2 # Copyright (C) 2014 Intel Corporation
4 # Permission is hereby granted, free of charge, to any person obtaining a
5 # copy of this software and associated documentation files (the "Software"),
6 # to deal in the Software without restriction, including without limitation
7 # the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 # and/or sell copies of the Software, and to permit persons to whom the
9 # Software is furnished to do so, subject to the following conditions:
11 # The above copyright notice and this permission notice (including the next
12 # paragraph) shall be included in all copies or substantial portions of the
15 # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 # THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 # FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 # Jason Ekstrand (jason@jlekstrand.net)
29 # Convenience variables
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.
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.
48 # Variable names are specified as "[#]name[@type][(cond)]" where "#" inicates
49 # that 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, and (cond) specifies an additional condition function
52 # (see nir_search_helpers.h).
54 # For constants, you have to be careful to make sure that it is the right
55 # type because python is unaware of the source and destination types of the
58 # All expression types can have a bit-size specified. For opcodes, this
59 # looks like "op@32", for variables it is "a@32" or "a@uint32" to specify a
60 # type and size, and for literals, you can write "2.0@32". In the search half
61 # of the expression this indicates that it should only match that particular
62 # bit-size. In the replace half of the expression this indicates that the
63 # constructed value should have that bit-size.
67 (('imul', a
, '#b@32(is_pos_power_of_two)'), ('ishl', a
, ('find_lsb', b
))),
68 (('imul', a
, '#b@32(is_neg_power_of_two)'), ('ineg', ('ishl', a
, ('find_lsb', ('iabs', b
))))),
73 (('udiv', a
, '#b@32(is_pos_power_of_two)'), ('ushr', a
, ('find_lsb', b
))),
74 (('idiv', a
, '#b@32(is_pos_power_of_two)'), ('imul', ('isign', a
), ('ushr', ('iabs', a
), ('find_lsb', b
))), 'options->lower_idiv'),
75 (('idiv', a
, '#b@32(is_neg_power_of_two)'), ('ineg', ('imul', ('isign', a
), ('ushr', ('iabs', a
), ('find_lsb', ('iabs', b
))))), 'options->lower_idiv'),
76 (('umod', a
, '#b(is_pos_power_of_two)'), ('iand', a
, ('isub', b
, 1))),
78 (('fneg', ('fneg', a
)), a
),
79 (('ineg', ('ineg', a
)), a
),
80 (('fabs', ('fabs', a
)), ('fabs', a
)),
81 (('fabs', ('fneg', a
)), ('fabs', a
)),
82 (('fabs', ('u2f32', a
)), ('u2f32', a
)),
83 (('iabs', ('iabs', a
)), ('iabs', a
)),
84 (('iabs', ('ineg', a
)), ('iabs', a
)),
85 (('~fadd', a
, 0.0), a
),
87 (('usadd_4x8', a
, 0), a
),
88 (('usadd_4x8', a
, ~
0), ~
0),
89 (('~fadd', ('fmul', a
, b
), ('fmul', a
, c
)), ('fmul', a
, ('fadd', b
, c
))),
90 (('iadd', ('imul', a
, b
), ('imul', a
, c
)), ('imul', a
, ('iadd', b
, c
))),
91 (('~fadd', ('fneg', a
), a
), 0.0),
92 (('iadd', ('ineg', a
), a
), 0),
93 (('iadd', ('ineg', a
), ('iadd', a
, b
)), b
),
94 (('iadd', a
, ('iadd', ('ineg', a
), b
)), b
),
95 (('~fadd', ('fneg', a
), ('fadd', a
, b
)), b
),
96 (('~fadd', a
, ('fadd', ('fneg', a
), b
)), b
),
97 (('~fmul', a
, 0.0), 0.0),
99 (('umul_unorm_4x8', a
, 0), 0),
100 (('umul_unorm_4x8', a
, ~
0), a
),
101 (('fmul', a
, 1.0), a
),
103 (('fmul', a
, -1.0), ('fneg', a
)),
104 (('imul', a
, -1), ('ineg', a
)),
105 (('~ffma', 0.0, a
, b
), b
),
106 (('~ffma', a
, 0.0, b
), b
),
107 (('~ffma', a
, b
, 0.0), ('fmul', a
, b
)),
108 (('ffma', a
, 1.0, b
), ('fadd', a
, b
)),
109 (('ffma', 1.0, a
, b
), ('fadd', a
, b
)),
110 (('~flrp', a
, b
, 0.0), a
),
111 (('~flrp', a
, b
, 1.0), b
),
112 (('~flrp', a
, a
, b
), a
),
113 (('~flrp', 0.0, a
, b
), ('fmul', a
, b
)),
114 (('~flrp', a
, b
, ('b2f', c
)), ('bcsel', c
, b
, a
), 'options->lower_flrp32'),
115 (('~flrp', a
, 0.0, c
), ('fadd', ('fmul', ('fneg', a
), c
), a
)),
116 (('flrp@32', a
, b
, c
), ('fadd', ('fmul', c
, ('fsub', b
, a
)), a
), 'options->lower_flrp32'),
117 (('flrp@64', a
, b
, c
), ('fadd', ('fmul', c
, ('fsub', b
, a
)), a
), 'options->lower_flrp64'),
118 (('ffract', a
