Bug Summary

File:llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
Warning:line 511, column 18
Called C++ object pointer is null

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name PPCISelDAGToDAG.cpp -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=cplusplus -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -setup-static-analyzer -analyzer-config-compatibility-mode=true -mrelocation-model pic -pic-level 2 -mframe-pointer=none -fmath-errno -fno-rounding-math -mconstructor-aliases -munwind-tables -target-cpu x86-64 -tune-cpu generic -debugger-tuning=gdb -ffunction-sections -fdata-sections -fcoverage-compilation-dir=/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/build-llvm/lib/Target/PowerPC -resource-dir /usr/lib/llvm-13/lib/clang/13.0.0 -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/build-llvm/lib/Target/PowerPC -I /build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC -I /build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/build-llvm/include -I /build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include -D NDEBUG -U NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/x86_64-linux-gnu/c++/10 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10/backward -internal-isystem /usr/lib/llvm-13/lib/clang/13.0.0/include -internal-isystem /usr/local/include -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../x86_64-linux-gnu/include -internal-externc-isystem /usr/include/x86_64-linux-gnu -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -Wno-unused-parameter -Wwrite-strings -Wno-missing-field-initializers -Wno-long-long -Wno-maybe-uninitialized -Wno-class-memaccess -Wno-redundant-move -Wno-pessimizing-move -Wno-noexcept-type -Wno-comment -std=c++14 -fdeprecated-macro -fdebug-compilation-dir=/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/build-llvm/lib/Target/PowerPC -fdebug-prefix-map=/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367=. -ferror-limit 19 -fvisibility hidden -fvisibility-inlines-hidden -stack-protector 2 -fgnuc-version=4.2.1 -vectorize-loops -vectorize-slp -analyzer-output=html -analyzer-config stable-report-filename=true -faddrsig -D__GCC_HAVE_DWARF2_CFI_ASM=1 -o /tmp/scan-build-2021-07-11-231758-18690-1 -x c++ /build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp

/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp

1//===-- PPCISelDAGToDAG.cpp - PPC --pattern matching inst selector --------===//
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//
9// This file defines a pattern matching instruction selector for PowerPC,
10// converting from a legalized dag to a PPC dag.
11//
12//===----------------------------------------------------------------------===//
13
14#include "MCTargetDesc/PPCMCTargetDesc.h"
15#include "MCTargetDesc/PPCPredicates.h"
16#include "PPC.h"
17#include "PPCISelLowering.h"
18#include "PPCMachineFunctionInfo.h"
19#include "PPCSubtarget.h"
20#include "PPCTargetMachine.h"
21#include "llvm/ADT/APInt.h"
22#include "llvm/ADT/DenseMap.h"
23#include "llvm/ADT/STLExtras.h"
24#include "llvm/ADT/SmallPtrSet.h"
25#include "llvm/ADT/SmallVector.h"
26#include "llvm/ADT/Statistic.h"
27#include "llvm/Analysis/BranchProbabilityInfo.h"
28#include "llvm/CodeGen/FunctionLoweringInfo.h"
29#include "llvm/CodeGen/ISDOpcodes.h"
30#include "llvm/CodeGen/MachineBasicBlock.h"
31#include "llvm/CodeGen/MachineFunction.h"
32#include "llvm/CodeGen/MachineInstrBuilder.h"
33#include "llvm/CodeGen/MachineRegisterInfo.h"
34#include "llvm/CodeGen/SelectionDAG.h"
35#include "llvm/CodeGen/SelectionDAGISel.h"
36#include "llvm/CodeGen/SelectionDAGNodes.h"
37#include "llvm/CodeGen/TargetInstrInfo.h"
38#include "llvm/CodeGen/TargetRegisterInfo.h"
39#include "llvm/CodeGen/ValueTypes.h"
40#include "llvm/IR/BasicBlock.h"
41#include "llvm/IR/DebugLoc.h"
42#include "llvm/IR/Function.h"
43#include "llvm/IR/GlobalValue.h"
44#include "llvm/IR/InlineAsm.h"
45#include "llvm/IR/InstrTypes.h"
46#include "llvm/IR/IntrinsicsPowerPC.h"
47#include "llvm/IR/Module.h"
48#include "llvm/Support/Casting.h"
49#include "llvm/Support/CodeGen.h"
50#include "llvm/Support/CommandLine.h"
51#include "llvm/Support/Compiler.h"
52#include "llvm/Support/Debug.h"
53#include "llvm/Support/ErrorHandling.h"
54#include "llvm/Support/KnownBits.h"
55#include "llvm/Support/MachineValueType.h"
56#include "llvm/Support/MathExtras.h"
57#include "llvm/Support/raw_ostream.h"
58#include <algorithm>
59#include <cassert>
60#include <cstdint>
61#include <iterator>
62#include <limits>
63#include <memory>
64#include <new>
65#include <tuple>
66#include <utility>
67
68using namespace llvm;
69
70#define DEBUG_TYPE"ppc-codegen" "ppc-codegen"
71
72STATISTIC(NumSextSetcc,static llvm::Statistic NumSextSetcc = {"ppc-codegen", "NumSextSetcc"
, "Number of (sext(setcc)) nodes expanded into GPR sequence."
}
73 "Number of (sext(setcc)) nodes expanded into GPR sequence.")static llvm::Statistic NumSextSetcc = {"ppc-codegen", "NumSextSetcc"
, "Number of (sext(setcc)) nodes expanded into GPR sequence."
}
;
74STATISTIC(NumZextSetcc,static llvm::Statistic NumZextSetcc = {"ppc-codegen", "NumZextSetcc"
, "Number of (zext(setcc)) nodes expanded into GPR sequence."
}
75 "Number of (zext(setcc)) nodes expanded into GPR sequence.")static llvm::Statistic NumZextSetcc = {"ppc-codegen", "NumZextSetcc"
, "Number of (zext(setcc)) nodes expanded into GPR sequence."
}
;
76STATISTIC(SignExtensionsAdded,static llvm::Statistic SignExtensionsAdded = {"ppc-codegen", "SignExtensionsAdded"
, "Number of sign extensions for compare inputs added."}
77 "Number of sign extensions for compare inputs added.")static llvm::Statistic SignExtensionsAdded = {"ppc-codegen", "SignExtensionsAdded"
, "Number of sign extensions for compare inputs added."}
;
78STATISTIC(ZeroExtensionsAdded,static llvm::Statistic ZeroExtensionsAdded = {"ppc-codegen", "ZeroExtensionsAdded"
, "Number of zero extensions for compare inputs added."}
79 "Number of zero extensions for compare inputs added.")static llvm::Statistic ZeroExtensionsAdded = {"ppc-codegen", "ZeroExtensionsAdded"
, "Number of zero extensions for compare inputs added."}
;
80STATISTIC(NumLogicOpsOnComparison,static llvm::Statistic NumLogicOpsOnComparison = {"ppc-codegen"
, "NumLogicOpsOnComparison", "Number of logical ops on i1 values calculated in GPR."
}
81 "Number of logical ops on i1 values calculated in GPR.")static llvm::Statistic NumLogicOpsOnComparison = {"ppc-codegen"
, "NumLogicOpsOnComparison", "Number of logical ops on i1 values calculated in GPR."
}
;
82STATISTIC(OmittedForNonExtendUses,static llvm::Statistic OmittedForNonExtendUses = {"ppc-codegen"
, "OmittedForNonExtendUses", "Number of compares not eliminated as they have non-extending uses."
}
83 "Number of compares not eliminated as they have non-extending uses.")static llvm::Statistic OmittedForNonExtendUses = {"ppc-codegen"
, "OmittedForNonExtendUses", "Number of compares not eliminated as they have non-extending uses."
}
;
84STATISTIC(NumP9Setb,static llvm::Statistic NumP9Setb = {"ppc-codegen", "NumP9Setb"
, "Number of compares lowered to setb."}
85 "Number of compares lowered to setb.")static llvm::Statistic NumP9Setb = {"ppc-codegen", "NumP9Setb"
, "Number of compares lowered to setb."}
;
86
87// FIXME: Remove this once the bug has been fixed!
88cl::opt<bool> ANDIGlueBug("expose-ppc-andi-glue-bug",
89cl::desc("expose the ANDI glue bug on PPC"), cl::Hidden);
90
91static cl::opt<bool>
92 UseBitPermRewriter("ppc-use-bit-perm-rewriter", cl::init(true),
93 cl::desc("use aggressive ppc isel for bit permutations"),
94 cl::Hidden);
95static cl::opt<bool> BPermRewriterNoMasking(
96 "ppc-bit-perm-rewriter-stress-rotates",
97 cl::desc("stress rotate selection in aggressive ppc isel for "
98 "bit permutations"),
99 cl::Hidden);
100
101static cl::opt<bool> EnableBranchHint(
102 "ppc-use-branch-hint", cl::init(true),
103 cl::desc("Enable static hinting of branches on ppc"),
104 cl::Hidden);
105
106static cl::opt<bool> EnableTLSOpt(
107 "ppc-tls-opt", cl::init(true),
108 cl::desc("Enable tls optimization peephole"),
109 cl::Hidden);
110
111enum ICmpInGPRType { ICGPR_All, ICGPR_None, ICGPR_I32, ICGPR_I64,
112 ICGPR_NonExtIn, ICGPR_Zext, ICGPR_Sext, ICGPR_ZextI32,
113 ICGPR_SextI32, ICGPR_ZextI64, ICGPR_SextI64 };
114
115static cl::opt<ICmpInGPRType> CmpInGPR(
116 "ppc-gpr-icmps", cl::Hidden, cl::init(ICGPR_All),
117 cl::desc("Specify the types of comparisons to emit GPR-only code for."),
118 cl::values(clEnumValN(ICGPR_None, "none", "Do not modify integer comparisons.")llvm::cl::OptionEnumValue { "none", int(ICGPR_None), "Do not modify integer comparisons."
}
,
119 clEnumValN(ICGPR_All, "all", "All possible int comparisons in GPRs.")llvm::cl::OptionEnumValue { "all", int(ICGPR_All), "All possible int comparisons in GPRs."
}
,
120 clEnumValN(ICGPR_I32, "i32", "Only i32 comparisons in GPRs.")llvm::cl::OptionEnumValue { "i32", int(ICGPR_I32), "Only i32 comparisons in GPRs."
}
,
121 clEnumValN(ICGPR_I64, "i64", "Only i64 comparisons in GPRs.")llvm::cl::OptionEnumValue { "i64", int(ICGPR_I64), "Only i64 comparisons in GPRs."
}
,
122 clEnumValN(ICGPR_NonExtIn, "nonextin",llvm::cl::OptionEnumValue { "nonextin", int(ICGPR_NonExtIn), "Only comparisons where inputs don't need [sz]ext."
}
123 "Only comparisons where inputs don't need [sz]ext.")llvm::cl::OptionEnumValue { "nonextin", int(ICGPR_NonExtIn), "Only comparisons where inputs don't need [sz]ext."
}
,
124 clEnumValN(ICGPR_Zext, "zext", "Only comparisons with zext result.")llvm::cl::OptionEnumValue { "zext", int(ICGPR_Zext), "Only comparisons with zext result."
}
,
125 clEnumValN(ICGPR_ZextI32, "zexti32",llvm::cl::OptionEnumValue { "zexti32", int(ICGPR_ZextI32), "Only i32 comparisons with zext result."
}
126 "Only i32 comparisons with zext result.")llvm::cl::OptionEnumValue { "zexti32", int(ICGPR_ZextI32), "Only i32 comparisons with zext result."
}
,
127 clEnumValN(ICGPR_ZextI64, "zexti64",llvm::cl::OptionEnumValue { "zexti64", int(ICGPR_ZextI64), "Only i64 comparisons with zext result."
}
128 "Only i64 comparisons with zext result.")llvm::cl::OptionEnumValue { "zexti64", int(ICGPR_ZextI64), "Only i64 comparisons with zext result."
}
,
129 clEnumValN(ICGPR_Sext, "sext", "Only comparisons with sext result.")llvm::cl::OptionEnumValue { "sext", int(ICGPR_Sext), "Only comparisons with sext result."
}
,
130 clEnumValN(ICGPR_SextI32, "sexti32",llvm::cl::OptionEnumValue { "sexti32", int(ICGPR_SextI32), "Only i32 comparisons with sext result."
}
131 "Only i32 comparisons with sext result.")llvm::cl::OptionEnumValue { "sexti32", int(ICGPR_SextI32), "Only i32 comparisons with sext result."
}
,
132 clEnumValN(ICGPR_SextI64, "sexti64",llvm::cl::OptionEnumValue { "sexti64", int(ICGPR_SextI64), "Only i64 comparisons with sext result."
}
133 "Only i64 comparisons with sext result.")llvm::cl::OptionEnumValue { "sexti64", int(ICGPR_SextI64), "Only i64 comparisons with sext result."
}
));
134namespace {
135
136 //===--------------------------------------------------------------------===//
137 /// PPCDAGToDAGISel - PPC specific code to select PPC machine
138 /// instructions for SelectionDAG operations.
139 ///
140 class PPCDAGToDAGISel : public SelectionDAGISel {
141 const PPCTargetMachine &TM;
142 const PPCSubtarget *Subtarget = nullptr;
143 const PPCTargetLowering *PPCLowering = nullptr;
144 unsigned GlobalBaseReg = 0;
145
146 public:
147 explicit PPCDAGToDAGISel(PPCTargetMachine &tm, CodeGenOpt::Level OptLevel)
148 : SelectionDAGISel(tm, OptLevel), TM(tm) {}
149
150 bool runOnMachineFunction(MachineFunction &MF) override {
151 // Make sure we re-emit a set of the global base reg if necessary
152 GlobalBaseReg = 0;
153 Subtarget = &MF.getSubtarget<PPCSubtarget>();
154 PPCLowering = Subtarget->getTargetLowering();
155 if (Subtarget->hasROPProtect()) {
156 // Create a place on the stack for the ROP Protection Hash.
157 // The ROP Protection Hash will always be 8 bytes and aligned to 8
158 // bytes.
159 MachineFrameInfo &MFI = MF.getFrameInfo();
160 PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
161 const int Result = MFI.CreateStackObject(8, Align(8), false);
162 FI->setROPProtectionHashSaveIndex(Result);
163 }
164 SelectionDAGISel::runOnMachineFunction(MF);
165
166 return true;
167 }
168
169 void PreprocessISelDAG() override;
170 void PostprocessISelDAG() override;
171
172 /// getI16Imm - Return a target constant with the specified value, of type
173 /// i16.
174 inline SDValue getI16Imm(unsigned Imm, const SDLoc &dl) {
175 return CurDAG->getTargetConstant(Imm, dl, MVT::i16);
176 }
177
178 /// getI32Imm - Return a target constant with the specified value, of type
179 /// i32.
180 inline SDValue getI32Imm(unsigned Imm, const SDLoc &dl) {
181 return CurDAG->getTargetConstant(Imm, dl, MVT::i32);
182 }
183
184 /// getI64Imm - Return a target constant with the specified value, of type
185 /// i64.
186 inline SDValue getI64Imm(uint64_t Imm, const SDLoc &dl) {
187 return CurDAG->getTargetConstant(Imm, dl, MVT::i64);
188 }
189
190 /// getSmallIPtrImm - Return a target constant of pointer type.
191 inline SDValue getSmallIPtrImm(unsigned Imm, const SDLoc &dl) {
192 return CurDAG->getTargetConstant(
193 Imm, dl, PPCLowering->getPointerTy(CurDAG->getDataLayout()));
194 }
195
196 /// isRotateAndMask - Returns true if Mask and Shift can be folded into a
197 /// rotate and mask opcode and mask operation.
198 static bool isRotateAndMask(SDNode *N, unsigned Mask, bool isShiftMask,
199 unsigned &SH, unsigned &MB, unsigned &ME);
200
201 /// getGlobalBaseReg - insert code into the entry mbb to materialize the PIC
202 /// base register. Return the virtual register that holds this value.
203 SDNode *getGlobalBaseReg();
204
205 void selectFrameIndex(SDNode *SN, SDNode *N, unsigned Offset = 0);
206
207 // Select - Convert the specified operand from a target-independent to a
208 // target-specific node if it hasn't already been changed.
209 void Select(SDNode *N) override;
210
211 bool tryBitfieldInsert(SDNode *N);
212 bool tryBitPermutation(SDNode *N);
213 bool tryIntCompareInGPR(SDNode *N);
214
215 // tryTLSXFormLoad - Convert an ISD::LOAD fed by a PPCISD::ADD_TLS into
216 // an X-Form load instruction with the offset being a relocation coming from
217 // the PPCISD::ADD_TLS.
218 bool tryTLSXFormLoad(LoadSDNode *N);
219 // tryTLSXFormStore - Convert an ISD::STORE fed by a PPCISD::ADD_TLS into
220 // an X-Form store instruction with the offset being a relocation coming from
221 // the PPCISD::ADD_TLS.
222 bool tryTLSXFormStore(StoreSDNode *N);
223 /// SelectCC - Select a comparison of the specified values with the
224 /// specified condition code, returning the CR# of the expression.
225 SDValue SelectCC(SDValue LHS, SDValue RHS, ISD::CondCode CC,
226 const SDLoc &dl, SDValue Chain = SDValue());
227
228 /// SelectAddrImmOffs - Return true if the operand is valid for a preinc
229 /// immediate field. Note that the operand at this point is already the
230 /// result of a prior SelectAddressRegImm call.
231 bool SelectAddrImmOffs(SDValue N, SDValue &Out) const {
232 if (N.getOpcode() == ISD::TargetConstant ||
233 N.getOpcode() == ISD::TargetGlobalAddress) {
234 Out = N;
235 return true;
236 }
237
238 return false;
239 }
240
241 /// SelectDSForm - Returns true if address N can be represented by the
242 /// addressing mode of DSForm instructions (a base register, plus a signed
243 /// 16-bit displacement that is a multiple of 4.
244 bool SelectDSForm(SDNode *Parent, SDValue N, SDValue &Disp, SDValue &Base) {
245 return PPCLowering->SelectOptimalAddrMode(Parent, N, Disp, Base, *CurDAG,
246 Align(4)) == PPC::AM_DSForm;
247 }
248
249 /// SelectDQForm - Returns true if address N can be represented by the
250 /// addressing mode of DQForm instructions (a base register, plus a signed
251 /// 16-bit displacement that is a multiple of 16.
252 bool SelectDQForm(SDNode *Parent, SDValue N, SDValue &Disp, SDValue &Base) {
253 return PPCLowering->SelectOptimalAddrMode(Parent, N, Disp, Base, *CurDAG,
254 Align(16)) == PPC::AM_DQForm;
255 }
256
257 /// SelectDForm - Returns true if address N can be represented by
258 /// the addressing mode of DForm instructions (a base register, plus a
259 /// signed 16-bit immediate.
260 bool SelectDForm(SDNode *Parent, SDValue N, SDValue &Disp, SDValue &Base) {
261 return PPCLowering->SelectOptimalAddrMode(Parent, N, Disp, Base, *CurDAG,
262 None) == PPC::AM_DForm;
263 }
264
265 /// SelectXForm - Returns true if address N can be represented by the
266 /// addressing mode of XForm instructions (an indexed [r+r] operation).
267 bool SelectXForm(SDNode *Parent, SDValue N, SDValue &Disp, SDValue &Base) {
268 return PPCLowering->SelectOptimalAddrMode(Parent, N, Disp, Base, *CurDAG,
269 None) == PPC::AM_XForm;
270 }
271
272 /// SelectForceXForm - Given the specified address, force it to be
273 /// represented as an indexed [r+r] operation (an XForm instruction).
274 bool SelectForceXForm(SDNode *Parent, SDValue N, SDValue &Disp,
275 SDValue &Base) {
276 return PPCLowering->SelectForceXFormMode(N, Disp, Base, *CurDAG) ==
277 PPC::AM_XForm;
278 }
279
280 /// SelectAddrIdx - Given the specified address, check to see if it can be
281 /// represented as an indexed [r+r] operation.
282 /// This is for xform instructions whose associated displacement form is D.
283 /// The last parameter \p 0 means associated D form has no requirment for 16
284 /// bit signed displacement.
285 /// Returns false if it can be represented by [r+imm], which are preferred.
286 bool SelectAddrIdx(SDValue N, SDValue &Base, SDValue &Index) {
287 return PPCLowering->SelectAddressRegReg(N, Base, Index, *CurDAG, None);
288 }
289
290 /// SelectAddrIdx4 - Given the specified address, check to see if it can be
291 /// represented as an indexed [r+r] operation.
292 /// This is for xform instructions whose associated displacement form is DS.
293 /// The last parameter \p 4 means associated DS form 16 bit signed
294 /// displacement must be a multiple of 4.
295 /// Returns false if it can be represented by [r+imm], which are preferred.
296 bool SelectAddrIdxX4(SDValue N, SDValue &Base, SDValue &Index) {
297 return PPCLowering->SelectAddressRegReg(N, Base, Index, *CurDAG,
298 Align(4));
299 }
300
301 /// SelectAddrIdx16 - Given the specified address, check to see if it can be
302 /// represented as an indexed [r+r] operation.
303 /// This is for xform instructions whose associated displacement form is DQ.
304 /// The last parameter \p 16 means associated DQ form 16 bit signed
305 /// displacement must be a multiple of 16.
306 /// Returns false if it can be represented by [r+imm], which are preferred.
307 bool SelectAddrIdxX16(SDValue N, SDValue &Base, SDValue &Index) {
308 return PPCLowering->SelectAddressRegReg(N, Base, Index, *CurDAG,
309 Align(16));
310 }
311
312 /// SelectAddrIdxOnly - Given the specified address, force it to be
313 /// represented as an indexed [r+r] operation.
314 bool SelectAddrIdxOnly(SDValue N, SDValue &Base, SDValue &Index) {
315 return PPCLowering->SelectAddressRegRegOnly(N, Base, Index, *CurDAG);
316 }
317
318 /// SelectAddrImm - Returns true if the address N can be represented by
319 /// a base register plus a signed 16-bit displacement [r+imm].
320 /// The last parameter \p 0 means D form has no requirment for 16 bit signed
321 /// displacement.
322 bool SelectAddrImm(SDValue N, SDValue &Disp,
323 SDValue &Base) {
324 return PPCLowering->SelectAddressRegImm(N, Disp, Base, *CurDAG, None);
325 }
326
327 /// SelectAddrImmX4 - Returns true if the address N can be represented by
328 /// a base register plus a signed 16-bit displacement that is a multiple of
329 /// 4 (last parameter). Suitable for use by STD and friends.
330 bool SelectAddrImmX4(SDValue N, SDValue &Disp, SDValue &Base) {
331 return PPCLowering->SelectAddressRegImm(N, Disp, Base, *CurDAG, Align(4));
332 }
333
334 /// SelectAddrImmX16 - Returns true if the address N can be represented by
335 /// a base register plus a signed 16-bit displacement that is a multiple of
336 /// 16(last parameter). Suitable for use by STXV and friends.
337 bool SelectAddrImmX16(SDValue N, SDValue &Disp, SDValue &Base) {
338 return PPCLowering->SelectAddressRegImm(N, Disp, Base, *CurDAG,
339 Align(16));
340 }
341
342 /// SelectAddrImmX34 - Returns true if the address N can be represented by
343 /// a base register plus a signed 34-bit displacement. Suitable for use by
344 /// PSTXVP and friends.
345 bool SelectAddrImmX34(SDValue N, SDValue &Disp, SDValue &Base) {
346 return PPCLowering->SelectAddressRegImm34(N, Disp, Base, *CurDAG);
347 }
348
349 // Select an address into a single register.
350 bool SelectAddr(SDValue N, SDValue &Base) {
351 Base = N;
352 return true;
353 }
354
355 bool SelectAddrPCRel(SDValue N, SDValue &Base) {
356 return PPCLowering->SelectAddressPCRel(N, Base);
357 }
358
359 /// SelectInlineAsmMemoryOperand - Implement addressing mode selection for
360 /// inline asm expressions. It is always correct to compute the value into
361 /// a register. The case of adding a (possibly relocatable) constant to a
362 /// register can be improved, but it is wrong to substitute Reg+Reg for
363 /// Reg in an asm, because the load or store opcode would have to change.
364 bool SelectInlineAsmMemoryOperand(const SDValue &Op,
365 unsigned ConstraintID,
366 std::vector<SDValue> &OutOps) override {
367 switch(ConstraintID) {
368 default:
369 errs() << "ConstraintID: " << ConstraintID << "\n";
370 llvm_unreachable("Unexpected asm memory constraint")::llvm::llvm_unreachable_internal("Unexpected asm memory constraint"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 370)
;
371 case InlineAsm::Constraint_es:
372 case InlineAsm::Constraint_m:
373 case InlineAsm::Constraint_o:
374 case InlineAsm::Constraint_Q:
375 case InlineAsm::Constraint_Z:
376 case InlineAsm::Constraint_Zy:
377 // We need to make sure that this one operand does not end up in r0
378 // (because we might end up lowering this as 0(%op)).
379 const TargetRegisterInfo *TRI = Subtarget->getRegisterInfo();
380 const TargetRegisterClass *TRC = TRI->getPointerRegClass(*MF, /*Kind=*/1);
381 SDLoc dl(Op);
382 SDValue RC = CurDAG->getTargetConstant(TRC->getID(), dl, MVT::i32);
383 SDValue NewOp =
384 SDValue(CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS,
385 dl, Op.getValueType(),
386 Op, RC), 0);
387
388 OutOps.push_back(NewOp);
389 return false;
390 }
391 return true;
392 }
393
394 StringRef getPassName() const override {
395 return "PowerPC DAG->DAG Pattern Instruction Selection";
396 }
397
398// Include the pieces autogenerated from the target description.
399#include "PPCGenDAGISel.inc"
400
401private:
402 bool trySETCC(SDNode *N);
403 bool tryFoldSWTestBRCC(SDNode *N);
404 bool tryAsSingleRLDICL(SDNode *N);
405 bool tryAsSingleRLDICR(SDNode *N);
406 bool tryAsSingleRLWINM(SDNode *N);
407 bool tryAsSingleRLWINM8(SDNode *N);
408 bool tryAsSingleRLWIMI(SDNode *N);
409 bool tryAsPairOfRLDICL(SDNode *N);
410 bool tryAsSingleRLDIMI(SDNode *N);
411
412 void PeepholePPC64();
413 void PeepholePPC64ZExt();
414 void PeepholeCROps();
415
416 SDValue combineToCMPB(SDNode *N);
417 void foldBoolExts(SDValue &Res, SDNode *&N);
418
419 bool AllUsersSelectZero(SDNode *N);
420 void SwapAllSelectUsers(SDNode *N);
421
422 bool isOffsetMultipleOf(SDNode *N, unsigned Val) const;
423 void transferMemOperands(SDNode *N, SDNode *Result);
424 };
425
426} // end anonymous namespace
427
428/// getGlobalBaseReg - Output the instructions required to put the
429/// base address to use for accessing globals into a register.
430///
431SDNode *PPCDAGToDAGISel::getGlobalBaseReg() {
432 if (!GlobalBaseReg) {
433 const TargetInstrInfo &TII = *Subtarget->getInstrInfo();
434 // Insert the set of GlobalBaseReg into the first MBB of the function
435 MachineBasicBlock &FirstMBB = MF->front();
436 MachineBasicBlock::iterator MBBI = FirstMBB.begin();
437 const Module *M = MF->getFunction().getParent();
438 DebugLoc dl;
439
440 if (PPCLowering->getPointerTy(CurDAG->getDataLayout()) == MVT::i32) {
441 if (Subtarget->isTargetELF()) {
442 GlobalBaseReg = PPC::R30;
443 if (!Subtarget->isSecurePlt() &&
444 M->getPICLevel() == PICLevel::SmallPIC) {
445 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MoveGOTtoLR));
446 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MFLR), GlobalBaseReg);
447 MF->getInfo<PPCFunctionInfo>()->setUsesPICBase(true);
448 } else {
449 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MovePCtoLR));
450 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MFLR), GlobalBaseReg);
451 Register TempReg = RegInfo->createVirtualRegister(&PPC::GPRCRegClass);
452 BuildMI(FirstMBB, MBBI, dl,
453 TII.get(PPC::UpdateGBR), GlobalBaseReg)
454 .addReg(TempReg, RegState::Define).addReg(GlobalBaseReg);
455 MF->getInfo<PPCFunctionInfo>()->setUsesPICBase(true);
456 }
457 } else {
458 GlobalBaseReg =
459 RegInfo->createVirtualRegister(&PPC::GPRC_and_GPRC_NOR0RegClass);
460 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MovePCtoLR));
461 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MFLR), GlobalBaseReg);
462 }
463 } else {
464 // We must ensure that this sequence is dominated by the prologue.
465 // FIXME: This is a bit of a big hammer since we don't get the benefits
466 // of shrink-wrapping whenever we emit this instruction. Considering
467 // this is used in any function where we emit a jump table, this may be
468 // a significant limitation. We should consider inserting this in the
469 // block where it is used and then commoning this sequence up if it
470 // appears in multiple places.
471 // Note: on ISA 3.0 cores, we can use lnia (addpcis) instead of
472 // MovePCtoLR8.
473 MF->getInfo<PPCFunctionInfo>()->setShrinkWrapDisabled(true);
474 GlobalBaseReg = RegInfo->createVirtualRegister(&PPC::G8RC_and_G8RC_NOX0RegClass);
475 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MovePCtoLR8));
476 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MFLR8), GlobalBaseReg);
477 }
478 }
479 return CurDAG->getRegister(GlobalBaseReg,
480 PPCLowering->getPointerTy(CurDAG->getDataLayout()))
481 .getNode();
482}
483
484// Check if a SDValue has the toc-data attribute.
485static bool hasTocDataAttr(SDValue Val, unsigned PointerSize) {
486 GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Val);
1
Calling 'dyn_cast<llvm::GlobalAddressSDNode, llvm::SDValue>'
10
Returning from 'dyn_cast<llvm::GlobalAddressSDNode, llvm::SDValue>'
487 if (!GA)
11
Assuming 'GA' is non-null
12
Taking false branch
488 return false;
489
490 const GlobalVariable *GV = dyn_cast_or_null<GlobalVariable>(GA->getGlobal());
13
Assuming the object is a 'GlobalVariable'
491 if (!GV
13.1
'GV' is non-null
13.1
'GV' is non-null
)
14
Taking false branch
492 return false;
493
494 if (!GV->hasAttribute("toc-data"))
15
Assuming the condition is false
16
Taking false branch
495 return false;
496
497 // TODO: These asserts should be updated as more support for the toc data
498 // transformation is added (64 bit, struct support, etc.).
499
500 assert(PointerSize == 4 && "Only 32 Bit Codegen is currently supported by "(static_cast <bool> (PointerSize == 4 && "Only 32 Bit Codegen is currently supported by "
"the toc data transformation.") ? void (0) : __assert_fail (
"PointerSize == 4 && \"Only 32 Bit Codegen is currently supported by \" \"the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 501, __extension__ __PRETTY_FUNCTION__))
17
Assuming 'PointerSize' is equal to 4
18
'?' condition is true
501 "the toc data transformation.")(static_cast <bool> (PointerSize == 4 && "Only 32 Bit Codegen is currently supported by "
"the toc data transformation.") ? void (0) : __assert_fail (
"PointerSize == 4 && \"Only 32 Bit Codegen is currently supported by \" \"the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 501, __extension__ __PRETTY_FUNCTION__))
;
502
503 assert(PointerSize >= GV->getAlign().valueOrOne().value() &&(static_cast <bool> (PointerSize >= GV->getAlign(
).valueOrOne().value() && "GlobalVariables with an alignment requirement stricter then 4-bytes "
"not supported by the toc data transformation.") ? void (0) :
__assert_fail ("PointerSize >= GV->getAlign().valueOrOne().value() && \"GlobalVariables with an alignment requirement stricter then 4-bytes \" \"not supported by the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 505, __extension__ __PRETTY_FUNCTION__))
19
'?' condition is true
504 "GlobalVariables with an alignment requirement stricter then 4-bytes "(static_cast <bool> (PointerSize >= GV->getAlign(
).valueOrOne().value() && "GlobalVariables with an alignment requirement stricter then 4-bytes "
"not supported by the toc data transformation.") ? void (0) :
__assert_fail ("PointerSize >= GV->getAlign().valueOrOne().value() && \"GlobalVariables with an alignment requirement stricter then 4-bytes \" \"not supported by the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 505, __extension__ __PRETTY_FUNCTION__))
505 "not supported by the toc data transformation.")(static_cast <bool> (PointerSize >= GV->getAlign(
).valueOrOne().value() && "GlobalVariables with an alignment requirement stricter then 4-bytes "
"not supported by the toc data transformation.") ? void (0) :
__assert_fail ("PointerSize >= GV->getAlign().valueOrOne().value() && \"GlobalVariables with an alignment requirement stricter then 4-bytes \" \"not supported by the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 505, __extension__ __PRETTY_FUNCTION__))
;
506
507 Type *PtrType = GV->getType();
508 assert(PtrType->isPointerTy() &&(static_cast <bool> (PtrType->isPointerTy() &&
"GlobalVariables always have pointer type!.") ? void (0) : __assert_fail
("PtrType->isPointerTy() && \"GlobalVariables always have pointer type!.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 509, __extension__ __PRETTY_FUNCTION__))
20
'?' condition is true
509 "GlobalVariables always have pointer type!.")(static_cast <bool> (PtrType->isPointerTy() &&
"GlobalVariables always have pointer type!.") ? void (0) : __assert_fail
("PtrType->isPointerTy() && \"GlobalVariables always have pointer type!.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 509, __extension__ __PRETTY_FUNCTION__))
;
510
511 Type *GVType = dyn_cast<PointerType>(PtrType)->getElementType();
21
Assuming 'PtrType' is not a 'PointerType'
22
Called C++ object pointer is null
512
513 assert(GVType->isSized() && "A GlobalVariable's size must be known to be "(static_cast <bool> (GVType->isSized() && "A GlobalVariable's size must be known to be "
"supported by the toc data transformation.") ? void (0) : __assert_fail
("GVType->isSized() && \"A GlobalVariable's size must be known to be \" \"supported by the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 514, __extension__ __PRETTY_FUNCTION__))
514 "supported by the toc data transformation.")(static_cast <bool> (GVType->isSized() && "A GlobalVariable's size must be known to be "
"supported by the toc data transformation.") ? void (0) : __assert_fail
("GVType->isSized() && \"A GlobalVariable's size must be known to be \" \"supported by the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 514, __extension__ __PRETTY_FUNCTION__))
;
515
516 if (GVType->isVectorTy())
517 report_fatal_error("A GlobalVariable of Vector type is not currently "
518 "supported by the toc data transformation.");
519
520 if (GVType->isArrayTy())
521 report_fatal_error("A GlobalVariable of Array type is not currently "
522 "supported by the toc data transformation.");
523
524 if (GVType->isStructTy())
525 report_fatal_error("A GlobalVariable of Struct type is not currently "
526 "supported by the toc data transformation.");
527
528 assert(GVType->getPrimitiveSizeInBits() <= PointerSize * 8 &&(static_cast <bool> (GVType->getPrimitiveSizeInBits(
) <= PointerSize * 8 && "A GlobalVariable with size larger than 32 bits is not currently "
"supported by the toc data transformation.") ? void (0) : __assert_fail
("GVType->getPrimitiveSizeInBits() <= PointerSize * 8 && \"A GlobalVariable with size larger than 32 bits is not currently \" \"supported by the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 530, __extension__ __PRETTY_FUNCTION__))
529 "A GlobalVariable with size larger than 32 bits is not currently "(static_cast <bool> (GVType->getPrimitiveSizeInBits(
) <= PointerSize * 8 && "A GlobalVariable with size larger than 32 bits is not currently "
"supported by the toc data transformation.") ? void (0) : __assert_fail
("GVType->getPrimitiveSizeInBits() <= PointerSize * 8 && \"A GlobalVariable with size larger than 32 bits is not currently \" \"supported by the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 530, __extension__ __PRETTY_FUNCTION__))
530 "supported by the toc data transformation.")(static_cast <bool> (GVType->getPrimitiveSizeInBits(
) <= PointerSize * 8 && "A GlobalVariable with size larger than 32 bits is not currently "
"supported by the toc data transformation.") ? void (0) : __assert_fail
("GVType->getPrimitiveSizeInBits() <= PointerSize * 8 && \"A GlobalVariable with size larger than 32 bits is not currently \" \"supported by the toc data transformation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 530, __extension__ __PRETTY_FUNCTION__))
;
531
532 if (GV->hasLocalLinkage() || GV->hasPrivateLinkage())
533 report_fatal_error("A GlobalVariable with private or local linkage is not "
534 "currently supported by the toc data transformation.");
535
536 assert(!GV->hasCommonLinkage() &&(static_cast <bool> (!GV->hasCommonLinkage() &&
"Tentative definitions cannot have the mapping class XMC_TD."
) ? void (0) : __assert_fail ("!GV->hasCommonLinkage() && \"Tentative definitions cannot have the mapping class XMC_TD.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 537, __extension__ __PRETTY_FUNCTION__))
537 "Tentative definitions cannot have the mapping class XMC_TD.")(static_cast <bool> (!GV->hasCommonLinkage() &&
"Tentative definitions cannot have the mapping class XMC_TD."
) ? void (0) : __assert_fail ("!GV->hasCommonLinkage() && \"Tentative definitions cannot have the mapping class XMC_TD.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 537, __extension__ __PRETTY_FUNCTION__))
;
538
539 return true;
540}
541
542/// isInt32Immediate - This method tests to see if the node is a 32-bit constant
543/// operand. If so Imm will receive the 32-bit value.
544static bool isInt32Immediate(SDNode *N, unsigned &Imm) {
545 if (N->getOpcode() == ISD::Constant && N->getValueType(0) == MVT::i32) {
546 Imm = cast<ConstantSDNode>(N)->getZExtValue();
547 return true;
548 }
549 return false;
550}
551
552/// isInt64Immediate - This method tests to see if the node is a 64-bit constant
553/// operand. If so Imm will receive the 64-bit value.
554static bool isInt64Immediate(SDNode *N, uint64_t &Imm) {
555 if (N->getOpcode() == ISD::Constant && N->getValueType(0) == MVT::i64) {
556 Imm = cast<ConstantSDNode>(N)->getZExtValue();
557 return true;
558 }
559 return false;
560}
561
562// isInt32Immediate - This method tests to see if a constant operand.
563// If so Imm will receive the 32 bit value.
564static bool isInt32Immediate(SDValue N, unsigned &Imm) {
565 return isInt32Immediate(N.getNode(), Imm);
566}
567
568/// isInt64Immediate - This method tests to see if the value is a 64-bit
569/// constant operand. If so Imm will receive the 64-bit value.
570static bool isInt64Immediate(SDValue N, uint64_t &Imm) {
571 return isInt64Immediate(N.getNode(), Imm);
572}
573
574static unsigned getBranchHint(unsigned PCC,
575 const FunctionLoweringInfo &FuncInfo,
576 const SDValue &DestMBB) {
577 assert(isa<BasicBlockSDNode>(DestMBB))(static_cast <bool> (isa<BasicBlockSDNode>(DestMBB
)) ? void (0) : __assert_fail ("isa<BasicBlockSDNode>(DestMBB)"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 577, __extension__ __PRETTY_FUNCTION__))
;
578
579 if (!FuncInfo.BPI) return PPC::BR_NO_HINT;
580
581 const BasicBlock *BB = FuncInfo.MBB->getBasicBlock();
582 const Instruction *BBTerm = BB->getTerminator();
583
584 if (BBTerm->getNumSuccessors() != 2) return PPC::BR_NO_HINT;
585
586 const BasicBlock *TBB = BBTerm->getSuccessor(0);
587 const BasicBlock *FBB = BBTerm->getSuccessor(1);
588
589 auto TProb = FuncInfo.BPI->getEdgeProbability(BB, TBB);
590 auto FProb = FuncInfo.BPI->getEdgeProbability(BB, FBB);
591
592 // We only want to handle cases which are easy to predict at static time, e.g.
593 // C++ throw statement, that is very likely not taken, or calling never
594 // returned function, e.g. stdlib exit(). So we set Threshold to filter
595 // unwanted cases.
596 //
597 // Below is LLVM branch weight table, we only want to handle case 1, 2
598 //
599 // Case Taken:Nontaken Example
600 // 1. Unreachable 1048575:1 C++ throw, stdlib exit(),
601 // 2. Invoke-terminating 1:1048575
602 // 3. Coldblock 4:64 __builtin_expect
603 // 4. Loop Branch 124:4 For loop
604 // 5. PH/ZH/FPH 20:12
605 const uint32_t Threshold = 10000;
606
607 if (std::max(TProb, FProb) / Threshold < std::min(TProb, FProb))
608 return PPC::BR_NO_HINT;
609
610 LLVM_DEBUG(dbgs() << "Use branch hint for '" << FuncInfo.Fn->getName()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Use branch hint for '" <<
FuncInfo.Fn->getName() << "::" << BB->getName
() << "'\n" << " -> " << TBB->getName
() << ": " << TProb << "\n" << " -> "
<< FBB->getName() << ": " << FProb <<
"\n"; } } while (false)
611 << "::" << BB->getName() << "'\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Use branch hint for '" <<
FuncInfo.Fn->getName() << "::" << BB->getName
() << "'\n" << " -> " << TBB->getName
() << ": " << TProb << "\n" << " -> "
<< FBB->getName() << ": " << FProb <<
"\n"; } } while (false)
612 << " -> " << TBB->getName() << ": " << TProb << "\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Use branch hint for '" <<
FuncInfo.Fn->getName() << "::" << BB->getName
() << "'\n" << " -> " << TBB->getName
() << ": " << TProb << "\n" << " -> "
<< FBB->getName() << ": " << FProb <<
"\n"; } } while (false)
613 << " -> " << FBB->getName() << ": " << FProb << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Use branch hint for '" <<
FuncInfo.Fn->getName() << "::" << BB->getName
() << "'\n" << " -> " << TBB->getName
() << ": " << TProb << "\n" << " -> "
<< FBB->getName() << ": " << FProb <<
"\n"; } } while (false)
;
614
615 const BasicBlockSDNode *BBDN = cast<BasicBlockSDNode>(DestMBB);
616
617 // If Dest BasicBlock is False-BasicBlock (FBB), swap branch probabilities,
618 // because we want 'TProb' stands for 'branch probability' to Dest BasicBlock
619 if (BBDN->getBasicBlock()->getBasicBlock() != TBB)
620 std::swap(TProb, FProb);
621
622 return (TProb > FProb) ? PPC::BR_TAKEN_HINT : PPC::BR_NONTAKEN_HINT;
623}
624
625// isOpcWithIntImmediate - This method tests to see if the node is a specific
626// opcode and that it has a immediate integer right operand.
627// If so Imm will receive the 32 bit value.
628static bool isOpcWithIntImmediate(SDNode *N, unsigned Opc, unsigned& Imm) {
629 return N->getOpcode() == Opc
630 && isInt32Immediate(N->getOperand(1).getNode(), Imm);
631}
632
633void PPCDAGToDAGISel::selectFrameIndex(SDNode *SN, SDNode *N, unsigned Offset) {
634 SDLoc dl(SN);
635 int FI = cast<FrameIndexSDNode>(N)->getIndex();
636 SDValue TFI = CurDAG->getTargetFrameIndex(FI, N->getValueType(0));
637 unsigned Opc = N->getValueType(0) == MVT::i32 ? PPC::ADDI : PPC::ADDI8;
638 if (SN->hasOneUse())
639 CurDAG->SelectNodeTo(SN, Opc, N->getValueType(0), TFI,
640 getSmallIPtrImm(Offset, dl));
641 else
642 ReplaceNode(SN, CurDAG->getMachineNode(Opc, dl, N->getValueType(0), TFI,
643 getSmallIPtrImm(Offset, dl)));
644}
645
646bool PPCDAGToDAGISel::isRotateAndMask(SDNode *N, unsigned Mask,
647 bool isShiftMask, unsigned &SH,
648 unsigned &MB, unsigned &ME) {
649 // Don't even go down this path for i64, since different logic will be
650 // necessary for rldicl/rldicr/rldimi.
