def _(bs):
ints = len([x for x in arg_types if x is type_impl.Int or x is type_impl.Str or x is Raw or x is type_impl.NoneType or isinstance(x, _FuncPtr) or x is type_impl.Bool])
floats = len([x for x in arg_types if x is type_impl.Float])
floats_orig = floats
pos = 0
for arg_type in reversed(arg_types):
type = bs.flow.stack.pop()
assert arg_type is type, (type, arg_type)
if type is type_impl.Int or type is type_impl.Bool:
ints -= 1
bs.code += isa.mov(int_regs[ints], MemRef(registers.rsp, pos))
pos += 8
elif isinstance(type, _FuncPtr):
ints -= 1
bs.code += isa.mov(int_regs[ints], MemRef(registers.rsp, pos))
pos += 8
elif type is type_impl.NoneType:
ints -= 1
bs.code += isa.mov(int_regs[ints], 0)
elif type is Raw:
ints -= 1
bs.code += isa.mov(int_regs[ints], MemRef(registers.rsp, pos))
pos += 8
bs.code += isa.add(int_regs[ints], 8)
elif type is type_impl.Float:
floats -= 1
bs.code += isa.movsd(float_regs[floats], MemRef(registers.rsp, pos))
pos += 8
elif type is type_impl.Str:
ints -= 1
bs.code += isa.mov(int_regs[ints], MemRef(registers.rsp, pos))
bs.code += isa.test(int_regs[ints], 1)
short = bs.program.get_unique_label()
end = bs.program.get_unique_label()
bs.code += isa.jnz(short)
# long
bs.code += isa.add(int_regs[ints], 8)
bs.code += isa.jmp(end)
bs.code += short
# short
bs.code += isa.shr(MemRef(registers.rsp, pos), 8)
bs.code += isa.lea(int_regs[ints], MemRef(registers.rsp, pos, data_size=None))
bs.code += end
pos += 8
else:
assert False, type
assert bs.flow.stack.pop() is self
bs.code += isa.mov(registers.rbx, ctypes.cast(self.func, ctypes.c_void_p).value)
bs.code += isa.mov(registers.r12, registers.rsp)
bs.code += isa.and_(registers.rsp, -16)
bs.code += isa.mov(registers.rax, floats)
bs.code += isa.call(registers.rbx)
bs.code += isa.mov(registers.rsp, registers.r12)
bs.code += isa.add(registers.rsp, pos)
bs.code += isa.push(registers.rax)
bs.flow.stack.append(type_impl.Int)
def end(self):
"""Do post-loop iterator code"""
if self.mode == CTR:
self.code.add(x86_64.loop(self.start_label))
elif self.mode == DEC:
# branch if r_count is not zero (CR)
# Note that this relies on someone (e.g. cleanup()) setting the
# condition register properly.
if self.step_size() == 1:
self.code.add(x86_64.jnz(self.start_label))
else:
self.code.add(x86_64.cmp(self.r_count, 0))
self.code.add(x86_64.jg(self.start_label))
elif self.mode == INC:
if self.external_stop:
if isinstance(self.r_count, memory.MemRef) and isinstance(
self.r_stop, memory.MemRef):
if self.r_clobber == None:
raise Exception(
'Must specify clobber_reg if count and stop values are both stored in memory.'
)
#self.code.add(x86_64.push(registers.rax))
#if self.r_count.base != registers.rsp:
# self.code.add(x86_64.mov(registers.rax, self.r_count))
#else:
# oldm = self.r_count
# m = memory.MemRef(oldm.base, oldm.disp+8, oldm.index, oldm.scale, oldm.data_size)
# self.code.add(x86_64.mov(registers.rax, m))
#if self.r_stop.base != registers.rsp:
# self.code.add(x86_64.cmp(registers.rax, self.r_stop))
#else:
# oldm = self.r_stop
# m = memory.MemRef(oldm.base, oldm.disp+8, oldm.index, oldm.scale, oldm.data_size)
# self.code.add(x86_64.cmp(registers.rax, m))
#self.code.add(x86_64.pop(registers.rax))
else:
self.code.add(x86_64.mov(self.r_clobber, self.r_count))
self.code.add(x86_64.cmp(self.r_clobber, self.r_stop))
else:
self.code.add(x86_64.cmp(self.r_count, self.r_stop))
else:
self.code.add(x86_64.cmp(self.r_count, self.n))
self.code.add(x86_64.jnge(self.start_label))
return
def end(self):
"""Do post-loop iterator code"""
if self.mode == CTR:
self.code.add(x86_64.loop(self.start_label))
elif self.mode == DEC:
# branch if r_count is not zero (CR)
# Note that this relies on someone (e.g. cleanup()) setting the
# condition register properly.
if self.step_size() == 1:
self.code.add(x86_64.jnz(self.start_label))
else:
self.code.add(x86_64.cmp(self.r_count, 0))
self.code.add(x86_64.jg(self.start_label))
elif self.mode == INC:
if self.external_stop:
if isinstance(self.r_count, memory.MemRef) and isinstance(self.r_stop, memory.MemRef):
if self.r_clobber == None:
raise Exception('Must specify clobber_reg if count and stop values are both stored in memory.')
