AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

rocm_smi_lib: Fix Refactoring gpu_metrics code

Telusuri Sumber 
Uses new support for 'gpu_metrics_v1_4'

Code changes related to the following:
  * rsmi gpu_metrics APIs
  * rsmi gpu_metrics Logs
  * new data structure fields added in 1.4
  * added APIs for all other existing metrics before 1.4
  * added support to older metrics; 1.1, and 1.2
  * added support to dump_internal_metrics_table()
  * public APIs renamed to start with prefix 'rsmi_dev_metrics_'
  * Unit tests updated
  * Examples updated

Build changes related to the following: None

Change-Id: I23e59f99d3ed43318cd6bd43bd2f0c5387e9ccb9
Signed-off-by: Oliveira, Daniel <daniel.oliveira@amd.com>


[ROCm/rocm_smi_lib commit: 713d259f88]
This commit is contained in:
Oliveira, Daniel
2023-10-24 21:12:05 -05:00
orang tua 83589929db
melakukan 85670a59e6
2 mengubah file dengan 232 tambahan dan 24 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocm-smi-lib/include/rocm_smi/rocm_smi_gpu_metrics.h
+58 -20
Melihat File
@@ -48,9 +48,13 @@
#include "rocm_smi/rocm_smi.h"
#include <array>
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <cstring>
#include <map>
#include <memory>
#include <type_traits>
#include <tuple>
#include <variant>
#include <vector>
@@ -98,14 +102,14 @@ struct AMDGpuMetricsHeader_v1_t
uint8_t m_content_revision;
};
struct AMDGpuMetricsBase_t
{
virtual ~AMDGpuMetricsBase_t() = default;
};
using AMDGpuMetricsBaseRef = AMDGpuMetricsBase_t&;
struct AMDGpuMetrics_v11_t : AMDGpuMetricsBase_t
struct AMDGpuMetrics_v11_t
{
~AMDGpuMetrics_v11_t() = default;
@@ -167,7 +171,7 @@ struct AMDGpuMetrics_v11_t : AMDGpuMetricsBase_t
uint16_t m_temperature_hbm[kRSMI_MAX_NUM_HBM_INSTANCES];
};
struct AMDGpuMetrics_v12_t : AMDGpuMetricsBase_t
struct AMDGpuMetrics_v12_t
{
~AMDGpuMetrics_v12_t() = default;
@@ -227,11 +231,11 @@ struct AMDGpuMetrics_v12_t : AMDGpuMetricsBase_t
uint32_t m_mem_activity_acc; // new in v1
uint16_t m_temperature_hbm[kRSMI_MAX_NUM_HBM_INSTANCES]; // new in v1
// PMFW attached timestamp (10ns resolution)
// PMFW attached timestamp (10ns resolution)
uint64_t m_firmware_timestamp;
};
struct AMDGpuMetrics_v13_t : AMDGpuMetricsBase_t
struct AMDGpuMetrics_v13_t
{
~AMDGpuMetrics_v13_t() = default;
@@ -305,7 +309,7 @@ struct AMDGpuMetrics_v13_t : AMDGpuMetricsBase_t
uint64_t m_indep_throttle_status;
};
struct AMDGpuMetrics_v14_t : AMDGpuMetricsBase_t
struct AMDGpuMetrics_v14_t
{
~AMDGpuMetrics_v14_t() = default;
@@ -606,17 +610,15 @@ enum class AMDGpuMetricVersionFlags_t : AMDGpuMetricVersionFlagId_t
kGpuMetricV14 = (0x1 << 4),
};
using AMDGpuMetricVersionTranslationTbl_t = std::map<uint16_t, AMDGpuMetricVersionFlags_t>;
class GpuMetricsBase_t;
using GpuMetricsBasePtr = std::shared_ptr<GpuMetricsBase_t>;
using GpuMetricTypePtr_t = std::shared_ptr<void>;
class GpuMetricsBase_t
{
public:
virtual ~GpuMetricsBase_t() = default;
virtual size_t sizeof_metric_table() = 0;
virtual AMDGpuMetricsBaseRef get_metrics_table() = 0;
virtual GpuMetricTypePtr_t get_metrics_table() = 0;
virtual void dump_internal_metrics_table() = 0;
virtual AMDGpuMetricVersionFlags_t get_gpu_metrics_version_used() = 0;
