1 /*
2 * Copyright © 2019 Intel Corporation
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
21 * DEALINGS IN THE SOFTWARE.
22 */
23 #include <gtest/gtest.h>
24 #include "nir.h"
25 #include "nir_builder.h"
26
27 class comparison_pre_test : public ::testing::Test {
28 protected:
comparison_pre_test()29 comparison_pre_test()
30 {
31 glsl_type_singleton_init_or_ref();
32
33 static const nir_shader_compiler_options options = { };
34 nir_builder_init_simple_shader(&bld, NULL, MESA_SHADER_VERTEX, &options);
35
36 v1 = nir_imm_vec4(&bld, -2.0, -1.0, 1.0, 2.0);
37 v2 = nir_imm_vec4(&bld, 2.0, 1.0, -1.0, -2.0);
38 v3 = nir_imm_vec4(&bld, 3.0, 4.0, 5.0, 6.0);
39 }
40
~comparison_pre_test()41 ~comparison_pre_test()
42 {
43 ralloc_free(bld.shader);
44 glsl_type_singleton_decref();
45 }
46
47 struct nir_builder bld;
48
49 nir_ssa_def *v1;
50 nir_ssa_def *v2;
51 nir_ssa_def *v3;
52
53 const uint8_t xxxx[4] = { 0, 0, 0, 0 };
54 const uint8_t wwww[4] = { 3, 3, 3, 3 };
55 };
56
TEST_F(comparison_pre_test,a_lt_b_vs_neg_a_plus_b)57 TEST_F(comparison_pre_test, a_lt_b_vs_neg_a_plus_b)
58 {
59 /* Before:
60 *
61 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
62 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
63 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
64 * vec1 32 ssa_3 = load_const ( 1.0)
65 * vec4 32 ssa_4 = fadd ssa_0, ssa_2
66 * vec1 32 ssa_5 = mov ssa_4.x
67 * vec1 1 ssa_6 = flt ssa_5, ssa_3
68 *
69 * if ssa_6 {
70 * vec1 32 ssa_7 = fneg ssa_5
71 * vec1 32 ssa_8 = fadd ssa_7, ssa_3
72 * } else {
73 * }
74 *
75 * After:
76 *
77 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
78 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
79 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
80 * vec1 32 ssa_3 = load_const ( 1.0)
81 * vec4 32 ssa_4 = fadd ssa_0, ssa_2
82 * vec1 32 ssa_5 = mov ssa_4.x
83 * vec1 32 ssa_9 = fneg ssa_5
84 * vec1 32 ssa_10 = fadd ssa_3, ssa_9
85 * vec1 32 ssa_11 = load_const (0.0)
86 * vec1 1 ssa_12 = flt ssa_11, ssa_10
87 * vec1 32 ssa_13 = mov ssa_10
88 * vec1 1 ssa_14 = mov ssa_12
89 *
90 * if ssa_14 {
91 * vec1 32 ssa_7 = fneg ssa_5
92 * } else {
93 * }
94 */
95 nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
96 nir_ssa_def *a = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
97
98 nir_ssa_def *flt = nir_flt(&bld, a, one);
99
100 nir_if *nif = nir_push_if(&bld, flt);
101
102 nir_fadd(&bld, nir_fneg(&bld, a), one);
103
104 nir_pop_if(&bld, nif);
105
106 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
107 }
108
TEST_F(comparison_pre_test,a_lt_b_vs_a_minus_b)109 TEST_F(comparison_pre_test, a_lt_b_vs_a_minus_b)
110 {
111 /* Before:
112 *
113 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
114 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
115 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
116 * vec1 32 ssa_3 = load_const ( 1.0)
117 * vec4 32 ssa_4 = fadd ssa_0, ssa_2
118 * vec1 32 ssa_5 = mov ssa_4.x
119 * vec1 1 ssa_6 = flt ssa_3, ssa_5
120 *
121 * if ssa_6 {
122 * vec1 32 ssa_7 = fneg ssa_5
123 * vec1 32 ssa_8 = fadd ssa_3, ssa_7
124 * } else {
125 * }
126 *
127 * After:
128 *
129 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
130 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
131 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
132 * vec1 32 ssa_3 = load_const ( 1.0)
133 * vec4 32 ssa_4 = fadd ssa_0, ssa_2
134 * vec1 32 ssa_5 = mov ssa_4.x
