blob: a5e4efde82ced4c1873037b55685e137e92ee893 [file] [log] [blame]
/*
* Copyright (c) 2013, 2015, 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.amd64;
import static jdk.vm.ci.code.ValueUtil.asRegister;
import static org.graalvm.compiler.lir.LIRInstruction.OperandFlag.ILLEGAL;
import static org.graalvm.compiler.lir.LIRInstruction.OperandFlag.REG;
import org.graalvm.compiler.asm.Label;
import org.graalvm.compiler.asm.amd64.AMD64Address;
import org.graalvm.compiler.asm.amd64.AMD64Address.Scale;
import org.graalvm.compiler.asm.amd64.AMD64Assembler.ConditionFlag;
import org.graalvm.compiler.asm.amd64.AMD64Assembler.OperandSize;
import org.graalvm.compiler.asm.amd64.AMD64Assembler.SSEOp;
import org.graalvm.compiler.asm.amd64.AMD64MacroAssembler;
import org.graalvm.compiler.core.common.LIRKind;
import org.graalvm.compiler.core.common.NumUtil;
import org.graalvm.compiler.lir.LIRInstructionClass;
import org.graalvm.compiler.lir.Opcode;
import org.graalvm.compiler.lir.asm.CompilationResultBuilder;
import org.graalvm.compiler.lir.gen.LIRGeneratorTool;
import jdk.vm.ci.amd64.AMD64;
import jdk.vm.ci.amd64.AMD64.CPUFeature;
import jdk.vm.ci.amd64.AMD64Kind;
import jdk.vm.ci.code.Register;
import jdk.vm.ci.code.TargetDescription;
import jdk.vm.ci.meta.JavaKind;
import jdk.vm.ci.meta.Value;
/**
* Emits code which compares two arrays of the same length. If the CPU supports any vector
* instructions specialized code is emitted to leverage these instructions.
*/
@Opcode("ARRAY_EQUALS")
public final class AMD64ArrayEqualsOp extends AMD64LIRInstruction {
public static final LIRInstructionClass<AMD64ArrayEqualsOp> TYPE = LIRInstructionClass.create(AMD64ArrayEqualsOp.class);
private final JavaKind kind;
private final int arrayBaseOffset;
private final int arrayIndexScale;
@Def({REG}) protected Value resultValue;
@Alive({REG}) protected Value array1Value;
@Alive({REG}) protected Value array2Value;
@Alive({REG}) protected Value lengthValue;
@Temp({REG}) protected Value temp1;
@Temp({REG}) protected Value temp2;
@Temp({REG}) protected Value temp3;
@Temp({REG}) protected Value temp4;
@Temp({REG, ILLEGAL}) protected Value temp5;
@Temp({REG, ILLEGAL}) protected Value tempXMM;
@Temp({REG, ILLEGAL}) protected Value vectorTemp1;
@Temp({REG, ILLEGAL}) protected Value vectorTemp2;
public AMD64ArrayEqualsOp(LIRGeneratorTool tool, JavaKind kind, Value result, Value array1, Value array2, Value length) {
super(TYPE);
this.kind = kind;
this.arrayBaseOffset = tool.getProviders().getArrayOffsetProvider().arrayBaseOffset(kind);
this.arrayIndexScale = tool.getProviders().getArrayOffsetProvider().arrayScalingFactor(kind);
this.resultValue = result;
this.array1Value = array1;
this.array2Value = array2;
this.lengthValue = length;
// Allocate some temporaries.
this.temp1 = tool.newVariable(LIRKind.unknownReference(tool.target().arch.getWordKind()));
this.temp2 = tool.newVariable(LIRKind.unknownReference(tool.target().arch.getWordKind()));
this.temp3 = tool.newVariable(LIRKind.value(tool.target().arch.getWordKind()));
this.temp4 = tool.newVariable(LIRKind.value(tool.target().arch.getWordKind()));
this.temp5 = kind.isNumericFloat() ? tool.newVariable(LIRKind.value(tool.target().arch.getWordKind())) : Value.ILLEGAL;
if (kind == JavaKind.Float) {
this.tempXMM = tool.newVariable(LIRKind.value(AMD64Kind.SINGLE));
} else if (kind == JavaKind.Double) {
this.tempXMM = tool.newVariable(LIRKind.value(AMD64Kind.DOUBLE));
} else {
this.tempXMM = Value.ILLEGAL;
}
// We only need the vector temporaries if we generate SSE code.
