def _gen_address(self, result, base_loc, scaled_loc, scale=0, static_offset=0):
assert scaled_loc.is_core_reg()
assert base_loc.is_core_reg()
assert check_imm_arg(scale)
assert check_imm_arg(static_offset)
if scale > 0:
self.mc.LSL_ri(r.ip.value, scaled_loc.value, scale)
scaled_loc = r.ip
else:
scaled_loc = scaled_loc
self.mc.ADD_rr(result.value, base_loc.value, scaled_loc.value)
self.mc.ADD_ri(result.value, result.value, static_offset)
def _adjust_sp(self, n):
# adjust the current stack pointer by n bytes
if n > 0:
if check_imm_arg(n):
self.mc.ADD_ri(r.sp.value, r.sp.value, n)
else:
self.mc.gen_load_int(r.ip.value, n)
self.mc.ADD_rr(r.sp.value, r.sp.value, r.ip.value)
elif n < 0:
n = abs(n)
if check_imm_arg(n):
self.mc.SUB_ri(r.sp.value, r.sp.value, n)
else:
self.mc.gen_load_int(r.ip.value, n)
self.mc.SUB_rr(r.sp.value, r.sp.value, r.ip.value)
def _adjust_sp(self, n):
# adjust the current stack pointer by n bytes
if n > 0:
if check_imm_arg(n):
self.mc.ADD_ri(r.sp.value, r.sp.value, n)
else:
self.mc.gen_load_int(r.ip.value, n)
self.mc.ADD_rr(r.sp.value, r.sp.value, r.ip.value)
elif n < 0:
n = abs(n)
if check_imm_arg(n):
self.mc.SUB_ri(r.sp.value, r.sp.value, n)
else:
self.mc.gen_load_int(r.ip.value, n)
self.mc.SUB_rr(r.sp.value, r.sp.value, r.ip.value)
def _call_assembler_load_result(self, op, result_loc):
if op.result is not None:
# load the return value from (tmploc, 0)
kind = op.result.type
descr = self.cpu.getarraydescr_for_frame(kind)
if kind == FLOAT:
ofs = self.cpu.unpack_arraydescr(descr)
assert check_imm_arg(ofs)
assert result_loc.is_vfp_reg()
# we always have a register here, since we have to sync them
# before call_assembler
self.load_reg(self.mc, result_loc, r.r0, ofs=ofs)
else:
assert result_loc is r.r0
ofs = self.cpu.unpack_arraydescr(descr)
assert check_imm_arg(ofs)
self.mc.LDR_ri(result_loc.value, result_loc.value, imm=ofs)
def prepare_op_gc_store_indexed(self, op, fcond):
boxes = op.getarglist()
base_loc = self.make_sure_var_in_reg(boxes[0], boxes)
value_loc = self.make_sure_var_in_reg(boxes[2], boxes)
index_loc = self.make_sure_var_in_reg(boxes[1], boxes)
assert boxes[3].getint() == 1 # scale
ofs = boxes[4].getint()
size = boxes[5].getint()
assert check_imm_arg(ofs)
return [value_loc, base_loc, index_loc, imm(size), imm(ofs)]
def prepare_op_gc_store_indexed(self, op, fcond):
boxes = op.getarglist()
base_loc = self.make_sure_var_in_reg(boxes[0], boxes)
value_loc = self.make_sure_var_in_reg(boxes[2], boxes)
index_loc = self.make_sure_var_in_reg(boxes[1], boxes)
assert boxes[3].getint() == 1 # scale
ofs = boxes[4].getint()
size = boxes[5].getint()
assert check_imm_arg(ofs)
return [value_loc, base_loc, index_loc, imm(size), imm(ofs)]
def _call_assembler_check_descr(self, value, tmploc):
ofs = self.cpu.get_ofs_of_frame_field("jf_descr")
self.mc.LDR_ri(r.ip.value, tmploc.value, imm=ofs)
if check_imm_arg(value):
self.mc.CMP_ri(r.ip.value, imm=value)
else:
self.mc.gen_load_int(r.lr.value, value)
self.mc.CMP_rr(r.ip.value, r.lr.value)
pos = self.mc.currpos()
self.mc.BKPT()
return pos
def _prepare_op_gc_load_indexed(self, op, fcond):
boxes = op.getarglist()
base_loc = self.make_sure_var_in_reg(boxes[0], boxes)