135 * vec1 32 ssa_9 = fneg ssa_5
136 * vec1 32 ssa_10 = fadd ssa_3, ssa_9
137 * vec1 32 ssa_11 = load_const (0.0)
138 * vec1 1 ssa_12 = flt ssa_10, ssa_11
139 * vec1 32 ssa_13 = mov ssa_10
140 * vec1 1 ssa_14 = mov ssa_12
141 *
142 * if ssa_14 {
143 * vec1 32 ssa_7 = fneg ssa_5
144 * } else {
145 * }
146 */
147 nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
148 nir_ssa_def *b = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
149
150 nir_ssa_def *flt = nir_flt(&bld, one, b);
151
152 nir_if *nif = nir_push_if(&bld, flt);
153
154 nir_fadd(&bld, one, nir_fneg(&bld, b));
155
156 nir_pop_if(&bld, nif);
157
158 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
159 }
160
TEST_F(comparison_pre_test,neg_a_lt_b_vs_a_plus_b)161 TEST_F(comparison_pre_test, neg_a_lt_b_vs_a_plus_b)
162 {
163 /* Before:
164 *
165 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
166 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
167 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
168 * vec1 32 ssa_3 = load_const ( 1.0)
169 * vec4 32 ssa_4 = fadd ssa_0, ssa_2
170 * vec1 32 ssa_5 = mov ssa_4.x
171 * vec1 32 ssa_6 = fneg ssa_5
172 * vec1 1 ssa_7 = flt ssa_6, ssa_3
173 *
174 * if ssa_7 {
175 * vec1 32 ssa_8 = fadd ssa_5, ssa_3
176 * } else {
177 * }
178 *
179 * After:
180 *
181 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
182 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
183 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
184 * vec1 32 ssa_3 = load_const ( 1.0)
185 * vec4 32 ssa_4 = fadd ssa_0, ssa_2
186 * vec1 32 ssa_5 = mov ssa_4.x
187 * vec1 32 ssa_9 = fneg ssa_5
188 * vec1 32 ssa_9 = fneg ssa_6
189 * vec1 32 ssa_10 = fadd ssa_3, ssa_9
190 * vec1 32 ssa_11 = load_const ( 0.0)
191 * vec1 1 ssa_12 = flt ssa_11, ssa_10
192 * vec1 32 ssa_13 = mov ssa_10
193 * vec1 1 ssa_14 = mov ssa_12
194 *
195 * if ssa_14 {
196 * } else {
197 * }
198 */
199
200 nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
201 nir_ssa_def *a = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
202
203 nir_ssa_def *flt = nir_flt(&bld, nir_fneg(&bld, a), one);
204
205 nir_if *nif = nir_push_if(&bld, flt);
206
207 nir_fadd(&bld, a, one);
208
209 nir_pop_if(&bld, nif);
210
211 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
212 }
213
TEST_F(comparison_pre_test,a_lt_neg_b_vs_a_plus_b)214 TEST_F(comparison_pre_test, a_lt_neg_b_vs_a_plus_b)
215 {
216 /* Before:
217 *
218 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
219 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
220 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
221 * vec1 32 ssa_3 = load_const ( 1.0)
222 * vec4 32 ssa_4 = fadd ssa_0, ssa_2
223 * vec1 32 ssa_5 = mov ssa_4.x
224 * vec1 32 ssa_6 = fneg ssa_5
225 * vec1 1 ssa_7 = flt ssa_3, ssa_6
226 *
227 * if ssa_7 {
228 * vec1 32 ssa_8 = fadd ssa_3, ssa_5
229 * } else {
230 * }
231 *
232 * After:
233 *
234 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
235 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
236 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
237 * vec1 32 ssa_3 = load_const ( 1.0)
238 * vec4 32 ssa_4 = fadd ssa_0, ssa_2
239 * vec1 32 ssa_5 = mov ssa_4.x
240 * vec1 32 ssa_9 = fneg ssa_5
241 * vec1 32 ssa_9 = fneg ssa_6
242 * vec1 32 ssa_10 = fadd ssa_3, ssa_9
243 * vec1 32 ssa_11 = load_const ( 0.0)
244 * vec1 1 ssa_12 = flt ssa_10, ssa_11
245 * vec1 32 ssa_13 = mov ssa_10
246 * vec1 1 ssa_14 = mov ssa_12