651 if (N->getValueType(0) != MVT::i32)
652 return false;
653
654 unsigned Shift = 32;
655 unsigned Indeterminant = ~0; // bit mask marking indeterminant results
656 unsigned Opcode = N->getOpcode();
657 if (N->getNumOperands() != 2 ||
658 !isInt32Immediate(N->getOperand(1).getNode(), Shift) || (Shift > 31))
659 return false;
660
661 if (Opcode == ISD::SHL) {
662 // apply shift left to mask if it comes first
663 if (isShiftMask) Mask = Mask << Shift;
664 // determine which bits are made indeterminant by shift
665 Indeterminant = ~(0xFFFFFFFFu << Shift);
666 } else if (Opcode == ISD::SRL) {
667 // apply shift right to mask if it comes first
668 if (isShiftMask) Mask = Mask >> Shift;
669 // determine which bits are made indeterminant by shift
670 Indeterminant = ~(0xFFFFFFFFu >> Shift);
671 // adjust for the left rotate
672 Shift = 32 - Shift;
673 } else if (Opcode == ISD::ROTL) {
674 Indeterminant = 0;
675 } else {
676 return false;
677 }
678
679 // if the mask doesn't intersect any Indeterminant bits
680 if (Mask && !(Mask & Indeterminant)) {
681 SH = Shift & 31;
682 // make sure the mask is still a mask (wrap arounds may not be)
683 return isRunOfOnes(Mask, MB, ME);
684 }
685 return false;
686}
687
688bool PPCDAGToDAGISel::tryTLSXFormStore(StoreSDNode *ST) {
689 SDValue Base = ST->getBasePtr();
690 if (Base.getOpcode() != PPCISD::ADD_TLS)
691 return false;
692 SDValue Offset = ST->getOffset();
693 if (!Offset.isUndef())
694 return false;
695 if (Base.getOperand(1).getOpcode() == PPCISD::TLS_LOCAL_EXEC_MAT_ADDR)
696 return false;
697
698 SDLoc dl(ST);
699 EVT MemVT = ST->getMemoryVT();
700 EVT RegVT = ST->getValue().getValueType();
701
702 unsigned Opcode;
703 switch (MemVT.getSimpleVT().SimpleTy) {
704 default:
705 return false;
706 case MVT::i8: {
707 Opcode = (RegVT == MVT::i32) ? PPC::STBXTLS_32 : PPC::STBXTLS;
708 break;
709 }
710 case MVT::i16: {
711 Opcode = (RegVT == MVT::i32) ? PPC::STHXTLS_32 : PPC::STHXTLS;
712 break;
713 }
714 case MVT::i32: {
715 Opcode = (RegVT == MVT::i32) ? PPC::STWXTLS_32 : PPC::STWXTLS;
716 break;
717 }
718 case MVT::i64: {
719 Opcode = PPC::STDXTLS;
720 break;
721 }
722 }
723 SDValue Chain = ST->getChain();
724 SDVTList VTs = ST->getVTList();
725 SDValue Ops[] = {ST->getValue(), Base.getOperand(0), Base.getOperand(1),
726 Chain};
727 SDNode *MN = CurDAG->getMachineNode(Opcode, dl, VTs, Ops);
728 transferMemOperands(ST, MN);
729 ReplaceNode(ST, MN);
730 return true;
731}
732
733bool PPCDAGToDAGISel::tryTLSXFormLoad(LoadSDNode *LD) {
734 SDValue Base = LD->getBasePtr();
735 if (Base.getOpcode() != PPCISD::ADD_TLS)
736 return false;
737 SDValue Offset = LD->getOffset();
738 if (!Offset.isUndef())
739 return false;
740 if (Base.getOperand(1).getOpcode() == PPCISD::TLS_LOCAL_EXEC_MAT_ADDR)
741 return false;
742
743 SDLoc dl(LD);
744 EVT MemVT = LD->getMemoryVT();
745 EVT RegVT = LD->getValueType(0);
746 unsigned Opcode;
747 switch (MemVT.getSimpleVT().SimpleTy) {
748 default:
749 return false;
750 case MVT::i8: {
751 Opcode = (RegVT == MVT::i32) ? PPC::LBZXTLS_32 : PPC::LBZXTLS;
752 break;
753 }
754 case MVT::i16: {
755 Opcode = (RegVT == MVT::i32) ? PPC::LHZXTLS_32 : PPC::LHZXTLS;
756 break;
757 }
758 case MVT::i32: {
759 Opcode = (RegVT == MVT::i32) ? PPC::LWZXTLS_32 : PPC::LWZXTLS;
760 break;
761 }
762 case MVT::i64: {
763 Opcode = PPC::LDXTLS;
764 break;
765 }
766 }
767 SDValue Chain = LD->getChain();
768 SDVTList VTs = LD->getVTList();
769 SDValue Ops[] = {Base.getOperand(0), Base.getOperand(1), Chain};
770 SDNode *MN = CurDAG->getMachineNode(Opcode, dl, VTs, Ops);
771 transferMemOperands(LD, MN);
772 ReplaceNode(LD, MN);
773 return true;
774}
775
776/// Turn an or of two masked values into the rotate left word immediate then
777/// mask insert (rlwimi) instruction.
778bool PPCDAGToDAGISel::tryBitfieldInsert(SDNode *N) {
779 SDValue Op0 = N->getOperand(0);
780 SDValue Op1 = N->getOperand(1);
781 SDLoc dl(N);
782
783 KnownBits LKnown = CurDAG->computeKnownBits(Op0);
784 KnownBits RKnown = CurDAG->computeKnownBits(Op1);
785
786 unsigned TargetMask = LKnown.Zero.getZExtValue();
787 unsigned InsertMask = RKnown.Zero.getZExtValue();
788
789 if ((TargetMask | InsertMask) == 0xFFFFFFFF) {
790 unsigned Op0Opc = Op0.getOpcode();
791 unsigned Op1Opc = Op1.getOpcode();
792 unsigned Value, SH = 0;
793 TargetMask = ~TargetMask;
794 InsertMask = ~InsertMask;
795
796 // If the LHS has a foldable shift and the RHS does not, then swap it to the
797 // RHS so that we can fold the shift into the insert.
798 if (Op0Opc == ISD::AND && Op1Opc == ISD::AND) {
799 if (Op0.getOperand(0).getOpcode() == ISD::SHL ||
800 Op0.getOperand(0).getOpcode() == ISD::SRL) {
801 if (Op1.getOperand(0).getOpcode() != ISD::SHL &&
802 Op1.getOperand(0).getOpcode() != ISD::SRL) {
803 std::swap(Op0, Op1);
804 std::swap(Op0Opc, Op1Opc);
805 std::swap(TargetMask, InsertMask);
806 }
807 }
808 } else if (Op0Opc == ISD::SHL || Op0Opc == ISD::SRL) {
809 if (Op1Opc == ISD::AND && Op1.getOperand(0).getOpcode() != ISD::SHL &&
810 Op1.getOperand(0).getOpcode() != ISD::SRL) {
811 std::swap(Op0, Op1);
812 std::swap(Op0Opc, Op1Opc);
813 std::swap(TargetMask, InsertMask);
814 }
815 }
816
817 unsigned MB, ME;
818 if (isRunOfOnes(InsertMask, MB, ME)) {
819 if ((Op1Opc == ISD::SHL || Op1Opc == ISD::SRL) &&
820 isInt32Immediate(Op1.getOperand(1), Value)) {
821 Op1 = Op1.getOperand(0);
822 SH = (Op1Opc == ISD::SHL) ? Value : 32 - Value;
823 }
824 if (Op1Opc == ISD::AND) {
825 // The AND mask might not be a constant, and we need to make sure that
826 // if we're going to fold the masking with the insert, all bits not
827 // know to be zero in the mask are known to be one.
828 KnownBits MKnown = CurDAG->computeKnownBits(Op1.getOperand(1));
829 bool CanFoldMask = InsertMask == MKnown.One.getZExtValue();
830
831 unsigned SHOpc = Op1.getOperand(0).getOpcode();
832 if ((SHOpc == ISD::SHL || SHOpc == ISD::SRL) && CanFoldMask &&
833 isInt32Immediate(Op1.getOperand(0).getOperand(1), Value)) {
834 // Note that Value must be in range here (less than 32) because
835 // otherwise there would not be any bits set in InsertMask.
836 Op1 = Op1.getOperand(0).getOperand(0);
837 SH = (SHOpc == ISD::SHL) ? Value : 32 - Value;
838 }
839 }
840
841 SH &= 31;
842 SDValue Ops[] = { Op0, Op1, getI32Imm(SH, dl), getI32Imm(MB, dl),
843 getI32Imm(ME, dl) };
844 ReplaceNode(N, CurDAG->getMachineNode(PPC::RLWIMI, dl, MVT::i32, Ops));
845 return true;
846 }
847 }
848 return false;
849}
850
851static unsigned allUsesTruncate(SelectionDAG *CurDAG, SDNode *N) {
852 unsigned MaxTruncation = 0;
853 // Cannot use range-based for loop here as we need the actual use (i.e. we
854 // need the operand number corresponding to the use). A range-based for
855 // will unbox the use and provide an SDNode*.
856 for (SDNode::use_iterator Use = N->use_begin(), UseEnd = N->use_end();
857 Use != UseEnd; ++Use) {
858 unsigned Opc =
859 Use->isMachineOpcode() ? Use->getMachineOpcode() : Use->getOpcode();
860 switch (Opc) {
861 default: return 0;
862 case ISD::TRUNCATE:
863 if (Use->isMachineOpcode())
864 return 0;
865 MaxTruncation =
866 std::max(MaxTruncation, (unsigned)Use->getValueType(0).getSizeInBits());
867 continue;
868 case ISD::STORE: {
869 if (Use->isMachineOpcode())
870 return 0;
871 StoreSDNode *STN = cast<StoreSDNode>(*Use);
872 unsigned MemVTSize = STN->getMemoryVT().getSizeInBits();
873 if (MemVTSize == 64 || Use.getOperandNo() != 0)
874 return 0;
875 MaxTruncation = std::max(MaxTruncation, MemVTSize);
876 continue;
877 }
878 case PPC::STW8:
879 case PPC::STWX8:
880 case PPC::STWU8:
881 case PPC::STWUX8:
882 if (Use.getOperandNo() != 0)
883 return 0;
884 MaxTruncation = std::max(MaxTruncation, 32u);
885 continue;
886 case PPC::STH8:
887 case PPC::STHX8:
888 case PPC::STHU8:
889 case PPC::STHUX8:
890 if (Use.getOperandNo() != 0)
891 return 0;
892 MaxTruncation = std::max(MaxTruncation, 16u);
893 continue;
894 case PPC::STB8:
895 case PPC::STBX8:
896 case PPC::STBU8:
897 case PPC::STBUX8:
898 if (Use.getOperandNo() != 0)
899 return 0;
900 MaxTruncation = std::max(MaxTruncation, 8u);
901 continue;
902 }
903 }
904 return MaxTruncation;
905}
906
907// For any 32 < Num < 64, check if the Imm contains at least Num consecutive
908// zeros and return the number of bits by the left of these consecutive zeros.
909static int findContiguousZerosAtLeast(uint64_t Imm, unsigned Num) {
910 unsigned HiTZ = countTrailingZeros<uint32_t>(Hi_32(Imm));
911 unsigned LoLZ = countLeadingZeros<uint32_t>(Lo_32(Imm));
912 if ((HiTZ + LoLZ) >= Num)
913 return (32 + HiTZ);
914 return 0;
915}
916
917// Direct materialization of 64-bit constants by enumerated patterns.
918static SDNode *selectI64ImmDirect(SelectionDAG *CurDAG, const SDLoc &dl,
919 uint64_t Imm, unsigned &InstCnt) {
920 unsigned TZ = countTrailingZeros<uint64_t>(Imm);
921 unsigned LZ = countLeadingZeros<uint64_t>(Imm);
922 unsigned TO = countTrailingOnes<uint64_t>(Imm);
923 unsigned LO = countLeadingOnes<uint64_t>(Imm);
924 unsigned Hi32 = Hi_32(Imm);
925 unsigned Lo32 = Lo_32(Imm);
926 SDNode *Result = nullptr;
927 unsigned Shift = 0;
928
929 auto getI32Imm = [CurDAG, dl](unsigned Imm) {
930 return CurDAG->getTargetConstant(Imm, dl, MVT::i32);
931 };
932
933 // Following patterns use 1 instructions to materialize the Imm.
934 InstCnt = 1;
935 // 1-1) Patterns : {zeros}{15-bit valve}
936 // {ones}{15-bit valve}
937 if (isInt<16>(Imm)) {
938 SDValue SDImm = CurDAG->getTargetConstant(Imm, dl, MVT::i64);
939 return CurDAG->getMachineNode(PPC::LI8, dl, MVT::i64, SDImm);
940 }
941 // 1-2) Patterns : {zeros}{15-bit valve}{16 zeros}
942 // {ones}{15-bit valve}{16 zeros}
943 if (TZ > 15 && (LZ > 32 || LO > 32))
944 return CurDAG->getMachineNode(PPC::LIS8, dl, MVT::i64,
945 getI32Imm((Imm >> 16) & 0xffff));
946
947 // Following patterns use 2 instructions to materialize the Imm.
948 InstCnt = 2;
949 assert(LZ < 64 && "Unexpected leading zeros here.")(static_cast <bool> (LZ < 64 && "Unexpected leading zeros here."
) ? void (0) : __assert_fail ("LZ < 64 && \"Unexpected leading zeros here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 949, __extension__ __PRETTY_FUNCTION__))
;
950 // Count of ones follwing the leading zeros.
951 unsigned FO = countLeadingOnes<uint64_t>(Imm << LZ);
952 // 2-1) Patterns : {zeros}{31-bit value}
953 // {ones}{31-bit value}
954 if (isInt<32>(Imm)) {
955 uint64_t ImmHi16 = (Imm >> 16) & 0xffff;
956 unsigned Opcode = ImmHi16 ? PPC::LIS8 : PPC::LI8;
957 Result = CurDAG->getMachineNode(Opcode, dl, MVT::i64, getI32Imm(ImmHi16));
958 return CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, SDValue(Result, 0),
959 getI32Imm(Imm & 0xffff));
960 }
961 // 2-2) Patterns : {zeros}{ones}{15-bit value}{zeros}
962 // {zeros}{15-bit value}{zeros}
963 // {zeros}{ones}{15-bit value}
964 // {ones}{15-bit value}{zeros}
965 // We can take advantage of LI's sign-extension semantics to generate leading
966 // ones, and then use RLDIC to mask off the ones in both sides after rotation.
967 if ((LZ + FO + TZ) > 48) {
968 Result = CurDAG->getMachineNode(PPC::LI8, dl, MVT::i64,
969 getI32Imm((Imm >> TZ) & 0xffff));
970 return CurDAG->getMachineNode(PPC::RLDIC, dl, MVT::i64, SDValue(Result, 0),
971 getI32Imm(TZ), getI32Imm(LZ));
972 }
973 // 2-3) Pattern : {zeros}{15-bit value}{ones}
974 // Shift right the Imm by (48 - LZ) bits to construct a negtive 16 bits value,
975 // therefore we can take advantage of LI's sign-extension semantics, and then
976 // mask them off after rotation.
977 //
978 // +--LZ--||-15-bit-||--TO--+ +-------------|--16-bit--+
979 // |00000001bbbbbbbbb1111111| -> |00000000000001bbbbbbbbb1|
980 // +------------------------+ +------------------------+
981 // 63 0 63 0
982 // Imm (Imm >> (48 - LZ) & 0xffff)
983 // +----sext-----|--16-bit--+ +clear-|-----------------+
984 // |11111111111111bbbbbbbbb1| -> |00000001bbbbbbbbb1111111|
985 // +------------------------+ +------------------------+
986 // 63 0 63 0
987 // LI8: sext many leading zeros RLDICL: rotate left (48 - LZ), clear left LZ
988 if ((LZ + TO) > 48) {
989 // Since the immediates with (LZ > 32) have been handled by previous
990 // patterns, here we have (LZ <= 32) to make sure we will not shift right
991 // the Imm by a negative value.
992 assert(LZ <= 32 && "Unexpected shift value.")(static_cast <bool> (LZ <= 32 && "Unexpected shift value."
) ? void (0) : __assert_fail ("LZ <= 32 && \"Unexpected shift value.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 992, __extension__ __PRETTY_FUNCTION__))
;
993 Result = CurDAG->getMachineNode(PPC::LI8, dl, MVT::i64,
994 getI32Imm((Imm >> (48 - LZ) & 0xffff)));
995 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, SDValue(Result, 0),
996 getI32Imm(48 - LZ), getI32Imm(LZ));
997 }
998 // 2-4) Patterns : {zeros}{ones}{15-bit value}{ones}
999 // {ones}{15-bit value}{ones}
1000 // We can take advantage of LI's sign-extension semantics to generate leading
1001 // ones, and then use RLDICL to mask off the ones in left sides (if required)
1002 // after rotation.
1003 //
1004 // +-LZ-FO||-15-bit-||--TO--+ +-------------|--16-bit--+
1005 // |00011110bbbbbbbbb1111111| -> |000000000011110bbbbbbbbb|
1006 // +------------------------+ +------------------------+
1007 // 63 0 63 0
1008 // Imm (Imm >> TO) & 0xffff
1009 // +----sext-----|--16-bit--+ +LZ|---------------------+
1010 // |111111111111110bbbbbbbbb| -> |00011110bbbbbbbbb1111111|
1011 // +------------------------+ +------------------------+
1012 // 63 0 63 0
1013 // LI8: sext many leading zeros RLDICL: rotate left TO, clear left LZ
1014 if ((LZ + FO + TO) > 48) {
1015 Result = CurDAG->getMachineNode(PPC::LI8, dl, MVT::i64,
1016 getI32Imm((Imm >> TO) & 0xffff));
1017 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, SDValue(Result, 0),
1018 getI32Imm(TO), getI32Imm(LZ));
1019 }
1020 // 2-5) Pattern : {32 zeros}{****}{0}{15-bit value}
1021 // If Hi32 is zero and the Lo16(in Lo32) can be presented as a positive 16 bit
1022 // value, we can use LI for Lo16 without generating leading ones then add the
1023 // Hi16(in Lo32).
1024 if (LZ == 32 && ((Lo32 & 0x8000) == 0)) {
1025 Result = CurDAG->getMachineNode(PPC::LI8, dl, MVT::i64,
1026 getI32Imm(Lo32 & 0xffff));
1027 return CurDAG->getMachineNode(PPC::ORIS8, dl, MVT::i64, SDValue(Result, 0),
1028 getI32Imm(Lo32 >> 16));
1029 }
1030 // 2-6) Patterns : {******}{49 zeros}{******}
1031 // {******}{49 ones}{******}
1032 // If the Imm contains 49 consecutive zeros/ones, it means that a total of 15
1033 // bits remain on both sides. Rotate right the Imm to construct an int<16>
1034 // value, use LI for int<16> value and then use RLDICL without mask to rotate
1035 // it back.
1036 //
1037 // 1) findContiguousZerosAtLeast(Imm, 49)
1038 // +------|--zeros-|------+ +---ones--||---15 bit--+
1039 // |bbbbbb0000000000aaaaaa| -> |0000000000aaaaaabbbbbb|
1040 // +----------------------+ +----------------------+
1041 // 63 0 63 0
1042 //
1043 // 2) findContiguousZerosAtLeast(~Imm, 49)
1044 // +------|--ones--|------+ +---ones--||---15 bit--+
1045 // |bbbbbb1111111111aaaaaa| -> |1111111111aaaaaabbbbbb|
1046 // +----------------------+ +----------------------+
1047 // 63 0 63 0
1048 if ((Shift = findContiguousZerosAtLeast(Imm, 49)) ||
1049 (Shift = findContiguousZerosAtLeast(~Imm, 49))) {
1050 uint64_t RotImm = APInt(64, Imm).rotr(Shift).getZExtValue();
1051 Result = CurDAG->getMachineNode(PPC::LI8, dl, MVT::i64,
1052 getI32Imm(RotImm & 0xffff));
1053 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, SDValue(Result, 0),
1054 getI32Imm(Shift), getI32Imm(0));
1055 }
1056
1057 // Following patterns use 3 instructions to materialize the Imm.
1058 InstCnt = 3;
1059 // 3-1) Patterns : {zeros}{ones}{31-bit value}{zeros}
1060 // {zeros}{31-bit value}{zeros}
1061 // {zeros}{ones}{31-bit value}
1062 // {ones}{31-bit value}{zeros}
1063 // We can take advantage of LIS's sign-extension semantics to generate leading
1064 // ones, add the remaining bits with ORI, and then use RLDIC to mask off the
1065 // ones in both sides after rotation.
1066 if ((LZ + FO + TZ) > 32) {
1067 uint64_t ImmHi16 = (Imm >> (TZ + 16)) & 0xffff;
1068 unsigned Opcode = ImmHi16 ? PPC::LIS8 : PPC::LI8;
1069 Result = CurDAG->getMachineNode(Opcode, dl, MVT::i64, getI32Imm(ImmHi16));
1070 Result = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, SDValue(Result, 0),
1071 getI32Imm((Imm >> TZ) & 0xffff));
1072 return CurDAG->getMachineNode(PPC::RLDIC, dl, MVT::i64, SDValue(Result, 0),
1073 getI32Imm(TZ), getI32Imm(LZ));
1074 }
1075 // 3-2) Pattern : {zeros}{31-bit value}{ones}
1076 // Shift right the Imm by (32 - LZ) bits to construct a negtive 32 bits value,
1077 // therefore we can take advantage of LIS's sign-extension semantics, add
1078 // the remaining bits with ORI, and then mask them off after rotation.
1079 // This is similar to Pattern 2-3, please refer to the diagram there.
1080 if ((LZ + TO) > 32) {
1081 // Since the immediates with (LZ > 32) have been handled by previous
1082 // patterns, here we have (LZ <= 32) to make sure we will not shift right
1083 // the Imm by a negative value.
1084 assert(LZ <= 32 && "Unexpected shift value.")(static_cast <bool> (LZ <= 32 && "Unexpected shift value."
) ? void (0) : __assert_fail ("LZ <= 32 && \"Unexpected shift value.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 1084, __extension__ __PRETTY_FUNCTION__))
;
1085 Result = CurDAG->getMachineNode(PPC::LIS8, dl, MVT::i64,
1086 getI32Imm((Imm >> (48 - LZ)) & 0xffff));
1087 Result = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, SDValue(Result, 0),
1088 getI32Imm((Imm >> (32 - LZ)) & 0xffff));
1089 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, SDValue(Result, 0),
1090 getI32Imm(32 - LZ), getI32Imm(LZ));
1091 }
1092 // 3-3) Patterns : {zeros}{ones}{31-bit value}{ones}
1093 // {ones}{31-bit value}{ones}
1094 // We can take advantage of LIS's sign-extension semantics to generate leading
1095 // ones, add the remaining bits with ORI, and then use RLDICL to mask off the
1096 // ones in left sides (if required) after rotation.
1097 // This is similar to Pattern 2-4, please refer to the diagram there.
1098 if ((LZ + FO + TO) > 32) {
1099 Result = CurDAG->getMachineNode(PPC::LIS8, dl, MVT::i64,
1100 getI32Imm((Imm >> (TO + 16)) & 0xffff));
1101 Result = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, SDValue(Result, 0),
1102 getI32Imm((Imm >> TO) & 0xffff));
1103 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, SDValue(Result, 0),
1104 getI32Imm(TO), getI32Imm(LZ));
1105 }
1106 // 3-4) Patterns : High word == Low word
1107 if (Hi32 == Lo32) {
1108 // Handle the first 32 bits.
1109 uint64_t ImmHi16 = (Lo32 >> 16) & 0xffff;
1110 unsigned Opcode = ImmHi16 ? PPC::LIS8 : PPC::LI8;
1111 Result = CurDAG->getMachineNode(Opcode, dl, MVT::i64, getI32Imm(ImmHi16));
1112 Result = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, SDValue(Result, 0),
1113 getI32Imm(Lo32 & 0xffff));
1114 // Use rldimi to insert the Low word into High word.
1115 SDValue Ops[] = {SDValue(Result, 0), SDValue(Result, 0), getI32Imm(32),
1116 getI32Imm(0)};
1117 return CurDAG->getMachineNode(PPC::RLDIMI, dl, MVT::i64, Ops);
1118 }
1119 // 3-5) Patterns : {******}{33 zeros}{******}
1120 // {******}{33 ones}{******}
1121 // If the Imm contains 33 consecutive zeros/ones, it means that a total of 31
1122 // bits remain on both sides. Rotate right the Imm to construct an int<32>
1123 // value, use LIS + ORI for int<32> value and then use RLDICL without mask to
1124 // rotate it back.
1125 // This is similar to Pattern 2-6, please refer to the diagram there.
1126 if ((Shift = findContiguousZerosAtLeast(Imm, 33)) ||
1127 (Shift = findContiguousZerosAtLeast(~Imm, 33))) {
1128 uint64_t RotImm = APInt(64, Imm).rotr(Shift).getZExtValue();
1129 uint64_t ImmHi16 = (RotImm >> 16) & 0xffff;
1130 unsigned Opcode = ImmHi16 ? PPC::LIS8 : PPC::LI8;
1131 Result = CurDAG->getMachineNode(Opcode, dl, MVT::i64, getI32Imm(ImmHi16));
1132 Result = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, SDValue(Result, 0),
1133 getI32Imm(RotImm & 0xffff));
1134 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, SDValue(Result, 0),
1135 getI32Imm(Shift), getI32Imm(0));
1136 }
1137
1138 InstCnt = 0;
1139 return nullptr;
1140}
1141
1142// Try to select instructions to generate a 64 bit immediate using prefix as
1143// well as non prefix instructions. The function will return the SDNode
1144// to materialize that constant or it will return nullptr if it does not
1145// find one. The variable InstCnt is set to the number of instructions that
1146// were selected.
1147static SDNode *selectI64ImmDirectPrefix(SelectionDAG *CurDAG, const SDLoc &dl,
1148 uint64_t Imm, unsigned &InstCnt) {
1149 unsigned TZ = countTrailingZeros<uint64_t>(Imm);
1150 unsigned LZ = countLeadingZeros<uint64_t>(Imm);
1151 unsigned TO = countTrailingOnes<uint64_t>(Imm);
1152 unsigned FO = countLeadingOnes<uint64_t>(LZ == 64 ? 0 : (Imm << LZ));
1153 unsigned Hi32 = Hi_32(Imm);
1154 unsigned Lo32 = Lo_32(Imm);
1155
1156 auto getI32Imm = [CurDAG, dl](unsigned Imm) {
1157 return CurDAG->getTargetConstant(Imm, dl, MVT::i32);
1158 };
1159
1160 auto getI64Imm = [CurDAG, dl](uint64_t Imm) {
1161 return CurDAG->getTargetConstant(Imm, dl, MVT::i64);
1162 };
1163
1164 // Following patterns use 1 instruction to materialize Imm.
1165 InstCnt = 1;
1166
1167 // The pli instruction can materialize up to 34 bits directly.
1168 // If a constant fits within 34-bits, emit the pli instruction here directly.
1169 if (isInt<34>(Imm))
1170 return CurDAG->getMachineNode(PPC::PLI8, dl, MVT::i64,
1171 CurDAG->getTargetConstant(Imm, dl, MVT::i64));
1172
1173 // Require at least two instructions.
1174 InstCnt = 2;
1175 SDNode *Result = nullptr;
1176 // Patterns : {zeros}{ones}{33-bit value}{zeros}
1177 // {zeros}{33-bit value}{zeros}
1178 // {zeros}{ones}{33-bit value}
1179 // {ones}{33-bit value}{zeros}
1180 // We can take advantage of PLI's sign-extension semantics to generate leading
1181 // ones, and then use RLDIC to mask off the ones on both sides after rotation.
1182 if ((LZ + FO + TZ) > 30) {
1183 APInt SignedInt34 = APInt(34, (Imm >> TZ) & 0x3ffffffff);
1184 APInt Extended = SignedInt34.sext(64);
1185 Result = CurDAG->getMachineNode(PPC::PLI8, dl, MVT::i64,
1186 getI64Imm(*Extended.getRawData()));
1187 return CurDAG->getMachineNode(PPC::RLDIC, dl, MVT::i64, SDValue(Result, 0),
1188 getI32Imm(TZ), getI32Imm(LZ));
1189 }
1190 // Pattern : {zeros}{33-bit value}{ones}
1191 // Shift right the Imm by (30 - LZ) bits to construct a negative 34 bit value,
1192 // therefore we can take advantage of PLI's sign-extension semantics, and then
1193 // mask them off after rotation.
1194 //
1195 // +--LZ--||-33-bit-||--TO--+ +-------------|--34-bit--+
1196 // |00000001bbbbbbbbb1111111| -> |00000000000001bbbbbbbbb1|
1197 // +------------------------+ +------------------------+
1198 // 63 0 63 0
1199 //
1200 // +----sext-----|--34-bit--+ +clear-|-----------------+
1201 // |11111111111111bbbbbbbbb1| -> |00000001bbbbbbbbb1111111|
1202 // +------------------------+ +------------------------+
1203 // 63 0 63 0
1204 if ((LZ + TO) > 30) {
1205 APInt SignedInt34 = APInt(34, (Imm >> (30 - LZ)) & 0x3ffffffff);
1206 APInt Extended = SignedInt34.sext(64);
1207 Result = CurDAG->getMachineNode(PPC::PLI8, dl, MVT::i64,
1208 getI64Imm(*Extended.getRawData()));
1209 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, SDValue(Result, 0),
1210 getI32Imm(30 - LZ), getI32Imm(LZ));
1211 }
1212 // Patterns : {zeros}{ones}{33-bit value}{ones}
1213 // {ones}{33-bit value}{ones}
1214 // Similar to LI we can take advantage of PLI's sign-extension semantics to
1215 // generate leading ones, and then use RLDICL to mask off the ones in left
1216 // sides (if required) after rotation.
1217 if ((LZ + FO + TO) > 30) {
1218 APInt SignedInt34 = APInt(34, (Imm >> TO) & 0x3ffffffff);
1219 APInt Extended = SignedInt34.sext(64);
1220 Result = CurDAG->getMachineNode(PPC::PLI8, dl, MVT::i64,
1221 getI64Imm(*Extended.getRawData()));
1222 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, SDValue(Result, 0),
1223 getI32Imm(TO), getI32Imm(LZ));
1224 }
1225 // Patterns : {******}{31 zeros}{******}
1226 // : {******}{31 ones}{******}
1227 // If Imm contains 31 consecutive zeros/ones then the remaining bit count
1228 // is 33. Rotate right the Imm to construct a int<33> value, we can use PLI
1229 // for the int<33> value and then use RLDICL without a mask to rotate it back.
1230 //
1231 // +------|--ones--|------+ +---ones--||---33 bit--+
1232 // |bbbbbb1111111111aaaaaa| -> |1111111111aaaaaabbbbbb|
1233 // +----------------------+ +----------------------+
1234 // 63 0 63 0
1235 for (unsigned Shift = 0; Shift < 63; ++Shift) {
1236 uint64_t RotImm = APInt(64, Imm).rotr(Shift).getZExtValue();
1237 if (isInt<34>(RotImm)) {
1238 Result =
1239 CurDAG->getMachineNode(PPC::PLI8, dl, MVT::i64, getI64Imm(RotImm));
1240 return CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
1241 SDValue(Result, 0), getI32Imm(Shift),
1242 getI32Imm(0));
1243 }
1244 }
1245
1246 // Patterns : High word == Low word
1247 // This is basically a splat of a 32 bit immediate.
1248 if (Hi32 == Lo32) {
1249 Result = CurDAG->getMachineNode(PPC::PLI8, dl, MVT::i64, getI64Imm(Hi32));
1250 SDValue Ops[] = {SDValue(Result, 0), SDValue(Result, 0), getI32Imm(32),
1251 getI32Imm(0)};
1252 return CurDAG->getMachineNode(PPC::RLDIMI, dl, MVT::i64, Ops);
1253 }
1254
1255 InstCnt = 3;
1256 // Catch-all
1257 // This pattern can form any 64 bit immediate in 3 instructions.
1258 SDNode *ResultHi =
1259 CurDAG->getMachineNode(PPC::PLI8, dl, MVT::i64, getI64Imm(Hi32));
1260 SDNode *ResultLo =
1261 CurDAG->getMachineNode(PPC::PLI8, dl, MVT::i64, getI64Imm(Lo32));
1262 SDValue Ops[] = {SDValue(ResultLo, 0), SDValue(ResultHi, 0), getI32Imm(32),
1263 getI32Imm(0)};
1264 return CurDAG->getMachineNode(PPC::RLDIMI, dl, MVT::i64, Ops);
1265}
1266
1267static SDNode *selectI64Imm(SelectionDAG *CurDAG, const SDLoc &dl, uint64_t Imm,
1268 unsigned *InstCnt = nullptr) {
1269 unsigned InstCntDirect = 0;
1270 // No more than 3 instructions is used if we can select the i64 immediate
1271 // directly.
1272 SDNode *Result = selectI64ImmDirect(CurDAG, dl, Imm, InstCntDirect);
1273
1274 const PPCSubtarget &Subtarget =
1275 CurDAG->getMachineFunction().getSubtarget<PPCSubtarget>();
1276
1277 // If we have prefixed instructions and there is a chance we can
1278 // materialize the constant with fewer prefixed instructions than
1279 // non-prefixed, try that.
1280 if (Subtarget.hasPrefixInstrs() && InstCntDirect != 1) {
1281 unsigned InstCntDirectP = 0;
1282 SDNode *ResultP = selectI64ImmDirectPrefix(CurDAG, dl, Imm, InstCntDirectP);
1283 // Use the prefix case in either of two cases:
1284 // 1) We have no result from the non-prefix case to use.
1285 // 2) The non-prefix case uses more instructions than the prefix case.
1286 // If the prefix and non-prefix cases use the same number of instructions
1287 // we will prefer the non-prefix case.
1288 if (ResultP && (!Result || InstCntDirectP < InstCntDirect)) {
1289 if (InstCnt)
1290 *InstCnt = InstCntDirectP;
1291 return ResultP;
1292 }
1293 }
1294
1295 if (Result) {
1296 if (InstCnt)
1297 *InstCnt = InstCntDirect;
1298 return Result;
1299 }
1300 auto getI32Imm = [CurDAG, dl](unsigned Imm) {
1301 return CurDAG->getTargetConstant(Imm, dl, MVT::i32);
1302 };
1303 // Handle the upper 32 bit value.
1304 Result =
1305 selectI64ImmDirect(CurDAG, dl, Imm & 0xffffffff00000000, InstCntDirect);
1306 // Add in the last bits as required.
1307 if (uint32_t Hi16 = (Lo_32(Imm) >> 16) & 0xffff) {
1308 Result = CurDAG->getMachineNode(PPC::ORIS8, dl, MVT::i64,
1309 SDValue(Result, 0), getI32Imm(Hi16));
1310 ++InstCntDirect;
1311 }
1312 if (uint32_t Lo16 = Lo_32(Imm) & 0xffff) {
1313 Result = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, SDValue(Result, 0),
1314 getI32Imm(Lo16));
1315 ++InstCntDirect;
1316 }
1317 if (InstCnt)
1318 *InstCnt = InstCntDirect;
1319 return Result;
1320}
1321
1322// Select a 64-bit constant.
1323static SDNode *selectI64Imm(SelectionDAG *CurDAG, SDNode *N) {
1324 SDLoc dl(N);
1325
1326 // Get 64 bit value.
1327 int64_t Imm = cast<ConstantSDNode>(N)->getZExtValue();
1328 if (unsigned MinSize = allUsesTruncate(CurDAG, N)) {
1329 uint64_t SextImm = SignExtend64(Imm, MinSize);
1330 SDValue SDImm = CurDAG->getTargetConstant(SextImm, dl, MVT::i64);
1331 if (isInt<16>(SextImm))
1332 return CurDAG->getMachineNode(PPC::LI8, dl, MVT::i64, SDImm);
1333 }
1334 return selectI64Imm(CurDAG, dl, Imm);
1335}
1336
1337namespace {
1338
1339class BitPermutationSelector {
1340 struct ValueBit {
1341 SDValue V;
1342
1343 // The bit number in the value, using a convention where bit 0 is the
1344 // lowest-order bit.
1345 unsigned Idx;
1346
1347 // ConstZero means a bit we need to mask off.
1348 // Variable is a bit comes from an input variable.
1349 // VariableKnownToBeZero is also a bit comes from an input variable,
1350 // but it is known to be already zero. So we do not need to mask them.
1351 enum Kind {
1352 ConstZero,
1353 Variable,
1354 VariableKnownToBeZero
1355 } K;
1356
1357 ValueBit(SDValue V, unsigned I, Kind K = Variable)
1358 : V(V), Idx(I), K(K) {}
1359 ValueBit(Kind K = Variable)
1360 : V(SDValue(nullptr, 0)), Idx(UINT32_MAX(4294967295U)), K(K) {}
1361
1362 bool isZero() const {
1363 return K == ConstZero || K == VariableKnownToBeZero;
1364 }
1365
1366 bool hasValue() const {
1367 return K == Variable || K == VariableKnownToBeZero;
1368 }
1369
1370 SDValue getValue() const {
1371 assert(hasValue() && "Cannot get the value of a constant bit")(static_cast <bool> (hasValue() && "Cannot get the value of a constant bit"
) ? void (0) : __assert_fail ("hasValue() && \"Cannot get the value of a constant bit\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 1371, __extension__ __PRETTY_FUNCTION__))
;
1372 return V;
1373 }
1374
1375 unsigned getValueBitIndex() const {
1376 assert(hasValue() && "Cannot get the value bit index of a constant bit")(static_cast <bool> (hasValue() && "Cannot get the value bit index of a constant bit"
) ? void (0) : __assert_fail ("hasValue() && \"Cannot get the value bit index of a constant bit\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 1376, __extension__ __PRETTY_FUNCTION__))
;
1377 return Idx;
1378 }
1379 };
1380
1381 // A bit group has the same underlying value and the same rotate factor.
1382 struct BitGroup {
1383 SDValue V;
1384 unsigned RLAmt;
1385 unsigned StartIdx, EndIdx;
1386
1387 // This rotation amount assumes that the lower 32 bits of the quantity are
1388 // replicated in the high 32 bits by the rotation operator (which is done
1389 // by rlwinm and friends in 64-bit mode).
1390 bool Repl32;
1391 // Did converting to Repl32 == true change the rotation factor? If it did,
1392 // it decreased it by 32.
1393 bool Repl32CR;
1394 // Was this group coalesced after setting Repl32 to true?
1395 bool Repl32Coalesced;
1396
1397 BitGroup(SDValue V, unsigned R, unsigned S, unsigned E)
1398 : V(V), RLAmt(R), StartIdx(S), EndIdx(E), Repl32(false), Repl32CR(false),
1399 Repl32Coalesced(false) {
1400 LLVM_DEBUG(dbgs() << "\tbit group for " << V.getNode() << " RLAmt = " << Rdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tbit group for " <<
V.getNode() << " RLAmt = " << R << " [" <<
S << ", " << E << "]\n"; } } while (false)
1401 << " [" << S << ", " << E << "]\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tbit group for " <<
V.getNode() << " RLAmt = " << R << " [" <<
S << ", " << E << "]\n"; } } while (false)
;
1402 }
1403 };
1404
1405 // Information on each (Value, RLAmt) pair (like the number of groups
1406 // associated with each) used to choose the lowering method.
1407 struct ValueRotInfo {
1408 SDValue V;
1409 unsigned RLAmt = std::numeric_limits<unsigned>::max();
1410 unsigned NumGroups = 0;
1411 unsigned FirstGroupStartIdx = std::numeric_limits<unsigned>::max();
1412 bool Repl32 = false;
1413
1414 ValueRotInfo() = default;
1415
1416 // For sorting (in reverse order) by NumGroups, and then by
1417 // FirstGroupStartIdx.
1418 bool operator < (const ValueRotInfo &Other) const {
1419 // We need to sort so that the non-Repl32 come first because, when we're
1420 // doing masking, the Repl32 bit groups might be subsumed into the 64-bit
1421 // masking operation.
1422 if (Repl32 < Other.Repl32)
1423 return true;
1424 else if (Repl32 > Other.Repl32)
1425 return false;
1426 else if (NumGroups > Other.NumGroups)
1427 return true;
1428 else if (NumGroups < Other.NumGroups)
1429 return false;
1430 else if (RLAmt == 0 && Other.RLAmt != 0)
1431 return true;
1432 else if (RLAmt != 0 && Other.RLAmt == 0)
1433 return false;
1434 else if (FirstGroupStartIdx < Other.FirstGroupStartIdx)
1435 return true;
1436 return false;
1437 }
1438 };
1439
1440 using ValueBitsMemoizedValue = std::pair<bool, SmallVector<ValueBit, 64>>;
1441 using ValueBitsMemoizer =
1442 DenseMap<SDValue, std::unique_ptr<ValueBitsMemoizedValue>>;
1443 ValueBitsMemoizer Memoizer;
1444
1445 // Return a pair of bool and a SmallVector pointer to a memoization entry.
1446 // The bool is true if something interesting was deduced, otherwise if we're
1447 // providing only a generic representation of V (or something else likewise
1448 // uninteresting for instruction selection) through the SmallVector.
1449 std::pair<bool, SmallVector<ValueBit, 64> *> getValueBits(SDValue V,
1450 unsigned NumBits) {
1451 auto &ValueEntry = Memoizer[V];
1452 if (ValueEntry)
1453 return std::make_pair(ValueEntry->first, &ValueEntry->second);
1454 ValueEntry.reset(new ValueBitsMemoizedValue());
1455 bool &Interesting = ValueEntry->first;
1456 SmallVector<ValueBit, 64> &Bits = ValueEntry->second;
1457 Bits.resize(NumBits);
1458
1459 switch (V.getOpcode()) {
1460 default: break;
1461 case ISD::ROTL:
1462 if (isa<ConstantSDNode>(V.getOperand(1))) {
1463 unsigned RotAmt = V.getConstantOperandVal(1);
1464
1465 const auto &LHSBits = *getValueBits(V.getOperand(0), NumBits).second;
1466
1467 for (unsigned i = 0; i < NumBits; ++i)
1468 Bits[i] = LHSBits[i < RotAmt ? i + (NumBits - RotAmt) : i - RotAmt];
1469
1470 return std::make_pair(Interesting = true, &Bits);
1471 }
1472 break;
1473 case ISD::SHL:
1474 case PPCISD::SHL:
1475 if (isa<ConstantSDNode>(V.getOperand(1))) {
1476 unsigned ShiftAmt = V.getConstantOperandVal(1);
1477
1478 const auto &LHSBits = *getValueBits(V.getOperand(0), NumBits).second;
1479
1480 for (unsigned i = ShiftAmt; i < NumBits; ++i)
1481 Bits[i] = LHSBits[i - ShiftAmt];
1482
1483 for (unsigned i = 0; i < ShiftAmt; ++i)
1484 Bits[i] = ValueBit(ValueBit::ConstZero);
1485
1486 return std::make_pair(Interesting = true, &Bits);
1487 }
1488 break;
1489 case ISD::SRL:
1490 case PPCISD::SRL:
1491 if (isa<ConstantSDNode>(V.getOperand(1))) {
1492 unsigned ShiftAmt = V.getConstantOperandVal(1);
1493
1494 const auto &LHSBits = *getValueBits(V.getOperand(0), NumBits).second;
1495
1496 for (unsigned i = 0; i < NumBits - ShiftAmt; ++i)
1497 Bits[i] = LHSBits[i + ShiftAmt];
1498
1499 for (unsigned i = NumBits - ShiftAmt; i < NumBits; ++i)
1500 Bits[i] = ValueBit(ValueBit::ConstZero);
1501
1502 return std::make_pair(Interesting = true, &Bits);
1503 }
1504 break;
1505 case ISD::AND:
1506 if (isa<ConstantSDNode>(V.getOperand(1))) {
1507 uint64_t Mask = V.getConstantOperandVal(1);
1508
1509 const SmallVector<ValueBit, 64> *LHSBits;
1510 // Mark this as interesting, only if the LHS was also interesting. This
1511 // prevents the overall procedure from matching a single immediate 'and'
1512 // (which is non-optimal because such an and might be folded with other
1513 // things if we don't select it here).
1514 std::tie(Interesting, LHSBits) = getValueBits(V.getOperand(0), NumBits);
1515
1516 for (unsigned i = 0; i < NumBits; ++i)
1517 if (((Mask >> i) & 1) == 1)
1518 Bits[i] = (*LHSBits)[i];
1519 else {
1520 // AND instruction masks this bit. If the input is already zero,
1521 // we have nothing to do here. Otherwise, make the bit ConstZero.
1522 if ((*LHSBits)[i].isZero())
1523 Bits[i] = (*LHSBits)[i];
1524 else
1525 Bits[i] = ValueBit(ValueBit::ConstZero);
1526 }
1527
1528 return std::make_pair(Interesting, &Bits);
1529 }
1530 break;
1531 case ISD::OR: {
1532 const auto &LHSBits = *getValueBits(V.getOperand(0), NumBits).second;
1533 const auto &RHSBits = *getValueBits(V.getOperand(1), NumBits).second;
1534
1535 bool AllDisjoint = true;
1536 SDValue LastVal = SDValue();
1537 unsigned LastIdx = 0;
1538 for (unsigned i = 0; i < NumBits; ++i) {
1539 if (LHSBits[i].isZero() && RHSBits[i].isZero()) {
1540 // If both inputs are known to be zero and one is ConstZero and
1541 // another is VariableKnownToBeZero, we can select whichever
1542 // we like. To minimize the number of bit groups, we select
1543 // VariableKnownToBeZero if this bit is the next bit of the same
1544 // input variable from the previous bit. Otherwise, we select
1545 // ConstZero.
1546 if (LHSBits[i].hasValue() && LHSBits[i].getValue() == LastVal &&
1547 LHSBits[i].getValueBitIndex() == LastIdx + 1)
1548 Bits[i] = LHSBits[i];
1549 else if (RHSBits[i].hasValue() && RHSBits[i].getValue() == LastVal &&
1550 RHSBits[i].getValueBitIndex() == LastIdx + 1)
1551 Bits[i] = RHSBits[i];
1552 else
1553 Bits[i] = ValueBit(ValueBit::ConstZero);
1554 }
1555 else if (LHSBits[i].isZero())
1556 Bits[i] = RHSBits[i];
1557 else if (RHSBits[i].isZero())
1558 Bits[i] = LHSBits[i];
1559 else {
1560 AllDisjoint = false;
1561 break;
1562 }
1563 // We remember the value and bit index of this bit.
1564 if (Bits[i].hasValue()) {
1565 LastVal = Bits[i].getValue();
1566 LastIdx = Bits[i].getValueBitIndex();
1567 }
1568 else {
1569 if (LastVal) LastVal = SDValue();
1570 LastIdx = 0;
1571 }
1572 }
1573
1574 if (!AllDisjoint)
1575 break;
1576
1577 return std::make_pair(Interesting = true, &Bits);
1578 }
1579 case ISD::ZERO_EXTEND: {
1580 // We support only the case with zero extension from i32 to i64 so far.
1581 if (V.getValueType() != MVT::i64 ||
1582 V.getOperand(0).getValueType() != MVT::i32)
1583 break;
1584
1585 const SmallVector<ValueBit, 64> *LHSBits;
1586 const unsigned NumOperandBits = 32;
1587 std::tie(Interesting, LHSBits) = getValueBits(V.getOperand(0),
1588 NumOperandBits);
1589
1590 for (unsigned i = 0; i < NumOperandBits; ++i)
1591 Bits[i] = (*LHSBits)[i];
1592
1593 for (unsigned i = NumOperandBits; i < NumBits; ++i)
1594 Bits[i] = ValueBit(ValueBit::ConstZero);
1595
1596 return std::make_pair(Interesting, &Bits);
1597 }
1598 case ISD::TRUNCATE: {
1599 EVT FromType = V.getOperand(0).getValueType();
1600 EVT ToType = V.getValueType();
1601 // We support only the case with truncate from i64 to i32.
1602 if (FromType != MVT::i64 || ToType != MVT::i32)
1603 break;
1604 const unsigned NumAllBits = FromType.getSizeInBits();
1605 SmallVector<ValueBit, 64> *InBits;
1606 std::tie(Interesting, InBits) = getValueBits(V.getOperand(0),
1607 NumAllBits);
1608 const unsigned NumValidBits = ToType.getSizeInBits();
1609
1610 // A 32-bit instruction cannot touch upper 32-bit part of 64-bit value.
1611 // So, we cannot include this truncate.
