1 /*
2 * Copyright (c) 2009, 2019, 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
26 package org.graalvm.compiler.core.amd64;
27
28 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64BinaryArithmetic.ADD;
29 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64BinaryArithmetic.AND;
30 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64BinaryArithmetic.CMP;
31 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64BinaryArithmetic.OR;
32 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64BinaryArithmetic.SUB;
33 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64BinaryArithmetic.XOR;
34 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64MOp.NEG;
35 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64MOp.NOT;
36 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.BSF;
37 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.BSR;
38 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.LZCNT;
39 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.MOV;
40 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.MOVSD;
41 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.MOVSS;
42 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.MOVSX;
43 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.MOVSXB;
44 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.MOVSXD;
45 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.MOVZX;
46 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.MOVZXB;
47 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.POPCNT;
48 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.TEST;
49 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.TESTB;
50 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp.TZCNT;
51 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64Shift.ROL;
52 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64Shift.ROR;
53 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64Shift.SAR;
54 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64Shift.SHL;
55 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64Shift.SHR;
56 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VADDSD;
57 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VADDSS;
58 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VDIVSD;
59 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VDIVSS;
60 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VMULSD;
61 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VMULSS;
62 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VORPD;
63 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VORPS;
64 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VSUBSD;
65 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VSUBSS;
66 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VXORPD;
67 import static org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp.VXORPS;
68 import static org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize.BYTE;
69 import static org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize.DWORD;
70 import static org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize.PD;
71 import static org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize.PS;
72 import static org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize.QWORD;
73 import static org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize.SD;
74 import static org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize.SS;
75 import static org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize.WORD;
76 import static org.graalvm.compiler.core.common.GraalOptions.GeneratePIC;
77 import static org.graalvm.compiler.lir.LIRValueUtil.asConstantValue;
78 import static org.graalvm.compiler.lir.LIRValueUtil.asJavaConstant;
79 import static org.graalvm.compiler.lir.LIRValueUtil.isConstantValue;
80 import static org.graalvm.compiler.lir.LIRValueUtil.isJavaConstant;
81 import static org.graalvm.compiler.lir.amd64.AMD64Arithmetic.DREM;
82 import static org.graalvm.compiler.lir.amd64.AMD64Arithmetic.FREM;
83
84 import org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64BinaryArithmetic;
85 import org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64MIOp;
86 import org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64MOp;
87 import org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64MROp;
88 import org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMIOp;
89 import org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64RMOp;
90 import org.graalvm.compiler.asm.amd64.AMD64Assembler.AMD64Shift;
91 import org.graalvm.compiler.asm.amd64.AMD64Assembler.SSEOp;
92 import org.graalvm.compiler.asm.amd64.AMD64Assembler.VexGeneralPurposeRMOp;
93 import org.graalvm.compiler.asm.amd64.AMD64Assembler.VexGeneralPurposeRVMOp;
94 import org.graalvm.compiler.asm.amd64.AMD64Assembler.VexRVMOp;
95 import org.graalvm.compiler.asm.amd64.AMD64BaseAssembler.OperandSize;
96 import org.graalvm.compiler.asm.amd64.AVXKind;
97 import org.graalvm.compiler.asm.amd64.AVXKind.AVXSize;
98 import org.graalvm.compiler.core.common.LIRKind;
99 import org.graalvm.compiler.core.common.NumUtil;
100 import org.graalvm.compiler.core.common.calc.FloatConvert;
101 import org.graalvm.compiler.debug.GraalError;
102 import org.graalvm.compiler.lir.ConstantValue;
103 import org.graalvm.compiler.lir.LIRFrameState;
104 import org.graalvm.compiler.lir.LIRValueUtil;
105 import org.graalvm.compiler.lir.Variable;
106 import org.graalvm.compiler.lir.amd64.AMD64AddressValue;
107 import org.graalvm.compiler.lir.amd64.AMD64Arithmetic.FPDivRemOp;
108 import org.graalvm.compiler.lir.amd64.AMD64ArithmeticLIRGeneratorTool;
109 import org.graalvm.compiler.lir.amd64.AMD64Binary;
110 import org.graalvm.compiler.lir.amd64.AMD64BinaryConsumer;
111 import org.graalvm.compiler.lir.amd64.AMD64ClearRegisterOp;
112 import org.graalvm.compiler.lir.amd64.AMD64MathCosOp;
113 import org.graalvm.compiler.lir.amd64.AMD64MathExpOp;
114 import org.graalvm.compiler.lir.amd64.AMD64MathLog10Op;
115 import org.graalvm.compiler.lir.amd64.AMD64MathLogOp;
116 import org.graalvm.compiler.lir.amd64.AMD64MathPowOp;
117 import org.graalvm.compiler.lir.amd64.AMD64MathSinOp;
118 import org.graalvm.compiler.lir.amd64.AMD64MathTanOp;
119 import org.graalvm.compiler.lir.amd64.AMD64Move;
120 import org.graalvm.compiler.lir.amd64.AMD64MulDivOp;
121 import org.graalvm.compiler.lir.amd64.AMD64ShiftOp;
122 import org.graalvm.compiler.lir.amd64.AMD64SignExtendOp;
123 import org.graalvm.compiler.lir.amd64.AMD64Unary;
124 import org.graalvm.compiler.lir.amd64.vector.AMD64VectorBinary;
125 import org.graalvm.compiler.lir.amd64.vector.AMD64VectorBinary.AVXBinaryOp;
126 import org.graalvm.compiler.lir.amd64.vector.AMD64VectorUnary;
127 import org.graalvm.compiler.lir.gen.ArithmeticLIRGenerator;
128
129 import jdk.vm.ci.amd64.AMD64;
130 import jdk.vm.ci.amd64.AMD64.CPUFeature;
131 import jdk.vm.ci.amd64.AMD64Kind;
132 import jdk.vm.ci.code.CodeUtil;
133 import jdk.vm.ci.code.Register;
134 import jdk.vm.ci.code.RegisterValue;
135 import jdk.vm.ci.code.TargetDescription;
136 import jdk.vm.ci.meta.AllocatableValue;
137 import jdk.vm.ci.meta.Constant;
138 import jdk.vm.ci.meta.JavaConstant;
139 import jdk.vm.ci.meta.JavaKind;
140 import jdk.vm.ci.meta.PlatformKind;
141 import jdk.vm.ci.meta.VMConstant;
142 import jdk.vm.ci.meta.Value;
143 import jdk.vm.ci.meta.ValueKind;
144
145 /**
146 * This class implements the AMD64 specific portion of the LIR generator.
