1 /*
2 * Copyright (c) 2023, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // Note Lilliput:
26 // Tests rely on array members starting at the same offset, otherwise vectorization does not kick in. Not true
27 // for Lilliput.
28 // For now I just enforce -CompactObjectHeaders.
29
30 /*
31 * @test
32 * @bug 8304042
33 * @summary Test some examples with independent packs with cyclic dependency
34 * between the packs.
35 * @requires vm.bits == 64
36 * @requires vm.compiler2.enabled
37 * @modules java.base/jdk.internal.misc
38 * @library /test/lib /
39 * @run driver compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency
40 */
41
42 package compiler.loopopts.superword;
43
44 import jdk.internal.misc.Unsafe;
45 import jdk.test.lib.Asserts;
46 import compiler.lib.ir_framework.*;
47
48 public class TestIndependentPacksWithCyclicDependency {
49 static final int RANGE = 1024;
50 static final int ITER = 10_000;
51 static Unsafe unsafe = Unsafe.getUnsafe();
52
53 int[] goldI0 = new int[RANGE];
54 float[] goldF0 = new float[RANGE];
55 int[] goldI1 = new int[RANGE];
56 float[] goldF1 = new float[RANGE];
57 int[] goldI2 = new int[RANGE];
58 float[] goldF2 = new float[RANGE];
59 int[] goldI3 = new int[RANGE];
60 float[] goldF3 = new float[RANGE];
61 int[] goldI4 = new int[RANGE];
62 float[] goldF4 = new float[RANGE];
63 int[] goldI5 = new int[RANGE];
64 float[] goldF5 = new float[RANGE];
65 int[] goldI6 = new int[RANGE];
66 float[] goldF6 = new float[RANGE];
67 long[] goldL6 = new long[RANGE];
68 int[] goldI7 = new int[RANGE];
69 float[] goldF7 = new float[RANGE];
70 long[] goldL7 = new long[RANGE];
71 int[] goldI8 = new int[RANGE];
72 float[] goldF8 = new float[RANGE];
73 long[] goldL8 = new long[RANGE];
74 int[] goldI9 = new int[RANGE];
75 float[] goldF9 = new float[RANGE];
76 long[] goldL9 = new long[RANGE];
77 int[] goldI10 = new int[RANGE];
78 float[] goldF10 = new float[RANGE];
79 long[] goldL10 = new long[RANGE];
80
81 public static void main(String args[]) {
82 TestFramework.runWithFlags("--add-modules", "java.base", "--add-exports", "java.base/jdk.internal.misc=ALL-UNNAMED",
83 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::test*",
84 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::verify",
85 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::init",
86 "-XX:LoopUnrollLimit=1000", "-XX:+UnlockExperimentalVMOptions", "-XX:-UseCompactObjectHeaders");
87 }
88
89 TestIndependentPacksWithCyclicDependency() {
90 // compute the gold standard in interpreter mode
91 init(goldI0, goldF0);
92 test0(goldI0, goldI0, goldF0, goldF0);
93 init(goldI1, goldF1);
94 test1(goldI1, goldI1, goldF1, goldF1);
95 init(goldI2, goldF2);
96 test2(goldI2, goldI2, goldF2, goldF2);
97 init(goldI3, goldF3);
98 test3(goldI3, goldI3, goldF3, goldF3);
99 init(goldI4, goldF4);
100 test4(goldI4, goldI4, goldF4, goldF4);
101 init(goldI5, goldF5);
102 test5(goldI5, goldI5, goldF5, goldF5);
103 init(goldI6, goldF6, goldL6);
104 test6(goldI6, goldI6, goldF6, goldF6, goldL6, goldL6);
105 init(goldI7, goldF7, goldL7);
106 test7(goldI7, goldI7, goldF7, goldF7, goldL7, goldL7);
107 init(goldI8, goldF8, goldL8);
108 test8(goldI8, goldI8, goldF8, goldF8, goldL8, goldL8);
109 init(goldI9, goldF9, goldL9);
110 test9(goldI9, goldI9, goldF9, goldF9, goldL9, goldL9);
111 init(goldI10, goldF10, goldL10);
112 test10(goldI10, goldI10, goldF10, goldF10, goldL10, goldL10);
113 }
114
115 @Run(test = "test0")
116 @Warmup(100)
117 public void runTest0() {
118 int[] dataI = new int[RANGE];
119 float[] dataF = new float[RANGE];
120 init(dataI, dataF);
121 test0(dataI, dataI, dataF, dataF);
122 verify("test0", dataI, goldI0);
123 verify("test0", dataF, goldF0);
124 }
125
126 @Test
127 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0"},
128 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
129 static void test0(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
130 for (int i = 0; i < RANGE; i+=2) {
131 // Hand-unrolled 2x. Int and Float slice are completely separate.
