AArch64ArrayEqualsOp.java revision 12968:4d8a004e5c6d
1/* 2 * Copyright (c) 2013, 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 */ 23package org.graalvm.compiler.lir.aarch64; 24 25import static jdk.vm.ci.aarch64.AArch64.zr; 26import static jdk.vm.ci.code.ValueUtil.asRegister; 27import static org.graalvm.compiler.lir.LIRInstruction.OperandFlag.REG; 28 29import java.lang.reflect.Array; 30import java.lang.reflect.Field; 31 32import org.graalvm.compiler.asm.Label; 33import org.graalvm.compiler.asm.aarch64.AArch64Address; 34import org.graalvm.compiler.asm.aarch64.AArch64Assembler.ConditionFlag; 35import org.graalvm.compiler.asm.aarch64.AArch64MacroAssembler; 36import org.graalvm.compiler.asm.aarch64.AArch64MacroAssembler.ScratchRegister; 37import org.graalvm.compiler.core.common.LIRKind; 38import org.graalvm.compiler.lir.LIRInstructionClass; 39import org.graalvm.compiler.lir.Opcode; 40import org.graalvm.compiler.lir.asm.CompilationResultBuilder; 41import org.graalvm.compiler.lir.gen.LIRGeneratorTool; 42 43import jdk.vm.ci.code.Register; 44import jdk.vm.ci.meta.JavaKind; 45import jdk.vm.ci.meta.Value; 46import sun.misc.Unsafe; 47 48/** 49 * Emits code which compares two arrays of the same length. If the CPU supports any vector 50 * instructions specialized code is emitted to leverage these instructions. 51 */ 52@Opcode("ARRAY_EQUALS") 53public final class AArch64ArrayEqualsOp extends AArch64LIRInstruction { 54 public static final LIRInstructionClass<AArch64ArrayEqualsOp> TYPE = LIRInstructionClass.create(AArch64ArrayEqualsOp.class); 55 56 private final JavaKind kind; 57 private final int arrayBaseOffset; 58 private final int arrayIndexScale; 59 60 @Def({REG}) protected Value resultValue; 61 @Alive({REG}) protected Value array1Value; 62 @Alive({REG}) protected Value array2Value; 63 @Alive({REG}) protected Value lengthValue; 64 @Temp({REG}) protected Value temp1; 65 @Temp({REG}) protected Value temp2; 66 @Temp({REG}) protected Value temp3; 67 @Temp({REG}) protected Value temp4; 68 69 public AArch64ArrayEqualsOp(LIRGeneratorTool tool, JavaKind kind, Value result, Value array1, Value array2, Value length) { 70 super(TYPE); 71 this.kind = kind; 72 73 Class<?> arrayClass = Array.newInstance(kind.toJavaClass(), 0).getClass(); 74 this.arrayBaseOffset = UNSAFE.arrayBaseOffset(arrayClass); 75 this.arrayIndexScale = UNSAFE.arrayIndexScale(arrayClass); 76 77 this.resultValue = result; 78 this.array1Value = array1; 79 this.array2Value = array2; 80 this.lengthValue = length; 81 82 // Allocate some temporaries. 83 this.temp1 = tool.newVariable(LIRKind.unknownReference(tool.target().arch.getWordKind())); 84 this.temp2 = tool.newVariable(LIRKind.unknownReference(tool.target().arch.getWordKind())); 85 this.temp3 = tool.newVariable(LIRKind.value(tool.target().arch.getWordKind())); 86 this.temp4 = tool.newVariable(LIRKind.value(tool.target().arch.getWordKind())); 87 } 88 89 @Override 90 public void emitCode(CompilationResultBuilder crb, AArch64MacroAssembler masm) { 91 Register result = asRegister(resultValue); 92 Register array1 = asRegister(temp1); 93 Register array2 = asRegister(temp2); 94 Register length = asRegister(temp3); 95 96 Label breakLabel = new Label(); 97 98 try (ScratchRegister sc1 = masm.getScratchRegister()) { 99 Register rscratch1 = sc1.getRegister(); 100 // Load array base addresses. 101 masm.lea(array1, AArch64Address.createUnscaledImmediateAddress(asRegister(array1Value), arrayBaseOffset)); 102 masm.lea(array2, AArch64Address.createUnscaledImmediateAddress(asRegister(array2Value), arrayBaseOffset)); 103 104 // Get array length in bytes. 105 masm.mov(rscratch1, arrayIndexScale); 106 masm.smaddl(length, asRegister(lengthValue), rscratch1, zr); 107 masm.mov(64, result, length); // copy 108 109 emit8ByteCompare(crb, masm, result, array1, array2, length, breakLabel, rscratch1); 110 emitTailCompares(masm, result, array1, array2, breakLabel, rscratch1); 111 112 // Return: rscratch1 is non-zero iff the arrays differ 113 masm.bind(breakLabel); 114 masm.cmp(64, rscratch1, zr); 115 masm.cset(result, ConditionFlag.EQ); 116 } 117 } 118 119 /** 120 * Vector size used in {@link #emit8ByteCompare}. 121 */ 122 private static final int VECTOR_SIZE = 8; 123 124 /** 125 * Emits code that uses 8-byte vector compares. 