LLVM 19.0.0git
AMDGPUInstructionSelector.h
Go to the documentation of this file.
1//===- AMDGPUInstructionSelector --------------------------------*- C++ -*-==//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8/// \file
9/// This file declares the targeting of the InstructionSelector class for
10/// AMDGPU.
11//===----------------------------------------------------------------------===//
12
13#ifndef LLVM_LIB_TARGET_AMDGPU_AMDGPUINSTRUCTIONSELECTOR_H
14#define LLVM_LIB_TARGET_AMDGPU_AMDGPUINSTRUCTIONSELECTOR_H
15
16#include "SIDefines.h"
17#include "llvm/CodeGen/GlobalISel/InstructionSelector.h"
18#include "llvm/IR/InstrTypes.h"
19
20namespace {
21#define GET_GLOBALISEL_PREDICATE_BITSET
22#define AMDGPUSubtarget GCNSubtarget
23#include "AMDGPUGenGlobalISel.inc"
24#undef GET_GLOBALISEL_PREDICATE_BITSET
25#undef AMDGPUSubtarget
26}
27
28namespace llvm {
29
30namespace AMDGPU {
31struct ImageDimIntrinsicInfo;
32}
33
34class AMDGPURegisterBankInfo;
35class AMDGPUTargetMachine;
36class BlockFrequencyInfo;
37class ProfileSummaryInfo;
38class GCNSubtarget;
39class MachineInstr;
40class MachineIRBuilder;
41class MachineOperand;
42class MachineRegisterInfo;
43class RegisterBank;
44class SIInstrInfo;
45class SIRegisterInfo;
46class TargetRegisterClass;
47
48class AMDGPUInstructionSelector final : public InstructionSelector {
49private:
50 MachineRegisterInfo *MRI;
51 const GCNSubtarget *Subtarget;
52
53public:
54 AMDGPUInstructionSelector(const GCNSubtarget &STI,
55 const AMDGPURegisterBankInfo &RBI,
56 const AMDGPUTargetMachine &TM);
57
58 bool select(MachineInstr &I) override;
59 static const char *getName();
60
61 void setupMF(MachineFunction &MF, GISelKnownBits *KB,
62 CodeGenCoverage *CoverageInfo, ProfileSummaryInfo *PSI,
63 BlockFrequencyInfo *BFI) override;
64
65private:
66 struct GEPInfo {
67 SmallVector<unsigned, 2> SgprParts;
68 SmallVector<unsigned, 2> VgprParts;
69 int64_t Imm = 0;
70 };
71
72 bool isSGPR(Register Reg) const;
73
74 bool isInstrUniform(const MachineInstr &MI) const;
75 bool isVCC(Register Reg, const MachineRegisterInfo &MRI) const;
76
77 const RegisterBank *getArtifactRegBank(
78 Register Reg, const MachineRegisterInfo &MRI,
79 const TargetRegisterInfo &TRI) const;
80
81 /// tblgen-erated 'select' implementation.
82 bool selectImpl(MachineInstr &I, CodeGenCoverage &CoverageInfo) const;
83
84 MachineOperand getSubOperand64(MachineOperand &MO,
85 const TargetRegisterClass &SubRC,
86 unsigned SubIdx) const;
87
88 bool constrainCopyLikeIntrin(MachineInstr &MI, unsigned NewOpc) const;
89 bool selectCOPY(MachineInstr &I) const;
90 bool selectPHI(MachineInstr &I) const;
91 bool selectG_TRUNC(MachineInstr &I) const;
92 bool selectG_SZA_EXT(MachineInstr &I) const;
93 bool selectG_FPEXT(MachineInstr &I) const;
94 bool selectG_CONSTANT(MachineInstr &I) const;
95 bool selectG_FNEG(MachineInstr &I) const;
96 bool selectG_FABS(MachineInstr &I) const;
97 bool selectG_AND_OR_XOR(MachineInstr &I) const;
98 bool selectG_ADD_SUB(MachineInstr &I) const;
99 bool selectG_UADDO_USUBO_UADDE_USUBE(MachineInstr &I) const;
100 bool selectG_AMDGPU_MAD_64_32(MachineInstr &I) const;
101 bool selectG_EXTRACT(MachineInstr &I) const;
102 bool selectG_FMA_FMAD(MachineInstr &I) const;
103 bool selectG_MERGE_VALUES(MachineInstr &I) const;
104 bool selectG_UNMERGE_VALUES(MachineInstr &I) const;
105 bool selectG_BUILD_VECTOR(MachineInstr &I) const;
106 bool selectG_PTR_ADD(MachineInstr &I) const;
107 bool selectG_IMPLICIT_DEF(MachineInstr &I) const;
