e72b4a791cb052fb9e3518ba36d909c2761a28ff
[mesa.git] / src / compiler / nir / nir_opt_algebraic.py
1 #! /usr/bin/env python
2 #
3 # Copyright (C) 2014 Intel Corporation
4 #
5 # Permission is hereby granted, free of charge, to any person obtaining a
6 # copy of this software and associated documentation files (the "Software"),
7 # to deal in the Software without restriction, including without limitation
8 # the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 # and/or sell copies of the Software, and to permit persons to whom the
10 # Software is furnished to do so, subject to the following conditions:
11 #
12 # The above copyright notice and this permission notice (including the next
13 # paragraph) shall be included in all copies or substantial portions of the
14 # Software.
15 #
16 # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 # FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 # IN THE SOFTWARE.
23 #
24 # Authors:
25 # Jason Ekstrand (jason@jlekstrand.net)
26
27 import nir_algebraic
28
29 # Convenience variables
30 a = 'a'
31 b = 'b'
32 c = 'c'
33 d = 'd'
34
35 # Written in the form (<search>, <replace>) where <search> is an expression
36 # and <replace> is either an expression or a value. An expression is
37 # defined as a tuple of the form ([~]<op>, <src0>, <src1>, <src2>, <src3>)
38 # where each source is either an expression or a value. A value can be
39 # either a numeric constant or a string representing a variable name.
40 #
41 # If the opcode in a search expression is prefixed by a '~' character, this
42 # indicates that the operation is inexact. Such operations will only get
43 # applied to SSA values that do not have the exact bit set. This should be
44 # used by by any optimizations that are not bit-for-bit exact. It should not,
45 # however, be used for backend-requested lowering operations as those need to
46 # happen regardless of precision.
47 #
48 # Variable names are specified as "[#]name[@type]" where "#" inicates that
49 # the given variable will only match constants and the type indicates that
50 # the given variable will only match values from ALU instructions with the
51 # given output type.
52 #
53 # For constants, you have to be careful to make sure that it is the right
54 # type because python is unaware of the source and destination types of the
55 # opcodes.
56
57 optimizations = [
58 (('fneg', ('fneg', a)), a),
59 (('ineg', ('ineg', a)), a),
60 (('fabs', ('fabs', a)), ('fabs', a)),
61 (('fabs', ('fneg', a)), ('fabs', a)),
62 (('iabs', ('iabs', a)), ('iabs', a)),
63 (('iabs', ('ineg', a)), ('iabs', a)),
64 (('~fadd', a, 0.0), a),
65 (('iadd', a, 0), a),
66 (('usadd_4x8', a, 0), a),
67 (('usadd_4x8', a, ~0), ~0),
68 (('~fadd', ('fmul', a, b), ('fmul', a, c)), ('fmul', a, ('fadd', b, c))),
69 (('iadd', ('imul', a, b), ('imul', a, c)), ('imul', a, ('iadd', b, c))),
70 (('~fadd', ('fneg', a), a), 0.0),
71 (('iadd', ('ineg', a), a), 0),
72 (('iadd', ('ineg', a), ('iadd', a, b)), b),
73 (('iadd', a, ('iadd', ('ineg', a), b)), b),
