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 * @test
26 * @bug 8310886 8325252 8320622
27 * @summary Test MulAddS2I vectorization.
28 * @library /test/lib /
29 * @run driver compiler.loopopts.superword.TestMulAddS2I
30 */
31
32 package compiler.loopopts.superword;
33
34 import compiler.lib.ir_framework.*;
35 import jdk.test.lib.Asserts;
36 import jdk.test.lib.Platform;
37
38 public class TestMulAddS2I {
39 static final int RANGE = 1024*16;
40 static final int ITER = RANGE/2 - 1;
41
42 static short[] sArr1 = new short[RANGE];
43 static short[] sArr2 = new short[RANGE];
44 static final int[] GOLDEN_A;
45 static final int[] GOLDEN_B;
46 static final int[] GOLDEN_C;
47 static final int[] GOLDEN_D;
48 static final int[] GOLDEN_E;
49 static final int[] GOLDEN_F;
50 static final int[] GOLDEN_G;
51 static final int[] GOLDEN_H;
52 static final int[] GOLDEN_I;
53 static final int[] GOLDEN_J;
54 static final int[] GOLDEN_K;
55 static final int[] GOLDEN_L;
56 static final int[] GOLDEN_M;
57
58 static {
59 for (int i = 0; i < RANGE; i++) {
60 sArr1[i] = (short)(AbstractInfo.getRandom().nextInt());
61 sArr2[i] = (short)(AbstractInfo.getRandom().nextInt());
62 }
63 GOLDEN_A = testa();
64 GOLDEN_B = testb();
65 GOLDEN_C = testc(new int[ITER]);
66 GOLDEN_D = testd(new int[ITER]);
67 GOLDEN_E = teste(new int[ITER]);
68 GOLDEN_F = testf(new int[ITER]);
69 GOLDEN_G = testg(new int[ITER]);
70 GOLDEN_H = testh(new int[ITER]);
71 GOLDEN_I = testi(new int[ITER]);
72 GOLDEN_J = testj(new int[ITER]);
73 GOLDEN_K = testk(new int[ITER]);
74 GOLDEN_L = testl(new int[ITER]);
75 GOLDEN_M = testm(new int[ITER]);
76 }
77
78
79 public static void main(String[] args) {
80 TestFramework.runWithFlags("-XX:+IgnoreUnrecognizedVMOptions", "-XX:-AlignVector", "-XX:+UnlockExperimentalVMOptions", "-XX:-UseCompactObjectHeaers");
81 TestFramework.runWithFlags("-XX:+IgnoreUnrecognizedVMOptions", "-XX:+AlignVector", "-XX:+UnlockExperimentalVMOptions", "-XX:-UseCompactObjectHeaers");
82 TestFramework.runWithFlags("-XX:+IgnoreUnrecognizedVMOptions", "-XX:-AlignVector", "-XX:+UnlockExperimentalVMOptions", "-XX:+UseCompactObjectHeaers");
83 TestFramework.runWithFlags("-XX:+IgnoreUnrecognizedVMOptions", "-XX:+AlignVector", "-XX:+UnlockExperimentalVMOptions", "-XX:+UseCompactObjectHeaers");
84 }
85
86 @Run(test = {"testa", "testb", "testc", "testd", "teste", "testf", "testg", "testh",
87 "testi", "testj", "testk", "testl", "testm"})
88 @Warmup(0)
89 public static void run() {
90 compare(testa(), GOLDEN_A, "testa");
91 compare(testb(), GOLDEN_B, "testb");
92 compare(testc(new int[ITER]), GOLDEN_C, "testc");
93 compare(testd(new int[ITER]), GOLDEN_D, "testd");
94 compare(teste(new int[ITER]), GOLDEN_E, "teste");
95 compare(testf(new int[ITER]), GOLDEN_F, "testf");
96 compare(testg(new int[ITER]), GOLDEN_G, "testg");
97 compare(testh(new int[ITER]), GOLDEN_H, "testh");
98 compare(testi(new int[ITER]), GOLDEN_I, "testi");
99 compare(testj(new int[ITER]), GOLDEN_J, "testj");
100 compare(testk(new int[ITER]), GOLDEN_K, "testk");
101 compare(testl(new int[ITER]), GOLDEN_L, "testl");
102 compare(testm(new int[ITER]), GOLDEN_M, "testm");
103 }
104
105 public static void compare(int[] out, int[] golden, String name) {
106 for (int i = 0; i < ITER; i++) {
107 Asserts.assertEQ(out[i], golden[i], "wrong result for '" + name + "' out[" + i + "]");
108 }
109 }
110
111 @Test
112 @IR(applyIfCPUFeature = {"sse2", "true"},
113 applyIfPlatform = {"64-bit", "true"},
114 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
115 @IR(applyIfCPUFeature = {"asimd", "true"},
116 applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
117 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
118 @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