1612 bool UseUpper32bit = false;
1613 for (unsigned i = 0; i < NumValidBits; ++i)
1614 if ((*InBits)[i].hasValue() && (*InBits)[i].getValueBitIndex() >= 32) {
1615 UseUpper32bit = true;
1616 break;
1617 }
1618 if (UseUpper32bit)
1619 break;
1620
1621 for (unsigned i = 0; i < NumValidBits; ++i)
1622 Bits[i] = (*InBits)[i];
1623
1624 return std::make_pair(Interesting, &Bits);
1625 }
1626 case ISD::AssertZext: {
1627 // For AssertZext, we look through the operand and
1628 // mark the bits known to be zero.
1629 const SmallVector<ValueBit, 64> *LHSBits;
1630 std::tie(Interesting, LHSBits) = getValueBits(V.getOperand(0),
1631 NumBits);
1632
1633 EVT FromType = cast<VTSDNode>(V.getOperand(1))->getVT();
1634 const unsigned NumValidBits = FromType.getSizeInBits();
1635 for (unsigned i = 0; i < NumValidBits; ++i)
1636 Bits[i] = (*LHSBits)[i];
1637
1638 // These bits are known to be zero but the AssertZext may be from a value
1639 // that already has some constant zero bits (i.e. from a masking and).
1640 for (unsigned i = NumValidBits; i < NumBits; ++i)
1641 Bits[i] = (*LHSBits)[i].hasValue()
1642 ? ValueBit((*LHSBits)[i].getValue(),
1643 (*LHSBits)[i].getValueBitIndex(),
1644 ValueBit::VariableKnownToBeZero)
1645 : ValueBit(ValueBit::ConstZero);
1646
1647 return std::make_pair(Interesting, &Bits);
1648 }
1649 case ISD::LOAD:
1650 LoadSDNode *LD = cast<LoadSDNode>(V);
1651 if (ISD::isZEXTLoad(V.getNode()) && V.getResNo() == 0) {
1652 EVT VT = LD->getMemoryVT();
1653 const unsigned NumValidBits = VT.getSizeInBits();
1654
1655 for (unsigned i = 0; i < NumValidBits; ++i)
1656 Bits[i] = ValueBit(V, i);
1657
1658 // These bits are known to be zero.
1659 for (unsigned i = NumValidBits; i < NumBits; ++i)
1660 Bits[i] = ValueBit(V, i, ValueBit::VariableKnownToBeZero);
1661
1662 // Zero-extending load itself cannot be optimized. So, it is not
1663 // interesting by itself though it gives useful information.
1664 return std::make_pair(Interesting = false, &Bits);
1665 }
1666 break;
1667 }
1668
1669 for (unsigned i = 0; i < NumBits; ++i)
1670 Bits[i] = ValueBit(V, i);
1671
1672 return std::make_pair(Interesting = false, &Bits);
1673 }
1674
1675 // For each value (except the constant ones), compute the left-rotate amount
1676 // to get it from its original to final position.
1677 void computeRotationAmounts() {
1678 NeedMask = false;
1679 RLAmt.resize(Bits.size());
1680 for (unsigned i = 0; i < Bits.size(); ++i)
1681 if (Bits[i].hasValue()) {
1682 unsigned VBI = Bits[i].getValueBitIndex();
1683 if (i >= VBI)
1684 RLAmt[i] = i - VBI;
1685 else
1686 RLAmt[i] = Bits.size() - (VBI - i);
1687 } else if (Bits[i].isZero()) {
1688 NeedMask = true;
1689 RLAmt[i] = UINT32_MAX(4294967295U);
1690 } else {
1691 llvm_unreachable("Unknown value bit type")::llvm::llvm_unreachable_internal("Unknown value bit type", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 1691)
;
1692 }
1693 }
1694
1695 // Collect groups of consecutive bits with the same underlying value and
1696 // rotation factor. If we're doing late masking, we ignore zeros, otherwise
1697 // they break up groups.
1698 void collectBitGroups(bool LateMask) {
1699 BitGroups.clear();
1700
1701 unsigned LastRLAmt = RLAmt[0];
1702 SDValue LastValue = Bits[0].hasValue() ? Bits[0].getValue() : SDValue();
1703 unsigned LastGroupStartIdx = 0;
1704 bool IsGroupOfZeros = !Bits[LastGroupStartIdx].hasValue();
1705 for (unsigned i = 1; i < Bits.size(); ++i) {
1706 unsigned ThisRLAmt = RLAmt[i];
1707 SDValue ThisValue = Bits[i].hasValue() ? Bits[i].getValue() : SDValue();
1708 if (LateMask && !ThisValue) {
1709 ThisValue = LastValue;
1710 ThisRLAmt = LastRLAmt;
1711 // If we're doing late masking, then the first bit group always starts
1712 // at zero (even if the first bits were zero).
1713 if (BitGroups.empty())
1714 LastGroupStartIdx = 0;
1715 }
1716
1717 // If this bit is known to be zero and the current group is a bit group
1718 // of zeros, we do not need to terminate the current bit group even the
1719 // Value or RLAmt does not match here. Instead, we terminate this group
1720 // when the first non-zero bit appears later.
1721 if (IsGroupOfZeros && Bits[i].isZero())
1722 continue;
1723
1724 // If this bit has the same underlying value and the same rotate factor as
1725 // the last one, then they're part of the same group.
1726 if (ThisRLAmt == LastRLAmt && ThisValue == LastValue)
1727 // We cannot continue the current group if this bits is not known to
1728 // be zero in a bit group of zeros.
1729 if (!(IsGroupOfZeros && ThisValue && !Bits[i].isZero()))
1730 continue;
1731
1732 if (LastValue.getNode())
1733 BitGroups.push_back(BitGroup(LastValue, LastRLAmt, LastGroupStartIdx,
1734 i-1));
1735 LastRLAmt = ThisRLAmt;
1736 LastValue = ThisValue;
1737 LastGroupStartIdx = i;
1738 IsGroupOfZeros = !Bits[LastGroupStartIdx].hasValue();
1739 }
1740 if (LastValue.getNode())
1741 BitGroups.push_back(BitGroup(LastValue, LastRLAmt, LastGroupStartIdx,
1742 Bits.size()-1));
1743
1744 if (BitGroups.empty())
1745 return;
1746
1747 // We might be able to combine the first and last groups.
1748 if (BitGroups.size() > 1) {
1749 // If the first and last groups are the same, then remove the first group
1750 // in favor of the last group, making the ending index of the last group
1751 // equal to the ending index of the to-be-removed first group.
1752 if (BitGroups[0].StartIdx == 0 &&
1753 BitGroups[BitGroups.size()-1].EndIdx == Bits.size()-1 &&
1754 BitGroups[0].V == BitGroups[BitGroups.size()-1].V &&
1755 BitGroups[0].RLAmt == BitGroups[BitGroups.size()-1].RLAmt) {
1756 LLVM_DEBUG(dbgs() << "\tcombining final bit group with initial one\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining final bit group with initial one\n"
; } } while (false)
;
1757 BitGroups[BitGroups.size()-1].EndIdx = BitGroups[0].EndIdx;
1758 BitGroups.erase(BitGroups.begin());
1759 }
1760 }
1761 }
1762
1763 // Take all (SDValue, RLAmt) pairs and sort them by the number of groups
1764 // associated with each. If the number of groups are same, we prefer a group
1765 // which does not require rotate, i.e. RLAmt is 0, to avoid the first rotate
1766 // instruction. If there is a degeneracy, pick the one that occurs
1767 // first (in the final value).
1768 void collectValueRotInfo() {
1769 ValueRots.clear();
1770
1771 for (auto &BG : BitGroups) {
1772 unsigned RLAmtKey = BG.RLAmt + (BG.Repl32 ? 64 : 0);
1773 ValueRotInfo &VRI = ValueRots[std::make_pair(BG.V, RLAmtKey)];
1774 VRI.V = BG.V;
1775 VRI.RLAmt = BG.RLAmt;
1776 VRI.Repl32 = BG.Repl32;
1777 VRI.NumGroups += 1;
1778 VRI.FirstGroupStartIdx = std::min(VRI.FirstGroupStartIdx, BG.StartIdx);
1779 }
1780
1781 // Now that we've collected the various ValueRotInfo instances, we need to
1782 // sort them.
1783 ValueRotsVec.clear();
1784 for (auto &I : ValueRots) {
1785 ValueRotsVec.push_back(I.second);
1786 }
1787 llvm::sort(ValueRotsVec);
1788 }
1789
1790 // In 64-bit mode, rlwinm and friends have a rotation operator that
1791 // replicates the low-order 32 bits into the high-order 32-bits. The mask
1792 // indices of these instructions can only be in the lower 32 bits, so they
1793 // can only represent some 64-bit bit groups. However, when they can be used,
1794 // the 32-bit replication can be used to represent, as a single bit group,
1795 // otherwise separate bit groups. We'll convert to replicated-32-bit bit
1796 // groups when possible. Returns true if any of the bit groups were
1797 // converted.
1798 void assignRepl32BitGroups() {
1799 // If we have bits like this:
1800 //
1801 // Indices: 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
1802 // V bits: ... 7 6 5 4 3 2 1 0 31 30 29 28 27 26 25 24
1803 // Groups: | RLAmt = 8 | RLAmt = 40 |
1804 //
1805 // But, making use of a 32-bit operation that replicates the low-order 32
1806 // bits into the high-order 32 bits, this can be one bit group with a RLAmt
1807 // of 8.
1808
1809 auto IsAllLow32 = [this](BitGroup & BG) {
1810 if (BG.StartIdx <= BG.EndIdx) {
1811 for (unsigned i = BG.StartIdx; i <= BG.EndIdx; ++i) {
1812 if (!Bits[i].hasValue())
1813 continue;
1814 if (Bits[i].getValueBitIndex() >= 32)
1815 return false;
1816 }
1817 } else {
1818 for (unsigned i = BG.StartIdx; i < Bits.size(); ++i) {
1819 if (!Bits[i].hasValue())
1820 continue;
1821 if (Bits[i].getValueBitIndex() >= 32)
1822 return false;
1823 }
1824 for (unsigned i = 0; i <= BG.EndIdx; ++i) {
1825 if (!Bits[i].hasValue())
1826 continue;
1827 if (Bits[i].getValueBitIndex() >= 32)
1828 return false;
1829 }
1830 }
1831
1832 return true;
1833 };
1834
1835 for (auto &BG : BitGroups) {
1836 // If this bit group has RLAmt of 0 and will not be merged with
1837 // another bit group, we don't benefit from Repl32. We don't mark
1838 // such group to give more freedom for later instruction selection.
1839 if (BG.RLAmt == 0) {
1840 auto PotentiallyMerged = [this](BitGroup & BG) {
1841 for (auto &BG2 : BitGroups)
1842 if (&BG != &BG2 && BG.V == BG2.V &&
1843 (BG2.RLAmt == 0 || BG2.RLAmt == 32))
1844 return true;
1845 return false;
1846 };
1847 if (!PotentiallyMerged(BG))
1848 continue;
1849 }
1850 if (BG.StartIdx < 32 && BG.EndIdx < 32) {
1851 if (IsAllLow32(BG)) {
1852 if (BG.RLAmt >= 32) {
1853 BG.RLAmt -= 32;
1854 BG.Repl32CR = true;
1855 }
1856
1857 BG.Repl32 = true;
1858
1859 LLVM_DEBUG(dbgs() << "\t32-bit replicated bit group for "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t32-bit replicated bit group for "
<< BG.V.getNode() << " RLAmt = " << BG.RLAmt
<< " [" << BG.StartIdx << ", " << BG
.EndIdx << "]\n"; } } while (false)
1860 << BG.V.getNode() << " RLAmt = " << BG.RLAmt << " ["do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t32-bit replicated bit group for "
<< BG.V.getNode() << " RLAmt = " << BG.RLAmt
<< " [" << BG.StartIdx << ", " << BG
.EndIdx << "]\n"; } } while (false)
1861 << BG.StartIdx << ", " << BG.EndIdx << "]\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t32-bit replicated bit group for "
<< BG.V.getNode() << " RLAmt = " << BG.RLAmt
<< " [" << BG.StartIdx << ", " << BG
.EndIdx << "]\n"; } } while (false)
;
1862 }
1863 }
1864 }
1865
1866 // Now walk through the bit groups, consolidating where possible.
1867 for (auto I = BitGroups.begin(); I != BitGroups.end();) {
1868 // We might want to remove this bit group by merging it with the previous
1869 // group (which might be the ending group).
1870 auto IP = (I == BitGroups.begin()) ?
1871 std::prev(BitGroups.end()) : std::prev(I);
1872 if (I->Repl32 && IP->Repl32 && I->V == IP->V && I->RLAmt == IP->RLAmt &&
1873 I->StartIdx == (IP->EndIdx + 1) % 64 && I != IP) {
1874
1875 LLVM_DEBUG(dbgs() << "\tcombining 32-bit replicated bit group for "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining 32-bit replicated bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with group with range [" << IP
->StartIdx << ", " << IP->EndIdx << "]\n"
; } } while (false)
1876 << I->V.getNode() << " RLAmt = " << I->RLAmt << " ["do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining 32-bit replicated bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with group with range [" << IP
->StartIdx << ", " << IP->EndIdx << "]\n"
; } } while (false)
1877 << I->StartIdx << ", " << I->EndIdxdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining 32-bit replicated bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with group with range [" << IP
->StartIdx << ", " << IP->EndIdx << "]\n"
; } } while (false)
1878 << "] with group with range [" << IP->StartIdx << ", "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining 32-bit replicated bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with group with range [" << IP
->StartIdx << ", " << IP->EndIdx << "]\n"
; } } while (false)
1879 << IP->EndIdx << "]\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining 32-bit replicated bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with group with range [" << IP
->StartIdx << ", " << IP->EndIdx << "]\n"
; } } while (false)
;
1880
1881 IP->EndIdx = I->EndIdx;
1882 IP->Repl32CR = IP->Repl32CR || I->Repl32CR;
1883 IP->Repl32Coalesced = true;
1884 I = BitGroups.erase(I);
1885 continue;
1886 } else {
1887 // There is a special case worth handling: If there is a single group
1888 // covering the entire upper 32 bits, and it can be merged with both
1889 // the next and previous groups (which might be the same group), then
1890 // do so. If it is the same group (so there will be only one group in
1891 // total), then we need to reverse the order of the range so that it
1892 // covers the entire 64 bits.
1893 if (I->StartIdx == 32 && I->EndIdx == 63) {
1894 assert(std::next(I) == BitGroups.end() &&(static_cast <bool> (std::next(I) == BitGroups.end() &&
"bit group ends at index 63 but there is another?") ? void (
0) : __assert_fail ("std::next(I) == BitGroups.end() && \"bit group ends at index 63 but there is another?\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 1895, __extension__ __PRETTY_FUNCTION__))
1895 "bit group ends at index 63 but there is another?")(static_cast <bool> (std::next(I) == BitGroups.end() &&
"bit group ends at index 63 but there is another?") ? void (
0) : __assert_fail ("std::next(I) == BitGroups.end() && \"bit group ends at index 63 but there is another?\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 1895, __extension__ __PRETTY_FUNCTION__))
;
1896 auto IN = BitGroups.begin();
1897
1898 if (IP->Repl32 && IN->Repl32 && I->V == IP->V && I->V == IN->V &&
1899 (I->RLAmt % 32) == IP->RLAmt && (I->RLAmt % 32) == IN->RLAmt &&
1900 IP->EndIdx == 31 && IN->StartIdx == 0 && I != IP &&
1901 IsAllLow32(*I)) {
1902
1903 LLVM_DEBUG(dbgs() << "\tcombining bit group for " << I->V.getNode()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with 32-bit replicated groups with ranges ["
<< IP->StartIdx << ", " << IP->EndIdx
<< "] and [" << IN->StartIdx << ", " <<
IN->EndIdx << "]\n"; } } while (false)
1904 << " RLAmt = " << I->RLAmt << " [" << I->StartIdxdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with 32-bit replicated groups with ranges ["
<< IP->StartIdx << ", " << IP->EndIdx
<< "] and [" << IN->StartIdx << ", " <<
IN->EndIdx << "]\n"; } } while (false)
1905 << ", " << I->EndIdxdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with 32-bit replicated groups with ranges ["
<< IP->StartIdx << ", " << IP->EndIdx
<< "] and [" << IN->StartIdx << ", " <<
IN->EndIdx << "]\n"; } } while (false)
1906 << "] with 32-bit replicated groups with ranges ["do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with 32-bit replicated groups with ranges ["
<< IP->StartIdx << ", " << IP->EndIdx
<< "] and [" << IN->StartIdx << ", " <<
IN->EndIdx << "]\n"; } } while (false)
1907 << IP->StartIdx << ", " << IP->EndIdx << "] and ["do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with 32-bit replicated groups with ranges ["
<< IP->StartIdx << ", " << IP->EndIdx
<< "] and [" << IN->StartIdx << ", " <<
IN->EndIdx << "]\n"; } } while (false)
1908 << IN->StartIdx << ", " << IN->EndIdx << "]\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tcombining bit group for "
<< I->V.getNode() << " RLAmt = " << I->
RLAmt << " [" << I->StartIdx << ", " <<
I->EndIdx << "] with 32-bit replicated groups with ranges ["
<< IP->StartIdx << ", " << IP->EndIdx
<< "] and [" << IN->StartIdx << ", " <<
IN->EndIdx << "]\n"; } } while (false)
;
1909
1910 if (IP == IN) {
1911 // There is only one other group; change it to cover the whole
1912 // range (backward, so that it can still be Repl32 but cover the
1913 // whole 64-bit range).
1914 IP->StartIdx = 31;
1915 IP->EndIdx = 30;
1916 IP->Repl32CR = IP->Repl32CR || I->RLAmt >= 32;
1917 IP->Repl32Coalesced = true;
1918 I = BitGroups.erase(I);
1919 } else {
1920 // There are two separate groups, one before this group and one
1921 // after us (at the beginning). We're going to remove this group,
1922 // but also the group at the very beginning.
1923 IP->EndIdx = IN->EndIdx;
1924 IP->Repl32CR = IP->Repl32CR || IN->Repl32CR || I->RLAmt >= 32;
1925 IP->Repl32Coalesced = true;
1926 I = BitGroups.erase(I);
1927 BitGroups.erase(BitGroups.begin());
1928 }
1929
1930 // This must be the last group in the vector (and we might have
1931 // just invalidated the iterator above), so break here.
1932 break;
1933 }
1934 }
1935 }
1936
1937 ++I;
1938 }
1939 }
1940
1941 SDValue getI32Imm(unsigned Imm, const SDLoc &dl) {
1942 return CurDAG->getTargetConstant(Imm, dl, MVT::i32);
1943 }
1944
1945 uint64_t getZerosMask() {
1946 uint64_t Mask = 0;
1947 for (unsigned i = 0; i < Bits.size(); ++i) {
1948 if (Bits[i].hasValue())
1949 continue;
1950 Mask |= (UINT64_C(1)1UL << i);
1951 }
1952
1953 return ~Mask;
1954 }
1955
1956 // This method extends an input value to 64 bit if input is 32-bit integer.
1957 // While selecting instructions in BitPermutationSelector in 64-bit mode,
1958 // an input value can be a 32-bit integer if a ZERO_EXTEND node is included.
1959 // In such case, we extend it to 64 bit to be consistent with other values.
1960 SDValue ExtendToInt64(SDValue V, const SDLoc &dl) {
1961 if (V.getValueSizeInBits() == 64)
1962 return V;
1963
1964 assert(V.getValueSizeInBits() == 32)(static_cast <bool> (V.getValueSizeInBits() == 32) ? void
(0) : __assert_fail ("V.getValueSizeInBits() == 32", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 1964, __extension__ __PRETTY_FUNCTION__))
;
1965 SDValue SubRegIdx = CurDAG->getTargetConstant(PPC::sub_32, dl, MVT::i32);
1966 SDValue ImDef = SDValue(CurDAG->getMachineNode(PPC::IMPLICIT_DEF, dl,
1967 MVT::i64), 0);
1968 SDValue ExtVal = SDValue(CurDAG->getMachineNode(PPC::INSERT_SUBREG, dl,
1969 MVT::i64, ImDef, V,
1970 SubRegIdx), 0);
1971 return ExtVal;
1972 }
1973
1974 SDValue TruncateToInt32(SDValue V, const SDLoc &dl) {
1975 if (V.getValueSizeInBits() == 32)
1976 return V;
1977
1978 assert(V.getValueSizeInBits() == 64)(static_cast <bool> (V.getValueSizeInBits() == 64) ? void
(0) : __assert_fail ("V.getValueSizeInBits() == 64", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 1978, __extension__ __PRETTY_FUNCTION__))
;
1979 SDValue SubRegIdx = CurDAG->getTargetConstant(PPC::sub_32, dl, MVT::i32);
1980 SDValue SubVal = SDValue(CurDAG->getMachineNode(PPC::EXTRACT_SUBREG, dl,
1981 MVT::i32, V, SubRegIdx), 0);
1982 return SubVal;
1983 }
1984
1985 // Depending on the number of groups for a particular value, it might be
1986 // better to rotate, mask explicitly (using andi/andis), and then or the
1987 // result. Select this part of the result first.
1988 void SelectAndParts32(const SDLoc &dl, SDValue &Res, unsigned *InstCnt) {
1989 if (BPermRewriterNoMasking)
1990 return;
1991
1992 for (ValueRotInfo &VRI : ValueRotsVec) {
1993 unsigned Mask = 0;
1994 for (unsigned i = 0; i < Bits.size(); ++i) {
1995 if (!Bits[i].hasValue() || Bits[i].getValue() != VRI.V)
1996 continue;
1997 if (RLAmt[i] != VRI.RLAmt)
1998 continue;
1999 Mask |= (1u << i);
2000 }
2001
2002 // Compute the masks for andi/andis that would be necessary.
2003 unsigned ANDIMask = (Mask & UINT16_MAX(65535)), ANDISMask = Mask >> 16;
2004 assert((ANDIMask != 0 || ANDISMask != 0) &&(static_cast <bool> ((ANDIMask != 0 || ANDISMask != 0) &&
"No set bits in mask for value bit groups") ? void (0) : __assert_fail
("(ANDIMask != 0 || ANDISMask != 0) && \"No set bits in mask for value bit groups\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2005, __extension__ __PRETTY_FUNCTION__))
2005 "No set bits in mask for value bit groups")(static_cast <bool> ((ANDIMask != 0 || ANDISMask != 0) &&
"No set bits in mask for value bit groups") ? void (0) : __assert_fail
("(ANDIMask != 0 || ANDISMask != 0) && \"No set bits in mask for value bit groups\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2005, __extension__ __PRETTY_FUNCTION__))
;
2006 bool NeedsRotate = VRI.RLAmt != 0;
2007
2008 // We're trying to minimize the number of instructions. If we have one
2009 // group, using one of andi/andis can break even. If we have three
2010 // groups, we can use both andi and andis and break even (to use both
2011 // andi and andis we also need to or the results together). We need four
2012 // groups if we also need to rotate. To use andi/andis we need to do more
2013 // than break even because rotate-and-mask instructions tend to be easier
2014 // to schedule.
2015
2016 // FIXME: We've biased here against using andi/andis, which is right for
2017 // POWER cores, but not optimal everywhere. For example, on the A2,
2018 // andi/andis have single-cycle latency whereas the rotate-and-mask
2019 // instructions take two cycles, and it would be better to bias toward
2020 // andi/andis in break-even cases.
2021
2022 unsigned NumAndInsts = (unsigned) NeedsRotate +
2023 (unsigned) (ANDIMask != 0) +
2024 (unsigned) (ANDISMask != 0) +
2025 (unsigned) (ANDIMask != 0 && ANDISMask != 0) +
2026 (unsigned) (bool) Res;
2027
2028 LLVM_DEBUG(dbgs() << "\t\trotation groups for " << VRI.V.getNode()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\trotation groups for "
<< VRI.V.getNode() << " RL: " << VRI.RLAmt
<< ":" << "\n\t\t\tisel using masking: " <<
NumAndInsts << " using rotates: " << VRI.NumGroups
<< "\n"; } } while (false)
2029 << " RL: " << VRI.RLAmt << ":"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\trotation groups for "
<< VRI.V.getNode() << " RL: " << VRI.RLAmt
<< ":" << "\n\t\t\tisel using masking: " <<
NumAndInsts << " using rotates: " << VRI.NumGroups
<< "\n"; } } while (false)
2030 << "\n\t\t\tisel using masking: " << NumAndInstsdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\trotation groups for "
<< VRI.V.getNode() << " RL: " << VRI.RLAmt
<< ":" << "\n\t\t\tisel using masking: " <<
NumAndInsts << " using rotates: " << VRI.NumGroups
<< "\n"; } } while (false)
2031 << " using rotates: " << VRI.NumGroups << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\trotation groups for "
<< VRI.V.getNode() << " RL: " << VRI.RLAmt
<< ":" << "\n\t\t\tisel using masking: " <<
NumAndInsts << " using rotates: " << VRI.NumGroups
<< "\n"; } } while (false)
;
2032
2033 if (NumAndInsts >= VRI.NumGroups)
2034 continue;
2035
2036 LLVM_DEBUG(dbgs() << "\t\t\t\tusing masking\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\t\t\tusing masking\n";
} } while (false)
;
2037
2038 if (InstCnt) *InstCnt += NumAndInsts;
2039
2040 SDValue VRot;
2041 if (VRI.RLAmt) {
2042 SDValue Ops[] =
2043 { TruncateToInt32(VRI.V, dl), getI32Imm(VRI.RLAmt, dl),
2044 getI32Imm(0, dl), getI32Imm(31, dl) };
2045 VRot = SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32,
2046 Ops), 0);
2047 } else {
2048 VRot = TruncateToInt32(VRI.V, dl);
2049 }
2050
2051 SDValue ANDIVal, ANDISVal;
2052 if (ANDIMask != 0)
2053 ANDIVal = SDValue(CurDAG->getMachineNode(PPC::ANDI_rec, dl, MVT::i32,
2054 VRot, getI32Imm(ANDIMask, dl)),
2055 0);
2056 if (ANDISMask != 0)
2057 ANDISVal =
2058 SDValue(CurDAG->getMachineNode(PPC::ANDIS_rec, dl, MVT::i32, VRot,
2059 getI32Imm(ANDISMask, dl)),
2060 0);
2061
2062 SDValue TotalVal;
2063 if (!ANDIVal)
2064 TotalVal = ANDISVal;
2065 else if (!ANDISVal)
2066 TotalVal = ANDIVal;
2067 else
2068 TotalVal = SDValue(CurDAG->getMachineNode(PPC::OR, dl, MVT::i32,
2069 ANDIVal, ANDISVal), 0);
2070
2071 if (!Res)
2072 Res = TotalVal;
2073 else
2074 Res = SDValue(CurDAG->getMachineNode(PPC::OR, dl, MVT::i32,
2075 Res, TotalVal), 0);
2076
2077 // Now, remove all groups with this underlying value and rotation
2078 // factor.
2079 eraseMatchingBitGroups([VRI](const BitGroup &BG) {
2080 return BG.V == VRI.V && BG.RLAmt == VRI.RLAmt;
2081 });
2082 }
2083 }
2084
2085 // Instruction selection for the 32-bit case.
2086 SDNode *Select32(SDNode *N, bool LateMask, unsigned *InstCnt) {
2087 SDLoc dl(N);
2088 SDValue Res;
2089
2090 if (InstCnt) *InstCnt = 0;
2091
2092 // Take care of cases that should use andi/andis first.
2093 SelectAndParts32(dl, Res, InstCnt);
2094
2095 // If we've not yet selected a 'starting' instruction, and we have no zeros
2096 // to fill in, select the (Value, RLAmt) with the highest priority (largest
2097 // number of groups), and start with this rotated value.
2098 if ((!NeedMask || LateMask) && !Res) {
2099 ValueRotInfo &VRI = ValueRotsVec[0];
2100 if (VRI.RLAmt) {
2101 if (InstCnt) *InstCnt += 1;
2102 SDValue Ops[] =
2103 { TruncateToInt32(VRI.V, dl), getI32Imm(VRI.RLAmt, dl),
2104 getI32Imm(0, dl), getI32Imm(31, dl) };
2105 Res = SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, Ops),
2106 0);
2107 } else {
2108 Res = TruncateToInt32(VRI.V, dl);
2109 }
2110
2111 // Now, remove all groups with this underlying value and rotation factor.
2112 eraseMatchingBitGroups([VRI](const BitGroup &BG) {
2113 return BG.V == VRI.V && BG.RLAmt == VRI.RLAmt;
2114 });
2115 }
2116
2117 if (InstCnt) *InstCnt += BitGroups.size();
2118
2119 // Insert the other groups (one at a time).
2120 for (auto &BG : BitGroups) {
2121 if (!Res) {
2122 SDValue Ops[] =
2123 { TruncateToInt32(BG.V, dl), getI32Imm(BG.RLAmt, dl),
2124 getI32Imm(Bits.size() - BG.EndIdx - 1, dl),
2125 getI32Imm(Bits.size() - BG.StartIdx - 1, dl) };
2126 Res = SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, Ops), 0);
2127 } else {
2128 SDValue Ops[] =
2129 { Res, TruncateToInt32(BG.V, dl), getI32Imm(BG.RLAmt, dl),
2130 getI32Imm(Bits.size() - BG.EndIdx - 1, dl),
2131 getI32Imm(Bits.size() - BG.StartIdx - 1, dl) };
2132 Res = SDValue(CurDAG->getMachineNode(PPC::RLWIMI, dl, MVT::i32, Ops), 0);
2133 }
2134 }
2135
2136 if (LateMask) {
2137 unsigned Mask = (unsigned) getZerosMask();
2138
2139 unsigned ANDIMask = (Mask & UINT16_MAX(65535)), ANDISMask = Mask >> 16;
2140 assert((ANDIMask != 0 || ANDISMask != 0) &&(static_cast <bool> ((ANDIMask != 0 || ANDISMask != 0) &&
"No set bits in zeros mask?") ? void (0) : __assert_fail ("(ANDIMask != 0 || ANDISMask != 0) && \"No set bits in zeros mask?\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2141, __extension__ __PRETTY_FUNCTION__))
2141 "No set bits in zeros mask?")(static_cast <bool> ((ANDIMask != 0 || ANDISMask != 0) &&
"No set bits in zeros mask?") ? void (0) : __assert_fail ("(ANDIMask != 0 || ANDISMask != 0) && \"No set bits in zeros mask?\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2141, __extension__ __PRETTY_FUNCTION__))
;
2142
2143 if (InstCnt) *InstCnt += (unsigned) (ANDIMask != 0) +
2144 (unsigned) (ANDISMask != 0) +
2145 (unsigned) (ANDIMask != 0 && ANDISMask != 0);
2146
2147 SDValue ANDIVal, ANDISVal;
2148 if (ANDIMask != 0)
2149 ANDIVal = SDValue(CurDAG->getMachineNode(PPC::ANDI_rec, dl, MVT::i32,
2150 Res, getI32Imm(ANDIMask, dl)),
2151 0);
2152 if (ANDISMask != 0)
2153 ANDISVal =
2154 SDValue(CurDAG->getMachineNode(PPC::ANDIS_rec, dl, MVT::i32, Res,
2155 getI32Imm(ANDISMask, dl)),
2156 0);
2157
2158 if (!ANDIVal)
2159 Res = ANDISVal;
2160 else if (!ANDISVal)
2161 Res = ANDIVal;
2162 else
2163 Res = SDValue(CurDAG->getMachineNode(PPC::OR, dl, MVT::i32,
2164 ANDIVal, ANDISVal), 0);
2165 }
2166
2167 return Res.getNode();
2168 }
2169
2170 unsigned SelectRotMask64Count(unsigned RLAmt, bool Repl32,
2171 unsigned MaskStart, unsigned MaskEnd,
2172 bool IsIns) {
2173 // In the notation used by the instructions, 'start' and 'end' are reversed
2174 // because bits are counted from high to low order.
2175 unsigned InstMaskStart = 64 - MaskEnd - 1,
2176 InstMaskEnd = 64 - MaskStart - 1;
2177
2178 if (Repl32)
2179 return 1;
2180
2181 if ((!IsIns && (InstMaskEnd == 63 || InstMaskStart == 0)) ||
2182 InstMaskEnd == 63 - RLAmt)
2183 return 1;
2184
2185 return 2;
2186 }
2187
2188 // For 64-bit values, not all combinations of rotates and masks are
2189 // available. Produce one if it is available.
2190 SDValue SelectRotMask64(SDValue V, const SDLoc &dl, unsigned RLAmt,
2191 bool Repl32, unsigned MaskStart, unsigned MaskEnd,
2192 unsigned *InstCnt = nullptr) {
2193 // In the notation used by the instructions, 'start' and 'end' are reversed
2194 // because bits are counted from high to low order.
2195 unsigned InstMaskStart = 64 - MaskEnd - 1,
2196 InstMaskEnd = 64 - MaskStart - 1;
2197
2198 if (InstCnt) *InstCnt += 1;
2199
2200 if (Repl32) {
2201 // This rotation amount assumes that the lower 32 bits of the quantity
2202 // are replicated in the high 32 bits by the rotation operator (which is
2203 // done by rlwinm and friends).
2204 assert(InstMaskStart >= 32 && "Mask cannot start out of range")(static_cast <bool> (InstMaskStart >= 32 && "Mask cannot start out of range"
) ? void (0) : __assert_fail ("InstMaskStart >= 32 && \"Mask cannot start out of range\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2204, __extension__ __PRETTY_FUNCTION__))
;
2205 assert(InstMaskEnd >= 32 && "Mask cannot end out of range")(static_cast <bool> (InstMaskEnd >= 32 && "Mask cannot end out of range"
) ? void (0) : __assert_fail ("InstMaskEnd >= 32 && \"Mask cannot end out of range\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2205, __extension__ __PRETTY_FUNCTION__))
;
2206 SDValue Ops[] =
2207 { ExtendToInt64(V, dl), getI32Imm(RLAmt, dl),
2208 getI32Imm(InstMaskStart - 32, dl), getI32Imm(InstMaskEnd - 32, dl) };
2209 return SDValue(CurDAG->getMachineNode(PPC::RLWINM8, dl, MVT::i64,
2210 Ops), 0);
2211 }
2212
2213 if (InstMaskEnd == 63) {
2214 SDValue Ops[] =
2215 { ExtendToInt64(V, dl), getI32Imm(RLAmt, dl),
2216 getI32Imm(InstMaskStart, dl) };
2217 return SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, Ops), 0);
2218 }
2219
2220 if (InstMaskStart == 0) {
2221 SDValue Ops[] =
2222 { ExtendToInt64(V, dl), getI32Imm(RLAmt, dl),
2223 getI32Imm(InstMaskEnd, dl) };
2224 return SDValue(CurDAG->getMachineNode(PPC::RLDICR, dl, MVT::i64, Ops), 0);
2225 }
2226
2227 if (InstMaskEnd == 63 - RLAmt) {
2228 SDValue Ops[] =
2229 { ExtendToInt64(V, dl), getI32Imm(RLAmt, dl),
2230 getI32Imm(InstMaskStart, dl) };
2231 return SDValue(CurDAG->getMachineNode(PPC::RLDIC, dl, MVT::i64, Ops), 0);
2232 }
2233
2234 // We cannot do this with a single instruction, so we'll use two. The
2235 // problem is that we're not free to choose both a rotation amount and mask
2236 // start and end independently. We can choose an arbitrary mask start and
2237 // end, but then the rotation amount is fixed. Rotation, however, can be
2238 // inverted, and so by applying an "inverse" rotation first, we can get the
2239 // desired result.
2240 if (InstCnt) *InstCnt += 1;
2241
2242 // The rotation mask for the second instruction must be MaskStart.
2243 unsigned RLAmt2 = MaskStart;
2244 // The first instruction must rotate V so that the overall rotation amount
2245 // is RLAmt.
2246 unsigned RLAmt1 = (64 + RLAmt - RLAmt2) % 64;
2247 if (RLAmt1)
2248 V = SelectRotMask64(V, dl, RLAmt1, false, 0, 63);
2249 return SelectRotMask64(V, dl, RLAmt2, false, MaskStart, MaskEnd);
2250 }
2251
2252 // For 64-bit values, not all combinations of rotates and masks are
2253 // available. Produce a rotate-mask-and-insert if one is available.
2254 SDValue SelectRotMaskIns64(SDValue Base, SDValue V, const SDLoc &dl,
2255 unsigned RLAmt, bool Repl32, unsigned MaskStart,
2256 unsigned MaskEnd, unsigned *InstCnt = nullptr) {
2257 // In the notation used by the instructions, 'start' and 'end' are reversed
2258 // because bits are counted from high to low order.
2259 unsigned InstMaskStart = 64 - MaskEnd - 1,
2260 InstMaskEnd = 64 - MaskStart - 1;
2261
2262 if (InstCnt) *InstCnt += 1;
2263
2264 if (Repl32) {
2265 // This rotation amount assumes that the lower 32 bits of the quantity
2266 // are replicated in the high 32 bits by the rotation operator (which is
2267 // done by rlwinm and friends).
2268 assert(InstMaskStart >= 32 && "Mask cannot start out of range")(static_cast <bool> (InstMaskStart >= 32 && "Mask cannot start out of range"
) ? void (0) : __assert_fail ("InstMaskStart >= 32 && \"Mask cannot start out of range\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2268, __extension__ __PRETTY_FUNCTION__))
;
2269 assert(InstMaskEnd >= 32 && "Mask cannot end out of range")(static_cast <bool> (InstMaskEnd >= 32 && "Mask cannot end out of range"
) ? void (0) : __assert_fail ("InstMaskEnd >= 32 && \"Mask cannot end out of range\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2269, __extension__ __PRETTY_FUNCTION__))
;
2270 SDValue Ops[] =
2271 { ExtendToInt64(Base, dl), ExtendToInt64(V, dl), getI32Imm(RLAmt, dl),
2272 getI32Imm(InstMaskStart - 32, dl), getI32Imm(InstMaskEnd - 32, dl) };
2273 return SDValue(CurDAG->getMachineNode(PPC::RLWIMI8, dl, MVT::i64,
2274 Ops), 0);
2275 }
2276
2277 if (InstMaskEnd == 63 - RLAmt) {
2278 SDValue Ops[] =
2279 { ExtendToInt64(Base, dl), ExtendToInt64(V, dl), getI32Imm(RLAmt, dl),
2280 getI32Imm(InstMaskStart, dl) };
2281 return SDValue(CurDAG->getMachineNode(PPC::RLDIMI, dl, MVT::i64, Ops), 0);
2282 }
2283
2284 // We cannot do this with a single instruction, so we'll use two. The
2285 // problem is that we're not free to choose both a rotation amount and mask
2286 // start and end independently. We can choose an arbitrary mask start and
2287 // end, but then the rotation amount is fixed. Rotation, however, can be
2288 // inverted, and so by applying an "inverse" rotation first, we can get the
2289 // desired result.
2290 if (InstCnt) *InstCnt += 1;
2291
2292 // The rotation mask for the second instruction must be MaskStart.
2293 unsigned RLAmt2 = MaskStart;
2294 // The first instruction must rotate V so that the overall rotation amount
2295 // is RLAmt.
2296 unsigned RLAmt1 = (64 + RLAmt - RLAmt2) % 64;
2297 if (RLAmt1)
2298 V = SelectRotMask64(V, dl, RLAmt1, false, 0, 63);
2299 return SelectRotMaskIns64(Base, V, dl, RLAmt2, false, MaskStart, MaskEnd);
2300 }
2301
2302 void SelectAndParts64(const SDLoc &dl, SDValue &Res, unsigned *InstCnt) {
2303 if (BPermRewriterNoMasking)
2304 return;
2305
2306 // The idea here is the same as in the 32-bit version, but with additional
2307 // complications from the fact that Repl32 might be true. Because we
2308 // aggressively convert bit groups to Repl32 form (which, for small
2309 // rotation factors, involves no other change), and then coalesce, it might
2310 // be the case that a single 64-bit masking operation could handle both
2311 // some Repl32 groups and some non-Repl32 groups. If converting to Repl32
2312 // form allowed coalescing, then we must use a 32-bit rotaton in order to
2313 // completely capture the new combined bit group.
2314
2315 for (ValueRotInfo &VRI : ValueRotsVec) {
2316 uint64_t Mask = 0;
2317
2318 // We need to add to the mask all bits from the associated bit groups.
2319 // If Repl32 is false, we need to add bits from bit groups that have
2320 // Repl32 true, but are trivially convertable to Repl32 false. Such a
2321 // group is trivially convertable if it overlaps only with the lower 32
2322 // bits, and the group has not been coalesced.
2323 auto MatchingBG = [VRI](const BitGroup &BG) {
2324 if (VRI.V != BG.V)
2325 return false;
2326
2327 unsigned EffRLAmt = BG.RLAmt;
2328 if (!VRI.Repl32 && BG.Repl32) {
2329 if (BG.StartIdx < 32 && BG.EndIdx < 32 && BG.StartIdx <= BG.EndIdx &&
2330 !BG.Repl32Coalesced) {
2331 if (BG.Repl32CR)
2332 EffRLAmt += 32;
2333 } else {
2334 return false;
2335 }
2336 } else if (VRI.Repl32 != BG.Repl32) {
2337 return false;
2338 }
2339
2340 return VRI.RLAmt == EffRLAmt;
2341 };
2342
2343 for (auto &BG : BitGroups) {
2344 if (!MatchingBG(BG))
2345 continue;
2346
2347 if (BG.StartIdx <= BG.EndIdx) {
2348 for (unsigned i = BG.StartIdx; i <= BG.EndIdx; ++i)
2349 Mask |= (UINT64_C(1)1UL << i);
2350 } else {
2351 for (unsigned i = BG.StartIdx; i < Bits.size(); ++i)
2352 Mask |= (UINT64_C(1)1UL << i);
2353 for (unsigned i = 0; i <= BG.EndIdx; ++i)
2354 Mask |= (UINT64_C(1)1UL << i);
2355 }
2356 }
2357
2358 // We can use the 32-bit andi/andis technique if the mask does not
2359 // require any higher-order bits. This can save an instruction compared
2360 // to always using the general 64-bit technique.
2361 bool Use32BitInsts = isUInt<32>(Mask);
2362 // Compute the masks for andi/andis that would be necessary.
2363 unsigned ANDIMask = (Mask & UINT16_MAX(65535)),
2364 ANDISMask = (Mask >> 16) & UINT16_MAX(65535);
2365
2366 bool NeedsRotate = VRI.RLAmt || (VRI.Repl32 && !isUInt<32>(Mask));
2367
2368 unsigned NumAndInsts = (unsigned) NeedsRotate +
2369 (unsigned) (bool) Res;
2370 unsigned NumOfSelectInsts = 0;
2371 selectI64Imm(CurDAG, dl, Mask, &NumOfSelectInsts);
2372 assert(NumOfSelectInsts > 0 && "Failed to select an i64 constant.")(static_cast <bool> (NumOfSelectInsts > 0 &&
"Failed to select an i64 constant.") ? void (0) : __assert_fail
("NumOfSelectInsts > 0 && \"Failed to select an i64 constant.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2372, __extension__ __PRETTY_FUNCTION__))
;
2373 if (Use32BitInsts)
2374 NumAndInsts += (unsigned) (ANDIMask != 0) + (unsigned) (ANDISMask != 0) +
2375 (unsigned) (ANDIMask != 0 && ANDISMask != 0);
2376 else
2377 NumAndInsts += NumOfSelectInsts + /* and */ 1;
2378
2379 unsigned NumRLInsts = 0;
2380 bool FirstBG = true;
2381 bool MoreBG = false;
2382 for (auto &BG : BitGroups) {
2383 if (!MatchingBG(BG)) {
2384 MoreBG = true;
2385 continue;
2386 }
2387 NumRLInsts +=
2388 SelectRotMask64Count(BG.RLAmt, BG.Repl32, BG.StartIdx, BG.EndIdx,
2389 !FirstBG);
2390 FirstBG = false;
2391 }
2392
2393 LLVM_DEBUG(dbgs() << "\t\trotation groups for " << VRI.V.getNode()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\trotation groups for "
<< VRI.V.getNode() << " RL: " << VRI.RLAmt
<< (VRI.Repl32 ? " (32):" : ":") << "\n\t\t\tisel using masking: "
<< NumAndInsts << " using rotates: " << NumRLInsts
<< "\n"; } } while (false)
2394 << " RL: " << VRI.RLAmt << (VRI.Repl32 ? " (32):" : ":")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\trotation groups for "
<< VRI.V.getNode() << " RL: " << VRI.RLAmt
<< (VRI.Repl32 ? " (32):" : ":") << "\n\t\t\tisel using masking: "
<< NumAndInsts << " using rotates: " << NumRLInsts
<< "\n"; } } while (false)
2395 << "\n\t\t\tisel using masking: " << NumAndInstsdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\trotation groups for "
<< VRI.V.getNode() << " RL: " << VRI.RLAmt
<< (VRI.Repl32 ? " (32):" : ":") << "\n\t\t\tisel using masking: "
<< NumAndInsts << " using rotates: " << NumRLInsts
<< "\n"; } } while (false)
2396 << " using rotates: " << NumRLInsts << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\trotation groups for "
<< VRI.V.getNode() << " RL: " << VRI.RLAmt
<< (VRI.Repl32 ? " (32):" : ":") << "\n\t\t\tisel using masking: "
<< NumAndInsts << " using rotates: " << NumRLInsts
<< "\n"; } } while (false)
;
2397
2398 // When we'd use andi/andis, we bias toward using the rotates (andi only
2399 // has a record form, and is cracked on POWER cores). However, when using
2400 // general 64-bit constant formation, bias toward the constant form,
2401 // because that exposes more opportunities for CSE.
2402 if (NumAndInsts > NumRLInsts)
2403 continue;
2404 // When merging multiple bit groups, instruction or is used.
2405 // But when rotate is used, rldimi can inert the rotated value into any
2406 // register, so instruction or can be avoided.
2407 if ((Use32BitInsts || MoreBG) && NumAndInsts == NumRLInsts)
2408 continue;
2409
2410 LLVM_DEBUG(dbgs() << "\t\t\t\tusing masking\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\t\t\tusing masking\n";
} } while (false)
;
2411
2412 if (InstCnt) *InstCnt += NumAndInsts;
2413
2414 SDValue VRot;
2415 // We actually need to generate a rotation if we have a non-zero rotation
2416 // factor or, in the Repl32 case, if we care about any of the
2417 // higher-order replicated bits. In the latter case, we generate a mask
2418 // backward so that it actually includes the entire 64 bits.