147 */
148 public class AMD64ArithmeticLIRGenerator extends ArithmeticLIRGenerator implements AMD64ArithmeticLIRGeneratorTool {
149
150 private static final RegisterValue RCX_I = AMD64.rcx.asValue(LIRKind.value(AMD64Kind.DWORD));
151
152 public AMD64ArithmeticLIRGenerator(AllocatableValue nullRegisterValue) {
153 this.nullRegisterValue = nullRegisterValue;
154 }
155
156 private final AllocatableValue nullRegisterValue;
157
158 @Override
159 public Variable emitNegate(Value inputVal) {
160 AllocatableValue input = asAllocatable(inputVal);
161 Variable result = getLIRGen().newVariable(LIRKind.combine(input));
162 boolean isAvx = supportAVX();
163 switch ((AMD64Kind) input.getPlatformKind()) {
164 case DWORD:
165 getLIRGen().append(new AMD64Unary.MOp(NEG, DWORD, result, input));
166 break;
167 case QWORD:
168 getLIRGen().append(new AMD64Unary.MOp(NEG, QWORD, result, input));
169 break;
170 case SINGLE:
171 JavaConstant floatMask = JavaConstant.forFloat(Float.intBitsToFloat(0x80000000));
172 if (isAvx) {
173 getLIRGen().append(new AVXBinaryOp(VXORPS, getRegisterSize(result), result, asAllocatable(input), asAllocatable(getLIRGen().emitJavaConstant(floatMask))));
174 } else {
175 getLIRGen().append(new AMD64Binary.DataTwoOp(SSEOp.XOR, PS, result, input, floatMask, 16));
176 }
177 break;
178 case DOUBLE:
179 JavaConstant doubleMask = JavaConstant.forDouble(Double.longBitsToDouble(0x8000000000000000L));
180 if (isAvx) {
181 getLIRGen().append(new AVXBinaryOp(VXORPD, getRegisterSize(result), result, asAllocatable(input), asAllocatable(getLIRGen().emitJavaConstant(doubleMask))));
182 } else {
183 getLIRGen().append(new AMD64Binary.DataTwoOp(SSEOp.XOR, PD, result, input, doubleMask, 16));
184 }
185 break;
186 default:
187 throw GraalError.shouldNotReachHere(input.getPlatformKind().toString());
188 }
189 return result;
190 }
191
192 @Override
193 public Variable emitNot(Value inputVal) {
194 AllocatableValue input = asAllocatable(inputVal);
195 Variable result = getLIRGen().newVariable(LIRKind.combine(input));
196 switch ((AMD64Kind) input.getPlatformKind()) {
197 case DWORD:
198 getLIRGen().append(new AMD64Unary.MOp(NOT, DWORD, result, input));
199 break;
200 case QWORD:
201 getLIRGen().append(new AMD64Unary.MOp(NOT, QWORD, result, input));
202 break;
203 default:
204 throw GraalError.shouldNotReachHere();
205 }
206 return result;
207 }
208
209 private Variable emitBinary(LIRKind resultKind, AMD64BinaryArithmetic op, OperandSize size, boolean commutative, Value a, Value b, boolean setFlags) {
210 if (isJavaConstant(b)) {
211 return emitBinaryConst(resultKind, op, size, commutative, asAllocatable(a), asConstantValue(b), setFlags);
212 } else if (commutative && isJavaConstant(a)) {
213 return emitBinaryConst(resultKind, op, size, commutative, asAllocatable(b), asConstantValue(a), setFlags);
214 } else {
215 return emitBinaryVar(resultKind, op.getRMOpcode(size), size, commutative, asAllocatable(a), asAllocatable(b));
216 }
217 }
218
219 private Variable emitBinary(LIRKind resultKind, AMD64RMOp op, OperandSize size, boolean commutative, Value a, Value b) {
220 if (isJavaConstant(b)) {
221 return emitBinaryConst(resultKind, op, size, asAllocatable(a), asJavaConstant(b));
222 } else if (commutative && isJavaConstant(a)) {
223 return emitBinaryConst(resultKind, op, size, asAllocatable(b), asJavaConstant(a));
224 } else {
225 return emitBinaryVar(resultKind, op, size, commutative, asAllocatable(a), asAllocatable(b));
226 }
227 }
228
229 private Variable emitBinaryConst(LIRKind resultKind, AMD64BinaryArithmetic op, OperandSize size, boolean commutative, AllocatableValue a, ConstantValue b, boolean setFlags) {
230 long value = b.getJavaConstant().asLong();
231 if (NumUtil.isInt(value)) {
232 Variable result = getLIRGen().newVariable(resultKind);
233 int constant = (int) value;
234
235 if (!setFlags) {
236 AMD64MOp mop = getMOp(op, constant);
237 if (mop != null) {
238 getLIRGen().append(new AMD64Unary.MOp(mop, size, result, a));
239 return result;
240 }
241 }
242
243 getLIRGen().append(new AMD64Binary.ConstOp(op, size, result, a, constant));
244 return result;
245 } else {
246 return emitBinaryVar(resultKind, op.getRMOpcode(size), size, commutative, a, asAllocatable(b));
247 }
248 }
249
250 private static AMD64MOp getMOp(AMD64BinaryArithmetic op, int constant) {
251 if (constant == 1) {
252 if (op.equals(AMD64BinaryArithmetic.ADD)) {
253 return AMD64MOp.INC;
254 }
255 if (op.equals(AMD64BinaryArithmetic.SUB)) {
256 return AMD64MOp.DEC;
257 }
258 } else if (constant == -1) {
259 if (op.equals(AMD64BinaryArithmetic.ADD)) {
260 return AMD64MOp.DEC;
261 }
262 if (op.equals(AMD64BinaryArithmetic.SUB)) {
263 return AMD64MOp.INC;
264 }
265 }
266 return null;
267 }
268
269 private Variable emitBinaryConst(LIRKind resultKind, AMD64RMOp op, OperandSize size, AllocatableValue a, JavaConstant b) {
270 Variable result = getLIRGen().newVariable(resultKind);
271 getLIRGen().append(new AMD64Binary.DataTwoOp(op, size, result, a, b));
272 return result;
273 }
274
275 private Variable emitBinaryVar(LIRKind resultKind, AMD64RMOp op, OperandSize size, boolean commutative, AllocatableValue a, AllocatableValue b) {
276 Variable result = getLIRGen().newVariable(resultKind);
277 if (commutative) {
278 getLIRGen().append(new AMD64Binary.CommutativeTwoOp(op, size, result, a, b));
279 } else {
280 getLIRGen().append(new AMD64Binary.TwoOp(op, size, result, a, b));
281 }
282 return result;
283 }
284
285 @Override
286 protected boolean isNumericInteger(PlatformKind kind) {
287 return ((AMD64Kind) kind).isInteger();
288 }
289
290 private Variable emitBaseOffsetLea(LIRKind resultKind, Value base, int offset, OperandSize size) {
291 Variable result = getLIRGen().newVariable(resultKind);
292 AMD64AddressValue address = new AMD64AddressValue(resultKind, asAllocatable(base), offset);
