1 /* 2 * Copyright (c) 2015, 2017, 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 8073583 27 * @summary C2 support for CRC32C on SPARC 28 * 29 * @run main/othervm/timeout=600 -Xbatch compiler.intrinsics.zip.TestCRC32C -m 30 */ 31 32 package compiler.intrinsics.zip; 33 34 import java.nio.ByteBuffer; 35 import java.util.zip.CRC32C; 36 import java.util.zip.Checksum; 37 38 public class TestCRC32C { 39 // CRC32C (Castagnoli) polynomial 40 // coefficients in different forms 41 // normal: polyBits = 0x1edc6f41 = 0b0001 1110 1101 1100 0110 1111 0100 0001 42 // reversed: polybits = 0x82f63b78 = 0b1000 0010 1111 0110 0011 1011 0111 1000 43 // reversed reciprocal polybits = 0x8f6e37a0 = 0b1000 1111 0110 1110 0011 0111 1010 0000 44 // 45 // 0 5 9 13 17 21 25 29 46 // | | | | | | | | 47 // reversed shiftL 1 polyBits = 0x105ec76f1L = 0b1 0000 0101 1110 1100 0111 0110 1111 0001 48 final static long polyBits = (1L<<(32-32)) + (1L<<(32-28)) + (1L<<(32-27)) 49 + (1L<<(32-26)) + (1L<<(32-25)) + (1L<<(32-23)) + (1L<<(32-22)) 50 + (1L<<(32-20)) + (1L<<(32-19)) + (1L<<(32-18)) + (1L<<(32-14)) 51 + (1L<<(32-13)) + (1L<<(32-11)) + (1L<<(32-10)) + (1L<<(32-9)) 52 + (1L<<(32-8)) + (1L<<(32-6)) + (1L<<(32-0)); 53 final static long polyBitsShifted = polyBits>>1; 54 main(String[] args)55 public static void main(String[] args) throws Exception { 56 int offset = Integer.getInteger("offset", 0); 57 int msgSize = Integer.getInteger("msgSize", 512); 58 boolean multi = false; 59 int iters = 20000; 60 int warmupIters = 20000; 61 62 if (args.length > 0) { 63 if (args[0].equals("-m")) { 64 multi = true; 65 } else { 66 iters = Integer.valueOf(args[0]); 67 } 68 if (args.length > 1) { 69 warmupIters = Integer.valueOf(args[1]); 70 } 71 } 72 73 if (multi) { 74 test_multi(warmupIters); 75 return; 76 } 77 78 System.out.println(" offset = " + offset); 79 System.out.println("msgSize = " + msgSize + " bytes"); 80 System.out.println(" iters = " + iters); 81 82 byte[] b = initializedBytes(msgSize, offset); 83 84 final long crcReference = update_byteLoop(0, b, offset); 85 86 CRC32C crc0 = new CRC32C(); 87 CRC32C crc1 = new CRC32C(); 88 CRC32C crc2 = new CRC32C(); 89 90 crc0.update(b, offset, msgSize); 91 check(crc0, crcReference); 92 93 System.out.println("-------------------------------------------------------"); 94 95 /* warm up */ 96 for (int i = 0; i < warmupIters; i++) { 97 crc1.reset(); 98 crc1.update(b, offset, msgSize); 99 check(crc1, crcReference); 100 } 101 102 /* check correctness 103 * Do that before measuring performance 104 * to even better heat up involved methods. 105 */ 106 for (int i = 0; i < iters; i++) { 107 crc1.reset(); 108 crc1.update(b, offset, msgSize); 109 check(crc1, crcReference); 110 } 111 report("CRCs", crc1, crcReference); 112 113 /* measure performance 114 * Don't spoil times with error checking. 