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 /*
26 * @test
27 * @bug 8304042
28 * @summary Test some examples with independent packs with cyclic dependency
29 * between the packs.
30 * @modules java.base/jdk.internal.misc
31 * @library /test/lib /
32 * @run driver compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency
33 */
34
35 package compiler.loopopts.superword;
36
37 import jdk.internal.misc.Unsafe;
38 import jdk.test.lib.Asserts;
39 import compiler.lib.ir_framework.*;
40
41 public class TestIndependentPacksWithCyclicDependency {
42 static final int RANGE = 1024;
43 static final int ITER = 10_000;
44 static Unsafe unsafe = Unsafe.getUnsafe();
45
46 int[] goldI0 = new int[RANGE];
47 float[] goldF0 = new float[RANGE];
48 int[] goldI1 = new int[RANGE];
49 float[] goldF1 = new float[RANGE];
50 int[] goldI2 = new int[RANGE];
51 float[] goldF2 = new float[RANGE];
52 int[] goldI3 = new int[RANGE];
53 float[] goldF3 = new float[RANGE];
54 int[] goldI4 = new int[RANGE];
55 float[] goldF4 = new float[RANGE];
56 int[] goldI5 = new int[RANGE];
57 float[] goldF5 = new float[RANGE];
58 int[] goldI6 = new int[RANGE];
59 float[] goldF6 = new float[RANGE];
60 long[] goldL6 = new long[RANGE];
61 int[] goldI7 = new int[RANGE];
62 float[] goldF7 = new float[RANGE];
63 long[] goldL7 = new long[RANGE];
64 int[] goldI8 = new int[RANGE];
65 float[] goldF8 = new float[RANGE];
66 long[] goldL8 = new long[RANGE];
67 int[] goldI9 = new int[RANGE];
68 float[] goldF9 = new float[RANGE];
69 long[] goldL9 = new long[RANGE];
70 int[] goldI10 = new int[RANGE];
71 float[] goldF10 = new float[RANGE];
72 long[] goldL10 = new long[RANGE];
73
74 public static void main(String args[]) {
75 TestFramework.runWithFlags("--add-modules", "java.base", "--add-exports", "java.base/jdk.internal.misc=ALL-UNNAMED",
76 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::test*",
77 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::verify",
78 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::init",
79 "-XX:+IgnoreUnrecognizedVMOptions", "-XX:LoopUnrollLimit=1000");
80 }
81
82 TestIndependentPacksWithCyclicDependency() {
83 // compute the gold standard in interpreter mode
84 init(goldI0, goldF0);
85 test0(goldI0, goldI0, goldF0, goldF0);
86 init(goldI1, goldF1);
87 test1(goldI1, goldI1, goldF1, goldF1);
88 init(goldI2, goldF2);
89 test2(goldI2, goldI2, goldF2, goldF2);
90 init(goldI3, goldF3);
91 test3(goldI3, goldI3, goldF3, goldF3);
92 init(goldI4, goldF4);
93 test4(goldI4, goldI4, goldF4, goldF4);
94 init(goldI5, goldF5);
95 test5(goldI5, goldI5, goldF5, goldF5);
96 init(goldI6, goldF6, goldL6);
97 test6(goldI6, goldI6, goldF6, goldF6, goldL6, goldL6);
98 init(goldI7, goldF7, goldL7);
99 test7(goldI7, goldI7, goldF7, goldF7, goldL7, goldL7);
100 init(goldI8, goldF8, goldL8);
101 test8(goldI8, goldI8, goldF8, goldF8, goldL8, goldL8);
102 init(goldI9, goldF9, goldL9);
103 test9(goldI9, goldI9, goldF9, goldF9, goldL9, goldL9);
104 init(goldI10, goldF10, goldL10);
105 test10(goldI10, goldI10, goldF10, goldF10, goldL10, goldL10);
106 }
107
108 @Run(test = "test0")
109 @Warmup(100)
110 public void runTest0() {
111 int[] dataI = new int[RANGE];
112 float[] dataF = new float[RANGE];
113 init(dataI, dataF);
114 test0(dataI, dataI, dataF, dataF);
115 verify("test0", dataI, goldI0);
116 verify("test0", dataF, goldF0);
117 }
118
119 @Test
120 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0"},
121 applyIfPlatform = {"64-bit", "true"},
122 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
123 static void test0(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
124 for (int i = 0; i < RANGE; i+=2) {
125 // Hand-unrolled 2x. Int and Float slice are completely separate.
