geopm  3.1.1.dev296+g5916b956
GEOPM - Global Extensible Open Power Manager
LevelZeroDevicePoolImp.hpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2015 - 2024 Intel Corporation
3  * SPDX-License-Identifier: BSD-3-Clause
4  */
5 
6 #ifndef LEVELZERODEVICEPOOLIMP_HPP_INCLUDE
7 #define LEVELZERODEVICEPOOLIMP_HPP_INCLUDE
8 
9 #include <string>
10 #include <cstdint>
11 #include <map>
12 
13 #include "LevelZeroDevicePool.hpp"
14 #include "LevelZero.hpp"
15 
16 #include "geopm_time.h"
17 
18 namespace geopm
19 {
21  {
22  public:
25  virtual ~LevelZeroDevicePoolImp() = default;
26  int num_gpu(int domain_type) const override;
27  double frequency_status(int domain, unsigned int domain_idx,
28  int l0_domain) const override;
29  double frequency_efficient(int domain, unsigned int domain_idx,
30  int l0_domain) const override;
31  double frequency_min(int domain, unsigned int domain_idx,
32  int l0_domain) const override;
33  double frequency_max(int domain, unsigned int domain_idx,
34  int l0_domain) const override;
35  double frequency_step(int domain, unsigned int domain_idx,
36  int l0_domain) const override;
37  uint32_t frequency_throttle_reasons(int domain, unsigned int domain_idx,
38  int l0_domain) const override;
39  std::pair <double, double> frequency_range(int domain,
40  unsigned int domain_idx,
41  int l0_domain) const override;
42  double temperature_max(int domain, unsigned int domain_idx,
43  int l0_domain) const override;
44  std::pair<uint64_t, uint64_t> active_time_pair(int domain,
45  unsigned int device_idx,
46  int l0_domain) const override;
47  double active_time(int domain, unsigned int device_idx,
48  int l0_domain) const override;
49  double active_time_timestamp(int domain, unsigned int device_idx,
50  int l0_domain) const override;
51  int32_t power_limit_tdp(int domain, unsigned int domain_idx,
52  int l0_domain) const override;
53  int32_t power_limit_min(int domain, unsigned int domain_idx,
54  int l0_domain) const override;
55  int32_t power_limit_max(int domain, unsigned int domain_idx,
56  int l0_domain) const override;
57  std::pair<uint64_t, uint64_t> energy_pair(int domain, unsigned int domain_idx,
58  int l0_domain) const override;
59  uint64_t energy(int domain, unsigned int domain_idx, int l0_domain) const override;
60  uint64_t energy_timestamp(int domain, unsigned int domain_idx,
61  int l0_domain) const override;
62  double performance_factor(int domain,
63  unsigned int domain_idx,
64  int l0_domain) const override;
65  void frequency_control(int domain, unsigned int domain_idx,
66  int l0_domain, double range_min,
67  double range_max) const override;
68  void performance_factor_control(int domain, unsigned int domain_idx,
69  int l0_domain,
70  double setting) const override;
71  double ras_reset_count_correctable(int domain, unsigned int domain_idx,
72  int l0_domain) const override;
73  double ras_programming_errcount_correctable(int domain, unsigned int domain_idx,
74  int l0_domain) const override;
75  double ras_driver_errcount_correctable(int domain, unsigned int domain_idx,
76  int l0_domain) const override;
77  double ras_compute_errcount_correctable(int domain, unsigned int domain_idx,
78  int l0_domain) const override;
79  double ras_noncompute_errcount_correctable(int domain, unsigned int domain_idx,
80  int l0_domain) const override;
81  double ras_cache_errcount_correctable(int domain, unsigned int domain_idx,
82  int l0_domain) const override;
83  double ras_display_errcount_correctable(int domain, unsigned int domain_idx,
84  int l0_domain) const override;
85  double ras_reset_count_uncorrectable(int domain, unsigned int domain_idx,
86  int l0_domain) const override;
87  double ras_programming_errcount_uncorrectable(int domain, unsigned int domain_idx,
88  int l0_domain) const override;
89  double ras_driver_errcount_uncorrectable(int domain, unsigned int domain_idx,
90  int l0_domain) const override;
91  double ras_compute_errcount_uncorrectable(int domain, unsigned int domain_idx,
92  int l0_domain) const override;
93  double ras_noncompute_errcount_uncorrectable(int domain, unsigned int domain_idx,
94  int l0_domain) const override;
95  double ras_cache_errcount_uncorrectable(int domain, unsigned int domain_idx,
96  int l0_domain) const override;
97  double ras_display_errcount_uncorrectable(int domain, unsigned int domain_idx,
98  int l0_domain) const override;
99  private:
100  const LevelZero &m_levelzero;
101 
102  void check_idx_range(int domain, unsigned int domain_idx) const;
103  void check_domain_exists(int size, const char *func, int line) const;
104  std::pair<unsigned int, unsigned int> subdevice_device_conversion(unsigned int idx) const;
105  mutable std::map<int, std::vector<uint64_t> > m_active_time_last; // Map from l0_domain to vector over gpu chips
106  mutable std::map<int, std::vector<uint64_t> > m_active_time_rollover; // Map from l0_domain to vector over gpu chips
107  };
108 }
109 #endif
Definition: LevelZeroDevicePool.hpp:19
Definition: LevelZeroDevicePoolImp.hpp:21
double frequency_min(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device mininmum frequency in MHz.
