linuxSmartie/src/gpu.cpp

#include <cstdint>
#include <fstream>
#include <gpu.hpp>
#include <iostream>
#include <string>
#include <utils.hpp>

struct GpuMetrics_v1_1_t {
    uint16_t temperature_edge;
    uint16_t temperature_hotspot;
    uint16_t temperature_mem;
    uint16_t temperature_vrgfx;
    uint16_t temperature_vrsoc;
    uint16_t temperature_vrmem;
    uint16_t average_gfx_activity;
    uint16_t average_umc_activity;
    uint16_t average_mm_activity;
    uint16_t average_socket_power;
    uint64_t energy_accumulator;
    uint64_t system_clock_counter;
    uint16_t average_gfxclk_frequency;
    uint16_t average_socclk_frequency;
    uint16_t average_uclk_frequency;
    uint16_t average_vclk0_frequency;
    uint16_t average_dclk0_frequency;
    uint16_t average_vclk1_frequency;
    uint16_t average_dclk1_frequency;
    uint16_t current_gfxclk;
    uint16_t current_socclk;
    uint16_t current_uclk;
    uint16_t current_vclk0;
    uint16_t current_dclk0;
    uint16_t current_vclk1;
    uint16_t current_dclk1;
    uint32_t throttle_status;
    uint16_t current_fan_speed;
    uint16_t pcie_link_width;
    uint16_t pcie_link_speed;
    uint16_t padding;
    uint32_t gfx_activity_acc;
    uint32_t mem_activity_acc;
    uint16_t temperature_hbm;
};

struct GpuMetrics_v1_3_t {
    uint16_t temperature_edge;
    uint16_t temperature_hotspot;
    uint16_t temperature_mem;
    uint16_t temperature_vrgfx;
    uint16_t temperature_vrsoc;
    uint16_t temperature_vrmem;
    uint16_t average_gfx_activity;
    uint16_t average_umc_activity;
    uint16_t average_mm_activity;
    uint16_t average_socket_power;
    uint64_t energy_accumulator;
    uint64_t system_clock_counter;
    uint16_t average_gfxclk_frequency;
    uint16_t average_socclk_frequency;
    uint16_t average_uclk_frequency;
    uint16_t average_vclk0_frequency;
    uint16_t average_dclk0_frequency;
    uint16_t average_vclk1_frequency;
    uint16_t average_dclk1_frequency;
    uint16_t current_gfxclk;
    uint16_t current_socclk;
    uint16_t current_uclk;
    uint16_t current_vclk0;
    uint16_t current_dclk0;
    uint16_t current_vclk1;
    uint16_t current_dclk1;
    uint32_t throttle_status;
    uint16_t current_fan_speed;
    uint16_t pcie_link_width;
    uint16_t pcie_link_speed;
    uint16_t padding;
    uint32_t gfx_activity_acc;
    uint32_t mem_activity_acc;
    uint16_t temperature_hbm;
    uint64_t firmware_timestamp;
    uint16_t voltage_soc;
    uint16_t voltage_gfx;
    uint16_t voltage_mem;
    uint8_t padding1;
};

GpuMetricsHeader_t getGpuMetricsHeader(std::string path) {
    GpuMetricsHeader_t header;
    std::ifstream f(path, std::ios::binary);
    if (f.good()) {
        f.read(reinterpret_cast<char *>(&header), sizeof(header));
        std::cout << "[GpuMetrics] header at path " + path + " read, GpuMetrics_v" << std::to_string(header.format_revision) << "_" << std::to_string(header.content_revision);
        std::cout << ", size " << std::to_string(header.structure_size) << std::endl;
        f.close();
    } else {
        std::cerr << "[GpuMetrics] error reading header at path " + path + " (wrong card # ?)" << std::endl;
    }

    return header;
}

auto getGpuMetrics(GpuMetricsHeader_t h, std::string path) {
    GpuMetrics_v1_3_t metrics;
    if (h.format_revision == 1 && h.content_revision == 3) {
        std::ifstream f(path, std::ios::binary);
        if (f.good()) {
            f.ignore(sizeof(h));  // skip header
            f.read(reinterpret_cast<char *>(&metrics), sizeof(metrics));
            f.close();
        }
    } else {
        std::cout << "[GpuMetrics] read failed, unsupported format" << std::endl;
    }

    return metrics;
}

// clean abstraction of gpu_metrics + extra metrics from fs
gpu_t getGpu(GpuMetricsHeader_t header, std::string hw) {
    gpu_t gpu;
    auto metrics = getGpuMetrics(header, hw + "/gpu_metrics");

    // we want as much data as posible from gpu_metrics so we dont read filesystem as much
    gpu.fan_speed = std::stol(readFile(hw + "/hwmon/hwmon2/fan1_input"));
    gpu.core_clock = metrics.average_vclk1_frequency;  // why not gfxclock?
    gpu.mem_clock = metrics.average_gfxclk_frequency;  // why gfxclock? min mem clock seems to be 96mhz, this is probably including time sleeping
    gpu.vcore = std::stod(readFile(hw + "/hwmon/hwmon2/in0_input")) / 1000.0;
    gpu.power_avg = metrics.average_socket_power;
    gpu.power_cap = std::stol(readFile(hw + "/hwmon/hwmon2/power1_cap", 6));
    gpu.power_max = std::stol(readFile(hw + "/hwmon/hwmon2/power1_cap_max", 6));
    // gpu.fan_pwm = std::stoi(readFile(hw + "/hwmon/hwmon2/pwm1")); // not super useful
    gpu.edge_temp = metrics.temperature_edge;
    gpu.hotspot_temp = metrics.temperature_hotspot;
    gpu.mem_temp = metrics.temperature_mem;
    gpu.gfx_vrm_temp = metrics.temperature_vrgfx;
    gpu.soc_vrm_temp = metrics.temperature_vrsoc;
    gpu.mem_vrm_temp = metrics.temperature_vrmem;
    gpu.vram_used = std::stoul(readFile(hw + "/mem_info_vram_used"));
    gpu.vram_total = std::stoul(readFile(hw + "/mem_info_vram_total"));
    gpu.core_usage = metrics.average_gfx_activity;
    gpu.media_usage = metrics.average_mm_activity;
    gpu.umc_usage = metrics.average_umc_activity;
    return gpu;
}