), ('fsub', a
, ('ffloor', a
)), 'options->lower_ffract'),
119 (('~fadd', ('fmul', a
, ('fadd', 1.0, ('fneg', ('b2f', c
)))), ('fmul', b
, ('b2f', c
))), ('bcsel', c
, b
, a
), 'options->lower_flrp32'),
120 (('~fadd@32', ('fmul', a
, ('fadd', 1.0, ('fneg', c
))), ('fmul', b
, c
)), ('flrp', a
, b
, c
), '!options->lower_flrp32'),
121 (('~fadd@64', ('fmul', a
, ('fadd', 1.0, ('fneg', c
))), ('fmul', b
, c
)), ('flrp', a
, b
, c
), '!options->lower_flrp64'),
122 (('~fadd', a
, ('fmul', ('b2f', c
), ('fadd', b
, ('fneg', a
)))), ('bcsel', c
, b
, a
), 'options->lower_flrp32'),
123 (('~fadd@32', a
, ('fmul', c
, ('fadd', b
, ('fneg', a
)))), ('flrp', a
, b
, c
), '!options->lower_flrp32'),
124 (('~fadd@64', a
, ('fmul', c
, ('fadd', b
, ('fneg', a
)))), ('flrp', a
, b
, c
), '!options->lower_flrp64'),
125 (('ffma', a
, b
, c
), ('fadd', ('fmul', a
, b
), c
), 'options->lower_ffma'),
126 (('~fadd', ('fmul', a
, b
), c
), ('ffma', a
, b
, c
), 'options->fuse_ffma'),
128 (('fdot4', ('vec4', a
, b
, c
, 1.0), d
), ('fdph', ('vec3', a
, b
, c
), d
)),
129 (('fdot4', ('vec4', a
, 0.0, 0.0, 0.0), b
), ('fmul', a
, b
)),
130 (('fdot4', ('vec4', a
, b
, 0.0, 0.0), c
), ('fdot2', ('vec2', a
, b
), c
)),
131 (('fdot4', ('vec4', a
, b
, c
, 0.0), d
), ('fdot3', ('vec3', a
, b
, c
), d
)),
133 (('fdot3', ('vec3', a
, 0.0, 0.0), b
), ('fmul', a
, b
)),
134 (('fdot3', ('vec3', a
, b
, 0.0), c
), ('fdot2', ('vec2', a
, b
), c
)),
136 # (a * #b + #c) << #d
137 # ((a * #b) << #d) + (#c << #d)
138 # (a * (#b << #d)) + (#c << #d)
139 (('ishl', ('iadd', ('imul', a
, '#b'), '#c'), '#d'),
140 ('iadd', ('imul', a
, ('ishl', b
, d
)), ('ishl', c
, d
))),
144 (('ishl', ('imul', a
, '#b'), '#c'), ('imul', a
, ('ishl', b
, c
))),
146 # Comparison simplifications
147 (('~inot', ('flt', a
, b
)), ('fge', a
, b
)),
148 (('~inot', ('fge', a
, b
)), ('flt', a
, b
)),
149 (('~inot', ('feq', a
, b
)), ('fne', a
, b
)),
150 (('~inot', ('fne', a
, b
)), ('feq', a
, b
)),
151 (('inot', ('ilt', a
, b
)), ('ige', a
, b
)),
152 (('inot', ('ige', a
, b
)), ('ilt', a
, b
)),
153 (('inot', ('ieq', a
, b
)), ('ine', a
, b
)),
154 (('inot', ('ine', a
, b
)), ('ieq', a
, b
)),
158 # b2f(a) == 0.0 because b2f(a) can only be 0 or 1
160 (('fge', 0.0, ('b2f', a
)), ('inot', a
)),
162 (('fge', ('fneg', ('b2f', a
)), 0.0), ('inot', a
)),
164 (('~flt', ('fadd', a
, b
), a
), ('flt', b
, 0.0)),
165 (('~fge', ('fadd', a
, b
), a
), ('fge', b
, 0.0)),
166 (('~feq', ('fadd', a
, b
), a
), ('feq', b
, 0.0)),
167 (('~fne', ('fadd', a
, b
), a
), ('fne', b
, 0.0)),
169 # Cannot remove the addition from ilt or ige due to overflow.
170 (('ieq', ('iadd', a
, b
), a
), ('ieq', b
, 0)),
171 (('ine', ('iadd', a
, b
), a
), ('ine', b
, 0)),
173 # fmin(-b2f(a), b) >= 0.0
174 # -b2f(a) >= 0.0 && b >= 0.0
175 # -b2f(a) == 0.0 && b >= 0.0 -b2f can only be 0 or -1, never >0
176 # b2f(a) == 0.0 && b >= 0.0
177 # a == False && b >= 0.0
180 # The fge in the second replacement is not a typo. I leave the proof that
181 # "fmin(-b2f(a), b) >= 0 <=> fmin(-b2f(a), b) == 0" as an exercise for the
183 (('fge', ('fmin', ('fneg', ('b2f', a
)), b
), 0.0), ('iand', ('inot', a
), ('fge', b
, 0.0))),
184 (('feq', ('fmin', ('fneg', ('b2f', a
)), b
), 0.0), ('iand', ('inot', a
), ('fge', b
, 0.0))),
186 (('feq', ('b2f', a
), 0.0), ('inot', a
)),
187 (('fne', ('b2f', a
), 0.0), a
),
188 (('ieq', ('b2i', a
), 0), ('inot', a
)),
189 (('ine', ('b2i', a
), 0), a
),
193 # fabs(a) != 0.0 because fabs(a) must be >= 0
195 (('flt', 0.0, ('fabs', a
)), ('fne', a
, 0.0)),
197 (('fmax', ('b2f(is_used_once)', a
), ('b2f', b
)), ('b2f', ('ior', a
, b
))),
198 (('fmax', ('fneg(is_used_once)', ('b2f(is_used_once)', a
)), ('fneg', ('b2f', b
))), ('fneg', ('b2f', ('ior', a
, b
)))),
199 (('fmin', ('b2f(is_used_once)', a
), ('b2f', b
)), ('b2f', ('iand', a
, b
))),
200 (('fmin', ('fneg(is_used_once)', ('b2f(is_used_once)', a
)), ('fneg', ('b2f', b
))), ('fneg', ('b2f', ('iand', a
, b
)))),
203 # bcsel(a, fmin(b2f(a), b), fmin(b2f(a), b))
204 # bcsel(a, fmin(b2f(True), b), fmin(b2f(False), b))
205 # bcsel(a, fmin(1.0, b), fmin(0.0, b))
207 # Since b is a constant, constant folding will eliminate the fmin and the
208 # fmax. If b is > 1.0, the bcsel will be replaced with a b2f.