index_loc = self.make_sure_var_in_reg(boxes[1], boxes)
assert boxes[2].getint() == 1 # scale
ofs = boxes[3].getint()
nsize = boxes[4].getint()
assert check_imm_arg(ofs)
self.possibly_free_vars_for_op(op)
self.free_temp_vars()
res_loc = self.force_allocate_reg(op)
return [res_loc, base_loc, index_loc, imm(nsize), imm(ofs)]
def _prepare_op_gc_load_indexed(self, op, fcond):
boxes = op.getarglist()
base_loc = self.make_sure_var_in_reg(boxes[0], boxes)
index_loc = self.make_sure_var_in_reg(boxes[1], boxes)
assert boxes[2].getint() == 1 # scale
ofs = boxes[3].getint()
nsize = boxes[4].getint()
assert check_imm_arg(ofs)
self.possibly_free_vars_for_op(op)
self.free_temp_vars()
res_loc = self.force_allocate_reg(op)
return [res_loc, base_loc, index_loc, imm(nsize), imm(ofs)]
def prepare_op_gc_store(self, op, fcond):
boxes = op.getarglist()
base_loc = self.make_sure_var_in_reg(boxes[0], boxes)
ofs = boxes[1].getint()
value_loc = self.make_sure_var_in_reg(boxes[2], boxes)
size = boxes[3].getint()
ofs_size = default_imm_size if size < 8 else VMEM_imm_size
if check_imm_arg(ofs, size=ofs_size):
ofs_loc = imm(ofs)
else:
ofs_loc = self.get_scratch_reg(INT, boxes)
self.assembler.load(ofs_loc, imm(ofs))
return [value_loc, base_loc, ofs_loc, imm(size)]
def prepare_op_gc_store(self, op, fcond):
boxes = op.getarglist()
base_loc = self.make_sure_var_in_reg(boxes[0], boxes)
ofs = boxes[1].getint()
value_loc = self.make_sure_var_in_reg(boxes[2], boxes)
size = boxes[3].getint()
ofs_size = default_imm_size if size < 8 else VMEM_imm_size
if check_imm_arg(ofs, size=ofs_size):
ofs_loc = imm(ofs)
else:
ofs_loc = self.get_scratch_reg(INT, boxes)
self.assembler.load(ofs_loc, imm(ofs))
return [value_loc, base_loc, ofs_loc, imm(size)]
def _prepare_op_gc_load(self, op, fcond):
a0 = op.getarg(0)
ofs = op.getarg(1).getint()
nsize = op.getarg(2).getint() # negative for "signed"
base_loc = self.make_sure_var_in_reg(a0)
immofs = imm(ofs)
ofs_size = default_imm_size if abs(nsize) < 8 else VMEM_imm_size
if check_imm_arg(ofs, size=ofs_size):
ofs_loc = immofs
else:
ofs_loc = self.get_scratch_reg(INT, [a0])
self.assembler.load(ofs_loc, immofs)
self.possibly_free_vars_for_op(op)
self.free_temp_vars()
res_loc = self.force_allocate_reg(op)
return [base_loc, ofs_loc, res_loc, imm(nsize)]
def _prepare_op_gc_load(self, op, fcond):
a0 = op.getarg(0)
ofs = op.getarg(1).getint()
nsize = op.getarg(2).getint() # negative for "signed"
base_loc = self.make_sure_var_in_reg(a0)
immofs = imm(ofs)
ofs_size = default_imm_size if abs(nsize) < 8 else VMEM_imm_size
if check_imm_arg(ofs, size=ofs_size):
ofs_loc = immofs
else:
ofs_loc = self.get_scratch_reg(INT, [a0])
self.assembler.load(ofs_loc, immofs)
self.possibly_free_vars_for_op(op)
self.free_temp_vars()
res_loc = self.force_allocate_reg(op)
return [base_loc, ofs_loc, res_loc, imm(nsize)]
def emit_op_zero_array(self, op, arglocs, regalloc, fcond):
from rpython.jit.backend.llsupport.descr import unpack_arraydescr
assert len(arglocs) == 0
size_box = op.getarg(2)
if isinstance(size_box, ConstInt) and size_box.getint() == 0:
return fcond # nothing to do
itemsize, baseofs, _ = unpack_arraydescr(op.getdescr())
args = op.getarglist()