2419 if (VRI.RLAmt || (VRI.Repl32 && !isUInt<32>(Mask)))
2420 VRot = SelectRotMask64(VRI.V, dl, VRI.RLAmt, VRI.Repl32,
2421 VRI.Repl32 ? 31 : 0, VRI.Repl32 ? 30 : 63);
2422 else
2423 VRot = VRI.V;
2424
2425 SDValue TotalVal;
2426 if (Use32BitInsts) {
2427 assert((ANDIMask != 0 || ANDISMask != 0) &&(static_cast <bool> ((ANDIMask != 0 || ANDISMask != 0) &&
"No set bits in mask when using 32-bit ands for 64-bit value"
) ? void (0) : __assert_fail ("(ANDIMask != 0 || ANDISMask != 0) && \"No set bits in mask when using 32-bit ands for 64-bit value\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2428, __extension__ __PRETTY_FUNCTION__))
2428 "No set bits in mask when using 32-bit ands for 64-bit value")(static_cast <bool> ((ANDIMask != 0 || ANDISMask != 0) &&
"No set bits in mask when using 32-bit ands for 64-bit value"
) ? void (0) : __assert_fail ("(ANDIMask != 0 || ANDISMask != 0) && \"No set bits in mask when using 32-bit ands for 64-bit value\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2428, __extension__ __PRETTY_FUNCTION__))
;
2429
2430 SDValue ANDIVal, ANDISVal;
2431 if (ANDIMask != 0)
2432 ANDIVal = SDValue(CurDAG->getMachineNode(PPC::ANDI8_rec, dl, MVT::i64,
2433 ExtendToInt64(VRot, dl),
2434 getI32Imm(ANDIMask, dl)),
2435 0);
2436 if (ANDISMask != 0)
2437 ANDISVal =
2438 SDValue(CurDAG->getMachineNode(PPC::ANDIS8_rec, dl, MVT::i64,
2439 ExtendToInt64(VRot, dl),
2440 getI32Imm(ANDISMask, dl)),
2441 0);
2442
2443 if (!ANDIVal)
2444 TotalVal = ANDISVal;
2445 else if (!ANDISVal)
2446 TotalVal = ANDIVal;
2447 else
2448 TotalVal = SDValue(CurDAG->getMachineNode(PPC::OR8, dl, MVT::i64,
2449 ExtendToInt64(ANDIVal, dl), ANDISVal), 0);
2450 } else {
2451 TotalVal = SDValue(selectI64Imm(CurDAG, dl, Mask), 0);
2452 TotalVal =
2453 SDValue(CurDAG->getMachineNode(PPC::AND8, dl, MVT::i64,
2454 ExtendToInt64(VRot, dl), TotalVal),
2455 0);
2456 }
2457
2458 if (!Res)
2459 Res = TotalVal;
2460 else
2461 Res = SDValue(CurDAG->getMachineNode(PPC::OR8, dl, MVT::i64,
2462 ExtendToInt64(Res, dl), TotalVal),
2463 0);
2464
2465 // Now, remove all groups with this underlying value and rotation
2466 // factor.
2467 eraseMatchingBitGroups(MatchingBG);
2468 }
2469 }
2470
2471 // Instruction selection for the 64-bit case.
2472 SDNode *Select64(SDNode *N, bool LateMask, unsigned *InstCnt) {
2473 SDLoc dl(N);
2474 SDValue Res;
2475
2476 if (InstCnt) *InstCnt = 0;
2477
2478 // Take care of cases that should use andi/andis first.
2479 SelectAndParts64(dl, Res, InstCnt);
2480
2481 // If we've not yet selected a 'starting' instruction, and we have no zeros
2482 // to fill in, select the (Value, RLAmt) with the highest priority (largest
2483 // number of groups), and start with this rotated value.
2484 if ((!NeedMask || LateMask) && !Res) {
2485 // If we have both Repl32 groups and non-Repl32 groups, the non-Repl32
2486 // groups will come first, and so the VRI representing the largest number
2487 // of groups might not be first (it might be the first Repl32 groups).
2488 unsigned MaxGroupsIdx = 0;
2489 if (!ValueRotsVec[0].Repl32) {
2490 for (unsigned i = 0, ie = ValueRotsVec.size(); i < ie; ++i)
2491 if (ValueRotsVec[i].Repl32) {
2492 if (ValueRotsVec[i].NumGroups > ValueRotsVec[0].NumGroups)
2493 MaxGroupsIdx = i;
2494 break;
2495 }
2496 }
2497
2498 ValueRotInfo &VRI = ValueRotsVec[MaxGroupsIdx];
2499 bool NeedsRotate = false;
2500 if (VRI.RLAmt) {
2501 NeedsRotate = true;
2502 } else if (VRI.Repl32) {
2503 for (auto &BG : BitGroups) {
2504 if (BG.V != VRI.V || BG.RLAmt != VRI.RLAmt ||
2505 BG.Repl32 != VRI.Repl32)
2506 continue;
2507
2508 // We don't need a rotate if the bit group is confined to the lower
2509 // 32 bits.
2510 if (BG.StartIdx < 32 && BG.EndIdx < 32 && BG.StartIdx < BG.EndIdx)
2511 continue;
2512
2513 NeedsRotate = true;
2514 break;
2515 }
2516 }
2517
2518 if (NeedsRotate)
2519 Res = SelectRotMask64(VRI.V, dl, VRI.RLAmt, VRI.Repl32,
2520 VRI.Repl32 ? 31 : 0, VRI.Repl32 ? 30 : 63,
2521 InstCnt);
2522 else
2523 Res = VRI.V;
2524
2525 // Now, remove all groups with this underlying value and rotation factor.
2526 if (Res)
2527 eraseMatchingBitGroups([VRI](const BitGroup &BG) {
2528 return BG.V == VRI.V && BG.RLAmt == VRI.RLAmt &&
2529 BG.Repl32 == VRI.Repl32;
2530 });
2531 }
2532
2533 // Because 64-bit rotates are more flexible than inserts, we might have a
2534 // preference regarding which one we do first (to save one instruction).
2535 if (!Res)
2536 for (auto I = BitGroups.begin(), IE = BitGroups.end(); I != IE; ++I) {
2537 if (SelectRotMask64Count(I->RLAmt, I->Repl32, I->StartIdx, I->EndIdx,
2538 false) <
2539 SelectRotMask64Count(I->RLAmt, I->Repl32, I->StartIdx, I->EndIdx,
2540 true)) {
2541 if (I != BitGroups.begin()) {
2542 BitGroup BG = *I;
2543 BitGroups.erase(I);
2544 BitGroups.insert(BitGroups.begin(), BG);
2545 }
2546
2547 break;
2548 }
2549 }
2550
2551 // Insert the other groups (one at a time).
2552 for (auto &BG : BitGroups) {
2553 if (!Res)
2554 Res = SelectRotMask64(BG.V, dl, BG.RLAmt, BG.Repl32, BG.StartIdx,
2555 BG.EndIdx, InstCnt);
2556 else
2557 Res = SelectRotMaskIns64(Res, BG.V, dl, BG.RLAmt, BG.Repl32,
2558 BG.StartIdx, BG.EndIdx, InstCnt);
2559 }
2560
2561 if (LateMask) {
2562 uint64_t Mask = getZerosMask();
2563
2564 // We can use the 32-bit andi/andis technique if the mask does not
2565 // require any higher-order bits. This can save an instruction compared
2566 // to always using the general 64-bit technique.
2567 bool Use32BitInsts = isUInt<32>(Mask);
2568 // Compute the masks for andi/andis that would be necessary.
2569 unsigned ANDIMask = (Mask & UINT16_MAX(65535)),
2570 ANDISMask = (Mask >> 16) & UINT16_MAX(65535);
2571
2572 if (Use32BitInsts) {
2573 assert((ANDIMask != 0 || ANDISMask != 0) &&(static_cast <bool> ((ANDIMask != 0 || ANDISMask != 0) &&
"No set bits in mask when using 32-bit ands for 64-bit value"
) ? void (0) : __assert_fail ("(ANDIMask != 0 || ANDISMask != 0) && \"No set bits in mask when using 32-bit ands for 64-bit value\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2574, __extension__ __PRETTY_FUNCTION__))
2574 "No set bits in mask when using 32-bit ands for 64-bit value")(static_cast <bool> ((ANDIMask != 0 || ANDISMask != 0) &&
"No set bits in mask when using 32-bit ands for 64-bit value"
) ? void (0) : __assert_fail ("(ANDIMask != 0 || ANDISMask != 0) && \"No set bits in mask when using 32-bit ands for 64-bit value\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2574, __extension__ __PRETTY_FUNCTION__))
;
2575
2576 if (InstCnt) *InstCnt += (unsigned) (ANDIMask != 0) +
2577 (unsigned) (ANDISMask != 0) +
2578 (unsigned) (ANDIMask != 0 && ANDISMask != 0);
2579
2580 SDValue ANDIVal, ANDISVal;
2581 if (ANDIMask != 0)
2582 ANDIVal = SDValue(CurDAG->getMachineNode(PPC::ANDI8_rec, dl, MVT::i64,
2583 ExtendToInt64(Res, dl),
2584 getI32Imm(ANDIMask, dl)),
2585 0);
2586 if (ANDISMask != 0)
2587 ANDISVal =
2588 SDValue(CurDAG->getMachineNode(PPC::ANDIS8_rec, dl, MVT::i64,
2589 ExtendToInt64(Res, dl),
2590 getI32Imm(ANDISMask, dl)),
2591 0);
2592
2593 if (!ANDIVal)
2594 Res = ANDISVal;
2595 else if (!ANDISVal)
2596 Res = ANDIVal;
2597 else
2598 Res = SDValue(CurDAG->getMachineNode(PPC::OR8, dl, MVT::i64,
2599 ExtendToInt64(ANDIVal, dl), ANDISVal), 0);
2600 } else {
2601 unsigned NumOfSelectInsts = 0;
2602 SDValue MaskVal =
2603 SDValue(selectI64Imm(CurDAG, dl, Mask, &NumOfSelectInsts), 0);
2604 Res = SDValue(CurDAG->getMachineNode(PPC::AND8, dl, MVT::i64,
2605 ExtendToInt64(Res, dl), MaskVal),
2606 0);
2607 if (InstCnt)
2608 *InstCnt += NumOfSelectInsts + /* and */ 1;
2609 }
2610 }
2611
2612 return Res.getNode();
2613 }
2614
2615 SDNode *Select(SDNode *N, bool LateMask, unsigned *InstCnt = nullptr) {
2616 // Fill in BitGroups.
2617 collectBitGroups(LateMask);
2618 if (BitGroups.empty())
2619 return nullptr;
2620
2621 // For 64-bit values, figure out when we can use 32-bit instructions.
2622 if (Bits.size() == 64)
2623 assignRepl32BitGroups();
2624
2625 // Fill in ValueRotsVec.
2626 collectValueRotInfo();
2627
2628 if (Bits.size() == 32) {
2629 return Select32(N, LateMask, InstCnt);
2630 } else {
2631 assert(Bits.size() == 64 && "Not 64 bits here?")(static_cast <bool> (Bits.size() == 64 && "Not 64 bits here?"
) ? void (0) : __assert_fail ("Bits.size() == 64 && \"Not 64 bits here?\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2631, __extension__ __PRETTY_FUNCTION__))
;
2632 return Select64(N, LateMask, InstCnt);
2633 }
2634
2635 return nullptr;
2636 }
2637
2638 void eraseMatchingBitGroups(function_ref<bool(const BitGroup &)> F) {
2639 erase_if(BitGroups, F);
2640 }
2641
2642 SmallVector<ValueBit, 64> Bits;
2643
2644 bool NeedMask = false;
2645 SmallVector<unsigned, 64> RLAmt;
2646
2647 SmallVector<BitGroup, 16> BitGroups;
2648
2649 DenseMap<std::pair<SDValue, unsigned>, ValueRotInfo> ValueRots;
2650 SmallVector<ValueRotInfo, 16> ValueRotsVec;
2651
2652 SelectionDAG *CurDAG = nullptr;
2653
2654public:
2655 BitPermutationSelector(SelectionDAG *DAG)
2656 : CurDAG(DAG) {}
2657
2658 // Here we try to match complex bit permutations into a set of
2659 // rotate-and-shift/shift/and/or instructions, using a set of heuristics
2660 // known to produce optimal code for common cases (like i32 byte swapping).
2661 SDNode *Select(SDNode *N) {
2662 Memoizer.clear();
2663 auto Result =
2664 getValueBits(SDValue(N, 0), N->getValueType(0).getSizeInBits());
2665 if (!Result.first)
2666 return nullptr;
2667 Bits = std::move(*Result.second);
2668
2669 LLVM_DEBUG(dbgs() << "Considering bit-permutation-based instruction"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Considering bit-permutation-based instruction"
" selection for: "; } } while (false)
2670 " selection for: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Considering bit-permutation-based instruction"
" selection for: "; } } while (false)
;
2671 LLVM_DEBUG(N->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { N->dump(CurDAG); } } while (false)
;
2672
2673 // Fill it RLAmt and set NeedMask.
2674 computeRotationAmounts();
2675
2676 if (!NeedMask)
2677 return Select(N, false);
2678
2679 // We currently have two techniques for handling results with zeros: early
2680 // masking (the default) and late masking. Late masking is sometimes more
2681 // efficient, but because the structure of the bit groups is different, it
2682 // is hard to tell without generating both and comparing the results. With
2683 // late masking, we ignore zeros in the resulting value when inserting each
2684 // set of bit groups, and then mask in the zeros at the end. With early
2685 // masking, we only insert the non-zero parts of the result at every step.
2686
2687 unsigned InstCnt = 0, InstCntLateMask = 0;
2688 LLVM_DEBUG(dbgs() << "\tEarly masking:\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tEarly masking:\n"; } } while
(false)
;
2689 SDNode *RN = Select(N, false, &InstCnt);
2690 LLVM_DEBUG(dbgs() << "\t\tisel would use " << InstCnt << " instructions\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\tisel would use " <<
InstCnt << " instructions\n"; } } while (false)
;
2691
2692 LLVM_DEBUG(dbgs() << "\tLate masking:\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tLate masking:\n"; } } while
(false)
;
2693 SDNode *RNLM = Select(N, true, &InstCntLateMask);
2694 LLVM_DEBUG(dbgs() << "\t\tisel would use " << InstCntLateMaskdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\tisel would use " <<
InstCntLateMask << " instructions\n"; } } while (false
)
2695 << " instructions\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\t\tisel would use " <<
InstCntLateMask << " instructions\n"; } } while (false
)
;
2696
2697 if (InstCnt <= InstCntLateMask) {
2698 LLVM_DEBUG(dbgs() << "\tUsing early-masking for isel\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tUsing early-masking for isel\n"
; } } while (false)
;
2699 return RN;
2700 }
2701
2702 LLVM_DEBUG(dbgs() << "\tUsing late-masking for isel\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\tUsing late-masking for isel\n"
; } } while (false)
;
2703 return RNLM;
2704 }
2705};
2706
2707class IntegerCompareEliminator {
2708 SelectionDAG *CurDAG;
2709 PPCDAGToDAGISel *S;
2710 // Conversion type for interpreting results of a 32-bit instruction as
2711 // a 64-bit value or vice versa.
2712 enum ExtOrTruncConversion { Ext, Trunc };
2713
2714 // Modifiers to guide how an ISD::SETCC node's result is to be computed
2715 // in a GPR.
2716 // ZExtOrig - use the original condition code, zero-extend value
2717 // ZExtInvert - invert the condition code, zero-extend value
2718 // SExtOrig - use the original condition code, sign-extend value
2719 // SExtInvert - invert the condition code, sign-extend value
2720 enum SetccInGPROpts { ZExtOrig, ZExtInvert, SExtOrig, SExtInvert };
2721
2722 // Comparisons against zero to emit GPR code sequences for. Each of these
2723 // sequences may need to be emitted for two or more equivalent patterns.
2724 // For example (a >= 0) == (a > -1). The direction of the comparison (</>)
2725 // matters as well as the extension type: sext (-1/0), zext (1/0).
2726 // GEZExt - (zext (LHS >= 0))
2727 // GESExt - (sext (LHS >= 0))
2728 // LEZExt - (zext (LHS <= 0))
2729 // LESExt - (sext (LHS <= 0))
2730 enum ZeroCompare { GEZExt, GESExt, LEZExt, LESExt };
2731
2732 SDNode *tryEXTEND(SDNode *N);
2733 SDNode *tryLogicOpOfCompares(SDNode *N);
2734 SDValue computeLogicOpInGPR(SDValue LogicOp);
2735 SDValue signExtendInputIfNeeded(SDValue Input);
2736 SDValue zeroExtendInputIfNeeded(SDValue Input);
2737 SDValue addExtOrTrunc(SDValue NatWidthRes, ExtOrTruncConversion Conv);
2738 SDValue getCompoundZeroComparisonInGPR(SDValue LHS, SDLoc dl,
2739 ZeroCompare CmpTy);
2740 SDValue get32BitZExtCompare(SDValue LHS, SDValue RHS, ISD::CondCode CC,
2741 int64_t RHSValue, SDLoc dl);
2742 SDValue get32BitSExtCompare(SDValue LHS, SDValue RHS, ISD::CondCode CC,
2743 int64_t RHSValue, SDLoc dl);
2744 SDValue get64BitZExtCompare(SDValue LHS, SDValue RHS, ISD::CondCode CC,
2745 int64_t RHSValue, SDLoc dl);
2746 SDValue get64BitSExtCompare(SDValue LHS, SDValue RHS, ISD::CondCode CC,
2747 int64_t RHSValue, SDLoc dl);
2748 SDValue getSETCCInGPR(SDValue Compare, SetccInGPROpts ConvOpts);
2749
2750public:
2751 IntegerCompareEliminator(SelectionDAG *DAG,
2752 PPCDAGToDAGISel *Sel) : CurDAG(DAG), S(Sel) {
2753 assert(CurDAG->getTargetLoweringInfo()(static_cast <bool> (CurDAG->getTargetLoweringInfo()
.getPointerTy(CurDAG->getDataLayout()).getSizeInBits() ==
64 && "Only expecting to use this on 64 bit targets."
) ? void (0) : __assert_fail ("CurDAG->getTargetLoweringInfo() .getPointerTy(CurDAG->getDataLayout()).getSizeInBits() == 64 && \"Only expecting to use this on 64 bit targets.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2755, __extension__ __PRETTY_FUNCTION__))
2754 .getPointerTy(CurDAG->getDataLayout()).getSizeInBits() == 64 &&(static_cast <bool> (CurDAG->getTargetLoweringInfo()
.getPointerTy(CurDAG->getDataLayout()).getSizeInBits() ==
64 && "Only expecting to use this on 64 bit targets."
) ? void (0) : __assert_fail ("CurDAG->getTargetLoweringInfo() .getPointerTy(CurDAG->getDataLayout()).getSizeInBits() == 64 && \"Only expecting to use this on 64 bit targets.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2755, __extension__ __PRETTY_FUNCTION__))
2755 "Only expecting to use this on 64 bit targets.")(static_cast <bool> (CurDAG->getTargetLoweringInfo()
.getPointerTy(CurDAG->getDataLayout()).getSizeInBits() ==
64 && "Only expecting to use this on 64 bit targets."
) ? void (0) : __assert_fail ("CurDAG->getTargetLoweringInfo() .getPointerTy(CurDAG->getDataLayout()).getSizeInBits() == 64 && \"Only expecting to use this on 64 bit targets.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2755, __extension__ __PRETTY_FUNCTION__))
;
2756 }
2757 SDNode *Select(SDNode *N) {
2758 if (CmpInGPR == ICGPR_None)
2759 return nullptr;
2760 switch (N->getOpcode()) {
2761 default: break;
2762 case ISD::ZERO_EXTEND:
2763 if (CmpInGPR == ICGPR_Sext || CmpInGPR == ICGPR_SextI32 ||
2764 CmpInGPR == ICGPR_SextI64)
2765 return nullptr;
2766 LLVM_FALLTHROUGH[[gnu::fallthrough]];
2767 case ISD::SIGN_EXTEND:
2768 if (CmpInGPR == ICGPR_Zext || CmpInGPR == ICGPR_ZextI32 ||
2769 CmpInGPR == ICGPR_ZextI64)
2770 return nullptr;
2771 return tryEXTEND(N);
2772 case ISD::AND:
2773 case ISD::OR:
2774 case ISD::XOR:
2775 return tryLogicOpOfCompares(N);
2776 }
2777 return nullptr;
2778 }
2779};
2780
2781static bool isLogicOp(unsigned Opc) {
2782 return Opc == ISD::AND || Opc == ISD::OR || Opc == ISD::XOR;
2783}
2784// The obvious case for wanting to keep the value in a GPR. Namely, the
2785// result of the comparison is actually needed in a GPR.
2786SDNode *IntegerCompareEliminator::tryEXTEND(SDNode *N) {
2787 assert((N->getOpcode() == ISD::ZERO_EXTEND ||(static_cast <bool> ((N->getOpcode() == ISD::ZERO_EXTEND
|| N->getOpcode() == ISD::SIGN_EXTEND) && "Expecting a zero/sign extend node!"
) ? void (0) : __assert_fail ("(N->getOpcode() == ISD::ZERO_EXTEND || N->getOpcode() == ISD::SIGN_EXTEND) && \"Expecting a zero/sign extend node!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2789, __extension__ __PRETTY_FUNCTION__))
2788 N->getOpcode() == ISD::SIGN_EXTEND) &&(static_cast <bool> ((N->getOpcode() == ISD::ZERO_EXTEND
|| N->getOpcode() == ISD::SIGN_EXTEND) && "Expecting a zero/sign extend node!"
) ? void (0) : __assert_fail ("(N->getOpcode() == ISD::ZERO_EXTEND || N->getOpcode() == ISD::SIGN_EXTEND) && \"Expecting a zero/sign extend node!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2789, __extension__ __PRETTY_FUNCTION__))
2789 "Expecting a zero/sign extend node!")(static_cast <bool> ((N->getOpcode() == ISD::ZERO_EXTEND
|| N->getOpcode() == ISD::SIGN_EXTEND) && "Expecting a zero/sign extend node!"
) ? void (0) : __assert_fail ("(N->getOpcode() == ISD::ZERO_EXTEND || N->getOpcode() == ISD::SIGN_EXTEND) && \"Expecting a zero/sign extend node!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2789, __extension__ __PRETTY_FUNCTION__))
;
2790 SDValue WideRes;
2791 // If we are zero-extending the result of a logical operation on i1
2792 // values, we can keep the values in GPRs.
2793 if (isLogicOp(N->getOperand(0).getOpcode()) &&
2794 N->getOperand(0).getValueType() == MVT::i1 &&
2795 N->getOpcode() == ISD::ZERO_EXTEND)
2796 WideRes = computeLogicOpInGPR(N->getOperand(0));
2797 else if (N->getOperand(0).getOpcode() != ISD::SETCC)
2798 return nullptr;
2799 else
2800 WideRes =
2801 getSETCCInGPR(N->getOperand(0),
2802 N->getOpcode() == ISD::SIGN_EXTEND ?
2803 SetccInGPROpts::SExtOrig : SetccInGPROpts::ZExtOrig);
2804
2805 if (!WideRes)
2806 return nullptr;
2807
2808 SDLoc dl(N);
2809 bool Input32Bit = WideRes.getValueType() == MVT::i32;
2810 bool Output32Bit = N->getValueType(0) == MVT::i32;
2811
2812 NumSextSetcc += N->getOpcode() == ISD::SIGN_EXTEND ? 1 : 0;
2813 NumZextSetcc += N->getOpcode() == ISD::SIGN_EXTEND ? 0 : 1;
2814
2815 SDValue ConvOp = WideRes;
2816 if (Input32Bit != Output32Bit)
2817 ConvOp = addExtOrTrunc(WideRes, Input32Bit ? ExtOrTruncConversion::Ext :
2818 ExtOrTruncConversion::Trunc);
2819 return ConvOp.getNode();
2820}
2821
2822// Attempt to perform logical operations on the results of comparisons while
2823// keeping the values in GPRs. Without doing so, these would end up being
2824// lowered to CR-logical operations which suffer from significant latency and
2825// low ILP.
2826SDNode *IntegerCompareEliminator::tryLogicOpOfCompares(SDNode *N) {
2827 if (N->getValueType(0) != MVT::i1)
2828 return nullptr;
2829 assert(isLogicOp(N->getOpcode()) &&(static_cast <bool> (isLogicOp(N->getOpcode()) &&
"Expected a logic operation on setcc results.") ? void (0) :
__assert_fail ("isLogicOp(N->getOpcode()) && \"Expected a logic operation on setcc results.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2830, __extension__ __PRETTY_FUNCTION__))
2830 "Expected a logic operation on setcc results.")(static_cast <bool> (isLogicOp(N->getOpcode()) &&
"Expected a logic operation on setcc results.") ? void (0) :
__assert_fail ("isLogicOp(N->getOpcode()) && \"Expected a logic operation on setcc results.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2830, __extension__ __PRETTY_FUNCTION__))
;
2831 SDValue LoweredLogical = computeLogicOpInGPR(SDValue(N, 0));
2832 if (!LoweredLogical)
2833 return nullptr;
2834
2835 SDLoc dl(N);
2836 bool IsBitwiseNegate = LoweredLogical.getMachineOpcode() == PPC::XORI8;
2837 unsigned SubRegToExtract = IsBitwiseNegate ? PPC::sub_eq : PPC::sub_gt;
2838 SDValue CR0Reg = CurDAG->getRegister(PPC::CR0, MVT::i32);
2839 SDValue LHS = LoweredLogical.getOperand(0);
2840 SDValue RHS = LoweredLogical.getOperand(1);
2841 SDValue WideOp;
2842 SDValue OpToConvToRecForm;
2843
2844 // Look through any 32-bit to 64-bit implicit extend nodes to find the
2845 // opcode that is input to the XORI.
2846 if (IsBitwiseNegate &&
2847 LoweredLogical.getOperand(0).getMachineOpcode() == PPC::INSERT_SUBREG)
2848 OpToConvToRecForm = LoweredLogical.getOperand(0).getOperand(1);
2849 else if (IsBitwiseNegate)
2850 // If the input to the XORI isn't an extension, that's what we're after.
2851 OpToConvToRecForm = LoweredLogical.getOperand(0);
2852 else
2853 // If this is not an XORI, it is a reg-reg logical op and we can convert
2854 // it to record-form.
2855 OpToConvToRecForm = LoweredLogical;
2856
2857 // Get the record-form version of the node we're looking to use to get the
2858 // CR result from.
2859 uint16_t NonRecOpc = OpToConvToRecForm.getMachineOpcode();
2860 int NewOpc = PPCInstrInfo::getRecordFormOpcode(NonRecOpc);
2861
2862 // Convert the right node to record-form. This is either the logical we're
2863 // looking at or it is the input node to the negation (if we're looking at
2864 // a bitwise negation).
2865 if (NewOpc != -1 && IsBitwiseNegate) {
2866 // The input to the XORI has a record-form. Use it.
2867 assert(LoweredLogical.getConstantOperandVal(1) == 1 &&(static_cast <bool> (LoweredLogical.getConstantOperandVal
(1) == 1 && "Expected a PPC::XORI8 only for bitwise negation."
) ? void (0) : __assert_fail ("LoweredLogical.getConstantOperandVal(1) == 1 && \"Expected a PPC::XORI8 only for bitwise negation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2868, __extension__ __PRETTY_FUNCTION__))
2868 "Expected a PPC::XORI8 only for bitwise negation.")(static_cast <bool> (LoweredLogical.getConstantOperandVal
(1) == 1 && "Expected a PPC::XORI8 only for bitwise negation."
) ? void (0) : __assert_fail ("LoweredLogical.getConstantOperandVal(1) == 1 && \"Expected a PPC::XORI8 only for bitwise negation.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2868, __extension__ __PRETTY_FUNCTION__))
;
2869 // Emit the record-form instruction.
2870 std::vector<SDValue> Ops;
2871 for (int i = 0, e = OpToConvToRecForm.getNumOperands(); i < e; i++)
2872 Ops.push_back(OpToConvToRecForm.getOperand(i));
2873
2874 WideOp =
2875 SDValue(CurDAG->getMachineNode(NewOpc, dl,
2876 OpToConvToRecForm.getValueType(),
2877 MVT::Glue, Ops), 0);
2878 } else {
2879 assert((NewOpc != -1 || !IsBitwiseNegate) &&(static_cast <bool> ((NewOpc != -1 || !IsBitwiseNegate)
&& "No record form available for AND8/OR8/XOR8?") ? void
(0) : __assert_fail ("(NewOpc != -1 || !IsBitwiseNegate) && \"No record form available for AND8/OR8/XOR8?\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2880, __extension__ __PRETTY_FUNCTION__))
2880 "No record form available for AND8/OR8/XOR8?")(static_cast <bool> ((NewOpc != -1 || !IsBitwiseNegate)
&& "No record form available for AND8/OR8/XOR8?") ? void
(0) : __assert_fail ("(NewOpc != -1 || !IsBitwiseNegate) && \"No record form available for AND8/OR8/XOR8?\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2880, __extension__ __PRETTY_FUNCTION__))
;
2881 WideOp =
2882 SDValue(CurDAG->getMachineNode(NewOpc == -1 ? PPC::ANDI8_rec : NewOpc,
2883 dl, MVT::i64, MVT::Glue, LHS, RHS),
2884 0);
2885 }
2886
2887 // Select this node to a single bit from CR0 set by the record-form node
2888 // just created. For bitwise negation, use the EQ bit which is the equivalent
2889 // of negating the result (i.e. it is a bit set when the result of the
2890 // operation is zero).
2891 SDValue SRIdxVal =
2892 CurDAG->getTargetConstant(SubRegToExtract, dl, MVT::i32);
2893 SDValue CRBit =
2894 SDValue(CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG, dl,
2895 MVT::i1, CR0Reg, SRIdxVal,
2896 WideOp.getValue(1)), 0);
2897 return CRBit.getNode();
2898}
2899
2900// Lower a logical operation on i1 values into a GPR sequence if possible.
2901// The result can be kept in a GPR if requested.
2902// Three types of inputs can be handled:
2903// - SETCC
2904// - TRUNCATE
2905// - Logical operation (AND/OR/XOR)
2906// There is also a special case that is handled (namely a complement operation
2907// achieved with xor %a, -1).
2908SDValue IntegerCompareEliminator::computeLogicOpInGPR(SDValue LogicOp) {
2909 assert(isLogicOp(LogicOp.getOpcode()) &&(static_cast <bool> (isLogicOp(LogicOp.getOpcode()) &&
"Can only handle logic operations here.") ? void (0) : __assert_fail
("isLogicOp(LogicOp.getOpcode()) && \"Can only handle logic operations here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2910, __extension__ __PRETTY_FUNCTION__))
2910 "Can only handle logic operations here.")(static_cast <bool> (isLogicOp(LogicOp.getOpcode()) &&
"Can only handle logic operations here.") ? void (0) : __assert_fail
("isLogicOp(LogicOp.getOpcode()) && \"Can only handle logic operations here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2910, __extension__ __PRETTY_FUNCTION__))
;
2911 assert(LogicOp.getValueType() == MVT::i1 &&(static_cast <bool> (LogicOp.getValueType() == MVT::i1 &&
"Can only handle logic operations on i1 values here.") ? void
(0) : __assert_fail ("LogicOp.getValueType() == MVT::i1 && \"Can only handle logic operations on i1 values here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2912, __extension__ __PRETTY_FUNCTION__))
2912 "Can only handle logic operations on i1 values here.")(static_cast <bool> (LogicOp.getValueType() == MVT::i1 &&
"Can only handle logic operations on i1 values here.") ? void
(0) : __assert_fail ("LogicOp.getValueType() == MVT::i1 && \"Can only handle logic operations on i1 values here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2912, __extension__ __PRETTY_FUNCTION__))
;
2913 SDLoc dl(LogicOp);
2914 SDValue LHS, RHS;
2915
2916 // Special case: xor %a, -1
2917 bool IsBitwiseNegation = isBitwiseNot(LogicOp);
2918
2919 // Produces a GPR sequence for each operand of the binary logic operation.
2920 // For SETCC, it produces the respective comparison, for TRUNCATE it truncates
2921 // the value in a GPR and for logic operations, it will recursively produce
2922 // a GPR sequence for the operation.
2923 auto getLogicOperand = [&] (SDValue Operand) -> SDValue {
2924 unsigned OperandOpcode = Operand.getOpcode();
2925 if (OperandOpcode == ISD::SETCC)
2926 return getSETCCInGPR(Operand, SetccInGPROpts::ZExtOrig);
2927 else if (OperandOpcode == ISD::TRUNCATE) {
2928 SDValue InputOp = Operand.getOperand(0);
2929 EVT InVT = InputOp.getValueType();
2930 return SDValue(CurDAG->getMachineNode(InVT == MVT::i32 ? PPC::RLDICL_32 :
2931 PPC::RLDICL, dl, InVT, InputOp,
2932 S->getI64Imm(0, dl),
2933 S->getI64Imm(63, dl)), 0);
2934 } else if (isLogicOp(OperandOpcode))
2935 return computeLogicOpInGPR(Operand);
2936 return SDValue();
2937 };
2938 LHS = getLogicOperand(LogicOp.getOperand(0));
2939 RHS = getLogicOperand(LogicOp.getOperand(1));
2940
2941 // If a GPR sequence can't be produced for the LHS we can't proceed.
2942 // Not producing a GPR sequence for the RHS is only a problem if this isn't
2943 // a bitwise negation operation.
2944 if (!LHS || (!RHS && !IsBitwiseNegation))
2945 return SDValue();
2946
2947 NumLogicOpsOnComparison++;
2948
2949 // We will use the inputs as 64-bit values.
2950 if (LHS.getValueType() == MVT::i32)
2951 LHS = addExtOrTrunc(LHS, ExtOrTruncConversion::Ext);
2952 if (!IsBitwiseNegation && RHS.getValueType() == MVT::i32)
2953 RHS = addExtOrTrunc(RHS, ExtOrTruncConversion::Ext);
2954
2955 unsigned NewOpc;
2956 switch (LogicOp.getOpcode()) {
2957 default: llvm_unreachable("Unknown logic operation.")::llvm::llvm_unreachable_internal("Unknown logic operation.",
"/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2957)
;
2958 case ISD::AND: NewOpc = PPC::AND8; break;
2959 case ISD::OR: NewOpc = PPC::OR8; break;
2960 case ISD::XOR: NewOpc = PPC::XOR8; break;
2961 }
2962
2963 if (IsBitwiseNegation) {
2964 RHS = S->getI64Imm(1, dl);
2965 NewOpc = PPC::XORI8;
2966 }
2967
2968 return SDValue(CurDAG->getMachineNode(NewOpc, dl, MVT::i64, LHS, RHS), 0);
2969
2970}
2971
2972/// If the value isn't guaranteed to be sign-extended to 64-bits, extend it.
2973/// Otherwise just reinterpret it as a 64-bit value.
2974/// Useful when emitting comparison code for 32-bit values without using
2975/// the compare instruction (which only considers the lower 32-bits).
2976SDValue IntegerCompareEliminator::signExtendInputIfNeeded(SDValue Input) {
2977 assert(Input.getValueType() == MVT::i32 &&(static_cast <bool> (Input.getValueType() == MVT::i32 &&
"Can only sign-extend 32-bit values here.") ? void (0) : __assert_fail
("Input.getValueType() == MVT::i32 && \"Can only sign-extend 32-bit values here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2978, __extension__ __PRETTY_FUNCTION__))
2978 "Can only sign-extend 32-bit values here.")(static_cast <bool> (Input.getValueType() == MVT::i32 &&
"Can only sign-extend 32-bit values here.") ? void (0) : __assert_fail
("Input.getValueType() == MVT::i32 && \"Can only sign-extend 32-bit values here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 2978, __extension__ __PRETTY_FUNCTION__))
;
2979 unsigned Opc = Input.getOpcode();
2980
2981 // The value was sign extended and then truncated to 32-bits. No need to
2982 // sign extend it again.
2983 if (Opc == ISD::TRUNCATE &&
2984 (Input.getOperand(0).getOpcode() == ISD::AssertSext ||
2985 Input.getOperand(0).getOpcode() == ISD::SIGN_EXTEND))
2986 return addExtOrTrunc(Input, ExtOrTruncConversion::Ext);
2987
2988 LoadSDNode *InputLoad = dyn_cast<LoadSDNode>(Input);
2989 // The input is a sign-extending load. All ppc sign-extending loads
2990 // sign-extend to the full 64-bits.
2991 if (InputLoad && InputLoad->getExtensionType() == ISD::SEXTLOAD)
2992 return addExtOrTrunc(Input, ExtOrTruncConversion::Ext);
2993
2994 ConstantSDNode *InputConst = dyn_cast<ConstantSDNode>(Input);
2995 // We don't sign-extend constants.
2996 if (InputConst)
2997 return addExtOrTrunc(Input, ExtOrTruncConversion::Ext);
2998
2999 SDLoc dl(Input);
3000 SignExtensionsAdded++;
3001 return SDValue(CurDAG->getMachineNode(PPC::EXTSW_32_64, dl,
3002 MVT::i64, Input), 0);
3003}
3004
3005/// If the value isn't guaranteed to be zero-extended to 64-bits, extend it.
3006/// Otherwise just reinterpret it as a 64-bit value.
3007/// Useful when emitting comparison code for 32-bit values without using
3008/// the compare instruction (which only considers the lower 32-bits).
3009SDValue IntegerCompareEliminator::zeroExtendInputIfNeeded(SDValue Input) {
3010 assert(Input.getValueType() == MVT::i32 &&(static_cast <bool> (Input.getValueType() == MVT::i32 &&
"Can only zero-extend 32-bit values here.") ? void (0) : __assert_fail
("Input.getValueType() == MVT::i32 && \"Can only zero-extend 32-bit values here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3011, __extension__ __PRETTY_FUNCTION__))
3011 "Can only zero-extend 32-bit values here.")(static_cast <bool> (Input.getValueType() == MVT::i32 &&
"Can only zero-extend 32-bit values here.") ? void (0) : __assert_fail
("Input.getValueType() == MVT::i32 && \"Can only zero-extend 32-bit values here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3011, __extension__ __PRETTY_FUNCTION__))
;
3012 unsigned Opc = Input.getOpcode();
3013
3014 // The only condition under which we can omit the actual extend instruction:
3015 // - The value is a positive constant
3016 // - The value comes from a load that isn't a sign-extending load
3017 // An ISD::TRUNCATE needs to be zero-extended unless it is fed by a zext.
3018 bool IsTruncateOfZExt = Opc == ISD::TRUNCATE &&
3019 (Input.getOperand(0).getOpcode() == ISD::AssertZext ||
3020 Input.getOperand(0).getOpcode() == ISD::ZERO_EXTEND);
3021 if (IsTruncateOfZExt)
3022 return addExtOrTrunc(Input, ExtOrTruncConversion::Ext);
3023
3024 ConstantSDNode *InputConst = dyn_cast<ConstantSDNode>(Input);
3025 if (InputConst && InputConst->getSExtValue() >= 0)
3026 return addExtOrTrunc(Input, ExtOrTruncConversion::Ext);
3027
3028 LoadSDNode *InputLoad = dyn_cast<LoadSDNode>(Input);
3029 // The input is a load that doesn't sign-extend (it will be zero-extended).
3030 if (InputLoad && InputLoad->getExtensionType() != ISD::SEXTLOAD)
3031 return addExtOrTrunc(Input, ExtOrTruncConversion::Ext);
3032
3033 // None of the above, need to zero-extend.
3034 SDLoc dl(Input);
3035 ZeroExtensionsAdded++;
3036 return SDValue(CurDAG->getMachineNode(PPC::RLDICL_32_64, dl, MVT::i64, Input,
3037 S->getI64Imm(0, dl),
3038 S->getI64Imm(32, dl)), 0);
3039}
3040
3041// Handle a 32-bit value in a 64-bit register and vice-versa. These are of
3042// course not actual zero/sign extensions that will generate machine code,
3043// they're just a way to reinterpret a 32 bit value in a register as a
3044// 64 bit value and vice-versa.
3045SDValue IntegerCompareEliminator::addExtOrTrunc(SDValue NatWidthRes,
3046 ExtOrTruncConversion Conv) {
3047 SDLoc dl(NatWidthRes);
3048
3049 // For reinterpreting 32-bit values as 64 bit values, we generate
3050 // INSERT_SUBREG IMPLICIT_DEF:i64, <input>, TargetConstant:i32<1>
3051 if (Conv == ExtOrTruncConversion::Ext) {
3052 SDValue ImDef(CurDAG->getMachineNode(PPC::IMPLICIT_DEF, dl, MVT::i64), 0);
3053 SDValue SubRegIdx =
3054 CurDAG->getTargetConstant(PPC::sub_32, dl, MVT::i32);
3055 return SDValue(CurDAG->getMachineNode(PPC::INSERT_SUBREG, dl, MVT::i64,
3056 ImDef, NatWidthRes, SubRegIdx), 0);
3057 }
3058
3059 assert(Conv == ExtOrTruncConversion::Trunc &&(static_cast <bool> (Conv == ExtOrTruncConversion::Trunc
&& "Unknown convertion between 32 and 64 bit values."
) ? void (0) : __assert_fail ("Conv == ExtOrTruncConversion::Trunc && \"Unknown convertion between 32 and 64 bit values.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3060, __extension__ __PRETTY_FUNCTION__))
3060 "Unknown convertion between 32 and 64 bit values.")(static_cast <bool> (Conv == ExtOrTruncConversion::Trunc
&& "Unknown convertion between 32 and 64 bit values."
) ? void (0) : __assert_fail ("Conv == ExtOrTruncConversion::Trunc && \"Unknown convertion between 32 and 64 bit values.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3060, __extension__ __PRETTY_FUNCTION__))
;
3061 // For reinterpreting 64-bit values as 32-bit values, we just need to
3062 // EXTRACT_SUBREG (i.e. extract the low word).
3063 SDValue SubRegIdx =
3064 CurDAG->getTargetConstant(PPC::sub_32, dl, MVT::i32);
3065 return SDValue(CurDAG->getMachineNode(PPC::EXTRACT_SUBREG, dl, MVT::i32,
3066 NatWidthRes, SubRegIdx), 0);
3067}
3068
3069// Produce a GPR sequence for compound comparisons (<=, >=) against zero.
3070// Handle both zero-extensions and sign-extensions.
3071SDValue
3072IntegerCompareEliminator::getCompoundZeroComparisonInGPR(SDValue LHS, SDLoc dl,
3073 ZeroCompare CmpTy) {
3074 EVT InVT = LHS.getValueType();
3075 bool Is32Bit = InVT == MVT::i32;
3076 SDValue ToExtend;
3077
3078 // Produce the value that needs to be either zero or sign extended.
3079 switch (CmpTy) {
3080 case ZeroCompare::GEZExt:
3081 case ZeroCompare::GESExt:
3082 ToExtend = SDValue(CurDAG->getMachineNode(Is32Bit ? PPC::NOR : PPC::NOR8,
3083 dl, InVT, LHS, LHS), 0);
3084 break;
3085 case ZeroCompare::LEZExt:
3086 case ZeroCompare::LESExt: {
3087 if (Is32Bit) {
3088 // Upper 32 bits cannot be undefined for this sequence.
3089 LHS = signExtendInputIfNeeded(LHS);
3090 SDValue Neg =
3091 SDValue(CurDAG->getMachineNode(PPC::NEG8, dl, MVT::i64, LHS), 0);
3092 ToExtend =
3093 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3094 Neg, S->getI64Imm(1, dl),
3095 S->getI64Imm(63, dl)), 0);
3096 } else {
3097 SDValue Addi =
3098 SDValue(CurDAG->getMachineNode(PPC::ADDI8, dl, MVT::i64, LHS,
3099 S->getI64Imm(~0ULL, dl)), 0);
3100 ToExtend = SDValue(CurDAG->getMachineNode(PPC::OR8, dl, MVT::i64,
3101 Addi, LHS), 0);
3102 }
3103 break;
3104 }
3105 }
3106
3107 // For 64-bit sequences, the extensions are the same for the GE/LE cases.
3108 if (!Is32Bit &&
3109 (CmpTy == ZeroCompare::GEZExt || CmpTy == ZeroCompare::LEZExt))
3110 return SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3111 ToExtend, S->getI64Imm(1, dl),
3112 S->getI64Imm(63, dl)), 0);
3113 if (!Is32Bit &&
3114 (CmpTy == ZeroCompare::GESExt || CmpTy == ZeroCompare::LESExt))
3115 return SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64, ToExtend,
3116 S->getI64Imm(63, dl)), 0);
3117
3118 assert(Is32Bit && "Should have handled the 32-bit sequences above.")(static_cast <bool> (Is32Bit && "Should have handled the 32-bit sequences above."
) ? void (0) : __assert_fail ("Is32Bit && \"Should have handled the 32-bit sequences above.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3118, __extension__ __PRETTY_FUNCTION__))
;
3119 // For 32-bit sequences, the extensions differ between GE/LE cases.
3120 switch (CmpTy) {
3121 case ZeroCompare::GEZExt: {
3122 SDValue ShiftOps[] = { ToExtend, S->getI32Imm(1, dl), S->getI32Imm(31, dl),
3123 S->getI32Imm(31, dl) };
3124 return SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32,
3125 ShiftOps), 0);
3126 }
3127 case ZeroCompare::GESExt:
3128 return SDValue(CurDAG->getMachineNode(PPC::SRAWI, dl, MVT::i32, ToExtend,
3129 S->getI32Imm(31, dl)), 0);
3130 case ZeroCompare::LEZExt:
3131 return SDValue(CurDAG->getMachineNode(PPC::XORI8, dl, MVT::i64, ToExtend,
3132 S->getI32Imm(1, dl)), 0);
3133 case ZeroCompare::LESExt:
3134 return SDValue(CurDAG->getMachineNode(PPC::ADDI8, dl, MVT::i64, ToExtend,
3135 S->getI32Imm(-1, dl)), 0);
3136 }
3137
3138 // The above case covers all the enumerators so it can't have a default clause
3139 // to avoid compiler warnings.
3140 llvm_unreachable("Unknown zero-comparison type.")::llvm::llvm_unreachable_internal("Unknown zero-comparison type."
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3140)
;
3141}
3142
3143/// Produces a zero-extended result of comparing two 32-bit values according to
3144/// the passed condition code.
