1 /*
2 * Copyright (c) 2023, 2024, 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 nCOH_nAV
33 * @run driver compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency nCOH_yAV
34 * @run driver compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency yCOH_nAV
35 * @run driver compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency yCOH_yAV
36 */
37
38 package compiler.loopopts.superword;
39
40 import jdk.internal.misc.Unsafe;
41 import jdk.test.lib.Asserts;
42 import compiler.lib.ir_framework.*;
43
44 public class TestIndependentPacksWithCyclicDependency {
45 static final int RANGE = 1024;
46 static final int ITER = 10_000;
47 static Unsafe unsafe = Unsafe.getUnsafe();
48
49 int[] goldI0 = new int[RANGE];
50 float[] goldF0 = new float[RANGE];
51 int[] goldI1 = new int[RANGE];
52 float[] goldF1 = new float[RANGE];
53 int[] goldI2 = new int[RANGE];
54 float[] goldF2 = new float[RANGE];
55 int[] goldI3 = new int[RANGE];
56 float[] goldF3 = new float[RANGE];
57 int[] goldI4 = new int[RANGE];
58 float[] goldF4 = new float[RANGE];
59 int[] goldI5 = new int[RANGE];
60 float[] goldF5 = new float[RANGE];
61 int[] goldI6 = new int[RANGE];
62 float[] goldF6 = new float[RANGE];
63 long[] goldL6 = new long[RANGE];
64 int[] goldI7 = new int[RANGE];
65 float[] goldF7 = new float[RANGE];
66 long[] goldL7 = new long[RANGE];
67 int[] goldI8 = new int[RANGE];
68 float[] goldF8 = new float[RANGE];
69 long[] goldL8 = new long[RANGE];
70 int[] goldI9 = new int[RANGE];
71 float[] goldF9 = new float[RANGE];
72 long[] goldL9 = new long[RANGE];
73 int[] goldI10 = new int[RANGE];
74 float[] goldF10 = new float[RANGE];
75 long[] goldL10 = new long[RANGE];
76
77 public static void main(String args[]) {
78 TestFramework framework = new TestFramework(TestIndependentPacksWithCyclicDependency.class);
79 framework.addFlags("--add-modules", "java.base", "--add-exports", "java.base/jdk.internal.misc=ALL-UNNAMED",
80 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::test*",
81 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::verify",
82 "-XX:CompileCommand=compileonly,compiler.loopopts.superword.TestIndependentPacksWithCyclicDependency::init",
83 "-XX:+IgnoreUnrecognizedVMOptions", "-XX:LoopUnrollLimit=1000");
84 switch (args[0]) {
85 case "nCOH_nAV" -> { framework.addFlags("-XX:+UnlockExperimentalVMOptions", "-XX:-UseCompactObjectHeaders", "-XX:-AlignVector"); }
86 case "nCOH_yAV" -> { framework.addFlags("-XX:+UnlockExperimentalVMOptions", "-XX:-UseCompactObjectHeaders", "-XX:+AlignVector"); }
87 case "yCOH_nAV" -> { framework.addFlags("-XX:+UnlockExperimentalVMOptions", "-XX:+UseCompactObjectHeaders", "-XX:-AlignVector"); }
88 case "yCOH_yAV" -> { framework.addFlags("-XX:+UnlockExperimentalVMOptions", "-XX:+UseCompactObjectHeaders", "-XX:+AlignVector"); }
89 default -> { throw new RuntimeException("Test argument not recognized: " + args[0]); }
90 };
91 framework.start();
92 }
93
94 TestIndependentPacksWithCyclicDependency() {
95 // compute the gold standard in interpreter mode
96 init(goldI0, goldF0);
97 test0(goldI0, goldI0, goldF0, goldF0);
98 init(goldI1, goldF1);
99 test1(goldI1, goldI1, goldF1, goldF1);
100 init(goldI2, goldF2);
101 test2(goldI2, goldI2, goldF2, goldF2);
102 init(goldI3, goldF3);
103 test3(goldI3, goldI3, goldF3, goldF3);
104 init(goldI4, goldF4);
105 test4(goldI4, goldI4, goldF4, goldF4);
106 init(goldI5, goldF5);
107 test5(goldI5, goldI5, goldF5, goldF5);
108 init(goldI6, goldF6, goldL6);
109 test6(goldI6, goldI6, goldF6, goldF6, goldL6, goldL6);
110 init(goldI7, goldF7, goldL7);
111 test7(goldI7, goldI7, goldF7, goldF7, goldL7, goldL7);
112 init(goldI8, goldF8, goldL8);
113 test8(goldI8, goldI8, goldF8, goldF8, goldL8, goldL8);
114 init(goldI9, goldF9, goldL9);
115 test9(goldI9, goldI9, goldF9, goldF9, goldL9, goldL9);
116 init(goldI10, goldF10, goldL10);
117 test10(goldI10, goldI10, goldF10, goldF10, goldL10, goldL10);
118 }
119
120 @Run(test = "test0")
121 @Warmup(100)
122 public void runTest0() {
123 int[] dataI = new int[RANGE];
124 float[] dataF = new float[RANGE];
125 init(dataI, dataF);
126 test0(dataI, dataI, dataF, dataF);
127 verify("test0", dataI, goldI0);
128 verify("test0", dataF, goldF0);
129 }
130
131 @Test
132 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0"},
133 applyIfPlatform = {"64-bit", "true"},
134 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true", "rvv", "true"})
135 static void test0(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
136 for (int i = 0; i < RANGE; i+=2) {
137 // Hand-unrolled 2x. Int and Float slice are completely separate.