126 * 127 */ 128 private void emit8ByteCompare(CompilationResultBuilder crb, AArch64MacroAssembler masm, Register result, Register array1, Register array2, Register length, Label breakLabel, 129 Register rscratch1) { 130 Label loop = new Label(); 131 Label compareTail = new Label(); 132 133 Register temp = asRegister(temp4); 134 135 masm.and(64, result, result, VECTOR_SIZE - 1); // tail count (in bytes) 136 masm.ands(64, length, length, ~(VECTOR_SIZE - 1)); // vector count (in bytes) 137 masm.branchConditionally(ConditionFlag.EQ, compareTail); 138 139 masm.lea(array1, AArch64Address.createRegisterOffsetAddress(array1, length, false)); 140 masm.lea(array2, AArch64Address.createRegisterOffsetAddress(array2, length, false)); 141 masm.sub(64, length, zr, length); 142 143 // Align the main loop 144 masm.align(crb.target.wordSize * 2); 145 masm.bind(loop); 146 masm.ldr(64, temp, AArch64Address.createRegisterOffsetAddress(array1, length, false)); 147 masm.ldr(64, rscratch1, AArch64Address.createRegisterOffsetAddress(array2, length, false)); 148 masm.eor(64, rscratch1, temp, rscratch1); 149 masm.cbnz(64, rscratch1, breakLabel); 150 masm.add(64, length, length, VECTOR_SIZE); 151 masm.cbnz(64, length, loop); 152 153 masm.cbz(64, result, breakLabel); 154 155 /* 156 * Compare the remaining bytes with an unaligned memory load aligned to the end of the 157 * array. 158 */ 159 masm.lea(array1, AArch64Address.createUnscaledImmediateAddress(array1, -VECTOR_SIZE)); 160 masm.lea(array2, AArch64Address.createUnscaledImmediateAddress(array2, -VECTOR_SIZE)); 161 masm.ldr(64, temp, AArch64Address.createRegisterOffsetAddress(array1, result, false)); 162 masm.ldr(64, rscratch1, AArch64Address.createRegisterOffsetAddress(array2, result, false)); 163 masm.eor(64, rscratch1, temp, rscratch1); 164 masm.jmp(breakLabel); 165 166 masm.bind(compareTail); 167 } 168 169 /** 170 * Emits code to compare the remaining 1 to 4 bytes. 171 * 172 */ 173 private void emitTailCompares(AArch64MacroAssembler masm, Register result, Register array1, Register array2, Label breakLabel, Register rscratch1) { 174 Label compare2Bytes = new Label(); 175 Label compare1Byte = new Label(); 176 Label end = new Label(); 177 178 Register temp = asRegister(temp4); 179 180 if (kind.getByteCount() <= 4) { 181 // Compare trailing 4 bytes, if any. 182 masm.ands(32, zr, result, 4); 183 masm.branchConditionally(ConditionFlag.EQ, compare2Bytes); 184 masm.ldr(32, temp, AArch64Address.createPostIndexedImmediateAddress(array1, 4)); 185 masm.ldr(32, rscratch1, AArch64Address.createPostIndexedImmediateAddress(array2, 4)); 186 masm.eor(32, rscratch1, temp, rscratch1); 187 masm.cbnz(32, rscratch1, breakLabel); 188 189 if (kind.getByteCount() <= 2) { 190 // Compare trailing 2 bytes, if any. 191 masm.bind(compare2Bytes); 192 masm.ands(32, zr, result, 2); 193 masm.branchConditionally(ConditionFlag.EQ, compare1Byte); 194 masm.ldr(16, temp, AArch64Address.createPostIndexedImmediateAddress(array1, 2)); 195 masm.ldr(16, rscratch1, AArch64Address.createPostIndexedImmediateAddress(array2, 2)); 196 masm.eor(32, rscratch1, temp, rscratch1); 197 masm.cbnz(32, rscratch1, breakLabel); 198 199 // The one-byte tail compare is only required for boolean and byte arrays. 200 if (kind.getByteCount() <= 1) { 201 // Compare trailing byte, if any. 202 masm.bind(compare1Byte); 203 masm.ands(32, zr, result, 1); 204 masm.branchConditionally(ConditionFlag.EQ, end); 205 masm.ldr(8, temp, AArch64Address.createBaseRegisterOnlyAddress(array1)); 206 masm.ldr(8, rscratch1, AArch64Address.createBaseRegisterOnlyAddress(array2)); 207 masm.eor(32, rscratch1, temp, rscratch1); 208 masm.cbnz(32, rscratch1, breakLabel); 209 } else { 210 masm.bind(compare1Byte); 211 } 212 } else { 213 masm.bind(compare2Bytes); 214 } 215 } 216 masm.bind(end); 217 masm.mov(64, rscratch1, zr); 218 } 219 220 private static final Unsafe UNSAFE = initUnsafe(); 221 222 private static Unsafe initUnsafe() { 223 try { 224 return Unsafe.getUnsafe(); 225 } catch (SecurityException se) { 226 try { 227 Field theUnsafe = Unsafe.class.getDeclaredField("theUnsafe"); 228 theUnsafe.setAccessible(true); 229 return (Unsafe) theUnsafe.get(Unsafe.class); 230 } catch (Exception e) { 231 throw new RuntimeException("exception while trying to get Unsafe", e); 232 } 233 } 234 } 235} 236