVICE_GET_PROPERTIES_PTR", res); - } - do_tracepoint(lttng_ust_ze_sampling, deviceProperties, - (zes_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], deviceIdx, - &deviceProps); - } - } - intializeFrequency(); - intializePower(); - intializeEngines(); - intializeFabricPorts(); - intializeMemModules(); - return 0; -} - -static void readFrequency_dump(uint32_t driverIdx, uint32_t deviceIdx) { - if (!_sampling_freq_initialized) - return; - ze_result_t result; - for (uint32_t domainIdx = 0; domainIdx < _sampling_freqDomainCounts[driverIdx][deviceIdx]; - domainIdx++) { - zes_freq_state_t freqState = {0}; - result = ZES_FREQUENCY_GET_STATE_PTR(_sampling_hFrequencies[driverIdx][deviceIdx][domainIdx], - &freqState); - if (result != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_FREQUENCY_GET_STATE_PTR", result); - continue; - } - do_tracepoint(lttng_ust_ze_sampling, gpu_frequency, - (zes_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_freq_handle_t)_sampling_hFrequencies[driverIdx][deviceIdx][domainIdx], - domainIdx, &freqState); - } -} - -static void readFabricPorts_dump(uint32_t driverIdx, uint32_t deviceIdx) { - if (!_sampling_fabricPorts_initialized) - return; - ze_result_t result; - for (uint32_t portIdx = 0; portIdx < _sampling_fabricPortCount[driverIdx][deviceIdx]; portIdx++) { - zes_fabric_port_state_t portState = {0}; - portState.pNext = NULL; - portState.stype = ZES_STRUCTURE_TYPE_FABRIC_PORT_STATE; - result = ZES_FABRIC_PORT_GET_STATE_PTR(_sampling_hFabricPort[driverIdx][deviceIdx][portIdx], - &portState); - if (result != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_FABRIC_PORT_GET_STATE_PTR", result); - continue; - } - zes_fabric_port_throughput_t throughput = {0}; - result = ZES_FABRIC_PORT_GET_THROUGHPUT_PTR( - _sampling_hFabricPort[driverIdx][deviceIdx][portIdx], &throughput); - if (result != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_FABRIC_PORT_GET_THROUGHPUT_PTR", result); - continue; - } - do_tracepoint(lttng_ust_ze_sampling, fabricPort, - (zes_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_fabric_port_handle_t)_sampling_hFabricPort[driverIdx][deviceIdx][portIdx], - &portState, &throughput); - } -} - -static void readMemModules_dump(uint32_t driverIdx, uint32_t deviceIdx) { - if (!_sampling_memModules_initialized) - return; - ze_result_t result; - for (uint32_t memModuleIdx = 0; memModuleIdx < _sampling_memModuleCount[driverIdx][deviceIdx]; - ++memModuleIdx) { - zes_mem_state_t memState = {0}; - memState.stype = ZES_STRUCTURE_TYPE_MEM_STATE; - zes_mem_bandwidth_t memBandwidth = {0}; - result = ZES_MEMORY_GET_STATE_PTR(_sampling_hMemModule[driverIdx][deviceIdx][memModuleIdx], - &memState); - if (result != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_MEMORY_GET_STATE_PTR", result); - continue; - } - result = ZES_MEMORY_GET_BANDWIDTH_PTR(_sampling_hMemModule[driverIdx][deviceIdx][memModuleIdx], - &memBandwidth); - if (result != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_MEMORY_GET_BANDWIDTH_PTR", result); - continue; - } - do_tracepoint(lttng_ust_ze_sampling, memStats, - (zes_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_mem_handle_t)_sampling_hMemModule[driverIdx][deviceIdx][memModuleIdx], - &memState, &memBandwidth); - } -} - -static void readEnergy_dump(uint32_t driverIdx, uint32_t deviceIdx) { - if (!_sampling_pwr_initialized) - return; - ze_result_t result; - for (uint32_t domainIdx = 0; domainIdx < _sampling_powerDomainCounts[driverIdx][deviceIdx]; - domainIdx++) { - zes_power_energy_counter_t energyCounter = {0}; - result = ZES_POWER_GET_ENERGY_COUNTER_PTR(_sampling_hPowers[driverIdx][deviceIdx][domainIdx], - &energyCounter); - if (result != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_POWER_GET_ENERGY_COUNTER_PTR", result); - continue; - } - do_tracepoint(lttng_ust_ze_sampling, gpu_energy, - (ze_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_pwr_handle_t)_sampling_hPowers[driverIdx][deviceIdx][domainIdx], domainIdx, - &energyCounter); - } -} -static void readEngines_dump(uint32_t driverIdx, uint32_t deviceIdx) { - if (!_sampling_engines_initialized) - return; - ze_result_t result; - for (uint32_t engineIdx = 0; engineIdx < _sampling_engineCounts[driverIdx][deviceIdx]; - ++engineIdx) { - zes_engine_stats_t engineStats = {0}; - result = ZES_ENGINE_GET_ACTIVITY_PTR(_sampling_engineHandles[driverIdx][deviceIdx][engineIdx], - &engineStats); - if (result != ZE_RESULT_SUCCESS) { - _ZE_ERROR_MSG("ZES_ENGINE_GET_ACTIVITY_PTR", result); - continue; - } - do_tracepoint(lttng_ust_ze_sampling, engineStats, - (zes_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], - (zes_engine_handle_t)_sampling_engineHandles[driverIdx][deviceIdx][engineIdx], - &engineStats); - } -} - -static void thapi_sampling_energy() { - for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) { - for (uint32_t deviceIdx = 0; deviceIdx < _sampling_deviceCount[driverIdx]; deviceIdx++) { - if (tracepoint_enabled(lttng_ust_ze_sampling, gpu_frequency)) { - readFrequency_dump(driverIdx, deviceIdx); - } - if (tracepoint_enabled(lttng_ust_ze_sampling, gpu_energy)) { - readEnergy_dump(driverIdx, deviceIdx); - } - if (tracepoint_enabled(lttng_ust_ze_sampling, engineStats)) { - readEngines_dump(driverIdx, deviceIdx); - } - if (tracepoint_enabled(lttng_ust_ze_sampling, fabricPort)) { - readFabricPorts_dump(driverIdx, deviceIdx); - } - if (tracepoint_enabled(lttng_ust_ze_sampling, memStats)) { - readMemModules_dump(driverIdx, deviceIdx); - } - } - } -} static void _load_tracer(void) { char *s = NULL; void *handle = NULL; int verbose = 0; - struct timespec interval; + //struct timespec interval; thapi_sampling_init(); s = getenv("LTTNG_UST_ZE_LIBZE_LOADER"); @@ -1319,16 +850,12 @@ static void _load_tracer(void) { _do_paranoid_memory_location = 1; s = getenv("LTTNG_UST_SAMPLING_ENERGY"); - if (s) { - initializeHandles(); - /* TODO: make it configurable */ - interval.tv_sec = 0; - interval.tv_nsec = 50000000; - thapi_sampling_energy(); - _sampling_handle = thapi_register_sampling(&thapi_sampling_energy, &interval); - } + if (s) + initialize_sampling(); _do_cleanup = 1; + cleanup_sampling(); + #ifndef THAPI_USE_DESTRUCTORS atexit(_lib_cleanup); #endif