ampere_research/analysis/data_stat.py

import re

from enum import Enum

class Stat(Enum):
    CPU = 'cpu'

    THREADS = 'threads'
    ITERATIONS = 'iterations'
    BASELINE_TIME_S = 'baseline time (sec)'
    BASELINE_DELAY_S = 'baseline delay (sec)'

    SOLVER = 'solver'

    MATRIX_FILE = 'matrix file'
    MATRIX_SHAPE = 'matrix shape'
    MATRIX_SIZE = 'matrix size'
    MATRIX_NNZ = 'matrix nnz'
    MATRIX_DENSITY = 'matrix density %'
 
    POWER_BEFORE = 'power before'
    POWER = 'power'
    POWER_AFTER = 'power after'
    TIME_S = 'time (sec)'
    DELTA_WATT = 'Δ watt'
    JOULES = 'joules'

    TASK_CLK = 'task clock (msec)'
    PAGE_FAULTS = 'page faults'
    CYCLES = 'cycles'
    INSTS = 'instructions'

    BR = 'branches'
    BR_MISS = 'branch mispredictions'
    ITLB = 'ITLB accesses'
    ITLB_MISS = 'ITLB misses'
    DTLB = 'DTLB accesses'
    DTLB_MISS = 'DTLB misses'
    L2D_TLB = 'L2D TLB accesses'
    L2D_TLB_MISS = 'L2D TLB misses'
    L1I_CACHE = 'L1I cache accesses'
    L1I_CACHE_MISS = 'L1I cache misses'
    L1D_CACHE = 'L1D cache accesses'
    L1D_CACHE_MISS = 'L1D cache misses'
    L2D_CACHE = 'L2D cache accesses'
    L2D_CACHE_MISS = 'L2D cache misses'
    LL_CACHE = 'LL cache accesses'
    LL_CACHE_MISS = 'LL cache misses'

    IPC = "instructions per cycle"
    BRANCH_MISS_RATE = 'branch miss rate'
    ITLB_MISS_RATE = 'ITLB miss rate'
    DTLB_MISS_RATE = 'DTLB miss rate'
    L2D_TLB_MISS_RATE = 'L2D TLB miss rate'
    L1I_CACHE_MISS_RATE = 'L1I cache miss rate'
    L1D_CACHE_MISS_RATE = 'L1D cache miss rate'
    L2D_CACHE_MISS_RATE = 'L2D cache miss rate'
    LL_CACHE_MISS_RATE = 'LL cache miss rate'

class Cpu(Enum):
    #ALTRA = altra_names
    #XEON = xeon_names
    ALTRA = 'Altra'
    EPYC_7313P = 'Epyc 7313P'

names = {
        Cpu.ALTRA: {
            Stat.TASK_CLK: 'task-clock:u',
            Stat.PAGE_FAULTS: 'page-faults:u',
            Stat.CYCLES: 'cycles:u',
            Stat.INSTS: 'instructions:u',

            Stat.BR: 'BR_RETIRED:u',
            Stat.BR_MISS: 'BR_MIS_PRED_RETIRED:u',
            Stat.ITLB: 'L1I_TLB:u',
            Stat.ITLB_MISS: 'ITLB_WALK:u',
            Stat.DTLB: 'L1D_TLB:u',
            Stat.DTLB_MISS: 'DTLB_WALK:u',
            Stat.L2D_TLB: 'L2D_TLB:u',
            Stat.L2D_TLB_MISS: 'L2D_TLB_REFILL:u',
            Stat.L1I_CACHE: 'L1I_CACHE:u',
            Stat.L1I_CACHE_MISS: 'L1I_CACHE_REFILL:u',
            Stat.L1D_CACHE: 'L1D_CACHE:u',
            Stat.L1D_CACHE_MISS: 'L1D_CACHE_REFILL:u',
            Stat.L2D_CACHE: 'L2D_CACHE:u',
            Stat.L2D_CACHE_MISS: 'L2D_CACHE_REFILL:u',
            Stat.LL_CACHE: 'LL_CACHE_RD:u',
            Stat.LL_CACHE_MISS: 'LL_CACHE_MISS_RD:u',
        },
        Cpu.EPYC_7313P: {
            Stat.TASK_CLK: 'task-clock:u',
            Stat.PAGE_FAULTS: 'page-faults:u',
            Stat.CYCLES: 'cycles:u',
            Stat.INSTS: 'instructions:u',