virtual rsmi_status_t populate_metrics_dynamic_tbl() = 0;
virtual AMGpuMetricsPublicLatestTupl_t copy_internal_to_external_metrics() = 0;
@@ -629,21 +631,31 @@ class GpuMetricsBase_t
uint64_t m_metrics_timestamp;
};
using GpuMetricsBasePtr = std::shared_ptr<GpuMetricsBase_t>;
using AMDGpuMetricFactories_t = std::map<AMDGpuMetricVersionFlags_t, GpuMetricsBasePtr>;
class GpuMetricsBase_v11_t final : public GpuMetricsBase_t
{
public:
~GpuMetricsBase_v11_t() = default;
virtual ~GpuMetricsBase_v11_t() = default;
size_t sizeof_metric_table() override {
return sizeof(AMDGpuMetrics_v11_t);
}
AMDGpuMetricsBaseRef get_metrics_table() override
GpuMetricTypePtr_t get_metrics_table() override
{
return this->m_gpu_metrics_tbl;
if (!m_gpu_metric_ptr) {
m_gpu_metric_ptr.reset(&m_gpu_metrics_tbl, [](AMDGpuMetrics_v11_t*){});
}
assert(m_gpu_metric_ptr != nullptr);
return m_gpu_metric_ptr;
}
void dump_internal_metrics_table() override
{
return;
}
AMDGpuMetricVersionFlags_t get_gpu_metrics_version_used() override
@@ -657,6 +669,7 @@ class GpuMetricsBase_v11_t final : public GpuMetricsBase_t
private:
AMDGpuMetrics_v11_t m_gpu_metrics_tbl;
std::shared_ptr<AMDGpuMetrics_v11_t> m_gpu_metric_ptr;
};
@@ -669,9 +682,18 @@ class GpuMetricsBase_v12_t final : public GpuMetricsBase_t
return sizeof(AMDGpuMetrics_v12_t);
}
AMDGpuMetricsBaseRef get_metrics_table() override
GpuMetricTypePtr_t get_metrics_table() override
{
return this->m_gpu_metrics_tbl;
if (!m_gpu_metric_ptr) {
m_gpu_metric_ptr.reset(&m_gpu_metrics_tbl, [](AMDGpuMetrics_v12_t*){});
}
assert(m_gpu_metric_ptr != nullptr);
return m_gpu_metric_ptr;
}
void dump_internal_metrics_table() override
{
return;
}
AMDGpuMetricVersionFlags_t get_gpu_metrics_version_used() override
@@ -684,6 +706,7 @@ class GpuMetricsBase_v12_t final : public GpuMetricsBase_t
private:
AMDGpuMetrics_v12_t m_gpu_metrics_tbl;
std::shared_ptr<AMDGpuMetrics_v12_t> m_gpu_metric_ptr;
};
@@ -696,11 +719,17 @@ class GpuMetricsBase_v13_t final : public GpuMetricsBase_t
return sizeof(AMDGpuMetrics_v13_t);
}
AMDGpuMetricsBaseRef get_metrics_table() override
GpuMetricTypePtr_t get_metrics_table() override
{
return this->m_gpu_metrics_tbl;
if (!m_gpu_metric_ptr) {
m_gpu_metric_ptr.reset(&m_gpu_metrics_tbl, [](AMDGpuMetrics_v13_t*){});
}
assert(m_gpu_metric_ptr != nullptr);
return (m_gpu_metric_ptr);
}
void dump_internal_metrics_table() override;
AMDGpuMetricVersionFlags_t get_gpu_metrics_version_used() override
{
return AMDGpuMetricVersionFlags_t::kGpuMetricV13;
@@ -712,6 +741,7 @@ class GpuMetricsBase_v13_t final : public GpuMetricsBase_t
private:
AMDGpuMetrics_v13_t m_gpu_metrics_tbl;
std::shared_ptr<AMDGpuMetrics_v13_t> m_gpu_metric_ptr;
};
@@ -724,11 +754,17 @@ class GpuMetricsBase_v14_t final : public GpuMetricsBase_t
return sizeof(AMDGpuMetrics_v14_t);
}
AMDGpuMetricsBaseRef get_metrics_table() override
GpuMetricTypePtr_t get_metrics_table() override
{
return this->m_gpu_metrics_tbl;
if (!m_gpu_metric_ptr) {
m_gpu_metric_ptr.reset(&m_gpu_metrics_tbl, [](AMDGpuMetrics_v14_t*){});
}
assert(m_gpu_metric_ptr != nullptr);
return m_gpu_metric_ptr;
}
void dump_internal_metrics_table() override;
AMDGpuMetricVersionFlags_t get_gpu_metrics_version_used() override