if (supportsSSE41(tool.target())) {
this.vectorTemp1 = tool.newVariable(LIRKind.value(AMD64Kind.DOUBLE));
this.vectorTemp2 = tool.newVariable(LIRKind.value(AMD64Kind.DOUBLE));
} else {
this.vectorTemp1 = Value.ILLEGAL;
this.vectorTemp2 = Value.ILLEGAL;
}
}
@Override
public void emitCode(CompilationResultBuilder crb, AMD64MacroAssembler masm) {
Register result = asRegister(resultValue);
Register array1 = asRegister(temp1);
Register array2 = asRegister(temp2);
Register length = asRegister(temp3);
Label trueLabel = new Label();
Label falseLabel = new Label();
Label done = new Label();
// Load array base addresses.
masm.leaq(array1, new AMD64Address(asRegister(array1Value), arrayBaseOffset));
masm.leaq(array2, new AMD64Address(asRegister(array2Value), arrayBaseOffset));
// Get array length in bytes.
masm.movl(length, asRegister(lengthValue));
if (arrayIndexScale > 1) {
masm.shll(length, NumUtil.log2Ceil(arrayIndexScale)); // scale length
}
masm.movl(result, length); // copy
if (supportsAVX2(crb.target)) {
emitAVXCompare(crb, masm, result, array1, array2, length, trueLabel, falseLabel);
} else if (supportsSSE41(crb.target)) {
// this code is used for AVX as well because our backend correctly ensures that
// VEX-prefixed instructions are emitted if AVX is supported
emitSSE41Compare(crb, masm, result, array1, array2, length, trueLabel, falseLabel);
}
emit8ByteCompare(crb, masm, result, array1, array2, length, trueLabel, falseLabel);
emitTailCompares(masm, result, array1, array2, length, trueLabel, falseLabel);
// Return true
masm.bind(trueLabel);
masm.movl(result, 1);
masm.jmpb(done);
// Return false
masm.bind(falseLabel);
masm.xorl(result, result);
// That's it
masm.bind(done);
}
/**
* Returns if the underlying AMD64 architecture supports SSE 4.1 instructions.
*
* @param target target description of the underlying architecture
* @return true if the underlying architecture supports SSE 4.1
*/
private static boolean supportsSSE41(TargetDescription target) {
AMD64 arch = (AMD64) target.arch;
return arch.getFeatures().contains(CPUFeature.SSE4_1);
}
/**
* Vector size used in {@link #emitSSE41Compare}.
*/
private static final int SSE4_1_VECTOR_SIZE = 16;
/**
* Emits code that uses SSE4.1 128-bit (16-byte) vector compares.
*/
private void emitSSE41Compare(CompilationResultBuilder crb, AMD64MacroAssembler masm, Register result, Register array1, Register array2, Register length, Label trueLabel, Label falseLabel) {
assert supportsSSE41(crb.target);
Register vector1 = asRegister(vectorTemp1, AMD64Kind.DOUBLE);
Register vector2 = asRegister(vectorTemp2, AMD64Kind.DOUBLE);
Label loop = new Label();
Label compareTail = new Label();
boolean requiresNaNCheck = kind.isNumericFloat();
Label loopCheck = new Label();
Label nanCheck = new Label();
// Compare 16-byte vectors
masm.andl(result, SSE4_1_VECTOR_SIZE - 1); // tail count (in bytes)
masm.andl(length, ~(SSE4_1_VECTOR_SIZE - 1)); // vector count (in bytes)
masm.jcc(ConditionFlag.Zero, compareTail);
masm.leaq(array1, new AMD64Address(array1, length, Scale.Times1, 0));
masm.leaq(array2, new AMD64Address(array2, length, Scale.Times1, 0));
masm.negq(length);
// Align the main loop
masm.align(crb.target.wordSize * 2);
masm.bind(loop);
masm.movdqu(vector1, new AMD64Address(array1, length, Scale.Times1, 0));
masm.movdqu(vector2, new AMD64Address(array2, length, Scale.Times1, 0));
masm.pxor(vector1, vector2);
masm.ptest(vector1, vector1);
masm.jcc(ConditionFlag.NotZero, requiresNaNCheck ? nanCheck : falseLabel);
masm.bind(loopCheck);
masm.addq(length, SSE4_1_VECTOR_SIZE);
masm.jcc(ConditionFlag.NotZero, loop);
masm.testl(result, result);
masm.jcc(ConditionFlag.Zero, trueLabel);
if (requiresNaNCheck) {
Label unalignedCheck = new Label();
masm.jmpb(unalignedCheck);
masm.bind(nanCheck);
emitFloatCompareWithinRange(crb, masm, array1, array2, length, 0, falseLabel, SSE4_1_VECTOR_SIZE);
masm.jmpb(loopCheck);
masm.bind(unalignedCheck);
}
/*
* Compare the remaining bytes with an unaligned memory load aligned to the end of the
* array.