74 (('~fadd', ('fneg', a), ('fadd', a, b)), b),
75 (('~fadd', a, ('fadd', ('fneg', a), b)), b),
76 (('~fmul', a, 0.0), 0.0),
77 (('imul', a, 0), 0),
78 (('umul_unorm_4x8', a, 0), 0),
79 (('umul_unorm_4x8', a, ~0), a),
80 (('fmul', a, 1.0), a),
81 (('imul', a, 1), a),
82 (('fmul', a, -1.0), ('fneg', a)),
83 (('imul', a, -1), ('ineg', a)),
84 (('~ffma', 0.0, a, b), b),
85 (('~ffma', a, 0.0, b), b),
86 (('~ffma', a, b, 0.0), ('fmul', a, b)),
87 (('ffma', a, 1.0, b), ('fadd', a, b)),
88 (('ffma', 1.0, a, b), ('fadd', a, b)),
89 (('~flrp', a, b, 0.0), a),
90 (('~flrp', a, b, 1.0), b),
91 (('~flrp', a, a, b), a),
92 (('~flrp', 0.0, a, b), ('fmul', a, b)),
93 (('~flrp', a, b, ('b2f', c)), ('bcsel', c, b, a), 'options->lower_flrp'),
94 (('flrp', a, b, c), ('fadd', ('fmul', c, ('fsub', b, a)), a), 'options->lower_flrp'),
95 (('ffract', a), ('fsub', a, ('ffloor', a)), 'options->lower_ffract'),
96 (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', ('b2f', c)))), ('fmul', b, ('b2f', c))), ('bcsel', c, b, a), 'options->lower_flrp'),
97 (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', c ))), ('fmul', b, c )), ('flrp', a, b, c), '!options->lower_flrp'),
98 (('~fadd', a, ('fmul', ('b2f', c), ('fadd', b, ('fneg', a)))), ('bcsel', c, b, a), 'options->lower_flrp'),
99 (('~fadd', a, ('fmul', c , ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp'),
100 (('ffma', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma'),
101 (('~fadd', ('fmul', a, b), c), ('ffma', a, b, c), '!options->lower_ffma'),
102 # Comparison simplifications
103 (('~inot', ('flt', a, b)), ('fge', a, b)),
104 (('~inot', ('fge', a, b)), ('flt', a, b)),
105 (('~inot', ('feq', a, b)), ('fne', a, b)),
106 (('~inot', ('fne', a, b)), ('feq', a, b)),
107 (('inot', ('ilt', a, b)), ('ige', a, b)),
108 (('inot', ('ige', a, b)), ('ilt', a, b)),
109 (('inot', ('ieq', a, b)), ('ine', a, b)),
110 (('inot', ('ine', a, b)), ('ieq', a, b)),
111
112 # 0.0 >= b2f(a)
113 # b2f(a) <= 0.0
114 # b2f(a) == 0.0 because b2f(a) can only be 0 or 1
115 # inot(a)
116 (('fge', 0.0, ('b2f', a)), ('inot', a)),
117
118 # 0.0 < fabs(a)
119 # fabs(a) > 0.0
120 # fabs(a) != 0.0 because fabs(a) must be >= 0
121 # a != 0.0
122 (('flt', 0.0, ('fabs', a)), ('fne', a, 0.0)),
123
124 (('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
125 (('bcsel', ('flt', b, a), b, a), ('fmin', a, b)),
126 (('bcsel', ('flt', a, b), b, a), ('fmax', a, b)),
127 (('bcsel', ('inot', 'a@bool'), b, c), ('bcsel', a, c, b)),
128 (('bcsel', a, ('bcsel', a, b, c), d), ('bcsel', a, b, d)),
129 (('bcsel', a, True, 'b@bool'), ('ior', a, b)),
130 (('fmin', a, a), a),
131 (('fmax', a, a), a),
132 (('imin', a, a), a),
133 (('imax', a, a), a),
134 (('umin', a, a), a),
135 (('umax', a, a), a),
136 (('~fmin', ('fmax', a, 0.0), 1.0), ('fsat', a), '!options->lower_fsat'),
137 (('~fmax', ('fmin', a, 1.0), 0.0), ('fsat', a), '!options->lower_fsat'),
138 (('fsat', a), ('fmin', ('fmax', a, 0.0), 1.0), 'options->lower_fsat'),
139 (('fsat', ('fsat', a)), ('fsat', a)),
