correlateRC.py

#!/usr/bin/env python3

# Read grt/rcx capacitance/wire length files for multiple designs and
# use linear regression to fit layer capacitances to rcx net capacitances.
# Use ORFS 'make write_net_rc' to write cap files.

import os
from sys import exit
from collections import defaultdict

import collections
import argparse
import re
import numpy as np
from sklearn.linear_model import LinearRegression
import matplotlib.pyplot as plt

# Parse and validate arguments
# =============================================================================

def parse_args():
  parser = argparse.ArgumentParser(
      description='Determines layer capacitance and resistance values for OpenROAD flow designs')
  parser.add_argument('-cap_unit',
                      required=False,
                      default='pf',
                      help='capacitance unit ff|pf')
  parser.add_argument('-res_unit',
                      required=False,
                      default='kohm',
                      help='resistance unit ohm|kohm')
  parser.add_argument('-plot_cap',
                      required=False,
                      action='store_true',
                      default=False,
                      help='Plot grt/rcx capacitance differences')
  parser.add_argument('-plot_res',
                      required=False,
                      action='store_true',
                      default=False,
                      help='Plot grt/rcx resistance differences')
  parser.add_argument('rc_file',
                      nargs='+',
                      help='rc csv file written by make compare_rc')
  args = parser.parse_args()

  return args

args = parse_args()

# kohm/ff nangate45, asap7
# kohm/pf sky130hd, sky130hs

res_unit = args.res_unit
if res_unit == 'ohm':
  res_scale = 1
elif res_unit == 'kohm':
  res_scale = 1e+3
else:
  print('unknown resistance unit')
  exit

cap_unit = args.cap_unit
if cap_unit == 'ff':
  cap_scale = 1e-15
elif cap_unit == 'pf':
  cap_scale = 1e-12
else:
  print('unknown capacitance unit')
  exit

def makeDict():
  return collections.defaultdict(makeDict)

data = makeDict()

# Parse the cap CSV file generated by compare_rc_script.tcl
for rc_file in args.rc_file:
  design = rc_file
  print("reading", design)
  with open(rc_file) as f:
    nonGrtNets = 0
    for line in f:
      tokens = line.strip().split(',')
      netName = tokens[0]
      gpl_res = float(tokens[1])
      gpl_cap = float(tokens[2])
      grt_res = float(tokens[3])
      grt_cap = float(tokens[4])
      rcx_res = float(tokens[5])
      rcx_cap = float(tokens[6])

      data[design][netName]["gpl_res"] = gpl_res
      data[design][netName]["gpl_cap"] = gpl_cap
      data[design][netName]["grt_res"] = grt_res
      data[design][netName]["grt_cap"] = grt_cap
      data[design][netName]["rcx_res"] = rcx_res
      data[design][netName]["rcx_cap"] = rcx_cap

      layer_lengths = []
      layer_names = []
      wire_length = 0.0
      for i in range(7, len(tokens), 2):
        layer_names.append(tokens[i])
        layer_length = float(tokens[i + 1])
        layer_lengths.append(layer_length)
        wire_length += layer_length
      data[design][netName]["layer_lengths"] = layer_lengths
      data[design][netName]["wire_length"] = wire_length

################################################################

if args.plot_cap:
  # Compare the GRT cap estimate vs. OpenRCX SPEF cap

  diff_x = []
  diff_percent_x = []
  for design in data:
    for net in data[design]:
      grt_cap = data[design][net]["grt_cap"]
      rcx_cap = data[design][net]["rcx_cap"]
      diff = grt_cap - rcx_cap
      if abs(diff) > 1e-12:
        print("large discrapancy:", design, net, diff)
      if rcx_cap != 0.0:
        normDiff = (diff / rcx_cap) * 100
        diff_x.append(diff / cap_scale )
        diff_percent_x.append(normDiff)

  # Generate histograms
  num_bins = 200
  fig = plt.figure()
  fig.suptitle("Difference between GRT est. Cap and RCX Cap")
  plt.subplot(2,2,1)
  plt.hist(diff_x, num_bins, facecolor='blue', alpha=0.5)
  plt.ylabel("# Nets")
  plt.xlabel("Capacitance ({})\n\nMean: {:.3f}{}\nStd. dev: {:.3f}fF".format(cap_unit, np.mean(diff_x), cap_unit, np.std(diff_x)))

  plt.subplot(2,2,2)
  plt.hist(diff_percent_x, num_bins, range={-1000, 1000}, facecolor='blue', alpha=0.5)
  plt.ylabel("# Nets")
  plt.xlabel("%\n\nMean: {:.3f}%\nStd. dev: {:.3f}%".format(np.mean(diff_percent_x),
                                                            np.std(diff_percent_x)))
  plt.show()