#self.code.add(x86_64.push(registers.rax))
#if self.r_count.base != registers.rsp:
# self.code.add(x86_64.mov(registers.rax, self.r_count))
#else:
# oldm = self.r_count
# m = memory.MemRef(oldm.base, oldm.disp+8, oldm.index, oldm.scale, oldm.data_size)
# self.code.add(x86_64.mov(registers.rax, m))
#if self.r_stop.base != registers.rsp:
# self.code.add(x86_64.cmp(registers.rax, self.r_stop))
#else:
# oldm = self.r_stop
# m = memory.MemRef(oldm.base, oldm.disp+8, oldm.index, oldm.scale, oldm.data_size)
# self.code.add(x86_64.cmp(registers.rax, m))
#self.code.add(x86_64.pop(registers.rax))
else:
self.code.add(x86_64.mov(self.r_clobber, self.r_count))
self.code.add(x86_64.cmp(self.r_clobber, self.r_stop))
else:
self.code.add(x86_64.cmp(self.r_count, self.r_stop))
else:
self.code.add(x86_64.cmp(self.r_count, self.n))
self.code.add(x86_64.jnge(self.start_label))
return
data[0] = 0
prgm = env.Program()
code = prgm.get_stream()
x86.set_active_code(code)
x86.mov(rax, 1)
x86.mov(rcx, ITERS)
x86.mov(rdi, dbi[0])
lbl_loop = prgm.get_unique_label("loop")
code.add(lbl_loop)
x86.add(MemRef(rdi), rax)
x86.dec(rcx)
x86.jnz(lbl_loop)
prgm += code
prgm.print_code(hex = True)
proc = env.Processor()
t1 = time.time()
ids = [proc.execute(prgm, async = True) for i in xrange(0, THREADS)]
[proc.join(i) for i in ids]
t2 = time.time()
print "time", t2 - t1
print "val", data[0], ITERS * THREADS
print "passed?", data[0] == ITERS * THREADS
data[0] = 0
prgm = env.Program()
code = prgm.get_stream()
x86.set_active_code(code)
x86.mov(rax, 1)
x86.mov(rcx, ITERS)
x86.mov(rdi, dbi[0])
lbl_loop = prgm.get_unique_label("loop")
code.add(lbl_loop)
x86.add(MemRef(rdi), rax)
x86.dec(rcx)
x86.jnz(lbl_loop)
prgm += code
prgm.print_code(hex=True)
proc = env.Processor()
t1 = time.time()
ids = [proc.execute(prgm, async=True) for i in xrange(0, THREADS)]
[proc.join(i) for i in ids]
t2 = time.time()
print "time", t2 - t1
print "val", data[0], ITERS * THREADS
print "passed?", data[0] == ITERS * THREADS
# This case locks like it should, so should be correct.
def Test():
code = env.InstructionStream()
proc = env.Processor()
params = env.ExecParams()
params.p1 = 3
mr32 = MemRef(rbp, 16, data_size=32)
mr8 = MemRef(rbp, 16, data_size=8)
lbl1 = code.get_label("lbl1")
lbl2 = code.get_label("lbl2")
code.add(x86.xor(rax, rax))
code.add(x86.cmp(rax, 1))
code.add(x86.jne(lbl1))
code.add(x86.ud2())
code.add(x86.ud2())
code.add(lbl1)
code.add(x86.cmp(rax, 1))
code.add(x86.je(lbl2))
code.add(x86.add(rax, 12))
code.add(lbl2)
# printer.PrintInstructionStream(code, printer.x86_64_Nasm(function_name="foobar"))
ret = proc.execute(code)
print "ret", ret
assert (ret == 12)
print "W00T"
code.reset()
code.add(x86.xor(rax, rax))
code.add(x86.cmp(rax, 1))
code.add(x86.jne(28))
code.add(x86.ud2())
code.add(x86.ud2())
code.add(x86.cmp(eax, 1))
code.add(x86.je(37))
code.add(x86.add(rax, 12))
code.print_code(hex=True, pro=True, epi=True)
print "a"
ret = proc.execute(code)
print "b"
print "ret", ret
assert (ret == 12)
print "w00t 2"
code.reset()
call_lbl = code.get_label("call_fn")
code.add(x86.xor(rax, rax))
code.add(x86.call(call_lbl))
code.add(x86.jmp(code.lbl_epilogue))
code.add(x86.mov(rax, 75))
code.add(x86.mov(rax, 42))
code.add(call_lbl)
code.add(x86.mov(rax, 15))
code.add(x86.ret())
code.print_code()
ret = proc.execute(code)
print "ret", ret
assert (ret == 15)
code.reset()
fwd_lbl = code.get_label("FORWARD")
bck_lbl = code.get_label("BACKWARD")
code.add(x86.xor(rax, rax))
code.add(bck_lbl)
code.add(x86.cmp(rax, 1))
code.add(x86.jne(fwd_lbl))
for i in xrange(0, 65):
code.add(x86.pop(r15))
code.add(fwd_lbl)
ret = proc.execute(code, mode='int')
assert (ret == 0)
code.reset()
loop_lbl = code.get_label("LOOP")
out_lbl = code.get_label("OUT")
skip_lbl = code.get_label("SKIP")
code.add(x86.xor(rax, rax))
code.add(loop_lbl)
for i in range(0, 1):