126 dataIb[i+0] = dataIa[i+0] + 3;
127 dataIb[i+1] = dataIa[i+1] + 3;
128 dataFb[i+0] = dataFa[i+0] * 1.3f;
129 dataFb[i+1] = dataFa[i+1] * 1.3f;
130 }
131 }
132
133 @Run(test = "test1")
134 @Warmup(100)
135 public void runTest1() {
136 int[] dataI = new int[RANGE];
137 float[] dataF = new float[RANGE];
138 init(dataI, dataF);
139 test1(dataI, dataI, dataF, dataF);
140 verify("test1", dataI, goldI1);
141 verify("test1", dataF, goldF1);
142 }
143
144 @Test
145 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0", IRNode.VECTOR_CAST_F2I, "> 0", IRNode.VECTOR_CAST_I2F, "> 0"},
146 applyIfPlatform = {"64-bit", "true"},
147 applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true"})
148 static void test1(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
149 for (int i = 0; i < RANGE; i+=2) {
150 // Hand-unrolled 2x. Converst to and from. StoreF -> LoadF dependency.
151 dataFa[i+0] = dataIa[i+0] + 3;
152 dataFa[i+1] = dataIa[i+1] + 3;
153 dataIb[i+0] = (int)(dataFb[i+0] * 1.3f);
154 dataIb[i+1] = (int)(dataFb[i+1] * 1.3f);
155 }
156 }
157
158 @Run(test = "test2")
159 public void runTest2() {
160 int[] dataI = new int[RANGE];
161 float[] dataF = new float[RANGE];
162 init(dataI, dataF);
163 test2(dataI, dataI, dataF, dataF);
164 verify("test2", dataI, goldI2);
165 verify("test2", dataF, goldF2);
166 }
167
168 @Test
169 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VI, "> 0"},
170 applyIfPlatform = {"64-bit", "true"},
171 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
172 static void test2(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
173 for (int i = 0; i < RANGE; i+=2) {
174 // int and float arrays are two slices. But we pretend both are of type int.
175 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, dataIa[i+0] + 1);
176 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, dataIa[i+1] + 1);
177 dataIb[i+0] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0);
178 dataIb[i+1] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4);
179 }
180 }
181
182 @Run(test = "test3")
183 @Warmup(100)
184 public void runTest3() {
185 int[] dataI = new int[RANGE];
186 float[] dataF = new float[RANGE];
187 init(dataI, dataF);
188 test3(dataI, dataI, dataF, dataF);
189 verify("test3", dataI, goldI3);
190 verify("test3", dataF, goldF3);
191 }
192
193 @Test
194 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0"},
195 applyIfPlatform = {"64-bit", "true"},
196 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
197 static void test3(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
198 for (int i = 0; i < RANGE; i+=2) {
199 // Inversion of orders. But because we operate on separate slices, this should
200 // safely vectorize. It should detect that each line is independent, so it can
201 // reorder them.
202 dataIb[i+0] = dataIa[i+0] + 3;
203 dataFb[i+1] = dataFa[i+1] * 1.3f;
204 dataFb[i+0] = dataFa[i+0] * 1.3f;
205 dataIb[i+1] = dataIa[i+1] + 3;
206 }
207 }
208
209 @Run(test = "test4")
210 @Warmup(100)
211 public void runTest4() {
212 int[] dataI = new int[RANGE];
213 float[] dataF = new float[RANGE];
214 init(dataI, dataF);
215 test4(dataI, dataI, dataF, dataF);
216 verify("test4", dataI, goldI4);
217 verify("test4", dataF, goldF4);
218 }
219
220 @Test
221 static void test4(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
222 for (int i = 0; i < RANGE; i+=2) {
223 // same as test1, except that reordering leads to different semantics
224 // [A,B] and [X,Y] are both packs that are internally independent
225 // But we have dependencies A -> X (StoreF -> LoadF)
226 // and Y -> B (StoreI -> LoadI)
227 // Hence the two packs have a cyclic dependency, we cannot schedule
228 // one before the other.