209 (('fmin', ('b2f', a
), '#b'), ('bcsel', a
, ('fmin', b
, 1.0), ('fmin', b
, 0.0))),
211 (('flt', ('fadd(is_used_once)', a
, ('fneg', b
)), 0.0), ('flt', a
, b
)),
213 (('fge', ('fneg', ('fabs', a
)), 0.0), ('feq', a
, 0.0)),
214 (('~bcsel', ('flt', b
, a
), b
, a
), ('fmin', a
, b
)),
215 (('~bcsel', ('flt', a
, b
), b
, a
), ('fmax', a
, b
)),
216 (('~bcsel', ('fge', a
, b
), b
, a
), ('fmin', a
, b
)),
217 (('~bcsel', ('fge', b
, a
), b
, a
), ('fmax', a
, b
)),
218 (('bcsel', ('inot', a
), b
, c
), ('bcsel', a
, c
, b
)),
219 (('bcsel', a
, ('bcsel', a
, b
, c
), d
), ('bcsel', a
, b
, d
)),
220 (('bcsel', a
, True, 'b@bool'), ('ior', a
, b
)),
227 (('fmin', a
, ('fneg', a
)), ('fneg', ('fabs', a
))),
228 (('imin', a
, ('ineg', a
)), ('ineg', ('iabs', a
))),
229 (('fmin', a
, ('fneg', ('fabs', a
))), ('fneg', ('fabs', a
))),
230 (('imin', a
, ('ineg', ('iabs', a
))), ('ineg', ('iabs', a
))),
231 (('fmin', a
, ('fabs', a
)), a
),
232 (('imin', a
, ('iabs', a
)), a
),
233 (('fmax', a
, ('fneg', ('fabs', a
))), a
),
234 (('imax', a
, ('ineg', ('iabs', a
))), a
),
235 (('fmax', a
, ('fabs', a
)), ('fabs', a
)),
236 (('imax', a
, ('iabs', a
)), ('iabs', a
)),
237 (('fmax', a
, ('fneg', a
)), ('fabs', a
)),
238 (('imax', a
, ('ineg', a
)), ('iabs', a
)),
239 (('~fmin', ('fmax', a
, 0.0), 1.0), ('fsat', a
), '!options->lower_fsat'),
240 (('~fmax', ('fmin', a
, 1.0), 0.0), ('fsat', a
), '!options->lower_fsat'),
241 (('fsat', a
), ('fmin', ('fmax', a
, 0.0), 1.0), 'options->lower_fsat'),
242 (('fsat', ('fsat', a
)), ('fsat', a
)),
243 (('fmin', ('fmax', ('fmin', ('fmax', a
, b
), c
), b
), c
), ('fmin', ('fmax', a
, b
), c
)),
244 (('imin', ('imax', ('imin', ('imax', a
, b
), c
), b
), c
), ('imin', ('imax', a
, b
), c
)),
245 (('umin', ('umax', ('umin', ('umax', a
, b
), c
), b
), c
), ('umin', ('umax', a
, b
), c
)),
246 (('fmax', ('fsat', a
), '#b@32(is_zero_to_one)'), ('fsat', ('fmax', a
, b
))),
247 (('fmin', ('fsat', a
), '#b@32(is_zero_to_one)'), ('fsat', ('fmin', a
, b
))),
248 (('extract_u8', ('imin', ('imax', a
, 0), 0xff), 0), ('imin', ('imax', a
, 0), 0xff)),
249 (('~ior', ('flt(is_used_once)', a
, b
), ('flt', a
, c
)), ('flt', a
, ('fmax', b
, c
))),
250 (('~ior', ('flt(is_used_once)', a
, c
), ('flt', b
, c
)), ('flt', ('fmin', a
, b
), c
)),
251 (('~ior', ('fge(is_used_once)', a
, b
), ('fge', a
, c
)), ('fge', a
, ('fmin', b
, c
))),
252 (('~ior', ('fge(is_used_once)', a
, c
), ('fge', b
, c
)), ('fge', ('fmax', a
, b
), c
)),
253 (('~ior', ('flt', a
, '#b'), ('flt', a
, '#c')), ('flt', a
, ('fmax', b
, c
))),
254 (('~ior', ('flt', '#a', c
), ('flt', '#b', c
)), ('flt', ('fmin', a
, b
), c
)),
255 (('~ior', ('fge', a
, '#b'), ('fge', a
, '#c')), ('fge', a
, ('fmin', b
, c
))),
256 (('~ior', ('fge', '#a', c
), ('fge', '#b', c
)), ('fge', ('fmax', a
, b
), c
)),
257 (('~iand', ('flt(is_used_once)', a
, b
), ('flt', a
, c
)), ('flt', a
, ('fmin', b
, c
))),
258 (('~iand', ('flt(is_used_once)', a
, c
), ('flt', b
, c
)), ('flt', ('fmax', a
, b
), c
)),
259 (('~iand', ('fge(is_used_once)', a
, b
), ('fge', a
, c
)), ('fge', a
, ('fmax', b
, c
))),
260 (('~iand', ('fge(is_used_once)', a
, c
), ('fge', b
, c
)), ('fge', ('fmin', a
, b
), c
)),
261 (('~iand', ('flt', a
, '#b'), ('flt', a
, '#c')), ('flt', a
, ('fmin', b
, c
))),
262 (('~iand', ('flt', '#a', c
), ('flt', '#b', c
)), ('flt', ('fmax', a
, b
), c
)),
263 (('~iand', ('fge', a
, '#b'), ('fge', a
, '#c')), ('fge', a
, ('fmax', b
, c
))),
264 (('~iand', ('fge', '#a', c
), ('fge', '#b', c
)), ('fge', ('fmin', a
, b
), c
)),
266 (('ior', ('ilt(is_used_once)', a
, b
), ('ilt', a
, c
)), ('ilt', a
, ('imax', b
, c
))),
267 (('ior', ('ilt(is_used_once)', a
, c
), ('ilt', b
, c
)), ('ilt', ('imin', a
, b
), c
)),
268 (('ior', ('ige(is_used_once)', a
, b
), ('ige', a
, c
)), ('ige', a
, ('imin', b
, c
))),
269 (('ior', ('ige(is_used_once)', a
, c
), ('ige', b
, c
)), ('ige', ('imax', a
, b
), c
)),
270 (('ior', ('ult(is_used_once)', a
, b
), ('ult', a
, c
)), ('ult', a
, ('umax', b
, c
))),
271 (('ior', ('ult(is_used_once)', a
, c
), ('ult', b
, c
)), ('ult', ('umin', a
, b
), c
)),
272 (('ior', ('uge(is_used_once)', a
, b
), ('uge', a
, c
)), ('uge', a
, ('umin', b
, c
))),
273 (('ior', ('uge(is_used_once)', a
, c
), ('uge', b
, c
)), ('uge', ('umax', a
, b
), c
)),
274 (('iand', ('ilt(is_used_once)', a
, b
), ('ilt', a
, c
)), ('ilt', a
, ('imin', b
, c
))),
275 (('iand', ('ilt(is_used_once)', a
, c
), ('ilt', b
, c
)), ('ilt', ('imax', a
, b
), c
)),
276 (('iand', ('ige(is_used_once)', a
, b
), ('ige', a
, c
)), ('ige', a
, ('imax', b
, c
))),
277 (('iand', ('ige(is_used_once)', a
, c
), ('ige', b
, c
)), ('ige', ('imin', a
, b
), c
)),
278 (('iand', ('ult(is_used_once)', a
, b
), ('ult', a
, c
)), ('ult', a
, ('umin', b
, c
))),
279 (('iand', ('ult(is_used_once)', a
, c
), ('ult', b
, c
)), ('ult', ('umax', a
, b
), c
)),
280 (('iand', ('uge(is_used_once)', a
, b
), ('uge', a
, c
)), ('uge', a
, ('umax', b
, c
))),
281 (('iand', ('uge(is_used_once)', a
, c
), ('uge', b
, c
)), ('uge', ('umin', a
, b
), c
)),
283 # These patterns can result when (a < b || a < c) => (a < min(b, c))
284 # transformations occur before constant propagation and loop-unrolling.