293 getLIRGen().append(new AMD64Move.LeaOp(result, address, size));
294 return result;
295 }
296
297 @Override
298 public Variable emitAdd(LIRKind resultKind, Value a, Value b, boolean setFlags) {
299 boolean isAvx = supportAVX();
300 switch ((AMD64Kind) a.getPlatformKind()) {
301 case DWORD:
302 if (isJavaConstant(b) && !setFlags) {
303 long displacement = asJavaConstant(b).asLong();
304 if (NumUtil.isInt(displacement) && displacement != 1 && displacement != -1) {
305 return emitBaseOffsetLea(resultKind, a, (int) displacement, OperandSize.DWORD);
306 }
307 }
308 return emitBinary(resultKind, ADD, DWORD, true, a, b, setFlags);
309 case QWORD:
310 if (isJavaConstant(b) && !setFlags) {
311 long displacement = asJavaConstant(b).asLong();
312 if (NumUtil.isInt(displacement) && displacement != 1 && displacement != -1) {
313 return emitBaseOffsetLea(resultKind, a, (int) displacement, OperandSize.QWORD);
314 }
315 }
316 return emitBinary(resultKind, ADD, QWORD, true, a, b, setFlags);
317 case SINGLE:
318 if (isAvx) {
319 return emitBinary(resultKind, VADDSS, a, b);
320 } else {
321 return emitBinary(resultKind, SSEOp.ADD, SS, true, a, b);
322 }
323 case DOUBLE:
324 if (isAvx) {
325 return emitBinary(resultKind, VADDSD, a, b);
326 } else {
327 return emitBinary(resultKind, SSEOp.ADD, SD, true, a, b);
328 }
329 default:
330 throw GraalError.shouldNotReachHere();
331 }
332 }
333
334 @Override
335 public Variable emitSub(LIRKind resultKind, Value a, Value b, boolean setFlags) {
336 boolean isAvx = supportAVX();
337 switch ((AMD64Kind) a.getPlatformKind()) {
338 case DWORD:
339 return emitBinary(resultKind, SUB, DWORD, false, a, b, setFlags);
340 case QWORD:
341 return emitBinary(resultKind, SUB, QWORD, false, a, b, setFlags);
342 case SINGLE:
343 if (isAvx) {
344 return emitBinary(resultKind, VSUBSS, a, b);
345 } else {
346 return emitBinary(resultKind, SSEOp.SUB, SS, false, a, b);
347 }
348 case DOUBLE:
349 if (isAvx) {
350 return emitBinary(resultKind, VSUBSD, a, b);
351 } else {
352 return emitBinary(resultKind, SSEOp.SUB, SD, false, a, b);
353 }
354 default:
355 throw GraalError.shouldNotReachHere();
356 }
357 }
358
359 private Variable emitIMULConst(OperandSize size, AllocatableValue a, ConstantValue b) {
360 long value = b.getJavaConstant().asLong();
361 if (NumUtil.isInt(value)) {
362 int imm = (int) value;
363 AMD64RMIOp op;
364 if (NumUtil.isByte(imm)) {
365 op = AMD64RMIOp.IMUL_SX;
366 } else {
367 op = AMD64RMIOp.IMUL;
368 }
369
370 Variable ret = getLIRGen().newVariable(LIRKind.combine(a, b));
371 getLIRGen().append(new AMD64Binary.RMIOp(op, size, ret, a, imm));
372 return ret;
373 } else {
374 return emitBinaryVar(LIRKind.combine(a, b), AMD64RMOp.IMUL, size, true, a, asAllocatable(b));
375 }
376 }
377
378 private Variable emitIMUL(OperandSize size, Value a, Value b) {
379 if (isJavaConstant(b)) {
380 return emitIMULConst(size, asAllocatable(a), asConstantValue(b));
381 } else if (isJavaConstant(a)) {
382 return emitIMULConst(size, asAllocatable(b), asConstantValue(a));
383 } else {
384 return emitBinaryVar(LIRKind.combine(a, b), AMD64RMOp.IMUL, size, true, asAllocatable(a), asAllocatable(b));
385 }
386 }
387
388 @Override
389 public Variable emitMul(Value a, Value b, boolean setFlags) {
390 boolean isAvx = supportAVX();
391 LIRKind resultKind = LIRKind.combine(a, b);
392 switch ((AMD64Kind) a.getPlatformKind()) {
393 case DWORD:
394 return emitIMUL(DWORD, a, b);
395 case QWORD:
396 return emitIMUL(QWORD, a, b);
397 case SINGLE:
398 if (isAvx) {
399 return emitBinary(resultKind, VMULSS, a, b);
400 } else {
401 return emitBinary(resultKind, SSEOp.MUL, SS, true, a, b);
402 }
403 case DOUBLE:
404 if (isAvx) {
405 return emitBinary(resultKind, VMULSD, a, b);
406 } else {
407 return emitBinary(resultKind, SSEOp.MUL, SD, true, a, b);
408 }
409 default:
410 throw GraalError.shouldNotReachHere();
411 }
412 }
413
414 private RegisterValue moveToReg(Register reg, Value v) {
415 RegisterValue ret = reg.asValue(v.getValueKind());
416 getLIRGen().emitMove(ret, v);
417 return ret;
418 }
419
420 private Value emitMulHigh(AMD64MOp opcode, OperandSize size, Value a, Value b) {
421 AMD64MulDivOp mulHigh = getLIRGen().append(new AMD64MulDivOp(opcode, size, LIRKind.combine(a, b), moveToReg(AMD64.rax, a), asAllocatable(b)));
422 return getLIRGen().emitMove(mulHigh.getHighResult());
423 }
424
425 @Override
426 public Value emitMulHigh(Value a, Value b) {
427 switch ((AMD64Kind) a.getPlatformKind()) {
428 case DWORD:
429 return emitMulHigh(AMD64MOp.IMUL, DWORD, a, b);
430 case QWORD:
431 return emitMulHigh(AMD64MOp.IMUL, QWORD, a, b);
432 default:
433 throw GraalError.shouldNotReachHere();
434 }
435 }
436
437 @Override
438 public Value emitUMulHigh(Value a, Value b) {
439 switch ((AMD64Kind) a.getPlatformKind()) {
440 case DWORD:
441 return emitMulHigh(AMD64MOp.MUL, DWORD, a, b);
442 case QWORD:
443 return emitMulHigh(AMD64MOp.MUL, QWORD, a, b);
444 default:
445 throw GraalError.shouldNotReachHere();
446 }
447 }
448
449 public Value emitBinaryMemory(AMD64RMOp op, OperandSize size, AllocatableValue a, AMD64AddressValue location, LIRFrameState state) {
450 Variable result = getLIRGen().newVariable(LIRKind.combine(a));
451 getLIRGen().append(new AMD64Binary.MemoryTwoOp(op, size, result, a, location, state));
452 return result;
453 }
454
455 protected Value emitConvertMemoryOp(PlatformKind kind, AMD64RMOp op, OperandSize size, AMD64AddressValue address, LIRFrameState state) {
456 Variable result = getLIRGen().newVariable(LIRKind.value(kind));
457 getLIRGen().append(new AMD64Unary.MemoryOp(op, size, result, address, state));
458 return result;
459 }
460
461 protected Value emitZeroExtendMemory(AMD64Kind memoryKind, int resultBits, AMD64AddressValue address, LIRFrameState state) {
462 // Issue a zero extending load of the proper bit size and set the result to
463 // the proper kind.