132 dataIb[i+0] = dataIa[i+0] + 3;
133 dataIb[i+1] = dataIa[i+1] + 3;
134 dataFb[i+0] = dataFa[i+0] * 1.3f;
135 dataFb[i+1] = dataFa[i+1] * 1.3f;
136 }
137 }
138
139 @Run(test = "test1")
140 @Warmup(100)
141 public void runTest1() {
142 int[] dataI = new int[RANGE];
143 float[] dataF = new float[RANGE];
144 init(dataI, dataF);
145 test1(dataI, dataI, dataF, dataF);
146 verify("test1", dataI, goldI1);
147 verify("test1", dataF, goldF1);
148 }
149
150 @Test
151 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0", IRNode.VECTOR_CAST_F2X, "> 0", IRNode.VECTOR_CAST_I2X, "> 0"},
152 applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true"})
153 static void test1(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
154 for (int i = 0; i < RANGE; i+=2) {
155 // Hand-unrolled 2x. Converst to and from. StoreF -> LoadF dependency.
156 dataFa[i+0] = dataIa[i+0] + 3;
157 dataFa[i+1] = dataIa[i+1] + 3;
158 dataIb[i+0] = (int)(dataFb[i+0] * 1.3f);
159 dataIb[i+1] = (int)(dataFb[i+1] * 1.3f);
160 }
161 }
162
163 @Run(test = "test2")
164 public void runTest2() {
165 int[] dataI = new int[RANGE];
166 float[] dataF = new float[RANGE];
167 init(dataI, dataF);
168 test2(dataI, dataI, dataF, dataF);
169 verify("test2", dataI, goldI2);
170 verify("test2", dataF, goldF2);
171 }
172
173 @Test
174 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VI, "> 0"},
175 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
176 static void test2(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
177 for (int i = 0; i < RANGE; i+=2) {
178 // int and float arrays are two slices. But we pretend both are of type int.
179 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, dataIa[i+0] + 1);
180 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, dataIa[i+1] + 1);
181 dataIb[i+0] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0);
182 dataIb[i+1] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4);
183 }
184 }
185
186 @Run(test = "test3")
187 @Warmup(100)
188 public void runTest3() {
189 int[] dataI = new int[RANGE];
190 float[] dataF = new float[RANGE];
191 init(dataI, dataF);
192 test3(dataI, dataI, dataF, dataF);
193 verify("test3", dataI, goldI3);
194 verify("test3", dataF, goldF3);
195 }
196
197 @Test
198 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0"},
199 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
200 static void test3(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
201 for (int i = 0; i < RANGE; i+=2) {
202 // Inversion of orders. But because we operate on separate slices, this should
203 // safely vectorize. It should detect that each line is independent, so it can
204 // reorder them.
205 dataIb[i+0] = dataIa[i+0] + 3;
206 dataFb[i+1] = dataFa[i+1] * 1.3f;
207 dataFb[i+0] = dataFa[i+0] * 1.3f;
208 dataIb[i+1] = dataIa[i+1] + 3;
209 }
210 }
211
212 @Run(test = "test4")
213 @Warmup(100)
214 public void runTest4() {
215 int[] dataI = new int[RANGE];
216 float[] dataF = new float[RANGE];
217 init(dataI, dataF);
218 test4(dataI, dataI, dataF, dataF);
219 verify("test4", dataI, goldI4);
220 verify("test4", dataF, goldF4);
221 }
222
223 @Test
224 static void test4(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
225 for (int i = 0; i < RANGE; i+=2) {
226 // same as test1, except that reordering leads to different semantics
227 // [A,B] and [X,Y] are both packs that are internally independent
228 // But we have dependencies A -> X (StoreF -> LoadF)
229 // and Y -> B (StoreI -> LoadI)
230 // Hence the two packs have a cyclic dependency, we cannot schedule
231 // one before the other.
