/*
 * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
 */


package org.graalvm.compiler.lir.aarch64;

import static jdk.vm.ci.code.ValueUtil.asRegister;
import static org.graalvm.compiler.lir.LIRInstruction.OperandFlag.CONST;
import static org.graalvm.compiler.lir.LIRInstruction.OperandFlag.REG;

import org.graalvm.compiler.asm.Label;
import org.graalvm.compiler.asm.aarch64.AArch64Assembler;
import org.graalvm.compiler.asm.aarch64.AArch64MacroAssembler;
import org.graalvm.compiler.asm.aarch64.AArch64MacroAssembler.ScratchRegister;
import org.graalvm.compiler.lir.LIRInstructionClass;
import org.graalvm.compiler.lir.LIRValueUtil;
import org.graalvm.compiler.lir.Opcode;
import org.graalvm.compiler.lir.asm.CompilationResultBuilder;

import jdk.vm.ci.aarch64.AArch64Kind;
import jdk.vm.ci.code.Register;
import jdk.vm.ci.meta.AllocatableValue;
import jdk.vm.ci.meta.Value;

public class AArch64AtomicMove {
    
Compare and swap instruction. Does the following atomically: CAS(newVal, expected, address): oldVal = *address if oldVal == expected: *address = newVal return oldVal
/** * Compare and swap instruction. Does the following atomically: <code> * CAS(newVal, expected, address): * oldVal = *address * if oldVal == expected: * *address = newVal * return oldVal * </code> */
@Opcode("CAS") public static class CompareAndSwapOp extends AArch64LIRInstruction { public static final LIRInstructionClass<CompareAndSwapOp> TYPE = LIRInstructionClass.create(CompareAndSwapOp.class); @Def protected AllocatableValue resultValue; @Alive protected Value expectedValue; @Alive protected AllocatableValue newValue; @Alive protected AllocatableValue addressValue; @Temp protected AllocatableValue scratchValue; public CompareAndSwapOp(AllocatableValue result, Value expectedValue, AllocatableValue newValue, AllocatableValue addressValue, AllocatableValue scratch) { super(TYPE); this.resultValue = result; this.expectedValue = expectedValue; this.newValue = newValue; this.addressValue = addressValue; this.scratchValue = scratch; } @Override public void emitCode(CompilationResultBuilder crb, AArch64MacroAssembler masm) { AArch64Kind kind = (AArch64Kind) expectedValue.getPlatformKind(); assert kind.isInteger(); final int size = kind.getSizeInBytes() * Byte.SIZE; Register address = asRegister(addressValue); Register result = asRegister(resultValue); Register newVal = asRegister(newValue); if (AArch64LIRFlagsVersioned.useLSE(masm.target.arch)) { Register expected = asRegister(expectedValue); masm.mov(size, result, expected); masm.cas(size, result, newVal, address, true /* acquire */, true /* release */); AArch64Compare.gpCompare(masm, resultValue, expectedValue); } else { // We could avoid using a scratch register here, by reusing resultValue for the // stlxr success flag and issue a mov resultValue, expectedValue in case of success // before returning. Register scratch = asRegister(scratchValue); Label retry = new Label(); Label fail = new Label(); masm.bind(retry); masm.ldaxr(size, result, address); AArch64Compare.gpCompare(masm, resultValue, expectedValue); masm.branchConditionally(AArch64Assembler.ConditionFlag.NE, fail); masm.stlxr(size, scratch, newVal, address); // if scratch == 0 then write successful, else retry. masm.cbnz(32, scratch, retry); masm.bind(fail); } } }
Load (Read) and Add instruction. Does the following atomically: ATOMIC_READ_AND_ADD(addend, result, address): result = *address *address = result + addend return result
/** * Load (Read) and Add instruction. Does the following atomically: <code> * ATOMIC_READ_AND_ADD(addend, result, address): * result = *address * *address = result + addend * return result * </code> */
@Opcode("ATOMIC_READ_AND_ADD") public static final class AtomicReadAndAddOp extends AArch64LIRInstruction { public static final LIRInstructionClass<AtomicReadAndAddOp> TYPE = LIRInstructionClass.create(AtomicReadAndAddOp.class); private final AArch64Kind accessKind; @Def({REG}) protected AllocatableValue resultValue; @Alive({REG}) protected AllocatableValue addressValue; @Alive({REG, CONST}) protected Value deltaValue; public AtomicReadAndAddOp(AArch64Kind kind, AllocatableValue result, AllocatableValue address, Value delta) { super(TYPE); this.accessKind = kind; this.resultValue = result; this.addressValue = address; this.deltaValue = delta; } @Override public void emitCode(CompilationResultBuilder crb, AArch64MacroAssembler masm) { assert accessKind.isInteger(); final int size = accessKind.getSizeInBytes() * Byte.SIZE; Register address = asRegister(addressValue); Register result = asRegister(resultValue); Label retry = new Label(); masm.bind(retry); masm.ldaxr(size, result, address); try (ScratchRegister scratchRegister1 = masm.getScratchRegister()) { Register scratch1 = scratchRegister1.getRegister(); if (LIRValueUtil.isConstantValue(deltaValue)) { long delta = LIRValueUtil.asConstantValue(deltaValue).getJavaConstant().asLong(); masm.add(size, scratch1, result, delta); } else { // must be a register then masm.add(size, scratch1, result, asRegister(deltaValue)); } try (ScratchRegister scratchRegister2 = masm.getScratchRegister()) { Register scratch2 = scratchRegister2.getRegister(); masm.stlxr(size, scratch2, scratch1, address); // if scratch2 == 0 then write successful, else retry masm.cbnz(32, scratch2, retry); } } } }