3145SDValue
3146IntegerCompareEliminator::get32BitZExtCompare(SDValue LHS, SDValue RHS,
3147 ISD::CondCode CC,
3148 int64_t RHSValue, SDLoc dl) {
3149 if (CmpInGPR == ICGPR_I64 || CmpInGPR == ICGPR_SextI64 ||
3150 CmpInGPR == ICGPR_ZextI64 || CmpInGPR == ICGPR_Sext)
3151 return SDValue();
3152 bool IsRHSZero = RHSValue == 0;
3153 bool IsRHSOne = RHSValue == 1;
3154 bool IsRHSNegOne = RHSValue == -1LL;
3155 switch (CC) {
3156 default: return SDValue();
3157 case ISD::SETEQ: {
3158 // (zext (setcc %a, %b, seteq)) -> (lshr (cntlzw (xor %a, %b)), 5)
3159 // (zext (setcc %a, 0, seteq)) -> (lshr (cntlzw %a), 5)
3160 SDValue Xor = IsRHSZero ? LHS :
3161 SDValue(CurDAG->getMachineNode(PPC::XOR, dl, MVT::i32, LHS, RHS), 0);
3162 SDValue Clz =
3163 SDValue(CurDAG->getMachineNode(PPC::CNTLZW, dl, MVT::i32, Xor), 0);
3164 SDValue ShiftOps[] = { Clz, S->getI32Imm(27, dl), S->getI32Imm(5, dl),
3165 S->getI32Imm(31, dl) };
3166 return SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32,
3167 ShiftOps), 0);
3168 }
3169 case ISD::SETNE: {
3170 // (zext (setcc %a, %b, setne)) -> (xor (lshr (cntlzw (xor %a, %b)), 5), 1)
3171 // (zext (setcc %a, 0, setne)) -> (xor (lshr (cntlzw %a), 5), 1)
3172 SDValue Xor = IsRHSZero ? LHS :
3173 SDValue(CurDAG->getMachineNode(PPC::XOR, dl, MVT::i32, LHS, RHS), 0);
3174 SDValue Clz =
3175 SDValue(CurDAG->getMachineNode(PPC::CNTLZW, dl, MVT::i32, Xor), 0);
3176 SDValue ShiftOps[] = { Clz, S->getI32Imm(27, dl), S->getI32Imm(5, dl),
3177 S->getI32Imm(31, dl) };
3178 SDValue Shift =
3179 SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, ShiftOps), 0);
3180 return SDValue(CurDAG->getMachineNode(PPC::XORI, dl, MVT::i32, Shift,
3181 S->getI32Imm(1, dl)), 0);
3182 }
3183 case ISD::SETGE: {
3184 // (zext (setcc %a, %b, setge)) -> (xor (lshr (sub %a, %b), 63), 1)
3185 // (zext (setcc %a, 0, setge)) -> (lshr (~ %a), 31)
3186 if(IsRHSZero)
3187 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::GEZExt);
3188
3189 // Not a special case (i.e. RHS == 0). Handle (%a >= %b) as (%b <= %a)
3190 // by swapping inputs and falling through.
3191 std::swap(LHS, RHS);
3192 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3193 IsRHSZero = RHSConst && RHSConst->isNullValue();
3194 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3195 }
3196 case ISD::SETLE: {
3197 if (CmpInGPR == ICGPR_NonExtIn)
3198 return SDValue();
3199 // (zext (setcc %a, %b, setle)) -> (xor (lshr (sub %b, %a), 63), 1)
3200 // (zext (setcc %a, 0, setle)) -> (xor (lshr (- %a), 63), 1)
3201 if(IsRHSZero) {
3202 if (CmpInGPR == ICGPR_NonExtIn)
3203 return SDValue();
3204 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::LEZExt);
3205 }
3206
3207 // The upper 32-bits of the register can't be undefined for this sequence.
3208 LHS = signExtendInputIfNeeded(LHS);
3209 RHS = signExtendInputIfNeeded(RHS);
3210 SDValue Sub =
3211 SDValue(CurDAG->getMachineNode(PPC::SUBF8, dl, MVT::i64, LHS, RHS), 0);
3212 SDValue Shift =
3213 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, Sub,
3214 S->getI64Imm(1, dl), S->getI64Imm(63, dl)),
3215 0);
3216 return
3217 SDValue(CurDAG->getMachineNode(PPC::XORI8, dl,
3218 MVT::i64, Shift, S->getI32Imm(1, dl)), 0);
3219 }
3220 case ISD::SETGT: {
3221 // (zext (setcc %a, %b, setgt)) -> (lshr (sub %b, %a), 63)
3222 // (zext (setcc %a, -1, setgt)) -> (lshr (~ %a), 31)
3223 // (zext (setcc %a, 0, setgt)) -> (lshr (- %a), 63)
3224 // Handle SETLT -1 (which is equivalent to SETGE 0).
3225 if (IsRHSNegOne)
3226 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::GEZExt);
3227
3228 if (IsRHSZero) {
3229 if (CmpInGPR == ICGPR_NonExtIn)
3230 return SDValue();
3231 // The upper 32-bits of the register can't be undefined for this sequence.
3232 LHS = signExtendInputIfNeeded(LHS);
3233 RHS = signExtendInputIfNeeded(RHS);
3234 SDValue Neg =
3235 SDValue(CurDAG->getMachineNode(PPC::NEG8, dl, MVT::i64, LHS), 0);
3236 return SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3237 Neg, S->getI32Imm(1, dl), S->getI32Imm(63, dl)), 0);
3238 }
3239 // Not a special case (i.e. RHS == 0 or RHS == -1). Handle (%a > %b) as
3240 // (%b < %a) by swapping inputs and falling through.
3241 std::swap(LHS, RHS);
3242 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3243 IsRHSZero = RHSConst && RHSConst->isNullValue();
3244 IsRHSOne = RHSConst && RHSConst->getSExtValue() == 1;
3245 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3246 }
3247 case ISD::SETLT: {
3248 // (zext (setcc %a, %b, setlt)) -> (lshr (sub %a, %b), 63)
3249 // (zext (setcc %a, 1, setlt)) -> (xor (lshr (- %a), 63), 1)
3250 // (zext (setcc %a, 0, setlt)) -> (lshr %a, 31)
3251 // Handle SETLT 1 (which is equivalent to SETLE 0).
3252 if (IsRHSOne) {
3253 if (CmpInGPR == ICGPR_NonExtIn)
3254 return SDValue();
3255 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::LEZExt);
3256 }
3257
3258 if (IsRHSZero) {
3259 SDValue ShiftOps[] = { LHS, S->getI32Imm(1, dl), S->getI32Imm(31, dl),
3260 S->getI32Imm(31, dl) };
3261 return SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32,
3262 ShiftOps), 0);
3263 }
3264
3265 if (CmpInGPR == ICGPR_NonExtIn)
3266 return SDValue();
3267 // The upper 32-bits of the register can't be undefined for this sequence.
3268 LHS = signExtendInputIfNeeded(LHS);
3269 RHS = signExtendInputIfNeeded(RHS);
3270 SDValue SUBFNode =
3271 SDValue(CurDAG->getMachineNode(PPC::SUBF8, dl, MVT::i64, RHS, LHS), 0);
3272 return SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3273 SUBFNode, S->getI64Imm(1, dl),
3274 S->getI64Imm(63, dl)), 0);
3275 }
3276 case ISD::SETUGE:
3277 // (zext (setcc %a, %b, setuge)) -> (xor (lshr (sub %b, %a), 63), 1)
3278 // (zext (setcc %a, %b, setule)) -> (xor (lshr (sub %a, %b), 63), 1)
3279 std::swap(LHS, RHS);
3280 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3281 case ISD::SETULE: {
3282 if (CmpInGPR == ICGPR_NonExtIn)
3283 return SDValue();
3284 // The upper 32-bits of the register can't be undefined for this sequence.
3285 LHS = zeroExtendInputIfNeeded(LHS);
3286 RHS = zeroExtendInputIfNeeded(RHS);
3287 SDValue Subtract =
3288 SDValue(CurDAG->getMachineNode(PPC::SUBF8, dl, MVT::i64, LHS, RHS), 0);
3289 SDValue SrdiNode =
3290 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3291 Subtract, S->getI64Imm(1, dl),
3292 S->getI64Imm(63, dl)), 0);
3293 return SDValue(CurDAG->getMachineNode(PPC::XORI8, dl, MVT::i64, SrdiNode,
3294 S->getI32Imm(1, dl)), 0);
3295 }
3296 case ISD::SETUGT:
3297 // (zext (setcc %a, %b, setugt)) -> (lshr (sub %b, %a), 63)
3298 // (zext (setcc %a, %b, setult)) -> (lshr (sub %a, %b), 63)
3299 std::swap(LHS, RHS);
3300 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3301 case ISD::SETULT: {
3302 if (CmpInGPR == ICGPR_NonExtIn)
3303 return SDValue();
3304 // The upper 32-bits of the register can't be undefined for this sequence.
3305 LHS = zeroExtendInputIfNeeded(LHS);
3306 RHS = zeroExtendInputIfNeeded(RHS);
3307 SDValue Subtract =
3308 SDValue(CurDAG->getMachineNode(PPC::SUBF8, dl, MVT::i64, RHS, LHS), 0);
3309 return SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3310 Subtract, S->getI64Imm(1, dl),
3311 S->getI64Imm(63, dl)), 0);
3312 }
3313 }
3314}
3315
3316/// Produces a sign-extended result of comparing two 32-bit values according to
3317/// the passed condition code.
3318SDValue
3319IntegerCompareEliminator::get32BitSExtCompare(SDValue LHS, SDValue RHS,
3320 ISD::CondCode CC,
3321 int64_t RHSValue, SDLoc dl) {
3322 if (CmpInGPR == ICGPR_I64 || CmpInGPR == ICGPR_SextI64 ||
3323 CmpInGPR == ICGPR_ZextI64 || CmpInGPR == ICGPR_Zext)
3324 return SDValue();
3325 bool IsRHSZero = RHSValue == 0;
3326 bool IsRHSOne = RHSValue == 1;
3327 bool IsRHSNegOne = RHSValue == -1LL;
3328
3329 switch (CC) {
3330 default: return SDValue();
3331 case ISD::SETEQ: {
3332 // (sext (setcc %a, %b, seteq)) ->
3333 // (ashr (shl (ctlz (xor %a, %b)), 58), 63)
3334 // (sext (setcc %a, 0, seteq)) ->
3335 // (ashr (shl (ctlz %a), 58), 63)
3336 SDValue CountInput = IsRHSZero ? LHS :
3337 SDValue(CurDAG->getMachineNode(PPC::XOR, dl, MVT::i32, LHS, RHS), 0);
3338 SDValue Cntlzw =
3339 SDValue(CurDAG->getMachineNode(PPC::CNTLZW, dl, MVT::i32, CountInput), 0);
3340 SDValue SHLOps[] = { Cntlzw, S->getI32Imm(27, dl),
3341 S->getI32Imm(5, dl), S->getI32Imm(31, dl) };
3342 SDValue Slwi =
3343 SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, SHLOps), 0);
3344 return SDValue(CurDAG->getMachineNode(PPC::NEG, dl, MVT::i32, Slwi), 0);
3345 }
3346 case ISD::SETNE: {
3347 // Bitwise xor the operands, count leading zeros, shift right by 5 bits and
3348 // flip the bit, finally take 2's complement.
3349 // (sext (setcc %a, %b, setne)) ->
3350 // (neg (xor (lshr (ctlz (xor %a, %b)), 5), 1))
3351 // Same as above, but the first xor is not needed.
3352 // (sext (setcc %a, 0, setne)) ->
3353 // (neg (xor (lshr (ctlz %a), 5), 1))
3354 SDValue Xor = IsRHSZero ? LHS :
3355 SDValue(CurDAG->getMachineNode(PPC::XOR, dl, MVT::i32, LHS, RHS), 0);
3356 SDValue Clz =
3357 SDValue(CurDAG->getMachineNode(PPC::CNTLZW, dl, MVT::i32, Xor), 0);
3358 SDValue ShiftOps[] =
3359 { Clz, S->getI32Imm(27, dl), S->getI32Imm(5, dl), S->getI32Imm(31, dl) };
3360 SDValue Shift =
3361 SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, ShiftOps), 0);
3362 SDValue Xori =
3363 SDValue(CurDAG->getMachineNode(PPC::XORI, dl, MVT::i32, Shift,
3364 S->getI32Imm(1, dl)), 0);
3365 return SDValue(CurDAG->getMachineNode(PPC::NEG, dl, MVT::i32, Xori), 0);
3366 }
3367 case ISD::SETGE: {
3368 // (sext (setcc %a, %b, setge)) -> (add (lshr (sub %a, %b), 63), -1)
3369 // (sext (setcc %a, 0, setge)) -> (ashr (~ %a), 31)
3370 if (IsRHSZero)
3371 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::GESExt);
3372
3373 // Not a special case (i.e. RHS == 0). Handle (%a >= %b) as (%b <= %a)
3374 // by swapping inputs and falling through.
3375 std::swap(LHS, RHS);
3376 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3377 IsRHSZero = RHSConst && RHSConst->isNullValue();
3378 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3379 }
3380 case ISD::SETLE: {
3381 if (CmpInGPR == ICGPR_NonExtIn)
3382 return SDValue();
3383 // (sext (setcc %a, %b, setge)) -> (add (lshr (sub %b, %a), 63), -1)
3384 // (sext (setcc %a, 0, setle)) -> (add (lshr (- %a), 63), -1)
3385 if (IsRHSZero)
3386 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::LESExt);
3387
3388 // The upper 32-bits of the register can't be undefined for this sequence.
3389 LHS = signExtendInputIfNeeded(LHS);
3390 RHS = signExtendInputIfNeeded(RHS);
3391 SDValue SUBFNode =
3392 SDValue(CurDAG->getMachineNode(PPC::SUBF8, dl, MVT::i64, MVT::Glue,
3393 LHS, RHS), 0);
3394 SDValue Srdi =
3395 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3396 SUBFNode, S->getI64Imm(1, dl),
3397 S->getI64Imm(63, dl)), 0);
3398 return SDValue(CurDAG->getMachineNode(PPC::ADDI8, dl, MVT::i64, Srdi,
3399 S->getI32Imm(-1, dl)), 0);
3400 }
3401 case ISD::SETGT: {
3402 // (sext (setcc %a, %b, setgt)) -> (ashr (sub %b, %a), 63)
3403 // (sext (setcc %a, -1, setgt)) -> (ashr (~ %a), 31)
3404 // (sext (setcc %a, 0, setgt)) -> (ashr (- %a), 63)
3405 if (IsRHSNegOne)
3406 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::GESExt);
3407 if (IsRHSZero) {
3408 if (CmpInGPR == ICGPR_NonExtIn)
3409 return SDValue();
3410 // The upper 32-bits of the register can't be undefined for this sequence.
3411 LHS = signExtendInputIfNeeded(LHS);
3412 RHS = signExtendInputIfNeeded(RHS);
3413 SDValue Neg =
3414 SDValue(CurDAG->getMachineNode(PPC::NEG8, dl, MVT::i64, LHS), 0);
3415 return SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64, Neg,
3416 S->getI64Imm(63, dl)), 0);
3417 }
3418 // Not a special case (i.e. RHS == 0 or RHS == -1). Handle (%a > %b) as
3419 // (%b < %a) by swapping inputs and falling through.
3420 std::swap(LHS, RHS);
3421 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3422 IsRHSZero = RHSConst && RHSConst->isNullValue();
3423 IsRHSOne = RHSConst && RHSConst->getSExtValue() == 1;
3424 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3425 }
3426 case ISD::SETLT: {
3427 // (sext (setcc %a, %b, setgt)) -> (ashr (sub %a, %b), 63)
3428 // (sext (setcc %a, 1, setgt)) -> (add (lshr (- %a), 63), -1)
3429 // (sext (setcc %a, 0, setgt)) -> (ashr %a, 31)
3430 if (IsRHSOne) {
3431 if (CmpInGPR == ICGPR_NonExtIn)
3432 return SDValue();
3433 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::LESExt);
3434 }
3435 if (IsRHSZero)
3436 return SDValue(CurDAG->getMachineNode(PPC::SRAWI, dl, MVT::i32, LHS,
3437 S->getI32Imm(31, dl)), 0);
3438
3439 if (CmpInGPR == ICGPR_NonExtIn)
3440 return SDValue();
3441 // The upper 32-bits of the register can't be undefined for this sequence.
3442 LHS = signExtendInputIfNeeded(LHS);
3443 RHS = signExtendInputIfNeeded(RHS);
3444 SDValue SUBFNode =
3445 SDValue(CurDAG->getMachineNode(PPC::SUBF8, dl, MVT::i64, RHS, LHS), 0);
3446 return SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64,
3447 SUBFNode, S->getI64Imm(63, dl)), 0);
3448 }
3449 case ISD::SETUGE:
3450 // (sext (setcc %a, %b, setuge)) -> (add (lshr (sub %a, %b), 63), -1)
3451 // (sext (setcc %a, %b, setule)) -> (add (lshr (sub %b, %a), 63), -1)
3452 std::swap(LHS, RHS);
3453 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3454 case ISD::SETULE: {
3455 if (CmpInGPR == ICGPR_NonExtIn)
3456 return SDValue();
3457 // The upper 32-bits of the register can't be undefined for this sequence.
3458 LHS = zeroExtendInputIfNeeded(LHS);
3459 RHS = zeroExtendInputIfNeeded(RHS);
3460 SDValue Subtract =
3461 SDValue(CurDAG->getMachineNode(PPC::SUBF8, dl, MVT::i64, LHS, RHS), 0);
3462 SDValue Shift =
3463 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, Subtract,
3464 S->getI32Imm(1, dl), S->getI32Imm(63,dl)),
3465 0);
3466 return SDValue(CurDAG->getMachineNode(PPC::ADDI8, dl, MVT::i64, Shift,
3467 S->getI32Imm(-1, dl)), 0);
3468 }
3469 case ISD::SETUGT:
3470 // (sext (setcc %a, %b, setugt)) -> (ashr (sub %b, %a), 63)
3471 // (sext (setcc %a, %b, setugt)) -> (ashr (sub %a, %b), 63)
3472 std::swap(LHS, RHS);
3473 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3474 case ISD::SETULT: {
3475 if (CmpInGPR == ICGPR_NonExtIn)
3476 return SDValue();
3477 // The upper 32-bits of the register can't be undefined for this sequence.
3478 LHS = zeroExtendInputIfNeeded(LHS);
3479 RHS = zeroExtendInputIfNeeded(RHS);
3480 SDValue Subtract =
3481 SDValue(CurDAG->getMachineNode(PPC::SUBF8, dl, MVT::i64, RHS, LHS), 0);
3482 return SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64,
3483 Subtract, S->getI64Imm(63, dl)), 0);
3484 }
3485 }
3486}
3487
3488/// Produces a zero-extended result of comparing two 64-bit values according to
3489/// the passed condition code.
3490SDValue
3491IntegerCompareEliminator::get64BitZExtCompare(SDValue LHS, SDValue RHS,
3492 ISD::CondCode CC,
3493 int64_t RHSValue, SDLoc dl) {
3494 if (CmpInGPR == ICGPR_I32 || CmpInGPR == ICGPR_SextI32 ||
3495 CmpInGPR == ICGPR_ZextI32 || CmpInGPR == ICGPR_Sext)
3496 return SDValue();
3497 bool IsRHSZero = RHSValue == 0;
3498 bool IsRHSOne = RHSValue == 1;
3499 bool IsRHSNegOne = RHSValue == -1LL;
3500 switch (CC) {
3501 default: return SDValue();
3502 case ISD::SETEQ: {
3503 // (zext (setcc %a, %b, seteq)) -> (lshr (ctlz (xor %a, %b)), 6)
3504 // (zext (setcc %a, 0, seteq)) -> (lshr (ctlz %a), 6)
3505 SDValue Xor = IsRHSZero ? LHS :
3506 SDValue(CurDAG->getMachineNode(PPC::XOR8, dl, MVT::i64, LHS, RHS), 0);
3507 SDValue Clz =
3508 SDValue(CurDAG->getMachineNode(PPC::CNTLZD, dl, MVT::i64, Xor), 0);
3509 return SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, Clz,
3510 S->getI64Imm(58, dl),
3511 S->getI64Imm(63, dl)), 0);
3512 }
3513 case ISD::SETNE: {
3514 // {addc.reg, addc.CA} = (addcarry (xor %a, %b), -1)
3515 // (zext (setcc %a, %b, setne)) -> (sube addc.reg, addc.reg, addc.CA)
3516 // {addcz.reg, addcz.CA} = (addcarry %a, -1)
3517 // (zext (setcc %a, 0, setne)) -> (sube addcz.reg, addcz.reg, addcz.CA)
3518 SDValue Xor = IsRHSZero ? LHS :
3519 SDValue(CurDAG->getMachineNode(PPC::XOR8, dl, MVT::i64, LHS, RHS), 0);
3520 SDValue AC =
3521 SDValue(CurDAG->getMachineNode(PPC::ADDIC8, dl, MVT::i64, MVT::Glue,
3522 Xor, S->getI32Imm(~0U, dl)), 0);
3523 return SDValue(CurDAG->getMachineNode(PPC::SUBFE8, dl, MVT::i64, AC,
3524 Xor, AC.getValue(1)), 0);
3525 }
3526 case ISD::SETGE: {
3527 // {subc.reg, subc.CA} = (subcarry %a, %b)
3528 // (zext (setcc %a, %b, setge)) ->
3529 // (adde (lshr %b, 63), (ashr %a, 63), subc.CA)
3530 // (zext (setcc %a, 0, setge)) -> (lshr (~ %a), 63)
3531 if (IsRHSZero)
3532 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::GEZExt);
3533 std::swap(LHS, RHS);
3534 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3535 IsRHSZero = RHSConst && RHSConst->isNullValue();
3536 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3537 }
3538 case ISD::SETLE: {
3539 // {subc.reg, subc.CA} = (subcarry %b, %a)
3540 // (zext (setcc %a, %b, setge)) ->
3541 // (adde (lshr %a, 63), (ashr %b, 63), subc.CA)
3542 // (zext (setcc %a, 0, setge)) -> (lshr (or %a, (add %a, -1)), 63)
3543 if (IsRHSZero)
3544 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::LEZExt);
3545 SDValue ShiftL =
3546 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, LHS,
3547 S->getI64Imm(1, dl),
3548 S->getI64Imm(63, dl)), 0);
3549 SDValue ShiftR =
3550 SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64, RHS,
3551 S->getI64Imm(63, dl)), 0);
3552 SDValue SubtractCarry =
3553 SDValue(CurDAG->getMachineNode(PPC::SUBFC8, dl, MVT::i64, MVT::Glue,
3554 LHS, RHS), 1);
3555 return SDValue(CurDAG->getMachineNode(PPC::ADDE8, dl, MVT::i64, MVT::Glue,
3556 ShiftR, ShiftL, SubtractCarry), 0);
3557 }
3558 case ISD::SETGT: {
3559 // {subc.reg, subc.CA} = (subcarry %b, %a)
3560 // (zext (setcc %a, %b, setgt)) ->
3561 // (xor (adde (lshr %a, 63), (ashr %b, 63), subc.CA), 1)
3562 // (zext (setcc %a, 0, setgt)) -> (lshr (nor (add %a, -1), %a), 63)
3563 if (IsRHSNegOne)
3564 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::GEZExt);
3565 if (IsRHSZero) {
3566 SDValue Addi =
3567 SDValue(CurDAG->getMachineNode(PPC::ADDI8, dl, MVT::i64, LHS,
3568 S->getI64Imm(~0ULL, dl)), 0);
3569 SDValue Nor =
3570 SDValue(CurDAG->getMachineNode(PPC::NOR8, dl, MVT::i64, Addi, LHS), 0);
3571 return SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, Nor,
3572 S->getI64Imm(1, dl),
3573 S->getI64Imm(63, dl)), 0);
3574 }
3575 std::swap(LHS, RHS);
3576 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3577 IsRHSZero = RHSConst && RHSConst->isNullValue();
3578 IsRHSOne = RHSConst && RHSConst->getSExtValue() == 1;
3579 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3580 }
3581 case ISD::SETLT: {
3582 // {subc.reg, subc.CA} = (subcarry %a, %b)
3583 // (zext (setcc %a, %b, setlt)) ->
3584 // (xor (adde (lshr %b, 63), (ashr %a, 63), subc.CA), 1)
3585 // (zext (setcc %a, 0, setlt)) -> (lshr %a, 63)
3586 if (IsRHSOne)
3587 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::LEZExt);
3588 if (IsRHSZero)
3589 return SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, LHS,
3590 S->getI64Imm(1, dl),
3591 S->getI64Imm(63, dl)), 0);
3592 SDValue SRADINode =
3593 SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64,
3594 LHS, S->getI64Imm(63, dl)), 0);
3595 SDValue SRDINode =
3596 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3597 RHS, S->getI64Imm(1, dl),
3598 S->getI64Imm(63, dl)), 0);
3599 SDValue SUBFC8Carry =
3600 SDValue(CurDAG->getMachineNode(PPC::SUBFC8, dl, MVT::i64, MVT::Glue,
3601 RHS, LHS), 1);
3602 SDValue ADDE8Node =
3603 SDValue(CurDAG->getMachineNode(PPC::ADDE8, dl, MVT::i64, MVT::Glue,
3604 SRDINode, SRADINode, SUBFC8Carry), 0);
3605 return SDValue(CurDAG->getMachineNode(PPC::XORI8, dl, MVT::i64,
3606 ADDE8Node, S->getI64Imm(1, dl)), 0);
3607 }
3608 case ISD::SETUGE:
3609 // {subc.reg, subc.CA} = (subcarry %a, %b)
3610 // (zext (setcc %a, %b, setuge)) -> (add (sube %b, %b, subc.CA), 1)
3611 std::swap(LHS, RHS);
3612 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3613 case ISD::SETULE: {
3614 // {subc.reg, subc.CA} = (subcarry %b, %a)
3615 // (zext (setcc %a, %b, setule)) -> (add (sube %a, %a, subc.CA), 1)
3616 SDValue SUBFC8Carry =
3617 SDValue(CurDAG->getMachineNode(PPC::SUBFC8, dl, MVT::i64, MVT::Glue,
3618 LHS, RHS), 1);
3619 SDValue SUBFE8Node =
3620 SDValue(CurDAG->getMachineNode(PPC::SUBFE8, dl, MVT::i64, MVT::Glue,
3621 LHS, LHS, SUBFC8Carry), 0);
3622 return SDValue(CurDAG->getMachineNode(PPC::ADDI8, dl, MVT::i64,
3623 SUBFE8Node, S->getI64Imm(1, dl)), 0);
3624 }
3625 case ISD::SETUGT:
3626 // {subc.reg, subc.CA} = (subcarry %b, %a)
3627 // (zext (setcc %a, %b, setugt)) -> -(sube %b, %b, subc.CA)
3628 std::swap(LHS, RHS);
3629 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3630 case ISD::SETULT: {
3631 // {subc.reg, subc.CA} = (subcarry %a, %b)
3632 // (zext (setcc %a, %b, setult)) -> -(sube %a, %a, subc.CA)
3633 SDValue SubtractCarry =
3634 SDValue(CurDAG->getMachineNode(PPC::SUBFC8, dl, MVT::i64, MVT::Glue,
3635 RHS, LHS), 1);
3636 SDValue ExtSub =
3637 SDValue(CurDAG->getMachineNode(PPC::SUBFE8, dl, MVT::i64,
3638 LHS, LHS, SubtractCarry), 0);
3639 return SDValue(CurDAG->getMachineNode(PPC::NEG8, dl, MVT::i64,
3640 ExtSub), 0);
3641 }
3642 }
3643}
3644
3645/// Produces a sign-extended result of comparing two 64-bit values according to
3646/// the passed condition code.
3647SDValue
3648IntegerCompareEliminator::get64BitSExtCompare(SDValue LHS, SDValue RHS,
3649 ISD::CondCode CC,
3650 int64_t RHSValue, SDLoc dl) {
3651 if (CmpInGPR == ICGPR_I32 || CmpInGPR == ICGPR_SextI32 ||
3652 CmpInGPR == ICGPR_ZextI32 || CmpInGPR == ICGPR_Zext)
3653 return SDValue();
3654 bool IsRHSZero = RHSValue == 0;
3655 bool IsRHSOne = RHSValue == 1;
3656 bool IsRHSNegOne = RHSValue == -1LL;
3657 switch (CC) {
3658 default: return SDValue();
3659 case ISD::SETEQ: {
3660 // {addc.reg, addc.CA} = (addcarry (xor %a, %b), -1)
3661 // (sext (setcc %a, %b, seteq)) -> (sube addc.reg, addc.reg, addc.CA)
3662 // {addcz.reg, addcz.CA} = (addcarry %a, -1)
3663 // (sext (setcc %a, 0, seteq)) -> (sube addcz.reg, addcz.reg, addcz.CA)
3664 SDValue AddInput = IsRHSZero ? LHS :
3665 SDValue(CurDAG->getMachineNode(PPC::XOR8, dl, MVT::i64, LHS, RHS), 0);
3666 SDValue Addic =
3667 SDValue(CurDAG->getMachineNode(PPC::ADDIC8, dl, MVT::i64, MVT::Glue,
3668 AddInput, S->getI32Imm(~0U, dl)), 0);
3669 return SDValue(CurDAG->getMachineNode(PPC::SUBFE8, dl, MVT::i64, Addic,
3670 Addic, Addic.getValue(1)), 0);
3671 }
3672 case ISD::SETNE: {
3673 // {subfc.reg, subfc.CA} = (subcarry 0, (xor %a, %b))
3674 // (sext (setcc %a, %b, setne)) -> (sube subfc.reg, subfc.reg, subfc.CA)
3675 // {subfcz.reg, subfcz.CA} = (subcarry 0, %a)
3676 // (sext (setcc %a, 0, setne)) -> (sube subfcz.reg, subfcz.reg, subfcz.CA)
3677 SDValue Xor = IsRHSZero ? LHS :
3678 SDValue(CurDAG->getMachineNode(PPC::XOR8, dl, MVT::i64, LHS, RHS), 0);
3679 SDValue SC =
3680 SDValue(CurDAG->getMachineNode(PPC::SUBFIC8, dl, MVT::i64, MVT::Glue,
3681 Xor, S->getI32Imm(0, dl)), 0);
3682 return SDValue(CurDAG->getMachineNode(PPC::SUBFE8, dl, MVT::i64, SC,
3683 SC, SC.getValue(1)), 0);
3684 }
3685 case ISD::SETGE: {
3686 // {subc.reg, subc.CA} = (subcarry %a, %b)
3687 // (zext (setcc %a, %b, setge)) ->
3688 // (- (adde (lshr %b, 63), (ashr %a, 63), subc.CA))
3689 // (zext (setcc %a, 0, setge)) -> (~ (ashr %a, 63))
3690 if (IsRHSZero)
3691 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::GESExt);
3692 std::swap(LHS, RHS);
3693 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3694 IsRHSZero = RHSConst && RHSConst->isNullValue();
3695 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3696 }
3697 case ISD::SETLE: {
3698 // {subc.reg, subc.CA} = (subcarry %b, %a)
3699 // (zext (setcc %a, %b, setge)) ->
3700 // (- (adde (lshr %a, 63), (ashr %b, 63), subc.CA))
3701 // (zext (setcc %a, 0, setge)) -> (ashr (or %a, (add %a, -1)), 63)
3702 if (IsRHSZero)
3703 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::LESExt);
3704 SDValue ShiftR =
3705 SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64, RHS,
3706 S->getI64Imm(63, dl)), 0);
3707 SDValue ShiftL =
3708 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64, LHS,
3709 S->getI64Imm(1, dl),
3710 S->getI64Imm(63, dl)), 0);
3711 SDValue SubtractCarry =
3712 SDValue(CurDAG->getMachineNode(PPC::SUBFC8, dl, MVT::i64, MVT::Glue,
3713 LHS, RHS), 1);
3714 SDValue Adde =
3715 SDValue(CurDAG->getMachineNode(PPC::ADDE8, dl, MVT::i64, MVT::Glue,
3716 ShiftR, ShiftL, SubtractCarry), 0);
3717 return SDValue(CurDAG->getMachineNode(PPC::NEG8, dl, MVT::i64, Adde), 0);
3718 }
3719 case ISD::SETGT: {
3720 // {subc.reg, subc.CA} = (subcarry %b, %a)
3721 // (zext (setcc %a, %b, setgt)) ->
3722 // -(xor (adde (lshr %a, 63), (ashr %b, 63), subc.CA), 1)
3723 // (zext (setcc %a, 0, setgt)) -> (ashr (nor (add %a, -1), %a), 63)
3724 if (IsRHSNegOne)
3725 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::GESExt);
3726 if (IsRHSZero) {
3727 SDValue Add =
3728 SDValue(CurDAG->getMachineNode(PPC::ADDI8, dl, MVT::i64, LHS,
3729 S->getI64Imm(-1, dl)), 0);
3730 SDValue Nor =
3731 SDValue(CurDAG->getMachineNode(PPC::NOR8, dl, MVT::i64, Add, LHS), 0);
3732 return SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64, Nor,
3733 S->getI64Imm(63, dl)), 0);
3734 }
3735 std::swap(LHS, RHS);
3736 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3737 IsRHSZero = RHSConst && RHSConst->isNullValue();
3738 IsRHSOne = RHSConst && RHSConst->getSExtValue() == 1;
3739 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3740 }
3741 case ISD::SETLT: {
3742 // {subc.reg, subc.CA} = (subcarry %a, %b)
3743 // (zext (setcc %a, %b, setlt)) ->
3744 // -(xor (adde (lshr %b, 63), (ashr %a, 63), subc.CA), 1)
3745 // (zext (setcc %a, 0, setlt)) -> (ashr %a, 63)
3746 if (IsRHSOne)
3747 return getCompoundZeroComparisonInGPR(LHS, dl, ZeroCompare::LESExt);
3748 if (IsRHSZero) {
3749 return SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64, LHS,
3750 S->getI64Imm(63, dl)), 0);
3751 }
3752 SDValue SRADINode =
3753 SDValue(CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64,
3754 LHS, S->getI64Imm(63, dl)), 0);
3755 SDValue SRDINode =
3756 SDValue(CurDAG->getMachineNode(PPC::RLDICL, dl, MVT::i64,
3757 RHS, S->getI64Imm(1, dl),
3758 S->getI64Imm(63, dl)), 0);
3759 SDValue SUBFC8Carry =
3760 SDValue(CurDAG->getMachineNode(PPC::SUBFC8, dl, MVT::i64, MVT::Glue,
3761 RHS, LHS), 1);
3762 SDValue ADDE8Node =
3763 SDValue(CurDAG->getMachineNode(PPC::ADDE8, dl, MVT::i64,
3764 SRDINode, SRADINode, SUBFC8Carry), 0);
3765 SDValue XORI8Node =
3766 SDValue(CurDAG->getMachineNode(PPC::XORI8, dl, MVT::i64,
3767 ADDE8Node, S->getI64Imm(1, dl)), 0);
3768 return SDValue(CurDAG->getMachineNode(PPC::NEG8, dl, MVT::i64,
3769 XORI8Node), 0);
3770 }
3771 case ISD::SETUGE:
3772 // {subc.reg, subc.CA} = (subcarry %a, %b)
3773 // (sext (setcc %a, %b, setuge)) -> ~(sube %b, %b, subc.CA)
3774 std::swap(LHS, RHS);
3775 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3776 case ISD::SETULE: {
3777 // {subc.reg, subc.CA} = (subcarry %b, %a)
3778 // (sext (setcc %a, %b, setule)) -> ~(sube %a, %a, subc.CA)
3779 SDValue SubtractCarry =
3780 SDValue(CurDAG->getMachineNode(PPC::SUBFC8, dl, MVT::i64, MVT::Glue,
3781 LHS, RHS), 1);
3782 SDValue ExtSub =
3783 SDValue(CurDAG->getMachineNode(PPC::SUBFE8, dl, MVT::i64, MVT::Glue, LHS,
3784 LHS, SubtractCarry), 0);
3785 return SDValue(CurDAG->getMachineNode(PPC::NOR8, dl, MVT::i64,
3786 ExtSub, ExtSub), 0);
3787 }
3788 case ISD::SETUGT:
3789 // {subc.reg, subc.CA} = (subcarry %b, %a)
3790 // (sext (setcc %a, %b, setugt)) -> (sube %b, %b, subc.CA)
3791 std::swap(LHS, RHS);
3792 LLVM_FALLTHROUGH[[gnu::fallthrough]];
3793 case ISD::SETULT: {
3794 // {subc.reg, subc.CA} = (subcarry %a, %b)
3795 // (sext (setcc %a, %b, setult)) -> (sube %a, %a, subc.CA)
3796 SDValue SubCarry =
3797 SDValue(CurDAG->getMachineNode(PPC::SUBFC8, dl, MVT::i64, MVT::Glue,
3798 RHS, LHS), 1);
3799 return SDValue(CurDAG->getMachineNode(PPC::SUBFE8, dl, MVT::i64,
3800 LHS, LHS, SubCarry), 0);
3801 }
3802 }
3803}
3804
3805/// Do all uses of this SDValue need the result in a GPR?
3806/// This is meant to be used on values that have type i1 since
3807/// it is somewhat meaningless to ask if values of other types
3808/// should be kept in GPR's.
3809static bool allUsesExtend(SDValue Compare, SelectionDAG *CurDAG) {
3810 assert(Compare.getOpcode() == ISD::SETCC &&(static_cast <bool> (Compare.getOpcode() == ISD::SETCC &&
"An ISD::SETCC node required here.") ? void (0) : __assert_fail
("Compare.getOpcode() == ISD::SETCC && \"An ISD::SETCC node required here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3811, __extension__ __PRETTY_FUNCTION__))
3811 "An ISD::SETCC node required here.")(static_cast <bool> (Compare.getOpcode() == ISD::SETCC &&
"An ISD::SETCC node required here.") ? void (0) : __assert_fail
("Compare.getOpcode() == ISD::SETCC && \"An ISD::SETCC node required here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3811, __extension__ __PRETTY_FUNCTION__))
;
3812
3813 // For values that have a single use, the caller should obviously already have
3814 // checked if that use is an extending use. We check the other uses here.
3815 if (Compare.hasOneUse())
3816 return true;
3817 // We want the value in a GPR if it is being extended, used for a select, or
3818 // used in logical operations.
3819 for (auto CompareUse : Compare.getNode()->uses())
3820 if (CompareUse->getOpcode() != ISD::SIGN_EXTEND &&
3821 CompareUse->getOpcode() != ISD::ZERO_EXTEND &&
3822 CompareUse->getOpcode() != ISD::SELECT &&
3823 !isLogicOp(CompareUse->getOpcode())) {
3824 OmittedForNonExtendUses++;
3825 return false;
3826 }
3827 return true;
3828}
3829
3830/// Returns an equivalent of a SETCC node but with the result the same width as
3831/// the inputs. This can also be used for SELECT_CC if either the true or false
3832/// values is a power of two while the other is zero.
3833SDValue IntegerCompareEliminator::getSETCCInGPR(SDValue Compare,
3834 SetccInGPROpts ConvOpts) {
3835 assert((Compare.getOpcode() == ISD::SETCC ||(static_cast <bool> ((Compare.getOpcode() == ISD::SETCC
|| Compare.getOpcode() == ISD::SELECT_CC) && "An ISD::SETCC node required here."
) ? void (0) : __assert_fail ("(Compare.getOpcode() == ISD::SETCC || Compare.getOpcode() == ISD::SELECT_CC) && \"An ISD::SETCC node required here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3837, __extension__ __PRETTY_FUNCTION__))
3836 Compare.getOpcode() == ISD::SELECT_CC) &&(static_cast <bool> ((Compare.getOpcode() == ISD::SETCC
|| Compare.getOpcode() == ISD::SELECT_CC) && "An ISD::SETCC node required here."
) ? void (0) : __assert_fail ("(Compare.getOpcode() == ISD::SETCC || Compare.getOpcode() == ISD::SELECT_CC) && \"An ISD::SETCC node required here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3837, __extension__ __PRETTY_FUNCTION__))
3837 "An ISD::SETCC node required here.")(static_cast <bool> ((Compare.getOpcode() == ISD::SETCC
|| Compare.getOpcode() == ISD::SELECT_CC) && "An ISD::SETCC node required here."
) ? void (0) : __assert_fail ("(Compare.getOpcode() == ISD::SETCC || Compare.getOpcode() == ISD::SELECT_CC) && \"An ISD::SETCC node required here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 3837, __extension__ __PRETTY_FUNCTION__))
;
3838
3839 // Don't convert this comparison to a GPR sequence because there are uses
3840 // of the i1 result (i.e. uses that require the result in the CR).
3841 if ((Compare.getOpcode() == ISD::SETCC) && !allUsesExtend(Compare, CurDAG))
3842 return SDValue();
3843
3844 SDValue LHS = Compare.getOperand(0);
3845 SDValue RHS = Compare.getOperand(1);
3846
3847 // The condition code is operand 2 for SETCC and operand 4 for SELECT_CC.
3848 int CCOpNum = Compare.getOpcode() == ISD::SELECT_CC ? 4 : 2;
3849 ISD::CondCode CC =
3850 cast<CondCodeSDNode>(Compare.getOperand(CCOpNum))->get();
3851 EVT InputVT = LHS.getValueType();
3852 if (InputVT != MVT::i32 && InputVT != MVT::i64)
3853 return SDValue();
3854
3855 if (ConvOpts == SetccInGPROpts::ZExtInvert ||
3856 ConvOpts == SetccInGPROpts::SExtInvert)
3857 CC = ISD::getSetCCInverse(CC, InputVT);
3858
3859 bool Inputs32Bit = InputVT == MVT::i32;
3860
3861 SDLoc dl(Compare);
3862 ConstantSDNode *RHSConst = dyn_cast<ConstantSDNode>(RHS);
3863 int64_t RHSValue = RHSConst ? RHSConst->getSExtValue() : INT64_MAX(9223372036854775807L);
3864 bool IsSext = ConvOpts == SetccInGPROpts::SExtOrig ||
3865 ConvOpts == SetccInGPROpts::SExtInvert;
3866
3867 if (IsSext && Inputs32Bit)
3868 return get32BitSExtCompare(LHS, RHS, CC, RHSValue, dl);
3869 else if (Inputs32Bit)
3870 return get32BitZExtCompare(LHS, RHS, CC, RHSValue, dl);
3871 else if (IsSext)
3872 return get64BitSExtCompare(LHS, RHS, CC, RHSValue, dl);
3873 return get64BitZExtCompare(LHS, RHS, CC, RHSValue, dl);
3874}
3875
3876} // end anonymous namespace
3877
3878bool PPCDAGToDAGISel::tryIntCompareInGPR(SDNode *N) {
3879 if (N->getValueType(0) != MVT::i32 &&
3880 N->getValueType(0) != MVT::i64)
3881 return false;
3882
3883 // This optimization will emit code that assumes 64-bit registers
3884 // so we don't want to run it in 32-bit mode. Also don't run it
3885 // on functions that are not to be optimized.
3886 if (TM.getOptLevel() == CodeGenOpt::None || !TM.isPPC64())
3887 return false;
3888
3889 // For POWER10, it is more profitable to use the set boolean extension
3890 // instructions rather than the integer compare elimination codegen.
3891 // Users can override this via the command line option, `--ppc-gpr-icmps`.
3892 if (!(CmpInGPR.getNumOccurrences() > 0) && Subtarget->isISA3_1())
3893 return false;
3894
3895 switch (N->getOpcode()) {
3896 default: break;
3897 case ISD::ZERO_EXTEND:
3898 case ISD::SIGN_EXTEND:
3899 case ISD::AND:
3900 case ISD::OR:
3901 case ISD::XOR: {
3902 IntegerCompareEliminator ICmpElim(CurDAG, this);
3903 if (SDNode *New = ICmpElim.Select(N)) {
3904 ReplaceNode(N, New);
3905 return true;
3906 }
3907 }
3908 }
3909 return false;
3910}
3911
3912bool PPCDAGToDAGISel::tryBitPermutation(SDNode *N) {
3913 if (N->getValueType(0) != MVT::i32 &&
3914 N->getValueType(0) != MVT::i64)
3915 return false;
3916
3917 if (!UseBitPermRewriter)
3918 return false;
3919
3920 switch (N->getOpcode()) {
3921 default: break;
3922 case ISD::ROTL:
3923 case ISD::SHL:
3924 case ISD::SRL:
3925 case ISD::AND:
3926 case ISD::OR: {
3927 BitPermutationSelector BPS(CurDAG);
3928 if (SDNode *New = BPS.Select(N)) {
3929 ReplaceNode(N, New);
3930 return true;
3931 }
3932 return false;
3933 }
3934 }
3935
3936 return false;
3937}
3938
3939/// SelectCC - Select a comparison of the specified values with the specified
3940/// condition code, returning the CR# of the expression.
3941SDValue PPCDAGToDAGISel::SelectCC(SDValue LHS, SDValue RHS, ISD::CondCode CC,
3942 const SDLoc &dl, SDValue Chain) {
3943 // Always select the LHS.
3944 unsigned Opc;
3945
3946 if (LHS.getValueType() == MVT::i32) {
3947 unsigned Imm;
3948 if (CC == ISD::SETEQ || CC == ISD::SETNE) {
3949 if (isInt32Immediate(RHS, Imm)) {
3950 // SETEQ/SETNE comparison with 16-bit immediate, fold it.
3951 if (isUInt<16>(Imm))
3952 return SDValue(CurDAG->getMachineNode(PPC::CMPLWI, dl, MVT::i32, LHS,
3953 getI32Imm(Imm & 0xFFFF, dl)),
3954 0);
3955 // If this is a 16-bit signed immediate, fold it.
