src/ieee754/part/test/test_partsig.py

   1 #!/usr/bin/env python3
   2 # SPDX-License-Identifier: LGPL-2.1-or-later
   3 # See Notices.txt for copyright information
   4
   5 from nmigen import Signal, Module, Elaboratable
   6 from nmigen.back.pysim import Simulator, Delay
   7 from nmigen.cli import rtlil
   8
   9 from ieee754.part.partsig import PartitionedSignal
  10 from ieee754.part_mux.part_mux import PMux
  11
  12 from random import randint
  13 import unittest
  14 import itertools
  15 import math
  16
  17 def first_zero(x):
  18     res = 0
  19     for i in range(16):
  20         if x & (1<<i):
  21             return res
  22         res += 1
  23
  24 def count_bits(x):
  25     res = 0
  26     for i in range(16):
  27         if x & (1<<i):
  28             res += 1
  29     return res
  30
  31
  32 def perms(k):
  33     return map(''.join, itertools.product('01', repeat=k))
  34
  35
  36 def create_ilang(dut, traces, test_name):
  37     vl = rtlil.convert(dut, ports=traces)
  38     with open("%s.il" % test_name, "w") as f:
  39         f.write(vl)
  40
  41
  42 def create_simulator(module, traces, test_name):
  43     create_ilang(module, traces, test_name)
  44     return Simulator(module)
  45
  46
  47 # XXX this is for coriolis2 experimentation
  48 class TestAddMod2(Elaboratable):
  49     def __init__(self, width, partpoints):
  50         self.partpoints = partpoints
  51         self.a = PartitionedSignal(partpoints, width)
  52         self.b = PartitionedSignal(partpoints, width)
  53         self.bsig = Signal(width)
  54         self.add_output = Signal(width)
  55         self.ls_output = Signal(width) # left shift
  56         self.ls_scal_output = Signal(width) # left shift
  57         self.rs_output = Signal(width) # right shift
  58         self.rs_scal_output = Signal(width) # right shift
  59         self.sub_output = Signal(width)
  60         self.eq_output = Signal(len(partpoints)+1)
  61         self.gt_output = Signal(len(partpoints)+1)
  62         self.ge_output = Signal(len(partpoints)+1)
  63         self.ne_output = Signal(len(partpoints)+1)
  64         self.lt_output = Signal(len(partpoints)+1)
  65         self.le_output = Signal(len(partpoints)+1)
  66         self.mux_sel = Signal(len(partpoints)+1)
  67         self.mux_out = Signal(width)
  68         self.carry_in = Signal(len(partpoints)+1)
  69         self.add_carry_out = Signal(len(partpoints)+1)
  70         self.sub_carry_out = Signal(len(partpoints)+1)
  71         self.neg_output = Signal(width)
  72
  73     def elaborate(self, platform):
  74         m = Module()
  75         comb = m.d.comb
  76         sync = m.d.sync
  77         self.a.set_module(m)
  78         self.b.set_module(m)
  79         # compares
  80         sync += self.lt_output.eq(self.a < self.b)
  81         sync += self.ne_output.eq(self.a != self.b)
  82         sync += self.le_output.eq(self.a <= self.b)
  83         sync += self.gt_output.eq(self.a > self.b)
  84         sync += self.eq_output.eq(self.a == self.b)
  85         sync += self.ge_output.eq(self.a >= self.b)
  86         # add
  87         add_out, add_carry = self.a.add_op(self.a, self.b,
  88                                            self.carry_in)
  89         sync += self.add_output.eq(add_out)
  90         sync += self.add_carry_out.eq(add_carry)
  91         # sub
  92         sub_out, sub_carry = self.a.sub_op(self.a, self.b,
  93                                            self.carry_in)
  94         sync += self.sub_output.eq(sub_out)
  95         sync += self.sub_carry_out.eq(sub_carry)
  96         # neg
  97         sync += self.neg_output.eq(-self.a)
  98         # left shift
  99         sync += self.ls_output.eq(self.a << self.b)
 100         sync += self.rs_output.eq(self.a >> self.b)
 101         ppts = self.partpoints
 102         sync += self.mux_out.eq(PMux(m, ppts, self.mux_sel, self.a, self.b))
 103         # scalar left shift
 104         comb += self.bsig.eq(self.b.sig)
 105         sync += self.ls_scal_output.eq(self.a << self.bsig)
 106         sync += self.rs_scal_output.eq(self.a >> self.bsig)
 107
 108         return m
 109
 110
 111 class TestAddMod(Elaboratable):
 112     def __init__(self, width, partpoints):
 113         self.partpoints = partpoints
 114         self.a = PartitionedSignal(partpoints, width)
