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 
 57     static {
 58         for (int i = 0; i < RANGE; i++) {
 59             sArr1[i] = (short)(AbstractInfo.getRandom().nextInt());
 60             sArr2[i] = (short)(AbstractInfo.getRandom().nextInt());
 61         }
 62         GOLDEN_A = testa();
 63         GOLDEN_B = testb();
 64         GOLDEN_C = testc(new int[ITER]);
 65         GOLDEN_D = testd(new int[ITER]);
 66         GOLDEN_E = teste(new int[ITER]);
 67         GOLDEN_F = testf(new int[ITER]);
 68         GOLDEN_G = testg(new int[ITER]);
 69         GOLDEN_H = testh(new int[ITER]);
 70         GOLDEN_I = testi(new int[ITER]);
 71         GOLDEN_J = testj(new int[ITER]);
 72         GOLDEN_K = testk(new int[ITER]);
 73         GOLDEN_L = testl(new int[ITER]);
 74     }
 75 
 76 
 77     public static void main(String[] args) {
 78         TestFramework.runWithFlags("-XX:+IgnoreUnrecognizedVMOptions", "-XX:+AlignVector");
 79         TestFramework.runWithFlags("-XX:+IgnoreUnrecognizedVMOptions", "-XX:-AlignVector");
 80     }
 81 
 82     @Run(test = {"testa", "testb", "testc", "testd", "teste", "testf", "testg", "testh",
 83                  "testi", "testj", "testk", "testl"})
 84     @Warmup(0)
 85     public static void run() {
 86         compare(testa(), GOLDEN_A, "testa");
 87         compare(testb(), GOLDEN_B, "testb");
 88         compare(testc(new int[ITER]), GOLDEN_C, "testc");
 89         compare(testd(new int[ITER]), GOLDEN_D, "testd");
 90         compare(teste(new int[ITER]), GOLDEN_E, "teste");
 91         compare(testf(new int[ITER]), GOLDEN_F, "testf");
 92         compare(testg(new int[ITER]), GOLDEN_G, "testg");
 93         compare(testh(new int[ITER]), GOLDEN_H, "testh");
 94         compare(testi(new int[ITER]), GOLDEN_I, "testi");
 95         compare(testj(new int[ITER]), GOLDEN_J, "testj");
 96         compare(testk(new int[ITER]), GOLDEN_K, "testk");
 97         compare(testl(new int[ITER]), GOLDEN_L, "testl");
 98     }
 99 
100     public static void compare(int[] out, int[] golden, String name) {
101         for (int i = 0; i < ITER; i++) {
102             Asserts.assertEQ(out[i], golden[i], "wrong result for '" + name + "' out[" + i + "]");
103         }
104     }
105 
106     @Test
107     @IR(applyIfCPUFeature = {"sse2", "true"},
108         applyIfPlatform = {"64-bit", "true"},
109         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
110     @IR(applyIfCPUFeature = {"asimd", "true"},
111         applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
112         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
113     @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
114         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
115     public static int[] testa() {
116         int[] out = new int[ITER];
117         int[] out2 = new int[ITER];
118         for (int i = 0; i < ITER; i++) {
119             out[i] += ((sArr1[2*i] * sArr1[2*i]) + (sArr1[2*i+1] * sArr1[2*i+1]));
120             out2[i] += out[i];
121         }
122         return out;
123     }
124 
125     @Test
126     @IR(applyIfCPUFeature = {"sse2", "true"},
127         applyIfPlatform = {"64-bit", "true"},
128         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
129     @IR(applyIfCPUFeature = {"asimd", "true"},
130         applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
131         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
132     @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
133         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
134     public static int[] testb() {
135         int[] out = new int[ITER];
136         int[] out2 = new int[ITER];
137         for (int i = 0; i < ITER; i++) {
138             out[i] += ((sArr1[2*i] * sArr2[2*i]) + (sArr1[2*i+1] * sArr2[2*i+1]));
139             out2[i] += out[i];
140         }
141         return out;
142     }
143 
144     @Test
145     @IR(applyIfCPUFeature = {"sse2", "true"},
146         applyIfPlatform = {"64-bit", "true"},
147         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
148     @IR(applyIfCPUFeature = {"asimd", "true"},
149         applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
150         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
151     @IR(applyIfCPUFeature = {"avx512_vnni", "true"},
152         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
153     public static int[] testc(int[] out) {
154         for (int i = 0; i < ITER; i++) {
155             out[i] += ((sArr1[2*i] * sArr2[2*i]) + (sArr1[2*i+1] * sArr2[2*i+1]));
156         }
157         return out;
158     }
159 
160     @Test
161     @IR(applyIfCPUFeature = {"sse2", "true"},
162         applyIfPlatform = {"64-bit", "true"},

163         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
164     @IR(applyIfCPUFeature = {"asimd", "true"},
165         applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
166         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
167     @IR(applyIfCPUFeature = {"avx512_vnni", "true"},