285 (('~flt', a
, ('fmax', b
, a
)), ('flt', a
, b
)),
286 (('~flt', ('fmin', a
, b
), a
), ('flt', b
, a
)),
287 (('~fge', a
, ('fmin', b
, a
)), True),
288 (('~fge', ('fmax', a
, b
), a
), True),
289 (('~flt', a
, ('fmin', b
, a
)), False),
290 (('~flt', ('fmax', a
, b
), a
), False),
292 (('ilt', a
, ('imax', b
, a
)), ('ilt', a
, b
)),
293 (('ilt', ('imin', a
, b
), a
), ('ilt', b
, a
)),
294 (('ige', a
, ('imin', b
, a
)), True),
295 (('ige', ('imax', a
, b
), a
), True),
296 (('ult', a
, ('umax', b
, a
)), ('ult', a
, b
)),
297 (('ult', ('umin', a
, b
), a
), ('ult', b
, a
)),
298 (('uge', a
, ('umin', b
, a
)), True),
299 (('uge', ('umax', a
, b
), a
), True),
301 (('ilt', '#a', ('imin', '#b', c
)), ('iand', ('ilt', a
, b
), ('ilt', a
, c
))),
302 (('ilt', ('imax', '#a', b
), '#c'), ('iand', ('ilt', a
, c
), ('ilt', b
, c
))),
303 (('ige', '#a', ('imax', '#b', c
)), ('iand', ('ige', a
, b
), ('ige', a
, c
))),
304 (('ige', ('imin', '#a', b
), '#c'), ('iand', ('ige', a
, c
), ('ige', b
, c
))),
305 (('ult', '#a', ('umin', '#b', c
)), ('iand', ('ult', a
, b
), ('ult', a
, c
))),
306 (('ult', ('umax', '#a', b
), '#c'), ('iand', ('ult', a
, c
), ('ult', b
, c
))),
307 (('uge', '#a', ('umax', '#b', c
)), ('iand', ('uge', a
, b
), ('uge', a
, c
))),
308 (('uge', ('umin', '#a', b
), '#c'), ('iand', ('uge', a
, c
), ('uge', b
, c
))),
310 (('fabs', ('slt', a
, b
)), ('slt', a
, b
)),
311 (('fabs', ('sge', a
, b
)), ('sge', a
, b
)),
312 (('fabs', ('seq', a
, b
)), ('seq', a
, b
)),
313 (('fabs', ('sne', a
, b
)), ('sne', a
, b
)),
314 (('slt', a
, b
), ('b2f', ('flt', a
, b
)), 'options->lower_scmp'),
315 (('sge', a
, b
), ('b2f', ('fge', a
, b
)), 'options->lower_scmp'),
316 (('seq', a
, b
), ('b2f', ('feq', a
, b
)), 'options->lower_scmp'),
317 (('sne', a
, b
), ('b2f', ('fne', a
, b
)), 'options->lower_scmp'),
318 (('fne', ('fneg', a
), a
), ('fne', a
, 0.0)),
319 (('feq', ('fneg', a
), a
), ('feq', a
, 0.0)),
321 (('imul', ('b2i', a
), ('b2i', b
)), ('b2i', ('iand', a
, b
))),
322 (('fmul', ('b2f', a
), ('b2f', b
)), ('b2f', ('iand', a
, b
))),
323 (('fsat', ('fadd', ('b2f', a
), ('b2f', b
))), ('b2f', ('ior', a
, b
))),
324 (('iand', 'a@bool', 1.0), ('b2f', a
), '!options->lower_b2f'),
325 # True/False are ~0 and 0 in NIR. b2i of True is 1, and -1 is ~0 (True).
326 (('ineg', ('b2i@32', a
)), a
),
327 (('flt', ('fneg', ('b2f', a
)), 0), a
), # Generated by TGSI KILL_IF.
328 (('flt', ('fsub', 0.0, ('b2f', a
)), 0), a
), # Generated by TGSI KILL_IF.
