def _genop_vec_getarrayitem(self, op, arglocs, resloc):
# considers item scale (raw_load does not)
base_loc, ofs_loc, size_loc, ofs, integer_loc, aligned_loc = arglocs
scale = get_scale(size_loc.value)
src_addr = addr_add(base_loc, ofs_loc, ofs.value, scale)
self._vec_load(resloc, src_addr, integer_loc.value,
size_loc.value, aligned_loc.value)
def _genop_discard_vec_setarrayitem(self, op, arglocs):
# considers item scale (raw_store does not)
base_loc, ofs_loc, value_loc, size_loc, baseofs, integer_loc, aligned_loc = arglocs
scale = get_scale(size_loc.value)
dest_loc = addr_add(base_loc, ofs_loc, baseofs.value, scale)
self._vec_store(dest_loc, value_loc, integer_loc.value,
size_loc.value, aligned_loc.value)
def _genop_vec_getarrayitem(self, op, arglocs, resloc):
# considers item scale (raw_load does not)
base_loc, ofs_loc, size_loc, ofs, integer_loc, aligned_loc = arglocs
scale = get_scale(size_loc.value)
src_addr = addr_add(base_loc, ofs_loc, ofs.value, scale)
self._vec_load(resloc, src_addr, integer_loc.value,
size_loc.value, aligned_loc.value)
def _emit_op_gc_load(self, op, arglocs, regalloc, fcond):
base_loc, ofs_loc, res_loc, nsize_loc = arglocs
nsize = nsize_loc.value
signed = nsize < 0
scale = get_scale(abs(nsize))
self._load_from_mem(res_loc, base_loc, ofs_loc, imm(scale), signed, fcond)
return fcond
def _genop_discard_vec_setarrayitem(self, op, arglocs):
# considers item scale (raw_store does not)
base_loc, ofs_loc, value_loc, size_loc, baseofs, integer_loc, aligned_loc = arglocs
scale = get_scale(size_loc.value)
dest_loc = addr_add(base_loc, ofs_loc, baseofs.value, scale)
self._vec_store(dest_loc, value_loc, integer_loc.value,
size_loc.value, aligned_loc.value)
def emit_op_gc_store_indexed(self, op, arglocs):
value_loc, base_loc, index_loc, size_loc, ofs_loc = arglocs
assert index_loc.is_core_reg()
# add the base offset
if ofs_loc.value > 0:
self.mc.ADD_ri(r.ip0.value, index_loc.value, ofs_loc.value)
index_loc = r.ip0
scale = get_scale(size_loc.value)
self._write_to_mem(value_loc, base_loc, index_loc, scale)
def threadlocalref_get(self, op, arglocs):
res_loc, = arglocs
ofs_loc = self.imm(op.getarg(1).getint())
calldescr = op.getdescr()
ofs = self.saved_threadlocal_addr
self.load_reg(self.mc, res_loc, r.sp, ofs)
scale = get_scale(calldescr.get_result_size())
signed = (calldescr.is_result_signed() != 0)
self._load_from_mem(res_loc, res_loc, ofs_loc, scale, signed)
def emit_op_gc_store_indexed(self, op, arglocs, regalloc, fcond):
value_loc, base_loc, index_loc, size_loc, ofs_loc = arglocs
assert index_loc.is_core_reg()
# add the base offset
if ofs_loc.value > 0:
self.mc.ADD_ri(r.ip.value, index_loc.value, imm=ofs_loc.value)
index_loc = r.ip
scale = get_scale(size_loc.value)
self._write_to_mem(value_loc, base_loc, index_loc, imm(scale), fcond)
return fcond
def emit_opx_threadlocalref_get(self, op, arglocs, regalloc, fcond):
ofs_loc, size_loc, sign_loc, res_loc = arglocs
assert ofs_loc.is_imm()
assert size_loc.is_imm()
assert sign_loc.is_imm()
ofs = self.saved_threadlocal_addr
self.load_reg(self.mc, res_loc, r.sp, ofs)
scale = get_scale(size_loc.value)
signed = sign_loc.value != 0
self._load_from_mem(res_loc, res_loc, ofs_loc, imm(scale), signed, fcond)
return fcond
def _emit_op_gc_load_indexed(self, op, arglocs):
res_loc, base_loc, index_loc, nsize_loc, ofs_loc = arglocs
assert index_loc.is_core_reg()
nsize = nsize_loc.value
signed = (nsize < 0)
# add the base offset
if ofs_loc.value > 0:
self.mc.ADD_ri(r.ip0.value, index_loc.value, ofs_loc.value)
index_loc = r.ip0
#
scale = get_scale(abs(nsize))
self._load_from_mem(res_loc, base_loc, index_loc, scale, signed)
def _emit_op_gc_load_indexed(self, op, arglocs, regalloc, fcond):
res_loc, base_loc, index_loc, nsize_loc, ofs_loc = arglocs
assert index_loc.is_core_reg()
nsize = nsize_loc.value
signed = nsize < 0
# add the base offset
if ofs_loc.value > 0:
self.mc.ADD_ri(r.ip.value, index_loc.value, imm=ofs_loc.value)
index_loc = r.ip