*/
masm.movdqu(vector1, new AMD64Address(array1, result, Scale.Times1, -SSE4_1_VECTOR_SIZE));
masm.movdqu(vector2, new AMD64Address(array2, result, Scale.Times1, -SSE4_1_VECTOR_SIZE));
masm.pxor(vector1, vector2);
masm.ptest(vector1, vector1);
if (requiresNaNCheck) {
masm.jcc(ConditionFlag.Zero, trueLabel);
emitFloatCompareWithinRange(crb, masm, array1, array2, result, -SSE4_1_VECTOR_SIZE, falseLabel, SSE4_1_VECTOR_SIZE);
} else {
masm.jcc(ConditionFlag.NotZero, falseLabel);
}
masm.jmp(trueLabel);
masm.bind(compareTail);
masm.movl(length, result);
}
/**
* Returns if the underlying AMD64 architecture supports AVX instructions.
*
* @param target target description of the underlying architecture
* @return true if the underlying architecture supports AVX
*/
private static boolean supportsAVX2(TargetDescription target) {
AMD64 arch = (AMD64) target.arch;
return arch.getFeatures().contains(CPUFeature.AVX2);
}
/**
* Vector size used in {@link #emitAVXCompare}.
*/
private static final int AVX_VECTOR_SIZE = 32;
private void emitAVXCompare(CompilationResultBuilder crb, AMD64MacroAssembler masm, Register result, Register array1, Register array2, Register length, Label trueLabel, Label falseLabel) {
assert supportsAVX2(crb.target);
Register vector1 = asRegister(vectorTemp1, AMD64Kind.DOUBLE);
Register vector2 = asRegister(vectorTemp2, AMD64Kind.DOUBLE);
Label loop = new Label();
Label compareTail = new Label();
boolean requiresNaNCheck = kind.isNumericFloat();
Label loopCheck = new Label();
Label nanCheck = new Label();
// Compare 16-byte vectors
masm.andl(result, AVX_VECTOR_SIZE - 1); // tail count (in bytes)
masm.andl(length, ~(AVX_VECTOR_SIZE - 1)); // vector count (in bytes)
masm.jcc(ConditionFlag.Zero, compareTail);
masm.leaq(array1, new AMD64Address(array1, length, Scale.Times1, 0));
masm.leaq(array2, new AMD64Address(array2, length, Scale.Times1, 0));
masm.negq(length);
// Align the main loop
masm.align(crb.target.wordSize * 2);
masm.bind(loop);
masm.vmovdqu(vector1, new AMD64Address(array1, length, Scale.Times1, 0));
masm.vmovdqu(vector2, new AMD64Address(array2, length, Scale.Times1, 0));
masm.vpxor(vector1, vector1, vector2);
masm.vptest(vector1, vector1);
masm.jcc(ConditionFlag.NotZero, requiresNaNCheck ? nanCheck : falseLabel);
masm.bind(loopCheck);
masm.addq(length, AVX_VECTOR_SIZE);
masm.jcc(ConditionFlag.NotZero, loop);
masm.testl(result, result);
masm.jcc(ConditionFlag.Zero, trueLabel);
if (requiresNaNCheck) {
Label unalignedCheck = new Label();
masm.jmpb(unalignedCheck);
masm.bind(nanCheck);
emitFloatCompareWithinRange(crb, masm, array1, array2, length, 0, falseLabel, AVX_VECTOR_SIZE);
masm.jmpb(loopCheck);
masm.bind(unalignedCheck);
}
/*
* Compare the remaining bytes with an unaligned memory load aligned to the end of the
* array.