329 # Comparison with the same args. Note that these are not done for
330 # the float versions because NaN always returns false on float
332 (('ilt', a
, a
), False),
333 (('ige', a
, a
), True),
334 (('ieq', a
, a
), True),
335 (('ine', a
, a
), False),
336 (('ult', a
, a
), False),
337 (('uge', a
, a
), True),
338 # Logical and bit operations
339 (('fand', a
, 0.0), 0.0),
341 (('iand', a
, ~
0), a
),
345 (('ior', a
, True), True),
346 (('fxor', a
, a
), 0.0),
349 (('inot', ('inot', a
)), a
),
351 (('iand', ('inot', a
), ('inot', b
)), ('inot', ('ior', a
, b
))),
352 (('ior', ('inot', a
), ('inot', b
)), ('inot', ('iand', a
, b
))),
353 # Shift optimizations
360 (('iand', 0xff, ('ushr@32', a
, 24)), ('ushr', a
, 24)),
361 (('iand', 0xffff, ('ushr@32', a
, 16)), ('ushr', a
, 16)),
362 # Exponential/logarithmic identities
363 (('~fexp2', ('flog2', a
)), a
), # 2^lg2(a) = a
364 (('~flog2', ('fexp2', a
)), a
), # lg2(2^a) = a
365 (('fpow', a
, b
), ('fexp2', ('fmul', ('flog2', a
), b
)), 'options->lower_fpow'), # a^b = 2^(lg2(a)*b)
366 (('~fexp2', ('fmul', ('flog2', a
), b
)), ('fpow', a
, b
), '!options->lower_fpow'), # 2^(lg2(a)*b) = a^b
367 (('~fexp2', ('fadd', ('fmul', ('flog2', a
), b
), ('fmul', ('flog2', c
), d
))),
368 ('~fmul', ('fpow', a
, b
), ('fpow', c
, d
)), '!options->lower_fpow'), # 2^(lg2(a) * b + lg2(c) + d) = a^b * c^d
369 (('~fexp2', ('fmul', ('flog2', a
), 2.0)), ('fmul', a
, a
)),
370 (('~fexp2', ('fmul', ('flog2', a
), 4.0)), ('fmul', ('fmul', a
, a
), ('fmul', a
, a
))),
371 (('~fpow', a
, 1.0), a
),
372 (('~fpow', a
, 2.0), ('fmul', a
, a
)),
373 (('~fpow', a
, 4.0), ('fmul', ('fmul', a
, a
), ('fmul', a
, a
))),
374 (('~fpow', 2.0, a
), ('fexp2', a
)),
375 (('~fpow', ('fpow', a
, 2.2), 0.454545), a
),
376 (('~fpow', ('fabs', ('fpow', a
, 2.2)), 0.454545), ('fabs', a
)),
377 (('~fsqrt', ('fexp2', a
)), ('fexp2', ('fmul', 0.5, a
))),
378 (('~frcp', ('fexp2', a
)), ('fexp2', ('fneg', a
))),
379 (('~frsq', ('fexp2', a
)), ('fexp2', ('fmul', -0.5, a
))),
380 (('~flog2', ('fsqrt', a
)), ('fmul', 0.5, ('flog2', a
))),
381 (('~flog2', ('frcp', a
)), ('fneg', ('flog2', a
))),
382 (('~flog2', ('frsq', a
)), ('fmul', -0.5, ('flog2', a
))),
383 (('~flog2', ('fpow', a
, b
)), ('fmul', b
, ('flog2', a
))),
384 (('~fmul', ('fexp2(is_used_once)', a
), ('fexp2(is_used_once)', b
)), ('fexp2', ('fadd', a
, b
))),
385 # Division and reciprocal
386 (('~fdiv', 1.0, a
), ('frcp', a
)),
387 (('fdiv', a
, b
), ('fmul', a
, ('frcp', b
)), 'options->lower_fdiv'),
388 (('~frcp', ('frcp', a
)), a
),
389 (('~frcp', ('fsqrt', a
)), ('frsq', a
)),
390 (('fsqrt', a
), ('frcp', ('frsq', a
)), 'options->lower_fsqrt'),
391 (('~frcp', ('frsq', a
)), ('fsqrt', a
), '!options->lower_fsqrt'),
392 # Boolean simplifications
393 (('ieq', 'a@bool', True), a
),
394 (('ine(is_not_used_by_if)', 'a@bool', True), ('inot', a
)),
395 (('ine', 'a@bool', False), a
),
396 (('ieq(is_not_used_by_if)', 'a@bool', False), ('inot', 'a')),
397 (('bcsel', a
, True, False), a
),
398 (('bcsel', a
, False, True), ('inot', a
)),
399 (('bcsel@32', a
, 1.0, 0.0), ('b2f', a
)),
400 (('bcsel@32', a
, 0.0, 1.0), ('b2f', ('inot', a
))),
401 (('bcsel@32', a
, -1.0, -0.0), ('fneg', ('b2f', a
))),
402 (('bcsel@32', a
, -0.0, -1.0), ('fneg', ('b2f', ('inot', a
)))),
403 (('bcsel', True, b
, c
), b
),
404 (('bcsel', False, b
, c
), c
),
405 (('bcsel', a
, ('b2f(is_used_once)', b
), ('b2f', c
)), ('b2f', ('bcsel', a
, b
, c
))),
406 # The result of this should be hit by constant propagation and, in the
407 # next round of opt_algebraic, get picked up by one of the above two.
