Capstone disassembly/disassembler framework: Core (Arm, Arm64, BPF, EVM, M68K, M680X, MOS65xx, Mips, PPC, RISCV, Sparc, SystemZ, TMS320C64x, Web Assembly, X86, X86_64, XCore) + bindings. (bloaty 依赖)
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

124 lines
3.4 KiB

#!/usr/bin/python
# Simple benchmark for Capstone by disassembling random code. By Nguyen Anh Quynh, 2014
# Syntax:
# ./suite/benchmark.py --> Benchmark all archs
# ./suite/benchmark.py x86 --> Benchmark all X86 (all 16bit, 32bit, 64bit)
# ./suite/benchmark.py x86-32 --> Benchmark X86-32 arch only
# ./suite/benchmark.py arm --> Benchmark all ARM (arm, thumb)
# ./suite/benchmark.py aarch64 --> Benchmark ARM-64
# ./suite/benchmark.py mips --> Benchmark all Mips (32bit, 64bit)
# ./suite/benchmark.py ppc --> Benchmark PPC
from capstone import *
from time import time
from random import randint
import sys
# file providing code to disassemble
FILE = '/usr/bin/python'
all_tests = (
(CS_ARCH_X86, CS_MODE_16, "X86-16bit (Intel syntax)", 0),
(CS_ARCH_X86, CS_MODE_32, "X86-32bit (ATT syntax)", CS_OPT_SYNTAX_ATT),
(CS_ARCH_X86, CS_MODE_32, "X86-32 (Intel syntax)", 0),
(CS_ARCH_X86, CS_MODE_64, "X86-64 (Intel syntax)", 0),
(CS_ARCH_ARM, CS_MODE_ARM, "ARM", 0),
(CS_ARCH_ARM, CS_MODE_THUMB, "THUMB (ARM)", 0),
(CS_ARCH_MIPS, CS_MODE_32 + CS_MODE_BIG_ENDIAN, "MIPS-32 (Big-endian)", 0),
(CS_ARCH_MIPS, CS_MODE_64 + CS_MODE_LITTLE_ENDIAN, "MIPS-64-EL (Little-endian)", 0),
(CS_ARCH_ARM64, CS_MODE_ARM, "ARM-64 (AArch64)", 0),
(CS_ARCH_PPC, CS_MODE_BIG_ENDIAN, "PPC-64", 0),
(CS_ARCH_PPC, CS_MODE_BIG_ENDIAN, "PPC-64, print register with number only", CS_OPT_SYNTAX_NOREGNAME),
)
# for debugging
def to_hex(s):
return " ".join("0x" + "{0:x}".format(ord(c)).zfill(2) for c in s) # <-- Python 3 is OK
def get_code(f, size):
code = f.read(size)
if len(code) != size: # reached end-of-file?
# then reset file position to begin-of-file
f.seek(0)
code = f.read(size)
return code
def cs(md, code):
insns = md.disasm(code, 0)
# uncomment below line to speed up this function 200 times!
# return
for i in insns:
if i.address == 0x100000:
print i
def cs_lite(md, code):
insns = md.disasm_lite(code, 0)
for (addr, size, mnem, ops) in insns:
if addr == 0x100000:
print i
cfile = open(FILE)
for (arch, mode, comment, syntax) in all_tests:
try:
request = sys.argv[1]
if not request in comment.lower():
continue
except:
pass
print("Platform: %s" %comment)
try:
md = Cs(arch, mode)
#md.detail = True
if syntax != 0:
md.syntax = syntax
# warm up few times
cfile.seek(0)
for i in xrange(3):
code = get_code(cfile, 128)
#print to_hex(code)
#print
cs(md, code)
# start real benchmark
c_t = 0
for i in xrange(50000):
code = get_code(cfile, 128)
#print to_hex(code)
#print
t1 = time()
cs(md, code)
c_t += time() - t1
print "Benchmark - full obj:", c_t, "seconds"
print
cfile.seek(0)
c_t = 0
for i in xrange(50000):
code = get_code(cfile, 128)
#print to_hex(code)
#print
t1 = time()
cs_lite(md, code)
c_t += time() - t1
print "Benchmark - lite:", c_t, "seconds"
print
except CsError as e:
print("ERROR: %s" %e)