247 *
248 * if ssa_14 {
249 * } else {
250 * }
251 */
252 nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
253 nir_ssa_def *b = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
254
255 nir_ssa_def *flt = nir_flt(&bld, one, nir_fneg(&bld, b));
256
257 nir_if *nif = nir_push_if(&bld, flt);
258
259 nir_fadd(&bld, one, b);
260
261 nir_pop_if(&bld, nif);
262
263 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
264 }
265
TEST_F(comparison_pre_test,imm_lt_b_vs_neg_imm_plus_b)266 TEST_F(comparison_pre_test, imm_lt_b_vs_neg_imm_plus_b)
267 {
268 /* Before:
269 *
270 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
271 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
272 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
273 * vec1 32 ssa_3 = load_const ( 1.0)
274 * vec1 32 ssa_4 = load_const (-1.0)
275 * vec4 32 ssa_5 = fadd ssa_0, ssa_2
276 * vec1 32 ssa_6 = mov ssa_5.x
277 * vec1 1 ssa_7 = flt ssa_3, ssa_6
278 *
279 * if ssa_7 {
280 * vec1 32 ssa_8 = fadd ssa_4, ssa_6
281 * } else {
282 * }
283 *
284 * After:
285 *
286 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
287 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
288 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
289 * vec1 32 ssa_3 = load_const ( 1.0)
290 * vec1 32 ssa_4 = load_const (-1.0)
291 * vec4 32 ssa_5 = fadd ssa_0, ssa_2
292 * vec1 32 ssa_6 = mov ssa_5.x
293 * vec1 32 ssa_9 = fneg ssa_3
294 * vec1 32 ssa_10 = fadd ssa_6, ssa_9
295 * vec1 32 ssa_11 = load_const ( 0.0)
296 * vec1 1 ssa_12 = flt ssa_11, ssa_10
297 * vec1 32 ssa_13 = mov ssa_10
298 * vec1 1 ssa_14 = mov ssa_12
299 *
300 * if ssa_14 {
301 * } else {
302 * }
303 */
304 nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
305 nir_ssa_def *neg_one = nir_imm_float(&bld, -1.0f);
306 nir_ssa_def *a = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
307
308 nir_ssa_def *flt = nir_flt(&bld, one, a);
309
310 nir_if *nif = nir_push_if(&bld, flt);
311
312 nir_fadd(&bld, neg_one, a);
313
314 nir_pop_if(&bld, nif);
315
316 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
317 }
318
TEST_F(comparison_pre_test,a_lt_imm_vs_a_minus_imm)319 TEST_F(comparison_pre_test, a_lt_imm_vs_a_minus_imm)
320 {
321 /* Before:
322 *
323 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
324 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
325 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
326 * vec1 32 ssa_3 = load_const ( 1.0)
327 * vec1 32 ssa_4 = load_const (-1.0)
328 * vec4 32 ssa_5 = fadd ssa_0, ssa_2
329 * vec1 32 ssa_6 = mov ssa_5.x
330 * vec1 1 ssa_7 = flt ssa_6, ssa_3
331 *
332 * if ssa_6 {
333 * vec1 32 ssa_8 = fadd ssa_6, ssa_4
334 * } else {
335 * }
336 *
337 * After:
338 *
339 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
340 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
341 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
342 * vec1 32 ssa_3 = load_const ( 1.0)
343 * vec1 32 ssa_4 = load_const (-1.0)
344 * vec4 32 ssa_5 = fadd ssa_0, ssa_2
345 * vec1 32 ssa_6 = mov ssa_5.x
346 * vec1 32 ssa_9 = fneg ssa_3
347 * vec1 32 ssa_10 = fadd ssa_6, ssa_9
348 * vec1 32 ssa_11 = load_const ( 0.0)
349 * vec1 1 ssa_12 = flt ssa_10, ssa_11
350 * vec1 32 ssa_13 = mov ssa_10
351 * vec1 1 ssa_14 = mov ssa_12
352 *
353 * if ssa_14 {
354 * } else {
355 * }
356 */
357 nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
358 nir_ssa_def *neg_one = nir_imm_float(&bld, -1.0f);