232 dataFa[i+0] = dataIa[i+0] + 3; // A
233 dataIb[i+0] = (int)(dataFb[i+0] * 1.3f); // X
234 dataIb[i+1] = (int)(dataFb[i+1] * 1.3f); // Y
235 dataFa[i+1] = dataIa[i+1] + 3; // B
236 }
237 }
238
239 @Run(test = "test5")
240 public void runTest5() {
241 int[] dataI = new int[RANGE];
242 float[] dataF = new float[RANGE];
243 init(dataI, dataF);
244 test5(dataI, dataI, dataF, dataF);
245 verify("test5", dataI, goldI5);
246 verify("test5", dataF, goldF5);
247 }
248
249 @Test
250 static void test5(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
251 for (int i = 0; i < RANGE; i+=2) {
252 // same as test2, except that reordering leads to different semantics
253 // explanation analogue to test4
254 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, dataIa[i+0] + 1); // A
255 dataIb[i+0] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0); // X
256 dataIb[i+1] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4); // Y
257 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, dataIa[i+1] + 1); // B
258 }
259 }
260
261 @Run(test = "test6")
262 public void runTest6() {
263 int[] dataI = new int[RANGE];
264 float[] dataF = new float[RANGE];
265 long[] dataL = new long[RANGE];
266 init(dataI, dataF, dataL);
267 test6(dataI, dataI, dataF, dataF, dataL, dataL);
268 verify("test6", dataI, goldI6);
269 verify("test6", dataF, goldF6);
270 verify("test6", dataL, goldL6);
271 }
272
273 @Test
274 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VI, "> 0", IRNode.ADD_VF, "> 0"},
275 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
276 static void test6(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
277 long[] dataLa, long[] dataLb) {
278 for (int i = 0; i < RANGE; i+=2) {
279 // Chain of parallelizable op and conversion
280 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3;
281 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3;
282 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
283 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
284 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45;
285 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) * 45;
286 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
287 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
288 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f;
289 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f;
290 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
291 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
292 }
293 }
294
295 @Run(test = "test7")
296 public void runTest7() {
297 int[] dataI = new int[RANGE];
298 float[] dataF = new float[RANGE];
299 long[] dataL = new long[RANGE];
300 init(dataI, dataF, dataL);
301 test7(dataI, dataI, dataF, dataF, dataL, dataL);
302 verify("test7", dataI, goldI7);
303 verify("test7", dataF, goldF7);
304 verify("test7", dataL, goldL7);
305 }
306
307 @Test
308 static void test7(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
309 long[] dataLa, long[] dataLb) {
310 for (int i = 0; i < RANGE; i+=2) {
311 // Cycle involving 3 memory slices
312 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3;
313 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
314 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45;
315 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) * 45;
316 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
317 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
318 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f;
319 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f;
320 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
321 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
322 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3; // moved down
323 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
324 }
325 }
326
327
328 @Run(test = "test8")
329 public void runTest8() {
330 int[] dataI = new int[RANGE];
331 float[] dataF = new float[RANGE];
332 long[] dataL = new long[RANGE];
333 init(dataI, dataF, dataL);
334 test8(dataI, dataI, dataF, dataF, dataL, dataL);
335 verify("test8", dataI, goldI8);
336 verify("test8", dataF, goldF8);
337 verify("test8", dataL, goldL8);
338 }
339
340 @Test
341 static void test8(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
342 long[] dataLa, long[] dataLb) {
343 for (int i = 0; i < RANGE; i+=2) {
344 // 2-cycle, with more ops after
345 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3;
346 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
347 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45;
348 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) * 45;
349 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
350 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
351 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3;