140 (('fmin', ('fmax', ('fmin', ('fmax', a, 0.0), 1.0), 0.0), 1.0), ('fmin', ('fmax', a, 0.0), 1.0)),
141 (('~ior', ('flt', a, b), ('flt', a, c)), ('flt', a, ('fmax', b, c))),
142 (('~ior', ('flt', a, c), ('flt', b, c)), ('flt', ('fmin', a, b), c)),
143 (('~ior', ('fge', a, b), ('fge', a, c)), ('fge', a, ('fmin', b, c))),
144 (('~ior', ('fge', a, c), ('fge', b, c)), ('fge', ('fmax', a, b), c)),
145 (('fabs', ('slt', a, b)), ('slt', a, b)),
146 (('fabs', ('sge', a, b)), ('sge', a, b)),
147 (('fabs', ('seq', a, b)), ('seq', a, b)),
148 (('fabs', ('sne', a, b)), ('sne', a, b)),
149 (('slt', a, b), ('b2f', ('flt', a, b)), 'options->lower_scmp'),
150 (('sge', a, b), ('b2f', ('fge', a, b)), 'options->lower_scmp'),
151 (('seq', a, b), ('b2f', ('feq', a, b)), 'options->lower_scmp'),
152 (('sne', a, b), ('b2f', ('fne', a, b)), 'options->lower_scmp'),
153 (('fne', ('fneg', a), a), ('fne', a, 0.0)),
154 (('feq', ('fneg', a), a), ('feq', a, 0.0)),
155 # Emulating booleans
156 (('imul', ('b2i', a), ('b2i', b)), ('b2i', ('iand', a, b))),
157 (('fmul', ('b2f', a), ('b2f', b)), ('b2f', ('iand', a, b))),
158 (('fsat', ('fadd', ('b2f', a), ('b2f', b))), ('b2f', ('ior', a, b))),
159 (('iand', 'a@bool', 1.0), ('b2f', a)),
160 (('flt', ('fneg', ('b2f', a)), 0), a), # Generated by TGSI KILL_IF.
161 (('flt', ('fsub', 0.0, ('b2f', a)), 0), a), # Generated by TGSI KILL_IF.
162 # Comparison with the same args. Note that these are not done for
163 # the float versions because NaN always returns false on float
164 # inequalities.
165 (('ilt', a, a), False),
166 (('ige', a, a), True),
167 (('ieq', a, a), True),
168 (('ine', a, a), False),
169 (('ult', a, a), False),
170 (('uge', a, a), True),
171 # Logical and bit operations
172 (('fand', a, 0.0), 0.0),
173 (('iand', a, a), a),
174 (('iand', a, ~0), a),
175 (('iand', a, 0), 0),
176 (('ior', a, a), a),
177 (('ior', a, 0), a),
178 (('fxor', a, a), 0.0),
179 (('ixor', a, a), 0),
180 (('ixor', a, 0), a),
181 (('inot', ('inot', a)), a),
182 # DeMorgan's Laws
183 (('iand', ('inot', a), ('inot', b)), ('inot', ('ior', a, b))),
184 (('ior', ('inot', a), ('inot', b)), ('inot', ('iand', a, b))),
185 # Shift optimizations
186 (('ishl', 0, a), 0),
187 (('ishl', a, 0), a),
188 (('ishr', 0, a), 0),
189 (('ishr', a, 0), a),
190 (('ushr', 0, a), 0),
191 (('ushr', a, 0), a),
192 (('iand', 0xff, ('ushr', a, 24)), ('ushr', a, 24)),
193 (('iand', 0xffff, ('ushr', a, 16)), ('ushr', a, 16)),
194 # Exponential/logarithmic identities
195 (('~fexp2', ('flog2', a)), a), # 2^lg2(a) = a
196 (('~flog2', ('fexp2', a)), a), # lg2(2^a) = a
197 (('fpow', a, b), ('fexp2', ('fmul', ('flog2', a), b)), 'options->lower_fpow'), # a^b = 2^(lg2(a)*b)
198 (('~fexp2', ('fmul', ('flog2', a), b)), ('fpow', a, b), '!options->lower_fpow'), # 2^(lg2(a)*b) = a^b
199 (('~fexp2', ('fadd', ('fmul', ('flog2', a), b), ('fmul', ('flog2', c), d))),
200 ('~fmul', ('fpow', a, b), ('fpow', c, d)), '!options->lower_fpow'), # 2^(lg2(a) * b + lg2(c) + d) = a^b * c^d
201 (('~fpow', a, 1.0), a),
202 (('~fpow', a, 2.0), ('fmul', a, a)),