119 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
120 public static int[] testa() {
121 int[] out = new int[ITER];
122 int[] out2 = new int[ITER];
123 for (int i = 0; i < ITER; i++) {
124 out[i] += ((sArr1[2*i] * sArr1[2*i]) + (sArr1[2*i+1] * sArr1[2*i+1]));
125 out2[i] += out[i];
126 }
127 return out;
128 }
129
130 @Test
131 @IR(applyIfCPUFeature = {"sse2", "true"},
132 applyIfPlatform = {"64-bit", "true"},
133 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
134 @IR(applyIfCPUFeature = {"asimd", "true"},
135 applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
136 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
137 @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
138 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
139 public static int[] testb() {
140 int[] out = new int[ITER];
141 int[] out2 = new int[ITER];
142 for (int i = 0; i < ITER; i++) {
143 out[i] += ((sArr1[2*i] * sArr2[2*i]) + (sArr1[2*i+1] * sArr2[2*i+1]));
144 out2[i] += out[i];
145 }
146 return out;
147 }
148
149 @Test
150 @IR(applyIfCPUFeature = {"sse2", "true"},
151 applyIfPlatform = {"64-bit", "true"},
152 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
153 @IR(applyIfCPUFeature = {"asimd", "true"},
154 applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
155 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
156 @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
157 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
158 public static int[] testc(int[] out) {
159 for (int i = 0; i < ITER; i++) {
160 out[i] += ((sArr1[2*i] * sArr2[2*i]) + (sArr1[2*i+1] * sArr2[2*i+1]));
161 }
162 return out;
163 }
164
165 @Test
166 @IR(applyIfCPUFeature = {"sse2", "true"},
167 applyIfPlatform = {"64-bit", "true"},
168 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
169 @IR(applyIfCPUFeature = {"asimd", "true"},
170 applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
171 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
172 @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
173 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
174 public static int[] testd(int[] out) {
175 for (int i = 0; i < ITER-2; i+=2) {
176 // Unrolled, with the same structure.
177 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
178 out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+2]) + (sArr1[2*i+3] * sArr2[2*i+3]));
179 }
180 return out;
181 }
182
183 @Test
184 @IR(applyIfCPUFeature = {"sse2", "true"},
185 applyIfPlatform = {"64-bit", "true"},
186 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
187 @IR(applyIfCPUFeature = {"asimd", "true"},
188 applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
189 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
190 @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
191 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
192 public static int[] teste(int[] out) {
193 for (int i = 0; i < ITER-2; i+=2) {
194 // Unrolled, with some swaps.
195 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
196 out[i+1] += ((sArr2[2*i+2] * sArr1[2*i+2]) + (sArr1[2*i+3] * sArr2[2*i+3])); // swap(1 2)
197 }
198 return out;
199 }
200
201 @Test
202 @IR(applyIfCPUFeature = {"sse2", "true"},
203 applyIfPlatform = {"64-bit", "true"},
204 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
205 @IR(applyIfCPUFeature = {"asimd", "true"},
206 applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
207 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
208 @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
209 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
210 public static int[] testf(int[] out) {
211 for (int i = 0; i < ITER-2; i+=2) {
212 // Unrolled, with some swaps.
