1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
|
#
# Copyright (C) 2018 Alyssa Rosenzweig
#
# Copyright (C) 2016 Intel Corporation
#
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the "Software"),
# to deal in the Software without restriction, including without limitation
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
# and/or sell copies of the Software, and to permit persons to whom the
# Software is furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice (including the next
# paragraph) shall be included in all copies or substantial portions of the
# Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
# IN THE SOFTWARE.
import argparse
import sys
import math
a = 'a'
b = 'b'
c = 'c'
algebraic = [
(('pack_unorm_4x8', a), ('pack_32_4x8', ('f2u8', ('fround_even', ('fmul', ('fsat', a), 255.0)))))
]
algebraic_late = [
# ineg must be lowered late, but only for integers; floats will try to
# have modifiers attached... hence why this has to be here rather than
# a more standard lower_negate approach
(('ineg', a), ('isub', 0, a)),
# Likewise we want fsub lowered but not isub
(('fsub', a, b), ('fadd', a, ('fneg', b))),
# These two special-cases save space/an op than the actual csel op +
# scheduler flexibility
(('b32csel', a, 'b@32', 0), ('iand', a, b)),
(('b32csel', a, 0, 'b@32'), ('iand', ('inot', a), b)),
# Fuse sat_signed. This should probably be shared with Bifrost
(('~fmin', ('fmax', a, -1.0), 1.0), ('fsat_signed', a)),
(('~fmax', ('fmin', a, 1.0), -1.0), ('fsat_signed', a)),
# Fuse clamp_positive. This should probably be shared with Utgard/bifrost
(('fmax', a, 0.0), ('fclamp_pos', a)),
(('ishl', 'a@16', b), ('u2u16', ('ishl', ('u2u32', a), b))),
(('ishr', 'a@16', b), ('i2i16', ('ishr', ('i2i32', a), b))),
(('ushr', 'a@16', b), ('u2u16', ('ushr', ('u2u32', a), b))),
(('ishl', 'a@8', b), ('u2u8', ('u2u16', ('ishl', ('u2u32', ('u2u16', a)), b)))),
(('ishr', 'a@8', b), ('i2i8', ('i2i16', ('ishr', ('i2i32', ('i2i16', a)), b)))),
(('ushr', 'a@8', b), ('u2u8', ('u2u16', ('ushr', ('u2u32', ('u2u16', a)), b)))),
]
# Midgard is able to type convert down by only one "step" per instruction; if
# NIR wants more than one step, we need to break up into multiple instructions
converts = []
for op in ('u2u', 'i2i', 'f2f', 'i2f', 'u2f', 'f2i', 'f2u'):
srcsz_max = 64
dstsz_max = 64
# 8 bit float doesn't exist
srcsz_min = 8 if op[0] != 'f' else 16
dstsz_min = 8 if op[2] != 'f' else 16
dstsz = dstsz_min
# Iterate over all possible destination and source sizes
while dstsz <= dstsz_max:
srcsz = srcsz_min
while srcsz <= srcsz_max:
# Size converter lowering is only needed if src and dst sizes are
# spaced by a factor > 2.
# Type converter lowering is needed as soon as src_size != dst_size
if srcsz != dstsz and ((srcsz * 2 != dstsz and srcsz != dstsz * 2) or op[0] != op[2]):
cursz = srcsz
rule = a
# When converting down we first do the type conversion followed
# by one or more size conversions. When converting up, we do
# the type conversion at the end. This way we don't have to
# deal with the fact that f2f8 doesn't exists.
sizeconvop = op[0] + '2' + op[0] if srcsz < dstsz else op[2] + '2' + op[2]
if srcsz > dstsz and op[0] != op[2]:
rule = (op + str(int(cursz)), rule)
while cursz != dstsz:
cursz = cursz / 2 if dstsz < srcsz else cursz * 2
rule = (sizeconvop + str(int(cursz)), rule)
if srcsz < dstsz and op[0] != op[2]:
rule = (op + str(int(cursz)), rule)
converts += [((op + str(int(dstsz)), 'a@' + str(int(srcsz))), rule)]
srcsz *= 2
dstsz *= 2
# Try to force constants to the right
constant_switch = [
# fge gets flipped to fle, so we invert to keep the order
(('fge', 'a', '#b'), (('inot', ('flt', a, b)))),
(('fge32', 'a', '#b'), (('inot', ('flt32', a, b)))),
(('ige32', 'a', '#b'), (('inot', ('ilt32', a, b)))),
(('uge32', 'a', '#b'), (('inot', ('ult32', a, b)))),
# fge gets mapped to fle with a flip
(('flt32', '#a', 'b'), ('inot', ('fge32', a, b))),
(('ilt32', '#a', 'b'), ('inot', ('ige32', a, b))),
(('ult32', '#a', 'b'), ('inot', ('uge32', a, b)))
]
# ..since the above switching happens after algebraic stuff is done
cancel_inot = [
(('inot', ('inot', a)), a)
]
# Midgard scales fsin/fcos arguments by pi.
# Pass must be run only once, after the main loop
scale_trig = [
(('fsin', a), ('fsin', ('fdiv', a, math.pi))),
(('fcos', a), ('fcos', ('fdiv', a, math.pi))),
]
def main():
parser = argparse.ArgumentParser()
parser.add_argument('-p', '--import-path', required=True)
args = parser.parse_args()
sys.path.insert(0, args.import_path)
run()
def run():
import nir_algebraic # pylint: disable=import-error
print('#include "midgard_nir.h"')
print(nir_algebraic.AlgebraicPass("midgard_nir_lower_algebraic_early",
algebraic).render())
print(nir_algebraic.AlgebraicPass("midgard_nir_lower_algebraic_late",
algebraic_late + converts + constant_switch).render())
print(nir_algebraic.AlgebraicPass("midgard_nir_scale_trig",
scale_trig).render())
print(nir_algebraic.AlgebraicPass("midgard_nir_cancel_inot",
cancel_inot).render())
if __name__ == '__main__':
main()
|