408 (('bcsel', '#a', b
, c
), ('bcsel', ('ine', 'a', 0), b
, c
)),
410 (('bcsel', a
, b
, b
), b
),
411 (('fcsel', a
, b
, b
), b
),
414 (('i2b', ('b2i', a
)), a
),
415 (('i2b', 'a@bool'), a
),
416 (('f2i32', ('ftrunc', a
)), ('f2i32', a
)),
417 (('f2u32', ('ftrunc', a
)), ('f2u32', a
)),
418 (('i2b', ('ineg', a
)), ('i2b', a
)),
419 (('i2b', ('iabs', a
)), ('i2b', a
)),
420 (('fabs', ('b2f', a
)), ('b2f', a
)),
421 (('iabs', ('b2i', a
)), ('b2i', a
)),
423 # Packing and then unpacking does nothing
424 (('unpack_64_2x32_split_x', ('pack_64_2x32_split', a
, b
)), a
),
425 (('unpack_64_2x32_split_y', ('pack_64_2x32_split', a
, b
)), b
),
426 (('pack_64_2x32_split', ('unpack_64_2x32_split_x', a
),
427 ('unpack_64_2x32_split_y', a
)), a
),
430 (('ushr', a
, 24), ('extract_u8', a
, 3), '!options->lower_extract_byte'),
431 (('iand', 0xff, ('ushr', a
, 16)), ('extract_u8', a
, 2), '!options->lower_extract_byte'),
432 (('iand', 0xff, ('ushr', a
, 8)), ('extract_u8', a
, 1), '!options->lower_extract_byte'),
433 (('iand', 0xff, a
), ('extract_u8', a
, 0), '!options->lower_extract_byte'),
436 (('ushr', a
, 16), ('extract_u16', a
, 1), '!options->lower_extract_word'),
437 (('iand', 0xffff, a
), ('extract_u16', a
, 0), '!options->lower_extract_word'),
440 (('~fsub', a
, ('fsub', 0.0, b
)), ('fadd', a
, b
)),
441 (('isub', a
, ('isub', 0, b
)), ('iadd', a
, b
)),
442 (('ussub_4x8', a
, 0), a
),
443 (('ussub_4x8', a
, ~
0), 0),
444 (('fsub', a
, b
), ('fadd', a
, ('fneg', b
)), 'options->lower_sub'),
445 (('isub', a
, b
), ('iadd', a
, ('ineg', b
)), 'options->lower_sub'),
446 (('fneg', a
), ('fsub', 0.0, a
), 'options->lower_negate'),
447 (('ineg', a
), ('isub', 0, a
), 'options->lower_negate'),
448 (('~fadd', a
, ('fsub', 0.0, b
)), ('fsub', a
, b
)),
449 (('iadd', a
, ('isub', 0, b
)), ('isub', a
, b
)),
450 (('fabs', ('fsub', 0.0, a
)), ('fabs', a
)),
451 (('iabs', ('isub', 0, a
)), ('iabs', a
)),
453 # Propagate negation up multiplication chains
454 (('fmul', ('fneg', a
), b
), ('fneg', ('fmul', a
, b
))),
455 (('imul', ('ineg', a
), b
), ('ineg', ('imul', a
, b
))),
457 # Propagate constants up multiplication chains
458 (('~fmul(is_used_once)', ('fmul(is_used_once)', 'a(is_not_const)', 'b(is_not_const)'), '#c'), ('fmul', ('fmul', a
, c
), b
)),
459 (('imul(is_used_once)', ('imul(is_used_once)', 'a(is_not_const)', 'b(is_not_const)'), '#c'), ('imul', ('imul', a
, c
), b
)),
460 (('~fadd(is_used_once)', ('fadd(is_used_once)', 'a(is_not_const)', 'b(is_not_const)'), '#c'), ('fadd', ('fadd', a
, c
), b
)),
461 (('iadd(is_used_once)', ('iadd(is_used_once)', 'a(is_not_const)', 'b(is_not_const)'), '#c'), ('iadd', ('iadd', a
, c
), b
)),
463 # Reassociate constants in add/mul chains so they can be folded together.
464 # For now, we mostly only handle cases where the constants are separated by
465 # a single non-constant. We could do better eventually.
466 (('~fmul', '#a', ('fmul', b
, '#c')), ('fmul', ('fmul', a
, c
), b
)),
467 (('imul', '#a', ('imul', b
, '#c')), ('imul', ('imul', a
, c
), b
)),
468 (('~fadd', '#a', ('fadd', b
, '#c')), ('fadd', ('fadd', a
, c
), b
)),
469 (('~fadd', '#a', ('fneg', ('fadd', b
, '#c'))), ('fadd', ('fadd', a
, ('fneg', c
)), ('fneg', b
))),
470 (('iadd', '#a', ('iadd', b
, '#c')), ('iadd', ('iadd', a
, c
), b
)),
473 (('bcsel', ('ige', ('find_lsb', a
), 0), ('find_lsb', a
), -1), ('find_lsb', a
)),
474 (('bcsel', ('ige', ('ifind_msb', a
), 0), ('ifind_msb', a
), -1), ('ifind_msb', a
)),
475 (('bcsel', ('ige', ('ufind_msb', a
), 0), ('ufind_msb', a
), -1), ('ufind_msb', a
)),
477 (('bcsel', ('ine', a
, 0), ('find_lsb', a
), -1), ('find_lsb', a
)),
478 (('bcsel', ('ine', a
, 0), ('ifind_msb', a
), -1), ('ifind_msb', a
)),
479 (('bcsel', ('ine', a
, 0), ('ufind_msb', a
), -1), ('ufind_msb', a
)),
481 (('bcsel', ('ine', a
, -1), ('ifind_msb', a
), -1), ('ifind_msb', a
)),
484 (('fmod@32', a
, b
), ('fsub', a
, ('fmul', b
, ('ffloor', ('fdiv', a
, b
)))), 'options->lower_fmod32'),
485 (('fmod@64', a
, b
), ('fsub', a
, ('fmul', b
, ('ffloor', ('fdiv', a
, b
)))), 'options->lower_fmod64'),
486 (('frem', a
, b
), ('fsub', a
, ('fmul', b
, ('ftrunc', ('fdiv', a
, b
)))), 'options->lower_fmod32'),
487 (('uadd_carry@32', a
, b
), ('b2i', ('ult', ('iadd', a
, b
), a
)), 'options->lower_uadd_carry'),
488 (('usub_borrow@32', a
, b
), ('b2i', ('ult', a
, b
)), 'options->lower_usub_borrow'),
490 (('bitfield_insert', 'base', 'insert', 'offset', 'bits'),
491 ('bcsel', ('ilt', 31, 'bits'), 'insert',
492 ('bfi', ('bfm', 'bits', 'offset'), 'insert', 'base')),
493 'options->lower_bitfield_insert'),
495 (('ibitfield_extract', 'value', 'offset', 'bits'),
496 ('bcsel', ('ilt', 31, 'bits'), 'value',
497 ('ibfe', 'value', 'offset', 'bits')),
498 'options->lower_bitfield_extract'),
500 (('ubitfield_extract', 'value', 'offset', 'bits'),
501 ('bcsel', ('ult', 31, 'bits'), 'value',
502 ('ubfe', 'value', 'offset', 'bits')),
503 'options->lower_bitfield_extract'),
505 (('extract_i8', a
, 'b@32'),
506 ('ishr', ('ishl', a
, ('imul', ('isub', 3, b
), 8)), 24),
507 'options->lower_extract_byte'),
509 (('extract_u8', a
, 'b@32'),
510 ('iand', ('ushr', a
, ('imul', b
, 8)), 0xff),
