Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(241)

Side by Side Diff: src/IceTargetLoweringARM32.cpp

Issue 1655313002: Subzero: ARM32: lowering of vector insert and extract. (Closed) Base URL: https://chromium.googlesource.com/native_client/pnacl-subzero.git@master
Patch Set: Incorporating review feedback. Created 4 years, 10 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
OLDNEW
1 //===- subzero/src/IceTargetLoweringARM32.cpp - ARM32 lowering ------------===// 1 //===- subzero/src/IceTargetLoweringARM32.cpp - ARM32 lowering ------------===//
2 // 2 //
3 // The Subzero Code Generator 3 // The Subzero Code Generator
4 // 4 //
5 // This file is distributed under the University of Illinois Open Source 5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details. 6 // License. See LICENSE.TXT for details.
7 // 7 //
8 //===----------------------------------------------------------------------===// 8 //===----------------------------------------------------------------------===//
9 /// 9 ///
10 /// \file 10 /// \file
(...skipping 279 matching lines...) Expand 10 before | Expand all | Expand 10 after
290 CPUFeatures(Func->getContext()->getFlags()) {} 290 CPUFeatures(Func->getContext()->getFlags()) {}
291 291
292 void TargetARM32::staticInit(GlobalContext *Ctx) { 292 void TargetARM32::staticInit(GlobalContext *Ctx) {
293 293
294 // Limit this size (or do all bitsets need to be the same width)??? 294 // Limit this size (or do all bitsets need to be the same width)???
295 llvm::SmallBitVector IntegerRegisters(RegARM32::Reg_NUM); 295 llvm::SmallBitVector IntegerRegisters(RegARM32::Reg_NUM);
296 llvm::SmallBitVector I64PairRegisters(RegARM32::Reg_NUM); 296 llvm::SmallBitVector I64PairRegisters(RegARM32::Reg_NUM);
297 llvm::SmallBitVector Float32Registers(RegARM32::Reg_NUM); 297 llvm::SmallBitVector Float32Registers(RegARM32::Reg_NUM);
298 llvm::SmallBitVector Float64Registers(RegARM32::Reg_NUM); 298 llvm::SmallBitVector Float64Registers(RegARM32::Reg_NUM);
299 llvm::SmallBitVector VectorRegisters(RegARM32::Reg_NUM); 299 llvm::SmallBitVector VectorRegisters(RegARM32::Reg_NUM);
300 llvm::SmallBitVector QtoSRegisters(RegARM32::Reg_NUM);
300 llvm::SmallBitVector InvalidRegisters(RegARM32::Reg_NUM); 301 llvm::SmallBitVector InvalidRegisters(RegARM32::Reg_NUM);
301 for (int i = 0; i < RegARM32::Reg_NUM; ++i) { 302 for (int i = 0; i < RegARM32::Reg_NUM; ++i) {
302 const auto &Entry = RegARM32::RegTable[i]; 303 const auto &Entry = RegARM32::RegTable[i];
304 const auto EncodedReg_q8 = RegARM32::RegTable[RegARM32::Reg_q8].Encoding;
303 IntegerRegisters[i] = Entry.IsInt; 305 IntegerRegisters[i] = Entry.IsInt;
304 I64PairRegisters[i] = Entry.IsI64Pair; 306 I64PairRegisters[i] = Entry.IsI64Pair;
305 Float32Registers[i] = Entry.IsFP32; 307 Float32Registers[i] = Entry.IsFP32;
306 Float64Registers[i] = Entry.IsFP64; 308 Float64Registers[i] = Entry.IsFP64;
307 VectorRegisters[i] = Entry.IsVec128; 309 VectorRegisters[i] = Entry.IsVec128;
308 RegisterAliases[i].resize(RegARM32::Reg_NUM); 310 RegisterAliases[i].resize(RegARM32::Reg_NUM);
311 // TODO: It would be better to store a QtoS flag in the IceRegistersARM32
Jim Stichnoth 2016/02/08 18:08:29 TODO(person):
Eric Holk 2016/02/08 18:44:11 Done.
312 // table than to compare their encodings here.