 115         self.b = PartitionedSignal(partpoints, width)
 116         self.bsig = Signal(width)
 117         self.add_output = Signal(width)
 118         self.ls_output = Signal(width) # left shift
 119         self.ls_scal_output = Signal(width) # left shift
 120         self.rs_output = Signal(width) # right shift
 121         self.rs_scal_output = Signal(width) # right shift
 122         self.sub_output = Signal(width)
 123         self.eq_output = Signal(len(partpoints)+1)
 124         self.gt_output = Signal(len(partpoints)+1)
 125         self.ge_output = Signal(len(partpoints)+1)
 126         self.ne_output = Signal(len(partpoints)+1)
 127         self.lt_output = Signal(len(partpoints)+1)
 128         self.le_output = Signal(len(partpoints)+1)
 129         self.mux_sel = Signal(len(partpoints)+1)
 130         self.mux_out = Signal(width)
 131         self.carry_in = Signal(len(partpoints)+1)
 132         self.add_carry_out = Signal(len(partpoints)+1)
 133         self.sub_carry_out = Signal(len(partpoints)+1)
 134         self.neg_output = Signal(width)
 135
 136     def elaborate(self, platform):
 137         m = Module()
 138         comb = m.d.comb
 139         sync = m.d.sync
 140         self.a.set_module(m)
 141         self.b.set_module(m)
 142         # compares
 143         comb += self.lt_output.eq(self.a < self.b)
 144         comb += self.ne_output.eq(self.a != self.b)
 145         comb += self.le_output.eq(self.a <= self.b)
 146         comb += self.gt_output.eq(self.a > self.b)
 147         comb += self.eq_output.eq(self.a == self.b)
 148         comb += self.ge_output.eq(self.a >= self.b)
 149         # add
 150         add_out, add_carry = self.a.add_op(self.a, self.b,
 151                                            self.carry_in)
 152         comb += self.add_output.eq(add_out)
 153         comb += self.add_carry_out.eq(add_carry)
 154         # sub
 155         sub_out, sub_carry = self.a.sub_op(self.a, self.b,
 156                                            self.carry_in)
 157         comb += self.sub_output.eq(sub_out)
 158         comb += self.sub_carry_out.eq(sub_carry)
 159         # neg
 160         comb += self.neg_output.eq(-self.a)
 161         # left shift
 162         comb += self.ls_output.eq(self.a << self.b)
 163         # right shift
 164         comb += self.rs_output.eq(self.a >> self.b)
 165         ppts = self.partpoints
 166         # mux
 167         comb += self.mux_out.eq(PMux(m, ppts, self.mux_sel, self.a, self.b))
 168         # scalar left shift
 169         comb += self.bsig.eq(self.b.sig)
 170         comb += self.ls_scal_output.eq(self.a << self.bsig)
 171         # scalar right shift
 172         comb += self.rs_scal_output.eq(self.a >> self.bsig)
 173
 174         return m
 175
 176
 177 class TestPartitionPoints(unittest.TestCase):
 178     def test(self):
 179         width = 16
 180         part_mask = Signal(4)  # divide into 4-bits
 181         module = TestAddMod(width, part_mask)
 182
 183         test_name = "part_sig_add"
 184         traces = [part_mask,
 185                   module.a.sig,
 186                   module.b.sig,
 187                   module.add_output,
 188                   module.eq_output]
 189         sim = create_simulator(module, traces, test_name)
 190
 191         def async_process():
 192
 193             def test_ls_scal_fn(carry_in, a, b, mask):
 194                 # reduce range of b
 195                 bits = count_bits(mask)
 196                 newb = b & ((bits-1))
 197                 print ("%x %x %x bits %d trunc %x" % \
 198                         (a, b, mask, bits, newb))
 199                 b = newb
 200                 # TODO: carry
 201                 carry_in = 0
 202                 lsb = mask & ~(mask-1) if carry_in else 0
 203                 sum = ((a & mask) << b)
 204                 result = mask & sum
 205                 carry = (sum & mask) != sum
 206                 carry = 0
 207                 print("res", hex(a), hex(b), hex(sum), hex(mask), hex(result))
 208                 return result, carry
 209
 210             def test_rs_scal_fn(carry_in, a, b, mask):
 211                 # reduce range of b
 212                 bits = count_bits(mask)
 213                 newb = b & ((bits-1))
 214                 print ("%x %x %x bits %d trunc %x" % \