352 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
353 // more stuff after
354 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f;
355 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f;
356 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
357 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
358 }
359 }
360
361 @Run(test = "test9")
362 public void runTest9() {
363 int[] dataI = new int[RANGE];
364 float[] dataF = new float[RANGE];
365 long[] dataL = new long[RANGE];
366 init(dataI, dataF, dataL);
367 test9(dataI, dataI, dataF, dataF, dataL, dataL);
368 verify("test9", dataI, goldI9);
369 verify("test9", dataF, goldF9);
370 verify("test9", dataL, goldL9);
371 }
372
373 @Test
374 static void test9(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
375 long[] dataLa, long[] dataLb) {
376 for (int i = 0; i < RANGE; i+=2) {
377 // 2-cycle, with more stuff before
378 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f;
379 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f;
380 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
381 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
382 // 2-cycle
383 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3;
384 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
385 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45;
386 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) * 45;
387 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
388 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
389 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3;
390 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
391 }
392 }
393
394 @Run(test = "test10")
395 public void runTest10() {
396 int[] dataI = new int[RANGE];
397 float[] dataF = new float[RANGE];
398 long[] dataL = new long[RANGE];
399 init(dataI, dataF, dataL);
400 test10(dataI, dataI, dataF, dataF, dataL, dataL);
401 verify("test10", dataI, goldI10);
402 verify("test10", dataF, goldF10);
403 verify("test10", dataL, goldL10);
404 }
405
406 @Test
407 static void test10(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
408 long[] dataLa, long[] dataLb) {
409 for (int i = 0; i < RANGE; i+=2) {
410 // This creates the following graph before SuperWord:
411 //
412 // A -> R -> U
413 // S -> V -> B
414 //
415 // SuperWord analyzes the graph, and sees that [A,B] and [U,V]
416 // are adjacent, isomorphic and independent packs. However,
417 // [R,S] are not isomorphic (R mul, S add).
418 // So it vectorizes [A,B] and [U,V] this gives us this graph:
419 //
420 // -> R
421 // [A,B] -> [U,V] -+
422 // ^ -> S |
423 // | |
424 // +------------------+
425 //
426 // The cycle thus does not only go via packs, but also scalar ops.
427 //
428 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3; // A
429 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
430 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45; // R: constant mismatch
431 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) + 43; // S
432 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
433 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
434 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f; // U
435 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f; // V
436 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
437 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
438 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3; // B: moved down
439 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
440 }
441 }
442
443 static void init(int[] dataI, float[] dataF) {
444 for (int i = 0; i < RANGE; i++) {
445 dataI[i] = i + 1;
446 dataF[i] = i + 0.1f;
447 }
448 }
449
450 static void init(int[] dataI, float[] dataF, long[] dataL) {
451 for (int i = 0; i < RANGE; i++) {
452 dataI[i] = i + 1;
453 dataF[i] = i + 0.1f;
454 dataL[i] = i + 1;
455 }
456 }
457
458 static void verify(String name, int[] data, int[] gold) {
459 for (int i = 0; i < RANGE; i++) {
460 if (data[i] != gold[i]) {
461 throw new RuntimeException(" Invalid " + name + " result: dataI[" + i + "]: " + data[i] + " != " + gold[i]);
462 }
463 }
464 }
465
466 static void verify(String name, float[] data, float[] gold) {
467 for (int i = 0; i < RANGE; i++) {
468 int datav = unsafe.getInt(data, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i);
469 int goldv = unsafe.getInt(gold, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i);
470 if (datav != goldv) {
471 throw new RuntimeException(" Invalid " + name + " result: dataF[" + i + "]: " + datav + " != " + goldv);
472 }
473 }
474 }
475
476 static void verify(String name, long[] data, long[] gold) {
477 for (int i = 0; i < RANGE; i++) {
478 if (data[i] != gold[i]) {
479 throw new RuntimeException(" Invalid " + name + " result: dataL[" + i + "]: " + data[i] + " != " + gold[i]);
480 }
481 }
482 }
483 }
484