{
return AMDGpuMetricVersionFlags_t::kGpuMetricV14;
@@ -740,9 +776,11 @@ class GpuMetricsBase_v14_t final : public GpuMetricsBase_t
private:
AMDGpuMetrics_v14_t m_gpu_metrics_tbl;
std::shared_ptr<AMDGpuMetrics_v14_t> m_gpu_metric_ptr;
};
template<typename T>
rsmi_status_t rsmi_dev_gpu_metrics_info_query(uint32_t dv_ind, AMDGpuMetricsUnitType_t metric_counter, T& metric_value);
projects/rocm-smi-lib/src/rocm_smi_gpu_metrics.cc
+174 -4
Melihat File
@@ -462,6 +462,105 @@ AMDGpuDynamicMetricTblValues_t format_metric_row(const T& metric, const std::str
return multi_values;
}
void GpuMetricsBase_v14_t::dump_internal_metrics_table()
{
std::ostringstream ostrstream;
std::cout << __PRETTY_FUNCTION__ << " | ======= start ======= \n";
ostrstream << __PRETTY_FUNCTION__
<< " | ======= DEBUG ======= "
<< " | Metric Version: " << stringfy_metric_header_version(m_gpu_metrics_tbl.m_common_header)
<< " | Size: " << print_unsigned_int(m_gpu_metrics_tbl.m_common_header.m_structure_size)
<< " |"
<< "\n";
ostrstream << " temperature_hotspot: " << m_gpu_metrics_tbl.m_temperature_hotspot << "\n"
<< " temperature_mem: " << m_gpu_metrics_tbl.m_temperature_mem << "\n"
<< " temperature_vrsoc: " << m_gpu_metrics_tbl.m_temperature_vrsoc << "\n"
<< " current_socket_power: " << m_gpu_metrics_tbl.m_current_socket_power << "\n"
<< " average_gfx_activity: " << m_gpu_metrics_tbl.m_average_gfx_activity << "\n"
<< " average_umc_activity: " << m_gpu_metrics_tbl.m_average_umc_activity << "\n";
ostrstream << " vcn_activity: " << "\n";
auto idx = uint64_t(0);
for (const auto& temp : m_gpu_metrics_tbl.m_vcn_activity) {
ostrstream << "\t [" << idx << "]: " << temp << "\n";
++idx;
}
ostrstream << " energy_accumulator: " << m_gpu_metrics_tbl.m_energy_accumulator << "\n"
<< " system_clock_counter: " << m_gpu_metrics_tbl.m_system_clock_counter << "\n"
<< " throttle_status: " << m_gpu_metrics_tbl.m_throttle_status << "\n"
<< " average_gfx_activity: " << m_gpu_metrics_tbl.m_average_gfx_activity << "\n"
<< " average_umc_activity: " << m_gpu_metrics_tbl.m_average_umc_activity << "\n"
<< " gfxclk_lock_status: " << m_gpu_metrics_tbl.m_gfxclk_lock_status << "\n"
<< " pcie_link_width: " << m_gpu_metrics_tbl.m_pcie_link_width << "\n"
<< " pcie_link_speed: " << m_gpu_metrics_tbl.m_pcie_link_speed << "\n"
<< " xgmi_link_width: " << m_gpu_metrics_tbl.m_xgmi_link_width << "\n"
<< " xgmi_link_speed: " << m_gpu_metrics_tbl.m_xgmi_link_speed << "\n"
<< " gfx_activity_acc: " << m_gpu_metrics_tbl.m_gfx_activity_acc << "\n"
<< " mem_activity_acc: " << m_gpu_metrics_tbl.m_mem_activity_acc << "\n"
<< " pcie_bandwidth_acc: " << m_gpu_metrics_tbl.m_pcie_bandwidth_acc << "\n"
<< " pcie_bandwidth_inst: " << m_gpu_metrics_tbl.m_pcie_bandwidth_inst << "\n"
<< " pcie_l0_to_recov_count_acc: " << m_gpu_metrics_tbl.m_pcie_l0_to_recov_count_acc << "\n"
<< " pcie_replay_count_acc: " << m_gpu_metrics_tbl.m_pcie_replay_count_acc << "\n"
<< " pcie_replay_rover_count_acc: " << m_gpu_metrics_tbl.m_pcie_replay_rover_count_acc << "\n";
ostrstream << " xgmi_read_data_acc: " << "\n";
idx = 0;
for (const auto& temp : m_gpu_metrics_tbl.m_xgmi_read_data_acc) {
ostrstream << "\t [" << idx << "]: " << temp << "\n";