for i in xrange(0, 24):
code.add(x86.add(r15, MemRef(rsp, 4)))
code.add(x86.add(rax, 4))
code.add(x86.cmp(rax, 20))
code.add(x86.je(out_lbl))
for i in xrange(0, 24):
code.add(x86.add(r15, MemRef(rsp, 4)))
code.add(x86.cmp(rax, 32))
code.add(x86.jne(loop_lbl))
code.add(out_lbl)
code.add(x86.jmp(skip_lbl))
for i in xrange(0, 2):
code.add(x86.add(r15, MemRef(rsp, 4)))
code.add(skip_lbl)
ret = proc.execute(code, mode='int')
print "ret", ret
assert (ret == 20)
code.reset()
loop_lbl = code.get_label("LOOP")
else_lbl = code.get_label("ELSE")
finish_lbl = code.get_label("finish")
code.add(x86.mov(rax, 0))
code.add(x86.mov(rdx, 0))
code.add(loop_lbl)
code.add(x86.add(rax, 1))
code.add(x86.cmp(rax, 16))
code.add(x86.jge(finish_lbl))
code.add(x86.add(rdx, rax))
code.add(x86.mov(r8, rdx))
code.add(x86.and_(r8, 0x1))
code.add(x86.jnz(else_lbl))
code.add(x86.add(rdx, 1))
code.add(x86.jmp(loop_lbl))
code.add(else_lbl)
code.add(x86.add(rdx, r8))
code.add(x86.jmp(loop_lbl))
code.add(finish_lbl)
code.add(x86.mov(rax, rdx))
ret = proc.execute(code, mode='int')
print "ret", ret
assert (ret == 135)
code.reset()
loop_lbl = code.get_label("LOOP")
code.add(x86.xor(rax, rax))
code.add(x86.xor(rcx, rcx))
code.add(x86.mov(rdx, 1))
code.add(loop_lbl)
code.add(x86.inc(rax))
code.add(x86.cmp(rax, 7))
code.add(x86.cmove(rcx, rdx))
code.add(x86.jrcxz(loop_lbl))
code.print_code(hex=True)
ret = proc.execute(code, mode='int')
print "ret", ret
assert (ret == 7)
code.reset()
code.add(x86.mov(rax, MemRef(rbp, 16)))
code.add(x86.xor(rbx, rbx))
code.add(x86.mov(rbx, -1))
code.add(x86.mov(cl, 1))
code.add(x86.shld(rax, rbx, cl))
code.print_code(hex=True)
ret = proc.execute(code, params=params, mode='int')
print "ret", ret
assert (ret == 7)
# code.reset()
# code.add(x86.add(eax, 200))
# code.add(x86.xor(eax, eax))
# code.add(x86.add(al, 32))
# code.add(x86.add(bl, 32))
# code.add(x86.xor(bl, bl))
# code.add(x86.mov(mr8, al))
# code.add(x86.add(mr32, 0))
# code.add(x86.mov(eax, mr32))
# code.add(x86.mov(al, mr8))
#
# code.add(x86.imul(ax, ax, 4))
# code.add(x86.imul(eax, ebx, 10))
# code.add(x86.mov(cx, 1232))
# code.add(x86.sub(ax, cx))
# code.add(x86.xor(eax,eax))
# code.add(x86.mov(eax,ebx))
# code.add(x86.clc())
# code.add(x86.rcl(eax, 1))
# code.add(x86.rcr(eax, 1))
# #ret = proc.execute(code, debug = True, params = params)
# id1 = proc.execute(code, params = params, mode = 'int', async = True)
# id2 = proc.execute(code, params = params, mode = 'int', async = True)
# ret = proc.execute(code, params = params, mode = 'int')
# print "Return main thread: %d" % (ret)
# assert(ret == 1280)
# ret = proc.join(id1)
# print "Return thread 1: %d" % (ret)
# assert(ret == 1280)
# ret = proc.join(id2)
# print "Return thread 2: %d" % (ret)
# assert(ret == 1280)
code.reset()
code.add(x86.fldpi())
code.add(x86.pxor(xmm0, xmm0))
code.add(x86.fld1())
code.add(x86.fadd(st0, st0))
code.add(x86.fmulp())
code.add(x86.fsin())
code.add(x86.fcos())
code.add(x86.fld1())
code.add(x86.fyl2xp1())
# x86_64 now uses xmm0 to return floats, not st0. So here, just make room
# on the stack, convert the FP result to an int and store it on the stack,
# then pop it into rax, the int return register.
code.add(x86.push(rax))
code.add(x86.fistp(MemRef(rsp)))
code.add(x86.pop(rax))
code.print_code(hex=True)
ret = proc.execute(code, params=params, mode='int')
assert (ret == 1)
print "Return main thread: %d" % (ret)
code.reset()
lbl_ok = code.get_label("OK")
code.add(x86.emms())
code.add(x86.movd(xmm0, mr32))
code.add(x86.mov(ebx, mr32))
code.add(x86.cmp(ebx, 3))
code.add(x86.je(lbl_ok))
code.add(x86.movd(eax, xmm0))
code.add(x86.cmp(eax, 3))
code.add(x86.je(lbl_ok))
code.add(x86.ud2())
code.add(lbl_ok)
code.add(x86.xor(eax, eax))
code.add(x86.movd(xmm1, ebx))
code.add(x86.paddq(xmm0, xmm1))
code.add(x86.pextrw(ecx, xmm0, 0))