203 (('~fpow', a, 4.0), ('fmul', ('fmul', a, a), ('fmul', a, a))),
204 (('~fpow', 2.0, a), ('fexp2', a)),
205 (('~fpow', ('fpow', a, 2.2), 0.454545), a),
206 (('~fpow', ('fabs', ('fpow', a, 2.2)), 0.454545), ('fabs', a)),
207 (('~fsqrt', ('fexp2', a)), ('fexp2', ('fmul', 0.5, a))),
208 (('~frcp', ('fexp2', a)), ('fexp2', ('fneg', a))),
209 (('~frsq', ('fexp2', a)), ('fexp2', ('fmul', -0.5, a))),
210 (('~flog2', ('fsqrt', a)), ('fmul', 0.5, ('flog2', a))),
211 (('~flog2', ('frcp', a)), ('fneg', ('flog2', a))),
212 (('~flog2', ('frsq', a)), ('fmul', -0.5, ('flog2', a))),
213 (('~flog2', ('fpow', a, b)), ('fmul', b, ('flog2', a))),
214 (('~fadd', ('flog2', a), ('flog2', b)), ('flog2', ('fmul', a, b))),
215 (('~fadd', ('flog2', a), ('fneg', ('flog2', b))), ('flog2', ('fdiv', a, b))),
216 (('~fmul', ('fexp2', a), ('fexp2', b)), ('fexp2', ('fadd', a, b))),
217 # Division and reciprocal
218 (('~fdiv', 1.0, a), ('frcp', a)),
219 (('fdiv', a, b), ('fmul', a, ('frcp', b)), 'options->lower_fdiv'),
220 (('~frcp', ('frcp', a)), a),
221 (('~frcp', ('fsqrt', a)), ('frsq', a)),
222 (('fsqrt', a), ('frcp', ('frsq', a)), 'options->lower_fsqrt'),
223 (('~frcp', ('frsq', a)), ('fsqrt', a), '!options->lower_fsqrt'),
224 # Boolean simplifications
225 (('ieq', 'a@bool', True), a),
226 (('ine', 'a@bool', True), ('inot', a)),
227 (('ine', 'a@bool', False), a),
228 (('ieq', 'a@bool', False), ('inot', 'a')),
229 (('bcsel', a, True, False), ('ine', a, 0)),
230 (('bcsel', a, False, True), ('ieq', a, 0)),
231 (('bcsel', True, b, c), b),
232 (('bcsel', False, b, c), c),
233 # The result of this should be hit by constant propagation and, in the
234 # next round of opt_algebraic, get picked up by one of the above two.
235 (('bcsel', '#a', b, c), ('bcsel', ('ine', 'a', 0), b, c)),
236
237 (('bcsel', a, b, b), b),
238 (('fcsel', a, b, b), b),
239
240 # Conversions
241 (('i2b', ('b2i', a)), a),
242 (('f2i', ('ftrunc', a)), ('f2i', a)),
243 (('f2u', ('ftrunc', a)), ('f2u', a)),
244 (('i2b', ('ineg', a)), ('i2b', a)),
245 (('i2b', ('iabs', a)), ('i2b', a)),
246 (('fabs', ('b2f', a)), ('b2f', a)),
247 (('iabs', ('b2i', a)), ('b2i', a)),
248
249 # Byte extraction
250 (('ushr', a, 24), ('extract_u8', a, 3), '!options->lower_extract_byte'),
251 (('iand', 0xff, ('ushr', a, 16)), ('extract_u8', a, 2), '!options->lower_extract_byte'),
252 (('iand', 0xff, ('ushr', a, 8)), ('extract_u8', a, 1), '!options->lower_extract_byte'),
253 (('iand', 0xff, a), ('extract_u8', a, 0), '!options->lower_extract_byte'),
254
255 # Word extraction
256 (('ushr', a, 16), ('extract_u16', a, 1), '!options->lower_extract_word'),
257 (('iand', 0xffff, a), ('extract_u16', a, 0), '!options->lower_extract_word'),
258
259 # Subtracts
260 (('~fsub', a, ('fsub', 0.0, b)), ('fadd', a, b)),
261 (('isub', a, ('isub', 0, b)), ('iadd', a, b)),
262 (('ussub_4x8', a, 0), a),
263 (('ussub_4x8', a, ~0), 0),
264 (('fsub', a, b), ('fadd', a, ('fneg', b)), 'options->lower_sub'),
265 (('isub', a, b), ('iadd', a, ('ineg', b)), 'options->lower_sub'),