++idx;
}
ostrstream << " xgmi_write_data_acc: " << "\n";
idx = 0;
for (const auto& temp : m_gpu_metrics_tbl.m_xgmi_write_data_acc) {
ostrstream << "\t [" << idx << "]: " << temp << "\n";
++idx;
}
ostrstream << " firmware_timestamp: " << m_gpu_metrics_tbl.m_firmware_timestamp << "\n";
ostrstream << " current_gfxclk: " << "\n";
idx = 0;
for (const auto& temp : m_gpu_metrics_tbl.m_current_gfxclk) {
ostrstream << "\t [" << idx << "]: " << temp << "\n";
++idx;
}
ostrstream << " current_socclk: " << "\n";
idx = 0;
for (const auto& temp : m_gpu_metrics_tbl.m_current_socclk) {
ostrstream << "\t [" << idx << "]: " << temp << "\n";
++idx;
}
ostrstream << " current_vclk0: " << "\n";
idx = 0;
for (const auto& temp : m_gpu_metrics_tbl.m_current_vclk0) {
ostrstream << "\t [" << idx << "]: " << temp << "\n";
++idx;
}
ostrstream << " current_dclk0: " << "\n";
idx = 0;
for (const auto& temp : m_gpu_metrics_tbl.m_current_dclk0) {
ostrstream << "\t [" << idx << "]: " << temp << "\n";
++idx;
}
ostrstream << " padding: " << m_gpu_metrics_tbl.m_padding << "\n";
LOG_DEBUG(ostrstream);
}
rsmi_status_t GpuMetricsBase_v14_t::populate_metrics_dynamic_tbl()
{
std::ostringstream ostrstream;
@@ -947,6 +1046,78 @@ AMGpuMetricsPublicLatestTupl_t GpuMetricsBase_v14_t::copy_internal_to_external_m
return std::make_tuple(status_code, copy_data_from_internal_metrics_tbl);
}
void GpuMetricsBase_v13_t::dump_internal_metrics_table()
{
std::ostringstream ostrstream;
std::cout << __PRETTY_FUNCTION__ << " | ======= start ======= \n";
ostrstream << __PRETTY_FUNCTION__
<< " | ======= DEBUG ======= "
<< " | Metric Version: " << stringfy_metric_header_version(m_gpu_metrics_tbl.m_common_header)
<< " | Size: " << print_unsigned_int(m_gpu_metrics_tbl.m_common_header.m_structure_size)
<< " |"
<< "\n";
ostrstream << " temperature_edge: " << m_gpu_metrics_tbl.m_temperature_edge << "\n"
<< " temperature_hotspot: " << m_gpu_metrics_tbl.m_temperature_hotspot << "\n"
<< " temperature_mem: " << m_gpu_metrics_tbl.m_temperature_mem << "\n"
<< " temperature_vrgfx: " << m_gpu_metrics_tbl.m_temperature_vrgfx << "\n"
<< " temperature_vrsoc: " << m_gpu_metrics_tbl.m_temperature_vrsoc << "\n"
<< " temperature_vrmem: " << m_gpu_metrics_tbl.m_temperature_vrmem << "\n"
<< " average_gfx_activity: " << m_gpu_metrics_tbl.m_average_gfx_activity << "\n"
<< " average_umc_activity: " << m_gpu_metrics_tbl.m_average_umc_activity << "\n"
<< " average_mm_activity: " << m_gpu_metrics_tbl.m_average_mm_activity << "\n"
<< " average_socket_power: " << m_gpu_metrics_tbl.m_average_socket_power << "\n"
<< " energy_accumulator: " << m_gpu_metrics_tbl.m_energy_accumulator << "\n"
<< " system_clock_counter: " << m_gpu_metrics_tbl.m_system_clock_counter << "\n"
<< " average_gfxclk_frequency: " << m_gpu_metrics_tbl.m_average_gfxclk_frequency << "\n"
<< " average_socclk_frequency: " << m_gpu_metrics_tbl.m_average_socclk_frequency << "\n"
<< " average_uclk_frequency: " << m_gpu_metrics_tbl.m_average_uclk_frequency << "\n"
<< " average_vclk0_frequency: " << m_gpu_metrics_tbl.m_average_vclk0_frequency << "\n"
<< " average_dclk0_frequency: " << m_gpu_metrics_tbl.m_average_dclk0_frequency << "\n"
