2 # SPDX-License-Identifier: LGPL-2.1-or-later
3 # See Notices.txt for copyright information
5 from ieee754
.part_mul_add
.multiply
import \
6 (PartitionPoints
, PartitionedAdder
, AddReduce
,
7 Mul8_16_32_64
, OP_MUL_LOW
, OP_MUL_SIGNED_HIGH
,
8 OP_MUL_SIGNED_UNSIGNED_HIGH
, OP_MUL_UNSIGNED_HIGH
)
9 from nmigen
import Signal
, Module
10 from nmigen
.back
.pysim
import Simulator
, Delay
, Tick
, Passive
11 from nmigen
.hdl
.ast
import Assign
, Value
12 from typing
import Any
, Generator
, List
, Union
, Optional
, Tuple
, Iterable
14 from hashlib
import sha256
17 from nmigen
.cli
import verilog
, rtlil
20 def create_ilang(dut
, traces
, test_name
):
21 vl
= rtlil
.convert(dut
, ports
=traces
)
22 with
open("%s.il" % test_name
, "w") as f
:
26 def create_simulator(module
: Any
,
28 test_name
: str) -> Simulator
:
29 create_ilang(module
, traces
, test_name
)
30 return Simulator(module
,
31 vcd_file
=open(test_name
+ ".vcd", "w"),
32 gtkw_file
=open(test_name
+ ".gtkw", "w"),
36 AsyncProcessCommand
= Union
[Delay
, Tick
, Passive
, Assign
, Value
]
37 ProcessCommand
= Optional
[AsyncProcessCommand
]
38 AsyncProcessGenerator
= Generator
[AsyncProcessCommand
, Union
[int, None], None]
39 ProcessGenerator
= Generator
[ProcessCommand
, Union
[int, None], None]
42 class TestPartitionPoints(unittest
.TestCase
):
43 def test(self
) -> None:
47 partition_point_10
= Signal()
48 partition_points
= PartitionPoints({1: True,
50 10: partition_point_10
})
51 module
.d
.comb
+= mask
.eq(partition_points
.as_mask(width
))
52 with
create_simulator(module
,
53 [mask
, partition_point_10
],
54 "partition_points") as sim
:
55 def async_process() -> AsyncProcessGenerator
:
56 self
.assertEqual((yield partition_points
[1]), True)
57 self
.assertEqual((yield partition_points
[5]), False)
58 yield partition_point_10
.eq(0)
60 self
.assertEqual((yield mask
), 0xFFFD)
61 yield partition_point_10
.eq(1)
63 self
.assertEqual((yield mask
), 0xFBFD)
65 sim
.add_process(async_process
)
69 class TestPartitionedAdder(unittest
.TestCase
):
70 def test(self
) -> None:
72 partition_nibbles
= Signal()
73 partition_bytes
= Signal()
74 module
= PartitionedAdder(width
,
75 {0x4: partition_nibbles
,
76 0x8: partition_bytes | partition_nibbles
,
77 0xC: partition_nibbles
})
78 with
create_simulator(module
,
84 "partitioned_adder") as sim
:
85 def async_process() -> AsyncProcessGenerator
:
86 def test_add(msg_prefix
: str,
87 *mask_list
: Tuple
[int, ...]) -> Any
:
88 for a
, b
in [(0x0000, 0x0000),
99 for mask
in mask_list
:
100 y |
= mask
& ((a
& mask
) + (b
& mask
))
101 output
= (yield module
.output
)
102 msg
= f
"{msg_prefix}: 0x{a:X} + 0x{b:X}" + \
103 f
" => 0x{y:X} != 0x{output:X}"
104 self
.assertEqual(y
, output
, msg
)
105 yield partition_nibbles
.eq(0)
106 yield partition_bytes
.eq(0)
107 yield from test_add("16-bit", 0xFFFF)
108 yield partition_nibbles
.eq(0)
109 yield partition_bytes
.eq(1)
110 yield from test_add("8-bit", 0xFF00, 0x00FF)
111 yield partition_nibbles
.eq(1)
112 yield partition_bytes
.eq(0)
113 yield from test_add("4-bit", 0xF000, 0x0F00, 0x00F0, 0x000F)
115 sim
.add_process(async_process
)
119 class GenOrCheck(enum
.Enum
):
120 Generate
= enum
.auto()
124 class TestAddReduce(unittest
.TestCase
):
125 def calculate_input_values(self
,
128 extra_keys
: List
[int] = []
129 ) -> (List
[int], List
[str]):
131 input_values_str
= []
132 for i
in range(input_count
):
140 hash_input
