1 /* 2 * Copyright (c) 2015, 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 8143012 27 * @summary CRC32 Intrinsics support on SPARC 28 * 29 * @run main/othervm/timeout=720 -Xbatch compiler.intrinsics.zip.TestCRC32 -m 30 */ 31 32 package compiler.intrinsics.zip; 33 34 import java.nio.ByteBuffer; 35 import java.util.zip.CRC32; 36 import java.util.zip.Checksum; 37 38 public class TestCRC32 { 39 // standard CRC32 polynomial 40 // coefficients in different forms 41 // normal: polyBits = 0x04c11db7 = 0b0000 0100 1100 0001 0001 1101 1011 0111 42 // reversed: polybits = 0xedb88320 = 0b1110 1101 1011 1000 1000 0011 0010 0000 43 // reversed reciprocal polybits = 0x82608edb = 0b1000 0010 0110 0000 1000 1110 1101 1011 44 // 45 // 0 5 9 13 17 21 25 29 46 // | | | | | | | | 47 // reversed shiftL 1 polyBits = 0x1db710641L = 0b1 1101 1011 0111 0001 0000 0110 0100 0001 48 final static long polyBits = (1L<<(32-32)) + (1L<<(32-26)) + (1L<<(32-23)) + (1L<<(32-22)) 49 + (1L<<(32-16)) + (1L<<(32-12)) + (1L<<(32-11)) + (1L<<(32-10)) 50 + (1L<<(32-8)) + (1L<<(32-7)) + (1L<<(32-5)) + (1L<<(32-4)) 51 + (1L<<(32-2)) + (1L<<(32-1)) + (1L<<(32-0)); 52 final static long polyBitsShifted = polyBits>>1; 53 54 public static void main(String[] args) throws Exception { 55 int offset = Integer.getInteger("offset", 0); 56 int msgSize = Integer.getInteger("msgSize", 512); 57 boolean multi = false; 58 int iters = 20000; 59 int warmupIters = 20000; 60 61 if (args.length > 0) { 62 if (args[0].equals("-m")) { 63 multi = true; 64 } else { 65 iters = Integer.valueOf(args[0]); 66 } 67 if (args.length > 1) { 68 warmupIters = Integer.valueOf(args[1]); 69 } 70 } 71 72 if (multi) { 73 test_multi(warmupIters); 74 return; 75 } 76 77 System.out.println(" offset = " + offset); 78 System.out.println("msgSize = " + msgSize + " bytes"); 79 System.out.println(" iters = " + iters); 80 81 byte[] b = initializedBytes(msgSize, offset); 82 83 final long crcReference = update_byteLoop(0, b, offset); 84 85 CRC32 crc0 = new CRC32(); 86 CRC32 crc1 = new CRC32(); 87 CRC32 crc2 = new CRC32(); 88 89 crc0.update(b, offset, msgSize); 90 check(crc0, crcReference); 91 92 System.out.println("-------------------------------------------------------"); 93 94 /* warm up */ 95 for (int i = 0; i < warmupIters; i++) { 96 crc1.reset(); 97 crc1.update(b, offset, msgSize); 98 check(crc1, crcReference); 99 } 100 101 /* check correctness 102 * Do that before measuring performance 103 * to even better heat up involved methods. 104 */ 105 for (int i = 0; i < iters; i++) { 106 crc1.reset(); 107 crc1.update(b, offset, msgSize); 108 check(crc1, crcReference); 109 } 110 report("CRCs", crc1, crcReference); 111 112 /* measure performance 113 * Don't spoil times with error checking. 114 */ 115 long start = System.nanoTime(); 116 for (int i = 0; i < iters; i++) { 117 crc1.reset(); 118 crc1.update(b, offset, msgSize); 119 } 120 long end = System.nanoTime(); 121 122 double total = (double)(end - start)/1e9; // in seconds 123 double thruput = (double)msgSize*iters/1e6/total; // in MB/s 124 System.out.println("CRC32.update(byte[]) runtime = " + total + " seconds"); 125 System.out.println("CRC32.update(byte[]) throughput = " + thruput + " MB/s"); 126 report("CRCs", crc1, crcReference); 127 128 System.out.println("-------------------------------------------------------"); 129 130 ByteBuffer buf = ByteBuffer.allocateDirect(msgSize); 131 buf.put(b, offset, msgSize); 132 buf.flip(); 133 134 /* warm up */ 135 for (int i = 0; i < warmupIters; i++) { 136 crc2.reset(); 137 crc2.update(buf); 138 buf.rewind(); 139 check(crc2, crcReference); 140 } 141 142 /* check correctness 143 * Do that before measuring performance 144 * to even better heat up involved methods. 145 */ 146 for (int i = 0; i < iters; i++) { 147 crc2.reset(); 148 crc2.update(buf); 149 buf.rewind(); 150 check(crc2, crcReference); 151 } 152 report("CRCs", crc2, crcReference); 153 154 /* measure performance 155 * Don't spoil times with error checking. 