115 */ 116 long start = System.nanoTime(); 117 for (int i = 0; i < iters; i++) { 118 crc1.reset(); 119 crc1.update(b, offset, msgSize); 120 } 121 long end = System.nanoTime(); 122 123 double total = (double)(end - start)/1e9; // in seconds 124 double thruput = (double)msgSize*iters/1e6/total; // in MB/s 125 System.out.println("CRC32C.update(byte[]) runtime = " + total + " seconds"); 126 System.out.println("CRC32C.update(byte[]) throughput = " + thruput + " MB/s"); 127 report("CRCs", crc1, crcReference); 128 129 System.out.println("-------------------------------------------------------"); 130 131 ByteBuffer buf = ByteBuffer.allocateDirect(msgSize); 132 buf.put(b, offset, msgSize); 133 buf.flip(); 134 135 /* warm up */ 136 for (int i = 0; i < warmupIters; i++) { 137 crc2.reset(); 138 crc2.update(buf); 139 buf.rewind(); 140 check(crc2, crcReference); 141 } 142 143 /* check correctness 144 * Do that before measuring performance 145 * to even better heat up involved methods. 146 */ 147 for (int i = 0; i < iters; i++) { 148 crc2.reset(); 149 crc2.update(buf); 150 buf.rewind(); 151 check(crc2, crcReference); 152 } 153 report("CRCs", crc2, crcReference); 154 155 /* measure performance 156 * Don't spoil times with error checking. 157 */ 158 start = System.nanoTime(); 159 for (int i = 0; i < iters; i++) { 160 crc2.reset(); 161 crc2.update(buf); 162 buf.rewind(); 163 } 164 end = System.nanoTime(); 165 total = (double)(end - start)/1e9; // in seconds 166 thruput = (double)msgSize*iters/1e6/total; // in MB/s 167 System.out.println("CRC32C.update(ByteBuffer) runtime = " + total + " seconds"); 168 System.out.println("CRC32C.update(ByteBuffer) throughput = " + thruput + " MB/s"); 169 report("CRCs", crc2, crcReference); 170 171 System.out.println("-------------------------------------------------------"); 172 } 173 174 // Just a loop over a byte array, updating the CRC byte by byte. update_byteLoop(long crc, byte[] buf, int offset)175 public static long update_byteLoop(long crc, byte[] buf, int offset) { 176 return update_byteLoop(crc, buf, offset, buf.length-offset); 177 } 178 179 // Just a loop over a byte array, with given length, updating the CRC byte by byte. update_byteLoop(long crc, byte[] buf, int offset, int length)180 public static long update_byteLoop(long crc, byte[] buf, int offset, int length) { 181 int end = length+offset; 182 for (int i = offset; i < end; i++) { 183 crc = update_singlebyte(crc, polyBitsShifted, buf[i]); 184 } 185 return crc; 186 } 187 188 // Straight-forward implementation of CRC update by one byte. 189 // We use this very basic implementation to calculate reference 190 // results. It is necessary to have full control over how the 191 // reference results are calculated. It is not sufficient to rely 192 // on the interpreter (or c1, or c2) to do the right thing. update_singlebyte(long crc, long polynomial, int val)193 public static long update_singlebyte(long crc, long polynomial, int val) { 194 crc = (crc ^ -1L) & 0x00000000ffffffffL; // use 1's complement of crc 195 crc = crc ^ (val&0xff); // XOR in next byte from stream 196 for (int i = 0; i < 8; i++) { 197 boolean bitset = (crc & 0x01L) != 0; 198 199 crc = crc>>1; 200 if (bitset) { 201 crc = crc ^ polynomial; 202 crc = crc & 0x00000000ffffffffL; 203 } 204 } 205 crc = (crc ^ -1L) & 0x00000000ffffffffL; // revert taking 1's complement 206 return crc; 207 } 208 report(String s, Checksum crc, long crcReference)209 private static void report(String s, Checksum crc, long crcReference) { 210 