#
# ZERO_ARRAY(base_loc, start, size, 1, 1)
# 'start' and 'size' are both expressed in bytes,
# and the two scaling arguments should always be ConstInt(1) on ARM.
assert args[3].getint() == 1
assert args[4].getint() == 1
#
base_loc = regalloc.rm.make_sure_var_in_reg(args[0], args)
startbyte_box = args[1]
if isinstance(startbyte_box, ConstInt):
startbyte_loc = None
startbyte = startbyte_box.getint()
assert startbyte >= 0
else:
startbyte_loc = regalloc.rm.make_sure_var_in_reg(startbyte_box, args)
startbyte = -1
# base_loc and startbyte_loc are in two regs here (or startbyte_loc
# is an immediate). Compute the dstaddr_loc, which is the raw
# address that we will pass as first argument to memset().
# It can be in the same register as either one, but not in
# args[2], because we're still needing the latter.
dstaddr_box = TempVar()
dstaddr_loc = regalloc.rm.force_allocate_reg(dstaddr_box, [args[2]])
if startbyte >= 0: # a constant
ofs = baseofs + startbyte
reg = base_loc.value
else:
self.mc.ADD_rr(dstaddr_loc.value, base_loc.value, startbyte_loc.value)
ofs = baseofs
reg = dstaddr_loc.value
if check_imm_arg(ofs):
self.mc.ADD_ri(dstaddr_loc.value, reg, imm=ofs)
else:
self.mc.gen_load_int(r.ip.value, ofs)
self.mc.ADD_rr(dstaddr_loc.value, reg, r.ip.value)
# We use STRB, STRH or STR based on whether we know the array
# item size is a multiple of 1, 2 or 4.
if itemsize & 1:
itemsize = 1
elif itemsize & 2:
itemsize = 2
else:
itemsize = 4
limit = itemsize
next_group = -1
if itemsize < 4 and startbyte >= 0:
# we optimize STRB/STRH into STR, but this needs care:
# it only works if startindex_loc is a constant, otherwise
# we'd be doing unaligned accesses.
next_group = (-startbyte) & 3
limit = 4
if isinstance(size_box, ConstInt) and size_box.getint() <= 14 * limit: # same limit as GCC
# Inline a series of STR operations, starting at 'dstaddr_loc'.
#
self.mc.gen_load_int(r.ip.value, 0)
i = 0
total_size = size_box.getint()
while i < total_size:
sz = itemsize
if i == next_group:
next_group += 4
if next_group <= total_size:
sz = 4
if sz == 4:
self.mc.STR_ri(r.ip.value, dstaddr_loc.value, imm=i)
elif sz == 2:
self.mc.STRH_ri(r.ip.value, dstaddr_loc.value, imm=i)
else:
self.mc.STRB_ri(r.ip.value, dstaddr_loc.value, imm=i)
i += sz
else:
if isinstance(size_box, ConstInt):
size_loc = imm(size_box.getint())
else:
# load size_loc in a register different than dstaddr_loc
size_loc = regalloc.rm.make_sure_var_in_reg(size_box, [dstaddr_box])
#
# call memset()
regalloc.before_call()
self.simple_call_no_collect(imm(self.memset_addr), [dstaddr_loc, imm(0), size_loc])
regalloc.rm.possibly_free_var(size_box)
regalloc.rm.possibly_free_var(dstaddr_box)
return fcond