3956 if (isInt<16>((int)Imm))
3957 return SDValue(CurDAG->getMachineNode(PPC::CMPWI, dl, MVT::i32, LHS,
3958 getI32Imm(Imm & 0xFFFF, dl)),
3959 0);
3960
3961 // For non-equality comparisons, the default code would materialize the
3962 // constant, then compare against it, like this:
3963 // lis r2, 4660
3964 // ori r2, r2, 22136
3965 // cmpw cr0, r3, r2
3966 // Since we are just comparing for equality, we can emit this instead:
3967 // xoris r0,r3,0x1234
3968 // cmplwi cr0,r0,0x5678
3969 // beq cr0,L6
3970 SDValue Xor(CurDAG->getMachineNode(PPC::XORIS, dl, MVT::i32, LHS,
3971 getI32Imm(Imm >> 16, dl)), 0);
3972 return SDValue(CurDAG->getMachineNode(PPC::CMPLWI, dl, MVT::i32, Xor,
3973 getI32Imm(Imm & 0xFFFF, dl)), 0);
3974 }
3975 Opc = PPC::CMPLW;
3976 } else if (ISD::isUnsignedIntSetCC(CC)) {
3977 if (isInt32Immediate(RHS, Imm) && isUInt<16>(Imm))
3978 return SDValue(CurDAG->getMachineNode(PPC::CMPLWI, dl, MVT::i32, LHS,
3979 getI32Imm(Imm & 0xFFFF, dl)), 0);
3980 Opc = PPC::CMPLW;
3981 } else {
3982 int16_t SImm;
3983 if (isIntS16Immediate(RHS, SImm))
3984 return SDValue(CurDAG->getMachineNode(PPC::CMPWI, dl, MVT::i32, LHS,
3985 getI32Imm((int)SImm & 0xFFFF,
3986 dl)),
3987 0);
3988 Opc = PPC::CMPW;
3989 }
3990 } else if (LHS.getValueType() == MVT::i64) {
3991 uint64_t Imm;
3992 if (CC == ISD::SETEQ || CC == ISD::SETNE) {
3993 if (isInt64Immediate(RHS.getNode(), Imm)) {
3994 // SETEQ/SETNE comparison with 16-bit immediate, fold it.
3995 if (isUInt<16>(Imm))
3996 return SDValue(CurDAG->getMachineNode(PPC::CMPLDI, dl, MVT::i64, LHS,
3997 getI32Imm(Imm & 0xFFFF, dl)),
3998 0);
3999 // If this is a 16-bit signed immediate, fold it.
4000 if (isInt<16>(Imm))
4001 return SDValue(CurDAG->getMachineNode(PPC::CMPDI, dl, MVT::i64, LHS,
4002 getI32Imm(Imm & 0xFFFF, dl)),
4003 0);
4004
4005 // For non-equality comparisons, the default code would materialize the
4006 // constant, then compare against it, like this:
4007 // lis r2, 4660
4008 // ori r2, r2, 22136
4009 // cmpd cr0, r3, r2
4010 // Since we are just comparing for equality, we can emit this instead:
4011 // xoris r0,r3,0x1234
4012 // cmpldi cr0,r0,0x5678
4013 // beq cr0,L6
4014 if (isUInt<32>(Imm)) {
4015 SDValue Xor(CurDAG->getMachineNode(PPC::XORIS8, dl, MVT::i64, LHS,
4016 getI64Imm(Imm >> 16, dl)), 0);
4017 return SDValue(CurDAG->getMachineNode(PPC::CMPLDI, dl, MVT::i64, Xor,
4018 getI64Imm(Imm & 0xFFFF, dl)),
4019 0);
4020 }
4021 }
4022 Opc = PPC::CMPLD;
4023 } else if (ISD::isUnsignedIntSetCC(CC)) {
4024 if (isInt64Immediate(RHS.getNode(), Imm) && isUInt<16>(Imm))
4025 return SDValue(CurDAG->getMachineNode(PPC::CMPLDI, dl, MVT::i64, LHS,
4026 getI64Imm(Imm & 0xFFFF, dl)), 0);
4027 Opc = PPC::CMPLD;
4028 } else {
4029 int16_t SImm;
4030 if (isIntS16Immediate(RHS, SImm))
4031 return SDValue(CurDAG->getMachineNode(PPC::CMPDI, dl, MVT::i64, LHS,
4032 getI64Imm(SImm & 0xFFFF, dl)),
4033 0);
4034 Opc = PPC::CMPD;
4035 }
4036 } else if (LHS.getValueType() == MVT::f32) {
4037 if (Subtarget->hasSPE()) {
4038 switch (CC) {
4039 default:
4040 case ISD::SETEQ:
4041 case ISD::SETNE:
4042 Opc = PPC::EFSCMPEQ;
4043 break;
4044 case ISD::SETLT:
4045 case ISD::SETGE:
4046 case ISD::SETOLT:
4047 case ISD::SETOGE:
4048 case ISD::SETULT:
4049 case ISD::SETUGE:
4050 Opc = PPC::EFSCMPLT;
4051 break;
4052 case ISD::SETGT:
4053 case ISD::SETLE:
4054 case ISD::SETOGT:
4055 case ISD::SETOLE:
4056 case ISD::SETUGT:
4057 case ISD::SETULE:
4058 Opc = PPC::EFSCMPGT;
4059 break;
4060 }
4061 } else
4062 Opc = PPC::FCMPUS;
4063 } else if (LHS.getValueType() == MVT::f64) {
4064 if (Subtarget->hasSPE()) {
4065 switch (CC) {
4066 default:
4067 case ISD::SETEQ:
4068 case ISD::SETNE:
4069 Opc = PPC::EFDCMPEQ;
4070 break;
4071 case ISD::SETLT:
4072 case ISD::SETGE:
4073 case ISD::SETOLT:
4074 case ISD::SETOGE:
4075 case ISD::SETULT:
4076 case ISD::SETUGE:
4077 Opc = PPC::EFDCMPLT;
4078 break;
4079 case ISD::SETGT:
4080 case ISD::SETLE:
4081 case ISD::SETOGT:
4082 case ISD::SETOLE:
4083 case ISD::SETUGT:
4084 case ISD::SETULE:
4085 Opc = PPC::EFDCMPGT;
4086 break;
4087 }
4088 } else
4089 Opc = Subtarget->hasVSX() ? PPC::XSCMPUDP : PPC::FCMPUD;
4090 } else {
4091 assert(LHS.getValueType() == MVT::f128 && "Unknown vt!")(static_cast <bool> (LHS.getValueType() == MVT::f128 &&
"Unknown vt!") ? void (0) : __assert_fail ("LHS.getValueType() == MVT::f128 && \"Unknown vt!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4091, __extension__ __PRETTY_FUNCTION__))
;
4092 assert(Subtarget->hasP9Vector() && "XSCMPUQP requires Power9 Vector")(static_cast <bool> (Subtarget->hasP9Vector() &&
"XSCMPUQP requires Power9 Vector") ? void (0) : __assert_fail
("Subtarget->hasP9Vector() && \"XSCMPUQP requires Power9 Vector\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4092, __extension__ __PRETTY_FUNCTION__))
;
4093 Opc = PPC::XSCMPUQP;
4094 }
4095 if (Chain)
4096 return SDValue(
4097 CurDAG->getMachineNode(Opc, dl, MVT::i32, MVT::Other, LHS, RHS, Chain),
4098 0);
4099 else
4100 return SDValue(CurDAG->getMachineNode(Opc, dl, MVT::i32, LHS, RHS), 0);
4101}
4102
4103static PPC::Predicate getPredicateForSetCC(ISD::CondCode CC, const EVT &VT,
4104 const PPCSubtarget *Subtarget) {
4105 // For SPE instructions, the result is in GT bit of the CR
4106 bool UseSPE = Subtarget->hasSPE() && VT.isFloatingPoint();
4107
4108 switch (CC) {
4109 case ISD::SETUEQ:
4110 case ISD::SETONE:
4111 case ISD::SETOLE:
4112 case ISD::SETOGE:
4113 llvm_unreachable("Should be lowered by legalize!")::llvm::llvm_unreachable_internal("Should be lowered by legalize!"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4113)
;
4114 default: llvm_unreachable("Unknown condition!")::llvm::llvm_unreachable_internal("Unknown condition!", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4114)
;
4115 case ISD::SETOEQ:
4116 case ISD::SETEQ:
4117 return UseSPE ? PPC::PRED_GT : PPC::PRED_EQ;
4118 case ISD::SETUNE:
4119 case ISD::SETNE:
4120 return UseSPE ? PPC::PRED_LE : PPC::PRED_NE;
4121 case ISD::SETOLT:
4122 case ISD::SETLT:
4123 return UseSPE ? PPC::PRED_GT : PPC::PRED_LT;
4124 case ISD::SETULE:
4125 case ISD::SETLE:
4126 return PPC::PRED_LE;
4127 case ISD::SETOGT:
4128 case ISD::SETGT:
4129 return PPC::PRED_GT;
4130 case ISD::SETUGE:
4131 case ISD::SETGE:
4132 return UseSPE ? PPC::PRED_LE : PPC::PRED_GE;
4133 case ISD::SETO: return PPC::PRED_NU;
4134 case ISD::SETUO: return PPC::PRED_UN;
4135 // These two are invalid for floating point. Assume we have int.
4136 case ISD::SETULT: return PPC::PRED_LT;
4137 case ISD::SETUGT: return PPC::PRED_GT;
4138 }
4139}
4140
4141/// getCRIdxForSetCC - Return the index of the condition register field
4142/// associated with the SetCC condition, and whether or not the field is
4143/// treated as inverted. That is, lt = 0; ge = 0 inverted.
4144static unsigned getCRIdxForSetCC(ISD::CondCode CC, bool &Invert) {
4145 Invert = false;
4146 switch (CC) {
4147 default: llvm_unreachable("Unknown condition!")::llvm::llvm_unreachable_internal("Unknown condition!", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4147)
;
4148 case ISD::SETOLT:
4149 case ISD::SETLT: return 0; // Bit #0 = SETOLT
4150 case ISD::SETOGT:
4151 case ISD::SETGT: return 1; // Bit #1 = SETOGT
4152 case ISD::SETOEQ:
4153 case ISD::SETEQ: return 2; // Bit #2 = SETOEQ
4154 case ISD::SETUO: return 3; // Bit #3 = SETUO
4155 case ISD::SETUGE:
4156 case ISD::SETGE: Invert = true; return 0; // !Bit #0 = SETUGE
4157 case ISD::SETULE:
4158 case ISD::SETLE: Invert = true; return 1; // !Bit #1 = SETULE
4159 case ISD::SETUNE:
4160 case ISD::SETNE: Invert = true; return 2; // !Bit #2 = SETUNE
4161 case ISD::SETO: Invert = true; return 3; // !Bit #3 = SETO
4162 case ISD::SETUEQ:
4163 case ISD::SETOGE:
4164 case ISD::SETOLE:
4165 case ISD::SETONE:
4166 llvm_unreachable("Invalid branch code: should be expanded by legalize")::llvm::llvm_unreachable_internal("Invalid branch code: should be expanded by legalize"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4166)
;
4167 // These are invalid for floating point. Assume integer.
4168 case ISD::SETULT: return 0;
4169 case ISD::SETUGT: return 1;
4170 }
4171}
4172
4173// getVCmpInst: return the vector compare instruction for the specified
4174// vector type and condition code. Since this is for altivec specific code,
4175// only support the altivec types (v16i8, v8i16, v4i32, v2i64, v1i128,
4176// and v4f32).
4177static unsigned int getVCmpInst(MVT VecVT, ISD::CondCode CC,
4178 bool HasVSX, bool &Swap, bool &Negate) {
4179 Swap = false;
4180 Negate = false;
4181
4182 if (VecVT.isFloatingPoint()) {
4183 /* Handle some cases by swapping input operands. */
4184 switch (CC) {
4185 case ISD::SETLE: CC = ISD::SETGE; Swap = true; break;
4186 case ISD::SETLT: CC = ISD::SETGT; Swap = true; break;
4187 case ISD::SETOLE: CC = ISD::SETOGE; Swap = true; break;
4188 case ISD::SETOLT: CC = ISD::SETOGT; Swap = true; break;
4189 case ISD::SETUGE: CC = ISD::SETULE; Swap = true; break;
4190 case ISD::SETUGT: CC = ISD::SETULT; Swap = true; break;
4191 default: break;
4192 }
4193 /* Handle some cases by negating the result. */
4194 switch (CC) {
4195 case ISD::SETNE: CC = ISD::SETEQ; Negate = true; break;
4196 case ISD::SETUNE: CC = ISD::SETOEQ; Negate = true; break;
4197 case ISD::SETULE: CC = ISD::SETOGT; Negate = true; break;
4198 case ISD::SETULT: CC = ISD::SETOGE; Negate = true; break;
4199 default: break;
4200 }
4201 /* We have instructions implementing the remaining cases. */
4202 switch (CC) {
4203 case ISD::SETEQ:
4204 case ISD::SETOEQ:
4205 if (VecVT == MVT::v4f32)
4206 return HasVSX ? PPC::XVCMPEQSP : PPC::VCMPEQFP;
4207 else if (VecVT == MVT::v2f64)
4208 return PPC::XVCMPEQDP;
4209 break;
4210 case ISD::SETGT:
4211 case ISD::SETOGT:
4212 if (VecVT == MVT::v4f32)
4213 return HasVSX ? PPC::XVCMPGTSP : PPC::VCMPGTFP;
4214 else if (VecVT == MVT::v2f64)
4215 return PPC::XVCMPGTDP;
4216 break;
4217 case ISD::SETGE:
4218 case ISD::SETOGE:
4219 if (VecVT == MVT::v4f32)
4220 return HasVSX ? PPC::XVCMPGESP : PPC::VCMPGEFP;
4221 else if (VecVT == MVT::v2f64)
4222 return PPC::XVCMPGEDP;
4223 break;
4224 default:
4225 break;
4226 }
4227 llvm_unreachable("Invalid floating-point vector compare condition")::llvm::llvm_unreachable_internal("Invalid floating-point vector compare condition"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4227)
;
4228 } else {
4229 /* Handle some cases by swapping input operands. */
4230 switch (CC) {
4231 case ISD::SETGE: CC = ISD::SETLE; Swap = true; break;
4232 case ISD::SETLT: CC = ISD::SETGT; Swap = true; break;
4233 case ISD::SETUGE: CC = ISD::SETULE; Swap = true; break;
4234 case ISD::SETULT: CC = ISD::SETUGT; Swap = true; break;
4235 default: break;
4236 }
4237 /* Handle some cases by negating the result. */
4238 switch (CC) {
4239 case ISD::SETNE: CC = ISD::SETEQ; Negate = true; break;
4240 case ISD::SETUNE: CC = ISD::SETUEQ; Negate = true; break;
4241 case ISD::SETLE: CC = ISD::SETGT; Negate = true; break;
4242 case ISD::SETULE: CC = ISD::SETUGT; Negate = true; break;
4243 default: break;
4244 }
4245 /* We have instructions implementing the remaining cases. */
4246 switch (CC) {
4247 case ISD::SETEQ:
4248 case ISD::SETUEQ:
4249 if (VecVT == MVT::v16i8)
4250 return PPC::VCMPEQUB;
4251 else if (VecVT == MVT::v8i16)
4252 return PPC::VCMPEQUH;
4253 else if (VecVT == MVT::v4i32)
4254 return PPC::VCMPEQUW;
4255 else if (VecVT == MVT::v2i64)
4256 return PPC::VCMPEQUD;
4257 else if (VecVT == MVT::v1i128)
4258 return PPC::VCMPEQUQ;
4259 break;
4260 case ISD::SETGT:
4261 if (VecVT == MVT::v16i8)
4262 return PPC::VCMPGTSB;
4263 else if (VecVT == MVT::v8i16)
4264 return PPC::VCMPGTSH;
4265 else if (VecVT == MVT::v4i32)
4266 return PPC::VCMPGTSW;
4267 else if (VecVT == MVT::v2i64)
4268 return PPC::VCMPGTSD;
4269 else if (VecVT == MVT::v1i128)
4270 return PPC::VCMPGTSQ;
4271 break;
4272 case ISD::SETUGT:
4273 if (VecVT == MVT::v16i8)
4274 return PPC::VCMPGTUB;
4275 else if (VecVT == MVT::v8i16)
4276 return PPC::VCMPGTUH;
4277 else if (VecVT == MVT::v4i32)
4278 return PPC::VCMPGTUW;
4279 else if (VecVT == MVT::v2i64)
4280 return PPC::VCMPGTUD;
4281 else if (VecVT == MVT::v1i128)
4282 return PPC::VCMPGTUQ;
4283 break;
4284 default:
4285 break;
4286 }
4287 llvm_unreachable("Invalid integer vector compare condition")::llvm::llvm_unreachable_internal("Invalid integer vector compare condition"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4287)
;
4288 }
4289}
4290
4291bool PPCDAGToDAGISel::trySETCC(SDNode *N) {
4292 SDLoc dl(N);
4293 unsigned Imm;
4294 bool IsStrict = N->isStrictFPOpcode();
4295 ISD::CondCode CC =
4296 cast<CondCodeSDNode>(N->getOperand(IsStrict ? 3 : 2))->get();
4297 EVT PtrVT =
4298 CurDAG->getTargetLoweringInfo().getPointerTy(CurDAG->getDataLayout());
4299 bool isPPC64 = (PtrVT == MVT::i64);
4300 SDValue Chain = IsStrict ? N->getOperand(0) : SDValue();
4301
4302 SDValue LHS = N->getOperand(IsStrict ? 1 : 0);
4303 SDValue RHS = N->getOperand(IsStrict ? 2 : 1);
4304
4305 if (!IsStrict && !Subtarget->useCRBits() && isInt32Immediate(RHS, Imm)) {
4306 // We can codegen setcc op, imm very efficiently compared to a brcond.
4307 // Check for those cases here.
4308 // setcc op, 0
4309 if (Imm == 0) {
4310 SDValue Op = LHS;
4311 switch (CC) {
4312 default: break;
4313 case ISD::SETEQ: {
4314 Op = SDValue(CurDAG->getMachineNode(PPC::CNTLZW, dl, MVT::i32, Op), 0);
4315 SDValue Ops[] = { Op, getI32Imm(27, dl), getI32Imm(5, dl),
4316 getI32Imm(31, dl) };
4317 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
4318 return true;
4319 }
4320 case ISD::SETNE: {
4321 if (isPPC64) break;
4322 SDValue AD =
4323 SDValue(CurDAG->getMachineNode(PPC::ADDIC, dl, MVT::i32, MVT::Glue,
4324 Op, getI32Imm(~0U, dl)), 0);
4325 CurDAG->SelectNodeTo(N, PPC::SUBFE, MVT::i32, AD, Op, AD.getValue(1));
4326 return true;
4327 }
4328 case ISD::SETLT: {
4329 SDValue Ops[] = { Op, getI32Imm(1, dl), getI32Imm(31, dl),
4330 getI32Imm(31, dl) };
4331 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
4332 return true;
4333 }
4334 case ISD::SETGT: {
4335 SDValue T =
4336 SDValue(CurDAG->getMachineNode(PPC::NEG, dl, MVT::i32, Op), 0);
4337 T = SDValue(CurDAG->getMachineNode(PPC::ANDC, dl, MVT::i32, T, Op), 0);
4338 SDValue Ops[] = { T, getI32Imm(1, dl), getI32Imm(31, dl),
4339 getI32Imm(31, dl) };
4340 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
4341 return true;
4342 }
4343 }
4344 } else if (Imm == ~0U) { // setcc op, -1
4345 SDValue Op = LHS;
4346 switch (CC) {
4347 default: break;
4348 case ISD::SETEQ:
4349 if (isPPC64) break;
4350 Op = SDValue(CurDAG->getMachineNode(PPC::ADDIC, dl, MVT::i32, MVT::Glue,
4351 Op, getI32Imm(1, dl)), 0);
4352 CurDAG->SelectNodeTo(N, PPC::ADDZE, MVT::i32,
4353 SDValue(CurDAG->getMachineNode(PPC::LI, dl,
4354 MVT::i32,
4355 getI32Imm(0, dl)),
4356 0), Op.getValue(1));
4357 return true;
4358 case ISD::SETNE: {
4359 if (isPPC64) break;
4360 Op = SDValue(CurDAG->getMachineNode(PPC::NOR, dl, MVT::i32, Op, Op), 0);
4361 SDNode *AD = CurDAG->getMachineNode(PPC::ADDIC, dl, MVT::i32, MVT::Glue,
4362 Op, getI32Imm(~0U, dl));
4363 CurDAG->SelectNodeTo(N, PPC::SUBFE, MVT::i32, SDValue(AD, 0), Op,
4364 SDValue(AD, 1));
4365 return true;
4366 }
4367 case ISD::SETLT: {
4368 SDValue AD = SDValue(CurDAG->getMachineNode(PPC::ADDI, dl, MVT::i32, Op,
4369 getI32Imm(1, dl)), 0);
4370 SDValue AN = SDValue(CurDAG->getMachineNode(PPC::AND, dl, MVT::i32, AD,
4371 Op), 0);
4372 SDValue Ops[] = { AN, getI32Imm(1, dl), getI32Imm(31, dl),
4373 getI32Imm(31, dl) };
4374 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
4375 return true;
4376 }
4377 case ISD::SETGT: {
4378 SDValue Ops[] = { Op, getI32Imm(1, dl), getI32Imm(31, dl),
4379 getI32Imm(31, dl) };
4380 Op = SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, Ops), 0);
4381 CurDAG->SelectNodeTo(N, PPC::XORI, MVT::i32, Op, getI32Imm(1, dl));
4382 return true;
4383 }
4384 }
4385 }
4386 }
4387
4388 // Altivec Vector compare instructions do not set any CR register by default and
4389 // vector compare operations return the same type as the operands.
4390 if (!IsStrict && LHS.getValueType().isVector()) {
4391 if (Subtarget->hasSPE())
4392 return false;
4393
4394 EVT VecVT = LHS.getValueType();
4395 bool Swap, Negate;
4396 unsigned int VCmpInst =
4397 getVCmpInst(VecVT.getSimpleVT(), CC, Subtarget->hasVSX(), Swap, Negate);
4398 if (Swap)
4399 std::swap(LHS, RHS);
4400
4401 EVT ResVT = VecVT.changeVectorElementTypeToInteger();
4402 if (Negate) {
4403 SDValue VCmp(CurDAG->getMachineNode(VCmpInst, dl, ResVT, LHS, RHS), 0);
4404 CurDAG->SelectNodeTo(N, Subtarget->hasVSX() ? PPC::XXLNOR : PPC::VNOR,
4405 ResVT, VCmp, VCmp);
4406 return true;
4407 }
4408
4409 CurDAG->SelectNodeTo(N, VCmpInst, ResVT, LHS, RHS);
4410 return true;
4411 }
4412
4413 if (Subtarget->useCRBits())
4414 return false;
4415
4416 bool Inv;
4417 unsigned Idx = getCRIdxForSetCC(CC, Inv);
4418 SDValue CCReg = SelectCC(LHS, RHS, CC, dl, Chain);
4419 if (IsStrict)
4420 CurDAG->ReplaceAllUsesOfValueWith(SDValue(N, 1), CCReg.getValue(1));
4421 SDValue IntCR;
4422
4423 // SPE e*cmp* instructions only set the 'gt' bit, so hard-code that
4424 // The correct compare instruction is already set by SelectCC()
4425 if (Subtarget->hasSPE() && LHS.getValueType().isFloatingPoint()) {
4426 Idx = 1;
4427 }
4428
4429 // Force the ccreg into CR7.
4430 SDValue CR7Reg = CurDAG->getRegister(PPC::CR7, MVT::i32);
4431
4432 SDValue InFlag(nullptr, 0); // Null incoming flag value.
4433 CCReg = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, CR7Reg, CCReg,
4434 InFlag).getValue(1);
4435
4436 IntCR = SDValue(CurDAG->getMachineNode(PPC::MFOCRF, dl, MVT::i32, CR7Reg,
4437 CCReg), 0);
4438
4439 SDValue Ops[] = { IntCR, getI32Imm((32 - (3 - Idx)) & 31, dl),
4440 getI32Imm(31, dl), getI32Imm(31, dl) };
4441 if (!Inv) {
4442 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
4443 return true;
4444 }
4445
4446 // Get the specified bit.
4447 SDValue Tmp =
4448 SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, Ops), 0);
4449 CurDAG->SelectNodeTo(N, PPC::XORI, MVT::i32, Tmp, getI32Imm(1, dl));
4450 return true;
4451}
4452
4453/// Does this node represent a load/store node whose address can be represented
4454/// with a register plus an immediate that's a multiple of \p Val:
4455bool PPCDAGToDAGISel::isOffsetMultipleOf(SDNode *N, unsigned Val) const {
4456 LoadSDNode *LDN = dyn_cast<LoadSDNode>(N);
4457 StoreSDNode *STN = dyn_cast<StoreSDNode>(N);
4458 SDValue AddrOp;
4459 if (LDN)
4460 AddrOp = LDN->getOperand(1);
4461 else if (STN)
4462 AddrOp = STN->getOperand(2);
4463
4464 // If the address points a frame object or a frame object with an offset,
4465 // we need to check the object alignment.
4466 short Imm = 0;
4467 if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(
4468 AddrOp.getOpcode() == ISD::ADD ? AddrOp.getOperand(0) :
4469 AddrOp)) {
4470 // If op0 is a frame index that is under aligned, we can't do it either,
4471 // because it is translated to r31 or r1 + slot + offset. We won't know the
4472 // slot number until the stack frame is finalized.
4473 const MachineFrameInfo &MFI = CurDAG->getMachineFunction().getFrameInfo();
4474 unsigned SlotAlign = MFI.getObjectAlign(FI->getIndex()).value();
4475 if ((SlotAlign % Val) != 0)
4476 return false;
4477
4478 // If we have an offset, we need further check on the offset.
4479 if (AddrOp.getOpcode() != ISD::ADD)
4480 return true;
4481 }
4482
4483 if (AddrOp.getOpcode() == ISD::ADD)
4484 return isIntS16Immediate(AddrOp.getOperand(1), Imm) && !(Imm % Val);
4485
4486 // If the address comes from the outside, the offset will be zero.
4487 return AddrOp.getOpcode() == ISD::CopyFromReg;
4488}
4489
4490void PPCDAGToDAGISel::transferMemOperands(SDNode *N, SDNode *Result) {
4491 // Transfer memoperands.
4492 MachineMemOperand *MemOp = cast<MemSDNode>(N)->getMemOperand();
4493 CurDAG->setNodeMemRefs(cast<MachineSDNode>(Result), {MemOp});
4494}
4495
4496static bool mayUseP9Setb(SDNode *N, const ISD::CondCode &CC, SelectionDAG *DAG,
4497 bool &NeedSwapOps, bool &IsUnCmp) {
4498
4499 assert(N->getOpcode() == ISD::SELECT_CC && "Expecting a SELECT_CC here.")(static_cast <bool> (N->getOpcode() == ISD::SELECT_CC
&& "Expecting a SELECT_CC here.") ? void (0) : __assert_fail
("N->getOpcode() == ISD::SELECT_CC && \"Expecting a SELECT_CC here.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4499, __extension__ __PRETTY_FUNCTION__))
;
4500
4501 SDValue LHS = N->getOperand(0);
4502 SDValue RHS = N->getOperand(1);
4503 SDValue TrueRes = N->getOperand(2);
4504 SDValue FalseRes = N->getOperand(3);
4505 ConstantSDNode *TrueConst = dyn_cast<ConstantSDNode>(TrueRes);
4506 if (!TrueConst || (N->getSimpleValueType(0) != MVT::i64 &&
4507 N->getSimpleValueType(0) != MVT::i32))
4508 return false;
4509
4510 // We are looking for any of:
4511 // (select_cc lhs, rhs, 1, (sext (setcc [lr]hs, [lr]hs, cc2)), cc1)
4512 // (select_cc lhs, rhs, -1, (zext (setcc [lr]hs, [lr]hs, cc2)), cc1)
4513 // (select_cc lhs, rhs, 0, (select_cc [lr]hs, [lr]hs, 1, -1, cc2), seteq)
4514 // (select_cc lhs, rhs, 0, (select_cc [lr]hs, [lr]hs, -1, 1, cc2), seteq)
4515 int64_t TrueResVal = TrueConst->getSExtValue();
4516 if ((TrueResVal < -1 || TrueResVal > 1) ||
4517 (TrueResVal == -1 && FalseRes.getOpcode() != ISD::ZERO_EXTEND) ||
4518 (TrueResVal == 1 && FalseRes.getOpcode() != ISD::SIGN_EXTEND) ||
4519 (TrueResVal == 0 &&
4520 (FalseRes.getOpcode() != ISD::SELECT_CC || CC != ISD::SETEQ)))
4521 return false;
4522
4523 SDValue SetOrSelCC = FalseRes.getOpcode() == ISD::SELECT_CC
4524 ? FalseRes
4525 : FalseRes.getOperand(0);
4526 bool InnerIsSel = SetOrSelCC.getOpcode() == ISD::SELECT_CC;
4527 if (SetOrSelCC.getOpcode() != ISD::SETCC &&
4528 SetOrSelCC.getOpcode() != ISD::SELECT_CC)
4529 return false;
4530
4531 // Without this setb optimization, the outer SELECT_CC will be manually
4532 // selected to SELECT_CC_I4/SELECT_CC_I8 Pseudo, then expand-isel-pseudos pass
4533 // transforms pseudo instruction to isel instruction. When there are more than
4534 // one use for result like zext/sext, with current optimization we only see
4535 // isel is replaced by setb but can't see any significant gain. Since
4536 // setb has longer latency than original isel, we should avoid this. Another
4537 // point is that setb requires comparison always kept, it can break the
4538 // opportunity to get the comparison away if we have in future.
4539 if (!SetOrSelCC.hasOneUse() || (!InnerIsSel && !FalseRes.hasOneUse()))
4540 return false;
4541
4542 SDValue InnerLHS = SetOrSelCC.getOperand(0);
4543 SDValue InnerRHS = SetOrSelCC.getOperand(1);
4544 ISD::CondCode InnerCC =
4545 cast<CondCodeSDNode>(SetOrSelCC.getOperand(InnerIsSel ? 4 : 2))->get();
4546 // If the inner comparison is a select_cc, make sure the true/false values are
4547 // 1/-1 and canonicalize it if needed.
4548 if (InnerIsSel) {
4549 ConstantSDNode *SelCCTrueConst =
4550 dyn_cast<ConstantSDNode>(SetOrSelCC.getOperand(2));
4551 ConstantSDNode *SelCCFalseConst =
4552 dyn_cast<ConstantSDNode>(SetOrSelCC.getOperand(3));
4553 if (!SelCCTrueConst || !SelCCFalseConst)
4554 return false;
4555 int64_t SelCCTVal = SelCCTrueConst->getSExtValue();
4556 int64_t SelCCFVal = SelCCFalseConst->getSExtValue();
4557 // The values must be -1/1 (requiring a swap) or 1/-1.
4558 if (SelCCTVal == -1 && SelCCFVal == 1) {
4559 std::swap(InnerLHS, InnerRHS);
4560 } else if (SelCCTVal != 1 || SelCCFVal != -1)
4561 return false;
4562 }
4563
4564 // Canonicalize unsigned case
4565 if (InnerCC == ISD::SETULT || InnerCC == ISD::SETUGT) {
4566 IsUnCmp = true;
4567 InnerCC = (InnerCC == ISD::SETULT) ? ISD::SETLT : ISD::SETGT;
4568 }
4569
4570 bool InnerSwapped = false;
4571 if (LHS == InnerRHS && RHS == InnerLHS)
4572 InnerSwapped = true;
4573 else if (LHS != InnerLHS || RHS != InnerRHS)
4574 return false;
4575
4576 switch (CC) {
4577 // (select_cc lhs, rhs, 0, \
4578 // (select_cc [lr]hs, [lr]hs, 1, -1, setlt/setgt), seteq)
4579 case ISD::SETEQ:
4580 if (!InnerIsSel)
4581 return false;
4582 if (InnerCC != ISD::SETLT && InnerCC != ISD::SETGT)
4583 return false;
4584 NeedSwapOps = (InnerCC == ISD::SETGT) ? InnerSwapped : !InnerSwapped;
4585 break;
4586
4587 // (select_cc lhs, rhs, -1, (zext (setcc [lr]hs, [lr]hs, setne)), setu?lt)
4588 // (select_cc lhs, rhs, -1, (zext (setcc lhs, rhs, setgt)), setu?lt)
4589 // (select_cc lhs, rhs, -1, (zext (setcc rhs, lhs, setlt)), setu?lt)
4590 // (select_cc lhs, rhs, 1, (sext (setcc [lr]hs, [lr]hs, setne)), setu?lt)
4591 // (select_cc lhs, rhs, 1, (sext (setcc lhs, rhs, setgt)), setu?lt)
4592 // (select_cc lhs, rhs, 1, (sext (setcc rhs, lhs, setlt)), setu?lt)
4593 case ISD::SETULT:
4594 if (!IsUnCmp && InnerCC != ISD::SETNE)
4595 return false;
4596 IsUnCmp = true;
4597 LLVM_FALLTHROUGH[[gnu::fallthrough]];
4598 case ISD::SETLT:
4599 if (InnerCC == ISD::SETNE || (InnerCC == ISD::SETGT && !InnerSwapped) ||
4600 (InnerCC == ISD::SETLT && InnerSwapped))
4601 NeedSwapOps = (TrueResVal == 1);
4602 else
4603 return false;
4604 break;
4605
4606 // (select_cc lhs, rhs, 1, (sext (setcc [lr]hs, [lr]hs, setne)), setu?gt)
4607 // (select_cc lhs, rhs, 1, (sext (setcc lhs, rhs, setlt)), setu?gt)
4608 // (select_cc lhs, rhs, 1, (sext (setcc rhs, lhs, setgt)), setu?gt)
4609 // (select_cc lhs, rhs, -1, (zext (setcc [lr]hs, [lr]hs, setne)), setu?gt)
4610 // (select_cc lhs, rhs, -1, (zext (setcc lhs, rhs, setlt)), setu?gt)
4611 // (select_cc lhs, rhs, -1, (zext (setcc rhs, lhs, setgt)), setu?gt)
4612 case ISD::SETUGT:
4613 if (!IsUnCmp && InnerCC != ISD::SETNE)
4614 return false;
4615 IsUnCmp = true;
4616 LLVM_FALLTHROUGH[[gnu::fallthrough]];
4617 case ISD::SETGT:
4618 if (InnerCC == ISD::SETNE || (InnerCC == ISD::SETLT && !InnerSwapped) ||
4619 (InnerCC == ISD::SETGT && InnerSwapped))
4620 NeedSwapOps = (TrueResVal == -1);
4621 else
4622 return false;
4623 break;
4624
4625 default:
4626 return false;
4627 }
4628
4629 LLVM_DEBUG(dbgs() << "Found a node that can be lowered to a SETB: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Found a node that can be lowered to a SETB: "
; } } while (false)
;
4630 LLVM_DEBUG(N->dump())do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { N->dump(); } } while (false)
;
4631
4632 return true;
4633}
4634
4635// Return true if it's a software square-root/divide operand.
4636static bool isSWTestOp(SDValue N) {
4637 if (N.getOpcode() == PPCISD::FTSQRT)
4638 return true;
4639 if (N.getNumOperands() < 1 || !isa<ConstantSDNode>(N.getOperand(0)))
4640 return false;
4641 switch (N.getConstantOperandVal(0)) {
4642 case Intrinsic::ppc_vsx_xvtdivdp:
4643 case Intrinsic::ppc_vsx_xvtdivsp:
4644 case Intrinsic::ppc_vsx_xvtsqrtdp:
4645 case Intrinsic::ppc_vsx_xvtsqrtsp:
4646 return true;
4647 }
4648 return false;
4649}
4650
4651bool PPCDAGToDAGISel::tryFoldSWTestBRCC(SDNode *N) {
4652 assert(N->getOpcode() == ISD::BR_CC && "ISD::BR_CC is expected.")(static_cast <bool> (N->getOpcode() == ISD::BR_CC &&
"ISD::BR_CC is expected.") ? void (0) : __assert_fail ("N->getOpcode() == ISD::BR_CC && \"ISD::BR_CC is expected.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4652, __extension__ __PRETTY_FUNCTION__))
;
4653 // We are looking for following patterns, where `truncate to i1` actually has
4654 // the same semantic with `and 1`.
4655 // (br_cc seteq, (truncateToi1 SWTestOp), 0) -> (BCC PRED_NU, SWTestOp)
4656 // (br_cc seteq, (and SWTestOp, 2), 0) -> (BCC PRED_NE, SWTestOp)
4657 // (br_cc seteq, (and SWTestOp, 4), 0) -> (BCC PRED_LE, SWTestOp)
4658 // (br_cc seteq, (and SWTestOp, 8), 0) -> (BCC PRED_GE, SWTestOp)
4659 // (br_cc setne, (truncateToi1 SWTestOp), 0) -> (BCC PRED_UN, SWTestOp)
4660 // (br_cc setne, (and SWTestOp, 2), 0) -> (BCC PRED_EQ, SWTestOp)
4661 // (br_cc setne, (and SWTestOp, 4), 0) -> (BCC PRED_GT, SWTestOp)
4662 // (br_cc setne, (and SWTestOp, 8), 0) -> (BCC PRED_LT, SWTestOp)
4663 ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(1))->get();
4664 if (CC != ISD::SETEQ && CC != ISD::SETNE)
4665 return false;
4666
4667 SDValue CmpRHS = N->getOperand(3);
4668 if (!isa<ConstantSDNode>(CmpRHS) ||
4669 cast<ConstantSDNode>(CmpRHS)->getSExtValue() != 0)
4670 return false;
4671
4672 SDValue CmpLHS = N->getOperand(2);
4673 if (CmpLHS.getNumOperands() < 1 || !isSWTestOp(CmpLHS.getOperand(0)))
4674 return false;
4675
4676 unsigned PCC = 0;
4677 bool IsCCNE = CC == ISD::SETNE;
4678 if (CmpLHS.getOpcode() == ISD::AND &&
4679 isa<ConstantSDNode>(CmpLHS.getOperand(1)))
4680 switch (CmpLHS.getConstantOperandVal(1)) {
4681 case 1:
4682 PCC = IsCCNE ? PPC::PRED_UN : PPC::PRED_NU;
4683 break;
4684 case 2:
4685 PCC = IsCCNE ? PPC::PRED_EQ : PPC::PRED_NE;
4686 break;
4687 case 4:
4688 PCC = IsCCNE ? PPC::PRED_GT : PPC::PRED_LE;
4689 break;
4690 case 8:
4691 PCC = IsCCNE ? PPC::PRED_LT : PPC::PRED_GE;
4692 break;
4693 default:
4694 return false;
4695 }
4696 else if (CmpLHS.getOpcode() == ISD::TRUNCATE &&
4697 CmpLHS.getValueType() == MVT::i1)
4698 PCC = IsCCNE ? PPC::PRED_UN : PPC::PRED_NU;
4699
4700 if (PCC) {
4701 SDLoc dl(N);
4702 SDValue Ops[] = {getI32Imm(PCC, dl), CmpLHS.getOperand(0), N->getOperand(4),
4703 N->getOperand(0)};
4704 CurDAG->SelectNodeTo(N, PPC::BCC, MVT::Other, Ops);
4705 return true;
4706 }
4707 return false;
4708}
4709
4710bool PPCDAGToDAGISel::tryAsSingleRLWINM(SDNode *N) {
4711 assert(N->getOpcode() == ISD::AND && "ISD::AND SDNode expected")(static_cast <bool> (N->getOpcode() == ISD::AND &&
"ISD::AND SDNode expected") ? void (0) : __assert_fail ("N->getOpcode() == ISD::AND && \"ISD::AND SDNode expected\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4711, __extension__ __PRETTY_FUNCTION__))
;
4712 unsigned Imm;
4713 if (!isInt32Immediate(N->getOperand(1), Imm))
4714 return false;
4715
4716 SDLoc dl(N);
4717 SDValue Val = N->getOperand(0);
4718 unsigned SH, MB, ME;
4719 // If this is an and of a value rotated between 0 and 31 bits and then and'd
4720 // with a mask, emit rlwinm
4721 if (isRotateAndMask(Val.getNode(), Imm, false, SH, MB, ME)) {
4722 Val = Val.getOperand(0);
4723 SDValue Ops[] = {Val, getI32Imm(SH, dl), getI32Imm(MB, dl),
4724 getI32Imm(ME, dl)};
4725 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
4726 return true;
4727 }
4728
4729 // If this is just a masked value where the input is not handled, and
4730 // is not a rotate-left (handled by a pattern in the .td file), emit rlwinm
4731 if (isRunOfOnes(Imm, MB, ME) && Val.getOpcode() != ISD::ROTL) {
4732 SDValue Ops[] = {Val, getI32Imm(0, dl), getI32Imm(MB, dl),
4733 getI32Imm(ME, dl)};
4734 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
4735 return true;
4736 }
4737
4738 // AND X, 0 -> 0, not "rlwinm 32".
4739 if (Imm == 0) {
4740 ReplaceUses(SDValue(N, 0), N->getOperand(1));
4741 return true;
4742 }
4743
4744 return false;
4745}
4746
4747bool PPCDAGToDAGISel::tryAsSingleRLWINM8(SDNode *N) {
4748 assert(N->getOpcode() == ISD::AND && "ISD::AND SDNode expected")(static_cast <bool> (N->getOpcode() == ISD::AND &&
"ISD::AND SDNode expected") ? void (0) : __assert_fail ("N->getOpcode() == ISD::AND && \"ISD::AND SDNode expected\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4748, __extension__ __PRETTY_FUNCTION__))
;
4749 uint64_t Imm64;
4750 if (!isInt64Immediate(N->getOperand(1).getNode(), Imm64))
4751 return false;
4752
4753 unsigned MB, ME;
4754 if (isRunOfOnes64(Imm64, MB, ME) && MB >= 32 && MB <= ME) {
4755 // MB ME
4756 // +----------------------+
4757 // |xxxxxxxxxxx00011111000|
4758 // +----------------------+
4759 // 0 32 64
4760 // We can only do it if the MB is larger than 32 and MB <= ME
4761 // as RLWINM will replace the contents of [0 - 32) with [32 - 64) even
4762 // we didn't rotate it.
4763 SDLoc dl(N);
4764 SDValue Ops[] = {N->getOperand(0), getI64Imm(0, dl), getI64Imm(MB - 32, dl),
4765 getI64Imm(ME - 32, dl)};
4766 CurDAG->SelectNodeTo(N, PPC::RLWINM8, MVT::i64, Ops);
4767 return true;
4768 }
4769
4770 return false;
4771}
4772
4773bool PPCDAGToDAGISel::tryAsPairOfRLDICL(SDNode *N) {
4774 assert(N->getOpcode() == ISD::AND && "ISD::AND SDNode expected")(static_cast <bool> (N->getOpcode() == ISD::AND &&
"ISD::AND SDNode expected") ? void (0) : __assert_fail ("N->getOpcode() == ISD::AND && \"ISD::AND SDNode expected\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4774, __extension__ __PRETTY_FUNCTION__))
;
4775 uint64_t Imm64;
4776 if (!isInt64Immediate(N->getOperand(1).getNode(), Imm64))
4777 return false;
4778
4779 // Do nothing if it is 16-bit imm as the pattern in the .td file handle
4780 // it well with "andi.".
4781 if (isUInt<16>(Imm64))
4782 return false;
4783
4784 SDLoc Loc(N);
4785 SDValue Val = N->getOperand(0);
4786
4787 // Optimized with two rldicl's as follows:
4788 // Add missing bits on left to the mask and check that the mask is a
4789 // wrapped run of ones, i.e.
4790 // Change pattern |0001111100000011111111|
4791 // to |1111111100000011111111|.
4792 unsigned NumOfLeadingZeros = countLeadingZeros(Imm64);
4793 if (NumOfLeadingZeros != 0)
4794 Imm64 |= maskLeadingOnes<uint64_t>(NumOfLeadingZeros);
4795
4796 unsigned MB, ME;
4797 if (!isRunOfOnes64(Imm64, MB, ME))
4798 return false;
4799
4800 // ME MB MB-ME+63
4801 // +----------------------+ +----------------------+
4802 // |1111111100000011111111| -> |0000001111111111111111|
4803 // +----------------------+ +----------------------+
4804 // 0 63 0 63
4805 // There are ME + 1 ones on the left and (MB - ME + 63) & 63 zeros in between.
4806 unsigned OnesOnLeft = ME + 1;
4807 unsigned ZerosInBetween = (MB - ME + 63) & 63;
4808 // Rotate left by OnesOnLeft (so leading ones are now trailing ones) and clear
4809 // on the left the bits that are already zeros in the mask.
4810 Val = SDValue(CurDAG->getMachineNode(PPC::RLDICL, Loc, MVT::i64, Val,
4811 getI64Imm(OnesOnLeft, Loc),
4812 getI64Imm(ZerosInBetween, Loc)),
4813 0);
4814 // MB-ME+63 ME MB
4815 // +----------------------+ +----------------------+
4816 // |0000001111111111111111| -> |0001111100000011111111|
4817 // +----------------------+ +----------------------+
4818 // 0 63 0 63
4819 // Rotate back by 64 - OnesOnLeft to undo previous rotate. Then clear on the
4820 // left the number of ones we previously added.
4821 SDValue Ops[] = {Val, getI64Imm(64 - OnesOnLeft, Loc),
4822 getI64Imm(NumOfLeadingZeros, Loc)};
4823 CurDAG->SelectNodeTo(N, PPC::RLDICL, MVT::i64, Ops);
4824 return true;
4825}
4826
4827bool PPCDAGToDAGISel::tryAsSingleRLWIMI(SDNode *N) {
4828 assert(N->getOpcode() == ISD::AND && "ISD::AND SDNode expected")(static_cast <bool> (N->getOpcode() == ISD::AND &&
"ISD::AND SDNode expected") ? void (0) : __assert_fail ("N->getOpcode() == ISD::AND && \"ISD::AND SDNode expected\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4828, __extension__ __PRETTY_FUNCTION__))
;
4829 unsigned Imm;
4830 if (!isInt32Immediate(N->getOperand(1), Imm))
4831 return false;
4832
4833 SDValue Val = N->getOperand(0);
4834 unsigned Imm2;
4835 // ISD::OR doesn't get all the bitfield insertion fun.
4836 // (and (or x, c1), c2) where isRunOfOnes(~(c1^c2)) might be a
4837 // bitfield insert.
4838 if (Val.getOpcode() != ISD::OR || !isInt32Immediate(Val.getOperand(1), Imm2))
4839 return false;
4840
4841 // The idea here is to check whether this is equivalent to:
4842 // (c1 & m) | (x & ~m)
4843 // where m is a run-of-ones mask. The logic here is that, for each bit in
4844 // c1 and c2:
4845 // - if both are 1, then the output will be 1.
4846 // - if both are 0, then the output will be 0.
4847 // - if the bit in c1 is 0, and the bit in c2 is 1, then the output will
4848 // come from x.
4849 // - if the bit in c1 is 1, and the bit in c2 is 0, then the output will
4850 // be 0.
4851 // If that last condition is never the case, then we can form m from the
4852 // bits that are the same between c1 and c2.
