1 /*
2 * Copyright © 2018 Red Hat Inc.
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * IN THE SOFTWARE.
22 */
23
24 #ifndef NIR_BUILTIN_BUILDER_H
25 #define NIR_BUILTIN_BUILDER_H
26
27 #include "util/u_math.h"
28 #include "nir_builder.h"
29
30 #ifdef __cplusplus
31 extern "C" {
32 #endif
33
34 /*
35 * Functions are sorted alphabetically with removed type and "fast" prefix.
36 * Definitions for functions in the C file come first.
37 */
38
39 nir_ssa_def* nir_cross3(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y);
40 nir_ssa_def* nir_cross4(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y);
41 nir_ssa_def* nir_fast_length(nir_builder *b, nir_ssa_def *vec);
42 nir_ssa_def* nir_nextafter(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y);
43 nir_ssa_def* nir_normalize(nir_builder *b, nir_ssa_def *vec);
44 nir_ssa_def* nir_smoothstep(nir_builder *b, nir_ssa_def *edge0,
45 nir_ssa_def *edge1, nir_ssa_def *x);
46 nir_ssa_def* nir_upsample(nir_builder *b, nir_ssa_def *hi, nir_ssa_def *lo);
47 nir_ssa_def* nir_atan(nir_builder *b, nir_ssa_def *y_over_x);
48 nir_ssa_def* nir_atan2(nir_builder *b, nir_ssa_def *y, nir_ssa_def *x);
49
50 nir_ssa_def *
51 nir_get_texture_lod(nir_builder *b, nir_tex_instr *tex);
52
53 nir_ssa_def *
54 nir_get_texture_size(nir_builder *b, nir_tex_instr *tex);
55
56 static inline nir_ssa_def *
nir_nan_check2(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y,nir_ssa_def * res)57 nir_nan_check2(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y, nir_ssa_def *res)
58 {
59 return nir_bcsel(b, nir_fneu(b, x, x), x, nir_bcsel(b, nir_fneu(b, y, y), y, res));
60 }
61
62 static inline nir_ssa_def *
nir_fmax_abs_vec_comp(nir_builder * b,nir_ssa_def * vec)63 nir_fmax_abs_vec_comp(nir_builder *b, nir_ssa_def *vec)
64 {
65 nir_ssa_def *res = nir_channel(b, vec, 0);
66 for (unsigned i = 1; i < vec->num_components; ++i)
67 res = nir_fmax(b, res, nir_fabs(b, nir_channel(b, vec, i)));
68 return res;
69 }
70
71 static inline nir_ssa_def *
nir_iabs_diff(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y)72 nir_iabs_diff(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
73 {
74 nir_ssa_def *cond = nir_ige(b, x, y);
75 nir_ssa_def *res0 = nir_isub(b, x, y);
76 nir_ssa_def *res1 = nir_isub(b, y, x);
77 return nir_bcsel(b, cond, res0, res1);
78 }
79
80 static inline nir_ssa_def *
nir_uabs_diff(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y)81 nir_uabs_diff(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
82 {
83 nir_ssa_def *cond = nir_uge(b, x, y);
84 nir_ssa_def *res0 = nir_isub(b, x, y);
85 nir_ssa_def *res1 = nir_isub(b, y, x);
86 return nir_bcsel(b, cond, res0, res1);
87 }
88
89 static inline nir_ssa_def *
nir_fexp(nir_builder * b,nir_ssa_def * x)90 nir_fexp(nir_builder *b, nir_ssa_def *x)
91 {
92 return nir_fexp2(b, nir_fmul_imm(b, x, M_LOG2E));
93 }
94
95 static inline nir_ssa_def *
nir_flog(nir_builder * b,nir_ssa_def * x)96 nir_flog(nir_builder *b, nir_ssa_def *x)
97 {
98 return nir_fmul_imm(b, nir_flog2(b, x), 1.0 / M_LOG2E);
99 }
100
101 static inline nir_ssa_def *
nir_imad24(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y,nir_ssa_def * z)102 nir_imad24(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y, nir_ssa_def *z)