 215                         (a, b, mask, bits, newb))
 216                 b = newb
 217                 # TODO: carry
 218                 carry_in = 0
 219                 lsb = mask & ~(mask-1) if carry_in else 0
 220                 sum = ((a & mask) >> b)
 221                 result = mask & sum
 222                 carry = (sum & mask) != sum
 223                 carry = 0
 224                 print("res", hex(a), hex(b), hex(sum), hex(mask), hex(result))
 225                 return result, carry
 226
 227             def test_ls_fn(carry_in, a, b, mask):
 228                 # reduce range of b
 229                 bits = count_bits(mask)
 230                 fz = first_zero(mask)
 231                 newb = b & ((bits-1)<<fz)
 232                 print ("%x %x %x bits %d zero %d trunc %x" % \
 233                         (a, b, mask, bits, fz, newb))
 234                 b = newb
 235                 # TODO: carry
 236                 carry_in = 0
 237                 lsb = mask & ~(mask-1) if carry_in else 0
 238                 b = (b & mask)
 239                 b = b >>fz
 240                 sum = ((a & mask) << b)
 241                 result = mask & sum
 242                 carry = (sum & mask) != sum
 243                 carry = 0
 244                 print("res", hex(a), hex(b), hex(sum), hex(mask), hex(result))
 245                 return result, carry
 246
 247             def test_rs_fn(carry_in, a, b, mask):
 248                 # reduce range of b
 249                 bits = count_bits(mask)
 250                 fz = first_zero(mask)
 251                 newb = b & ((bits-1)<<fz)
 252                 print ("%x %x %x bits %d zero %d trunc %x" % \
 253                         (a, b, mask, bits, fz, newb))
 254                 b = newb
 255                 # TODO: carry
 256                 carry_in = 0
 257                 lsb = mask & ~(mask-1) if carry_in else 0
 258                 b = (b & mask)
 259                 b = b >>fz
 260                 sum = ((a & mask) >> b)
 261                 result = mask & sum
 262                 carry = (sum & mask) != sum
 263                 carry = 0
 264                 print("res", hex(a), hex(b), hex(sum), hex(mask), hex(result))
 265                 return result, carry
 266
 267             def test_add_fn(carry_in, a, b, mask):
 268                 lsb = mask & ~(mask-1) if carry_in else 0
 269                 sum = (a & mask) + (b & mask) + lsb
 270                 result = mask & sum
 271                 carry = (sum & mask) != sum
 272                 print(a, b, sum, mask)
 273                 return result, carry
 274
 275             def test_sub_fn(carry_in, a, b, mask):
 276                 lsb = mask & ~(mask-1) if carry_in else 0
 277                 sum = (a & mask) + (~b & mask) + lsb
 278                 result = mask & sum
 279                 carry = (sum & mask) != sum
 280                 return result, carry
 281
 282             def test_neg_fn(carry_in, a, b, mask):
 283                 return test_add_fn(0, a, ~0, mask)
 284
 285             def test_op(msg_prefix, carry, test_fn, mod_attr, *mask_list):
 286                 rand_data = []
 287                 for i in range(100):
 288                     a, b = randint(0, 1 << 16), randint(0, 1 << 16)
 289                     rand_data.append((a, b))
 290                 for a, b in [(0x0000, 0x0000),
 291                              (0x1234, 0x1234),
 292                              (0xABCD, 0xABCD),
 293                              (0xFFFF, 0x0000),
 294                              (0x0000, 0x0000),
 295                              (0xFFFF, 0xFFFF),
 296                              (0x0000, 0xFFFF)] + rand_data:
 297                     yield module.a.eq(a)
 298                     yield module.b.eq(b)
 299                     carry_sig = 0xf if carry else 0
 300                     yield module.carry_in.eq(carry_sig)
 301                     yield Delay(0.1e-6)
 302                     y = 0
 303                     carry_result = 0
 304                     for i, mask in enumerate(mask_list):
 305                         print ("i/mask", i, hex(mask))
 306                         res, c = test_fn(carry, a, b, mask)
 307                         y |= res
 308                         lsb = mask & ~(mask - 1)
 309                         bit_set = int(math.log2(lsb))