Load (Read) and Add instruction. Does the following atomically: ATOMIC_READ_AND_ADD(addend, result, address): result = *address *address = result + addend return result The LSE version has different properties with regards to the register allocator. To define these differences, we have to create a separate LIR instruction class. The difference to AtomicReadAndAddOp is:
  • deltaValue must be a register (@Use({REG}) instead @Alive({REG,CONST}))
  • resultValue may be an alias for the input registers (@Use instead of @Alive)
  • /** * Load (Read) and Add instruction. Does the following atomically: <code> * ATOMIC_READ_AND_ADD(addend, result, address): * result = *address * *address = result + addend * return result * </code> * * The LSE version has different properties with regards to the register allocator. To define * these differences, we have to create a separate LIR instruction class. * * The difference to {@linkplain AtomicReadAndAddOp} is: * <li>{@linkplain #deltaValue} must be a register (@Use({REG}) instead @Alive({REG,CONST})) * <li>{@linkplain #resultValue} may be an alias for the input registers (@Use instead * of @Alive) */
    @Opcode("ATOMIC_READ_AND_ADD") public static final class AtomicReadAndAddLSEOp extends AArch64LIRInstruction { public static final LIRInstructionClass<AtomicReadAndAddLSEOp> TYPE = LIRInstructionClass.create(AtomicReadAndAddLSEOp.class); private final AArch64Kind accessKind; @Def({REG}) protected AllocatableValue resultValue; @Use({REG}) protected AllocatableValue addressValue; @Use({REG}) protected AllocatableValue deltaValue; public AtomicReadAndAddLSEOp(AArch64Kind kind, AllocatableValue result, AllocatableValue address, AllocatableValue delta) { super(TYPE); this.accessKind = kind; this.resultValue = result; this.addressValue = address; this.deltaValue = delta; } @Override public void emitCode(CompilationResultBuilder crb, AArch64MacroAssembler masm) { assert accessKind.isInteger(); final int size = accessKind.getSizeInBytes() * Byte.SIZE; Register address = asRegister(addressValue); Register delta = asRegister(deltaValue); Register result = asRegister(resultValue); masm.ldadd(size, delta, result, address, true, true); } }
    Load (Read) and Write instruction. Does the following atomically: ATOMIC_READ_AND_WRITE(newValue, result, address): result = *address *address = newValue return result
    /** * Load (Read) and Write instruction. Does the following atomically: <code> * ATOMIC_READ_AND_WRITE(newValue, result, address): * result = *address * *address = newValue * return result * </code> */
    @Opcode("ATOMIC_READ_AND_WRITE") public static final class AtomicReadAndWriteOp extends AArch64LIRInstruction { public static final LIRInstructionClass<AtomicReadAndWriteOp> TYPE = LIRInstructionClass.create(AtomicReadAndWriteOp.class); private final AArch64Kind accessKind; @Def protected AllocatableValue resultValue; @Alive protected AllocatableValue addressValue; @Alive protected AllocatableValue newValue; @Temp protected AllocatableValue scratchValue; public AtomicReadAndWriteOp(AArch64Kind kind, AllocatableValue result, AllocatableValue address, AllocatableValue newValue, AllocatableValue scratch) { super(TYPE); this.accessKind = kind; this.resultValue = result; this.addressValue = address; this.newValue = newValue; this.scratchValue = scratch; } @Override public void emitCode(CompilationResultBuilder crb, AArch64MacroAssembler masm) { assert accessKind.isInteger(); final int size = accessKind.getSizeInBytes() * Byte.SIZE; Register address = asRegister(addressValue); Register value = asRegister(newValue); Register result = asRegister(resultValue); if (AArch64LIRFlagsVersioned.useLSE(masm.target.arch)) { masm.swp(size, value, result, address, true, true); } else { Register scratch = asRegister(scratchValue); Label retry = new Label(); masm.bind(retry); masm.ldaxr(size, result, address); masm.stlxr(size, scratch, value, address); // if scratch == 0 then write successful, else retry masm.cbnz(32, scratch, retry); } } } }