266 (('fneg', a), ('fsub', 0.0, a), 'options->lower_negate'),
267 (('ineg', a), ('isub', 0, a), 'options->lower_negate'),
268 (('~fadd', a, ('fsub', 0.0, b)), ('fsub', a, b)),
269 (('iadd', a, ('isub', 0, b)), ('isub', a, b)),
270 (('fabs', ('fsub', 0.0, a)), ('fabs', a)),
271 (('iabs', ('isub', 0, a)), ('iabs', a)),
272
273 # Propagate negation up multiplication chains
274 (('fmul', ('fneg', a), b), ('fneg', ('fmul', a, b))),
275 (('imul', ('ineg', a), b), ('ineg', ('imul', a, b))),
276
277 # Misc. lowering
278 (('fmod', a, b), ('fsub', a, ('fmul', b, ('ffloor', ('fdiv', a, b)))), 'options->lower_fmod'),
279 (('uadd_carry', a, b), ('b2i', ('ult', ('iadd', a, b), a)), 'options->lower_uadd_carry'),
280 (('usub_borrow', a, b), ('b2i', ('ult', a, b)), 'options->lower_usub_borrow'),
281
282 (('bitfield_insert', 'base', 'insert', 'offset', 'bits'),
283 ('bcsel', ('ilt', 31, 'bits'), 'insert',
284 ('bfi', ('bfm', 'bits', 'offset'), 'insert', 'base')),
285 'options->lower_bitfield_insert'),
286
287 (('ibitfield_extract', 'value', 'offset', 'bits'),
288 ('bcsel', ('ilt', 31, 'bits'), 'value',
289 ('ibfe', 'value', 'offset', 'bits')),
290 'options->lower_bitfield_extract'),
291
292 (('ubitfield_extract', 'value', 'offset', 'bits'),
293 ('bcsel', ('ult', 31, 'bits'), 'value',
294 ('ubfe', 'value', 'offset', 'bits')),
295 'options->lower_bitfield_extract'),
296
297 (('extract_i8', a, b),
298 ('ishr', ('ishl', a, ('imul', ('isub', 3, b), 8)), 24),
299 'options->lower_extract_byte'),
300
301 (('extract_u8', a, b),
302 ('iand', ('ushr', a, ('imul', b, 8)), 0xff),
303 'options->lower_extract_byte'),
304
305 (('extract_i16', a, b),
306 ('ishr', ('ishl', a, ('imul', ('isub', 1, b), 16)), 16),
307 'options->lower_extract_word'),
308
309 (('extract_u16', a, b),
310 ('iand', ('ushr', a, ('imul', b, 16)), 0xffff),
311 'options->lower_extract_word'),
312
313 (('pack_unorm_2x16', 'v'),
314 ('pack_uvec2_to_uint',
315 ('f2u', ('fround_even', ('fmul', ('fsat', 'v'), 65535.0)))),
316 'options->lower_pack_unorm_2x16'),
317
318 (('pack_unorm_4x8', 'v'),
319 ('pack_uvec4_to_uint',
320 ('f2u', ('fround_even', ('fmul', ('fsat', 'v'), 255.0)))),
321 'options->lower_pack_unorm_4x8'),
322
323 (('pack_snorm_2x16', 'v'),
324 ('pack_uvec2_to_uint',
325 ('f2i', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 32767.0)))),
326 'options->lower_pack_snorm_2x16'),
327
328 (('pack_snorm_4x8', 'v'),
329 ('pack_uvec4_to_uint',
330 ('f2i', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 127.0)))),
331 'options->lower_pack_snorm_4x8'),
332
333 (('unpack_unorm_2x16', 'v'),
334 ('fdiv', ('u2f', ('vec2', ('extract_u16', 'v', 0),
335 ('extract_u16', 'v', 1))),
336 65535.0),
337 'options->lower_unpack_unorm_2x16'),
338
339 (('unpack_unorm_4x8', 'v'),
340 ('fdiv', ('u2f', ('vec4', ('extract_u8', 'v', 0),
341 ('extract_u8', 'v', 1),
342 ('extract_u8', 'v', 2),
343 ('extract_u8', 'v', 3))),
344 255.0),
345 'options->lower_unpack_unorm_4x8'),
346
347 (('unpack_snorm_2x16', 'v'),