213 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
214 out[i+1] += ((sArr2[2*i+2] * sArr1[2*i+2]) + (sArr2[2*i+3] * sArr1[2*i+3])); // swap(1 2), swap(3 4)
215 }
216 return out;
217 }
218
219 @Test
220 @IR(applyIfCPUFeature = {"sse2", "true"},
221 applyIfPlatform = {"64-bit", "true"},
222 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
223 @IR(applyIfCPUFeature = {"asimd", "true"},
224 applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
225 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
226 @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
227 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
228 public static int[] testg(int[] out) {
229 for (int i = 0; i < ITER-2; i+=2) {
230 // Unrolled, with some swaps.
231 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
232 out[i+1] += ((sArr1[2*i+3] * sArr2[2*i+3]) + (sArr1[2*i+2] * sArr2[2*i+2])); // swap(1 3), swap(2 4)
233 }
234 return out;
235 }
236
237 @Test
238 @IR(applyIfCPUFeature = {"sse2", "true"},
239 applyIfPlatform = {"64-bit", "true"},
240 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
241 @IR(applyIfCPUFeature = {"asimd", "true"},
242 applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
243 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
244 @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
245 counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
246 public static int[] testh(int[] out) {
247 for (int i = 0; i < ITER-2; i+=2) {
248 // Unrolled, with some swaps.
249 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
250 out[i+1] += ((sArr2[2*i+3] * sArr1[2*i+3]) + (sArr2[2*i+2] * sArr1[2*i+2])); // swap(1 4), swap(2 3)
251 }
252 return out;
253 }
254
255 @Test
256 @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
257 applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
258 @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
259 public static int[] testi(int[] out) {
260 for (int i = 0; i < ITER-2; i+=2) {
261 // Unrolled, with some swaps that prevent vectorization.
262 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1])); // ok
263 out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+3]) + (sArr1[2*i+3] * sArr2[2*i+2])); // bad
264 }
265 return out;
266 }
267
268 @Test
269 @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
270 applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
271 @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
272 public static int[] testj(int[] out) {
273 for (int i = 0; i < ITER-2; i+=2) {
274 // Unrolled, with some swaps that prevent vectorization.
275 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+1]) + (sArr1[2*i+1] * sArr2[2*i+0])); // bad
276 out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+3]) + (sArr1[2*i+3] * sArr2[2*i+2])); // bad
277 }
278 return out;
279 }
280
281 @Test
282 @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
283 applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
284 @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
285 public static int[] testk(int[] out) {
286 for (int i = 0; i < ITER-2; i+=2) {
287 // Unrolled, with some swaps that prevent vectorization.
288 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+1]) + (sArr1[2*i+1] * sArr2[2*i+0])); // bad
289 out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+2]) + (sArr1[2*i+3] * sArr2[2*i+3])); // ok
290 }
291 return out;
292 }
293
294 @Test
295 @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
296 applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
297 @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
298 public static int[] testl(int[] out) {
299 for (int i = 0; i < ITER-2; i+=2) {
300 // Unrolled, with some swaps that prevent vectorization.
301 out[i+0] += ((sArr1[2*i+1] * sArr2[2*i+1]) + (sArr1[2*i+0] * sArr2[2*i+0])); // ok
302 out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+3]) + (sArr1[2*i+3] * sArr2[2*i+2])); // bad
303 }
304 return out;
305 }
306
307 @Test
308 @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
309 applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
310 @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
311 public static int[] testm(int[] out) {
312 for (int i = 0; i < ITER-4; i+=4) {
313 // Unrolled, with some swaps that prevent vectorization.
314 out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+1]) + (sArr1[2*i+1] * sArr2[2*i+0])); // bad
315 out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+2]) + (sArr1[2*i+3] * sArr2[2*i+3])); // ok
316 // 2-element gap
317 }
318 return out;
319 }
320 }