| #!/usr/bin/env python3 | |
| # | |
| #-------------------------------------------------------------------- | |
| # Parse the SPICE model files from sky130 and use the (commented out) | |
| # statistics block to generate the correct monte carlo expressions | |
| # in the parameters for process variation. | |
| #-------------------------------------------------------------------- | |
| import os | |
| import re | |
| import sys | |
| # Sort the process parameters so that names that are subsets of other | |
| # names appear later in the list. | |
| def getKey(item): | |
| return item[0] | |
| pr_switch_param = 'MC_PR_SWITCH' | |
| options = [] | |
| arguments = [] | |
| for item in sys.argv[1:]: | |
| if item.find('-', 0) == 0: | |
| options.append(item[1:]) | |
| else: | |
| arguments.append(item) | |
| variant = 'sky130A' | |
| walkpath = variant + '/libs.ref/sky130_fd_pr/spice' | |
| if len(options) > 0: | |
| for option in options: | |
| if option.startswith('variant'): | |
| variant = option.split('=')[1] | |
| walkpath = variant + '/libs.ref/sky130_fd_pr/spice' | |
| elif len(arguments) > 0: | |
| walkpath = arguments[0] | |
| process_params = [] | |
| parmrex = re.compile(r'^\.param[ \t]+') | |
| prrex = re.compile(r'^\*[ \t]*process[ \t]*\{') | |
| endrex = re.compile(r'^\*[ \t]*\}') | |
| filelist = [] | |
| #-------------------------------------------------------------------- | |
| # Step 1. Gather variables | |
| #-------------------------------------------------------------------- | |
| for dirpath, dirnames, filenames in os.walk(walkpath): | |
| for filename in filenames: | |
| if os.path.splitext(filename)[1] == '.spice': | |
| infile_name = os.path.join(dirpath, filename) | |
| filelist.append(infile_name) | |
| infile = open(infile_name, 'r') | |
| state = 'before_process' | |
| line_number = 0 | |
| replaced_something = False | |
| for line in infile: | |
| line_number += 1 | |
| if state == 'before_process': | |
| pmatch = prrex.match(line) | |
| if pmatch: | |
| state = 'in_process' | |
| elif state == 'in_process': | |
| ematch = endrex.match(line) | |
| if ematch: | |
| state = 'after_process' | |
| else: | |
| # Make sure all "A = B" syntax closes up around the equal sign. | |
| newline = re.sub(r'[ \t]*=[ \t]*', '=', line) | |
| tokens = newline.split() | |
| if 'vary' in tokens: | |
| if ('dist=gauss' in tokens) or ('gauss' in tokens): | |
| gtype = 'A' | |
| process_param = tokens[2] | |
| for token in tokens[3:]: | |
| gparam = token.split('=') | |
| if len(gparam) == 2: | |
| if gparam[0] == 'std': | |
| std_dev = float(gparam[1]) | |
| elif gparam[0] == 'percent' and gparam[1] == 'yes': | |
| gtype = '' | |
| if gtype == '': | |
| # Convert percentage to a fraction | |
| std_dev = std_dev / 100 | |
| repltext = ' + {}*' + gtype + 'GAUSS(0,{!s},1)' | |
| replacement = repltext.format(pr_switch_param, std_dev) | |
| process_params.append((process_param, replacement)) | |
| elif ('dist=lnorm' in tokens) or ('lnorm' in tokens): | |
| process_param = tokens[2] | |
| for token in tokens[3:]: | |
| gparam = token.split('=') | |
| if len(gparam) == 2: | |
| if gparam[0] == 'std': | |
| std_dev = float(gparam[1]) | |
| replacement = ' + {}*EXP(AGAUSS(0,{!s},1))'.format(pr_switch_param, std_dev) | |
| process_params.append((process_param, replacement)) | |
| infile.close() | |
| process_params.sort(reverse=True, key=getKey) | |
| print('') | |
| print('Process parameters found:') | |
| for (process_param, addendum) in process_params: | |
| print(process_param + ' :' + addendum) | |
| print('') | |
| #-------------------------------------------------------------------- | |
| # Step 2. Make replacements | |
| #-------------------------------------------------------------------- | |
| for infile_name in filelist: | |
| filepath = os.path.split(infile_name)[0] | |
| filename = os.path.split(infile_name)[1] | |
| fileroot = os.path.splitext(filename)[0] | |
| outfile_name = os.path.join(filepath, fileroot + '_temp') | |
| infile = open(infile_name, 'r') | |
| outfile = open(outfile_name, 'w') | |
| state = 'before_process' | |
| line_number = 0 | |
| replaced_something = False | |
| for line in infile: | |
| line_number += 1 | |
| if state == 'before_process': | |
| pmatch = prrex.match(line) | |
| if pmatch: | |
| state = 'in_process' | |
| outfile.write(line) | |
| else: | |
| pmatch = parmrex.match(line) | |
| if pmatch: | |
| newline = line | |
| for (param, replacement) in process_params: | |
| if ' ' + param + ' ' in newline: | |
| newline = newline.strip('\n') + replacement + '\n' | |
| print(" Line {}: Found process parameter '{}' and appended '{}'.".format(line_number, param, replacement)) | |
| replaced_something = True | |
| outfile.write(newline) | |
| else: | |
| outfile.write(line) | |
| elif state == 'in_process': | |
| outfile.write(line) | |
| ematch = endrex.match(line) | |
| if ematch: | |
| state = 'after_process' | |
| elif state == 'after_process': | |
| outfile.write(line) | |
| infile.close() | |
| outfile.close() | |
| if replaced_something: | |
| # print("Something was replaced in '{}', backed up original file" | |
| # + " and replaced with processed one.".format(infile_name)) | |
| print("Something was replaced in '{}'".format(infile_name)) | |
| # os.rename(infile_name, infile_name + '.orig') | |
| os.rename(outfile_name, infile_name) | |
| else: | |
| print("Nothing was replaced in '{}'.".format(infile_name)) | |
| os.remove(outfile_name) | |