229 dataFa[i+0] = dataIa[i+0] + 3; // A
230 dataIb[i+0] = (int)(dataFb[i+0] * 1.3f); // X
231 dataIb[i+1] = (int)(dataFb[i+1] * 1.3f); // Y
232 dataFa[i+1] = dataIa[i+1] + 3; // B
233 }
234 }
235
236 @Run(test = "test5")
237 public void runTest5() {
238 int[] dataI = new int[RANGE];
239 float[] dataF = new float[RANGE];
240 init(dataI, dataF);
241 test5(dataI, dataI, dataF, dataF);
242 verify("test5", dataI, goldI5);
243 verify("test5", dataF, goldF5);
244 }
245
246 @Test
247 static void test5(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
248 for (int i = 0; i < RANGE; i+=2) {
249 // same as test2, except that reordering leads to different semantics
250 // explanation analogue to test4
251 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, dataIa[i+0] + 1); // A
252 dataIb[i+0] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0); // X
253 dataIb[i+1] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4); // Y
254 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, dataIa[i+1] + 1); // B
255 }
256 }
257
258 @Run(test = "test6")
259 public void runTest6() {
260 int[] dataI = new int[RANGE];
261 float[] dataF = new float[RANGE];
262 long[] dataL = new long[RANGE];
263 init(dataI, dataF, dataL);
264 test6(dataI, dataI, dataF, dataF, dataL, dataL);
265 verify("test6", dataI, goldI6);
266 verify("test6", dataF, goldF6);
267 verify("test6", dataL, goldL6);
268 }
269
270 @Test
271 @IR(applyIf = {"UseCompactObjectHeaders", "false"},
272 counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VI, "> 0", IRNode.ADD_VF, "> 0"},
273 applyIfPlatform = {"64-bit", "true"},
274 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
275 static void test6(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
276 long[] dataLa, long[] dataLb) {
277 for (int i = 0; i < RANGE; i+=2) {
278 // Chain of parallelizable op and conversion
279 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3;
280 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3;
281 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
282 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
283 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45;
284 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) * 45;
285 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
286 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
287 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f;
288 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f;
289 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
290 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
291 }
292 }
293
294 @Run(test = "test7")
295 public void runTest7() {
296 int[] dataI = new int[RANGE];
297 float[] dataF = new float[RANGE];
298 long[] dataL = new long[RANGE];
299 init(dataI, dataF, dataL);
300 test7(dataI, dataI, dataF, dataF, dataL, dataL);
301 verify("test7", dataI, goldI7);
302 verify("test7", dataF, goldF7);
303 verify("test7", dataL, goldL7);
304 }
305
306 @Test
307 static void test7(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
308 long[] dataLa, long[] dataLb) {
309 for (int i = 0; i < RANGE; i+=2) {
310 // Cycle involving 3 memory slices
311 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3;
312 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
313 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45;
314 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) * 45;
315 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
316 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
317 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f;
318 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f;
319 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
320 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
321 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3; // moved down
322 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
323 }
324 }
325
326
327 @Run(test = "test8")
328 public void runTest8() {
329 int[] dataI = new int[RANGE];
330 float[] dataF = new float[RANGE];
331 long[] dataL = new long[RANGE];
332 init(dataI, dataF, dataL);
333 test8(dataI, dataI, dataF, dataF, dataL, dataL);
334 verify("test8", dataI, goldI8);
335 verify("test8", dataF, goldF8);
336 verify("test8", dataL, goldL8);
337 }
338
339 @Test
340 static void test8(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
341 long[] dataLa, long[] dataLb) {
342 for (int i = 0; i < RANGE; i+=2) {
343 // 2-cycle, with more ops after
344 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3;
345 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
346 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45;
347 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) * 45;
348 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
349 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
350 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3;