138 dataIb[i+0] = dataIa[i+0] + 3;
139 dataIb[i+1] = dataIa[i+1] + 3;
140 dataFb[i+0] = dataFa[i+0] * 1.3f;
141 dataFb[i+1] = dataFa[i+1] * 1.3f;
142 }
143 }
144
145 @Run(test = "test1")
146 @Warmup(100)
147 public void runTest1() {
148 int[] dataI = new int[RANGE];
149 float[] dataF = new float[RANGE];
150 init(dataI, dataF);
151 test1(dataI, dataI, dataF, dataF);
152 verify("test1", dataI, goldI1);
153 verify("test1", dataF, goldF1);
154 }
155
156 @Test
157 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0", IRNode.VECTOR_CAST_F2I, "> 0", IRNode.VECTOR_CAST_I2F, "> 0"},
158 applyIfPlatform = {"64-bit", "true"},
159 applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true", "rvv", "true"})
160 static void test1(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
161 for (int i = 0; i < RANGE; i+=2) {
162 // Hand-unrolled 2x. Converst to and from. StoreF -> LoadF dependency.
163 dataFa[i+0] = dataIa[i+0] + 3;
164 dataFa[i+1] = dataIa[i+1] + 3;
165 dataIb[i+0] = (int)(dataFb[i+0] * 1.3f);
166 dataIb[i+1] = (int)(dataFb[i+1] * 1.3f);
167 }
168 }
169
170 @Run(test = "test2")
171 public void runTest2() {
172 int[] dataI = new int[RANGE];
173 float[] dataF = new float[RANGE];
174 init(dataI, dataF);
175 test2(dataI, dataI, dataF, dataF);
176 verify("test2", dataI, goldI2);
177 verify("test2", dataF, goldF2);
178 }
179
180 @Test
181 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VI, "> 0"},
182 applyIfPlatform = {"64-bit", "true"},
183 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true", "rvv", "true"})
184 static void test2(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
185 for (int i = 0; i < RANGE; i+=2) {
186 // int and float arrays are two slices. But we pretend both are of type int.
187 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0, dataIa[i+0] + 1);
188 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4, dataIa[i+1] + 1);
189 dataIb[i+0] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0);
190 dataIb[i+1] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4);
191 }
192 }
193
194 @Run(test = "test3")
195 @Warmup(100)
196 public void runTest3() {
197 int[] dataI = new int[RANGE];
198 float[] dataF = new float[RANGE];
199 init(dataI, dataF);
200 test3(dataI, dataI, dataF, dataF);
201 verify("test3", dataI, goldI3);
202 verify("test3", dataF, goldF3);
203 }
204
205 @Test
206 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VF, "> 0"},
207 applyIfPlatform = {"64-bit", "true"},
208 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true", "rvv", "true"})
209 static void test3(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
210 for (int i = 0; i < RANGE; i+=2) {
211 // Inversion of orders. But because we operate on separate slices, this should
212 // safely vectorize. It should detect that each line is independent, so it can
213 // reorder them.
214 dataIb[i+0] = dataIa[i+0] + 3;
215 dataFb[i+1] = dataFa[i+1] * 1.3f;
216 dataFb[i+0] = dataFa[i+0] * 1.3f;
217 dataIb[i+1] = dataIa[i+1] + 3;
218 }
219 }
220
221 @Run(test = "test4")
222 @Warmup(100)
223 public void runTest4() {
224 int[] dataI = new int[RANGE];
225 float[] dataF = new float[RANGE];
226 init(dataI, dataF);
227 test4(dataI, dataI, dataF, dataF);
228 verify("test4", dataI, goldI4);
229 verify("test4", dataF, goldF4);
230 }
231
232 @Test
233 static void test4(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
234 for (int i = 0; i < RANGE; i+=2) {
235 // same as test1, except that reordering leads to different semantics
236 // [A,B] and [X,Y] are both packs that are internally independent
237 // But we have dependencies A -> X (StoreF -> LoadF)
238 // and Y -> B (StoreI -> LoadI)
239 // Hence the two packs have a cyclic dependency, we cannot schedule
240 // one before the other.