511 'options->lower_extract_byte'),
513 (('extract_i16', a
, 'b@32'),
514 ('ishr', ('ishl', a
, ('imul', ('isub', 1, b
), 16)), 16),
515 'options->lower_extract_word'),
517 (('extract_u16', a
, 'b@32'),
518 ('iand', ('ushr', a
, ('imul', b
, 16)), 0xffff),
519 'options->lower_extract_word'),
521 (('pack_unorm_2x16', 'v'),
522 ('pack_uvec2_to_uint',
523 ('f2u32', ('fround_even', ('fmul', ('fsat', 'v'), 65535.0)))),
524 'options->lower_pack_unorm_2x16'),
526 (('pack_unorm_4x8', 'v'),
527 ('pack_uvec4_to_uint',
528 ('f2u32', ('fround_even', ('fmul', ('fsat', 'v'), 255.0)))),
529 'options->lower_pack_unorm_4x8'),
531 (('pack_snorm_2x16', 'v'),
532 ('pack_uvec2_to_uint',
533 ('f2i32', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 32767.0)))),
534 'options->lower_pack_snorm_2x16'),
536 (('pack_snorm_4x8', 'v'),
537 ('pack_uvec4_to_uint',
538 ('f2i32', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 127.0)))),
539 'options->lower_pack_snorm_4x8'),
541 (('unpack_unorm_2x16', 'v'),
542 ('fdiv', ('u2f32', ('vec2', ('extract_u16', 'v', 0),
543 ('extract_u16', 'v', 1))),
545 'options->lower_unpack_unorm_2x16'),
547 (('unpack_unorm_4x8', 'v'),
548 ('fdiv', ('u2f32', ('vec4', ('extract_u8', 'v', 0),
549 ('extract_u8', 'v', 1),
550 ('extract_u8', 'v', 2),
551 ('extract_u8', 'v', 3))),
553 'options->lower_unpack_unorm_4x8'),
555 (('unpack_snorm_2x16', 'v'),
556 ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f32', ('vec2', ('extract_i16', 'v', 0),
557 ('extract_i16', 'v', 1))),
559 'options->lower_unpack_snorm_2x16'),
561 (('unpack_snorm_4x8', 'v'),
562 ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f32', ('vec4', ('extract_i8', 'v', 0),
563 ('extract_i8', 'v', 1),
564 ('extract_i8', 'v', 2),
565 ('extract_i8', 'v', 3))),
567 'options->lower_unpack_snorm_4x8'),
570 invert
= {'feq': 'fne', 'fne': 'feq', 'fge': 'flt', 'flt': 'fge' }
572 for left
, right
in list(itertools
.combinations(invert
.keys(), 2)) + zip(invert
.keys(), invert
.keys()):
573 optimizations
.append((('inot', ('ior(is_used_once)', (left
, a
, b
), (right
, c
, d
))),
574 ('iand', (invert
[left
], a
, b
), (invert
[right
], c
, d
))))
575 optimizations
.append((('inot', ('iand(is_used_once)', (left
, a
, b
), (right
, c
, d
))),
576 ('ior', (invert
[left
], a
, b
), (invert
[right
], c
, d
))))
578 def fexp2i(exp
, bits
):
579 # We assume that exp is already in the right range.
581 return ('ishl', ('iadd', exp
, 127), 23)
583 return ('pack_64_2x32_split', 0, ('ishl', ('iadd', exp
, 1023), 20))
587 def ldexp(f
, exp
, bits
):
588 # First, we clamp exp to a reasonable range. The maximum possible range
589 # for a normal exponent is [-126, 127] and, throwing in denormals, you get
590 # a maximum range of [-149, 127]. This means that we can potentially have
591 # a swing of +-276. If you start with FLT_MAX, you actually have to do
592 # ldexp(FLT_MAX, -278) to get it to flush all the way to zero. The GLSL
593 # spec, on the other hand, only requires that we handle an exponent value
594 # in the range [-126, 128]. This implementation is *mostly* correct; it
595 # handles a range on exp of [-252, 254] which allows you to create any
596 # value (including denorms if the hardware supports it) and to adjust the
597 # exponent of any normal value to anything you want.
599 exp
= ('imin', ('imax', exp
, -252), 254)
601 exp
= ('imin', ('imax', exp
, -2044), 2046)
605 # Now we compute two powers of 2, one for exp/2 and one for exp-exp/2.
606 # (We use ishr which isn't the same for -1, but the -1 case still works
607 # since we use exp-exp/2 as the second exponent.) While the spec
608 # technically defines ldexp as f * 2.0^exp, simply multiplying once doesn't
609 # work with denormals and doesn't allow for the full swing in exponents
610 # that you can get with normalized values. Instead, we create two powers
611 # of two and multiply by them each in turn. That way the effective range
612 # of our exponent is doubled.
613 pow2_1
= fexp2i(('ishr', exp
, 1), bits
)
614 pow2_2
= fexp2i(('isub', exp
, ('ishr', exp
, 1)), bits
)
615 return ('fmul', ('fmul', f
, pow2_1
), pow2_2
)
618 (('ldexp@32', 'x', 'exp'), ldexp('x', 'exp', 32), 'options->lower_ldexp'),
619 (('ldexp@64', 'x', 'exp'), ldexp('x', 'exp', 64), 'options->lower_ldexp'),
622 # Unreal Engine 4 demo applications open-codes bitfieldReverse()
623 def bitfield_reverse(u
):
624 step1
= ('ior', ('ishl', u
, 16), ('ushr', u
, 16))
625 step2
= ('ior', ('ishl', ('iand', step1
, 0x00ff00ff), 8), ('ushr', ('iand', step1
, 0xff00ff00), 8))
626 step3
= ('ior', ('ishl', ('iand', step2
, 0x0f0f0f0f), 4), ('ushr', ('iand', step2
, 0xf0f0f0f0), 4))
627 step4
= ('ior', ('ishl', ('iand', step3
, 0x33333333), 2), ('ushr', ('iand', step3
, 0xcccccccc), 2))
628 step5
= ('ior', ('ishl', ('iand', step4
, 0x55555555), 1), ('ushr', ('iand', step4
, 0xaaaaaaaa), 1))
632 optimizations
+= [(bitfield_reverse('x@32'), ('bitfield_reverse', 'x'))]
634 # For any float comparison operation, "cmp", if you have "a == a && a cmp b"
635 # then the "a == a" is redundant because it's equivalent to "a is not NaN"
636 # and, if a is a NaN then the second comparison will fail anyway.