464 Variable result = getLIRGen().newVariable(LIRKind.value(resultBits <= 32 ? AMD64Kind.DWORD : AMD64Kind.QWORD));
465 switch (memoryKind) {
466 case BYTE:
467 getLIRGen().append(new AMD64Unary.MemoryOp(MOVZXB, DWORD, result, address, state));
468 break;
469 case WORD:
470 getLIRGen().append(new AMD64Unary.MemoryOp(MOVZX, DWORD, result, address, state));
471 break;
472 case DWORD:
473 getLIRGen().append(new AMD64Unary.MemoryOp(MOV, DWORD, result, address, state));
474 break;
475 case QWORD:
476 getLIRGen().append(new AMD64Unary.MemoryOp(MOV, QWORD, result, address, state));
477 break;
478 default:
479 throw GraalError.shouldNotReachHere();
480 }
481 return result;
482 }
483
484 private AMD64MulDivOp emitIDIV(OperandSize size, Value a, Value b, LIRFrameState state) {
485 LIRKind kind = LIRKind.combine(a, b);
486
487 AMD64SignExtendOp sx = getLIRGen().append(new AMD64SignExtendOp(size, kind, moveToReg(AMD64.rax, a)));
488 return getLIRGen().append(new AMD64MulDivOp(AMD64MOp.IDIV, size, kind, sx.getHighResult(), sx.getLowResult(), asAllocatable(b), state));
489 }
490
491 private AMD64MulDivOp emitDIV(OperandSize size, Value a, Value b, LIRFrameState state) {
492 LIRKind kind = LIRKind.combine(a, b);
493
494 RegisterValue rax = moveToReg(AMD64.rax, a);
495 RegisterValue rdx = AMD64.rdx.asValue(kind);
496 getLIRGen().append(new AMD64ClearRegisterOp(size, rdx));
497 return getLIRGen().append(new AMD64MulDivOp(AMD64MOp.DIV, size, kind, rdx, rax, asAllocatable(b), state));
498 }
499
500 public Value[] emitSignedDivRem(Value a, Value b, LIRFrameState state) {
501 AMD64MulDivOp op;
502 switch ((AMD64Kind) a.getPlatformKind()) {
503 case DWORD:
504 op = emitIDIV(DWORD, a, b, state);
505 break;
506 case QWORD:
507 op = emitIDIV(QWORD, a, b, state);
508 break;
509 default:
510 throw GraalError.shouldNotReachHere();
511 }
512 return new Value[]{getLIRGen().emitMove(op.getQuotient()), getLIRGen().emitMove(op.getRemainder())};
513 }
514
515 public Value[] emitUnsignedDivRem(Value a, Value b, LIRFrameState state) {
516 AMD64MulDivOp op;
517 switch ((AMD64Kind) a.getPlatformKind()) {
518 case DWORD:
519 op = emitDIV(DWORD, a, b, state);
520 break;
521 case QWORD:
522 op = emitDIV(QWORD, a, b, state);
523 break;
524 default:
525 throw GraalError.shouldNotReachHere();
526 }
527 return new Value[]{getLIRGen().emitMove(op.getQuotient()), getLIRGen().emitMove(op.getRemainder())};
528 }
529
530 @Override
531 public Value emitDiv(Value a, Value b, LIRFrameState state) {
532 boolean isAvx = supportAVX();
533 LIRKind resultKind = LIRKind.combine(a, b);
534 switch ((AMD64Kind) a.getPlatformKind()) {
535 case DWORD:
536 AMD64MulDivOp op = emitIDIV(DWORD, a, b, state);
537 return getLIRGen().emitMove(op.getQuotient());
538 case QWORD:
539 AMD64MulDivOp lop = emitIDIV(QWORD, a, b, state);
540 return getLIRGen().emitMove(lop.getQuotient());
541 case SINGLE:
542 if (isAvx) {
543 return emitBinary(resultKind, VDIVSS, a, b);
544 } else {
545 return emitBinary(resultKind, SSEOp.DIV, SS, false, a, b);
546 }
547 case DOUBLE:
548 if (isAvx) {
549 return emitBinary(resultKind, VDIVSD, a, b);
550 } else {
551 return emitBinary(resultKind, SSEOp.DIV, SD, false, a, b);
552 }
553 default:
554 throw GraalError.shouldNotReachHere();
555 }
556 }
557
558 @Override
559 public Value emitRem(Value a, Value b, LIRFrameState state) {
560 switch ((AMD64Kind) a.getPlatformKind()) {
561 case DWORD:
562 AMD64MulDivOp op = emitIDIV(DWORD, a, b, state);
563 return getLIRGen().emitMove(op.getRemainder());
564 case QWORD:
565 AMD64MulDivOp lop = emitIDIV(QWORD, a, b, state);
566 return getLIRGen().emitMove(lop.getRemainder());
567 case SINGLE: {
568 Variable result = getLIRGen().newVariable(LIRKind.combine(a, b));
569 getLIRGen().append(new FPDivRemOp(FREM, result, getLIRGen().load(a), getLIRGen().load(b)));
570 return result;
571 }
572 case DOUBLE: {
573 Variable result = getLIRGen().newVariable(LIRKind.combine(a, b));
574 getLIRGen().append(new FPDivRemOp(DREM, result, getLIRGen().load(a), getLIRGen().load(b)));
575 return result;
576 }
577 default:
578 throw GraalError.shouldNotReachHere();
579 }
580 }
581
582 @Override
583 public Variable emitUDiv(Value a, Value b, LIRFrameState state) {
584 AMD64MulDivOp op;
585 switch ((AMD64Kind) a.getPlatformKind()) {
586 case DWORD:
587 op = emitDIV(DWORD, a, b, state);
588 break;
589 case QWORD:
590 op = emitDIV(QWORD, a, b, state);
591 break;
592 default:
593 throw GraalError.shouldNotReachHere();
594 }
595 return getLIRGen().emitMove(op.getQuotient());
596 }
597
598 @Override
599 public Variable emitURem(Value a, Value b, LIRFrameState state) {
600 AMD64MulDivOp op;
601 switch ((AMD64Kind) a.getPlatformKind()) {
602 case DWORD:
603 op = emitDIV(DWORD, a, b, state);
604 break;
605 case QWORD:
606 op = emitDIV(QWORD, a, b, state);
607 break;
608 default:
609 throw GraalError.shouldNotReachHere();
610 }
611 return getLIRGen().emitMove(op.getRemainder());
612 }
613
614 @Override
615 public Variable emitAnd(Value a, Value b) {
616 LIRKind resultKind = LIRKind.combine(a, b);
617 switch ((AMD64Kind) a.getPlatformKind()) {
618 case DWORD:
619 return emitBinary(resultKind, AND, DWORD, true, a, b, false);
620 case QWORD:
621 return emitBinary(resultKind, AND, QWORD, true, a, b, false);
622 case SINGLE:
623 return emitBinary(resultKind, SSEOp.AND, PS, true, a, b);
624 case DOUBLE:
625 return emitBinary(resultKind, SSEOp.AND, PD, true, a, b);
626 default:
627 throw GraalError.shouldNotReachHere();
628 }
629 }
630
631 @Override
632 public Variable emitOr(Value a, Value b) {
633 boolean isAvx = supportAVX();
634 LIRKind resultKind = LIRKind.combine(a, b);
635 switch ((AMD64Kind) a.getPlatformKind()) {
636 case DWORD:
637 return emitBinary(resultKind, OR, DWORD, true, a, b, false);
638 case QWORD:
639 return emitBinary(resultKind, OR, QWORD, true, a, b, false);
640 case SINGLE:
641 if (isAvx) {
642 return emitBinary(resultKind, VORPS, a, b);
643 } else {
644 return emitBinary(resultKind, SSEOp.OR, PS, true, a, b);
645 }
646 case DOUBLE:
647 if (isAvx) {
648 return emitBinary(resultKind, VORPD, a, b);
649 } else {
650 return emitBinary(resultKind, SSEOp.OR, PD, true, a, b);
651 }
652 default:
653 throw GraalError.shouldNotReachHere();
654 }
655 }
656
657 @Override
658 public Variable emitXor(Value a, Value b) {
659 boolean isAvx = supportAVX();
660 LIRKind resultKind = LIRKind.combine(a, b);
661 switch ((AMD64Kind) a.getPlatformKind()) {
662 case DWORD:
663 return emitBinary(resultKind, XOR, DWORD, true, a, b, false);
664 case QWORD:
665 return emitBinary(resultKind, XOR, QWORD, true, a, b, false);
666 case SINGLE:
667 if (isAvx) {
668 return emitBinary(resultKind, VXORPS, a, b);
669 } else {
670 return emitBinary(resultKind, SSEOp.XOR, PS, true, a, b);
671 }
672 case DOUBLE:
673 if (isAvx) {
674 return emitBinary(resultKind, VXORPD, a, b);
675 } else {
676 return emitBinary(resultKind, SSEOp.XOR, PD, true, a, b);
677 }
678 default:
679 throw GraalError.shouldNotReachHere();
680 }
681 }
682
683 private Variable emitShift(AMD64Shift op, OperandSize size, Value a, Value b) {
684 Variable result = getLIRGen().newVariable(LIRKind.combine(a, b).changeType(a.getPlatformKind()));
685 AllocatableValue input = asAllocatable(a);
686 if (isJavaConstant(b)) {
687 JavaConstant c = asJavaConstant(b);
688 if (c.asLong() == 1) {
689 getLIRGen().append(new AMD64Unary.MOp(op.m1Op, size, result, input));
690 } else {
691 /*
692 * c needs to be masked here, because shifts with immediate expect a byte.