108 bool selectG_INSERT(MachineInstr &I) const;
109 bool selectG_SBFX_UBFX(MachineInstr &I) const;
110
111 bool selectInterpP1F16(MachineInstr &MI) const;
112 bool selectWritelane(MachineInstr &MI) const;
113 bool selectDivScale(MachineInstr &MI) const;
114 bool selectIntrinsicCmp(MachineInstr &MI) const;
115 bool selectBallot(MachineInstr &I) const;
116 bool selectInverseBallot(MachineInstr &I) const;
117 bool selectRelocConstant(MachineInstr &I) const;
118 bool selectGroupStaticSize(MachineInstr &I) const;
119 bool selectReturnAddress(MachineInstr &I) const;
120 bool selectG_INTRINSIC(MachineInstr &I) const;
121
122 bool selectEndCfIntrinsic(MachineInstr &MI) const;
123 bool selectDSOrderedIntrinsic(MachineInstr &MI, Intrinsic::ID IID) const;
124 bool selectDSGWSIntrinsic(MachineInstr &MI, Intrinsic::ID IID) const;
125 bool selectDSAppendConsume(MachineInstr &MI, bool IsAppend) const;
126 bool selectSBarrier(MachineInstr &MI) const;
127 bool selectDSBvhStackIntrinsic(MachineInstr &MI) const;
128 bool selectPOPSExitingWaveID(MachineInstr &MI) const;
129
130 bool selectImageIntrinsic(MachineInstr &MI,
131 const AMDGPU::ImageDimIntrinsicInfo *Intr) const;
132 bool selectG_INTRINSIC_W_SIDE_EFFECTS(MachineInstr &I) const;
133 int getS_CMPOpcode(CmpInst::Predicate P, unsigned Size) const;
134 bool selectG_ICMP_or_FCMP(MachineInstr &I) const;
135 bool hasVgprParts(ArrayRef<GEPInfo> AddrInfo) const;
136 void getAddrModeInfo(const MachineInstr &Load, const MachineRegisterInfo &MRI,
137 SmallVectorImpl<GEPInfo> &AddrInfo) const;
138
139 void initM0(MachineInstr &I) const;
140 bool selectG_LOAD_STORE_ATOMICRMW(MachineInstr &I) const;
141 bool selectG_SELECT(MachineInstr &I) const;
142 bool selectG_BRCOND(MachineInstr &I) const;
143 bool selectG_GLOBAL_VALUE(MachineInstr &I) const;
144 bool selectG_PTRMASK(MachineInstr &I) const;
145 bool selectG_EXTRACT_VECTOR_ELT(MachineInstr &I) const;
146 bool selectG_INSERT_VECTOR_ELT(MachineInstr &I) const;
147 bool selectBufferLoadLds(MachineInstr &MI) const;
148 bool selectGlobalLoadLds(MachineInstr &MI) const;
149 bool selectBVHIntrinsic(MachineInstr &I) const;
150 bool selectSMFMACIntrin(MachineInstr &I) const;
151 bool selectWaveAddress(MachineInstr &I) const;
152 bool selectStackRestore(MachineInstr &MI) const;
153 bool selectNamedBarrierInst(MachineInstr &I, Intrinsic::ID IID) const;
154 bool selectSBarrierSignalIsfirst(MachineInstr &I, Intrinsic::ID IID) const;
155 bool selectSBarrierLeave(MachineInstr &I) const;
156
157 std::pair<Register, unsigned> selectVOP3ModsImpl(MachineOperand &Root,
158 bool IsCanonicalizing = true,
159 bool AllowAbs = true,
160 bool OpSel = false) const;
161
162 Register copyToVGPRIfSrcFolded(Register Src, unsigned Mods,
163 MachineOperand Root, MachineInstr *InsertPt,
164 bool ForceVGPR = false) const;
165
166 InstructionSelector::ComplexRendererFns
167 selectVCSRC(MachineOperand &Root) const;
168
169 InstructionSelector::ComplexRendererFns
170 selectVSRC0(MachineOperand &Root) const;
171
172 InstructionSelector::ComplexRendererFns
173 selectVOP3Mods0(MachineOperand &Root) const;
174 InstructionSelector::ComplexRendererFns
175 selectVOP3BMods0(MachineOperand &Root) const;
176 InstructionSelector::ComplexRendererFns
177 selectVOP3OMods(MachineOperand &Root) const;
178 InstructionSelector::ComplexRendererFns
179 selectVOP3Mods(MachineOperand &Root) const;
180 InstructionSelector::ComplexRendererFns
181 selectVOP3ModsNonCanonicalizing(MachineOperand &Root) const;
182 InstructionSelector::ComplexRendererFns