359 nir_ssa_def *a = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
360
361 nir_ssa_def *flt = nir_flt(&bld, a, one);
362
363 nir_if *nif = nir_push_if(&bld, flt);
364
365 nir_fadd(&bld, a, neg_one);
366
367 nir_pop_if(&bld, nif);
368
369 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
370 }
371
TEST_F(comparison_pre_test,neg_imm_lt_a_vs_a_plus_imm)372 TEST_F(comparison_pre_test, neg_imm_lt_a_vs_a_plus_imm)
373 {
374 /* Before:
375 *
376 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
377 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
378 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
379 * vec1 32 ssa_3 = load_const ( 1.0)
380 * vec1 32 ssa_4 = load_const (-1.0)
381 * vec4 32 ssa_5 = fadd ssa_0, ssa_2
382 * vec1 32 ssa_6 = mov ssa_5.x
383 * vec1 1 ssa_7 = flt ssa_4, ssa_6
384 *
385 * if ssa_7 {
386 * vec1 32 ssa_8 = fadd ssa_6, ssa_3
387 * } else {
388 * }
389 *
390 * After:
391 *
392 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
393 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
394 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
395 * vec1 32 ssa_3 = load_const ( 1.0)
396 * vec1 32 ssa_4 = load_const (-1.0)
397 * vec4 32 ssa_5 = fadd ssa_0, ssa_2
398 * vec1 32 ssa_6 = mov ssa_5.x
399 * vec1 32 ssa_9 = fneg ssa_4
400 * vec1 32 ssa_10 = fadd ssa_6, ssa_9
401 * vec1 32 ssa_11 = load_const ( 0.0)
402 * vec1 1 ssa_12 = flt ssa_11, ssa_10
403 * vec1 32 ssa_13 = mov ssa_10
404 * vec1 1 ssa_14 = mov ssa_12
405 *
406 * if ssa_14 {
407 * } else {
408 * }
409 */
410
411 nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
412 nir_ssa_def *neg_one = nir_imm_float(&bld, -1.0f);
413 nir_ssa_def *a = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
414
415 nir_ssa_def *flt = nir_flt(&bld, neg_one, a);
416
417 nir_if *nif = nir_push_if(&bld, flt);
418
419 nir_fadd(&bld, a, one);
420
421 nir_pop_if(&bld, nif);
422
423 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
424 }
425
TEST_F(comparison_pre_test,a_lt_neg_imm_vs_a_plus_imm)426 TEST_F(comparison_pre_test, a_lt_neg_imm_vs_a_plus_imm)
427 {
428 /* Before:
429 *
430 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
431 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
432 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
433 * vec1 32 ssa_3 = load_const ( 1.0)
434 * vec1 32 ssa_4 = load_const (-1.0)
435 * vec4 32 ssa_5 = fadd ssa_0, ssa_2
436 * vec1 32 ssa_6 = mov ssa_5.x
437 * vec1 1 ssa_7 = flt ssa_6, ssa_4
438 *
439 * if ssa_7 {
440 * vec1 32 ssa_8 = fadd ssa_6, ssa_3
441 * } else {
442 * }
443 *
444 * After:
445 *
446 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
447 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
448 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
449 * vec1 32 ssa_3 = load_const ( 1.0)
450 * vec1 32 ssa_4 = load_const (-1.0)
451 * vec4 32 ssa_5 = fadd ssa_0, ssa_2
452 * vec1 32 ssa_6 = mov ssa_5.x
453 * vec1 32 ssa_9 = fneg ssa_4
454 * vec1 32 ssa_10 = fadd ssa_6, ssa_9
455 * vec1 32 ssa_11 = load_const ( 0.0)
456 * vec1 1 ssa_12 = flt ssa_10, ssa_11
457 * vec1 32 ssa_13 = mov ssa_10
458 * vec1 1 ssa_14 = mov ssa_12
459 *
460 * if ssa_14 {
461 * } else {
462 * }
463 */
464 nir_ssa_def *one = nir_imm_float(&bld, 1.0f);
465 nir_ssa_def *neg_one = nir_imm_float(&bld, -1.0f);
466 nir_ssa_def *a = nir_channel(&bld, nir_fadd(&bld, v1, v3), 0);
467
468 nir_ssa_def *flt = nir_flt(&bld, a, neg_one);