637 for op
in ['flt', 'fge', 'feq']:
639 (('iand', ('feq', a
, a
), (op
, a
, b
)), (op
, a
, b
)),
640 (('iand', ('feq', a
, a
), (op
, b
, a
)), (op
, b
, a
)),
643 # Add optimizations to handle the case where the result of a ternary is
644 # compared to a constant. This way we can take things like
650 # a ? (0 > 0) : (1 > 0)
652 # which constant folding will eat for lunch. The resulting ternary will
653 # further get cleaned up by the boolean reductions above and we will be
654 # left with just the original variable "a".
655 for op
in ['flt', 'fge', 'feq', 'fne',
656 'ilt', 'ige', 'ieq', 'ine', 'ult', 'uge']:
658 ((op
, ('bcsel', 'a', '#b', '#c'), '#d'),
659 ('bcsel', 'a', (op
, 'b', 'd'), (op
, 'c', 'd'))),
660 ((op
, '#d', ('bcsel', a
, '#b', '#c')),
661 ('bcsel', 'a', (op
, 'd', 'b'), (op
, 'd', 'c'))),
665 # For example, this converts things like
667 # 1 + mix(0, a - 1, condition)
671 # mix(1, (a-1)+1, condition)
673 # Other optimizations will rearrange the constants.
674 for op
in ['fadd', 'fmul', 'iadd', 'imul']:
676 ((op
, ('bcsel(is_used_once)', a
, '#b', c
), '#d'), ('bcsel', a
, (op
, b
, d
), (op
, c
, d
)))
679 # This section contains "late" optimizations that should be run before
680 # creating ffmas and calling regular optimizations for the final time.
681 # Optimizations should go here if they help code generation and conflict
682 # with the regular optimizations.
683 before_ffma_optimizations
= [
684 # Propagate constants down multiplication chains
685 (('~fmul(is_used_once)', ('fmul(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('fmul', ('fmul', a
, c
), b
)),
686 (('imul(is_used_once)', ('imul(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('imul', ('imul', a
, c
), b
)),
687 (('~fadd(is_used_once)', ('fadd(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('fadd', ('fadd', a
, c
), b
)),
688 (('iadd(is_used_once)', ('iadd(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('iadd', ('iadd', a
, c
), b
)),
690 (('~fadd', ('fmul', a
, b
), ('fmul', a
, c
)), ('fmul', a
, ('fadd', b
, c
))),
691 (('iadd', ('imul', a
, b
), ('imul', a
, c
)), ('imul', a
, ('iadd', b
, c
))),
692 (('~fadd', ('fneg', a
), a
), 0.0),
693 (('iadd', ('ineg', a
), a
), 0),
694 (('iadd', ('ineg', a
), ('iadd', a
, b
)), b
),
695 (('iadd', a
, ('iadd', ('ineg', a
), b
)), b
),
696 (('~fadd', ('fneg', a
), ('fadd', a
, b
)), b
),
697 (('~fadd', a
, ('fadd', ('fneg', a
), b
)), b
),
700 # This section contains "late" optimizations that should be run after the
701 # regular optimizations have finished. Optimizations should go here if
702 # they help code generation but do not necessarily produce code that is
703 # more easily optimizable.
704 late_optimizations
= [
705 # Most of these optimizations aren't quite safe when you get infinity or
706 # Nan involved but the first one should be fine.
707 (('flt', ('fadd', a
, b
), 0.0), ('flt', a
, ('fneg', b
))),
708 (('flt', ('fneg', ('fadd', a
, b
)), 0.0), ('flt', ('fneg', a
), b
)),
709 (('~fge', ('fadd', a
, b
), 0.0), ('fge', a
, ('fneg', b
))),
710 (('~fge', ('fneg', ('fadd', a
, b
)), 0.0), ('fge', ('fneg', a
), b
)),
711 (('~feq', ('fadd', a
, b
), 0.0), ('feq', a
, ('fneg', b
))),
712 (('~fne', ('fadd', a
, b
), 0.0), ('fne', a
, ('fneg', b
))),
714 (('~fge', ('fmin(is_used_once)', ('fadd(is_used_once)', a
, b
), ('fadd', c
, d
)), 0.0), ('iand', ('fge', a
, ('fneg', b
)), ('fge', c
, ('fneg', d
)))),
716 (('fdot2', a
, b
), ('fdot_replicated2', a
, b
), 'options->fdot_replicates'),
717 (('fdot3', a
, b
), ('fdot_replicated3', a
, b
), 'options->fdot_replicates'),
718 (('fdot4', a
, b
), ('fdot_replicated4', a
, b
), 'options->fdot_replicates'),
719 (('fdph', a
, b
), ('fdph_replicated', a
, b
), 'options->fdot_replicates'),
721 (('b2f(is_used_more_than_once)', ('inot', a
)), ('bcsel', a
, 0.0, 1.0)),
722 (('fneg(is_used_more_than_once)', ('b2f', ('inot', a
))), ('bcsel', a
, -0.0, -1.0)),
724 # we do these late so that we don't get in the way of creating ffmas
725 (('fmin', ('fadd(is_used_once)', '#c', a
), ('fadd(is_used_once)', '#c', b
)), ('fadd', c
, ('fmin', a
, b
))),
726 (('fmax', ('fadd(is_used_once)', '#c', a
), ('fadd(is_used_once)', '#c', b
)), ('fadd', c
, ('fmax', a
, b
))),
728 # Lowered for backends without a dedicated b2f instruction
729 (('b2f@32', a
), ('iand', a
, 1.0), 'options->lower_b2f'),
732 print nir_algebraic
.AlgebraicPass("nir_opt_algebraic", optimizations
).render()
733 print nir_algebraic
.AlgebraicPass("nir_opt_algebraic_before_ffma",
734 before_ffma_optimizations
).render()
735 print nir_algebraic
.AlgebraicPass("nir_opt_algebraic_late",
736 late_optimizations
).render()
projects / mesa.git / blob