#
scale = get_scale(abs(nsize))
self._load_from_mem(res_loc, base_loc, index_loc, imm(scale), signed, fcond)
return fcond
def emit_op_setinteriorfield_gc(self, op, arglocs, regalloc, fcond):
(base_loc, index_loc, value_loc, ofs_loc, ofs, itemsize, fieldsize) = arglocs
scale = get_scale(fieldsize.value)
tmploc, save = self.get_tmp_reg([base_loc, index_loc, value_loc, ofs_loc])
assert not save
self.mc.gen_load_int(tmploc.value, itemsize.value)
self.mc.MUL(tmploc.value, index_loc.value, tmploc.value)
if ofs.value > 0:
if ofs_loc.is_imm():
self.mc.ADD_ri(tmploc.value, tmploc.value, ofs_loc.value)
else:
self.mc.ADD_rr(tmploc.value, tmploc.value, ofs_loc.value)
self._write_to_mem(value_loc, base_loc, tmploc, imm(scale), fcond)
return fcond
def emit_op_gc_store_indexed(self, op, arglocs):
value_loc, base_loc, index_loc, size_loc, ofs_loc = arglocs
assert index_loc.is_core_reg()
# add the base offset
if ofs_loc.value != 0:
if check_imm_arg(ofs_loc.value):
self.mc.ADD_ri(r.ip0.value, index_loc.value, ofs_loc.value)
else:
# ofs_loc.value is too large for an ADD_ri
self.load(r.ip0, ofs_loc)
self.mc.ADD_rr(r.ip0.value, r.ip0.value, index_loc.value)
index_loc = r.ip0
scale = get_scale(size_loc.value)
self._write_to_mem(value_loc, base_loc, index_loc, scale)
def _consider_vec_load(self, op):
descr = op.getdescr()
assert isinstance(descr, ArrayDescr)
assert not descr.is_array_of_pointers() and \
not descr.is_array_of_structs()
itemsize, _, _ = unpack_arraydescr(descr)
integer = not (descr.is_array_of_floats() or descr.getconcrete_type() == FLOAT)
args = op.getarglist()
scale = get_scale(op.getarg(2).getint())
ofs = op.getarg(3).getint()
base_loc = self.rm.make_sure_var_in_reg(op.getarg(0), args)
ofs_loc = self.rm.make_sure_var_in_reg(op.getarg(1), args)
result_loc = self.force_allocate_reg(op)
self.perform(op, [base_loc, ofs_loc, imm(itemsize), imm(scale),
imm(ofs), imm(integer)], result_loc)
def _consider_vec_load(self, op):
descr = op.getdescr()
assert isinstance(descr, ArrayDescr)
assert not descr.is_array_of_pointers() and \
not descr.is_array_of_structs()
itemsize, _, _ = unpack_arraydescr(descr)
integer = not (descr.is_array_of_floats() or descr.getconcrete_type() == FLOAT)
args = op.getarglist()
scale = get_scale(op.getarg(2).getint())
ofs = op.getarg(3).getint()
base_loc = self.rm.make_sure_var_in_reg(op.getarg(0), args)
ofs_loc = self.rm.make_sure_var_in_reg(op.getarg(1), args)
result_loc = self.force_allocate_reg(op)
self.perform(op, [base_loc, ofs_loc, imm(itemsize), imm(scale),
imm(ofs), imm(integer)], result_loc)
def flush_vec_cc(self, rev_cond, lhsloc, resloc, size):
# After emitting an instruction that leaves a boolean result in
# a condition code (cc), call this. In the common case, result_loc
# will be set to SPP by the regalloc, which in this case means
# "propagate it between this operation and the next guard by keeping
# it in the cc". In the uncommon case, result_loc is another
# register, and we emit a load from the cc into this register.
if resloc is ebp:
self.guard_success_cc = rev_cond
else:
assert lhsloc is xmm0
maskloc = X86_64_XMM_SCRATCH_REG
self.mc.MOVAPD(maskloc, heap(self.element_ones[get_scale(size)]))
self.mc.PXOR(resloc, resloc)
# note that resloc contains true false for each element by the last compare operation
self.mc.PBLENDVB_xx(resloc.value, maskloc.value)
def _emit_op_gc_load_indexed(self, op, arglocs):
res_loc, base_loc, index_loc, nsize_loc, ofs_loc = arglocs
assert index_loc.is_core_reg()
nsize = nsize_loc.value
signed = (nsize < 0)
# add the base offset
if ofs_loc.value != 0:
if check_imm_arg(ofs_loc.value):
self.mc.ADD_ri(r.ip0.value, index_loc.value, ofs_loc.value)
else:
# ofs_loc.value is too large for an ADD_ri
self.load(r.ip0, ofs_loc)
self.mc.ADD_rr(r.ip0.value, r.ip0.value, index_loc.value)
index_loc = r.ip0
#
scale = get_scale(abs(nsize))
self._load_from_mem(res_loc, base_loc, index_loc, scale, signed)
def flush_vec_cc(self, rev_cond, lhsloc, resloc, size):