469
470 nir_if *nif = nir_push_if(&bld, flt);
471
472 nir_fadd(&bld, a, one);
473
474 nir_pop_if(&bld, nif);
475
476 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
477 }
478
TEST_F(comparison_pre_test,swizzle_of_same_immediate_vector)479 TEST_F(comparison_pre_test, swizzle_of_same_immediate_vector)
480 {
481 /* Before:
482 *
483 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
484 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
485 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
486 * vec4 32 ssa_3 = fadd ssa_0, ssa_2
487 * vec1 1 ssa_4 = flt ssa_0.x, ssa_3.x
488 *
489 * if ssa_4 {
490 * vec1 32 ssa_5 = fadd ssa_0.w, ssa_3.x
491 * } else {
492 * }
493 */
494 nir_ssa_def *a = nir_fadd(&bld, v1, v3);
495
496 nir_alu_instr *flt = nir_alu_instr_create(bld.shader, nir_op_flt);
497
498 flt->src[0].src = nir_src_for_ssa(v1);
499 flt->src[1].src = nir_src_for_ssa(a);
500
501 memcpy(&flt->src[0].swizzle, xxxx, sizeof(xxxx));
502 memcpy(&flt->src[1].swizzle, xxxx, sizeof(xxxx));
503
504 nir_builder_alu_instr_finish_and_insert(&bld, flt);
505
506 flt->dest.dest.ssa.num_components = 1;
507 flt->dest.write_mask = 1;
508
509 nir_if *nif = nir_push_if(&bld, &flt->dest.dest.ssa);
510
511 nir_alu_instr *fadd = nir_alu_instr_create(bld.shader, nir_op_fadd);
512
513 fadd->src[0].src = nir_src_for_ssa(v1);
514 fadd->src[1].src = nir_src_for_ssa(a);
515
516 memcpy(&fadd->src[0].swizzle, wwww, sizeof(wwww));
517 memcpy(&fadd->src[1].swizzle, xxxx, sizeof(xxxx));
518
519 nir_builder_alu_instr_finish_and_insert(&bld, fadd);
520
521 fadd->dest.dest.ssa.num_components = 1;
522 fadd->dest.write_mask = 1;
523
524 nir_pop_if(&bld, nif);
525
526 EXPECT_TRUE(nir_opt_comparison_pre_impl(bld.impl));
527 }
528
TEST_F(comparison_pre_test,non_scalar_add_result)529 TEST_F(comparison_pre_test, non_scalar_add_result)
530 {
531 /* The optimization pass should not do anything because the result of the
532 * fadd is not a scalar.
533 *
534 * Before:
535 *
536 * vec4 32 ssa_0 = load_const (-2.0, -1.0, 1.0, 2.0)
537 * vec4 32 ssa_1 = load_const ( 2.0, 1.0, -1.0, -2.0)
538 * vec4 32 ssa_2 = load_const ( 3.0, 4.0, 5.0, 6.0)
539 * vec4 32 ssa_3 = fadd ssa_0, ssa_2
540 * vec1 1 ssa_4 = flt ssa_0.x, ssa_3.x
541 *
542 * if ssa_4 {
543 * vec2 32 ssa_5 = fadd ssa_1.xx, ssa_3.xx
544 * } else {
545 * }
546 *
547 * After:
548 *
549 * No change.
550 */
551 nir_ssa_def *a = nir_fadd(&bld, v1, v3);
552
553 nir_alu_instr *flt = nir_alu_instr_create(bld.shader, nir_op_flt);
554
555 flt->src[0].src = nir_src_for_ssa(v1);
556 flt->src[1].src = nir_src_for_ssa(a);
557
558 memcpy(&flt->src[0].swizzle, xxxx, sizeof(xxxx));
559 memcpy(&flt->src[1].swizzle, xxxx, sizeof(xxxx));
560
561 nir_builder_alu_instr_finish_and_insert(&bld, flt);
562
563 flt->dest.dest.ssa.num_components = 1;
564 flt->dest.write_mask = 1;
565
566 nir_if *nif = nir_push_if(&bld, &flt->dest.dest.ssa);
567
568 nir_alu_instr *fadd = nir_alu_instr_create(bld.shader, nir_op_fadd);
569
570 fadd->src[0].src = nir_src_for_ssa(v2);
571 fadd->src[1].src = nir_src_for_ssa(a);
572
573 memcpy(&fadd->src[0].swizzle, xxxx, sizeof(xxxx));
574 memcpy(&fadd->src[1].swizzle, xxxx, sizeof(xxxx));
575
576 nir_builder_alu_instr_finish_and_insert(&bld, fadd);
577
578 fadd->dest.dest.ssa.num_components = 2;
579 fadd->dest.write_mask = 3;
580
581 nir_pop_if(&bld, nif);
582
583 EXPECT_FALSE(nir_opt_comparison_pre_impl(bld.impl));
584 }
585