code.add(x86.pinsrw(mm1, ecx, 0))
code.add(x86.movq2dq(xmm0, mm1))
code.add(x86.movdq2q(mm2, xmm0))
code.add(x86.movd(edx, mm2))
code.add(x86.movd(xmm5, edx))
code.add(x86.movd(ecx, xmm5))
code.add(x86.pinsrw(xmm6, ecx, 0))
code.add(x86.movd(eax, xmm6))
code.print_code(hex=True)
ret = proc.execute(code, params=params, mode='int')
print "Return main thread: %d" % (ret)
assert (ret == 6)
code.reset()
# Test immediate size encodings
code.add(x86.add(eax, 300))
code.add(x86.add(ax, 300))
code.add(x86.add(ax, 30))
code.add(x86.mov(eax, 16))
code.add(x86.mov(eax, 300))
code.reset()
code.add(x86.add(eax, 0xDEADBEEF))
code.add(x86.add(ebx, 0xDEADBEEF))
code.print_code(hex=True)
# Try the LOCK prefix
code.reset()
code.add(x86.xor(eax, eax))
code.add(x86.add(mr32, eax))
code.add(x86.add(mr32, eax, lock=True))
#code.print_code(hex = True)
proc.execute(code, params=params)
code.reset()
code.add(x86.mov(edx, 0x1234))
code.add(x86.mov(eax, 0xFFFF))
code.add(x86.xchg(edx, eax))
code.print_code(hex=True)
ret = proc.execute(code, params=params)
print "ret:", ret
assert (ret == 0x1234)
code.reset()
code.add(x86.mov(rax, rsp))
code.add(x86.pushfq())
code.add(x86.sub(rax, rsp))
code.add(x86.add(rsp, rax))
code.print_code(hex=True)
ret = proc.execute(code, params=params)
print "ret:", ret
assert (ret == 8)
code.reset()
data = extarray.extarray('H', xrange(0, 16))
# code.add(x86.push(rdi))
code.add(x86.mov(rdi, data.buffer_info()[0]))
code.add(x86.movaps(xmm1, MemRef(rdi, data_size=128)))
code.add(x86.pextrw(rax, xmm1, 0))
code.add(x86.pextrw(rbx, xmm1, 1))
code.add(x86.pextrw(rcx, xmm1, 2))
code.add(x86.pextrw(rdx, xmm1, 3))
code.add(x86.shl(rbx, 16))
code.add(x86.shl(rcx, 32))
code.add(x86.shl(rdx, 48))
code.add(x86.or_(rax, rbx))
code.add(x86.or_(rax, rcx))
code.add(x86.or_(rax, rdx))
# code.add(x86.pop(rdi))
code.print_code(hex=True)
ret = proc.execute(code, mode='int')
print "ret %x" % ret
assert (ret == 0x0003000200010000)
code.reset()
L1 = code.get_label("L1")
code.add(x86.xor(rax, rax))
code.add(x86.mov(rcx, 3))
code.add(L1)
code.add(x86.add(rax, 1))
code.add(x86.loop(L1))
code.print_code(hex=True)
ret = proc.execute(code, mode='int')
print "ret %x" % ret
assert (ret == 0x03)
return
def Test():
prgm = env.Program()
code = prgm.get_stream()
proc = env.Processor()
params = env.ExecParams()
params.p1 = 3
lbl1 = prgm.get_label("lbl1")
lbl2 = prgm.get_label("lbl2")
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.jne(lbl1))
code.add(x86.ud2())
code.add(x86.ud2())
code.add(lbl1)
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.je(lbl2))
code.add(x86.add(prgm.gp_return, 12))
code.add(lbl2)
prgm.add(code)
#prgm.print_code(pro = True, epi = True, hex = True)
ret = proc.execute(prgm, mode='int')
print "ret", ret
assert (ret == 12)
prgm.reset()
code.reset()
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.jne(28))
code.add(x86.ud2())
code.add(x86.ud2())
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.je(37))
code.add(x86.add(prgm.gp_return, 12))
prgm.add(code)
prgm.print_code(hex=True, pro=True, epi=True)
ret = proc.execute(prgm)
print "ret", ret
assert (ret == 12)
prgm.reset()
code.reset()
call_lbl = prgm.get_label("call_fn")
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(x86.call(call_lbl))
code.add(x86.jmp(prgm.lbl_epilogue))
code.add(x86.mov(prgm.gp_return, 75))
code.add(x86.mov(prgm.gp_return, 42))
code.add(call_lbl)
code.add(x86.mov(prgm.gp_return, 15))
code.add(x86.ret())
prgm.add(code)
prgm.print_code()
ret = proc.execute(prgm, mode='int')
print "ret", ret
assert (ret == 15)
prgm.reset()
code.reset()
fwd_lbl = prgm.get_label("FORWARD")
bck_lbl = prgm.get_label("BACKWARD")
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(bck_lbl)
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.jne(fwd_lbl))
r_foo = prgm.acquire_register()
for i in xrange(0, 65):