= f
"{input_count} {i} {key} {extra_keys}"
141 hash = sha256(hash_input
.encode()).digest()
142 value
= int.from_bytes(hash, byteorder
="little")
144 input_values
.append(value
)
145 input_values_str
.append(f
"0x{value:04X}")
146 return input_values
, input_values_str
148 def subtest_value(self
,
149 inputs
: List
[Signal
],
151 mask_list
: List
[int],
152 gen_or_check
: GenOrCheck
,
153 values
: List
[int]) -> AsyncProcessGenerator
:
154 if gen_or_check
== GenOrCheck
.Generate
:
155 for i
, v
in zip(inputs
, values
):
159 for mask
in mask_list
:
164 output
= (yield module
.o
.output
)
165 if gen_or_check
== GenOrCheck
.Check
:
166 self
.assertEqual(y
, output
, f
"0x{y:X} != 0x{output:X}")
169 def subtest_key(self
,
171 inputs
: List
[Signal
],
174 mask_list
: List
[int],
175 gen_or_check
: GenOrCheck
) -> AsyncProcessGenerator
:
176 values
, values_str
= self
.calculate_input_values(input_count
, key
)
177 if gen_or_check
== GenOrCheck
.Check
:
178 with self
.subTest(inputs
=values_str
):
179 yield from self
.subtest_value(inputs
,
185 yield from self
.subtest_value(inputs
,
191 def subtest_run_sim(self
,
196 inputs
: List
[Signal
],
198 delay_cycles
: int) -> None:
199 def generic_process(gen_or_check
: GenOrCheck
) -> AsyncProcessGenerator
:
200 for partition_4_value
, partition_8_value
, mask_list
in [
202 (0, 1, [0xFF00, 0x00FF]),
203 (1, 0, [0xFFF0, 0x000F]),
204 (1, 1, [0xFF00, 0x00F0, 0x000F])]:
206 if gen_or_check
== GenOrCheck
.Check
:
207 with self
.subTest(partition_4
=partition_4_value
,
208 partition_8
=partition_8_value
):
209 for key
in range(key_count
):
210 with self
.subTest(key
=key
):
211 yield from self
.subtest_key(input_count
,
218 if gen_or_check
== GenOrCheck
.Generate
:
219 yield partition_4
.eq(partition_4_value
)
220 yield partition_8
.eq(partition_8_value
)
221 for key
in range(key_count
):
222 yield from self
.subtest_key(input_count
,
229 def generate_process() -> AsyncProcessGenerator
:
230 yield from generic_process(GenOrCheck
.Generate
)
232 def check_process() -> AsyncProcessGenerator
:
233 if delay_cycles
!= 0:
234 for _
in range(delay_cycles
):
236 yield from generic_process(GenOrCheck
.Check
)
238 if "sync" in sim
._domains
:
240 sim
.add_process(generate_process
)
241 sim
.add_process(check_process
)
244 def subtest_file(self
,
246 register_levels
: List
[int]) -> None:
247 max_level
= AddReduce
.get_max_level(input_count
)
248 for level
in register_levels
:
249 if level
> max_level
:
251 partition_4
= Signal()
252 partition_8
= Signal()
253 partition_points
= PartitionPoints()
254 partition_points
[4] = partition_4
255 partition_points
[8] = partition_8
257 inputs
= [Signal(width
, name
=f
"input_{i}")
258 for i
in range(input_count
)]
259 module
= AddReduce(inputs
,
264 file_name
= "add_reduce"
265 if len(register_levels
) != 0:
266 file_name
+= f
"-{'_'.join(map(repr, register_levels))}"
267 file_name
+= f
"-{input_count:02d}"
268 ports
= [partition_4
, partition_8
, *inputs
, module
.o
.output
]
269 #create_ilang(module, ports, file_name)
270 with
create_simulator(module
, ports
, file_name
) as sim
:
271 self
.subtest_run_sim(input_count
,
277 len(register_levels
))
279 def subtest_register_levels(self
, register_levels
: List
[int]) -> None:
280 for input_count
in range(0, 16):
281 with self
.subTest(input_count
=input_count
,
282 register_levels
=repr(register_levels
)):
283 self
.subtest_file(input_count