183 selectVOP3BMods(MachineOperand &Root) const;
184
185 ComplexRendererFns selectVOP3NoMods(MachineOperand &Root) const;
186
187 std::pair<Register, unsigned>
188 selectVOP3PModsImpl(Register Src, const MachineRegisterInfo &MRI,
189 bool IsDOT = false) const;
190
191 InstructionSelector::ComplexRendererFns
192 selectVOP3PMods(MachineOperand &Root) const;
193
194 InstructionSelector::ComplexRendererFns
195 selectVOP3PModsDOT(MachineOperand &Root) const;
196
197 InstructionSelector::ComplexRendererFns
198 selectVOP3PModsNeg(MachineOperand &Root) const;
199
200 InstructionSelector::ComplexRendererFns
201 selectWMMAOpSelVOP3PMods(MachineOperand &Root) const;
202
203 InstructionSelector::ComplexRendererFns
204 selectWMMAModsF32NegAbs(MachineOperand &Root) const;
205 InstructionSelector::ComplexRendererFns
206 selectWMMAModsF16Neg(MachineOperand &Root) const;
207 InstructionSelector::ComplexRendererFns
208 selectWMMAModsF16NegAbs(MachineOperand &Root) const;
209 InstructionSelector::ComplexRendererFns
210 selectWMMAVISrc(MachineOperand &Root) const;
211 InstructionSelector::ComplexRendererFns
212 selectSWMMACIndex8(MachineOperand &Root) const;
213 InstructionSelector::ComplexRendererFns
214 selectSWMMACIndex16(MachineOperand &Root) const;
215
216 InstructionSelector::ComplexRendererFns
217 selectVOP3OpSelMods(MachineOperand &Root) const;
218
219 InstructionSelector::ComplexRendererFns
220 selectVINTERPMods(MachineOperand &Root) const;
221 InstructionSelector::ComplexRendererFns
222 selectVINTERPModsHi(MachineOperand &Root) const;
223
224 bool selectSmrdOffset(MachineOperand &Root, Register &Base, Register *SOffset,
225 int64_t *Offset) const;
226 InstructionSelector::ComplexRendererFns
227 selectSmrdImm(MachineOperand &Root) const;
228 InstructionSelector::ComplexRendererFns
229 selectSmrdImm32(MachineOperand &Root) const;
230 InstructionSelector::ComplexRendererFns
231 selectSmrdSgpr(MachineOperand &Root) const;
232 InstructionSelector::ComplexRendererFns
233 selectSmrdSgprImm(MachineOperand &Root) const;
234
235 std::pair<Register, int> selectFlatOffsetImpl(MachineOperand &Root,
236 uint64_t FlatVariant) const;
237
238 InstructionSelector::ComplexRendererFns
239 selectFlatOffset(MachineOperand &Root) const;
240 InstructionSelector::ComplexRendererFns
241 selectGlobalOffset(MachineOperand &Root) const;
242 InstructionSelector::ComplexRendererFns
243 selectScratchOffset(MachineOperand &Root) const;
244
245 InstructionSelector::ComplexRendererFns
246 selectGlobalSAddr(MachineOperand &Root) const;
247
248 InstructionSelector::ComplexRendererFns
249 selectScratchSAddr(MachineOperand &Root) const;
250 bool checkFlatScratchSVSSwizzleBug(Register VAddr, Register SAddr,
251 uint64_t ImmOffset) const;
252 InstructionSelector::ComplexRendererFns
253 selectScratchSVAddr(MachineOperand &Root) const;
254
255 InstructionSelector::ComplexRendererFns
256 selectMUBUFScratchOffen(MachineOperand &Root) const;
257 InstructionSelector::ComplexRendererFns
258 selectMUBUFScratchOffset(MachineOperand &Root) const;
259
260 bool isDSOffsetLegal(Register Base, int64_t Offset) const;
261 bool isDSOffset2Legal(Register Base, int64_t Offset0, int64_t Offset1,
262 unsigned Size) const;
263 bool isFlatScratchBaseLegal(Register Addr) const;
264 bool isFlatScratchBaseLegalSV(Register Addr) const;
265 bool isFlatScratchBaseLegalSVImm(Register Addr) const;
266
267 std::pair<Register, unsigned>
268 selectDS1Addr1OffsetImpl(MachineOperand &Root) const;
269 InstructionSelector::ComplexRendererFns