# After emitting an instruction that leaves a boolean result in
# a condition code (cc), call this. In the common case, result_loc
# will be set to SPP by the regalloc, which in this case means
# "propagate it between this operation and the next guard by keeping
# it in the cc". In the uncommon case, result_loc is another
# register, and we emit a load from the cc into this register.
if resloc is ebp:
self.guard_success_cc = rev_cond
else:
assert lhsloc is xmm0
maskloc = X86_64_XMM_SCRATCH_REG
self.mc.MOVAPD(maskloc, heap(self.element_ones[get_scale(size)]))
self.mc.PXOR(resloc, resloc)
# note that resloc contains true false for each element by the last compare operation
self.mc.PBLENDVB_xx(resloc.value, maskloc.value)
def consider_vec_store(self, op):
# TODO
descr = op.getdescr()
assert isinstance(descr, ArrayDescr)
assert not descr.is_array_of_pointers() and \
not descr.is_array_of_structs()
itemsize, _, _ = unpack_arraydescr(descr)
args = op.getarglist()
base_loc = self.rm.make_sure_var_in_reg(op.getarg(0), args)
value_loc = self.make_sure_var_in_reg(op.getarg(2), args)
ofs_loc = self.rm.make_sure_var_in_reg(op.getarg(1), args)
scale = get_scale(op.getarg(3).getint())
ofs = op.getarg(4).getint()
integer = not (descr.is_array_of_floats() or \
descr.getconcrete_type() == FLOAT)
self.perform_discard(op, [base_loc, ofs_loc, value_loc, imm(itemsize),
imm(scale), imm(ofs), imm(integer)])
def consider_vec_store(self, op):
# TODO
descr = op.getdescr()
assert isinstance(descr, ArrayDescr)
assert not descr.is_array_of_pointers() and \
not descr.is_array_of_structs()
itemsize, _, _ = unpack_arraydescr(descr)
args = op.getarglist()
base_loc = self.rm.make_sure_var_in_reg(op.getarg(0), args)
value_loc = self.make_sure_var_in_reg(op.getarg(2), args)
ofs_loc = self.rm.make_sure_var_in_reg(op.getarg(1), args)
scale = get_scale(op.getarg(3).getint())
ofs = op.getarg(4).getint()
integer = not (descr.is_array_of_floats() or \
descr.getconcrete_type() == FLOAT)
self.perform_discard(op, [base_loc, ofs_loc, value_loc, imm(itemsize),
imm(scale), imm(ofs), imm(integer)])
def emit_op_getinteriorfield_gc(self, op, arglocs, regalloc, fcond):
(base_loc, index_loc, res_loc, ofs_loc, ofs, itemsize, fieldsize) = arglocs
scale = get_scale(fieldsize.value)
tmploc, save = self.get_tmp_reg([base_loc, ofs_loc])
assert not save
self.mc.gen_load_int(tmploc.value, itemsize.value)
self.mc.MUL(tmploc.value, index_loc.value, tmploc.value)
descr = op.getdescr()
assert isinstance(descr, InteriorFieldDescr)
signed = descr.fielddescr.is_field_signed()
if ofs.value > 0:
if ofs_loc.is_imm():
self.mc.ADD_ri(tmploc.value, tmploc.value, ofs_loc.value)
else:
self.mc.ADD_rr(tmploc.value, tmploc.value, ofs_loc.value)
ofs_loc = tmploc
self._load_from_mem(res_loc, base_loc, ofs_loc, imm(scale), signed, fcond)
return fcond
def emit_op_gc_store(self, op, arglocs, regalloc, fcond):
value_loc, base_loc, ofs_loc, size_loc = arglocs
scale = get_scale(size_loc.value)
self._write_to_mem(value_loc, base_loc, ofs_loc, imm(scale), fcond)
return fcond
def emit_op_getfield_gc(self, op, arglocs, regalloc, fcond):
base_loc, ofs, res, size = arglocs
signed = op.getdescr().is_field_signed()
scale = get_scale(size.value)
self._load_from_mem(res, base_loc, ofs, imm(scale), signed, fcond)
return fcond
def emit_op_gc_store(self, op, arglocs):
value_loc, base_loc, ofs_loc, size_loc = arglocs
scale = get_scale(size_loc.value)
self._write_to_mem(value_loc, base_loc, ofs_loc, scale)
def emit_op_setfield_gc(self, op, arglocs, regalloc, fcond):
value_loc, base_loc, ofs, size = arglocs
scale = get_scale(size.value)
self._write_to_mem(value_loc, base_loc, ofs, imm(scale), fcond)
return fcond
def _emit_op_gc_load(self, op, arglocs):
base_loc, ofs_loc, res_loc, nsize_loc = arglocs
nsize = nsize_loc.value
signed = (nsize < 0)
scale = get_scale(abs(nsize))
self._load_from_mem(res_loc, base_loc, ofs_loc, scale, signed)