            Stat.BR: 'branches:u',
            Stat.BR_MISS: 'branch-misses:u',
            Stat.ITLB: 'iTLB-loads:u',
            Stat.ITLB_MISS: 'iTLB-load-misses:u',
            Stat.DTLB: 'dTLB-loads:u',
            Stat.DTLB_MISS: 'dTLB-load-misses:u',
            Stat.L1I_CACHE: 'L1-icache-loads:u',
            Stat.L1I_CACHE_MISS: 'L1-icache-load-misses:u',
            Stat.L1D_CACHE: 'L1-dcache-loads:u',
            Stat.L1D_CACHE_MISS: 'L1-dcache-load-misses:u',
            Stat.LL_CACHE: 'LLC-loads:u',
            Stat.LL_CACHE_MISS: 'LLC-load-misses:u',
        }
}

def parse_output_old(filename: str, data: dict[str, str]) -> dict:
    result: dict[str, int | float] = dict()
    cpu: CPU = CPU[data['cpu'].upper()]

    with open(filename, 'r') as file:
        for line in file:
            for stat in [x for x in Stat if x in cpu.value]:
                regex = r'^\W*([\d+(,|\.)?]+)\W*.*' + cpu.value[stat]
                value = re.search(regex, line)

                if value is None:
                    continue
                elif stat == Stat.TASK_CLK: 
                    result[stat.value] = float(value.group(1).replace(',', ''))
                else:
                    result[stat.value] = int(value.group(1).replace(',', ''))

    return result | parse_power(filename, cpu)

def parse_output(output: str, cpu: Cpu) -> dict:
    result = dict()

    for line in output.split('\n'):
        for stat in [x for x in Stat if x in names[cpu]]:
            regex = r'^\W*([\d+(,|\.)?]+)\W*.*' + names[cpu][stat]
            value = re.search(regex, line)

            if value is None:
                continue
            elif stat == Stat.TASK_CLK: 
                result[stat.value] = float(value.group(1).replace(',', ''))
            else:
                result[stat.value] = int(value.group(1).replace(',', ''))

    return result

def derive_stats(data: dict) -> dict:
    result = dict()

    result[Stat.IPC.value] = data[Stat.INSTS.value] / data[Stat.CYCLES.value]
    result[Stat.BRANCH_MISS_RATE.value] = (
            data[Stat.BR_MISS.value] / data[Stat.BR.value])
    result[Stat.ITLB_MISS_RATE.value] = (
            data[Stat.ITLB_MISS.value] / data[Stat.ITLB.value])
    result[Stat.DTLB_MISS_RATE.value] = (
            data[Stat.DTLB_MISS.value] / data[Stat.DTLB.value])
    result[Stat.L2D_TLB_MISS_RATE.value] = (
            data[Stat.L2D_TLB_MISS.value] / data[Stat.L2D_TLB.value]
            if Stat.L2D_TLB_MISS.value in data and Stat.L2D_TLB.value in data
            else None)
    result[Stat.L1I_CACHE_MISS_RATE.value] = (
            data[Stat.L1I_CACHE_MISS.value] / data[Stat.L1I_CACHE.value])
    result[Stat.L1D_CACHE_MISS_RATE.value] = (
        data[Stat.L1D_CACHE_MISS.value] / data[Stat.L1D_CACHE.value])
    result[Stat.L2D_CACHE_MISS_RATE.value] = (
        data[Stat.L2D_CACHE_MISS.value] / data[Stat.L2D_CACHE.value]
        if Stat.L2D_CACHE_MISS.value in data and Stat.L2D_CACHE.value in data
        else None)
    result[Stat.LL_CACHE_MISS_RATE.value] = (
        data[Stat.LL_CACHE_MISS.value] / data[Stat.LL_CACHE.value])

    return result