*/
masm.vmovdqu(vector1, new AMD64Address(array1, result, Scale.Times1, -AVX_VECTOR_SIZE));
masm.vmovdqu(vector2, new AMD64Address(array2, result, Scale.Times1, -AVX_VECTOR_SIZE));
masm.vpxor(vector1, vector1, vector2);
masm.vptest(vector1, vector1);
if (requiresNaNCheck) {
masm.jcc(ConditionFlag.Zero, trueLabel);
emitFloatCompareWithinRange(crb, masm, array1, array2, result, -AVX_VECTOR_SIZE, falseLabel, AVX_VECTOR_SIZE);
} else {
masm.jcc(ConditionFlag.NotZero, falseLabel);
}
masm.jmp(trueLabel);
masm.bind(compareTail);
masm.movl(length, result);
}
/**
* Vector size used in {@link #emit8ByteCompare}.
*/
private static final int VECTOR_SIZE = 8;
/**
* Emits code that uses 8-byte vector compares.
*/
private void emit8ByteCompare(CompilationResultBuilder crb, AMD64MacroAssembler masm, Register result, Register array1, Register array2, Register length, Label trueLabel, Label falseLabel) {
Label loop = new Label();
Label compareTail = new Label();
boolean requiresNaNCheck = kind.isNumericFloat();
Label loopCheck = new Label();
Label nanCheck = new Label();
Register temp = asRegister(temp4);
masm.andl(result, VECTOR_SIZE - 1); // tail count (in bytes)
masm.andl(length, ~(VECTOR_SIZE - 1)); // vector count (in bytes)
masm.jcc(ConditionFlag.Zero, compareTail);
masm.leaq(array1, new AMD64Address(array1, length, Scale.Times1, 0));
masm.leaq(array2, new AMD64Address(array2, length, Scale.Times1, 0));
masm.negq(length);
// Align the main loop
masm.align(crb.target.wordSize * 2);
masm.bind(loop);
masm.movq(temp, new AMD64Address(array1, length, Scale.Times1, 0));
masm.cmpq(temp, new AMD64Address(array2, length, Scale.Times1, 0));
masm.jcc(ConditionFlag.NotEqual, requiresNaNCheck ? nanCheck : falseLabel);
masm.bind(loopCheck);
masm.addq(length, VECTOR_SIZE);
masm.jccb(ConditionFlag.NotZero, loop);
masm.testl(result, result);
masm.jcc(ConditionFlag.Zero, trueLabel);
if (requiresNaNCheck) {
// NaN check is slow path and hence placed outside of the main loop.
Label unalignedCheck = new Label();
masm.jmpb(unalignedCheck);
masm.bind(nanCheck);
// At most two iterations, unroll in the emitted code.
for (int offset = 0; offset < VECTOR_SIZE; offset += kind.getByteCount()) {
emitFloatCompare(masm, array1, array2, length, offset, falseLabel, kind.getByteCount() == VECTOR_SIZE);
}
masm.jmpb(loopCheck);
masm.bind(unalignedCheck);
}
/*
* Compare the remaining bytes with an unaligned memory load aligned to the end of the
* array.
*/
masm.movq(temp, new AMD64Address(array1, result, Scale.Times1, -VECTOR_SIZE));
masm.cmpq(temp, new AMD64Address(array2, result, Scale.Times1, -VECTOR_SIZE));
if (requiresNaNCheck) {
masm.jcc(ConditionFlag.Equal, trueLabel);
// At most two iterations, unroll in the emitted code.
for (int offset = 0; offset < VECTOR_SIZE; offset += kind.getByteCount()) {
emitFloatCompare(masm, array1, array2, result, -VECTOR_SIZE + offset, falseLabel, kind.getByteCount() == VECTOR_SIZE);
}
} else {
masm.jccb(ConditionFlag.NotEqual, falseLabel);
}
masm.jmpb(trueLabel);
masm.bind(compareTail);
masm.movl(length, result);
}
/**
* Emits code to compare the remaining 1 to 4 bytes.