270 selectDS1Addr1Offset(MachineOperand &Root) const;
271
272 InstructionSelector::ComplexRendererFns
273 selectDS64Bit4ByteAligned(MachineOperand &Root) const;
274
275 InstructionSelector::ComplexRendererFns
276 selectDS128Bit8ByteAligned(MachineOperand &Root) const;
277
278 std::pair<Register, unsigned> selectDSReadWrite2Impl(MachineOperand &Root,
279 unsigned size) const;
280 InstructionSelector::ComplexRendererFns
281 selectDSReadWrite2(MachineOperand &Root, unsigned size) const;
282
283 std::pair<Register, int64_t>
284 getPtrBaseWithConstantOffset(Register Root,
285 const MachineRegisterInfo &MRI) const;
286
287 // Parse out a chain of up to two g_ptr_add instructions.
288 // g_ptr_add (n0, _)
289 // g_ptr_add (n0, (n1 = g_ptr_add n2, n3))
290 struct MUBUFAddressData {
291 Register N0, N2, N3;
292 int64_t Offset = 0;
293 };
294
295 bool shouldUseAddr64(MUBUFAddressData AddrData) const;
296
297 void splitIllegalMUBUFOffset(MachineIRBuilder &B,
298 Register &SOffset, int64_t &ImmOffset) const;
299
300 MUBUFAddressData parseMUBUFAddress(Register Src) const;
301
302 bool selectMUBUFAddr64Impl(MachineOperand &Root, Register &VAddr,
303 Register &RSrcReg, Register &SOffset,
304 int64_t &Offset) const;
305
306 bool selectMUBUFOffsetImpl(MachineOperand &Root, Register &RSrcReg,
307 Register &SOffset, int64_t &Offset) const;
308
309 InstructionSelector::ComplexRendererFns
310 selectBUFSOffset(MachineOperand &Root) const;
311
312 InstructionSelector::ComplexRendererFns
313 selectMUBUFAddr64(MachineOperand &Root) const;
314
315 InstructionSelector::ComplexRendererFns
316 selectMUBUFOffset(MachineOperand &Root) const;
317
318 ComplexRendererFns selectSMRDBufferImm(MachineOperand &Root) const;
319 ComplexRendererFns selectSMRDBufferImm32(MachineOperand &Root) const;
320 ComplexRendererFns selectSMRDBufferSgprImm(MachineOperand &Root) const;
321
322 std::pair<Register, unsigned> selectVOP3PMadMixModsImpl(MachineOperand &Root,
323 bool &Matched) const;
324 ComplexRendererFns selectVOP3PMadMixModsExt(MachineOperand &Root) const;
325 ComplexRendererFns selectVOP3PMadMixMods(MachineOperand &Root) const;
326
327 void renderTruncImm32(MachineInstrBuilder &MIB, const MachineInstr &MI,
328 int OpIdx = -1) const;
329
330 void renderTruncTImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
331 int OpIdx) const;
332
333 void renderOpSelTImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
334 int OpIdx) const;
335
336 void renderNegateImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
337 int OpIdx) const;
338
339 void renderBitcastImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
340 int OpIdx) const;
341
342 void renderPopcntImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
343 int OpIdx) const;
344 void renderExtractCPol(MachineInstrBuilder &MIB, const MachineInstr &MI,
345 int OpIdx) const;
346 void renderExtractSWZ(MachineInstrBuilder &MIB, const MachineInstr &MI,
347 int OpIdx) const;
348 void renderExtractCpolSetGLC(MachineInstrBuilder &MIB, const MachineInstr &MI,
349 int OpIdx) const;
350
351 void renderFrameIndex(MachineInstrBuilder &MIB, const MachineInstr &MI,
352 int OpIdx) const;
353
354 void renderFPPow2ToExponent(MachineInstrBuilder &MIB, const MachineInstr &MI,
355 int OpIdx) const;
356
357 bool isInlineImmediate(const APInt &Imm) const;
358 bool isInlineImmediate(const APFloat &Imm) const;
359
360 // Returns true if TargetOpcode::G_AND MachineInstr `MI`'s masking of the
361 // shift amount operand's `ShAmtBits` bits is unneeded.