System.out.printf("%s: crc = %08x, crcReference = %08x\n", 211 s, crc.getValue(), crcReference); 212 } 213 check(Checksum crc, long crcReference)214 private static void check(Checksum crc, long crcReference) throws Exception { 215 if (crc.getValue() != crcReference) { 216 System.err.printf("ERROR: crc = %08x, crcReference = %08x\n", 217 crc.getValue(), crcReference); 218 throw new Exception("TestCRC32C Error"); 219 } 220 } 221 initializedBytes(int M, int offset)222 private static byte[] initializedBytes(int M, int offset) { 223 byte[] bytes = new byte[M + offset]; 224 for (int i = 0; i < offset; i++) { 225 bytes[i] = (byte) i; 226 } 227 for (int i = offset; i < bytes.length; i++) { 228 bytes[i] = (byte) (i - offset); 229 } 230 return bytes; 231 } 232 test_multi(int iters)233 private static void test_multi(int iters) throws Exception { 234 int len1 = 8; // the 8B/iteration loop 235 int len2 = 32; // the 32B/iteration loop 236 int len3 = 4096; // the 4KB/iteration loop 237 238 byte[] b = initializedBytes(len3*16, 0); 239 int[] offsets = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 16, 32, 64, 128, 256, 512 }; 240 int[] sizes = { 0, 1, 2, 3, 4, 5, 6, 7, 241 len1, len1+1, len1+2, len1+3, len1+4, len1+5, len1+6, len1+7, 242 len1*2, len1*2+1, len1*2+3, len1*2+5, len1*2+7, 243 len2, len2+1, len2+3, len2+5, len2+7, 244 len2*2, len2*4, len2*8, len2*16, len2*32, len2*64, 245 len3, len3+1, len3+3, len3+5, len3+7, 246 len3*2, len3*4, len3*8, 247 len1+len2, len1+len2+1, len1+len2+3, len1+len2+5, len1+len2+7, 248 len1+len3, len1+len3+1, len1+len3+3, len1+len3+5, len1+len3+7, 249 len2+len3, len2+len3+1, len2+len3+3, len2+len3+5, len2+len3+7, 250 len1+len2+len3, len1+len2+len3+1, len1+len2+len3+3, 251 len1+len2+len3+5, len1+len2+len3+7, 252 (len1+len2+len3)*2, (len1+len2+len3)*2+1, (len1+len2+len3)*2+3, 253 (len1+len2+len3)*2+5, (len1+len2+len3)*2+7, 254 (len1+len2+len3)*3, (len1+len2+len3)*3-1, (len1+len2+len3)*3-3, 255 (len1+len2+len3)*3-5, (len1+len2+len3)*3-7 }; 256 CRC32C[] crc1 = new CRC32C[offsets.length*sizes.length]; 257 long[] crcReference = new long[offsets.length*sizes.length]; 258 int i, j, k; 259 260 System.out.printf("testing %d cases ...\n", offsets.length*sizes.length); 261 262 // Initialize CRC32C result arrays, CRC32C reference array. 263 // Reference is calculated using a very basic Java implementation. 264 for (i = 0; i < offsets.length; i++) { 265 for (j = 0; j < sizes.length; j++) { 266 crc1[i*sizes.length + j] = new CRC32C(); 267 crcReference[i*sizes.length + j] = update_byteLoop(0, b, offsets[i], sizes[j]); 268 } 269 } 270 271 // Warm up the JIT compiler. Over time, all methods involved will 272 // be executed by the interpreter, then get compiled by c1 and 273 // finally by c2. Each calculated CRC value must, in each iteration, 274 // be equal to the precalculated reference value for the test to pass. 275 for (k = 0; k < iters; k++) { 276 for (i = 0; i < offsets.length; i++) { 277 for (j = 0; j < sizes.length; j++) { 278 crc1[i*sizes.length + j].reset(); 279 crc1[i*sizes.length + j].update(b, offsets[i], sizes[j]); 280 check(crc1[i*sizes.length + j], crcReference[i*sizes.length + j]); 281 } 282 } 283 } 284 } 285 } 286