*/
private void emitTailCompares(AMD64MacroAssembler masm, Register result, Register array1, Register array2, Register length, Label trueLabel, Label falseLabel) {
Label compare2Bytes = new Label();
Label compare1Byte = new Label();
Register temp = asRegister(temp4);
if (kind.getByteCount() <= 4) {
// Compare trailing 4 bytes, if any.
masm.testl(result, 4);
masm.jccb(ConditionFlag.Zero, compare2Bytes);
masm.movl(temp, new AMD64Address(array1, 0));
masm.cmpl(temp, new AMD64Address(array2, 0));
if (kind == JavaKind.Float) {
masm.jccb(ConditionFlag.Equal, trueLabel);
emitFloatCompare(masm, array1, array2, Register.None, 0, falseLabel, true);
masm.jmpb(trueLabel);
} else {
masm.jccb(ConditionFlag.NotEqual, falseLabel);
}
if (kind.getByteCount() <= 2) {
// Move array pointers forward.
masm.leaq(array1, new AMD64Address(array1, 4));
masm.leaq(array2, new AMD64Address(array2, 4));
// Compare trailing 2 bytes, if any.
masm.bind(compare2Bytes);
masm.testl(result, 2);
masm.jccb(ConditionFlag.Zero, compare1Byte);
masm.movzwl(temp, new AMD64Address(array1, 0));
masm.movzwl(length, new AMD64Address(array2, 0));
masm.cmpl(temp, length);
masm.jccb(ConditionFlag.NotEqual, falseLabel);
// The one-byte tail compare is only required for boolean and byte arrays.
if (kind.getByteCount() <= 1) {
// Move array pointers forward before we compare the last trailing byte.
masm.leaq(array1, new AMD64Address(array1, 2));
masm.leaq(array2, new AMD64Address(array2, 2));
// Compare trailing byte, if any.
masm.bind(compare1Byte);
masm.testl(result, 1);
masm.jccb(ConditionFlag.Zero, trueLabel);
masm.movzbl(temp, new AMD64Address(array1, 0));
masm.movzbl(length, new AMD64Address(array2, 0));
masm.cmpl(temp, length);
masm.jccb(ConditionFlag.NotEqual, falseLabel);
} else {
masm.bind(compare1Byte);
}
} else {
masm.bind(compare2Bytes);
}
}
}
/**
* Emits code to fall through if {@code src} is NaN, otherwise jump to {@code branchOrdered}.
*/
private void emitNaNCheck(AMD64MacroAssembler masm, AMD64Address src, Label branchIfNonNaN) {
assert kind.isNumericFloat();
Register tempXMMReg = asRegister(tempXMM);
if (kind == JavaKind.Float) {
masm.movflt(tempXMMReg, src);
} else {
masm.movdbl(tempXMMReg, src);
}
SSEOp.UCOMIS.emit(masm, kind == JavaKind.Float ? OperandSize.PS : OperandSize.PD, tempXMMReg, tempXMMReg);
masm.jcc(ConditionFlag.NoParity, branchIfNonNaN);
}
/**
* Emits code to compare if two floats are bitwise equal or both NaN.
*/
private void emitFloatCompare(AMD64MacroAssembler masm, Register base1, Register base2, Register index, int offset, Label falseLabel, boolean skipBitwiseCompare) {
AMD64Address address1 = new AMD64Address(base1, index, Scale.Times1, offset);
AMD64Address address2 = new AMD64Address(base2, index, Scale.Times1, offset);
Label bitwiseEqual = new Label();
if (!skipBitwiseCompare) {
// Bitwise compare
Register temp = asRegister(temp4);
if (kind == JavaKind.Float) {
masm.movl(temp, address1);
masm.cmpl(temp, address2);
} else {
masm.movq(temp, address1);
masm.cmpq(temp, address2);
}
masm.jccb(ConditionFlag.Equal, bitwiseEqual);
}
emitNaNCheck(masm, address1, falseLabel);
emitNaNCheck(masm, address2, falseLabel);
masm.bind(bitwiseEqual);
}
/**
* Emits code to compare float equality within a range.
*/
private void emitFloatCompareWithinRange(CompilationResultBuilder crb, AMD64MacroAssembler masm, Register base1, Register base2, Register index, int offset, Label falseLabel, int range) {
assert kind.isNumericFloat();
Label loop = new Label();
Register i = asRegister(temp5);
masm.movq(i, range);
masm.negq(i);
// Align the main loop
masm.align(crb.target.wordSize * 2);
masm.bind(loop);
emitFloatCompare(masm, base1, base2, index, offset, falseLabel, kind.getByteCount() == range);
masm.addq(index, kind.getByteCount());
masm.addq(i, kind.getByteCount());
masm.jccb(ConditionFlag.NotZero, loop);
// Floats within the range are equal, revert change to the register index
masm.subq(index, range);
}
}