, register_levels
)
285 def test_empty(self
) -> None:
286 self
.subtest_register_levels([])
288 def test_0(self
) -> None:
289 self
.subtest_register_levels([0])
291 def test_1(self
) -> None:
292 self
.subtest_register_levels([1])
294 def test_2(self
) -> None:
295 self
.subtest_register_levels([2])
297 def test_3(self
) -> None:
298 self
.subtest_register_levels([3])
300 def test_4(self
) -> None:
301 self
.subtest_register_levels([4])
303 def test_5(self
) -> None:
304 self
.subtest_register_levels([5])
306 def test_0(self
) -> None:
307 self
.subtest_register_levels([0])
309 def test_0_1(self
) -> None:
310 self
.subtest_register_levels([0, 1])
312 def test_0_1_2(self
) -> None:
313 self
.subtest_register_levels([0, 1, 2])
315 def test_0_1_2_3(self
) -> None:
316 self
.subtest_register_levels([0, 1, 2, 3])
318 def test_0_1_2_3_4(self
) -> None:
319 self
.subtest_register_levels([0, 1, 2, 3, 4])
321 def test_0_1_2_3_4_5(self
) -> None:
322 self
.subtest_register_levels([0, 1, 2, 3, 4, 5])
324 def test_0_2(self
) -> None:
325 self
.subtest_register_levels([0, 2])
327 def test_0_3(self
) -> None:
328 self
.subtest_register_levels([0, 3])
330 def test_0_4(self
) -> None:
331 self
.subtest_register_levels([0, 4])
333 def test_0_5(self
) -> None:
334 self
.subtest_register_levels([0, 5])
343 self
.a_signed
= a_signed
344 self
.b_signed
= b_signed
345 self
.bit_width
= bit_width
346 self
.high_half
= high_half
349 return f
"SIMDMulLane({self.a_signed}, {self.b_signed}, " +\
350 f
"{self.bit_width}, {self.high_half})"
353 class TestMul8_16_32_64(unittest
.TestCase
):
355 def simd_mul(a
: int, b
: int, lanes
: List
[SIMDMulLane
]) -> Tuple
[int, int]:
357 intermediate_output
= 0
360 a_signed
= lane
.a_signed
or not lane
.high_half
361 b_signed
= lane
.b_signed
or not lane
.high_half
362 mask
= (1 << lane
.bit_width
) - 1
363 sign_bit
= 1 << (lane
.bit_width
- 1)
364 a_part
= (a
>> shift
) & mask
365 if a_signed
and (a_part
& sign_bit
) != 0:
366 a_part
-= 1 << lane
.bit_width
367 b_part
= (b
>> shift
) & mask
368 if b_signed
and (b_part
& sign_bit
) != 0:
369 b_part
-= 1 << lane
.bit_width
370 value
= a_part
* b_part
371 value
&= (1 << (lane
.bit_width
* 2)) - 1
372 intermediate_output |
= value
<< (shift
* 2)
374 value
>>= lane
.bit_width
376 output |
= value
<< shift
377 shift
+= lane
.bit_width
378 return output
, intermediate_output
381 def get_test_cases(lanes
: List
[SIMDMulLane
],
382 keys
: Iterable
[int]) -> Iterable
[Tuple
[int, int]]:
385 hash_input
= f
"{i} {lanes} {list(keys)}"
386 hash = sha256(hash_input
.encode()).digest()
387 value
= int.from_bytes(hash, byteorder
="little")
388 yield (value
& mask
, value
>> 64)
393 a |
= 1 << (shift
+ lane
.bit_width
- 1)
394 b |
= 1 << (shift
+ lane
.bit_width
- 1)
395 shift
+= lane
.bit_width
398 def test_simd_mul_lane(self
):
399 self
.assertEqual(f
"{SIMDMulLane(True, True, 8, False)}",
400 "SIMDMulLane(True, True, 8, False)")
402 def test_simd_mul(self
):
403 lanes
= [SIMDMulLane(True,
419 a
= 0x0123456789ABCDEF
420 b
= 0xFEDCBA9876543210
421 output
= 0x0121FA00FE1C28FE
422 intermediate_output
= 0x0121FA0023E20B28C94DFE1C280AFEF0
423 self
.assertEqual(self
.simd_mul(a
, b
, lanes
),
424 (output
, intermediate_output
))
425 a
= 0x8123456789ABCDEF
426 b
= 0xFEDCBA9876543210
427 output
= 0x81B39CB4FE1C28FE
428 intermediate_output
= 0x81B39CB423E20B28C94DFE1C280AFEF0
429 self
.assertEqual(self
.simd_mul(a
