openpower/sv/insn-histogram.py

   1 #! /bin/env python3
   2 # see https://bugs.libre-soc.org/show_bug.cgi?id=532
   3
   4 # Print a per-opcode histogram for ppc asm.
   5
   6 # Copyright 2020 Alexandre Oliva
   7 # Copyright 2020 Luke Kenneth Casson Leighton
   8
   9 # This script is free software; you can redistribute it and/or modify
  10 # it under the terms of the GNU General Public License as published by
  11 # the Free Software Foundation; either version 3, or (at your option)
  12 # any later version.
  13
  14 # This script is distributed in the hope that it will be useful, but
  15 # WITHOUT ANY WARRANTY; without even the implied warranty of
  16 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17 # General Public License for more details.
  18
  19 # You should have received a copy of the GNU General Public License
  20 # along with this script; see the file COPYING3.  If not see
  21 # <http://www.gnu.org/licenses/>.
  22
  23 # Originally by Alexandre Oliva <oliva@gnu.org>.
  24
  25
  26 # Feed this script the output of:
  27 # powerpc64le-gnu-objdump -d -M raw --no-show-raw-insn ppc-prog
  28
  29 # It will print the occurrence count of each opcode,
  30 # and under it, indented by one character,
  31 # the occurrence count of each operand.
  32
  33 # Registers used as operands or as base addresses are counted
  34 # separately; immediates and offsets are grouped per bit length;
  35 # branch target offsets are grouped by range bit length.
  36
  37 import sys
  38 import re
  39
  40 insn = re.compile('\s+(?P<addr>[0-9a-f]+):\s+(?P<opcode>[^ \n]+) *(?P<operands>.*)[\n]?')
  41
  42 opkind = re.compile('(?P<immediate>-?[0-9]+)|(?P<branch>[0-9a-f]+)(?: <.*>)?|(?P<offset>-?[0-9]+)\((?P<basereg>r[0-9]+)\)')
  43
  44 histogram = {}
  45
  46 def count(ops, op):
  47     match = opkind.fullmatch(op)
  48
  49     if match is None:
  50         op = op
  51     elif match['immediate'] is not None:
  52         op = "%i-bit" % int (op).bit_length ()
  53     elif match['branch'] is not None:
  54         op = "%i-bit range" % (int (match['branch'], 16) -
  55                                int(addr, 16)).bit_length ()
  56     elif match['offset'] is not None:
  57         count(ops, match['offset'])
  58         op = match['basereg']
  59     else:
  60         raise "unrecognized operand kind"
  61
  62     if op not in ops:
  63         ops[op] = 1
  64     else:
  65         ops[op] += 1
  66
  67 for line in sys.stdin:
  68     match = insn.fullmatch(line)
  69     if match is None:
  70         continue
  71
  72     addr = match['addr']
  73     opcode = match['opcode']
  74     operands = match['operands']
  75
  76     if opcode not in histogram:
  77         ops = {}
  78         histogram[opcode] = [1,ops]
  79     else:
  80         histogram[opcode][0] += 1
  81         ops = histogram[opcode][1]
  82
  83     if len(operands) > 0:
  84         for operand in operands.split(','):
  85             count(ops, operand)
  86
  87 intregcounts = {}
  88 crregcounts = {}
  89
  90 # for each instruction print out a regcount.  first, sort by instr count
  91 hist = list(histogram.items())
  92 hist.sort(key = (lambda x : x[1][0]))
  93
  94 # now print each instruction and its register usage
  95 for x in hist:
  96     print('%6i %s:' % (x[1][0], x[0]))
  97     ops = list(x[1][1].items())
  98     ops.sort(key = (lambda x : x[1]))
  99
 100     # split out "-bit" from "-bit range" from "regs"
 101
 102     # first "rNs" or "anything-weird"
 103     for x in ops:
 104         if '-bit' in x[0]:
 105             continue
 106         if x[0].startswith('cr'):
 107             continue
 108         print('\t%6i %s' % (x[1], x[0]))
 109         # total up integer register counts
 110         if not x[0].startswith('r'):
 111             continue
 112         if not x[0] in intregcounts:
 113             intregcounts[x[0]] = 0
 114         intregcounts[x[0]] += x[1]
 115     print()
 116     # TODO: FP regs.
 117     # ...
 118
 119     # now Condition Registers
 120     for x in ops:
 121         if '-bit' in x[0]:
 122             continue
 123         if x[0].startswith('cr'):
 124             print('\t%6i %s' % (x[1], x[0]))
 125             if not x[0] in crregcounts:
 126                 crregcounts[x[0]] = 0
 127             crregcounts[x[0]] += x[1]
 128     print()
 129
 130     # now "N-bit immediates"
 131     for x in ops:
 132         if x[0].endswith('-bit'):
 133             print('\t%6i %s' % (x[1], x[0]))
 134     print()
 135
 136     # finally "bit-range" immediates
 137     for x in ops:
 138         if '-bit range' in x[0]:
 139             print('\t%6i %s' % (x[1], x[0]))
 140     print()
 141
 142 # print out regs usage totals (TODO: FP)
 143 for regcounts in [intregcounts, crregcounts]:
 144     regnums = list(regcounts.items())
 145     regnums.sort(key = (lambda x : x[1]))
 146     for x in regnums:
 147         print('%6i %s' % (x[1], x[0]))
 148     print()