code.add(x86.pop(r_foo))
prgm.release_register(r_foo)
code.add(fwd_lbl)
prgm.add(code)
ret = proc.execute(prgm, mode='int')
print "ret", ret
assert (ret == 0)
prgm.reset()
code.reset()
loop_lbl = prgm.get_label("LOOP")
out_lbl = prgm.get_label("OUT")
skip_lbl = prgm.get_label("SKIP")
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(loop_lbl)
r_foo = prgm.acquire_register()
for i in range(0, 1):
for i in xrange(0, 24):
code.add(x86.add(r_foo, MemRef(rsp, 4)))
code.add(x86.add(prgm.gp_return, 4))
code.add(x86.cmp(prgm.gp_return, 20))
code.add(x86.je(out_lbl))
for i in xrange(0, 24):
code.add(x86.add(r_foo, MemRef(rsp, 4)))
code.add(x86.cmp(prgm.gp_return, 32))
code.add(x86.jne(loop_lbl))
code.add(out_lbl)
code.add(x86.jmp(skip_lbl))
for i in xrange(0, 2):
code.add(x86.add(r_foo, MemRef(rsp, 4)))
code.add(skip_lbl)
prgm.release_register(r_foo)
prgm.add(code)
ret = proc.execute(prgm, mode='int')
print "ret", ret
assert (ret == 20)
prgm.reset()
code.reset()
r_tmp = prgm.acquire_registers(2)
loop_lbl = prgm.get_label("LOOP")
else_lbl = prgm.get_label("ELSE")
finish_lbl = prgm.get_label("finish")
code.add(x86.mov(prgm.gp_return, 0))
code.add(x86.mov(r_tmp[0], 0))
code.add(loop_lbl)
code.add(x86.add(prgm.gp_return, 1))
code.add(x86.cmp(prgm.gp_return, 16))
code.add(x86.jge(finish_lbl))
code.add(x86.add(r_tmp[0], prgm.gp_return))
code.add(x86.mov(r_tmp[1], r_tmp[0]))
code.add(x86.and_(r_tmp[1], 0x1))
code.add(x86.jnz(else_lbl))
code.add(x86.add(r_tmp[0], 1))
code.add(x86.jmp(loop_lbl))
code.add(else_lbl)
code.add(x86.add(r_tmp[0], r_tmp[1]))
code.add(x86.jmp(loop_lbl))
code.add(finish_lbl)
code.add(x86.mov(prgm.gp_return, r_tmp[0]))
prgm.release_registers(r_tmp)
prgm.add(code)
ret = proc.execute(prgm, mode='int')
print "ret", ret
assert (ret == 135)
prgm.reset()
code.reset()
loop_lbl = prgm.get_label("LOOP")
r_one = prgm.acquire_register()
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(x86.xor(rcx, rcx))
code.add(x86.mov(r_one, 1))
code.add(loop_lbl)
code.add(x86.inc(prgm.gp_return))
code.add(x86.cmp(prgm.gp_return, 7))
code.add(x86.cmove(rcx, r_one))
code.add(x86.jrcxz(loop_lbl))
prgm.release_register(r_one)
prgm.add(code)
prgm.print_code(hex=True)
ret = proc.execute(prgm, mode='int')
print "ret", ret
assert (ret == 7)
prgm.reset()
code.reset()
r_tmp = prgm.acquire_register()
code.add(x86.mov(prgm.gp_return, rdi))
code.add(x86.xor(r_tmp, r_tmp))
code.add(x86.mov(r_tmp, -1))
code.add(x86.mov(cl, 1))
code.add(x86.shld(prgm.gp_return, r_tmp, cl))
prgm.release_register(r_tmp)
prgm.add(code)
ret = proc.execute(prgm, params=params, mode='int')
print "ret", ret
assert (ret == 7)
prgm.reset()
code.reset()
code.add(x86.add(eax, 200))
code.add(x86.xor(eax, eax))
code.add(x86.add(al, 32))
code.add(x86.add(bl, 32))
code.add(x86.xor(bl, bl))
code.add(x86.mov(dil, al))
code.add(x86.add(rdi, 0))
code.add(x86.mov(eax, edi))
code.add(x86.mov(al, dil))
code.add(x86.imul(ax, ax, 4))
code.add(x86.imul(eax, ebx, 10))
code.add(x86.mov(cx, 1232))
code.add(x86.sub(ax, cx))
code.add(x86.xor(eax, eax))
code.add(x86.mov(eax, ebx))
code.add(x86.clc())
code.add(x86.rcl(eax, 1))
code.add(x86.rcr(eax, 1))
prgm.add(code)
#ret = proc.execute(prgm, debug = True, params = params)
id1 = proc.execute(prgm, params=params, mode='int', async=True)
id2 = proc.execute(prgm, params=params, mode='int', async=True)
ret = proc.execute(prgm, params=params, mode='int')
print "Return main thread: %d" % (ret)
assert (ret == 1280)
ret = proc.join(id1)
print "Return thread 1: %d" % (ret)
assert (ret == 1280)
ret = proc.join(id2)
print "Return thread 2: %d" % (ret)
assert (ret == 1280)
prgm.reset()
code.reset()
code.add(x86.fldpi())
code.add(x86.pxor(xmm0, xmm0))
code.add(x86.fld1())
code.add(x86.fadd(st0, st0))
code.add(x86.fmulp())
code.add(x86.fsin())
code.add(x86.fcos())
code.add(x86.fld1())
code.add(x86.fyl2xp1())