168         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
169     public static int[] testd(int[] out) {
170         for (int i = 0; i < ITER-2; i+=2) {
171             // Unrolled, with the same structure.
172             out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
173             out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+2]) + (sArr1[2*i+3] * sArr2[2*i+3]));
174         }
175         return out;
176     }
177 
178     @Test
179     @IR(applyIfCPUFeature = {"sse2", "true"},
180         applyIfPlatform = {"64-bit", "true"},

181         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
182     @IR(applyIfCPUFeature = {"asimd", "true"},
183         applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
184         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
185     @IR(applyIfCPUFeature = {"avx512_vnni", "true"},

186         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
187     public static int[] teste(int[] out) {
188         for (int i = 0; i < ITER-2; i+=2) {
189             // Unrolled, with some swaps.
190             out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
191             out[i+1] += ((sArr2[2*i+2] * sArr1[2*i+2]) + (sArr1[2*i+3] * sArr2[2*i+3])); // swap(1 2)
192         }
193         return out;
194     }
195 
196     @Test
197     @IR(applyIfCPUFeature = {"sse2", "true"},
198         applyIfPlatform = {"64-bit", "true"},

199         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
200     @IR(applyIfCPUFeature = {"asimd", "true"},
201         applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
202         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
203     @IR(applyIfCPUFeature = {"avx512_vnni", "true"},

204         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
205     public static int[] testf(int[] out) {
206         for (int i = 0; i < ITER-2; i+=2) {
207             // Unrolled, with some swaps.
208             out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
209             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)
210         }
211         return out;
212     }
213 
214     @Test
215     @IR(applyIfCPUFeature = {"sse2", "true"},
216         applyIfPlatform = {"64-bit", "true"},

217         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
218     @IR(applyIfCPUFeature = {"asimd", "true"},
219         applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
220         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
221     @IR(applyIfCPUFeature = {"avx512_vnni", "true"},

222         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
223     public static int[] testg(int[] out) {
224         for (int i = 0; i < ITER-2; i+=2) {
225             // Unrolled, with some swaps.
226             out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
227             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)
228         }
229         return out;
230     }
231 
232     @Test
233     @IR(applyIfCPUFeature = {"sse2", "true"},
234         applyIfPlatform = {"64-bit", "true"},

235         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
236     @IR(applyIfCPUFeature = {"asimd", "true"},
237         applyIf = {"MaxVectorSize", "16"}, // AD file requires vector_length = 16
238         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI, "> 0"})
239     @IR(applyIfCPUFeature = {"avx512_vnni", "true"},

240         counts = {IRNode.MUL_ADD_S2I, "> 0", IRNode.MUL_ADD_VS2VI_VNNI, "> 0"})
241     public static int[] testh(int[] out) {
242         for (int i = 0; i < ITER-2; i+=2) {
243             // Unrolled, with some swaps.
244             out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1]));
245             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)
246         }
247         return out;
248     }
249 
250     @Test
251     @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
252         applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
253     @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
254     public static int[] testi(int[] out) {
255         for (int i = 0; i < ITER-2; i+=2) {
256             // Unrolled, with some swaps that prevent vectorization.
257             out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+0]) + (sArr1[2*i+1] * sArr2[2*i+1])); // ok
258             out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+3]) + (sArr1[2*i+3] * sArr2[2*i+2])); // bad
259         }
260         return out;
261     }
262 
263     @Test
264     @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
265         applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
266     @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
267     public static int[] testj(int[] out) {
268         for (int i = 0; i < ITER-2; i+=2) {
269             // Unrolled, with some swaps that prevent vectorization.
270             out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+1]) + (sArr1[2*i+1] * sArr2[2*i+0])); // bad
271             out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+3]) + (sArr1[2*i+3] * sArr2[2*i+2])); // bad
272         }
273         return out;
274     }
275 
276     @Test
277     @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
278         applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
279     @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
280     public static int[] testk(int[] out) {
281         for (int i = 0; i < ITER-2; i+=2) {
282             // Unrolled, with some swaps that prevent vectorization.
283             out[i+0] += ((sArr1[2*i+0] * sArr2[2*i+1]) + (sArr1[2*i+1] * sArr2[2*i+0])); // bad
284             out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+2]) + (sArr1[2*i+3] * sArr2[2*i+3])); // ok
285         }
286         return out;
287     }
288 
289     @Test
290     @IR(counts = {IRNode.MUL_ADD_S2I, "> 0"},
291         applyIfCPUFeatureOr = {"sse2", "true", "asimd", "true"})
292     @IR(counts = {IRNode.MUL_ADD_VS2VI, "= 0"})
293     public static int[] testl(int[] out) {
294         for (int i = 0; i < ITER-2; i+=2) {
295             // Unrolled, with some swaps that prevent vectorization.
296             out[i+0] += ((sArr1[2*i+1] * sArr2[2*i+1]) + (sArr1[2*i+0] * sArr2[2*i+0])); // ok
297             out[i+1] += ((sArr1[2*i+2] * sArr2[2*i+3]) + (sArr1[2*i+3] * sArr2[2*i+2])); // bad
298         }
299         return out;
300     }
301 
302 }
--- EOF ---