693 */
694 getLIRGen().append(new AMD64Binary.ConstOp(op.miOp, size, result, input, (byte) c.asLong()));
695 }
696 } else {
697 getLIRGen().emitMove(RCX_I, b);
698 getLIRGen().append(new AMD64ShiftOp(op.mcOp, size, result, input, RCX_I));
699 }
700 return result;
701 }
702
703 @Override
704 public Variable emitShl(Value a, Value b) {
705 switch ((AMD64Kind) a.getPlatformKind()) {
706 case DWORD:
707 return emitShift(SHL, DWORD, a, b);
708 case QWORD:
709 return emitShift(SHL, QWORD, a, b);
710 default:
711 throw GraalError.shouldNotReachHere();
712 }
713 }
714
715 @Override
716 public Variable emitShr(Value a, Value b) {
717 switch ((AMD64Kind) a.getPlatformKind()) {
718 case DWORD:
719 return emitShift(SAR, DWORD, a, b);
720 case QWORD:
721 return emitShift(SAR, QWORD, a, b);
722 default:
723 throw GraalError.shouldNotReachHere();
724 }
725 }
726
727 @Override
728 public Variable emitUShr(Value a, Value b) {
729 switch ((AMD64Kind) a.getPlatformKind()) {
730 case DWORD:
731 return emitShift(SHR, DWORD, a, b);
732 case QWORD:
733 return emitShift(SHR, QWORD, a, b);
734 default:
735 throw GraalError.shouldNotReachHere();
736 }
737 }
738
739 public Variable emitRol(Value a, Value b) {
740 switch ((AMD64Kind) a.getPlatformKind()) {
741 case DWORD:
742 return emitShift(ROL, DWORD, a, b);
743 case QWORD:
744 return emitShift(ROL, QWORD, a, b);
745 default:
746 throw GraalError.shouldNotReachHere();
747 }
748 }
749
750 @Override
751 public Variable emitRor(Value a, Value b) {
752 switch ((AMD64Kind) a.getPlatformKind()) {
753 case DWORD:
754 return emitShift(ROR, DWORD, a, b);
755 case QWORD:
756 return emitShift(ROR, QWORD, a, b);
757 default:
758 throw GraalError.shouldNotReachHere();
759 }
760 }
761
762 private AllocatableValue emitConvertOp(LIRKind kind, AMD64RMOp op, OperandSize size, Value input) {
763 Variable result = getLIRGen().newVariable(kind);
764 getLIRGen().append(new AMD64Unary.RMOp(op, size, result, asAllocatable(input)));
765 return result;
766 }
767
768 private AllocatableValue emitConvertOp(LIRKind kind, AMD64MROp op, OperandSize size, Value input) {
769 Variable result = getLIRGen().newVariable(kind);
770 getLIRGen().append(new AMD64Unary.MROp(op, size, result, asAllocatable(input)));
771 return result;
772 }
773
774 @Override
775 public Value emitReinterpret(LIRKind to, Value inputVal) {
776 ValueKind<?> from = inputVal.getValueKind();
777 if (to.equals(from)) {
778 return inputVal;
779 }
780
781 AllocatableValue input = asAllocatable(inputVal);
782 /*
783 * Conversions between integer to floating point types require moves between CPU and FPU
784 * registers.
785 */
786 AMD64Kind fromKind = (AMD64Kind) from.getPlatformKind();
787 switch ((AMD64Kind) to.getPlatformKind()) {
788 case DWORD:
789 switch (fromKind) {
790 case SINGLE:
791 return emitConvertOp(to, AMD64MROp.MOVD, DWORD, input);
792 }
793 break;
794 case QWORD:
795 switch (fromKind) {
796 case DOUBLE:
797 return emitConvertOp(to, AMD64MROp.MOVQ, QWORD, input);
798 }
799 break;
800 case SINGLE:
801 switch (fromKind) {
802 case DWORD:
803 return emitConvertOp(to, AMD64RMOp.MOVD, DWORD, input);
804 }
805 break;
806 case DOUBLE:
807 switch (fromKind) {
808 case QWORD:
809 return emitConvertOp(to, AMD64RMOp.MOVQ, QWORD, input);
810 }
811 break;
812 }
813 throw GraalError.shouldNotReachHere();
814 }
815
816 @Override
817 public Value emitFloatConvert(FloatConvert op, Value input) {
818 switch (op) {
819 case D2F:
820 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.SINGLE), SSEOp.CVTSD2SS, SD, input);
821 case D2I:
822 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.DWORD), SSEOp.CVTTSD2SI, DWORD, input);
823 case D2L:
824 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.QWORD), SSEOp.CVTTSD2SI, QWORD, input);
825 case F2D:
826 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.DOUBLE), SSEOp.CVTSS2SD, SS, input);
827 case F2I:
828 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.DWORD), SSEOp.CVTTSS2SI, DWORD, input);
829 case F2L:
830 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.QWORD), SSEOp.CVTTSS2SI, QWORD, input);
831 case I2D:
832 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.DOUBLE), SSEOp.CVTSI2SD, DWORD, input);
833 case I2F:
834 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.SINGLE), SSEOp.CVTSI2SS, DWORD, input);
835 case L2D:
836 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.DOUBLE), SSEOp.CVTSI2SD, QWORD, input);
837 case L2F:
838 return emitConvertOp(LIRKind.combine(input).changeType(AMD64Kind.SINGLE), SSEOp.CVTSI2SS, QWORD, input);
839 default:
840 throw GraalError.shouldNotReachHere();
841 }
842 }
843
844 @Override
845 public Value emitNarrow(Value inputVal, int bits) {
846 if (inputVal.getPlatformKind() == AMD64Kind.QWORD && bits <= 32) {
847 // TODO make it possible to reinterpret Long as Int in LIR without move
848 return emitConvertOp(LIRKind.combine(inputVal).changeType(AMD64Kind.DWORD), AMD64RMOp.MOV, DWORD, inputVal);
849 } else {
850 return inputVal;
851 }
852 }
853
854 @Override
855 public Value emitSignExtend(Value inputVal, int fromBits, int toBits) {
856 assert fromBits <= toBits && toBits <= 64;
857 if (fromBits == toBits) {
858 return inputVal;
859 } else if (toBits > 32) {
860 // sign extend to 64 bits
861 switch (fromBits) {
862 case 8:
863 return emitConvertOp(LIRKind.combine(inputVal).changeType(AMD64Kind.QWORD), MOVSXB, QWORD, inputVal);
864 case 16:
865 return emitConvertOp(LIRKind.combine(inputVal).changeType(AMD64Kind.QWORD), MOVSX, QWORD, inputVal);
866 case 32:
867 return emitConvertOp(LIRKind.combine(inputVal).changeType(AMD64Kind.QWORD), MOVSXD, QWORD, inputVal);
868 default:
869 throw GraalError.unimplemented("unsupported sign extension (" + fromBits + " bit -> " + toBits + " bit)");
870 }
871 } else {
872 // sign extend to 32 bits (smaller values are internally represented as 32 bit values)
873 switch (fromBits) {
874 case 8:
875 return emitConvertOp(LIRKind.combine(inputVal).changeType(AMD64Kind.DWORD), MOVSXB, DWORD, inputVal);
876 case 16:
877 return emitConvertOp(LIRKind.combine(inputVal).changeType(AMD64Kind.DWORD), MOVSX, DWORD, inputVal);
878 case 32:
879 return inputVal;
880 default:
881 throw GraalError.unimplemented("unsupported sign extension (" + fromBits + " bit -> " + toBits + " bit)");
882 }
883 }
884 }
885
886 @Override
887 public Value emitZeroExtend(Value inputVal, int fromBits, int toBits) {
888 assert fromBits <= toBits && toBits <= 64;
889 if (fromBits == toBits) {
890 return inputVal;
891 } else if (fromBits > 32) {
892 assert inputVal.getPlatformKind() == AMD64Kind.QWORD;
893 Variable result = getLIRGen().newVariable(LIRKind.combine(inputVal));
894 long mask = CodeUtil.mask(fromBits);
895 getLIRGen().append(new AMD64Binary.DataTwoOp(AND.getRMOpcode(QWORD), QWORD, result, asAllocatable(inputVal), JavaConstant.forLong(mask)));
896 return result;
897 } else {
898 LIRKind resultKind = LIRKind.combine(inputVal);
899 if (toBits > 32) {
900 resultKind = resultKind.changeType(AMD64Kind.QWORD);
901 } else {
902 resultKind = resultKind.changeType(AMD64Kind.DWORD);
903 }
904
905 /*
906 * Always emit DWORD operations, even if the resultKind is Long. On AMD64, all DWORD
907 * operations implicitly set the upper half of the register to 0, which is what we want
908 * anyway. Compared to the QWORD oparations, the encoding of the DWORD operations is
909 * sometimes one byte shorter.