348 ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f', ('vec2', ('extract_i16', 'v', 0),
349 ('extract_i16', 'v', 1))),
350 32767.0))),
351 'options->lower_unpack_snorm_2x16'),
352
353 (('unpack_snorm_4x8', 'v'),
354 ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f', ('vec4', ('extract_i8', 'v', 0),
355 ('extract_i8', 'v', 1),
356 ('extract_i8', 'v', 2),
357 ('extract_i8', 'v', 3))),
358 127.0))),
359 'options->lower_unpack_snorm_4x8'),
360 ]
361
362 # Unreal Engine 4 demo applications open-codes bitfieldReverse()
363 def bitfield_reverse(u):
364 step1 = ('ior', ('ishl', u, 16), ('ushr', u, 16))
365 step2 = ('ior', ('ishl', ('iand', step1, 0x00ff00ff), 8), ('ushr', ('iand', step1, 0xff00ff00), 8))
366 step3 = ('ior', ('ishl', ('iand', step2, 0x0f0f0f0f), 4), ('ushr', ('iand', step2, 0xf0f0f0f0), 4))
367 step4 = ('ior', ('ishl', ('iand', step3, 0x33333333), 2), ('ushr', ('iand', step3, 0xcccccccc), 2))
368 step5 = ('ior', ('ishl', ('iand', step4, 0x55555555), 1), ('ushr', ('iand', step4, 0xaaaaaaaa), 1))
369
370 return step5
371
372 optimizations += [(bitfield_reverse('x'), ('bitfield_reverse', 'x'))]
373
374
375 # Add optimizations to handle the case where the result of a ternary is
376 # compared to a constant. This way we can take things like
377 #
378 # (a ? 0 : 1) > 0
379 #
380 # and turn it into
381 #
382 # a ? (0 > 0) : (1 > 0)
383 #
384 # which constant folding will eat for lunch. The resulting ternary will
385 # further get cleaned up by the boolean reductions above and we will be
386 # left with just the original variable "a".
387 for op in ['flt', 'fge', 'feq', 'fne',
388 'ilt', 'ige', 'ieq', 'ine', 'ult', 'uge']:
389 optimizations += [
390 ((op, ('bcsel', 'a', '#b', '#c'), '#d'),
391 ('bcsel', 'a', (op, 'b', 'd'), (op, 'c', 'd'))),
392 ((op, '#d', ('bcsel', a, '#b', '#c')),
393 ('bcsel', 'a', (op, 'd', 'b'), (op, 'd', 'c'))),
394 ]
395
396 # This section contains "late" optimizations that should be run after the
397 # regular optimizations have finished. Optimizations should go here if
398 # they help code generation but do not necessarily produce code that is
399 # more easily optimizable.
400 late_optimizations = [
401 # Most of these optimizations aren't quite safe when you get infinity or
402 # Nan involved but the first one should be fine.
403 (('flt', ('fadd', a, b), 0.0), ('flt', a, ('fneg', b))),
404 (('~fge', ('fadd', a, b), 0.0), ('fge', a, ('fneg', b))),
405 (('~feq', ('fadd', a, b), 0.0), ('feq', a, ('fneg', b))),
406 (('~fne', ('fadd', a, b), 0.0), ('fne', a, ('fneg', b))),
407
408 (('fdot2', a, b), ('fdot_replicated2', a, b), 'options->fdot_replicates'),
409 (('fdot3', a, b), ('fdot_replicated3', a, b), 'options->fdot_replicates'),
410 (('fdot4', a, b), ('fdot_replicated4', a, b), 'options->fdot_replicates'),
411 (('fdph', a, b), ('fdph_replicated', a, b), 'options->fdot_replicates'),
412 ]
413
414 print nir_algebraic.AlgebraicPass("nir_opt_algebraic", optimizations).render()
415 print nir_algebraic.AlgebraicPass("nir_opt_algebraic_late",
416 late_optimizations).render()