, b
, lanes
),
430 (output
, intermediate_output
))
432 def test_signed_mul_from_unsigned(self
):
433 for i
in range(0, 0x10):
434 for j
in range(0, 0x10):
435 si
= i
if i
& 8 else i
- 0x10 # signed i
436 sj
= j
if j
& 8 else j
- 0x10 # signed j
440 with self
.subTest(i
=i
, j
=j
, si
=si
, sj
=sj
,
441 mulu
=mulu
, mulsu
=mulsu
, mul
=mul
):
446 self
.assertEqual(mulsu
& 0xFF, mulsu2
& 0xFF)
451 self
.assertEqual(mul
& 0xFF, mul2
& 0xFF)
453 def subtest_value(self
,
456 module
: Mul8_16_32_64
,
457 lanes
: List
[SIMDMulLane
],
458 gen_or_check
: GenOrCheck
) -> AsyncProcessGenerator
:
459 if gen_or_check
== GenOrCheck
.Generate
:
462 output2
, intermediate_output2
= self
.simd_mul(a
, b
, lanes
)
464 if gen_or_check
== GenOrCheck
.Check
:
465 intermediate_output
= (yield module
.intermediate_output
)
466 self
.assertEqual(intermediate_output
,
467 intermediate_output2
,
468 f
"0x{intermediate_output:X} "
469 + f
"!= 0x{intermediate_output2:X}")
470 output
= (yield module
.output
)
471 self
.assertEqual(output
, output2
, f
"0x{output:X} != 0x{output2:X}")
474 def subtest_lanes_2(self
,
475 lanes
: List
[SIMDMulLane
],
476 module
: Mul8_16_32_64
,
477 gen_or_check
: GenOrCheck
) -> AsyncProcessGenerator
:
484 op
= OP_MUL_SIGNED_HIGH
486 op
= OP_MUL_SIGNED_UNSIGNED_HIGH
488 self
.assertFalse(lane
.b_signed
,
489 "unsigned * signed not supported")
490 op
= OP_MUL_UNSIGNED_HIGH
493 self
.assertEqual(lane
.bit_width
% 8, 0)
494 for i
in range(lane
.bit_width
// 8):
495 if gen_or_check
== GenOrCheck
.Generate
:
496 yield module
.part_ops
[part_index
].eq(op
)
498 for i
in range(lane
.bit_width
// 8 - 1):
499 if gen_or_check
== GenOrCheck
.Generate
:
500 yield module
.part_pts
[bit_index
].eq(0)
502 if bit_index
< 64 and gen_or_check
== GenOrCheck
.Generate
:
503 yield module
.part_pts
[bit_index
].eq(1)
505 self
.assertEqual(part_index
, 8)
506 for a
, b
in self
.get_test_cases(lanes
, ()):
507 if gen_or_check
== GenOrCheck
.Check
:
508 with self
.subTest(a
=f
"{a:X}", b
=f
"{b:X}"):
509 yield from self
.subtest_value(a
, b
, module
, lanes
, gen_or_check
)
511 yield from self
.subtest_value(a
, b
, module
, lanes
, gen_or_check
)
513 def subtest_lanes(self
,
514 lanes
: List
[SIMDMulLane
],
515 module
: Mul8_16_32_64
,
516 gen_or_check
: GenOrCheck
) -> AsyncProcessGenerator
:
517 if gen_or_check
== GenOrCheck
.Check
:
518 with self
.subTest(lanes
=repr(lanes
)):
519 yield from self
.subtest_lanes_2(lanes
, module
, gen_or_check
)
521 yield from self
.subtest_lanes_2(lanes
, module
, gen_or_check
)
523 def subtest_file(self
,
524 register_levels
: List
[int]) -> None:
525 module
= Mul8_16_32_64(register_levels
)
526 file_name
= "mul8_16_32_64"
527 if len(register_levels
) != 0:
528 file_name
+= f
"-{'_'.join(map(repr, register_levels))}"
531 module
.intermediate_output
,
533 ports
.extend(module
.part_ops
)
534 ports
.extend(module
.part_pts
.values())
535 with
create_simulator(module
, ports
, file_name
) as sim
:
536 def process(gen_or_check
: GenOrCheck
) -> AsyncProcessGenerator
:
537 for a_signed
in False, True:
538 for b_signed
in False, True:
539 if not a_signed
and b_signed
:
541 for high_half
in False, True:
542 if not high_half
and not (a_signed
and b_signed
):
544 yield from self
.subtest_lanes(
545 [SIMDMulLane(a_signed
,
551 yield from self
.subtest_lanes(
552 [SIMDMulLane(a_signed
,
558 yield from self
.subtest_lanes(
559 [SIMDMulLane(a_signed
,