# x86_64 now uses xmm0 to return floats, not st0. So here, just make room
# on the stack, convert the FP result to an int and store it on the stack,
# then pop it into rax, the int return register.
code.add(x86.push(prgm.gp_return))
code.add(x86.fistp(MemRef(rsp)))
code.add(x86.pop(prgm.gp_return))
prgm.add(code)
prgm.print_code(hex=True)
ret = proc.execute(prgm, params=params, mode='int')
assert (ret == 1)
print "Return main thread: %d" % (ret)
prgm.reset()
code.reset()
lbl_ok = prgm.get_label("OK")
code.add(x86.emms())
code.add(x86.movd(xmm0, edi))
code.add(x86.mov(ebx, edi))
code.add(x86.cmp(ebx, 3))
code.add(x86.je(lbl_ok))
code.add(x86.movd(eax, xmm0))
code.add(x86.cmp(eax, 3))
code.add(x86.je(lbl_ok))
code.add(x86.ud2())
code.add(lbl_ok)
code.add(x86.xor(eax, eax))
code.add(x86.movd(xmm1, ebx))
code.add(x86.paddq(xmm0, xmm1))
code.add(x86.pextrw(ecx, xmm0, 0))
code.add(x86.pxor(mm1, mm1))
code.add(x86.pinsrw(mm1, ecx, 0))
code.add(x86.movq2dq(xmm0, mm1))
code.add(x86.movdq2q(mm2, xmm0))
code.add(x86.movd(edx, mm2))
code.add(x86.movd(xmm5, edx))
code.add(x86.movd(ecx, xmm5))
code.add(x86.pxor(xmm6, xmm6))
code.add(x86.pinsrw(xmm6, ecx, 0))
code.add(x86.movd(eax, xmm6))
prgm.add(code)
prgm.print_code(hex=True)
ret = proc.execute(prgm, params=params, mode='int')
print "Return main thread: %d" % (ret)
assert (ret == 6)
prgm.reset()
code.reset()
code.add(x86.mov(edx, 0x1234))
code.add(x86.mov(eax, 0xFFFF))
code.add(x86.xchg(edx, eax))
prgm.add(code)
prgm.print_code(hex=True)
ret = proc.execute(prgm, params=params)
print "ret:", ret
assert (ret == 0x1234)
prgm.reset()
code.reset()
code.add(x86.mov(prgm.gp_return, rsp))
code.add(x86.pushfq())
code.add(x86.sub(prgm.gp_return, rsp))
code.add(x86.add(rsp, prgm.gp_return))
prgm.add(code)
prgm.print_code(hex=True)
ret = proc.execute(prgm, params=params)
print "ret:", ret
assert (ret == 8)
prgm.reset()
code.reset()
data = extarray.extarray('H', xrange(0, 16))
r_128 = prgm.acquire_register(reg_type=XMMRegister)
regs = prgm.acquire_registers(4)
code.add(x86.mov(regs[0], data.buffer_info()[0]))
code.add(x86.movaps(r_128, MemRef(regs[0], data_size=128)))
code.add(x86.pextrw(prgm.gp_return, r_128, 0))
code.add(x86.pextrw(regs[1], r_128, 1))
code.add(x86.pextrw(regs[2], r_128, 2))
code.add(x86.pextrw(regs[3], r_128, 3))
code.add(x86.shl(regs[1], 16))
code.add(x86.shl(regs[2], 32))
code.add(x86.shl(regs[3], 48))
code.add(x86.or_(prgm.gp_return, regs[1]))
code.add(x86.or_(prgm.gp_return, regs[2]))
code.add(x86.or_(prgm.gp_return, regs[3]))
prgm.release_register(r_128)
prgm.release_registers(regs)
prgm.add(code)
prgm.print_code()
ret = proc.execute(prgm, mode='int')
print "ret %x" % ret
assert (ret == 0x0003000200010000)
prgm.reset()
code.reset()
util.load_float(code, xmm0, 3.14159)
prgm.add(code)
ret = proc.execute(prgm, mode='fp')
print "ret", ret
assert (ret - 3.14159 < 0.00001)
return
def Test():
prgm = env.Program()
code = prgm.get_stream()
proc = env.Processor()
params = env.ExecParams()
params.p1 = 3
lbl1 = prgm.get_label("lbl1")
lbl2 = prgm.get_label("lbl2")
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.jne(lbl1))
code.add(x86.ud2())
code.add(x86.ud2())
code.add(lbl1)
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.je(lbl2))
code.add(x86.add(prgm.gp_return, 12))
code.add(lbl2)
prgm.add(code)
#prgm.print_code(pro = True, epi = True, hex = True)
ret = proc.execute(prgm, mode = 'int')
print "ret", ret
assert(ret == 12)
prgm.reset()
code.reset()
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.jne(28))
code.add(x86.ud2())
code.add(x86.ud2())
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.je(37))
code.add(x86.add(prgm.gp_return, 12))
prgm.add(code)
prgm.print_code(hex = True, pro = True, epi = True)
ret = proc.execute(prgm)
print "ret", ret
assert(ret == 12)
prgm.reset()
code.reset()
call_lbl = prgm.get_label("call_fn")
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(x86.call(call_lbl))
code.add(x86.jmp(prgm.lbl_epilogue))
code.add(x86.mov(prgm.gp_return, 75))
code.add(x86.mov(prgm.gp_return, 42))
code.add(call_lbl)
code.add(x86.mov(prgm.gp_return, 15))
code.add(x86.ret())
prgm.add(code)
prgm.print_code()
ret = proc.execute(prgm, mode = 'int')