313 QtoSRegisters[i] = Entry.IsVec128 && Entry.Encoding < EncodedReg_q8;
309 for (int j = 0; j < Entry.NumAliases; ++j) { 314 for (int j = 0; j < Entry.NumAliases; ++j) {
310 assert(i == j || !RegisterAliases[i][Entry.Aliases[j]]); 315 assert(i == j || !RegisterAliases[i][Entry.Aliases[j]]);
311 RegisterAliases[i].set(Entry.Aliases[j]); 316 RegisterAliases[i].set(Entry.Aliases[j]);
312 } 317 }
313 assert(RegisterAliases[i][i]); 318 assert(RegisterAliases[i][i]);
314 if (Entry.CCArg <= 0) { 319 if (Entry.CCArg <= 0) {
315 continue; 320 continue;
316 } 321 }
317 if (Entry.IsGPR) { 322 if (Entry.IsGPR) {
318 GPRArgInitializer[Entry.CCArg - 1] = i; 323 GPRArgInitializer[Entry.CCArg - 1] = i;
(...skipping 15 matching lines...) Expand all
334 TypeToRegisterSet[IceType_i64] = I64PairRegisters; 339 TypeToRegisterSet[IceType_i64] = I64PairRegisters;
335 TypeToRegisterSet[IceType_f32] = Float32Registers; 340 TypeToRegisterSet[IceType_f32] = Float32Registers;
336 TypeToRegisterSet[IceType_f64] = Float64Registers; 341 TypeToRegisterSet[IceType_f64] = Float64Registers;
337 TypeToRegisterSet[IceType_v4i1] = VectorRegisters; 342 TypeToRegisterSet[IceType_v4i1] = VectorRegisters;
338 TypeToRegisterSet[IceType_v8i1] = VectorRegisters; 343 TypeToRegisterSet[IceType_v8i1] = VectorRegisters;
339 TypeToRegisterSet[IceType_v16i1] = VectorRegisters; 344 TypeToRegisterSet[IceType_v16i1] = VectorRegisters;
340 TypeToRegisterSet[IceType_v16i8] = VectorRegisters; 345 TypeToRegisterSet[IceType_v16i8] = VectorRegisters;
341 TypeToRegisterSet[IceType_v8i16] = VectorRegisters; 346 TypeToRegisterSet[IceType_v8i16] = VectorRegisters;
342 TypeToRegisterSet[IceType_v4i32] = VectorRegisters; 347 TypeToRegisterSet[IceType_v4i32] = VectorRegisters;
343 TypeToRegisterSet[IceType_v4f32] = VectorRegisters; 348 TypeToRegisterSet[IceType_v4f32] = VectorRegisters;
349 TypeToRegisterSet[RegARM32::RCARM32_QtoS] = QtoSRegisters;
344 350
345 for (size_t i = 0; i < llvm::array_lengthof(TypeToRegisterSet); ++i) 351 for (size_t i = 0; i < llvm::array_lengthof(TypeToRegisterSet); ++i)
346 TypeToRegisterSetUnfiltered[i] = TypeToRegisterSet[i]; 352 TypeToRegisterSetUnfiltered[i] = TypeToRegisterSet[i];
347 353
348 filterTypeToRegisterSet( 354 filterTypeToRegisterSet(
349 Ctx, RegARM32::Reg_NUM, TypeToRegisterSet, 355 Ctx, RegARM32::Reg_NUM, TypeToRegisterSet,
350 llvm::array_lengthof(TypeToRegisterSet), [](int32_t RegNum) -> IceString { 356 llvm::array_lengthof(TypeToRegisterSet), [](int32_t RegNum) -> IceString {
351 // This function simply removes ", " from the register name. 357 // This function simply removes ", " from the register name.
352 IceString Name = RegARM32::getRegName(RegNum); 358 IceString Name = RegARM32::getRegName(RegNum);
353 constexpr const char RegSeparator[] = ", "; 359 constexpr const char RegSeparator[] = ", ";
(...skipping 3469 matching lines...) Expand 10 before | Expand all | Expand 10 after
3823 case IceType_v4i32: { 3829 case IceType_v4i32: {
3824 UnimplementedLoweringError(this, Inst); 3830 UnimplementedLoweringError(this, Inst);
3825 break; 3831 break;
3826 } 3832 }
3827 } 3833 }
3828 break; 3834 break;
3829 } 3835 }
3830 } 3836 }
3831 } 3837 }
3832 3838
3833 void TargetARM32::lowerExtractElement(const InstExtractElement *Inst) { 3839 void TargetARM32::lowerExtractElement(const InstExtractElement *Instr) {
3834 UnimplementedLoweringError(this, Inst); 3840 Variable *Dest = Instr->getDest();
3841 auto DestTy = Dest->getType();
3842
3843 Variable *Src0 = legalizeToReg(Instr->getSrc(0));
3844 Operand *Src1 = Instr->getSrc(1);
3845
3846 if (const auto *Imm = llvm::dyn_cast<ConstantInteger32>(Src1)) {
3847 const uint32_t Index = Imm->getValue();
3848 Variable *T = makeReg(DestTy);
3849 Variable *TSrc0 = makeReg(Src0->getType());
3850
3851 if (isFloatingType(DestTy)) {
3852 // We need to make sure the source is in a suitable register.