Definition: LevelZeroDevicePool.cpp:127
void performance_factor_control(int domain, unsigned int domain_idx, int l0_domain, double setting) const override
Set performance factor for LevelZero device.
Definition: LevelZeroDevicePool.cpp:493
double ras_compute_errcount_correctable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of correctable errors that have occurred in the compute accelerator...
Definition: LevelZeroDevicePool.cpp:568
double ras_display_errcount_correctable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of correctable errors that have occurred in the display.
Definition: LevelZeroDevicePool.cpp:622
double ras_driver_errcount_correctable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of low level driver communication correctable errors have occurred.
Definition: LevelZeroDevicePool.cpp:550
double frequency_status(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device actual frequency in MHz.
Definition: LevelZeroDevicePool.cpp:91
uint32_t frequency_throttle_reasons(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device frequency throttle reasons.
Definition: LevelZeroDevicePool.cpp:191
double temperature_max(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero domain maximum temperature in Celsius.
Definition: LevelZeroDevicePool.cpp:229
double performance_factor(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device performance factor.
Definition: LevelZeroDevicePool.cpp:457
double ras_cache_errcount_correctable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of correctable errors that have occurred in caches (L1/L3/register ...
Definition: LevelZeroDevicePool.cpp:604
double ras_cache_errcount_uncorrectable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of uncorrectable errors that have occurred in caches (L1/L3/registe...
Definition: LevelZeroDevicePool.cpp:732
LevelZeroDevicePoolImp()
Definition: LevelZeroDevicePool.cpp:31
double frequency_step(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device frequency step in MHz.
Definition: LevelZeroDevicePool.cpp:163
double ras_compute_errcount_uncorrectable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of uncorrectable errors that have occurred in the compute accelerat...
Definition: LevelZeroDevicePool.cpp:696
double frequency_efficient(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device efficient frequency in MHz.
Definition: LevelZeroDevicePool.cpp:109
void frequency_control(int domain, unsigned int domain_idx, int l0_domain, double range_min, double range_max) const override
Set min and max frequency for LevelZero device.
Definition: LevelZeroDevicePool.cpp:473
virtual ~LevelZeroDevicePoolImp()=default
std::pair< double, double > frequency_range(int domain, unsigned int domain_idx, int l0_domain) const override
Definition: LevelZeroDevicePool.cpp:209
uint64_t energy(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device energy in microjoules.
Definition: LevelZeroDevicePool.cpp:430
int32_t power_limit_min(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device minimum power limit in milliwatts.
Definition: LevelZeroDevicePool.cpp:328
double frequency_max(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device maximum frequency in MHz.
Definition: LevelZeroDevicePool.cpp:145
double ras_noncompute_errcount_uncorrectable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of uncorrectable errors that have occurred in the fixed-function ac...
Definition: LevelZeroDevicePool.cpp:714
double active_time_timestamp(int domain, unsigned int device_idx, int l0_domain) const override
Get the LevelZero device timestamp for the active time value in microseconds.
Definition: LevelZeroDevicePool.cpp:283
double ras_programming_errcount_correctable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of correctable hardware exceptions generated by the way workloads h...
Definition: LevelZeroDevicePool.cpp:532
std::pair< uint64_t, uint64_t > energy_pair(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device energy in microjoules and timestamp in microseconds.
Definition: LevelZeroDevicePool.cpp:369
double ras_reset_count_uncorrectable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of uncorrectable accelerator engine resets attempted by the driver.
Definition: LevelZeroDevicePool.cpp:642
double ras_reset_count_correctable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of correctable accelerator engine resets attempted by the driver.
Definition: LevelZeroDevicePool.cpp:514
double ras_programming_errcount_uncorrectable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of uncorrectable hardware exceptions generated by the way workloads...
Definition: LevelZeroDevicePool.cpp:660
uint64_t energy_timestamp(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device energy timestamp in microseconds.
Definition: LevelZeroDevicePool.cpp:399
double active_time(int domain, unsigned int device_idx, int l0_domain) const override
Get the LevelZero device timestamp for the active time value in microseconds.
Definition: LevelZeroDevicePool.cpp:304
int32_t power_limit_max(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device maximum power limit in milliwatts.
Definition: LevelZeroDevicePool.cpp:341
double ras_driver_errcount_uncorrectable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of low level driver communication uncorrectable errors have occurre...
Definition: LevelZeroDevicePool.cpp:678
double ras_noncompute_errcount_correctable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of correctable errors that have occurred in the fixed-function acce...
Definition: LevelZeroDevicePool.cpp:586
int32_t power_limit_tdp(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero device default power limit in milliwatts.
Definition: LevelZeroDevicePool.cpp:355
std::pair< uint64_t, uint64_t > active_time_pair(int domain, unsigned int device_idx, int l0_domain) const override
Get the LevelZero device active time and timestamp in microseconds.
Definition: LevelZeroDevicePool.cpp:248
int num_gpu(int domain_type) const override
Number of GPUs on the platform.
Definition: LevelZeroDevicePool.cpp:36
double ras_display_errcount_uncorrectable(int domain, unsigned int domain_idx, int l0_domain) const override
Get the LevelZero count of number of uncorrectable errors that have occurred in the display.
Definition: LevelZeroDevicePool.cpp:750
Definition: LevelZero.hpp:18
Definition: Agg.cpp:20
const LevelZero & levelzero()
Definition: LevelZero.cpp:28