4853 unsigned MB, ME;
4854 if (isRunOfOnes(~(Imm ^ Imm2), MB, ME) && !(~Imm & Imm2)) {
4855 SDLoc dl(N);
4856 SDValue Ops[] = {Val.getOperand(0), Val.getOperand(1), getI32Imm(0, dl),
4857 getI32Imm(MB, dl), getI32Imm(ME, dl)};
4858 ReplaceNode(N, CurDAG->getMachineNode(PPC::RLWIMI, dl, MVT::i32, Ops));
4859 return true;
4860 }
4861
4862 return false;
4863}
4864
4865bool PPCDAGToDAGISel::tryAsSingleRLDICL(SDNode *N) {
4866 assert(N->getOpcode() == ISD::AND && "ISD::AND SDNode expected")(static_cast <bool> (N->getOpcode() == ISD::AND &&
"ISD::AND SDNode expected") ? void (0) : __assert_fail ("N->getOpcode() == ISD::AND && \"ISD::AND SDNode expected\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4866, __extension__ __PRETTY_FUNCTION__))
;
4867 uint64_t Imm64;
4868 if (!isInt64Immediate(N->getOperand(1).getNode(), Imm64) || !isMask_64(Imm64))
4869 return false;
4870
4871 // If this is a 64-bit zero-extension mask, emit rldicl.
4872 unsigned MB = 64 - countTrailingOnes(Imm64);
4873 unsigned SH = 0;
4874 unsigned Imm;
4875 SDValue Val = N->getOperand(0);
4876 SDLoc dl(N);
4877
4878 if (Val.getOpcode() == ISD::ANY_EXTEND) {
4879 auto Op0 = Val.getOperand(0);
4880 if (Op0.getOpcode() == ISD::SRL &&
4881 isInt32Immediate(Op0.getOperand(1).getNode(), Imm) && Imm <= MB) {
4882
4883 auto ResultType = Val.getNode()->getValueType(0);
4884 auto ImDef = CurDAG->getMachineNode(PPC::IMPLICIT_DEF, dl, ResultType);
4885 SDValue IDVal(ImDef, 0);
4886
4887 Val = SDValue(CurDAG->getMachineNode(PPC::INSERT_SUBREG, dl, ResultType,
4888 IDVal, Op0.getOperand(0),
4889 getI32Imm(1, dl)),
4890 0);
4891 SH = 64 - Imm;
4892 }
4893 }
4894
4895 // If the operand is a logical right shift, we can fold it into this
4896 // instruction: rldicl(rldicl(x, 64-n, n), 0, mb) -> rldicl(x, 64-n, mb)
4897 // for n <= mb. The right shift is really a left rotate followed by a
4898 // mask, and this mask is a more-restrictive sub-mask of the mask implied
4899 // by the shift.
4900 if (Val.getOpcode() == ISD::SRL &&
4901 isInt32Immediate(Val.getOperand(1).getNode(), Imm) && Imm <= MB) {
4902 assert(Imm < 64 && "Illegal shift amount")(static_cast <bool> (Imm < 64 && "Illegal shift amount"
) ? void (0) : __assert_fail ("Imm < 64 && \"Illegal shift amount\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4902, __extension__ __PRETTY_FUNCTION__))
;
4903 Val = Val.getOperand(0);
4904 SH = 64 - Imm;
4905 }
4906
4907 SDValue Ops[] = {Val, getI32Imm(SH, dl), getI32Imm(MB, dl)};
4908 CurDAG->SelectNodeTo(N, PPC::RLDICL, MVT::i64, Ops);
4909 return true;
4910}
4911
4912bool PPCDAGToDAGISel::tryAsSingleRLDICR(SDNode *N) {
4913 assert(N->getOpcode() == ISD::AND && "ISD::AND SDNode expected")(static_cast <bool> (N->getOpcode() == ISD::AND &&
"ISD::AND SDNode expected") ? void (0) : __assert_fail ("N->getOpcode() == ISD::AND && \"ISD::AND SDNode expected\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4913, __extension__ __PRETTY_FUNCTION__))
;
4914 uint64_t Imm64;
4915 if (!isInt64Immediate(N->getOperand(1).getNode(), Imm64) ||
4916 !isMask_64(~Imm64))
4917 return false;
4918
4919 // If this is a negated 64-bit zero-extension mask,
4920 // i.e. the immediate is a sequence of ones from most significant side
4921 // and all zero for reminder, we should use rldicr.
4922 unsigned MB = 63 - countTrailingOnes(~Imm64);
4923 unsigned SH = 0;
4924 SDLoc dl(N);
4925 SDValue Ops[] = {N->getOperand(0), getI32Imm(SH, dl), getI32Imm(MB, dl)};
4926 CurDAG->SelectNodeTo(N, PPC::RLDICR, MVT::i64, Ops);
4927 return true;
4928}
4929
4930bool PPCDAGToDAGISel::tryAsSingleRLDIMI(SDNode *N) {
4931 assert(N->getOpcode() == ISD::OR && "ISD::OR SDNode expected")(static_cast <bool> (N->getOpcode() == ISD::OR &&
"ISD::OR SDNode expected") ? void (0) : __assert_fail ("N->getOpcode() == ISD::OR && \"ISD::OR SDNode expected\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4931, __extension__ __PRETTY_FUNCTION__))
;
4932 uint64_t Imm64;
4933 unsigned MB, ME;
4934 SDValue N0 = N->getOperand(0);
4935
4936 // We won't get fewer instructions if the imm is 32-bit integer.
4937 // rldimi requires the imm to have consecutive ones with both sides zero.
4938 // Also, make sure the first Op has only one use, otherwise this may increase
4939 // register pressure since rldimi is destructive.
4940 if (!isInt64Immediate(N->getOperand(1).getNode(), Imm64) ||
4941 isUInt<32>(Imm64) || !isRunOfOnes64(Imm64, MB, ME) || !N0.hasOneUse())
4942 return false;
4943
4944 unsigned SH = 63 - ME;
4945 SDLoc Dl(N);
4946 // Use select64Imm for making LI instr instead of directly putting Imm64
4947 SDValue Ops[] = {
4948 N->getOperand(0),
4949 SDValue(selectI64Imm(CurDAG, getI64Imm(-1, Dl).getNode()), 0),
4950 getI32Imm(SH, Dl), getI32Imm(MB, Dl)};
4951 CurDAG->SelectNodeTo(N, PPC::RLDIMI, MVT::i64, Ops);
4952 return true;
4953}
4954
4955// Select - Convert the specified operand from a target-independent to a
4956// target-specific node if it hasn't already been changed.
4957void PPCDAGToDAGISel::Select(SDNode *N) {
4958 SDLoc dl(N);
4959 if (N->isMachineOpcode()) {
4960 N->setNodeId(-1);
4961 return; // Already selected.
4962 }
4963
4964 // In case any misguided DAG-level optimizations form an ADD with a
4965 // TargetConstant operand, crash here instead of miscompiling (by selecting
4966 // an r+r add instead of some kind of r+i add).
4967 if (N->getOpcode() == ISD::ADD &&
4968 N->getOperand(1).getOpcode() == ISD::TargetConstant)
4969 llvm_unreachable("Invalid ADD with TargetConstant operand")::llvm::llvm_unreachable_internal("Invalid ADD with TargetConstant operand"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 4969)
;
4970
4971 // Try matching complex bit permutations before doing anything else.
4972 if (tryBitPermutation(N))
4973 return;
4974
4975 // Try to emit integer compares as GPR-only sequences (i.e. no use of CR).
4976 if (tryIntCompareInGPR(N))
4977 return;
4978
4979 switch (N->getOpcode()) {
4980 default: break;
4981
4982 case ISD::Constant:
4983 if (N->getValueType(0) == MVT::i64) {
4984 ReplaceNode(N, selectI64Imm(CurDAG, N));
4985 return;
4986 }
4987 break;
4988
4989 case ISD::INTRINSIC_WO_CHAIN: {
4990 if (!Subtarget->isISA3_1())
4991 break;
4992 unsigned Opcode = 0;
4993 switch (N->getConstantOperandVal(0)) {
4994 default:
4995 break;
4996 case Intrinsic::ppc_altivec_vstribr_p:
4997 Opcode = PPC::VSTRIBR_rec;
4998 break;
4999 case Intrinsic::ppc_altivec_vstribl_p:
5000 Opcode = PPC::VSTRIBL_rec;
5001 break;
5002 case Intrinsic::ppc_altivec_vstrihr_p:
5003 Opcode = PPC::VSTRIHR_rec;
5004 break;
5005 case Intrinsic::ppc_altivec_vstrihl_p:
5006 Opcode = PPC::VSTRIHL_rec;
5007 break;
5008 }
5009 if (!Opcode)
5010 break;
5011
5012 // Generate the appropriate vector string isolate intrinsic to match.
5013 EVT VTs[] = {MVT::v16i8, MVT::Glue};
5014 SDValue VecStrOp =
5015 SDValue(CurDAG->getMachineNode(Opcode, dl, VTs, N->getOperand(2)), 0);
5016 // Vector string isolate instructions update the EQ bit of CR6.
5017 // Generate a SETBC instruction to extract the bit and place it in a GPR.
5018 SDValue SubRegIdx = CurDAG->getTargetConstant(PPC::sub_eq, dl, MVT::i32);
5019 SDValue CR6Reg = CurDAG->getRegister(PPC::CR6, MVT::i32);
5020 SDValue CRBit = SDValue(
5021 CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG, dl, MVT::i1,
5022 CR6Reg, SubRegIdx, VecStrOp.getValue(1)),
5023 0);
5024 CurDAG->SelectNodeTo(N, PPC::SETBC, MVT::i32, CRBit);
5025 return;
5026 }
5027
5028 case ISD::SETCC:
5029 case ISD::STRICT_FSETCC:
5030 case ISD::STRICT_FSETCCS:
5031 if (trySETCC(N))
5032 return;
5033 break;
5034 // These nodes will be transformed into GETtlsADDR32 node, which
5035 // later becomes BL_TLS __tls_get_addr(sym at tlsgd)@PLT
5036 case PPCISD::ADDI_TLSLD_L_ADDR:
5037 case PPCISD::ADDI_TLSGD_L_ADDR: {
5038 const Module *Mod = MF->getFunction().getParent();
5039 if (PPCLowering->getPointerTy(CurDAG->getDataLayout()) != MVT::i32 ||
5040 !Subtarget->isSecurePlt() || !Subtarget->isTargetELF() ||
5041 Mod->getPICLevel() == PICLevel::SmallPIC)
5042 break;
5043 // Attach global base pointer on GETtlsADDR32 node in order to
5044 // generate secure plt code for TLS symbols.
5045 getGlobalBaseReg();
5046 } break;
5047 case PPCISD::CALL: {
5048 if (PPCLowering->getPointerTy(CurDAG->getDataLayout()) != MVT::i32 ||
5049 !TM.isPositionIndependent() || !Subtarget->isSecurePlt() ||
5050 !Subtarget->isTargetELF())
5051 break;
5052
5053 SDValue Op = N->getOperand(1);
5054
5055 if (GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op)) {
5056 if (GA->getTargetFlags() == PPCII::MO_PLT)
5057 getGlobalBaseReg();
5058 }
5059 else if (ExternalSymbolSDNode *ES = dyn_cast<ExternalSymbolSDNode>(Op)) {
5060 if (ES->getTargetFlags() == PPCII::MO_PLT)
5061 getGlobalBaseReg();
5062 }
5063 }
5064 break;
5065
5066 case PPCISD::GlobalBaseReg:
5067 ReplaceNode(N, getGlobalBaseReg());
5068 return;
5069
5070 case ISD::FrameIndex:
5071 selectFrameIndex(N, N);
5072 return;
5073
5074 case PPCISD::MFOCRF: {
5075 SDValue InFlag = N->getOperand(1);
5076 ReplaceNode(N, CurDAG->getMachineNode(PPC::MFOCRF, dl, MVT::i32,
5077 N->getOperand(0), InFlag));
5078 return;
5079 }
5080
5081 case PPCISD::READ_TIME_BASE:
5082 ReplaceNode(N, CurDAG->getMachineNode(PPC::ReadTB, dl, MVT::i32, MVT::i32,
5083 MVT::Other, N->getOperand(0)));
5084 return;
5085
5086 case PPCISD::SRA_ADDZE: {
5087 SDValue N0 = N->getOperand(0);
5088 SDValue ShiftAmt =
5089 CurDAG->getTargetConstant(*cast<ConstantSDNode>(N->getOperand(1))->
5090 getConstantIntValue(), dl,
5091 N->getValueType(0));
5092 if (N->getValueType(0) == MVT::i64) {
5093 SDNode *Op =
5094 CurDAG->getMachineNode(PPC::SRADI, dl, MVT::i64, MVT::Glue,
5095 N0, ShiftAmt);
5096 CurDAG->SelectNodeTo(N, PPC::ADDZE8, MVT::i64, SDValue(Op, 0),
5097 SDValue(Op, 1));
5098 return;
5099 } else {
5100 assert(N->getValueType(0) == MVT::i32 &&(static_cast <bool> (N->getValueType(0) == MVT::i32 &&
"Expecting i64 or i32 in PPCISD::SRA_ADDZE") ? void (0) : __assert_fail
("N->getValueType(0) == MVT::i32 && \"Expecting i64 or i32 in PPCISD::SRA_ADDZE\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5101, __extension__ __PRETTY_FUNCTION__))
5101 "Expecting i64 or i32 in PPCISD::SRA_ADDZE")(static_cast <bool> (N->getValueType(0) == MVT::i32 &&
"Expecting i64 or i32 in PPCISD::SRA_ADDZE") ? void (0) : __assert_fail
("N->getValueType(0) == MVT::i32 && \"Expecting i64 or i32 in PPCISD::SRA_ADDZE\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5101, __extension__ __PRETTY_FUNCTION__))
;
5102 SDNode *Op =
5103 CurDAG->getMachineNode(PPC::SRAWI, dl, MVT::i32, MVT::Glue,
5104 N0, ShiftAmt);
5105 CurDAG->SelectNodeTo(N, PPC::ADDZE, MVT::i32, SDValue(Op, 0),
5106 SDValue(Op, 1));
5107 return;
5108 }
5109 }
5110
5111 case ISD::STORE: {
5112 // Change TLS initial-exec D-form stores to X-form stores.
5113 StoreSDNode *ST = cast<StoreSDNode>(N);
5114 if (EnableTLSOpt && Subtarget->isELFv2ABI() &&
5115 ST->getAddressingMode() != ISD::PRE_INC)
5116 if (tryTLSXFormStore(ST))
5117 return;
5118 break;
5119 }
5120 case ISD::LOAD: {
5121 // Handle preincrement loads.
5122 LoadSDNode *LD = cast<LoadSDNode>(N);
5123 EVT LoadedVT = LD->getMemoryVT();
5124
5125 // Normal loads are handled by code generated from the .td file.
5126 if (LD->getAddressingMode() != ISD::PRE_INC) {
5127 // Change TLS initial-exec D-form loads to X-form loads.
5128 if (EnableTLSOpt && Subtarget->isELFv2ABI())
5129 if (tryTLSXFormLoad(LD))
5130 return;
5131 break;
5132 }
5133
5134 SDValue Offset = LD->getOffset();
5135 if (Offset.getOpcode() == ISD::TargetConstant ||
5136 Offset.getOpcode() == ISD::TargetGlobalAddress) {
5137
5138 unsigned Opcode;
5139 bool isSExt = LD->getExtensionType() == ISD::SEXTLOAD;
5140 if (LD->getValueType(0) != MVT::i64) {
5141 // Handle PPC32 integer and normal FP loads.
5142 assert((!isSExt || LoadedVT == MVT::i16) && "Invalid sext update load")(static_cast <bool> ((!isSExt || LoadedVT == MVT::i16) &&
"Invalid sext update load") ? void (0) : __assert_fail ("(!isSExt || LoadedVT == MVT::i16) && \"Invalid sext update load\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5142, __extension__ __PRETTY_FUNCTION__))
;
5143 switch (LoadedVT.getSimpleVT().SimpleTy) {
5144 default: llvm_unreachable("Invalid PPC load type!")::llvm::llvm_unreachable_internal("Invalid PPC load type!", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5144)
;
5145 case MVT::f64: Opcode = PPC::LFDU; break;
5146 case MVT::f32: Opcode = PPC::LFSU; break;
5147 case MVT::i32: Opcode = PPC::LWZU; break;
5148 case MVT::i16: Opcode = isSExt ? PPC::LHAU : PPC::LHZU; break;
5149 case MVT::i1:
5150 case MVT::i8: Opcode = PPC::LBZU; break;
5151 }
5152 } else {
5153 assert(LD->getValueType(0) == MVT::i64 && "Unknown load result type!")(static_cast <bool> (LD->getValueType(0) == MVT::i64
&& "Unknown load result type!") ? void (0) : __assert_fail
("LD->getValueType(0) == MVT::i64 && \"Unknown load result type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5153, __extension__ __PRETTY_FUNCTION__))
;
5154 assert((!isSExt || LoadedVT == MVT::i16) && "Invalid sext update load")(static_cast <bool> ((!isSExt || LoadedVT == MVT::i16) &&
"Invalid sext update load") ? void (0) : __assert_fail ("(!isSExt || LoadedVT == MVT::i16) && \"Invalid sext update load\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5154, __extension__ __PRETTY_FUNCTION__))
;
5155 switch (LoadedVT.getSimpleVT().SimpleTy) {
5156 default: llvm_unreachable("Invalid PPC load type!")::llvm::llvm_unreachable_internal("Invalid PPC load type!", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5156)
;
5157 case MVT::i64: Opcode = PPC::LDU; break;
5158 case MVT::i32: Opcode = PPC::LWZU8; break;
5159 case MVT::i16: Opcode = isSExt ? PPC::LHAU8 : PPC::LHZU8; break;
5160 case MVT::i1:
5161 case MVT::i8: Opcode = PPC::LBZU8; break;
5162 }
5163 }
5164
5165 SDValue Chain = LD->getChain();
5166 SDValue Base = LD->getBasePtr();
5167 SDValue Ops[] = { Offset, Base, Chain };
5168 SDNode *MN = CurDAG->getMachineNode(
5169 Opcode, dl, LD->getValueType(0),
5170 PPCLowering->getPointerTy(CurDAG->getDataLayout()), MVT::Other, Ops);
5171 transferMemOperands(N, MN);
5172 ReplaceNode(N, MN);
5173 return;
5174 } else {
5175 unsigned Opcode;
5176 bool isSExt = LD->getExtensionType() == ISD::SEXTLOAD;
5177 if (LD->getValueType(0) != MVT::i64) {
5178 // Handle PPC32 integer and normal FP loads.
5179 assert((!isSExt || LoadedVT == MVT::i16) && "Invalid sext update load")(static_cast <bool> ((!isSExt || LoadedVT == MVT::i16) &&
"Invalid sext update load") ? void (0) : __assert_fail ("(!isSExt || LoadedVT == MVT::i16) && \"Invalid sext update load\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5179, __extension__ __PRETTY_FUNCTION__))
;
5180 switch (LoadedVT.getSimpleVT().SimpleTy) {
5181 default: llvm_unreachable("Invalid PPC load type!")::llvm::llvm_unreachable_internal("Invalid PPC load type!", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5181)
;
5182 case MVT::f64: Opcode = PPC::LFDUX; break;
5183 case MVT::f32: Opcode = PPC::LFSUX; break;
5184 case MVT::i32: Opcode = PPC::LWZUX; break;
5185 case MVT::i16: Opcode = isSExt ? PPC::LHAUX : PPC::LHZUX; break;
5186 case MVT::i1:
5187 case MVT::i8: Opcode = PPC::LBZUX; break;
5188 }
5189 } else {
5190 assert(LD->getValueType(0) == MVT::i64 && "Unknown load result type!")(static_cast <bool> (LD->getValueType(0) == MVT::i64
&& "Unknown load result type!") ? void (0) : __assert_fail
("LD->getValueType(0) == MVT::i64 && \"Unknown load result type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5190, __extension__ __PRETTY_FUNCTION__))
;
5191 assert((!isSExt || LoadedVT == MVT::i16 || LoadedVT == MVT::i32) &&(static_cast <bool> ((!isSExt || LoadedVT == MVT::i16 ||
LoadedVT == MVT::i32) && "Invalid sext update load")
? void (0) : __assert_fail ("(!isSExt || LoadedVT == MVT::i16 || LoadedVT == MVT::i32) && \"Invalid sext update load\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5192, __extension__ __PRETTY_FUNCTION__))
5192 "Invalid sext update load")(static_cast <bool> ((!isSExt || LoadedVT == MVT::i16 ||
LoadedVT == MVT::i32) && "Invalid sext update load")
? void (0) : __assert_fail ("(!isSExt || LoadedVT == MVT::i16 || LoadedVT == MVT::i32) && \"Invalid sext update load\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5192, __extension__ __PRETTY_FUNCTION__))
;
5193 switch (LoadedVT.getSimpleVT().SimpleTy) {
5194 default: llvm_unreachable("Invalid PPC load type!")::llvm::llvm_unreachable_internal("Invalid PPC load type!", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5194)
;
5195 case MVT::i64: Opcode = PPC::LDUX; break;
5196 case MVT::i32: Opcode = isSExt ? PPC::LWAUX : PPC::LWZUX8; break;
5197 case MVT::i16: Opcode = isSExt ? PPC::LHAUX8 : PPC::LHZUX8; break;
5198 case MVT::i1:
5199 case MVT::i8: Opcode = PPC::LBZUX8; break;
5200 }
5201 }
5202
5203 SDValue Chain = LD->getChain();
5204 SDValue Base = LD->getBasePtr();
5205 SDValue Ops[] = { Base, Offset, Chain };
5206 SDNode *MN = CurDAG->getMachineNode(
5207 Opcode, dl, LD->getValueType(0),
5208 PPCLowering->getPointerTy(CurDAG->getDataLayout()), MVT::Other, Ops);
5209 transferMemOperands(N, MN);
5210 ReplaceNode(N, MN);
5211 return;
5212 }
5213 }
5214
5215 case ISD::AND:
5216 // If this is an 'and' with a mask, try to emit rlwinm/rldicl/rldicr
5217 if (tryAsSingleRLWINM(N) || tryAsSingleRLWIMI(N) || tryAsSingleRLDICL(N) ||
5218 tryAsSingleRLDICR(N) || tryAsSingleRLWINM8(N) || tryAsPairOfRLDICL(N))
5219 return;
5220
5221 // Other cases are autogenerated.
5222 break;
5223 case ISD::OR: {
5224 if (N->getValueType(0) == MVT::i32)
5225 if (tryBitfieldInsert(N))
5226 return;
5227
5228 int16_t Imm;
5229 if (N->getOperand(0)->getOpcode() == ISD::FrameIndex &&
5230 isIntS16Immediate(N->getOperand(1), Imm)) {
5231 KnownBits LHSKnown = CurDAG->computeKnownBits(N->getOperand(0));
5232
5233 // If this is equivalent to an add, then we can fold it with the
5234 // FrameIndex calculation.
5235 if ((LHSKnown.Zero.getZExtValue()|~(uint64_t)Imm) == ~0ULL) {
5236 selectFrameIndex(N, N->getOperand(0).getNode(), (int)Imm);
5237 return;
5238 }
5239 }
5240
5241 // If this is 'or' against an imm with consecutive ones and both sides zero,
5242 // try to emit rldimi
5243 if (tryAsSingleRLDIMI(N))
5244 return;
5245
5246 // OR with a 32-bit immediate can be handled by ori + oris
5247 // without creating an immediate in a GPR.
5248 uint64_t Imm64 = 0;
5249 bool IsPPC64 = Subtarget->isPPC64();
5250 if (IsPPC64 && isInt64Immediate(N->getOperand(1), Imm64) &&
5251 (Imm64 & ~0xFFFFFFFFuLL) == 0) {
5252 // If ImmHi (ImmHi) is zero, only one ori (oris) is generated later.
5253 uint64_t ImmHi = Imm64 >> 16;
5254 uint64_t ImmLo = Imm64 & 0xFFFF;
5255 if (ImmHi != 0 && ImmLo != 0) {
5256 SDNode *Lo = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64,
5257 N->getOperand(0),
5258 getI16Imm(ImmLo, dl));
5259 SDValue Ops1[] = { SDValue(Lo, 0), getI16Imm(ImmHi, dl)};
5260 CurDAG->SelectNodeTo(N, PPC::ORIS8, MVT::i64, Ops1);
5261 return;
5262 }
5263 }
5264
5265 // Other cases are autogenerated.
5266 break;
5267 }
5268 case ISD::XOR: {
5269 // XOR with a 32-bit immediate can be handled by xori + xoris
5270 // without creating an immediate in a GPR.
5271 uint64_t Imm64 = 0;
5272 bool IsPPC64 = Subtarget->isPPC64();
5273 if (IsPPC64 && isInt64Immediate(N->getOperand(1), Imm64) &&
5274 (Imm64 & ~0xFFFFFFFFuLL) == 0) {
5275 // If ImmHi (ImmHi) is zero, only one xori (xoris) is generated later.
5276 uint64_t ImmHi = Imm64 >> 16;
5277 uint64_t ImmLo = Imm64 & 0xFFFF;
5278 if (ImmHi != 0 && ImmLo != 0) {
5279 SDNode *Lo = CurDAG->getMachineNode(PPC::XORI8, dl, MVT::i64,
5280 N->getOperand(0),
5281 getI16Imm(ImmLo, dl));
5282 SDValue Ops1[] = { SDValue(Lo, 0), getI16Imm(ImmHi, dl)};
5283 CurDAG->SelectNodeTo(N, PPC::XORIS8, MVT::i64, Ops1);
5284 return;
5285 }
5286 }
5287
5288 break;
5289 }
5290 case ISD::ADD: {
5291 int16_t Imm;
5292 if (N->getOperand(0)->getOpcode() == ISD::FrameIndex &&
5293 isIntS16Immediate(N->getOperand(1), Imm)) {
5294 selectFrameIndex(N, N->getOperand(0).getNode(), (int)Imm);
5295 return;
5296 }
5297
5298 break;
5299 }
5300 case ISD::SHL: {
5301 unsigned Imm, SH, MB, ME;
5302 if (isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::AND, Imm) &&
5303 isRotateAndMask(N, Imm, true, SH, MB, ME)) {
5304 SDValue Ops[] = { N->getOperand(0).getOperand(0),
5305 getI32Imm(SH, dl), getI32Imm(MB, dl),
5306 getI32Imm(ME, dl) };
5307 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
5308 return;
5309 }
5310
5311 // Other cases are autogenerated.
5312 break;
5313 }
5314 case ISD::SRL: {
5315 unsigned Imm, SH, MB, ME;
5316 if (isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::AND, Imm) &&
5317 isRotateAndMask(N, Imm, true, SH, MB, ME)) {
5318 SDValue Ops[] = { N->getOperand(0).getOperand(0),
5319 getI32Imm(SH, dl), getI32Imm(MB, dl),
5320 getI32Imm(ME, dl) };
5321 CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops);
5322 return;
5323 }
5324
5325 // Other cases are autogenerated.
5326 break;
5327 }
5328 case ISD::MUL: {
5329 SDValue Op1 = N->getOperand(1);
5330 if (Op1.getOpcode() != ISD::Constant || Op1.getValueType() != MVT::i64)
5331 break;
5332
5333 // If the multiplier fits int16, we can handle it with mulli.
5334 int64_t Imm = cast<ConstantSDNode>(Op1)->getZExtValue();
5335 unsigned Shift = countTrailingZeros<uint64_t>(Imm);
5336 if (isInt<16>(Imm) || !Shift)
5337 break;
5338
5339 // If the shifted value fits int16, we can do this transformation:
5340 // (mul X, c1 << c2) -> (rldicr (mulli X, c1) c2). We do this in ISEL due to
5341 // DAGCombiner prefers (shl (mul X, c1), c2) -> (mul X, c1 << c2).
5342 uint64_t ImmSh = Imm >> Shift;
5343 if (isInt<16>(ImmSh)) {
5344 uint64_t SextImm = SignExtend64(ImmSh & 0xFFFF, 16);
5345 SDValue SDImm = CurDAG->getTargetConstant(SextImm, dl, MVT::i64);
5346 SDNode *MulNode = CurDAG->getMachineNode(PPC::MULLI8, dl, MVT::i64,
5347 N->getOperand(0), SDImm);
5348 CurDAG->SelectNodeTo(N, PPC::RLDICR, MVT::i64, SDValue(MulNode, 0),
5349 getI32Imm(Shift, dl), getI32Imm(63 - Shift, dl));
5350 return;
5351 }
5352 break;
5353 }
5354 // FIXME: Remove this once the ANDI glue bug is fixed:
5355 case PPCISD::ANDI_rec_1_EQ_BIT:
5356 case PPCISD::ANDI_rec_1_GT_BIT: {
5357 if (!ANDIGlueBug)
5358 break;
5359
5360 EVT InVT = N->getOperand(0).getValueType();
5361 assert((InVT == MVT::i64 || InVT == MVT::i32) &&(static_cast <bool> ((InVT == MVT::i64 || InVT == MVT::
i32) && "Invalid input type for ANDI_rec_1_EQ_BIT") ?
void (0) : __assert_fail ("(InVT == MVT::i64 || InVT == MVT::i32) && \"Invalid input type for ANDI_rec_1_EQ_BIT\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5362, __extension__ __PRETTY_FUNCTION__))
5362 "Invalid input type for ANDI_rec_1_EQ_BIT")(static_cast <bool> ((InVT == MVT::i64 || InVT == MVT::
i32) && "Invalid input type for ANDI_rec_1_EQ_BIT") ?
void (0) : __assert_fail ("(InVT == MVT::i64 || InVT == MVT::i32) && \"Invalid input type for ANDI_rec_1_EQ_BIT\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5362, __extension__ __PRETTY_FUNCTION__))
;
5363
5364 unsigned Opcode = (InVT == MVT::i64) ? PPC::ANDI8_rec : PPC::ANDI_rec;
5365 SDValue AndI(CurDAG->getMachineNode(Opcode, dl, InVT, MVT::Glue,
5366 N->getOperand(0),
5367 CurDAG->getTargetConstant(1, dl, InVT)),
5368 0);
5369 SDValue CR0Reg = CurDAG->getRegister(PPC::CR0, MVT::i32);
5370 SDValue SRIdxVal = CurDAG->getTargetConstant(
5371 N->getOpcode() == PPCISD::ANDI_rec_1_EQ_BIT ? PPC::sub_eq : PPC::sub_gt,
5372 dl, MVT::i32);
5373
5374 CurDAG->SelectNodeTo(N, TargetOpcode::EXTRACT_SUBREG, MVT::i1, CR0Reg,
5375 SRIdxVal, SDValue(AndI.getNode(), 1) /* glue */);
5376 return;
5377 }
5378 case ISD::SELECT_CC: {
5379 ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(4))->get();
5380 EVT PtrVT =
5381 CurDAG->getTargetLoweringInfo().getPointerTy(CurDAG->getDataLayout());
5382 bool isPPC64 = (PtrVT == MVT::i64);
5383
5384 // If this is a select of i1 operands, we'll pattern match it.
5385 if (Subtarget->useCRBits() && N->getOperand(0).getValueType() == MVT::i1)
5386 break;
5387
5388 if (Subtarget->isISA3_0() && Subtarget->isPPC64()) {
5389 bool NeedSwapOps = false;
5390 bool IsUnCmp = false;
5391 if (mayUseP9Setb(N, CC, CurDAG, NeedSwapOps, IsUnCmp)) {
5392 SDValue LHS = N->getOperand(0);
5393 SDValue RHS = N->getOperand(1);
5394 if (NeedSwapOps)
5395 std::swap(LHS, RHS);
5396
5397 // Make use of SelectCC to generate the comparison to set CR bits, for
5398 // equality comparisons having one literal operand, SelectCC probably
5399 // doesn't need to materialize the whole literal and just use xoris to
5400 // check it first, it leads the following comparison result can't
5401 // exactly represent GT/LT relationship. So to avoid this we specify
5402 // SETGT/SETUGT here instead of SETEQ.
5403 SDValue GenCC =
5404 SelectCC(LHS, RHS, IsUnCmp ? ISD::SETUGT : ISD::SETGT, dl);
5405 CurDAG->SelectNodeTo(
5406 N, N->getSimpleValueType(0) == MVT::i64 ? PPC::SETB8 : PPC::SETB,
5407 N->getValueType(0), GenCC);
5408 NumP9Setb++;
5409 return;
5410 }
5411 }
5412
5413 // Handle the setcc cases here. select_cc lhs, 0, 1, 0, cc
5414 if (!isPPC64)
5415 if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N->getOperand(1)))
5416 if (ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N->getOperand(2)))
5417 if (ConstantSDNode *N3C = dyn_cast<ConstantSDNode>(N->getOperand(3)))
5418 if (N1C->isNullValue() && N3C->isNullValue() &&
5419 N2C->getZExtValue() == 1ULL && CC == ISD::SETNE &&
5420 // FIXME: Implement this optzn for PPC64.
5421 N->getValueType(0) == MVT::i32) {
5422 SDNode *Tmp =
5423 CurDAG->getMachineNode(PPC::ADDIC, dl, MVT::i32, MVT::Glue,
5424 N->getOperand(0), getI32Imm(~0U, dl));
5425 CurDAG->SelectNodeTo(N, PPC::SUBFE, MVT::i32, SDValue(Tmp, 0),
5426 N->getOperand(0), SDValue(Tmp, 1));
5427 return;
5428 }
5429
5430 SDValue CCReg = SelectCC(N->getOperand(0), N->getOperand(1), CC, dl);
5431
5432 if (N->getValueType(0) == MVT::i1) {
5433 // An i1 select is: (c & t) | (!c & f).
5434 bool Inv;
5435 unsigned Idx = getCRIdxForSetCC(CC, Inv);
5436
5437 unsigned SRI;
5438 switch (Idx) {
5439 default: llvm_unreachable("Invalid CC index")::llvm::llvm_unreachable_internal("Invalid CC index", "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5439)
;
5440 case 0: SRI = PPC::sub_lt; break;
5441 case 1: SRI = PPC::sub_gt; break;
5442 case 2: SRI = PPC::sub_eq; break;
5443 case 3: SRI = PPC::sub_un; break;
5444 }
5445
5446 SDValue CCBit = CurDAG->getTargetExtractSubreg(SRI, dl, MVT::i1, CCReg);
5447
5448 SDValue NotCCBit(CurDAG->getMachineNode(PPC::CRNOR, dl, MVT::i1,
5449 CCBit, CCBit), 0);
5450 SDValue C = Inv ? NotCCBit : CCBit,
5451 NotC = Inv ? CCBit : NotCCBit;
5452
5453 SDValue CAndT(CurDAG->getMachineNode(PPC::CRAND, dl, MVT::i1,
5454 C, N->getOperand(2)), 0);
5455 SDValue NotCAndF(CurDAG->getMachineNode(PPC::CRAND, dl, MVT::i1,
5456 NotC, N->getOperand(3)), 0);
5457
5458 CurDAG->SelectNodeTo(N, PPC::CROR, MVT::i1, CAndT, NotCAndF);
5459 return;
5460 }
5461
5462 unsigned BROpc =
5463 getPredicateForSetCC(CC, N->getOperand(0).getValueType(), Subtarget);
5464
5465 unsigned SelectCCOp;
5466 if (N->getValueType(0) == MVT::i32)
5467 SelectCCOp = PPC::SELECT_CC_I4;
5468 else if (N->getValueType(0) == MVT::i64)
5469 SelectCCOp = PPC::SELECT_CC_I8;
5470 else if (N->getValueType(0) == MVT::f32) {
5471 if (Subtarget->hasP8Vector())
5472 SelectCCOp = PPC::SELECT_CC_VSSRC;
5473 else if (Subtarget->hasSPE())
5474 SelectCCOp = PPC::SELECT_CC_SPE4;
5475 else
5476 SelectCCOp = PPC::SELECT_CC_F4;
5477 } else if (N->getValueType(0) == MVT::f64) {
5478 if (Subtarget->hasVSX())
5479 SelectCCOp = PPC::SELECT_CC_VSFRC;
5480 else if (Subtarget->hasSPE())
5481 SelectCCOp = PPC::SELECT_CC_SPE;
5482 else
5483 SelectCCOp = PPC::SELECT_CC_F8;
5484 } else if (N->getValueType(0) == MVT::f128)
5485 SelectCCOp = PPC::SELECT_CC_F16;
5486 else if (Subtarget->hasSPE())
5487 SelectCCOp = PPC::SELECT_CC_SPE;
5488 else if (N->getValueType(0) == MVT::v2f64 ||
5489 N->getValueType(0) == MVT::v2i64)
5490 SelectCCOp = PPC::SELECT_CC_VSRC;
5491 else
5492 SelectCCOp = PPC::SELECT_CC_VRRC;
5493
5494 SDValue Ops[] = { CCReg, N->getOperand(2), N->getOperand(3),
5495 getI32Imm(BROpc, dl) };
5496 CurDAG->SelectNodeTo(N, SelectCCOp, N->getValueType(0), Ops);
5497 return;
5498 }
5499 case ISD::VECTOR_SHUFFLE:
5500 if (Subtarget->hasVSX() && (N->getValueType(0) == MVT::v2f64 ||
5501 N->getValueType(0) == MVT::v2i64)) {
5502 ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(N);
5503
5504 SDValue Op1 = N->getOperand(SVN->getMaskElt(0) < 2 ? 0 : 1),
5505 Op2 = N->getOperand(SVN->getMaskElt(1) < 2 ? 0 : 1);
5506 unsigned DM[2];
5507
5508 for (int i = 0; i < 2; ++i)
5509 if (SVN->getMaskElt(i) <= 0 || SVN->getMaskElt(i) == 2)
5510 DM[i] = 0;
5511 else
5512 DM[i] = 1;
5513
5514 if (Op1 == Op2 && DM[0] == 0 && DM[1] == 0 &&
5515 Op1.getOpcode() == ISD::SCALAR_TO_VECTOR &&
5516 isa<LoadSDNode>(Op1.getOperand(0))) {
5517 LoadSDNode *LD = cast<LoadSDNode>(Op1.getOperand(0));
5518 SDValue Base, Offset;
5519
5520 if (LD->isUnindexed() && LD->hasOneUse() && Op1.hasOneUse() &&
5521 (LD->getMemoryVT() == MVT::f64 ||
5522 LD->getMemoryVT() == MVT::i64) &&
5523 SelectAddrIdxOnly(LD->getBasePtr(), Base, Offset)) {
5524 SDValue Chain = LD->getChain();
5525 SDValue Ops[] = { Base, Offset, Chain };
5526 MachineMemOperand *MemOp = LD->getMemOperand();
5527 SDNode *NewN = CurDAG->SelectNodeTo(N, PPC::LXVDSX,
5528 N->getValueType(0), Ops);
5529 CurDAG->setNodeMemRefs(cast<MachineSDNode>(NewN), {MemOp});
5530 return;
5531 }
5532 }
5533
5534 // For little endian, we must swap the input operands and adjust
5535 // the mask elements (reverse and invert them).
5536 if (Subtarget->isLittleEndian()) {
5537 std::swap(Op1, Op2);
5538 unsigned tmp = DM[0];
5539 DM[0] = 1 - DM[1];
5540 DM[1] = 1 - tmp;
5541 }
5542
5543 SDValue DMV = CurDAG->getTargetConstant(DM[1] | (DM[0] << 1), dl,
5544 MVT::i32);
5545 SDValue Ops[] = { Op1, Op2, DMV };
5546 CurDAG->SelectNodeTo(N, PPC::XXPERMDI, N->getValueType(0), Ops);
5547 return;
5548 }
5549
5550 break;
5551 case PPCISD::BDNZ:
5552 case PPCISD::BDZ: {
5553 bool IsPPC64 = Subtarget->isPPC64();
5554 SDValue Ops[] = { N->getOperand(1), N->getOperand(0) };
5555 CurDAG->SelectNodeTo(N, N->getOpcode() == PPCISD::BDNZ
5556 ? (IsPPC64 ? PPC::BDNZ8 : PPC::BDNZ)
5557 : (IsPPC64 ? PPC::BDZ8 : PPC::BDZ),
5558 MVT::Other, Ops);
5559 return;
5560 }
5561 case PPCISD::COND_BRANCH: {
5562 // Op #0 is the Chain.
5563 // Op #1 is the PPC::PRED_* number.
5564 // Op #2 is the CR#
5565 // Op #3 is the Dest MBB
5566 // Op #4 is the Flag.
5567 // Prevent PPC::PRED_* from being selected into LI.
5568 unsigned PCC = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
5569 if (EnableBranchHint)
5570 PCC |= getBranchHint(PCC, *FuncInfo, N->getOperand(3));
5571
5572 SDValue Pred = getI32Imm(PCC, dl);
5573 SDValue Ops[] = { Pred, N->getOperand(2), N->getOperand(3),
5574 N->getOperand(0), N->getOperand(4) };
5575 CurDAG->SelectNodeTo(N, PPC::BCC, MVT::Other, Ops);
5576 return;
5577 }
5578 case ISD::BR_CC: {
5579 if (tryFoldSWTestBRCC(N))
5580 return;
5581 ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(1))->get();
5582 unsigned PCC =
5583 getPredicateForSetCC(CC, N->getOperand(2).getValueType(), Subtarget);
5584
5585 if (N->getOperand(2).getValueType() == MVT::i1) {
5586 unsigned Opc;
5587 bool Swap;
5588 switch (PCC) {
5589 default: llvm_unreachable("Unexpected Boolean-operand predicate")::llvm::llvm_unreachable_internal("Unexpected Boolean-operand predicate"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5589)
;
5590 case PPC::PRED_LT: Opc = PPC::CRANDC; Swap = true; break;
5591 case PPC::PRED_LE: Opc = PPC::CRORC; Swap = true; break;
5592 case PPC::PRED_EQ: Opc = PPC::CREQV; Swap = false; break;
5593 case PPC::PRED_GE: Opc = PPC::CRORC; Swap = false; break;
5594 case PPC::PRED_GT: Opc = PPC::CRANDC; Swap = false; break;
5595 case PPC::PRED_NE: Opc = PPC::CRXOR; Swap = false; break;
5596 }
5597
5598 // A signed comparison of i1 values produces the opposite result to an
5599 // unsigned one if the condition code includes less-than or greater-than.
5600 // This is because 1 is the most negative signed i1 number and the most
5601 // positive unsigned i1 number. The CR-logical operations used for such
5602 // comparisons are non-commutative so for signed comparisons vs. unsigned
5603 // ones, the input operands just need to be swapped.
5604 if (ISD::isSignedIntSetCC(CC))
5605 Swap = !Swap;
5606
5607 SDValue BitComp(CurDAG->getMachineNode(Opc, dl, MVT::i1,
5608 N->getOperand(Swap ? 3 : 2),
5609 N->getOperand(Swap ? 2 : 3)), 0);
5610 CurDAG->SelectNodeTo(N, PPC::BC, MVT::Other, BitComp, N->getOperand(4),
5611 N->getOperand(0));
5612 return;
5613 }
5614
5615 if (EnableBranchHint)
5616 PCC |= getBranchHint(PCC, *FuncInfo, N->getOperand(4));
5617
5618 SDValue CondCode = SelectCC(N->getOperand(2), N->getOperand(3), CC, dl);
5619 SDValue Ops[] = { getI32Imm(PCC, dl), CondCode,
5620 N->getOperand(4), N->getOperand(0) };
5621 CurDAG->SelectNodeTo(N, PPC::BCC, MVT::Other, Ops);
5622 return;
5623 }
5624 case ISD::BRIND: {
5625 // FIXME: Should custom lower this.
5626 SDValue Chain = N->getOperand(0);
5627 SDValue Target = N->getOperand(1);
5628 unsigned Opc = Target.getValueType() == MVT::i32 ? PPC::MTCTR : PPC::MTCTR8;
5629 unsigned Reg = Target.getValueType() == MVT::i32 ? PPC::BCTR : PPC::BCTR8;
5630 Chain = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Glue, Target,
5631 Chain), 0);
5632 CurDAG->SelectNodeTo(N, Reg, MVT::Other, Chain);
5633 return;
5634 }
5635 case PPCISD::TOC_ENTRY: {
5636 const bool isPPC64 = Subtarget->isPPC64();
5637 const bool isELFABI = Subtarget->isSVR4ABI();
5638 const bool isAIXABI = Subtarget->isAIXABI();
5639
5640 // PowerPC only support small, medium and large code model.
5641 const CodeModel::Model CModel = TM.getCodeModel();
5642 assert(!(CModel == CodeModel::Tiny || CModel == CodeModel::Kernel) &&(static_cast <bool> (!(CModel == CodeModel::Tiny || CModel
== CodeModel::Kernel) && "PowerPC doesn't support tiny or kernel code models."
) ? void (0) : __assert_fail ("!(CModel == CodeModel::Tiny || CModel == CodeModel::Kernel) && \"PowerPC doesn't support tiny or kernel code models.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5643, __extension__ __PRETTY_FUNCTION__))
5643 "PowerPC doesn't support tiny or kernel code models.")(static_cast <bool> (!(CModel == CodeModel::Tiny || CModel
== CodeModel::Kernel) && "PowerPC doesn't support tiny or kernel code models."
) ? void (0) : __assert_fail ("!(CModel == CodeModel::Tiny || CModel == CodeModel::Kernel) && \"PowerPC doesn't support tiny or kernel code models.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5643, __extension__ __PRETTY_FUNCTION__))
;
5644
5645 if (isAIXABI && CModel == CodeModel::Medium)
5646 report_fatal_error("Medium code model is not supported on AIX.");
5647
5648 // For 64-bit small code model, we allow SelectCodeCommon to handle this,
5649 // selecting one of LDtoc, LDtocJTI, LDtocCPT, and LDtocBA.
5650 if (isPPC64 && CModel == CodeModel::Small)
5651 break;
5652
5653 // Handle 32-bit small code model.