print "ret", ret
assert(ret == 15)
prgm.reset()
code.reset()
fwd_lbl = prgm.get_label("FORWARD")
bck_lbl = prgm.get_label("BACKWARD")
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(bck_lbl)
code.add(x86.cmp(prgm.gp_return, 1))
code.add(x86.jne(fwd_lbl))
r_foo = prgm.acquire_register()
for i in xrange(0, 65):
code.add(x86.pop(r_foo))
prgm.release_register(r_foo)
code.add(fwd_lbl)
prgm.add(code)
ret = proc.execute(prgm, mode = 'int')
print "ret", ret
assert(ret == 0)
prgm.reset()
code.reset()
loop_lbl = prgm.get_label("LOOP")
out_lbl = prgm.get_label("OUT")
skip_lbl = prgm.get_label("SKIP")
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(loop_lbl)
r_foo = prgm.acquire_register()
for i in range(0, 1):
for i in xrange(0, 24):
code.add(x86.add(r_foo, MemRef(rsp, 4)))
code.add(x86.add(prgm.gp_return, 4))
code.add(x86.cmp(prgm.gp_return, 20))
code.add(x86.je(out_lbl))
for i in xrange(0, 24):
code.add(x86.add(r_foo, MemRef(rsp, 4)))
code.add(x86.cmp(prgm.gp_return, 32))
code.add(x86.jne(loop_lbl))
code.add(out_lbl)
code.add(x86.jmp(skip_lbl))
for i in xrange(0, 2):
code.add(x86.add(r_foo, MemRef(rsp, 4)))
code.add(skip_lbl)
prgm.release_register(r_foo)
prgm.add(code)
ret = proc.execute(prgm, mode = 'int')
print "ret", ret
assert(ret == 20)
prgm.reset()
code.reset()
r_tmp = prgm.acquire_registers(2)
loop_lbl = prgm.get_label("LOOP")
else_lbl = prgm.get_label("ELSE")
finish_lbl = prgm.get_label("finish")
code.add(x86.mov(prgm.gp_return, 0))
code.add(x86.mov(r_tmp[0], 0))
code.add(loop_lbl)
code.add(x86.add(prgm.gp_return, 1))
code.add(x86.cmp(prgm.gp_return, 16))
code.add(x86.jge(finish_lbl))
code.add(x86.add(r_tmp[0], prgm.gp_return))
code.add(x86.mov(r_tmp[1], r_tmp[0]))
code.add(x86.and_(r_tmp[1], 0x1))
code.add(x86.jnz(else_lbl))
code.add(x86.add(r_tmp[0], 1))
code.add(x86.jmp(loop_lbl))
code.add(else_lbl)
code.add(x86.add(r_tmp[0], r_tmp[1]))
code.add(x86.jmp(loop_lbl))
code.add(finish_lbl)
code.add(x86.mov(prgm.gp_return, r_tmp[0]))
prgm.release_registers(r_tmp)
prgm.add(code)
ret = proc.execute(prgm, mode = 'int')
print "ret", ret
assert(ret == 135)
prgm.reset()
code.reset()
loop_lbl = prgm.get_label("LOOP")
r_one = prgm.acquire_register()
code.add(x86.xor(prgm.gp_return, prgm.gp_return))
code.add(x86.xor(rcx, rcx))
code.add(x86.mov(r_one, 1))
code.add(loop_lbl)
code.add(x86.inc(prgm.gp_return))
code.add(x86.cmp(prgm.gp_return, 7))
code.add(x86.cmove(rcx, r_one))
code.add(x86.jrcxz(loop_lbl))
prgm.release_register(r_one)
prgm.add(code)
prgm.print_code(hex = True)
ret = proc.execute(prgm, mode = 'int')
print "ret", ret
assert(ret == 7)
prgm.reset()
code.reset()
r_tmp = prgm.acquire_register()
code.add(x86.mov(prgm.gp_return, rdi))
code.add(x86.xor(r_tmp, r_tmp))
code.add(x86.mov(r_tmp, -1))
code.add(x86.mov(cl, 1))
code.add(x86.shld(prgm.gp_return, r_tmp, cl))
prgm.release_register(r_tmp)
prgm.add(code)
ret = proc.execute(prgm, params = params, mode = 'int')
print "ret", ret
assert(ret == 7)
prgm.reset()
code.reset()
code.add(x86.add(eax, 200))
code.add(x86.xor(eax, eax))
code.add(x86.add(al, 32))
code.add(x86.add(bl, 32))
code.add(x86.xor(bl, bl))
code.add(x86.mov(dil, al))
code.add(x86.add(rdi, 0))
code.add(x86.mov(eax, edi))
code.add(x86.mov(al, dil))
code.add(x86.imul(ax, ax, 4))
code.add(x86.imul(eax, ebx, 10))
code.add(x86.mov(cx, 1232))
code.add(x86.sub(ax, cx))
code.add(x86.xor(eax,eax))
code.add(x86.mov(eax,ebx))
code.add(x86.clc())
code.add(x86.rcl(eax, 1))
code.add(x86.rcr(eax, 1))
prgm.add(code)
#ret = proc.execute(prgm, debug = True, params = params)
id1 = proc.execute(prgm, params = params, mode = 'int', async = True)
id2 = proc.execute(prgm, params = params, mode = 'int', async = True)
ret = proc.execute(prgm, params = params, mode = 'int')
print "Return main thread: %d" % (ret)
assert(ret == 1280)
ret = proc.join(id1)
print "Return thread 1: %d" % (ret)
assert(ret == 1280)
ret = proc.join(id2)
print "Return thread 2: %d" % (ret)
assert(ret == 1280)
prgm.reset()
code.reset()
code.add(x86.fldpi())
code.add(x86.pxor(xmm0, xmm0))
code.add(x86.fld1())
code.add(x86.fadd(st0, st0))
code.add(x86.fmulp())
code.add(x86.fsin())
code.add(x86.fcos())
code.add(x86.fld1())
code.add(x86.fyl2xp1())
