Function parse

deps/v8/tools/regexp-sequences.py:19–92

(file, seqlen)

import collections

def parse(file, seqlen):
  # example:
  # pc = 00, sp = 0, curpos = 0, curchar = 0000000a ..., PushBacktrack, label: e4
  rx = re.compile(
      r'pc = (?P<pc>[0-9a-f]+), sp = (?P<sp>\d+), '
      r'curpos = (?P<curpos>\d+), curchar = (?P<char_hex>[0-9a-f]+) '
      r'(:?\.|\()(?P<char>\.|\w)(:?\.|\)), (?P<bc>\w+), .*')
  total = 0

  # stats[len][seq_tuple] = {'count': 0, 'loops': 0}
  # Using tuple as key for easier rotation
  stats = [{} for _ in range(seqlen)]

  # last[len] = deque of (bc, pc_int)
  last = [collections.deque(maxlen=i + 1) for i in range(seqlen)]

  with open(file) as f:
    l = f.readline()
    while l:
      l = l.strip()
      if l.startswith("Start bytecode interpreter"):
        for i in range(seqlen):
          last[i].clear()

      match = rx.search(l)
      if match:
        total += 1
        bc = match.group('bc')
        pc = int(match.group('pc'), 16)

        current_node = (bc, pc)

        for i in range(seqlen):
          deque = last[i]

          if len(deque) == i + 1:
            # Canonicalize based on PC: find the rotation that starts with the
            # lowest PC.  This aligns loops to their header.
            min_pc = float('inf')
            rotation_start = 0

            # Find the index with the minimum PC
            for idx, item in enumerate(deque):
              if item[1] < min_pc:
                min_pc = item[1]
                rotation_start = idx
              elif item[1] == min_pc:
                # Tie-break. If the same sequence appears twice (A->B->A->B),
                # we prefer the first.
                pass

            # Rotate to start at rotation_start
            canonical_seq = []
            for k in range(i + 1):
              idx = (rotation_start + k) % (i + 1)
              canonical_seq.append(deque[idx][0])

            seq_key = tuple(canonical_seq)