 310                         carry_result |= c << int(bit_set/4)
 311                     outval = (yield getattr(module, "%s_output" % mod_attr))
 312                     # TODO: get (and test) carry output as well
 313                     print(a, b, outval, carry)
 314                     msg = f"{msg_prefix}: 0x{a:X} {mod_attr} 0x{b:X}" + \
 315                         f" => 0x{y:X} != 0x{outval:X}"
 316                     self.assertEqual(y, outval, msg)
 317                     if hasattr(module, "%s_carry_out" % mod_attr):
 318                         c_outval = (yield getattr(module,
 319                                                   "%s_carry_out" % mod_attr))
 320                         msg = f"{msg_prefix}: 0x{a:X} {mod_attr} 0x{b:X}" + \
 321                             f" => 0x{carry_result:X} != 0x{c_outval:X}"
 322                         self.assertEqual(carry_result, c_outval, msg)
 323
 324             for (test_fn, mod_attr) in (
 325                                         (test_ls_scal_fn, "ls_scal"),
 326                                         (test_ls_fn, "ls"),
 327                                         (test_rs_scal_fn, "rs_scal"),
 328                                         (test_rs_fn, "rs"),
 329                                         (test_add_fn, "add"),
 330                                         (test_sub_fn, "sub"),
 331                                         (test_neg_fn, "neg"),
 332                                         ):
 333                 yield part_mask.eq(0)
 334                 yield from test_op("16-bit", 1, test_fn, mod_attr, 0xFFFF)
 335                 yield from test_op("16-bit", 0, test_fn, mod_attr, 0xFFFF)
 336                 yield part_mask.eq(0b10)
 337                 yield from test_op("8-bit", 0, test_fn, mod_attr,
 338                                    0xFF00, 0x00FF)
 339                 yield from test_op("8-bit", 1, test_fn, mod_attr,
 340                                    0xFF00, 0x00FF)
 341                 yield part_mask.eq(0b1111)
 342                 yield from test_op("4-bit", 0, test_fn, mod_attr,
 343                                    0xF000, 0x0F00, 0x00F0, 0x000F)
 344                 yield from test_op("4-bit", 1, test_fn, mod_attr,
 345                                    0xF000, 0x0F00, 0x00F0, 0x000F)
 346
 347             def test_ne_fn(a, b, mask):
 348                 return (a & mask) != (b & mask)
 349
 350             def test_lt_fn(a, b, mask):
 351                 return (a & mask) < (b & mask)
 352
 353             def test_le_fn(a, b, mask):
 354                 return (a & mask) <= (b & mask)
 355
 356             def test_eq_fn(a, b, mask):
 357                 return (a & mask) == (b & mask)
 358
 359             def test_gt_fn(a, b, mask):
 360                 return (a & mask) > (b & mask)
 361
 362             def test_ge_fn(a, b, mask):
 363                 return (a & mask) >= (b & mask)
 364
 365             def test_binop(msg_prefix, test_fn, mod_attr, *maskbit_list):
 366                 for a, b in [(0x0000, 0x0000),
 367                              (0x1234, 0x1234),
 368                              (0xABCD, 0xABCD),
 369                              (0xFFFF, 0x0000),
 370                              (0x0000, 0x0000),
 371                              (0xFFFF, 0xFFFF),
 372                              (0x0000, 0xFFFF),
 373                              (0xABCD, 0xABCE),
 374                              (0x8000, 0x0000),
 375                              (0xBEEF, 0xFEED)]:
 376                     yield module.a.eq(a)
 377                     yield module.b.eq(b)
 378                     yield Delay(0.1e-6)
 379                     # convert to mask_list
 380                     mask_list = []
 381                     for mb in maskbit_list:
 382                         v = 0
 383                         for i in range(4):
 384                             if mb & (1 << i):
 385                                 v |= 0xf << (i*4)
 386                         mask_list.append(v)
 387                     y = 0
 388                     # do the partitioned tests
 389                     for i, mask in enumerate(mask_list):
 390                         if test_fn(a, b, mask):
 391                             # OR y with the lowest set bit in the mask