3853 TSrc0->setRegClass(RegARM32::RCARM32_QtoS);
3854 }
3855
3856 _mov(TSrc0, Src0);
3857 _extractelement(T, TSrc0, Index);
3858 _mov(Dest, T);
3859 return;
3860 }
3861 assert(false && "extractelement requires a constant index");
3835 } 3862 }
3836 3863
3837 namespace { 3864 namespace {
3838 // Validates FCMPARM32_TABLE's declaration w.r.t. InstFcmp::FCondition ordering 3865 // Validates FCMPARM32_TABLE's declaration w.r.t. InstFcmp::FCondition ordering
3839 // (and naming). 3866 // (and naming).
3840 enum { 3867 enum {
3841 #define X(val, CC0, CC1) _fcmp_ll_##val, 3868 #define X(val, CC0, CC1) _fcmp_ll_##val,
3842 FCMPARM32_TABLE 3869 FCMPARM32_TABLE
3843 #undef X 3870 #undef X
3844 _fcmp_ll_NUM 3871 _fcmp_ll_NUM
(...skipping 373 matching lines...) Expand 10 before | Expand all | Expand 10 after
4218 _mov(T, _0); 4245 _mov(T, _0);
4219 CondWhenTrue Cond = lowerIcmpCond(Inst); 4246 CondWhenTrue Cond = lowerIcmpCond(Inst);
4220 _mov_redefined(T, _1, Cond.WhenTrue0); 4247 _mov_redefined(T, _1, Cond.WhenTrue0);
4221 _mov(Dest, T); 4248 _mov(Dest, T);
4222 4249
4223 assert(Cond.WhenTrue1 == CondARM32::kNone); 4250 assert(Cond.WhenTrue1 == CondARM32::kNone);
4224 4251
4225 return; 4252 return;
4226 } 4253 }
4227 4254
4228 void TargetARM32::lowerInsertElement(const InstInsertElement *Inst) { 4255 void TargetARM32::lowerInsertElement(const InstInsertElement *Instr) {
4229 UnimplementedLoweringError(this, Inst); 4256 Variable *Dest = Instr->getDest();
4257 auto DestTy = Dest->getType();
4258
4259 Variable *Src0 = legalizeToReg(Instr->getSrc(0));
4260 Variable *Src1 = legalizeToReg(Instr->getSrc(1));
4261 Operand *Src2 = Instr->getSrc(2);
4262
4263 if (const auto *Imm = llvm::dyn_cast<ConstantInteger32>(Src2)) {
4264 auto Index = Imm->getValue();
4265 Variable *T = makeReg(DestTy);
4266
4267 if (isFloatingType(DestTy)) {
4268 T->setRegClass(RegARM32::RCARM32_QtoS);
4269 }
4270
4271 _mov(T, Src0);
4272 _insertelement(T, Src1, Index);
4273 _set_dest_redefined();
4274 _mov(Dest, T);
4275 return;
4276 }
4277 assert(false && "insertelement requires a constant index");
4230 } 4278 }
4231 4279
4232 namespace { 4280 namespace {
4233 inline uint64_t getConstantMemoryOrder(Operand *Opnd) { 4281 inline uint64_t getConstantMemoryOrder(Operand *Opnd) {
4234 if (auto *Integer = llvm::dyn_cast<ConstantInteger32>(Opnd)) 4282 if (auto *Integer = llvm::dyn_cast<ConstantInteger32>(Opnd))
4235 return Integer->getValue(); 4283 return Integer->getValue();
4236 return Intrinsics::MemoryOrderInvalid; 4284 return Intrinsics::MemoryOrderInvalid;
4237 } 4285 }
4238 } // end of anonymous namespace 4286 } // end of anonymous namespace
4239 4287
(...skipping 2276 matching lines...) Expand 10 before | Expand all | Expand 10 after
6516 Str << ".eabi_attribute 14, 3 @ Tag_ABI_PCS_R9_use: Not used\n"; 6564 Str << ".eabi_attribute 14, 3 @ Tag_ABI_PCS_R9_use: Not used\n";
6517 } 6565 }
6518 6566
6519 llvm::SmallBitVector TargetARM32::TypeToRegisterSet[RegARM32::RCARM32_NUM]; 6567 llvm::SmallBitVector TargetARM32::TypeToRegisterSet[RegARM32::RCARM32_NUM];
6520 llvm::SmallBitVector 6568 llvm::SmallBitVector
6521 TargetARM32::TypeToRegisterSetUnfiltered[RegARM32::RCARM32_NUM]; 6569 TargetARM32::TypeToRegisterSetUnfiltered[RegARM32::RCARM32_NUM];
6522 llvm::SmallBitVector TargetARM32::RegisterAliases[RegARM32::Reg_NUM]; 6570 llvm::SmallBitVector TargetARM32::RegisterAliases[RegARM32::Reg_NUM];
6523 6571
6524 } // end of namespace ARM32 6572 } // end of namespace ARM32
6525 } // end of namespace Ice 6573 } // end of namespace Ice
OLDNEW

Powered by Google App Engine
This is Rietveld 408576698