103 {
104 nir_ssa_def *temp = nir_imul24(b, x, y);
105 return nir_iadd(b, temp, z);
106 }
107
108 static inline nir_ssa_def *
nir_imad_hi(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y,nir_ssa_def * z)109 nir_imad_hi(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y, nir_ssa_def *z)
110 {
111 nir_ssa_def *temp = nir_imul_high(b, x, y);
112 return nir_iadd(b, temp, z);
113 }
114
115 static inline nir_ssa_def *
nir_umad_hi(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y,nir_ssa_def * z)116 nir_umad_hi(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y, nir_ssa_def *z)
117 {
118 nir_ssa_def *temp = nir_umul_high(b, x, y);
119 return nir_iadd(b, temp, z);
120 }
121
122 static inline nir_ssa_def *
nir_bitselect(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y,nir_ssa_def * s)123 nir_bitselect(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y, nir_ssa_def *s)
124 {
125 return nir_ior(b, nir_iand(b, nir_inot(b, s), x), nir_iand(b, s, y));
126 }
127
128 static inline nir_ssa_def *
nir_fclamp(nir_builder * b,nir_ssa_def * x,nir_ssa_def * min_val,nir_ssa_def * max_val)129 nir_fclamp(nir_builder *b,
130 nir_ssa_def *x, nir_ssa_def *min_val, nir_ssa_def *max_val)
131 {
132 return nir_fmin(b, nir_fmax(b, x, min_val), max_val);
133 }
134
135 static inline nir_ssa_def *
nir_iclamp(nir_builder * b,nir_ssa_def * x,nir_ssa_def * min_val,nir_ssa_def * max_val)136 nir_iclamp(nir_builder *b,
137 nir_ssa_def *x, nir_ssa_def *min_val, nir_ssa_def *max_val)
138 {
139 return nir_imin(b, nir_imax(b, x, min_val), max_val);
140 }
141
142 static inline nir_ssa_def *
nir_uclamp(nir_builder * b,nir_ssa_def * x,nir_ssa_def * min_val,nir_ssa_def * max_val)143 nir_uclamp(nir_builder *b,
144 nir_ssa_def *x, nir_ssa_def *min_val, nir_ssa_def *max_val)
145 {
146 return nir_umin(b, nir_umax(b, x, min_val), max_val);
147 }
148
149 static inline nir_ssa_def *
nir_copysign(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y)150 nir_copysign(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
151 {
152 uint64_t masks = 1ull << (x->bit_size - 1);
153 uint64_t maskv = ~masks;
154
155 nir_ssa_def *s = nir_imm_intN_t(b, masks, x->bit_size);
156 nir_ssa_def *v = nir_imm_intN_t(b, maskv, x->bit_size);
157
158 return nir_ior(b, nir_iand(b, x, v), nir_iand(b, y, s));
159 }
160
161 static inline nir_ssa_def *
nir_degrees(nir_builder * b,nir_ssa_def * val)162 nir_degrees(nir_builder *b, nir_ssa_def *val)
163 {
164 return nir_fmul_imm(b, val, 180.0 / M_PI);
165 }
166
167 static inline nir_ssa_def *
nir_fdim(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y)168 nir_fdim(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
169 {
170 nir_ssa_def *cond = nir_flt(b, y, x);
171 nir_ssa_def *res = nir_fsub(b, x, y);
172 nir_ssa_def *zero = nir_imm_floatN_t(b, 0.0, x->bit_size);
173
174 // return NaN if either x or y are NaN, else x-y if x>y, else +0.0
175 return nir_nan_check2(b, x, y, nir_bcsel(b, cond, res, zero));
176 }
177
178 static inline nir_ssa_def *
nir_fast_distance(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y)179 nir_fast_distance(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
180 {
181 return nir_fast_length(b, nir_fsub(b, x, y));
182 }
183
184 static inline nir_ssa_def*