565 yield from self
.subtest_lanes(
566 [SIMDMulLane(a_signed
,
572 yield from self
.subtest_lanes([SIMDMulLane(False,
590 yield from self
.subtest_lanes([SIMDMulLane(True,
608 yield from self
.subtest_lanes([SIMDMulLane(True,
627 def generate_process() -> AsyncProcessGenerator
:
628 yield from process(GenOrCheck
.Generate
)
630 def check_process() -> AsyncProcessGenerator
:
631 if len(register_levels
) != 0:
632 for _
in register_levels
:
634 yield from process(GenOrCheck
.Check
)
636 if "sync" in sim
._domains
:
638 sim
.add_process(generate_process
)
639 sim
.add_process(check_process
)
642 def subtest_register_levels(self
, register_levels
: List
[int]) -> None:
643 with self
.subTest(register_levels
=repr(register_levels
)):
644 self
.subtest_file(register_levels
)
646 def test_empty(self
) -> None:
647 self
.subtest_register_levels([])
649 def test_0(self
) -> None:
650 self
.subtest_register_levels([0])
652 def test_1(self
) -> None:
653 self
.subtest_register_levels([1])
655 def test_2(self
) -> None:
656 self
.subtest_register_levels([2])
658 def test_3(self
) -> None:
659 self
.subtest_register_levels([3])
661 def test_4(self
) -> None:
662 self
.subtest_register_levels([4])
664 def test_5(self
) -> None:
665 self
.subtest_register_levels([5])
667 def test_6(self
) -> None:
668 self
.subtest_register_levels([6])
670 def test_7(self
) -> None:
671 self
.subtest_register_levels([7])
673 def test_8(self
) -> None:
674 self
.subtest_register_levels([8])
676 def test_9(self
) -> None:
677 self
.subtest_register_levels([9])
679 def test_10(self
) -> None:
680 self
.subtest_register_levels([10])
682 def test_0(self
) -> None:
683 self
.subtest_register_levels([0])
685 def test_0_1(self
) -> None:
686 self
.subtest_register_levels([0, 1])
688 def test_0_1_2(self
) -> None:
689 self
.subtest_register_levels([0, 1, 2])
691 def test_0_1_2_3(self
) -> None:
692 self
.subtest_register_levels([0, 1, 2, 3])
694 def test_0_1_2_3_4(self
) -> None:
695 self
.subtest_register_levels([0, 1, 2, 3, 4])
697 def test_0_1_2_3_4_5(self
) -> None:
698 self
.subtest_register_levels([0, 1, 2, 3, 4, 5])
700 def test_0_1_2_3_4_5_6(self
) -> None:
701 self
.subtest_register_levels([0, 1, 2, 3, 4, 5, 6])
703 def test_0_1_2_3_4_5_6_7(self
) -> None:
704 self
.subtest_register_levels([0, 1, 2, 3, 4, 5, 6, 7])
706 def test_0_1_2_3_4_5_6_7_8(self
) -> None:
707 self
.subtest_register_levels([0, 1, 2, 3, 4, 5, 6, 7, 8])
709 def test_0_1_2_3_4_5_6_7_8_9(self
) -> None:
710 self
.subtest_register_levels([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
712 def test_0_1_2_3_4_5_6_7_8_9_10(self
) -> None:
713 self
.subtest_register_levels([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
715 def test_0_2(self
) -> None:
716 self
.subtest_register_levels([0, 2])
718 def test_0_3(self
) -> None:
719 self
.subtest_register_levels([0, 3])
721 def test_0_4(self
) -> None:
722 self
.subtest_register_levels([0, 4])
724 def test_0_5(self
) -> None:
725 self
.subtest_register_levels([0, 5])
727 def test_0_6(self
) -> None:
728 self
.subtest_register_levels([0, 6])
730 def test_0_7(self
) -> None:
731 self
.subtest_register_levels([0, 7])
733 def test_0_8(self
) -> None:
734 self
.subtest_register_levels([0, 8])
736 def test_0_9(self
) -> None:
737 self
.subtest_register_levels([0, 9])
739 def test_0_10(self
) -> None:
740 self
.subtest_register_levels([0, 10])
742 if __name__
== '__main__':