351 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
352 // more stuff after
353 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f;
354 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f;
355 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
356 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
357 }
358 }
359
360 @Run(test = "test9")
361 public void runTest9() {
362 int[] dataI = new int[RANGE];
363 float[] dataF = new float[RANGE];
364 long[] dataL = new long[RANGE];
365 init(dataI, dataF, dataL);
366 test9(dataI, dataI, dataF, dataF, dataL, dataL);
367 verify("test9", dataI, goldI9);
368 verify("test9", dataF, goldF9);
369 verify("test9", dataL, goldL9);
370 }
371
372 @Test
373 static void test9(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
374 long[] dataLa, long[] dataLb) {
375 for (int i = 0; i < RANGE; i+=2) {
376 // 2-cycle, with more stuff before
377 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f;
378 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f;
379 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
380 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
381 // 2-cycle
382 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3;
383 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
384 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45;
385 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) * 45;
386 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
387 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
388 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3;
389 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
390 }
391 }
392
393 @Run(test = "test10")
394 public void runTest10() {
395 int[] dataI = new int[RANGE];
396 float[] dataF = new float[RANGE];
397 long[] dataL = new long[RANGE];
398 init(dataI, dataF, dataL);
399 test10(dataI, dataI, dataF, dataF, dataL, dataL);
400 verify("test10", dataI, goldI10);
401 verify("test10", dataF, goldF10);
402 verify("test10", dataL, goldL10);
403 }
404
405 @Test
406 static void test10(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
407 long[] dataLa, long[] dataLb) {
408 for (int i = 0; i < RANGE; i+=2) {
409 // This creates the following graph before SuperWord:
410 //
411 // A -> R -> U
412 // S -> V -> B
413 //
414 // SuperWord analyzes the graph, and sees that [A,B] and [U,V]
415 // are adjacent, isomorphic and independent packs. However,
416 // [R,S] are not isomorphic (R mul, S add).
417 // So it vectorizes [A,B] and [U,V] this gives us this graph:
418 //
419 // -> R
420 // [A,B] -> [U,V] -+
421 // ^ -> S |
422 // | |
423 // +------------------+
424 //
425 // The cycle thus does not only go via packs, but also scalar ops.
426 //
427 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0) + 3; // A
428 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0, v00);
429 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 0) * 45; // R: constant mismatch
430 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4) + 43; // S
431 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0, v10);
432 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4, v11);
433 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 0) + 0.55f; // U
434 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4 * i + 4) + 0.55f; // V
435 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 0, v20);
436 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4, v21);
437 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4 * i + 4) + 3; // B: moved down
438 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i + 4, v01);
439 }
440 }
441
442 static void init(int[] dataI, float[] dataF) {
443 for (int i = 0; i < RANGE; i++) {
444 dataI[i] = i + 1;
445 dataF[i] = i + 0.1f;
446 }
447 }
448
449 static void init(int[] dataI, float[] dataF, long[] dataL) {
450 for (int i = 0; i < RANGE; i++) {
451 dataI[i] = i + 1;
452 dataF[i] = i + 0.1f;
453 dataL[i] = i + 1;
454 }
455 }
456
457 static void verify(String name, int[] data, int[] gold) {
458 for (int i = 0; i < RANGE; i++) {
459 if (data[i] != gold[i]) {
460 throw new RuntimeException(" Invalid " + name + " result: dataI[" + i + "]: " + data[i] + " != " + gold[i]);
461 }
462 }
463 }
464
465 static void verify(String name, float[] data, float[] gold) {
466 for (int i = 0; i < RANGE; i++) {
467 int datav = unsafe.getInt(data, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i);
468 int goldv = unsafe.getInt(gold, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4 * i);
469 if (datav != goldv) {
470 throw new RuntimeException(" Invalid " + name + " result: dataF[" + i + "]: " + datav + " != " + goldv);
471 }
472 }
473 }
474
475 static void verify(String name, long[] data, long[] gold) {
476 for (int i = 0; i < RANGE; i++) {
477 if (data[i] != gold[i]) {
478 throw new RuntimeException(" Invalid " + name + " result: dataL[" + i + "]: " + data[i] + " != " + gold[i]);
479 }
480 }
481 }
482 }
483