geopm  3.1.1.dev214+gba4f9f6d
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 
12 #include "LevelZeroDevicePool.hpp"
13 #include "LevelZero.hpp"
14 
15 #include "geopm_time.h"
16 
17 namespace geopm
18 {
20  {
21  public:
24  virtual ~LevelZeroDevicePoolImp() = default;
25  int num_gpu(int domain_type) const override;
26  double frequency_status(int domain, unsigned int domain_idx,
27  int l0_domain) const override;
28  double frequency_efficient(int domain, unsigned int domain_idx,
29  int l0_domain) const override;
30  double frequency_min(int domain, unsigned int domain_idx,
31  int l0_domain) const override;
32  double frequency_max(int domain, unsigned int domain_idx,
33  int l0_domain) const override;
34  double frequency_step(int domain, unsigned int domain_idx,
35  int l0_domain) const override;
36  uint32_t frequency_throttle_reasons(int domain, unsigned int domain_idx,
37  int l0_domain) const override;
38  std::pair <double, double> frequency_range(int domain,
39  unsigned int domain_idx,
40  int l0_domain) const override;
41  double temperature_max(int domain, unsigned int domain_idx,
42  int l0_domain) const override;
43  std::pair<uint64_t, uint64_t> active_time_pair(int domain,
44  unsigned int device_idx,
45  int l0_domain) const override;
46  uint64_t active_time(int domain, unsigned int device_idx,
47  int l0_domain) const override;
48  uint64_t active_time_timestamp(int domain, unsigned int device_idx,
49  int l0_domain) const override;
50  int32_t power_limit_tdp(int domain, unsigned int domain_idx,
51  int l0_domain) const override;
52  int32_t power_limit_min(int domain, unsigned int domain_idx,
53  int l0_domain) const override;
54  int32_t power_limit_max(int domain, unsigned int domain_idx,
55  int l0_domain) const override;
56  std::pair<uint64_t, uint64_t> energy_pair(int domain, unsigned int domain_idx,
57  int l0_domain) const override;
58  uint64_t energy(int domain, unsigned int domain_idx, int l0_domain) const override;
59  uint64_t energy_timestamp(int domain, unsigned int domain_idx,
60  int l0_domain) const override;
61  double performance_factor(int domain,
62  unsigned int domain_idx,
63  int l0_domain) const override;
64  void frequency_control(int domain, unsigned int domain_idx,
65  int l0_domain, double range_min,
66  double range_max) const override;
67  void performance_factor_control(int domain, unsigned int domain_idx,
68  int l0_domain,
69  double setting) const override;
70  double ras_reset_count_correctable(int domain, unsigned int domain_idx,
71  int l0_domain) const override;
72  double ras_programming_errcount_correctable(int domain, unsigned int domain_idx,
73  int l0_domain) const override;
74  double ras_driver_errcount_correctable(int domain, unsigned int domain_idx,
75  int l0_domain) const override;
76  double ras_compute_errcount_correctable(int domain, unsigned int domain_idx,
77  int l0_domain) const override;
78  double ras_noncompute_errcount_correctable(int domain, unsigned int domain_idx,
79  int l0_domain) const override;
80  double ras_cache_errcount_correctable(int domain, unsigned int domain_idx,
81  int l0_domain) const override;
82  double ras_display_errcount_correctable(int domain, unsigned int domain_idx,
83  int l0_domain) const override;
84  double ras_reset_count_uncorrectable(int domain, unsigned int domain_idx,
85  int l0_domain) const override;
86  double ras_programming_errcount_uncorrectable(int domain, unsigned int domain_idx,
87  int l0_domain) const override;
88  double ras_driver_errcount_uncorrectable(int domain, unsigned int domain_idx,
89  int l0_domain) const override;
90  double ras_compute_errcount_uncorrectable(int domain, unsigned int domain_idx,
91  int l0_domain) const override;
92  double ras_noncompute_errcount_uncorrectable(int domain, unsigned int domain_idx,
93  int l0_domain) const override;
94  double ras_cache_errcount_uncorrectable(int domain, unsigned int domain_idx,
95  int l0_domain) const override;
96  double ras_display_errcount_uncorrectable(int domain, unsigned int domain_idx,
97  int l0_domain) const override;
98  private:
99  const LevelZero &m_levelzero;
100 
101  void check_idx_range(int domain, unsigned int domain_idx) const;
102  void check_domain_exists(int size, const char *func, int line) const;
103  std::pair<unsigned int, unsigned int> subdevice_device_conversion(unsigned int idx) const;
104  };
105 }
106 #endif
Definition: LevelZeroDevicePool.hpp:19
Definition: LevelZeroDevicePoolImp.hpp:20
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:474
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:549
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:603
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:531
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:438
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:585
uint64_t 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:290
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:713
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:677
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:454
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:411
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:309
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:695
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:513
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:350
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:623
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:495
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:641
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:380
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:322
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:659
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:567
uint64_t 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:269
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:336
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:731
Definition: LevelZero.hpp:18
Definition: Agg.cpp:20
const LevelZero & levelzero()
Definition: LevelZero.cpp:21