 392                             y |= maskbit_list[i]
 393                     # check the result
 394                     outval = (yield getattr(module, "%s_output" % mod_attr))
 395                     msg = f"{msg_prefix}: {mod_attr} 0x{a:X} == 0x{b:X}" + \
 396                         f" => 0x{y:X} != 0x{outval:X}, masklist %s"
 397                     print((msg % str(maskbit_list)).format(locals()))
 398                     self.assertEqual(y, outval, msg % str(maskbit_list))
 399
 400             for (test_fn, mod_attr) in ((test_eq_fn, "eq"),
 401                                         (test_gt_fn, "gt"),
 402                                         (test_ge_fn, "ge"),
 403                                         (test_lt_fn, "lt"),
 404                                         (test_le_fn, "le"),
 405                                         (test_ne_fn, "ne"),
 406                                         ):
 407                 yield part_mask.eq(0)
 408                 yield from test_binop("16-bit", test_fn, mod_attr, 0b1111)
 409                 yield part_mask.eq(0b10)
 410                 yield from test_binop("8-bit", test_fn, mod_attr,
 411                                       0b1100, 0b0011)
 412                 yield part_mask.eq(0b1111)
 413                 yield from test_binop("4-bit", test_fn, mod_attr,
 414                                       0b1000, 0b0100, 0b0010, 0b0001)
 415
 416             def test_muxop(msg_prefix, *maskbit_list):
 417                 for a, b in [(0x0000, 0x0000),
 418                              (0x1234, 0x1234),
 419                              (0xABCD, 0xABCD),
 420                              (0xFFFF, 0x0000),
 421                              (0x0000, 0x0000),
 422                              (0xFFFF, 0xFFFF),
 423                              (0x0000, 0xFFFF)]:
 424                     # convert to mask_list
 425                     mask_list = []
 426                     for mb in maskbit_list:
 427                         v = 0
 428                         for i in range(4):
 429                             if mb & (1 << i):
 430                                 v |= 0xf << (i*4)
 431                         mask_list.append(v)
 432
 433                     # TODO: sel needs to go through permutations of mask_list
 434                     for p in perms(len(mask_list)):
 435
 436                         sel = 0
 437                         selmask = 0
 438                         for i, v in enumerate(p):
 439                             if v == '1':
 440                                 sel |= maskbit_list[i]
 441                                 selmask |= mask_list[i]
 442
 443                         yield module.a.eq(a)
 444                         yield module.b.eq(b)
 445                         yield module.mux_sel.eq(sel)
 446                         yield Delay(0.1e-6)
 447                         y = 0
 448                         # do the partitioned tests
 449                         for i, mask in enumerate(mask_list):
 450                             if (selmask & mask):
 451                                 y |= (a & mask)
 452                             else:
 453                                 y |= (b & mask)
 454                         # check the result
 455                         outval = (yield module.mux_out)
 456                         msg = f"{msg_prefix}: mux " + \
 457                             f"0x{sel:X} ? 0x{a:X} : 0x{b:X}" + \
 458                             f" => 0x{y:X} != 0x{outval:X}, masklist %s"
 459                         # print ((msg % str(maskbit_list)).format(locals()))
 460                         self.assertEqual(y, outval, msg % str(maskbit_list))
 461
 462             yield part_mask.eq(0)
 463             yield from test_muxop("16-bit", 0b1111)
 464             yield part_mask.eq(0b10)
 465             yield from test_muxop("8-bit", 0b1100, 0b0011)
 466             yield part_mask.eq(0b1111)
 467             yield from test_muxop("4-bit", 0b1000, 0b0100, 0b0010, 0b0001)
 468
 469         sim.add_process(async_process)
 470         with sim.write_vcd(
 471                 vcd_file=open(test_name + ".vcd", "w"),
 472                 gtkw_file=open(test_name + ".gtkw", "w"),
 473                 traces=traces):
 474             sim.run()
 475
 476
 477 if __name__ == '__main__':
 478     unittest.main()