362 bool isUnneededShiftMask(const MachineInstr &MI, unsigned ShAmtBits) const;
363
364 const SIInstrInfo &TII;
365 const SIRegisterInfo &TRI;
366 const AMDGPURegisterBankInfo &RBI;
367 const AMDGPUTargetMachine &TM;
368 const GCNSubtarget &STI;
369 bool EnableLateStructurizeCFG;
370#define GET_GLOBALISEL_PREDICATES_DECL
371#define AMDGPUSubtarget GCNSubtarget
372#include "AMDGPUGenGlobalISel.inc"
373#undef GET_GLOBALISEL_PREDICATES_DECL
374#undef AMDGPUSubtarget
375
376#define GET_GLOBALISEL_TEMPORARIES_DECL
377#include "AMDGPUGenGlobalISel.inc"
378#undef GET_GLOBALISEL_TEMPORARIES_DECL
379};
380
381} // End llvm namespace.
382#endif
Intr
unsigned Intr
Definition: AMDGPUBaseInfo.cpp:2887
B
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
Addr
uint64_t Addr
Definition: ELFObjHandler.cpp:79
Size
uint64_t Size
Definition: ELFObjHandler.cpp:81
MI
IRTranslator LLVM IR MI
Definition: IRTranslator.cpp:113
I
#define I(x, y, z)
Definition: MD5.cpp:58
Reg
unsigned Reg
Definition: MachineSink.cpp:1874
P
#define P(N)
TM
const char LLVMTargetMachineRef TM
Definition: PassBuilderBindings.cpp:47
llvm::AMDGPUInstructionSelector::getName
static const char * getName()
llvm::AMDGPUInstructionSelector::select
bool select(MachineInstr &I) override
Select the (possibly generic) instruction I to only use target-specific opcodes.
Definition: AMDGPUInstructionSelector.cpp:3533
llvm::AMDGPUInstructionSelector::setupMF
void setupMF(MachineFunction &MF, GISelKnownBits *KB, CodeGenCoverage *CoverageInfo, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI) override
Setup per-MF executor state.
Definition: AMDGPUInstructionSelector.cpp:60
llvm::ArrayRef
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory),...
Definition: ArrayRef.h:41
llvm::BlockFrequencyInfo
BlockFrequencyInfo pass uses BlockFrequencyInfoImpl implementation to estimate IR basic block frequen...
Definition: BlockFrequencyInfo.h:38
llvm::CmpInst::Predicate
Predicate
This enumeration lists the possible predicates for CmpInst subclasses.
Definition: InstrTypes.h:993
llvm::GIMatchTableExecutor::ComplexRendererFns
std::optional< SmallVector< std::function< void(MachineInstrBuilder &)>, 4 > > ComplexRendererFns
Definition: GIMatchTableExecutor.h:597
llvm::MachineInstr
Representation of each machine instruction.
Definition: MachineInstr.h:69
llvm::MachineOperand
MachineOperand class - Representation of each machine instruction operand.
Definition: MachineOperand.h:48
llvm::MachineRegisterInfo
MachineRegisterInfo - Keep track of information for virtual and physical registers,...
Definition: MachineRegisterInfo.h:51
llvm::ProfileSummaryInfo
Analysis providing profile information.
Definition: ProfileSummaryInfo.h:42
llvm::RegisterBank
This class implements the register bank concept.
Definition: RegisterBank.h:28
llvm::Register
Wrapper class representing virtual and physical registers.
Definition: Register.h:19
llvm::SmallVectorImpl
This class consists of common code factored out of the SmallVector class to reduce code duplication b...
Definition: SmallVector.h:586
llvm::SmallVector
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
Definition: SmallVector.h:1209
llvm::TargetRegisterInfo
TargetRegisterInfo base class - We assume that the target defines a static array of TargetRegisterDes...
Definition: TargetRegisterInfo.h:238
uint64_t
llvm
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
llvm::Offset
@ Offset
Definition: DWP.cpp:456
Generated on Thu May 30 2024 23:53:05 for LLVM by doxygen 1.9.6