241 dataFa[i+0] = dataIa[i+0] + 3; // A
242 dataIb[i+0] = (int)(dataFb[i+0] * 1.3f); // X
243 dataIb[i+1] = (int)(dataFb[i+1] * 1.3f); // Y
244 dataFa[i+1] = dataIa[i+1] + 3; // B
245 }
246 }
247
248 @Run(test = "test5")
249 public void runTest5() {
250 int[] dataI = new int[RANGE];
251 float[] dataF = new float[RANGE];
252 init(dataI, dataF);
253 test5(dataI, dataI, dataF, dataF);
254 verify("test5", dataI, goldI5);
255 verify("test5", dataF, goldF5);
256 }
257
258 @Test
259 static void test5(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb) {
260 for (int i = 0; i < RANGE; i+=2) {
261 // same as test2, except that reordering leads to different semantics
262 // explanation analogue to test4
263 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0, dataIa[i+0] + 1); // A
264 dataIb[i+0] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0); // X
265 dataIb[i+1] = 11 * unsafe.getInt(dataFb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4); // Y
266 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4, dataIa[i+1] + 1); // B
267 }
268 }
269
270 @Run(test = "test6")
271 public void runTest6() {
272 int[] dataI = new int[RANGE];
273 float[] dataF = new float[RANGE];
274 long[] dataL = new long[RANGE];
275 init(dataI, dataF, dataL);
276 test6(dataI, dataI, dataF, dataF, dataL, dataL);
277 verify("test6", dataI, goldI6);
278 verify("test6", dataF, goldF6);
279 verify("test6", dataL, goldL6);
280 }
281
282 @Test
283 @IR(counts = {IRNode.ADD_VI, "> 0", IRNode.MUL_VI, "> 0", IRNode.ADD_VF, "> 0"},
284 applyIfPlatform = {"64-bit", "true"},
285 applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true", "rvv", "true"})
286 static void test6(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
287 long[] dataLa, long[] dataLb) {
288 for (int i = 0; i < RANGE; i+=2) {
289 // Chain of parallelizable op and conversion
290 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0) + 3;
291 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4) + 3;
292 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0, v00);
293 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4, v01);
294 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0) * 45;
295 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4) * 45;
296 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0, v10);
297 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4, v11);
298 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0) + 0.55f;
299 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4) + 0.55f;
300 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0, v20);
301 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4, v21);
302 }
303 }
304
305 @Run(test = "test7")
306 public void runTest7() {
307 int[] dataI = new int[RANGE];
308 float[] dataF = new float[RANGE];
309 long[] dataL = new long[RANGE];
310 init(dataI, dataF, dataL);
311 test7(dataI, dataI, dataF, dataF, dataL, dataL);
312 verify("test7", dataI, goldI7);
313 verify("test7", dataF, goldF7);
314 verify("test7", dataL, goldL7);
315 }
316
317 @Test
318 static void test7(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
319 long[] dataLa, long[] dataLb) {
320 for (int i = 0; i < RANGE; i+=2) {
321 // Cycle involving 3 memory slices
322 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0) + 3;
323 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0, v00);
324 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0) * 45;
325 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4) * 45;
326 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0, v10);
327 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4, v11);
328 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0) + 0.55f;
329 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4) + 0.55f;
330 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0, v20);
331 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4, v21);
332 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4) + 3; // moved down
333 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4, v01);
334 }
335 }
336
337
338 @Run(test = "test8")
339 public void runTest8() {
340 int[] dataI = new int[RANGE];
341 float[] dataF = new float[RANGE];
342 long[] dataL = new long[RANGE];
343 init(dataI, dataF, dataL);
344 test8(dataI, dataI, dataF, dataF, dataL, dataL);
345 verify("test8", dataI, goldI8);
346 verify("test8", dataF, goldF8);
347 verify("test8", dataL, goldL8);
348 }
349
350 @Test
351 static void test8(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
352 long[] dataLa, long[] dataLb) {
353 for (int i = 0; i < RANGE; i+=2) {
354 // 2-cycle, with more ops after
355 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0) + 3;
356 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0, v00);
357 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0) * 45;
358 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4) * 45;
359 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0, v10);
360 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4, v11);
361 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4) + 3;
362 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4, v01);