910 */
911 switch (fromBits) {
912 case 8:
913 return emitConvertOp(resultKind, MOVZXB, DWORD, inputVal);
914 case 16:
915 return emitConvertOp(resultKind, MOVZX, DWORD, inputVal);
916 case 32:
917 return emitConvertOp(resultKind, MOV, DWORD, inputVal);
918 }
919
920 // odd bit count, fall back on manual masking
921 Variable result = getLIRGen().newVariable(resultKind);
922 JavaConstant mask;
923 if (toBits > 32) {
924 mask = JavaConstant.forLong(CodeUtil.mask(fromBits));
925 } else {
926 mask = JavaConstant.forInt((int) CodeUtil.mask(fromBits));
927 }
928 getLIRGen().append(new AMD64Binary.DataTwoOp(AND.getRMOpcode(DWORD), DWORD, result, asAllocatable(inputVal), mask));
929 return result;
930 }
931 }
932
933 @Override
934 public Variable emitBitCount(Value value) {
935 Variable result = getLIRGen().newVariable(LIRKind.combine(value).changeType(AMD64Kind.DWORD));
936 assert ((AMD64Kind) value.getPlatformKind()).isInteger();
937 if (value.getPlatformKind() == AMD64Kind.QWORD) {
938 getLIRGen().append(new AMD64Unary.RMOp(POPCNT, QWORD, result, asAllocatable(value)));
939 } else {
940 getLIRGen().append(new AMD64Unary.RMOp(POPCNT, DWORD, result, asAllocatable(value)));
941 }
942 return result;
943 }
944
945 @Override
946 public Variable emitBitScanForward(Value value) {
947 Variable result = getLIRGen().newVariable(LIRKind.combine(value).changeType(AMD64Kind.DWORD));
948 getLIRGen().append(new AMD64Unary.RMOp(BSF, QWORD, result, asAllocatable(value)));
949 return result;
950 }
951
952 @Override
953 public Variable emitBitScanReverse(Value value) {
954 Variable result = getLIRGen().newVariable(LIRKind.combine(value).changeType(AMD64Kind.DWORD));
955 assert ((AMD64Kind) value.getPlatformKind()).isInteger();
956 if (value.getPlatformKind() == AMD64Kind.QWORD) {
957 getLIRGen().append(new AMD64Unary.RMOp(BSR, QWORD, result, asAllocatable(value)));
958 } else {
959 getLIRGen().append(new AMD64Unary.RMOp(BSR, DWORD, result, asAllocatable(value)));
960 }
961 return result;
962 }
963
964 @Override
965 public Value emitCountLeadingZeros(Value value) {
966 Variable result = getLIRGen().newVariable(LIRKind.combine(value).changeType(AMD64Kind.DWORD));
967 assert ((AMD64Kind) value.getPlatformKind()).isInteger();
968 if (value.getPlatformKind() == AMD64Kind.QWORD) {
969 getLIRGen().append(new AMD64Unary.RMOp(LZCNT, QWORD, result, asAllocatable(value)));
970 } else {
971 getLIRGen().append(new AMD64Unary.RMOp(LZCNT, DWORD, result, asAllocatable(value)));
972 }
973 return result;
974 }
975
976 @Override
977 public Value emitCountTrailingZeros(Value value) {
978 Variable result = getLIRGen().newVariable(LIRKind.combine(value).changeType(AMD64Kind.DWORD));
979 assert ((AMD64Kind) value.getPlatformKind()).isInteger();
980 if (value.getPlatformKind() == AMD64Kind.QWORD) {
981 getLIRGen().append(new AMD64Unary.RMOp(TZCNT, QWORD, result, asAllocatable(value)));
982 } else {
983 getLIRGen().append(new AMD64Unary.RMOp(TZCNT, DWORD, result, asAllocatable(value)));
984 }
985 return result;
986 }
987
988 @Override
989 public Value emitLogicalAndNot(Value value1, Value value2) {
990 Variable result = getLIRGen().newVariable(LIRKind.combine(value1, value2));
991
992 if (value1.getPlatformKind() == AMD64Kind.QWORD) {
993 getLIRGen().append(new AMD64VectorBinary.AVXBinaryOp(VexGeneralPurposeRVMOp.ANDN, AVXSize.QWORD, result, asAllocatable(value1), asAllocatable(value2)));
994 } else {
995 getLIRGen().append(new AMD64VectorBinary.AVXBinaryOp(VexGeneralPurposeRVMOp.ANDN, AVXSize.DWORD, result, asAllocatable(value1), asAllocatable(value2)));
996 }
997 return result;
998 }
999
1000 @Override
1001 public Value emitLowestSetIsolatedBit(Value value) {
1002 Variable result = getLIRGen().newVariable(LIRKind.combine(value));
1003
1004 if (value.getPlatformKind() == AMD64Kind.QWORD) {
1005 getLIRGen().append(new AMD64VectorUnary.AVXUnaryOp(VexGeneralPurposeRMOp.BLSI, AVXSize.QWORD, result, asAllocatable(value)));
1006 } else {
1007 getLIRGen().append(new AMD64VectorUnary.AVXUnaryOp(VexGeneralPurposeRMOp.BLSI, AVXSize.DWORD, result, asAllocatable(value)));
1008 }
1009
1010 return result;
1011 }
1012
1013 @Override
1014 public Value emitGetMaskUpToLowestSetBit(Value value) {
1015 Variable result = getLIRGen().newVariable(LIRKind.combine(value));
1016
1017 if (value.getPlatformKind() == AMD64Kind.QWORD) {
1018 getLIRGen().append(new AMD64VectorUnary.AVXUnaryOp(VexGeneralPurposeRMOp.BLSMSK, AVXSize.QWORD, result, asAllocatable(value)));
1019 } else {
1020 getLIRGen().append(new AMD64VectorUnary.AVXUnaryOp(VexGeneralPurposeRMOp.BLSMSK, AVXSize.DWORD, result, asAllocatable(value)));
1021 }
1022
1023 return result;
1024 }
1025
1026 @Override
1027 public Value emitResetLowestSetBit(Value value) {
1028 Variable result = getLIRGen().newVariable(LIRKind.combine(value));
1029
1030 if (value.getPlatformKind() == AMD64Kind.QWORD) {
1031 getLIRGen().append(new AMD64VectorUnary.AVXUnaryOp(VexGeneralPurposeRMOp.BLSR, AVXSize.QWORD, result, asAllocatable(value)));
1032 } else {
1033 getLIRGen().append(new AMD64VectorUnary.AVXUnaryOp(VexGeneralPurposeRMOp.BLSR, AVXSize.DWORD, result, asAllocatable(value)));
1034 }
1035
1036 return result;
1037 }
1038
1039 @Override