################################################################

if args.plot_res:
  # Compare the GRT res estimate vs. OpenRCX SPEF res

  diff_x = []
  diff_percent_x = []
  for design in data:
    for net in data[design]:
      grt_res = data[design][net]["grt_res"]
      rcx_res = data[design][net]["rcx_res"]
      if grt_res > 0 and rcx_res > 0:
        diff = grt_res - rcx_res
        if abs(diff) > 1e+3:
          print("large discrapancy:", design, net, diff)
        if rcx_res != 0.0:
          normDiff = (diff / rcx_res) * 100
          diff_x.append(diff / res_scale )
          diff_percent_x.append(normDiff)

  # Generate histograms
  num_bins = 200
  fig = plt.figure()
  fig.suptitle("Difference between GRT est. Res and RCX Res")
  plt.subplot(2,2,1)
  plt.hist(diff_x, num_bins, facecolor='blue', alpha=0.5)
  plt.ylabel("# Nets")
  plt.xlabel("Resacitance ({})\n\nMean: {:.3f}{}\nStd. dev: {:.3f}fF".format(res_unit, np.mean(diff_x), res_unit, np.std(diff_x)))

  plt.subplot(2,2,2)
  plt.hist(diff_percent_x, num_bins, range={-1000, 1000}, facecolor='blue', alpha=0.5)
  plt.ylabel("# Nets")
  plt.xlabel("%\n\nMean: {:.3f}%\nStd. dev: {:.3f}%".format(np.mean(diff_percent_x),
                                                            np.std(diff_percent_x)))
  plt.show()

################################################################

# Use linear regression to find updated layer resistances.

x = []
y = []
for design in data:
  for net in data[design]:
    rcx_res = data[design][net]["rcx_res"]
    if rcx_res > 0:
      layer_lengths = data[design][net]["layer_lengths"]
      x.append(layer_lengths)
      y.append(rcx_res)
    
x = np.array(x)
y = np.array(y)

res_model = LinearRegression(fit_intercept=False).fit(x, y)
r_sq = res_model.score(x, y)
print('Resistance coefficient of determination: {:.4f}'.format(r_sq))

################################################################

# Use linear regression to find updated layer capacitances.

x = []
y = []
for design in data:
  for net in data[design]:
    layer_lengths = data[design][net]["layer_lengths"]
    x.append(layer_lengths)
    y.append(data[design][net]["rcx_cap"])
    
x = np.array(x)
y = np.array(y)

cap_model = LinearRegression(fit_intercept=False).fit(x, y)
r_sq = cap_model.score(x, y)
print('Capacitance coefficient of determination: {:.4f}'.format(r_sq))

print('Updated layer resistance {}/um capacitance {}/um'.format(res_unit, cap_unit))
for layer, res_coeff, cap_coeff in zip(layer_names, res_model.coef_, cap_model.coef_):
  if res_coeff > 0.0 or cap_coeff > 0.0:
    print("set_layer_rc -layer {} -resistance {:.5E} -capacitance {:.5E}".format(layer, res_coeff / res_scale, cap_coeff / cap_scale))

################################################################

x = []
y = []
for design in data:
  for net in data[design]:
    wire_res = data[design][net]["rcx_res"]
    wire_length = data[design][net]["wire_length"]
    if wire_res != 0.0:
      x.append([wire_length])
      y.append(wire_res)

x = np.array(x)
y = np.array(y)

wire_res_model = LinearRegression(fit_intercept=False).fit(x, y)
wire_res = wire_res_model.coef_[0]

x = []
y = []
for design in data:
  for net in data[design]:
    wire_length = data[design][net]["wire_length"]
    x.append([wire_length])
    y.append(data[design][net]["rcx_cap"])

x = np.array(x)
y = np.array(y)

wire_cap_model = LinearRegression(fit_intercept=False).fit(x, y)
wire_cap = wire_cap_model.coef_[0]

print("set_wire_rc -resistance {:.5E} -capacitance {:.5E}".format(wire_res / res_scale, wire_cap / cap_scale))