363 // more stuff after
364 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0) + 0.55f;
365 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4) + 0.55f;
366 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0, v20);
367 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4, v21);
368 }
369 }
370
371 @Run(test = "test9")
372 public void runTest9() {
373 int[] dataI = new int[RANGE];
374 float[] dataF = new float[RANGE];
375 long[] dataL = new long[RANGE];
376 init(dataI, dataF, dataL);
377 test9(dataI, dataI, dataF, dataF, dataL, dataL);
378 verify("test9", dataI, goldI9);
379 verify("test9", dataF, goldF9);
380 verify("test9", dataL, goldL9);
381 }
382
383 @Test
384 static void test9(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
385 long[] dataLa, long[] dataLb) {
386 for (int i = 0; i < RANGE; i+=2) {
387 // 2-cycle, with more stuff before
388 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0) + 0.55f;
389 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4) + 0.55f;
390 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0, v20);
391 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4, v21);
392 // 2-cycle
393 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0) + 3;
394 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0, v00);
395 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0) * 45;
396 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4) * 45;
397 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0, v10);
398 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4, v11);
399 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4) + 3;
400 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4, v01);
401 }
402 }
403
404 @Run(test = "test10")
405 public void runTest10() {
406 int[] dataI = new int[RANGE];
407 float[] dataF = new float[RANGE];
408 long[] dataL = new long[RANGE];
409 init(dataI, dataF, dataL);
410 test10(dataI, dataI, dataF, dataF, dataL, dataL);
411 verify("test10", dataI, goldI10);
412 verify("test10", dataF, goldF10);
413 verify("test10", dataL, goldL10);
414 }
415
416 @Test
417 static void test10(int[] dataIa, int[] dataIb, float[] dataFa, float[] dataFb,
418 long[] dataLa, long[] dataLb) {
419 for (int i = 0; i < RANGE; i+=2) {
420 // This creates the following graph before SuperWord:
421 //
422 // A -> R -> U
423 // S -> V -> B
424 //
425 // SuperWord analyzes the graph, and sees that [A,B] and [U,V]
426 // are adjacent, isomorphic and independent packs. However,
427 // [R,S] are not isomorphic (R mul, S add).
428 // So it vectorizes [A,B] and [U,V] this gives us this graph:
429 //
430 // -> R
431 // [A,B] -> [U,V] -+
432 // ^ -> S |
433 // | |
434 // +------------------+
435 //
436 // The cycle thus does not only go via packs, but also scalar ops.
437 //
438 int v00 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0) + 3; // A
439 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0, v00);
440 int v10 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 0) * 45; // R: constant mismatch
441 int v11 = unsafe.getInt(dataFb, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4) + 43; // S
442 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0, v10);
443 unsafe.putInt(dataLa, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4, v11);
444 float v20 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 0) + 0.55f; // U
445 float v21 = unsafe.getFloat(dataLb, unsafe.ARRAY_LONG_BASE_OFFSET + 4L * i + 4) + 0.55f; // V
446 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 0, v20);
447 unsafe.putFloat(dataIb, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4, v21);
448 int v01 = unsafe.getInt(dataIa, unsafe.ARRAY_INT_BASE_OFFSET + 4L * i + 4) + 3; // B: moved down
449 unsafe.putInt(dataFa, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i + 4, v01);
450 }
451 }
452
453 static void init(int[] dataI, float[] dataF) {
454 for (int i = 0; i < RANGE; i++) {
455 dataI[i] = i + 1;
456 dataF[i] = i + 0.1f;
457 }
458 }
459
460 static void init(int[] dataI, float[] dataF, long[] dataL) {
461 for (int i = 0; i < RANGE; i++) {
462 dataI[i] = i + 1;
463 dataF[i] = i + 0.1f;
464 dataL[i] = i + 1;
465 }
466 }
467
468 static void verify(String name, int[] data, int[] gold) {
469 for (int i = 0; i < RANGE; i++) {
470 if (data[i] != gold[i]) {
471 throw new RuntimeException(" Invalid " + name + " result: dataI[" + i + "]: " + data[i] + " != " + gold[i]);
472 }
473 }
474 }
475
476 static void verify(String name, float[] data, float[] gold) {
477 for (int i = 0; i < RANGE; i++) {
478 int datav = unsafe.getInt(data, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i);
479 int goldv = unsafe.getInt(gold, unsafe.ARRAY_FLOAT_BASE_OFFSET + 4L * i);
480 if (datav != goldv) {
481 throw new RuntimeException(" Invalid " + name + " result: dataF[" + i + "]: " + datav + " != " + goldv);
482 }
483 }
484 }
485
486 static void verify(String name, long[] data, long[] gold) {
487 for (int i = 0; i < RANGE; i++) {
488 if (data[i] != gold[i]) {
489 throw new RuntimeException(" Invalid " + name + " result: dataL[" + i + "]: " + data[i] + " != " + gold[i]);
490 }
491 }
492 }
493 }
494