Consideration of addressing modes

[cod3monk]

Is it currently possible to differentiate between simple (base+offset) and complex (base+offset+index*scale) addressing as memory parameters in CSV the file syntax?

[JanLJL]

It is not yet possible to differentiate between different addressing modes in the CSV.
Nevertheless, when creating a parameter, every MemAddr object is checked for the addressing mode and contains the attributes

offset, base, index, scale

which are currently either True or None. This is done during the init of a MemAddr in param.py:

OSACA/osaca/param.py

Lines 27 to 42 in 961364c

	name = name.strip(', \t')
	self.offset = None
	self.base = None
	self.index = None
	self.scale = None
	m = re.match(r'(?P<offset>[x0-9a-fA-F]*)\((?P<base>[^,\)]+)(?:,\s*(?P<index>[^,\)]+)'
	r'(?:,\s*(?P<scale>[^,\)]+))?)?\)', name)
	if not m:
	raise ValueError('Type not supported: {!r}'.format(name))
	self.offset = m.group('offset') or None
	self.base = m.group('base') or None
	self.index = m.group('index') or None
	self.scale = m.group('scale') or None

Some examples:

>>> mem = MemAddr('8(%rsp,%rax,8)')
>>> print(mem)
MEM(offset(base, index, scale))
>>> mem = MemAddr('(%rsp,%rax,8)')
>>> print(mem)
MEM((base, index, scale))
>>> mem = MemAddr('(%rsp)')
>>> print(mem)
MEM((base))

[cod3monk]

Okay, the issue is that uops.info also has no notion of different addressing modes and seems to use simple addresses during their benchmarks and therefore consider Port7 as a viable target for all address generation uops. Not yet sure how this should be handled in future.

[cod3monk]

solved in new version (dev releases are available)