1040 public Value emitMathAbs(Value input) {
1041 Variable result = getLIRGen().newVariable(LIRKind.combine(input));
1042 switch ((AMD64Kind) input.getPlatformKind()) {
1043 case SINGLE:
1044 getLIRGen().append(new AMD64Binary.DataTwoOp(SSEOp.AND, PS, result, asAllocatable(input), JavaConstant.forFloat(Float.intBitsToFloat(0x7FFFFFFF)), 16));
1045 break;
1046 case DOUBLE:
1047 getLIRGen().append(new AMD64Binary.DataTwoOp(SSEOp.AND, PD, result, asAllocatable(input), JavaConstant.forDouble(Double.longBitsToDouble(0x7FFFFFFFFFFFFFFFL)), 16));
1048 break;
1049 default:
1050 throw GraalError.shouldNotReachHere();
1051 }
1052 return result;
1053 }
1054
1055 @Override
1056 public Value emitMathSqrt(Value input) {
1057 Variable result = getLIRGen().newVariable(LIRKind.combine(input));
1058 switch ((AMD64Kind) input.getPlatformKind()) {
1059 case SINGLE:
1060 getLIRGen().append(new AMD64Unary.RMOp(SSEOp.SQRT, SS, result, asAllocatable(input)));
1061 break;
1062 case DOUBLE:
1063 getLIRGen().append(new AMD64Unary.RMOp(SSEOp.SQRT, SD, result, asAllocatable(input)));
1064 break;
1065 default:
1066 throw GraalError.shouldNotReachHere();
1067 }
1068 return result;
1069 }
1070
1071 @Override
1072 public Value emitMathLog(Value input, boolean base10) {
1073 if (base10) {
1074 return new AMD64MathLog10Op().emitLIRWrapper(getLIRGen(), input);
1075 } else {
1076 return new AMD64MathLogOp().emitLIRWrapper(getLIRGen(), input);
1077 }
1078 }
1079
1080 @Override
1081 public Value emitMathCos(Value input) {
1082 return new AMD64MathCosOp().emitLIRWrapper(getLIRGen(), input);
1083 }
1084
1085 @Override
1086 public Value emitMathSin(Value input) {
1087 return new AMD64MathSinOp().emitLIRWrapper(getLIRGen(), input);
1088 }
1089
1090 @Override
1091 public Value emitMathTan(Value input) {
1092 return new AMD64MathTanOp().emitLIRWrapper(getLIRGen(), input);
1093 }
1094
1095 @Override
1096 public Value emitMathExp(Value input) {
1097 return new AMD64MathExpOp().emitLIRWrapper(getLIRGen(), input);
1098 }
1099
1100 @Override
1101 public Value emitMathPow(Value x, Value y) {
1102 return new AMD64MathPowOp().emitLIRWrapper(getLIRGen(), x, y);
1103 }
1104
1105 protected AMD64LIRGenerator getAMD64LIRGen() {
1106 return (AMD64LIRGenerator) getLIRGen();
1107 }
1108
1109 @Override
1110 public Variable emitLoad(LIRKind kind, Value address, LIRFrameState state) {
1111 AMD64AddressValue loadAddress = getAMD64LIRGen().asAddressValue(address);
1112 Variable result = getLIRGen().newVariable(getLIRGen().toRegisterKind(kind));
1113 switch ((AMD64Kind) kind.getPlatformKind()) {
1114 case BYTE:
1115 getLIRGen().append(new AMD64Unary.MemoryOp(MOVSXB, DWORD, result, loadAddress, state));
1116 break;
1117 case WORD:
1118 getLIRGen().append(new AMD64Unary.MemoryOp(MOVSX, DWORD, result, loadAddress, state));
1119 break;
1120 case DWORD:
1121 getLIRGen().append(new AMD64Unary.MemoryOp(MOV, DWORD, result, loadAddress, state));
1122 break;
1123 case QWORD:
1124 getLIRGen().append(new AMD64Unary.MemoryOp(MOV, QWORD, result, loadAddress, state));
1125 break;
1126 case SINGLE:
1127 getLIRGen().append(new AMD64Unary.MemoryOp(MOVSS, SS, result, loadAddress, state));
1128 break;
1129 case DOUBLE:
1130 getLIRGen().append(new AMD64Unary.MemoryOp(MOVSD, SD, result, loadAddress, state));
1131 break;
1132 default:
1133 throw GraalError.shouldNotReachHere();
1134 }
1135 return result;
1136 }
1137
1138 protected void emitStoreConst(AMD64Kind kind, AMD64AddressValue address, ConstantValue value, LIRFrameState state) {
1139 Constant c = value.getConstant();
1140 if (JavaConstant.isNull(c)) {
1141 assert kind == AMD64Kind.DWORD || kind == AMD64Kind.QWORD;
1142 OperandSize size = kind == AMD64Kind.DWORD ? DWORD : QWORD;
1143 getLIRGen().append(new AMD64BinaryConsumer.MemoryConstOp(AMD64MIOp.MOV, size, address, 0, state));
1144 return;
1145 } else if (c instanceof VMConstant) {
1146 // only 32-bit constants can be patched
1147 if (kind == AMD64Kind.DWORD) {
1148 if (getLIRGen().target().inlineObjects || !(c instanceof JavaConstant)) {
1149 // if c is a JavaConstant, it's an oop, otherwise it's a metaspace constant
1150 assert !(c instanceof JavaConstant) || ((JavaConstant) c).getJavaKind() == JavaKind.Object;
1151 getLIRGen().append(new AMD64BinaryConsumer.MemoryVMConstOp(AMD64MIOp.MOV, address, (VMConstant) c, state));
1152 return;
1153 }
1154 }
1155 } else {
1156 JavaConstant jc = (JavaConstant) c;
1157 assert jc.getJavaKind().isPrimitive();
1158
1159 AMD64MIOp op = AMD64MIOp.MOV;
1160 OperandSize size;
1161 long imm;
1162
1163 switch (kind) {
1164 case BYTE:
1165 op = AMD64MIOp.MOVB;
1166 size = BYTE;
1167 imm = jc.asInt();
1168 break;
1169 case WORD:
1170 size = WORD;
1171 imm = jc.asInt();
1172 break;
1173 case DWORD:
1174 size = DWORD;
1175 imm = jc.asInt();
1176 break;
1177 case QWORD:
1178 size = QWORD;
1179 imm = jc.asLong();
1180 break;
1181 case SINGLE:
1182 size = DWORD;
1183 imm = Float.floatToRawIntBits(jc.asFloat());
1184 break;
1185 case DOUBLE:
1186 size = QWORD;
1187 imm = Double.doubleToRawLongBits(jc.asDouble());
1188 break;
1189 default:
1190 throw GraalError.shouldNotReachHere("unexpected kind " + kind);
1191 }
1192
1193 if (NumUtil.isInt(imm)) {
1194 getLIRGen().append(new AMD64BinaryConsumer.MemoryConstOp(op, size, address, (int) imm, state));
