基于python机器学习的Soc静态压降预测源码+项目说明.zip

import os, re, bisect, gzip import numpy as np import pandas as pd from src.util import save import matplotlib import matplotlib.pyplot as plt import seaborn as sns class Paraser: def __init__(self, root_dir, arg, save_name): self.root_dir = root_dir self.save_name = save_name self.save_path = arg.save_path self.plot = arg.plot self.final_test = arg.final_test self.coordinate_x = [] self.coordinate_y = [] self.irdrop_map = None self.total_power_map = None # add new feature # leakage_power switching_power internal_power self.leakage_powe_map = None self.switching_power_map = None self.internal_power_map = None #//// self.eff_res_VDD_map = None self.eff_res_VSS_map = None self.instance_count = None self.instance_IR_drop = None def get_IR_drop_features(self): max_size = (1000, 1000) self.irdrop_map = np.zeros(max_size) self.instance_count = np.zeros(max_size) self.instance_IR_drop = np.empty(max_size, dtype=object) for i in np.ndindex(max_size): self.instance_IR_drop[i] = [] self.instance_name = np.empty(max_size, dtype=object) for i in np.ndindex(max_size): self.instance_name[i] = [] self.total_power_map = np.zeros(max_size) self.leakage_powe_map= np.zeros(max_size) self.switching_power_map = np.zeros(max_size) self.internal_power_map = np.zeros(max_size) self.eff_res_VDD_map = np.zeros(max_size) self.eff_res_VSS_map = np.zeros(max_size) self.coordinate_x = np.arange(0,max_size[0],1.44) self.coordinate_y = np.arange(0,max_size[1],1.152) # based on the row height from LEF # try: # if 'nvdla' in self.root_dir: # data_power = pd.read_csv(os.path.join(self.root_dir, 'NV_nvdla.inst.power.rpt.gz'),sep='\s+',header=1, compression='gzip') # elif 'Vortex' in self.root_dir: # data_power = pd.read_csv(os.path.join(self.root_dir, 'Vortex.inst.power.rpt.gz'),sep='\s+',header=1, compression='gzip') # else: # data_power = pd.read_csv(os.path.join(self.root_dir, 'pulpino_top.inst.power.rpt.gz'),sep='\s+',header=1, compression='gzip') # data_r = pd.read_csv(os.path.join(self.root_dir, 'eff_res.rpt.gz'),sep='\s+', low_memory=False, compression='gzip') # if not self.final_test: # data_ir = pd.read_csv(os.path.join(self.root_dir, 'static_ir.gz'),sep='\s+', compression='gzip') # except Exception as e: # print('one of the report not exists') # return 0 try: if 'nvdla' in self.root_dir: data_power = pd.read_csv(os.path.join(self.root_dir, 'NV_nvdla.inst.power.rpt'),sep='\s+',header=1) elif 'Vortex' in self.root_dir: data_power = pd.read_csv(os.path.join(self.root_dir, 'Vortex.inst.power.rpt'),sep='\s+',header=1) else: data_power = pd.read_csv(os.path.join(self.root_dir, 'pulpino_top.inst.power.rpt'),sep='\s+',header=1) data_r = pd.read_csv(os.path.join(self.root_dir, 'eff_res.rpt'),sep='\s+', low_memory=False) if not self.final_test: data_ir = pd.read_csv(os.path.join(self.root_dir, 'static_ir'),sep='\s+') except Exception as e: print('one of the report not exists') return 0 # try: # if 'nvdla' in self.root_dir: # data_power = pd.read_csv(os.path.join(self.root_dir, 'NV_nvdla.inst.power.rpt'),sep='\s+',header=1) # else: # # '\s+' 表示使用空格或多个空格作为分隔符, header=1 指定了CSV文件的第一行（索引从0开始）作为列名 # data_power = pd.read_csv(os.path.join(self.root_dir, 'pulpino_top.inst.power.rpt'),sep='\s+',header=1) # # low_memory=False会加快读取速度，占用更多内存; low_memory=True会降低读取速度，占用更少内存 # data_r = pd.read_csv(os.path.join(self.root_dir, 'eff_res.rpt'),sep='\s+', low_memory=False) # data_ir = pd.read_csv(os.path.join(self.root_dir, 'static_ir'),sep='\s+') max_x = 0 max_y = 0 # parse inst.power.rpt leakage_power switching_power internal_power total_power = data_power['total_power'] leakage_power = data_power['leakage_power'] switching_power = data_power['switching_power'] internal_power = data_power['internal_power'] bbox = data_power['bbox'] name = data_power['*inst_name'] if self.final_test: for i,j1,j2,j3,j4,k in zip(bbox, total_power,leakage_power,switching_power,internal_power, name): x1, y1, x2, y2 = i[1:-1].split(',') x = (float(x1)+float(x2))/2 y = (float(y1)+float(y2))/2 gcell_x = bisect.bisect_left(self.coordinate_x, float(x)-10) gcell_y = bisect.bisect_left(self.coordinate_y, float(y)-10) if gcell_x > max_x: max_x = gcell_x if gcell_y > max_y: max_y = gcell_y self.total_power_map[gcell_x, gcell_y] += j1 self.leakage_powe_map[gcell_x, gcell_y] +=j2 self.switching_power_map[gcell_x, gcell_y] +=j3 self.internal_power_map[gcell_x, gcell_y] +=j4 self.instance_name[gcell_x, gcell_y].append(k) self.instance_count[gcell_x, gcell_y] += 1 self.total_power_map = self.total_power_map[0:max_x+1,0:max_y+1] self.leakage_powe_map = self.leakage_powe_map[0:max_x+1,0:max_y+1] self.switching_power_map = self.switching_power_map[0:max_x+1,0:max_y+1] self.internal_power_map = self.internal_power_map[0:max_x+1,0:max_y+1] save(self.save_path, 'features/instance_count_from_power_rpt', self.save_name, self.instance_count) self.instance_name = np.concatenate(self.instance_name.ravel()) # 以npz形式保存以节省空间 instance_name_save_path = os.path.join(self.save_path, 'features/instance_name_from_power_rpt', self.save_name) if not os.path.exists(os.path.dirname(instance_name_save_path)): os.makedirs(os.path.dirname(instance_name_save_path)) np.savez_compressed(instance_name_save_path, instance_name=self.instance_name) else: for i,j1,j2,j3,j4 in zip(bbox, total_power,leakage_power,switching_power,internal_power): x1, y1, x2, y2 = i[1:-1].split(',') x = (float(x1)+float(x2))/2 y = (float(y1)+float(y2))/2 gcell_x = bisect.bisect_left(self.coordinate_x, float(x)-10) gcell_y = bisect.bisect_left(self.coordinate_y, float(y)-10) if gcell_x > max_x: max_x = gcell_x if gcell_y > max_y: max_y = gcell_y self.total_power_map[gcell_x, gcell_y] += j1 self.leakage_powe_map[gcell_x, gcell_y] +=j2 self.switching_power_map[gcell_x, gcell_y] +=j3 self.internal_power_map[gcell_x, gcell_y] +=j4 self.total_power_map = self.total_power_map[0:max_x+1,0:max_y+1] self.leakage_powe_map = self.leakage_powe_map[0:max_x+1,0:max_y+1] self.switching_power_map = self.switching_power_map[0:max_x+1,0:max_y+1] self.internal_power_map = self.internal_power_map[0:max_x+1,0:max_y+1] save(self.save_path, 'features/total_power', self.save_name, self.total_power_map) save(self.save_path, 'features/leakage_power', self.save_name, self.leakage_powe_map)