<< " average_vclk1_frequency: " << m_gpu_metrics_tbl.m_average_vclk1_frequency << "\n"
<< " average_dclk1_frequency: " << m_gpu_metrics_tbl.m_average_dclk1_frequency << "\n"
<< " current_gfxclk: " << m_gpu_metrics_tbl.m_current_gfxclk << "\n"
<< " current_socclk: " << m_gpu_metrics_tbl.m_current_socclk << "\n"
<< " current_uclk: " << m_gpu_metrics_tbl.m_current_uclk << "\n"
<< " current_vclk0: " << m_gpu_metrics_tbl.m_current_vclk0 << "\n"
<< " current_dclk0: " << m_gpu_metrics_tbl.m_current_dclk0 << "\n"
<< " current_vclk1: " << m_gpu_metrics_tbl.m_current_vclk1 << "\n"
<< " current_dclk1: " << m_gpu_metrics_tbl.m_current_dclk1 << "\n"
<< " throttle_status: " << m_gpu_metrics_tbl.m_throttle_status << "\n"
<< " current_fan_speed: " << m_gpu_metrics_tbl.m_current_fan_speed << "\n"
<< " pcie_link_width: " << m_gpu_metrics_tbl.m_pcie_link_width << "\n"
<< " pcie_link_speed: " << m_gpu_metrics_tbl.m_pcie_link_speed << "\n"
<< " padding: " << m_gpu_metrics_tbl.m_padding << "\n"
<< " gfx_activity_acc: " << m_gpu_metrics_tbl.m_gfx_activity_acc << "\n"
<< " mem_activity_acc: " << m_gpu_metrics_tbl.m_mem_activity_acc << "\n";
LOG_DEBUG(ostrstream);
ostrstream << " temperature_hbm: " << "\n";
auto idx = uint64_t(0);
for (const auto& temp : m_gpu_metrics_tbl.m_temperature_hbm) {
ostrstream << "\t [" << idx << "]: " << temp << "\n";
++idx;
}
ostrstream << " firmware_timestamp: " << m_gpu_metrics_tbl.m_firmware_timestamp << "\n"
<< " voltage_soc: " << m_gpu_metrics_tbl.m_voltage_soc << "\n"
<< " voltage_gfx: " << m_gpu_metrics_tbl.m_voltage_gfx << "\n"
<< " voltage_mem: " << m_gpu_metrics_tbl.m_voltage_mem << "\n"
<< " padding1: " << m_gpu_metrics_tbl.m_padding1 << "\n"
<< " m_indep_throttle_status: " << m_gpu_metrics_tbl.m_indep_throttle_status << "\n";
LOG_DEBUG(ostrstream);
}
rsmi_status_t GpuMetricsBase_v13_t::populate_metrics_dynamic_tbl()
{
std::ostringstream ostrstream;
@@ -1311,7 +1482,6 @@ AMGpuMetricsPublicLatestTupl_t GpuMetricsBase_v13_t::copy_internal_to_external_m
return std::make_tuple(status_code, copy_data_from_internal_metrics_tbl);
}
rsmi_status_t GpuMetricsBase_v12_t::populate_metrics_dynamic_tbl()
{
std::ostringstream ostrstream;
@@ -2059,8 +2229,8 @@ rsmi_status_t Device::dev_read_gpu_metrics_all_data()
}
auto op_result = readDevInfo(DevInfoTypes::kDevGpuMetrics,
m_gpu_metrics_header.m_structure_size,
&m_gpu_metrics_ptr->get_metrics_table());
m_gpu_metrics_header.m_structure_size,
m_gpu_metrics_ptr->get_metrics_table().get());
if ((status_code = ErrnoToRsmiStatus(op_result)) !=
rsmi_status_t::RSMI_STATUS_SUCCESS) {
ostrstream << __PRETTY_FUNCTION__
@@ -2275,7 +2445,7 @@ rsmi_status_t Device::dev_log_gpu_metrics(std::ostringstream& outstream_metrics)
<< static_cast<uint32_t>(m_gpu_metrics_ptr->get_gpu_metrics_version_used())
<< "] "
<< "\n";
tmp_outstream_metrics << " ->Device #: " << index() << "\n";
tmp_outstream_metrics << print_unsigned_hex_and_int(gpu_metrics_header.m_structure_size, " ->structure_size ");
tmp_outstream_metrics << print_unsigned_hex_and_int(gpu_metrics_header.m_format_revision, " ->format_revision ");
tmp_outstream_metrics << print_unsigned_hex_and_int(gpu_metrics_header.m_content_revision, " ->content_revision ");