1195 return;
1196 }
1197 }
1198
1199 // fallback: load, then store
1200 emitStore(kind, address, asAllocatable(value), state);
1201 }
1202
1203 protected void emitStore(AMD64Kind kind, AMD64AddressValue address, AllocatableValue value, LIRFrameState state) {
1204 switch (kind) {
1205 case BYTE:
1206 getLIRGen().append(new AMD64BinaryConsumer.MemoryMROp(AMD64MROp.MOVB, BYTE, address, value, state));
1207 break;
1208 case WORD:
1209 getLIRGen().append(new AMD64BinaryConsumer.MemoryMROp(AMD64MROp.MOV, WORD, address, value, state));
1210 break;
1211 case DWORD:
1212 getLIRGen().append(new AMD64BinaryConsumer.MemoryMROp(AMD64MROp.MOV, DWORD, address, value, state));
1213 break;
1214 case QWORD:
1215 getLIRGen().append(new AMD64BinaryConsumer.MemoryMROp(AMD64MROp.MOV, QWORD, address, value, state));
1216 break;
1217 case SINGLE:
1218 getLIRGen().append(new AMD64BinaryConsumer.MemoryMROp(AMD64MROp.MOVSS, SS, address, value, state));
1219 break;
1220 case DOUBLE:
1221 getLIRGen().append(new AMD64BinaryConsumer.MemoryMROp(AMD64MROp.MOVSD, SD, address, value, state));
1222 break;
1223 default:
1224 throw GraalError.shouldNotReachHere();
1225 }
1226 }
1227
1228 @Override
1229 public void emitStore(ValueKind<?> lirKind, Value address, Value input, LIRFrameState state) {
1230 AMD64AddressValue storeAddress = getAMD64LIRGen().asAddressValue(address);
1231 AMD64Kind kind = (AMD64Kind) lirKind.getPlatformKind();
1232 if (isConstantValue(input)) {
1233 emitStoreConst(kind, storeAddress, asConstantValue(input), state);
1234 } else {
1235 emitStore(kind, storeAddress, asAllocatable(input), state);
1236 }
1237 }
1238
1239 private boolean mustReplaceNullWithNullRegister(Constant nullConstant) {
1240 /* Uncompressed null pointers only */
1241 return nullRegisterValue != null && JavaConstant.NULL_POINTER.equals(nullConstant);
1242 }
1243
1244 @Override
1245 public void emitCompareOp(AMD64Kind cmpKind, Variable left, Value right) {
1246 OperandSize size;
1247 switch (cmpKind) {
1248 case BYTE:
1249 size = BYTE;
1250 break;
1251 case WORD:
1252 size = WORD;
1253 break;
1254 case DWORD:
1255 size = DWORD;
1256 break;
1257 case QWORD:
1258 size = QWORD;
1259 break;
1260 case SINGLE:
1261 getLIRGen().append(new AMD64BinaryConsumer.Op(SSEOp.UCOMIS, PS, left, asAllocatable(right)));
1262 return;
1263 case DOUBLE:
1264 getLIRGen().append(new AMD64BinaryConsumer.Op(SSEOp.UCOMIS, PD, left, asAllocatable(right)));
1265 return;
1266 default:
1267 throw GraalError.shouldNotReachHere("unexpected kind: " + cmpKind);
1268 }
1269
1270 if (isConstantValue(right)) {
1271 Constant c = LIRValueUtil.asConstant(right);
1272 if (JavaConstant.isNull(c)) {
1273 if (mustReplaceNullWithNullRegister(c)) {
1274 getLIRGen().append(new AMD64BinaryConsumer.Op(AMD64RMOp.CMP, size, left, nullRegisterValue));
1275 } else {
1276 getLIRGen().append(new AMD64BinaryConsumer.Op(TEST, size, left, left));
1277 }
1278 return;
1279 } else if (c instanceof VMConstant) {
1280 VMConstant vc = (VMConstant) c;
1281 if (size == DWORD && !GeneratePIC.getValue(getOptions())) {
1282 getLIRGen().append(new AMD64BinaryConsumer.VMConstOp(CMP.getMIOpcode(DWORD, false), left, vc));
1283 } else {
1284 getLIRGen().append(new AMD64BinaryConsumer.DataOp(CMP.getRMOpcode(size), size, left, vc));
1285 }
1286 return;
1287 } else if (c instanceof JavaConstant) {
1288 JavaConstant jc = (JavaConstant) c;
1289 if (jc.isDefaultForKind()) {
1290 AMD64RMOp op = size == BYTE ? TESTB : TEST;
1291 getLIRGen().append(new AMD64BinaryConsumer.Op(op, size, left, left));
1292 return;
1293 } else if (NumUtil.is32bit(jc.asLong())) {
1294 getLIRGen().append(new AMD64BinaryConsumer.ConstOp(CMP, size, left, (int) jc.asLong()));
1295 return;
1296 }
1297 }
1298 }
1299
1300 // fallback: load, then compare
1301 getLIRGen().append(new AMD64BinaryConsumer.Op(CMP.getRMOpcode(size), size, left, asAllocatable(right)));
1302 }
1303
1304 @Override
1305 public Value emitRound(Value value, RoundingMode mode) {
1306 Variable result = getLIRGen().newVariable(LIRKind.combine(value));
1307 assert ((AMD64Kind) value.getPlatformKind()).isXMM();
1308 if (value.getPlatformKind() == AMD64Kind.SINGLE) {
1309 getLIRGen().append(new AMD64Binary.RMIOp(AMD64RMIOp.ROUNDSS, OperandSize.PD, result, asAllocatable(value), mode.encoding));
1310 } else {
1311 getLIRGen().append(new AMD64Binary.RMIOp(AMD64RMIOp.ROUNDSD, OperandSize.PD, result, asAllocatable(value), mode.encoding));
1312 }
1313 return result;
1314 }
1315
1316 private boolean supportAVX() {
1317 TargetDescription target = getLIRGen().target();
1318 return ((AMD64) target.arch).getFeatures().contains(CPUFeature.AVX);
1319 }
1320
1321 private static AVXSize getRegisterSize(Value a) {
1322 AMD64Kind kind = (AMD64Kind) a.getPlatformKind();
1323 if (kind.isXMM()) {
1324 return AVXKind.getRegisterSize(kind);
1325 } else {
1326 return AVXSize.XMM;
1327 }
1328 }
1329
1330 private Variable emitBinary(LIRKind resultKind, VexRVMOp op, Value a, Value b) {
1331 Variable result = getLIRGen().newVariable(resultKind);
1332 getLIRGen().append(new AVXBinaryOp(op, getRegisterSize(result), result, asAllocatable(a), asAllocatable(b)));
1333 return result;
1334 }
1335
1336 }