nir_fast_normalize(nir_builder * b,nir_ssa_def * vec)185 nir_fast_normalize(nir_builder *b, nir_ssa_def *vec)
186 {
187 return nir_fdiv(b, vec, nir_fast_length(b, vec));
188 }
189
190 static inline nir_ssa_def*
nir_fmad(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y,nir_ssa_def * z)191 nir_fmad(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y, nir_ssa_def *z)
192 {
193 return nir_fadd(b, nir_fmul(b, x, y), z);
194 }
195
196 static inline nir_ssa_def*
nir_maxmag(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y)197 nir_maxmag(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
198 {
199 nir_ssa_def *xabs = nir_fabs(b, x);
200 nir_ssa_def *yabs = nir_fabs(b, y);
201
202 nir_ssa_def *condy = nir_flt(b, xabs, yabs);
203 nir_ssa_def *condx = nir_flt(b, yabs, xabs);
204
205 return nir_bcsel(b, condy, y, nir_bcsel(b, condx, x, nir_fmax(b, x, y)));
206 }
207
208 static inline nir_ssa_def*
nir_minmag(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y)209 nir_minmag(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
210 {
211 nir_ssa_def *xabs = nir_fabs(b, x);
212 nir_ssa_def *yabs = nir_fabs(b, y);
213
214 nir_ssa_def *condx = nir_flt(b, xabs, yabs);
215 nir_ssa_def *condy = nir_flt(b, yabs, xabs);
216
217 return nir_bcsel(b, condy, y, nir_bcsel(b, condx, x, nir_fmin(b, x, y)));
218 }
219
220 static inline nir_ssa_def*
nir_nan(nir_builder * b,nir_ssa_def * x)221 nir_nan(nir_builder *b, nir_ssa_def *x)
222 {
223 nir_ssa_def *nan = nir_imm_floatN_t(b, NAN, x->bit_size);
224 if (x->num_components == 1)
225 return nan;
226
227 nir_ssa_def *nans[NIR_MAX_VEC_COMPONENTS];
228 for (unsigned i = 0; i < x->num_components; ++i)
229 nans[i] = nan;
230
231 return nir_vec(b, nans, x->num_components);
232 }
233
234 static inline nir_ssa_def *
nir_radians(nir_builder * b,nir_ssa_def * val)235 nir_radians(nir_builder *b, nir_ssa_def *val)
236 {
237 return nir_fmul_imm(b, val, M_PI / 180.0);
238 }
239
240 static inline nir_ssa_def *
nir_select(nir_builder * b,nir_ssa_def * x,nir_ssa_def * y,nir_ssa_def * s)241 nir_select(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y, nir_ssa_def *s)
242 {
243 if (s->num_components != 1) {
244 uint64_t mask = 1ull << (s->bit_size - 1);
245 s = nir_iand(b, s, nir_imm_intN_t(b, mask, s->bit_size));
246 }
247 return nir_bcsel(b, nir_ieq_imm(b, s, 0), x, y);
248 }
249
250 static inline nir_ssa_def *
nir_ftan(nir_builder * b,nir_ssa_def * x)251 nir_ftan(nir_builder *b, nir_ssa_def *x)
252 {
253 return nir_fdiv(b, nir_fsin(b, x), nir_fcos(b, x));
254 }
255
256 static inline nir_ssa_def *
nir_clz_u(nir_builder * b,nir_ssa_def * a)257 nir_clz_u(nir_builder *b, nir_ssa_def *a)
258 {
259 nir_ssa_def *val;
260 val = nir_isub(b, nir_imm_intN_t(b, a->bit_size - 1, 32), nir_ufind_msb(b, a));
261 return nir_u2u(b, val, a->bit_size);
262 }
263
264 static inline nir_ssa_def *
nir_ctz_u(nir_builder * b,nir_ssa_def * a)265 nir_ctz_u(nir_builder *b, nir_ssa_def *a)
266 {
267 nir_ssa_def *cond = nir_ieq(b, a, nir_imm_intN_t(b, 0, a->bit_size));
268
269 return nir_bcsel(b, cond,
270 nir_imm_intN_t(b, a->bit_size, a->bit_size),
271 nir_u2u(b, nir_find_lsb(b, a), a->bit_size));
272 }
273
274 #ifdef __cplusplus
275 }
276 #endif
277
278 #endif /* NIR_BUILTIN_BUILDER_H */
279