156 */ 157 start = System.nanoTime(); 158 for (int i = 0; i < iters; i++) { 159 crc2.reset(); 160 crc2.update(buf); 161 buf.rewind(); 162 } 163 end = System.nanoTime(); 164 total = (double)(end - start)/1e9; // in seconds 165 thruput = (double)msgSize*iters/1e6/total; // in MB/s 166 System.out.println("CRC32.update(ByteBuffer) runtime = " + total + " seconds"); 167 System.out.println("CRC32.update(ByteBuffer) throughput = " + thruput + " MB/s"); 168 report("CRCs", crc2, crcReference); 169 170 System.out.println("-------------------------------------------------------"); 171 } 172 173 // Just a loop over a byte array, updating the CRC byte by byte. 174 public static long update_byteLoop(long crc, byte[] buf, int offset) { 175 return update_byteLoop(crc, buf, offset, buf.length-offset); 176 } 177 178 // Just a loop over a byte array, with given length, updating the CRC byte by byte. 179 public static long update_byteLoop(long crc, byte[] buf, int offset, int length) { 180 int end = length+offset; 181 for (int i = offset; i < end; i++) { 182 crc = update_singlebyte(crc, polyBitsShifted, buf[i]); 183 } 184 return crc; 185 } 186 187 // Straight-forward implementation of CRC update by one byte. 188 // We use this very basic implementation to calculate reference 189 // results. It is necessary to have full control over how the 190 // reference results are calculated. It is not sufficient to rely 191 // on the interpreter (or c1, or c2) to do the right thing. 192 public static long update_singlebyte(long crc, long polynomial, int val) { 193 crc = (crc ^ -1L) & 0x00000000ffffffffL; // use 1's complement of crc 194 crc = crc ^ (val&0xff); // XOR in next byte from stream 195 for (int i = 0; i < 8; i++) { 196 boolean bitset = (crc & 0x01L) != 0; 197 198 crc = crc>>1; 199 if (bitset) { 200 crc = crc ^ polynomial; 201 crc = crc & 0x00000000ffffffffL; 202 } 203 } 204 crc = (crc ^ -1L) & 0x00000000ffffffffL; // revert taking 1's complement 205 return crc; 206 } 207 208 private static void report(String s, Checksum crc, long crcReference) { 209 System.out.printf("%s: crc = %08x, crcReference = %08x\n", 210 s, crc.getValue(), crcReference); 211 } 212 213 private static void check(Checksum crc, long crcReference) throws Exception { 214 if (crc.getValue() != crcReference) { 215 System.err.printf("ERROR: crc = %08x, crcReference = %08x\n", 216 crc.getValue(), crcReference); 217 throw new Exception("TestCRC32 Error"); 218 } 219 } 220 221 private static byte[] initializedBytes(int M, int offset) { 222 byte[] bytes = new byte[M + offset]; 223 for (int i = 0; i < offset; i++) { 224 bytes[i] = (byte) i; 225 } 226 for (int i = offset; i < bytes.length; i++) { 227 bytes[i] = (byte) (i - offset); 228 } 229 return bytes; 230 } 231 232 private static void test_multi(int iters) throws Exception { 233 int len1 = 8; // the 8B/iteration loop 234 int len2 = 32; // the 32B/iteration loop 235 int len3 = 4096; // the 4KB/iteration loop 236 237 byte[] b = initializedBytes(len3*16, 0); 238 int[] offsets = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 16, 32, 64, 128, 256, 512 }; 239 int[] sizes = { 0, 1, 2, 3, 4, 5, 6, 7, 240 len1, len1+1, len1+2, len1+3, len1+4, len1+5, len1+6, len1+7, 241 len1*2, len1*2+1, len1*2+3, len1*2+5, len1*2+7, 242 len2, len2+1, len2+3, len2+5, len2+7, 243 len2*2, len2*4, len2*8, len2*16, len2*32, len2*64, 244 len3, len3+1, len3+3, len3+5, len3+7, 245 len3*2, len3*4, len3*8, 246 len1+len2, len1+len2+1, len1+len2+3, len1+len2+5, len1+len2+7, 247 len1+len3, len1+len3+1, len1+len3+3, len1+len3+5, len1+len3+7, 248 len2+len3, len2+len3+1, len2+len3+3, len2+len3+5, len2+len3+7, 249 len1+len2+len3, len1+len2+len3+1, len1+len2+len3+3, 250 len1+len2+len3+5, len1+len2+len3+7, 251 (len1+len2+len3)*2, (len1+len2+len3)*2+1, (len1+len2+len3)*2+3, 252 (len1+len2+len3)*2+5, (len1+len2+len3)*2+7, 253 (len1+len2+len3)*3, (len1+len2+len3)*3-1, (len1+len2+len3)*3-3, 254 (len1+len2+len3)*3-5, (len1+len2+len3)*3-7 }; 255 CRC32[] crc1 = new CRC32[offsets.length*sizes.length]; 256 long[] crcReference = new long[offsets.length*sizes.length]; 257 int i, j, k; 258 259 System.out.printf("testing %d cases ...\n", offsets.length*sizes.length); 260 261 // Initialize CRC32 result arrays, CRC32 reference array. 262 // Reference is calculated using a very basic Java implementation. 263 for (i = 0; i < offsets.length; i++) { 264 for (j = 0; j < sizes.length; j++) { 265 crc1[i*sizes.length + j] = new CRC32(); 266 crcReference[i*sizes.length + j] = update_byteLoop(0, b, offsets[i], sizes[j]); 267 } 268 } 269 270 // Warm up the JIT compiler. Over time, all methods involved will 271 // be executed by the interpreter, then get compiled by c1 and 272 // finally by c2. Each calculated CRC value must, in each iteration, 273 // be equal to the precalculated reference value for the test to pass. 274 for (k = 0; k < iters; k++) { 275 for (i = 0; i < offsets.length; i++) { 276 for (j = 0; j < sizes.length; j++) { 277 crc1[i*sizes.length + j].reset(); 278 crc1[i*sizes.length + j].update(b, offsets[i], sizes[j]); 279 check(crc1[i*sizes.length + j], crcReference[i*sizes.length + j]); 280 } 281 } 282 } 283 } 284 }