# x86_64 now uses xmm0 to return floats, not st0. So here, just make room
# on the stack, convert the FP result to an int and store it on the stack,
# then pop it into rax, the int return register.
code.add(x86.push(prgm.gp_return))
code.add(x86.fistp(MemRef(rsp)))
code.add(x86.pop(prgm.gp_return))
prgm.add(code)
prgm.print_code(hex = True)
ret = proc.execute(prgm, params = params, mode = 'int')
assert(ret == 1)
print "Return main thread: %d" % (ret)
prgm.reset()
code.reset()
lbl_ok = prgm.get_label("OK")
code.add(x86.emms())
code.add(x86.movd(xmm0, edi))
code.add(x86.mov(ebx, edi))
code.add(x86.cmp(ebx, 3))
code.add(x86.je(lbl_ok))
code.add(x86.movd(eax, xmm0))
code.add(x86.cmp(eax, 3))
code.add(x86.je(lbl_ok))
code.add(x86.ud2())
code.add(lbl_ok)
code.add(x86.xor(eax, eax))
code.add(x86.movd(xmm1, ebx))
code.add(x86.paddq(xmm0, xmm1))
code.add(x86.pextrw(ecx, xmm0, 0))
code.add(x86.pxor(mm1, mm1))
code.add(x86.pinsrw(mm1, ecx, 0))
code.add(x86.movq2dq(xmm0, mm1))
code.add(x86.movdq2q(mm2, xmm0))
code.add(x86.movd(edx,mm2))
code.add(x86.movd(xmm5,edx))
code.add(x86.movd(ecx, xmm5))
code.add(x86.pxor(xmm6, xmm6))
code.add(x86.pinsrw(xmm6, ecx, 0))
code.add(x86.movd(eax, xmm6))
prgm.add(code)
prgm.print_code(hex = True)
ret = proc.execute(prgm, params = params, mode = 'int')
print "Return main thread: %d" % (ret)
assert(ret == 6)
prgm.reset()
code.reset()
code.add(x86.mov(edx, 0x1234))
code.add(x86.mov(eax, 0xFFFF))
code.add(x86.xchg(edx, eax))
prgm.add(code)
prgm.print_code(hex = True)
ret = proc.execute(prgm, params = params)
print "ret:", ret
assert(ret == 0x1234)
prgm.reset()
code.reset()
code.add(x86.mov(prgm.gp_return, rsp))
code.add(x86.pushfq())
code.add(x86.sub(prgm.gp_return, rsp))
code.add(x86.add(rsp, prgm.gp_return))
prgm.add(code)
prgm.print_code(hex = True)
ret = proc.execute(prgm, params = params)
print "ret:", ret
assert(ret == 8)
prgm.reset()
code.reset()
data = extarray.extarray('H', xrange(0, 16))
r_128 = prgm.acquire_register(reg_type = XMMRegister)
regs = prgm.acquire_registers(4)
code.add(x86.mov(regs[0], data.buffer_info()[0]))
code.add(x86.movaps(r_128, MemRef(regs[0], data_size = 128)))
code.add(x86.pextrw(prgm.gp_return, r_128, 0))
code.add(x86.pextrw(regs[1], r_128, 1))
code.add(x86.pextrw(regs[2], r_128, 2))
code.add(x86.pextrw(regs[3], r_128, 3))
code.add(x86.shl(regs[1], 16))
code.add(x86.shl(regs[2], 32))
code.add(x86.shl(regs[3], 48))
code.add(x86.or_(prgm.gp_return, regs[1]))
code.add(x86.or_(prgm.gp_return, regs[2]))
code.add(x86.or_(prgm.gp_return, regs[3]))
prgm.release_register(r_128)
prgm.release_registers(regs)
prgm.add(code)
prgm.print_code()
ret = proc.execute(prgm, mode = 'int')
print "ret %x" % ret
assert(ret == 0x0003000200010000)
prgm.reset()
code.reset()
util.load_float(code, xmm0, 3.14159)
prgm.add(code)
ret = proc.execute(prgm, mode = 'fp')
print "ret", ret
assert(ret - 3.14159 < 0.00001)
return
def _(bs):
ints = len([
x for x in arg_types if x is type_impl.Int
or x is type_impl.Str or x is Raw or x is type_impl.NoneType
or isinstance(x, _FuncPtr) or x is type_impl.Bool
])
floats = len([x for x in arg_types if x is type_impl.Float])
floats_orig = floats
pos = 0
for arg_type in reversed(arg_types):
type = bs.flow.stack.pop()
assert arg_type is type, (type, arg_type)
if type is type_impl.Int or type is type_impl.Bool:
ints -= 1
bs.code += isa.mov(int_regs[ints],
MemRef(registers.rsp, pos))
pos += 8
elif isinstance(type, _FuncPtr):
ints -= 1
bs.code += isa.mov(int_regs[ints],
MemRef(registers.rsp, pos))
pos += 8
elif type is type_impl.NoneType:
ints -= 1
bs.code += isa.mov(int_regs[ints], 0)
elif type is Raw:
ints -= 1
bs.code += isa.mov(int_regs[ints],
MemRef(registers.rsp, pos))
pos += 8
bs.code += isa.add(int_regs[ints], 8)
elif type is type_impl.Float:
floats -= 1
bs.code += isa.movsd(float_regs[floats],
MemRef(registers.rsp, pos))
pos += 8
elif type is type_impl.Str:
ints -= 1
bs.code += isa.mov(int_regs[ints],
MemRef(registers.rsp, pos))
bs.code += isa.test(int_regs[ints], 1)
short = bs.program.get_unique_label()
end = bs.program.get_unique_label()
bs.code += isa.jnz(short)
# long
bs.code += isa.add(int_regs[ints], 8)
bs.code += isa.jmp(end)
bs.code += short
# short
bs.code += isa.shr(MemRef(registers.rsp, pos), 8)
bs.code += isa.lea(
int_regs[ints],
MemRef(registers.rsp, pos, data_size=None))
bs.code += end
pos += 8
else:
assert False, type
assert bs.flow.stack.pop() is self
bs.code += isa.mov(registers.rbx,
ctypes.cast(self.func, ctypes.c_void_p).value)
bs.code += isa.mov(registers.r12, registers.rsp)
bs.code += isa.and_(registers.rsp, -16)
bs.code += isa.mov(registers.rax, floats)
bs.code += isa.call(registers.rbx)
bs.code += isa.mov(registers.rsp, registers.r12)
bs.code += isa.add(registers.rsp, pos)
bs.code += isa.push(registers.rax)
bs.flow.stack.append(type_impl.Int)