5654 if (!isPPC64) {
5655 // Transforms the ISD::TOC_ENTRY node to passed in Opcode, either
5656 // PPC::ADDItoc, or PPC::LWZtoc
5657 auto replaceWith = [this, &dl](unsigned OpCode, SDNode *TocEntry) {
5658 SDValue GA = TocEntry->getOperand(0);
5659 SDValue TocBase = TocEntry->getOperand(1);
5660 SDNode *MN = CurDAG->getMachineNode(OpCode, dl, MVT::i32, GA, TocBase);
5661 transferMemOperands(TocEntry, MN);
5662 ReplaceNode(TocEntry, MN);
5663 };
5664
5665 if (isELFABI) {
5666 assert(TM.isPositionIndependent() &&(static_cast <bool> (TM.isPositionIndependent() &&
"32-bit ELF can only have TOC entries in position independent"
" code.") ? void (0) : __assert_fail ("TM.isPositionIndependent() && \"32-bit ELF can only have TOC entries in position independent\" \" code.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5668, __extension__ __PRETTY_FUNCTION__))
5667 "32-bit ELF can only have TOC entries in position independent"(static_cast <bool> (TM.isPositionIndependent() &&
"32-bit ELF can only have TOC entries in position independent"
" code.") ? void (0) : __assert_fail ("TM.isPositionIndependent() && \"32-bit ELF can only have TOC entries in position independent\" \" code.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5668, __extension__ __PRETTY_FUNCTION__))
5668 " code.")(static_cast <bool> (TM.isPositionIndependent() &&
"32-bit ELF can only have TOC entries in position independent"
" code.") ? void (0) : __assert_fail ("TM.isPositionIndependent() && \"32-bit ELF can only have TOC entries in position independent\" \" code.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5668, __extension__ __PRETTY_FUNCTION__))
;
5669 // 32-bit ELF always uses a small code model toc access.
5670 replaceWith(PPC::LWZtoc, N);
5671 return;
5672 }
5673
5674 if (isAIXABI && CModel == CodeModel::Small) {
5675 if (hasTocDataAttr(N->getOperand(0),
5676 CurDAG->getDataLayout().getPointerSize()))
5677 replaceWith(PPC::ADDItoc, N);
5678 else
5679 replaceWith(PPC::LWZtoc, N);
5680
5681 return;
5682 }
5683 }
5684
5685 assert(CModel != CodeModel::Small && "All small code models handled.")(static_cast <bool> (CModel != CodeModel::Small &&
"All small code models handled.") ? void (0) : __assert_fail
("CModel != CodeModel::Small && \"All small code models handled.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5685, __extension__ __PRETTY_FUNCTION__))
;
5686
5687 assert((isPPC64 || (isAIXABI && !isPPC64)) && "We are dealing with 64-bit"(static_cast <bool> ((isPPC64 || (isAIXABI && !
isPPC64)) && "We are dealing with 64-bit" " ELF/AIX or 32-bit AIX in the following."
) ? void (0) : __assert_fail ("(isPPC64 || (isAIXABI && !isPPC64)) && \"We are dealing with 64-bit\" \" ELF/AIX or 32-bit AIX in the following.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5688, __extension__ __PRETTY_FUNCTION__))
5688 " ELF/AIX or 32-bit AIX in the following.")(static_cast <bool> ((isPPC64 || (isAIXABI && !
isPPC64)) && "We are dealing with 64-bit" " ELF/AIX or 32-bit AIX in the following."
) ? void (0) : __assert_fail ("(isPPC64 || (isAIXABI && !isPPC64)) && \"We are dealing with 64-bit\" \" ELF/AIX or 32-bit AIX in the following.\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5688, __extension__ __PRETTY_FUNCTION__))
;
5689
5690 // Transforms the ISD::TOC_ENTRY node for 32-bit AIX large code model mode
5691 // or 64-bit medium (ELF-only) or large (ELF and AIX) code model code. We
5692 // generate two instructions as described below. The first source operand
5693 // is a symbol reference. If it must be toc-referenced according to
5694 // Subtarget, we generate:
5695 // [32-bit AIX]
5696 // LWZtocL(@sym, ADDIStocHA(%r2, @sym))
5697 // [64-bit ELF/AIX]
5698 // LDtocL(@sym, ADDIStocHA8(%x2, @sym))
5699 // Otherwise we generate:
5700 // ADDItocL(ADDIStocHA8(%x2, @sym), @sym)
5701 SDValue GA = N->getOperand(0);
5702 SDValue TOCbase = N->getOperand(1);
5703
5704 EVT VT = isPPC64 ? MVT::i64 : MVT::i32;
5705 SDNode *Tmp = CurDAG->getMachineNode(
5706 isPPC64 ? PPC::ADDIStocHA8 : PPC::ADDIStocHA, dl, VT, TOCbase, GA);
5707
5708 if (PPCLowering->isAccessedAsGotIndirect(GA)) {
5709 // If it is accessed as got-indirect, we need an extra LWZ/LD to load
5710 // the address.
5711 SDNode *MN = CurDAG->getMachineNode(
5712 isPPC64 ? PPC::LDtocL : PPC::LWZtocL, dl, VT, GA, SDValue(Tmp, 0));
5713
5714 transferMemOperands(N, MN);
5715 ReplaceNode(N, MN);
5716 return;
5717 }
5718
5719 // Build the address relative to the TOC-pointer.
5720 ReplaceNode(N, CurDAG->getMachineNode(PPC::ADDItocL, dl, MVT::i64,
5721 SDValue(Tmp, 0), GA));
5722 return;
5723 }
5724 case PPCISD::PPC32_PICGOT:
5725 // Generate a PIC-safe GOT reference.
5726 assert(Subtarget->is32BitELFABI() &&(static_cast <bool> (Subtarget->is32BitELFABI() &&
"PPCISD::PPC32_PICGOT is only supported for 32-bit SVR4") ? void
(0) : __assert_fail ("Subtarget->is32BitELFABI() && \"PPCISD::PPC32_PICGOT is only supported for 32-bit SVR4\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5727, __extension__ __PRETTY_FUNCTION__))
5727 "PPCISD::PPC32_PICGOT is only supported for 32-bit SVR4")(static_cast <bool> (Subtarget->is32BitELFABI() &&
"PPCISD::PPC32_PICGOT is only supported for 32-bit SVR4") ? void
(0) : __assert_fail ("Subtarget->is32BitELFABI() && \"PPCISD::PPC32_PICGOT is only supported for 32-bit SVR4\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5727, __extension__ __PRETTY_FUNCTION__))
;
5728 CurDAG->SelectNodeTo(N, PPC::PPC32PICGOT,
5729 PPCLowering->getPointerTy(CurDAG->getDataLayout()),
5730 MVT::i32);
5731 return;
5732
5733 case PPCISD::VADD_SPLAT: {
5734 // This expands into one of three sequences, depending on whether
5735 // the first operand is odd or even, positive or negative.
5736 assert(isa<ConstantSDNode>(N->getOperand(0)) &&(static_cast <bool> (isa<ConstantSDNode>(N->getOperand
(0)) && isa<ConstantSDNode>(N->getOperand(1)
) && "Invalid operand on VADD_SPLAT!") ? void (0) : __assert_fail
("isa<ConstantSDNode>(N->getOperand(0)) && isa<ConstantSDNode>(N->getOperand(1)) && \"Invalid operand on VADD_SPLAT!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5738, __extension__ __PRETTY_FUNCTION__))
5737 isa<ConstantSDNode>(N->getOperand(1)) &&(static_cast <bool> (isa<ConstantSDNode>(N->getOperand
(0)) && isa<ConstantSDNode>(N->getOperand(1)
) && "Invalid operand on VADD_SPLAT!") ? void (0) : __assert_fail
("isa<ConstantSDNode>(N->getOperand(0)) && isa<ConstantSDNode>(N->getOperand(1)) && \"Invalid operand on VADD_SPLAT!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5738, __extension__ __PRETTY_FUNCTION__))
5738 "Invalid operand on VADD_SPLAT!")(static_cast <bool> (isa<ConstantSDNode>(N->getOperand
(0)) && isa<ConstantSDNode>(N->getOperand(1)
) && "Invalid operand on VADD_SPLAT!") ? void (0) : __assert_fail
("isa<ConstantSDNode>(N->getOperand(0)) && isa<ConstantSDNode>(N->getOperand(1)) && \"Invalid operand on VADD_SPLAT!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5738, __extension__ __PRETTY_FUNCTION__))
;
5739
5740 int Elt = N->getConstantOperandVal(0);
5741 int EltSize = N->getConstantOperandVal(1);
5742 unsigned Opc1, Opc2, Opc3;
5743 EVT VT;
5744
5745 if (EltSize == 1) {
5746 Opc1 = PPC::VSPLTISB;
5747 Opc2 = PPC::VADDUBM;
5748 Opc3 = PPC::VSUBUBM;
5749 VT = MVT::v16i8;
5750 } else if (EltSize == 2) {
5751 Opc1 = PPC::VSPLTISH;
5752 Opc2 = PPC::VADDUHM;
5753 Opc3 = PPC::VSUBUHM;
5754 VT = MVT::v8i16;
5755 } else {
5756 assert(EltSize == 4 && "Invalid element size on VADD_SPLAT!")(static_cast <bool> (EltSize == 4 && "Invalid element size on VADD_SPLAT!"
) ? void (0) : __assert_fail ("EltSize == 4 && \"Invalid element size on VADD_SPLAT!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5756, __extension__ __PRETTY_FUNCTION__))
;
5757 Opc1 = PPC::VSPLTISW;
5758 Opc2 = PPC::VADDUWM;
5759 Opc3 = PPC::VSUBUWM;
5760 VT = MVT::v4i32;
5761 }
5762
5763 if ((Elt & 1) == 0) {
5764 // Elt is even, in the range [-32,-18] + [16,30].
5765 //
5766 // Convert: VADD_SPLAT elt, size
5767 // Into: tmp = VSPLTIS[BHW] elt
5768 // VADDU[BHW]M tmp, tmp
5769 // Where: [BHW] = B for size = 1, H for size = 2, W for size = 4
5770 SDValue EltVal = getI32Imm(Elt >> 1, dl);
5771 SDNode *Tmp = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
5772 SDValue TmpVal = SDValue(Tmp, 0);
5773 ReplaceNode(N, CurDAG->getMachineNode(Opc2, dl, VT, TmpVal, TmpVal));
5774 return;
5775 } else if (Elt > 0) {
5776 // Elt is odd and positive, in the range [17,31].
5777 //
5778 // Convert: VADD_SPLAT elt, size
5779 // Into: tmp1 = VSPLTIS[BHW] elt-16
5780 // tmp2 = VSPLTIS[BHW] -16
5781 // VSUBU[BHW]M tmp1, tmp2
5782 SDValue EltVal = getI32Imm(Elt - 16, dl);
5783 SDNode *Tmp1 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
5784 EltVal = getI32Imm(-16, dl);
5785 SDNode *Tmp2 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
5786 ReplaceNode(N, CurDAG->getMachineNode(Opc3, dl, VT, SDValue(Tmp1, 0),
5787 SDValue(Tmp2, 0)));
5788 return;
5789 } else {
5790 // Elt is odd and negative, in the range [-31,-17].
5791 //
5792 // Convert: VADD_SPLAT elt, size
5793 // Into: tmp1 = VSPLTIS[BHW] elt+16
5794 // tmp2 = VSPLTIS[BHW] -16
5795 // VADDU[BHW]M tmp1, tmp2
5796 SDValue EltVal = getI32Imm(Elt + 16, dl);
5797 SDNode *Tmp1 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
5798 EltVal = getI32Imm(-16, dl);
5799 SDNode *Tmp2 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
5800 ReplaceNode(N, CurDAG->getMachineNode(Opc2, dl, VT, SDValue(Tmp1, 0),
5801 SDValue(Tmp2, 0)));
5802 return;
5803 }
5804 }
5805 }
5806
5807 SelectCode(N);
5808}
5809
5810// If the target supports the cmpb instruction, do the idiom recognition here.
5811// We don't do this as a DAG combine because we don't want to do it as nodes
5812// are being combined (because we might miss part of the eventual idiom). We
5813// don't want to do it during instruction selection because we want to reuse
5814// the logic for lowering the masking operations already part of the
5815// instruction selector.
5816SDValue PPCDAGToDAGISel::combineToCMPB(SDNode *N) {
5817 SDLoc dl(N);
5818
5819 assert(N->getOpcode() == ISD::OR &&(static_cast <bool> (N->getOpcode() == ISD::OR &&
"Only OR nodes are supported for CMPB") ? void (0) : __assert_fail
("N->getOpcode() == ISD::OR && \"Only OR nodes are supported for CMPB\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5820, __extension__ __PRETTY_FUNCTION__))
5820 "Only OR nodes are supported for CMPB")(static_cast <bool> (N->getOpcode() == ISD::OR &&
"Only OR nodes are supported for CMPB") ? void (0) : __assert_fail
("N->getOpcode() == ISD::OR && \"Only OR nodes are supported for CMPB\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 5820, __extension__ __PRETTY_FUNCTION__))
;
5821
5822 SDValue Res;
5823 if (!Subtarget->hasCMPB())
5824 return Res;
5825
5826 if (N->getValueType(0) != MVT::i32 &&
5827 N->getValueType(0) != MVT::i64)
5828 return Res;
5829
5830 EVT VT = N->getValueType(0);
5831
5832 SDValue RHS, LHS;
5833 bool BytesFound[8] = {false, false, false, false, false, false, false, false};
5834 uint64_t Mask = 0, Alt = 0;
5835
5836 auto IsByteSelectCC = [this](SDValue O, unsigned &b,
5837 uint64_t &Mask, uint64_t &Alt,
5838 SDValue &LHS, SDValue &RHS) {
5839 if (O.getOpcode() != ISD::SELECT_CC)
5840 return false;
5841 ISD::CondCode CC = cast<CondCodeSDNode>(O.getOperand(4))->get();
5842
5843 if (!isa<ConstantSDNode>(O.getOperand(2)) ||
5844 !isa<ConstantSDNode>(O.getOperand(3)))
5845 return false;
5846
5847 uint64_t PM = O.getConstantOperandVal(2);
5848 uint64_t PAlt = O.getConstantOperandVal(3);
5849 for (b = 0; b < 8; ++b) {
5850 uint64_t Mask = UINT64_C(0xFF)0xFFUL << (8*b);
5851 if (PM && (PM & Mask) == PM && (PAlt & Mask) == PAlt)
5852 break;
5853 }
5854
5855 if (b == 8)
5856 return false;
5857 Mask |= PM;
5858 Alt |= PAlt;
5859
5860 if (!isa<ConstantSDNode>(O.getOperand(1)) ||
5861 O.getConstantOperandVal(1) != 0) {
5862 SDValue Op0 = O.getOperand(0), Op1 = O.getOperand(1);
5863 if (Op0.getOpcode() == ISD::TRUNCATE)
5864 Op0 = Op0.getOperand(0);
5865 if (Op1.getOpcode() == ISD::TRUNCATE)
5866 Op1 = Op1.getOperand(0);
5867
5868 if (Op0.getOpcode() == ISD::SRL && Op1.getOpcode() == ISD::SRL &&
5869 Op0.getOperand(1) == Op1.getOperand(1) && CC == ISD::SETEQ &&
5870 isa<ConstantSDNode>(Op0.getOperand(1))) {
5871
5872 unsigned Bits = Op0.getValueSizeInBits();
5873 if (b != Bits/8-1)
5874 return false;
5875 if (Op0.getConstantOperandVal(1) != Bits-8)
5876 return false;
5877
5878 LHS = Op0.getOperand(0);
5879 RHS = Op1.getOperand(0);
5880 return true;
5881 }
5882
5883 // When we have small integers (i16 to be specific), the form present
5884 // post-legalization uses SETULT in the SELECT_CC for the
5885 // higher-order byte, depending on the fact that the
5886 // even-higher-order bytes are known to all be zero, for example:
5887 // select_cc (xor $lhs, $rhs), 256, 65280, 0, setult
5888 // (so when the second byte is the same, because all higher-order
5889 // bits from bytes 3 and 4 are known to be zero, the result of the
5890 // xor can be at most 255)
5891 if (Op0.getOpcode() == ISD::XOR && CC == ISD::SETULT &&
5892 isa<ConstantSDNode>(O.getOperand(1))) {
5893
5894 uint64_t ULim = O.getConstantOperandVal(1);
5895 if (ULim != (UINT64_C(1)1UL << b*8))
5896 return false;
5897
5898 // Now we need to make sure that the upper bytes are known to be
5899 // zero.
5900 unsigned Bits = Op0.getValueSizeInBits();
5901 if (!CurDAG->MaskedValueIsZero(
5902 Op0, APInt::getHighBitsSet(Bits, Bits - (b + 1) * 8)))
5903 return false;
5904
5905 LHS = Op0.getOperand(0);
5906 RHS = Op0.getOperand(1);
5907 return true;
5908 }
5909
5910 return false;
5911 }
5912
5913 if (CC != ISD::SETEQ)
5914 return false;
5915
5916 SDValue Op = O.getOperand(0);
5917 if (Op.getOpcode() == ISD::AND) {
5918 if (!isa<ConstantSDNode>(Op.getOperand(1)))
5919 return false;
5920 if (Op.getConstantOperandVal(1) != (UINT64_C(0xFF)0xFFUL << (8*b)))
5921 return false;
5922
5923 SDValue XOR = Op.getOperand(0);
5924 if (XOR.getOpcode() == ISD::TRUNCATE)
5925 XOR = XOR.getOperand(0);
5926 if (XOR.getOpcode() != ISD::XOR)
5927 return false;
5928
5929 LHS = XOR.getOperand(0);
5930 RHS = XOR.getOperand(1);
5931 return true;
5932 } else if (Op.getOpcode() == ISD::SRL) {
5933 if (!isa<ConstantSDNode>(Op.getOperand(1)))
5934 return false;
5935 unsigned Bits = Op.getValueSizeInBits();
5936 if (b != Bits/8-1)
5937 return false;
5938 if (Op.getConstantOperandVal(1) != Bits-8)
5939 return false;
5940
5941 SDValue XOR = Op.getOperand(0);
5942 if (XOR.getOpcode() == ISD::TRUNCATE)
5943 XOR = XOR.getOperand(0);
5944 if (XOR.getOpcode() != ISD::XOR)
5945 return false;
5946
5947 LHS = XOR.getOperand(0);
5948 RHS = XOR.getOperand(1);
5949 return true;
5950 }
5951
5952 return false;
5953 };
5954
5955 SmallVector<SDValue, 8> Queue(1, SDValue(N, 0));
5956 while (!Queue.empty()) {
5957 SDValue V = Queue.pop_back_val();
5958
5959 for (const SDValue &O : V.getNode()->ops()) {
5960 unsigned b = 0;
5961 uint64_t M = 0, A = 0;
5962 SDValue OLHS, ORHS;
5963 if (O.getOpcode() == ISD::OR) {
5964 Queue.push_back(O);
5965 } else if (IsByteSelectCC(O, b, M, A, OLHS, ORHS)) {
5966 if (!LHS) {
5967 LHS = OLHS;
5968 RHS = ORHS;
5969 BytesFound[b] = true;
5970 Mask |= M;
5971 Alt |= A;
5972 } else if ((LHS == ORHS && RHS == OLHS) ||
5973 (RHS == ORHS && LHS == OLHS)) {
5974 BytesFound[b] = true;
5975 Mask |= M;
5976 Alt |= A;
5977 } else {
5978 return Res;
5979 }
5980 } else {
5981 return Res;
5982 }
5983 }
5984 }
5985
5986 unsigned LastB = 0, BCnt = 0;
5987 for (unsigned i = 0; i < 8; ++i)
5988 if (BytesFound[LastB]) {
5989 ++BCnt;
5990 LastB = i;
5991 }
5992
5993 if (!LastB || BCnt < 2)
5994 return Res;
5995
5996 // Because we'll be zero-extending the output anyway if don't have a specific
5997 // value for each input byte (via the Mask), we can 'anyext' the inputs.
5998 if (LHS.getValueType() != VT) {
5999 LHS = CurDAG->getAnyExtOrTrunc(LHS, dl, VT);
6000 RHS = CurDAG->getAnyExtOrTrunc(RHS, dl, VT);
6001 }
6002
6003 Res = CurDAG->getNode(PPCISD::CMPB, dl, VT, LHS, RHS);
6004
6005 bool NonTrivialMask = ((int64_t) Mask) != INT64_C(-1)-1L;
6006 if (NonTrivialMask && !Alt) {
6007 // Res = Mask & CMPB
6008 Res = CurDAG->getNode(ISD::AND, dl, VT, Res,
6009 CurDAG->getConstant(Mask, dl, VT));
6010 } else if (Alt) {
6011 // Res = (CMPB & Mask) | (~CMPB & Alt)
6012 // Which, as suggested here:
6013 // https://graphics.stanford.edu/~seander/bithacks.html#MaskedMerge
6014 // can be written as:
6015 // Res = Alt ^ ((Alt ^ Mask) & CMPB)
6016 // useful because the (Alt ^ Mask) can be pre-computed.
6017 Res = CurDAG->getNode(ISD::AND, dl, VT, Res,
6018 CurDAG->getConstant(Mask ^ Alt, dl, VT));
6019 Res = CurDAG->getNode(ISD::XOR, dl, VT, Res,
6020 CurDAG->getConstant(Alt, dl, VT));
6021 }
6022
6023 return Res;
6024}
6025
6026// When CR bit registers are enabled, an extension of an i1 variable to a i32
6027// or i64 value is lowered in terms of a SELECT_I[48] operation, and thus
6028// involves constant materialization of a 0 or a 1 or both. If the result of
6029// the extension is then operated upon by some operator that can be constant
6030// folded with a constant 0 or 1, and that constant can be materialized using
6031// only one instruction (like a zero or one), then we should fold in those
6032// operations with the select.
6033void PPCDAGToDAGISel::foldBoolExts(SDValue &Res, SDNode *&N) {
6034 if (!Subtarget->useCRBits())
6035 return;
6036
6037 if (N->getOpcode() != ISD::ZERO_EXTEND &&
6038 N->getOpcode() != ISD::SIGN_EXTEND &&
6039 N->getOpcode() != ISD::ANY_EXTEND)
6040 return;
6041
6042 if (N->getOperand(0).getValueType() != MVT::i1)
6043 return;
6044
6045 if (!N->hasOneUse())
6046 return;
6047
6048 SDLoc dl(N);
6049 EVT VT = N->getValueType(0);
6050 SDValue Cond = N->getOperand(0);
6051 SDValue ConstTrue =
6052 CurDAG->getConstant(N->getOpcode() == ISD::SIGN_EXTEND ? -1 : 1, dl, VT);
6053 SDValue ConstFalse = CurDAG->getConstant(0, dl, VT);
6054
6055 do {
6056 SDNode *User = *N->use_begin();
6057 if (User->getNumOperands() != 2)
6058 break;
6059
6060 auto TryFold = [this, N, User, dl](SDValue Val) {
6061 SDValue UserO0 = User->getOperand(0), UserO1 = User->getOperand(1);
6062 SDValue O0 = UserO0.getNode() == N ? Val : UserO0;
6063 SDValue O1 = UserO1.getNode() == N ? Val : UserO1;
6064
6065 return CurDAG->FoldConstantArithmetic(User->getOpcode(), dl,
6066 User->getValueType(0), {O0, O1});
6067 };
6068
6069 // FIXME: When the semantics of the interaction between select and undef
6070 // are clearly defined, it may turn out to be unnecessary to break here.
6071 SDValue TrueRes = TryFold(ConstTrue);
6072 if (!TrueRes || TrueRes.isUndef())
6073 break;
6074 SDValue FalseRes = TryFold(ConstFalse);
6075 if (!FalseRes || FalseRes.isUndef())
6076 break;
6077
6078 // For us to materialize these using one instruction, we must be able to
6079 // represent them as signed 16-bit integers.
6080 uint64_t True = cast<ConstantSDNode>(TrueRes)->getZExtValue(),
6081 False = cast<ConstantSDNode>(FalseRes)->getZExtValue();
6082 if (!isInt<16>(True) || !isInt<16>(False))
6083 break;
6084
6085 // We can replace User with a new SELECT node, and try again to see if we
6086 // can fold the select with its user.
6087 Res = CurDAG->getSelect(dl, User->getValueType(0), Cond, TrueRes, FalseRes);
6088 N = User;
6089 ConstTrue = TrueRes;
6090 ConstFalse = FalseRes;
6091 } while (N->hasOneUse());
6092}
6093
6094void PPCDAGToDAGISel::PreprocessISelDAG() {
6095 SelectionDAG::allnodes_iterator Position = CurDAG->allnodes_end();
6096
6097 bool MadeChange = false;
6098 while (Position != CurDAG->allnodes_begin()) {
6099 SDNode *N = &*--Position;
6100 if (N->use_empty())
6101 continue;
6102
6103 SDValue Res;
6104 switch (N->getOpcode()) {
6105 default: break;
6106 case ISD::OR:
6107 Res = combineToCMPB(N);
6108 break;
6109 }
6110
6111 if (!Res)
6112 foldBoolExts(Res, N);
6113
6114 if (Res) {
6115 LLVM_DEBUG(dbgs() << "PPC DAG preprocessing replacing:\nOld: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "PPC DAG preprocessing replacing:\nOld: "
; } } while (false)
;
6116 LLVM_DEBUG(N->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { N->dump(CurDAG); } } while (false)
;
6117 LLVM_DEBUG(dbgs() << "\nNew: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\nNew: "; } } while (false
)
;
6118 LLVM_DEBUG(Res.getNode()->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { Res.getNode()->dump(CurDAG); } } while (
false)
;
6119 LLVM_DEBUG(dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\n"; } } while (false)
;
6120
6121 CurDAG->ReplaceAllUsesOfValueWith(SDValue(N, 0), Res);
6122 MadeChange = true;
6123 }
6124 }
6125
6126 if (MadeChange)
6127 CurDAG->RemoveDeadNodes();
6128}
6129
6130/// PostprocessISelDAG - Perform some late peephole optimizations
6131/// on the DAG representation.
6132void PPCDAGToDAGISel::PostprocessISelDAG() {
6133 // Skip peepholes at -O0.
6134 if (TM.getOptLevel() == CodeGenOpt::None)
6135 return;
6136
6137 PeepholePPC64();
6138 PeepholeCROps();
6139 PeepholePPC64ZExt();
6140}
6141
6142// Check if all users of this node will become isel where the second operand
6143// is the constant zero. If this is so, and if we can negate the condition,
6144// then we can flip the true and false operands. This will allow the zero to
6145// be folded with the isel so that we don't need to materialize a register
6146// containing zero.
6147bool PPCDAGToDAGISel::AllUsersSelectZero(SDNode *N) {
6148 for (SDNode::use_iterator UI = N->use_begin(), UE = N->use_end();
6149 UI != UE; ++UI) {
6150 SDNode *User = *UI;
6151 if (!User->isMachineOpcode())
6152 return false;
6153 if (User->getMachineOpcode() != PPC::SELECT_I4 &&
6154 User->getMachineOpcode() != PPC::SELECT_I8)
6155 return false;
6156
6157 SDNode *Op1 = User->getOperand(1).getNode();
6158 SDNode *Op2 = User->getOperand(2).getNode();
6159 // If we have a degenerate select with two equal operands, swapping will
6160 // not do anything, and we may run into an infinite loop.
6161 if (Op1 == Op2)
6162 return false;
6163
6164 if (!Op2->isMachineOpcode())
6165 return false;
6166
6167 if (Op2->getMachineOpcode() != PPC::LI &&
6168 Op2->getMachineOpcode() != PPC::LI8)
6169 return false;
6170
6171 ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op2->getOperand(0));
6172 if (!C)
6173 return false;
6174
6175 if (!C->isNullValue())
6176 return false;
6177 }
6178
6179 return true;
6180}
6181
6182void PPCDAGToDAGISel::SwapAllSelectUsers(SDNode *N) {
6183 SmallVector<SDNode *, 4> ToReplace;
6184 for (SDNode::use_iterator UI = N->use_begin(), UE = N->use_end();
6185 UI != UE; ++UI) {
6186 SDNode *User = *UI;
6187 assert((User->getMachineOpcode() == PPC::SELECT_I4 ||(static_cast <bool> ((User->getMachineOpcode() == PPC
::SELECT_I4 || User->getMachineOpcode() == PPC::SELECT_I8)
&& "Must have all select users") ? void (0) : __assert_fail
("(User->getMachineOpcode() == PPC::SELECT_I4 || User->getMachineOpcode() == PPC::SELECT_I8) && \"Must have all select users\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 6189, __extension__ __PRETTY_FUNCTION__))
6188 User->getMachineOpcode() == PPC::SELECT_I8) &&(static_cast <bool> ((User->getMachineOpcode() == PPC
::SELECT_I4 || User->getMachineOpcode() == PPC::SELECT_I8)
&& "Must have all select users") ? void (0) : __assert_fail
("(User->getMachineOpcode() == PPC::SELECT_I4 || User->getMachineOpcode() == PPC::SELECT_I8) && \"Must have all select users\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 6189, __extension__ __PRETTY_FUNCTION__))
6189 "Must have all select users")(static_cast <bool> ((User->getMachineOpcode() == PPC
::SELECT_I4 || User->getMachineOpcode() == PPC::SELECT_I8)
&& "Must have all select users") ? void (0) : __assert_fail
("(User->getMachineOpcode() == PPC::SELECT_I4 || User->getMachineOpcode() == PPC::SELECT_I8) && \"Must have all select users\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 6189, __extension__ __PRETTY_FUNCTION__))
;
6190 ToReplace.push_back(User);
6191 }
6192
6193 for (SmallVector<SDNode *, 4>::iterator UI = ToReplace.begin(),
6194 UE = ToReplace.end(); UI != UE; ++UI) {
6195 SDNode *User = *UI;
6196 SDNode *ResNode =
6197 CurDAG->getMachineNode(User->getMachineOpcode(), SDLoc(User),
6198 User->getValueType(0), User->getOperand(0),
6199 User->getOperand(2),
6200 User->getOperand(1));
6201
6202 LLVM_DEBUG(dbgs() << "CR Peephole replacing:\nOld: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "CR Peephole replacing:\nOld: "
; } } while (false)
;
6203 LLVM_DEBUG(User->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { User->dump(CurDAG); } } while (false)
;
6204 LLVM_DEBUG(dbgs() << "\nNew: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\nNew: "; } } while (false
)
;
6205 LLVM_DEBUG(ResNode->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { ResNode->dump(CurDAG); } } while (false
)
;
6206 LLVM_DEBUG(dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\n"; } } while (false)
;
6207
6208 ReplaceUses(User, ResNode);
6209 }
6210}
6211
6212void PPCDAGToDAGISel::PeepholeCROps() {
6213 bool IsModified;
6214 do {
6215 IsModified = false;
6216 for (SDNode &Node : CurDAG->allnodes()) {
6217 MachineSDNode *MachineNode = dyn_cast<MachineSDNode>(&Node);
6218 if (!MachineNode || MachineNode->use_empty())
6219 continue;
6220 SDNode *ResNode = MachineNode;
6221
6222 bool Op1Set = false, Op1Unset = false,
6223 Op1Not = false,
6224 Op2Set = false, Op2Unset = false,
6225 Op2Not = false;
6226
6227 unsigned Opcode = MachineNode->getMachineOpcode();
6228 switch (Opcode) {
6229 default: break;
6230 case PPC::CRAND:
6231 case PPC::CRNAND:
6232 case PPC::CROR:
6233 case PPC::CRXOR:
6234 case PPC::CRNOR:
6235 case PPC::CREQV:
6236 case PPC::CRANDC:
6237 case PPC::CRORC: {
6238 SDValue Op = MachineNode->getOperand(1);
6239 if (Op.isMachineOpcode()) {
6240 if (Op.getMachineOpcode() == PPC::CRSET)
6241 Op2Set = true;
6242 else if (Op.getMachineOpcode() == PPC::CRUNSET)
6243 Op2Unset = true;
6244 else if (Op.getMachineOpcode() == PPC::CRNOR &&
6245 Op.getOperand(0) == Op.getOperand(1))
6246 Op2Not = true;
6247 }
6248 LLVM_FALLTHROUGH[[gnu::fallthrough]];
6249 }
6250 case PPC::BC:
6251 case PPC::BCn:
6252 case PPC::SELECT_I4:
6253 case PPC::SELECT_I8:
6254 case PPC::SELECT_F4:
6255 case PPC::SELECT_F8:
6256 case PPC::SELECT_SPE:
6257 case PPC::SELECT_SPE4:
6258 case PPC::SELECT_VRRC:
6259 case PPC::SELECT_VSFRC:
6260 case PPC::SELECT_VSSRC:
6261 case PPC::SELECT_VSRC: {
6262 SDValue Op = MachineNode->getOperand(0);
6263 if (Op.isMachineOpcode()) {
6264 if (Op.getMachineOpcode() == PPC::CRSET)
6265 Op1Set = true;
6266 else if (Op.getMachineOpcode() == PPC::CRUNSET)
6267 Op1Unset = true;
6268 else if (Op.getMachineOpcode() == PPC::CRNOR &&
6269 Op.getOperand(0) == Op.getOperand(1))
6270 Op1Not = true;
6271 }
6272 }
6273 break;
6274 }
6275
6276 bool SelectSwap = false;
6277 switch (Opcode) {
6278 default: break;
6279 case PPC::CRAND:
6280 if (MachineNode->getOperand(0) == MachineNode->getOperand(1))
6281 // x & x = x
6282 ResNode = MachineNode->getOperand(0).getNode();
6283 else if (Op1Set)
6284 // 1 & y = y
6285 ResNode = MachineNode->getOperand(1).getNode();
6286 else if (Op2Set)
6287 // x & 1 = x
6288 ResNode = MachineNode->getOperand(0).getNode();
6289 else if (Op1Unset || Op2Unset)
6290 // x & 0 = 0 & y = 0
6291 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode),
6292 MVT::i1);
6293 else if (Op1Not)
6294 // ~x & y = andc(y, x)
6295 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode),
6296 MVT::i1, MachineNode->getOperand(1),
6297 MachineNode->getOperand(0).
6298 getOperand(0));
6299 else if (Op2Not)
6300 // x & ~y = andc(x, y)
6301 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode),
6302 MVT::i1, MachineNode->getOperand(0),
6303 MachineNode->getOperand(1).
6304 getOperand(0));
6305 else if (AllUsersSelectZero(MachineNode)) {
6306 ResNode = CurDAG->getMachineNode(PPC::CRNAND, SDLoc(MachineNode),
6307 MVT::i1, MachineNode->getOperand(0),
6308 MachineNode->getOperand(1));
6309 SelectSwap = true;
6310 }
6311 break;
6312 case PPC::CRNAND:
6313 if (MachineNode->getOperand(0) == MachineNode->getOperand(1))
6314 // nand(x, x) -> nor(x, x)
6315 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6316 MVT::i1, MachineNode->getOperand(0),
6317 MachineNode->getOperand(0));
6318 else if (Op1Set)
6319 // nand(1, y) -> nor(y, y)
6320 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6321 MVT::i1, MachineNode->getOperand(1),
6322 MachineNode->getOperand(1));
6323 else if (Op2Set)
6324 // nand(x, 1) -> nor(x, x)
6325 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6326 MVT::i1, MachineNode->getOperand(0),
6327 MachineNode->getOperand(0));
6328 else if (Op1Unset || Op2Unset)
6329 // nand(x, 0) = nand(0, y) = 1
6330 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode),
6331 MVT::i1);
6332 else if (Op1Not)
6333 // nand(~x, y) = ~(~x & y) = x | ~y = orc(x, y)
6334 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode),
6335 MVT::i1, MachineNode->getOperand(0).
6336 getOperand(0),
6337 MachineNode->getOperand(1));
6338 else if (Op2Not)
6339 // nand(x, ~y) = ~x | y = orc(y, x)
6340 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode),
6341 MVT::i1, MachineNode->getOperand(1).
6342 getOperand(0),
6343 MachineNode->getOperand(0));
6344 else if (AllUsersSelectZero(MachineNode)) {
6345 ResNode = CurDAG->getMachineNode(PPC::CRAND, SDLoc(MachineNode),
6346 MVT::i1, MachineNode->getOperand(0),
6347 MachineNode->getOperand(1));
6348 SelectSwap = true;
6349 }
6350 break;
6351 case PPC::CROR:
6352 if (MachineNode->getOperand(0) == MachineNode->getOperand(1))
6353 // x | x = x
6354 ResNode = MachineNode->getOperand(0).getNode();
6355 else if (Op1Set || Op2Set)
6356 // x | 1 = 1 | y = 1
6357 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode),
6358 MVT::i1);
6359 else if (Op1Unset)
6360 // 0 | y = y
6361 ResNode = MachineNode->getOperand(1).getNode();
6362 else if (Op2Unset)
6363 // x | 0 = x
6364 ResNode = MachineNode->getOperand(0).getNode();
6365 else if (Op1Not)
6366 // ~x | y = orc(y, x)
6367 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode),
6368 MVT::i1, MachineNode->getOperand(1),
6369 MachineNode->getOperand(0).
6370 getOperand(0));
6371 else if (Op2Not)
6372 // x | ~y = orc(x, y)
6373 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode),
6374 MVT::i1, MachineNode->getOperand(0),
6375 MachineNode->getOperand(1).
6376 getOperand(0));
6377 else if (AllUsersSelectZero(MachineNode)) {
6378 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6379 MVT::i1, MachineNode->getOperand(0),
6380 MachineNode->getOperand(1));
6381 SelectSwap = true;
6382 }
6383 break;
6384 case PPC::CRXOR:
6385 if (MachineNode->getOperand(0) == MachineNode->getOperand(1))
6386 // xor(x, x) = 0
6387 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode),
6388 MVT::i1);
6389 else if (Op1Set)
6390 // xor(1, y) -> nor(y, y)
6391 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6392 MVT::i1, MachineNode->getOperand(1),
6393 MachineNode->getOperand(1));
6394 else if (Op2Set)
6395 // xor(x, 1) -> nor(x, x)
6396 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6397 MVT::i1, MachineNode->getOperand(0),
6398 MachineNode->getOperand(0));
6399 else if (Op1Unset)
6400 // xor(0, y) = y
6401 ResNode = MachineNode->getOperand(1).getNode();
6402 else if (Op2Unset)
6403 // xor(x, 0) = x
6404 ResNode = MachineNode->getOperand(0).getNode();
6405 else if (Op1Not)
6406 // xor(~x, y) = eqv(x, y)
6407 ResNode = CurDAG->getMachineNode(PPC::CREQV, SDLoc(MachineNode),
6408 MVT::i1, MachineNode->getOperand(0).
6409 getOperand(0),
6410 MachineNode->getOperand(1));
6411 else if (Op2Not)
6412 // xor(x, ~y) = eqv(x, y)
6413 ResNode = CurDAG->getMachineNode(PPC::CREQV, SDLoc(MachineNode),
6414 MVT::i1, MachineNode->getOperand(0),
6415 MachineNode->getOperand(1).
6416 getOperand(0));
6417 else if (AllUsersSelectZero(MachineNode)) {
6418 ResNode = CurDAG->getMachineNode(PPC::CREQV, SDLoc(MachineNode),
6419 MVT::i1, MachineNode->getOperand(0),
6420 MachineNode->getOperand(1));
6421 SelectSwap = true;
6422 }
6423 break;
6424 case PPC::CRNOR:
6425 if (Op1Set || Op2Set)
6426 // nor(1, y) -> 0
6427 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode),
6428 MVT::i1);
6429 else if (Op1Unset)
6430 // nor(0, y) = ~y -> nor(y, y)
6431 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6432 MVT::i1, MachineNode->getOperand(1),
6433 MachineNode->getOperand(1));
6434 else if (Op2Unset)
6435 // nor(x, 0) = ~x
6436 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6437 MVT::i1, MachineNode->getOperand(0),
6438 MachineNode->getOperand(0));
6439 else if (Op1Not)
6440 // nor(~x, y) = andc(x, y)
6441 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode),
6442 MVT::i1, MachineNode->getOperand(0).
6443 getOperand(0),
6444 MachineNode->getOperand(1));
6445 else if (Op2Not)
6446 // nor(x, ~y) = andc(y, x)
6447 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode),
6448 MVT::i1, MachineNode->getOperand(1).
6449 getOperand(0),
6450 MachineNode->getOperand(0));
6451 else if (AllUsersSelectZero(MachineNode)) {
6452 ResNode = CurDAG->getMachineNode(PPC::CROR, SDLoc(MachineNode),
6453 MVT::i1, MachineNode->getOperand(0),
6454 MachineNode->getOperand(1));
6455 SelectSwap = true;
6456 }
6457 break;
6458 case PPC::CREQV:
6459 if (MachineNode->getOperand(0) == MachineNode->getOperand(1))
6460 // eqv(x, x) = 1
6461 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode),
6462 MVT::i1);
6463 else if (Op1Set)
6464 // eqv(1, y) = y
6465 ResNode = MachineNode->getOperand(1).getNode();
6466 else if (Op2Set)
6467 // eqv(x, 1) = x
6468 ResNode = MachineNode->getOperand(0).getNode();
6469 else if (Op1Unset)
6470 // eqv(0, y) = ~y -> nor(y, y)
6471 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6472 MVT::i1, MachineNode->getOperand(1),
6473 MachineNode->getOperand(1));
6474 else if (Op2Unset)
6475 // eqv(x, 0) = ~x
6476 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6477 MVT::i1, MachineNode->getOperand(0),
6478 MachineNode->getOperand(0));
6479 else if (Op1Not)
6480 // eqv(~x, y) = xor(x, y)
6481 ResNode = CurDAG->getMachineNode(PPC::CRXOR, SDLoc(MachineNode),
6482 MVT::i1, MachineNode->getOperand(0).
6483 getOperand(0),
6484 MachineNode->getOperand(1));
6485 else if (Op2Not)
6486 // eqv(x, ~y) = xor(x, y)
6487 ResNode = CurDAG->getMachineNode(PPC::CRXOR, SDLoc(MachineNode),
6488 MVT::i1, MachineNode->getOperand(0),
6489 MachineNode->getOperand(1).
6490 getOperand(0));
6491 else if (AllUsersSelectZero(MachineNode)) {
6492 ResNode = CurDAG->getMachineNode(PPC::CRXOR, SDLoc(MachineNode),
6493 MVT::i1, MachineNode->getOperand(0),
6494 MachineNode->getOperand(1));
6495 SelectSwap = true;
6496 }
6497 break;
6498 case PPC::CRANDC:
6499 if (MachineNode->getOperand(0) == MachineNode->getOperand(1))
6500 // andc(x, x) = 0
6501 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode),
6502 MVT::i1);
6503 else if (Op1Set)
6504 // andc(1, y) = ~y
6505 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6506 MVT::i1, MachineNode->getOperand(1),
6507 MachineNode->getOperand(1));
6508 else if (Op1Unset || Op2Set)
6509 // andc(0, y) = andc(x, 1) = 0
6510 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode),
6511 MVT::i1);
6512 else if (Op2Unset)
6513 // andc(x, 0) = x
6514 ResNode = MachineNode->getOperand(0).getNode();
6515 else if (Op1Not)
6516 // andc(~x, y) = ~(x | y) = nor(x, y)
6517 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6518 MVT::i1, MachineNode->getOperand(0).
6519 getOperand(0),
6520 MachineNode->getOperand(1));
6521 else if (Op2Not)
6522 // andc(x, ~y) = x & y
6523 ResNode = CurDAG->getMachineNode(PPC::CRAND, SDLoc(MachineNode),
6524 MVT::i1, MachineNode->getOperand(0),
6525 MachineNode->getOperand(1).
6526 getOperand(0));
6527 else if (AllUsersSelectZero(MachineNode)) {
6528 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode),
6529 MVT::i1, MachineNode->getOperand(1),
6530 MachineNode->getOperand(0));
6531 SelectSwap = true;
6532 }
6533 break;
6534 case PPC::CRORC:
6535 if (MachineNode->getOperand(0) == MachineNode->getOperand(1))
6536 // orc(x, x) = 1
6537 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode),
6538 MVT::i1);
6539 else if (Op1Set || Op2Unset)
6540 // orc(1, y) = orc(x, 0) = 1
6541 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode),
6542 MVT::i1);
6543 else if (Op2Set)
6544 // orc(x, 1) = x
6545 ResNode = MachineNode->getOperand(0).getNode();
6546 else if (Op1Unset)
6547 // orc(0, y) = ~y
6548 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode),
6549 MVT::i1, MachineNode->getOperand(1),
6550 MachineNode->getOperand(1));
6551 else if (Op1Not)
6552 // orc(~x, y) = ~(x & y) = nand(x, y)
6553 ResNode = CurDAG->getMachineNode(PPC::CRNAND, SDLoc(MachineNode),
6554 MVT::i1, MachineNode->getOperand(0).
6555 getOperand(0),
6556 MachineNode->getOperand(1));
6557 else if (Op2Not)
6558 // orc(x, ~y) = x | y
6559 ResNode = CurDAG->getMachineNode(PPC::CROR, SDLoc(MachineNode),
6560 MVT::i1, MachineNode->getOperand(0),
6561 MachineNode->getOperand(1).
6562 getOperand(0));
6563 else if (AllUsersSelectZero(MachineNode)) {
6564 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode),
6565 MVT::i1, MachineNode->getOperand(1),
6566 MachineNode->getOperand(0));
6567 SelectSwap = true;
6568 }
6569 break;
6570 case PPC::SELECT_I4:
6571 case PPC::SELECT_I8:
6572 case PPC::SELECT_F4:
6573 case PPC::SELECT_F8:
6574 case PPC::SELECT_SPE:
6575 case PPC::SELECT_SPE4:
6576 case PPC::SELECT_VRRC:
6577 case PPC::SELECT_VSFRC:
6578 case PPC::SELECT_VSSRC:
6579 case PPC::SELECT_VSRC:
6580 if (Op1Set)
6581 ResNode = MachineNode->getOperand(1).getNode();
6582 else if (Op1Unset)
6583 ResNode = MachineNode->getOperand(2).getNode();
6584 else if (Op1Not)
6585 ResNode = CurDAG->getMachineNode(MachineNode->getMachineOpcode(),
6586 SDLoc(MachineNode),
6587 MachineNode->getValueType(0),
6588 MachineNode->getOperand(0).
6589 getOperand(0),
6590 MachineNode->getOperand(2),
6591 MachineNode->getOperand(1));
6592 break;
6593 case PPC::BC:
6594 case PPC::BCn:
6595 if (Op1Not)
6596 ResNode = CurDAG->getMachineNode(Opcode == PPC::BC ? PPC::BCn :
6597 PPC::BC,
6598 SDLoc(MachineNode),
6599 MVT::Other,
6600 MachineNode->getOperand(0).
6601 getOperand(0),
6602 MachineNode->getOperand(1),
6603 MachineNode->getOperand(2));
6604 // FIXME: Handle Op1Set, Op1Unset here too.
6605 break;
6606 }
6607
6608 // If we're inverting this node because it is used only by selects that
6609 // we'd like to swap, then swap the selects before the node replacement.
6610 if (SelectSwap)
6611 SwapAllSelectUsers(MachineNode);
6612
6613 if (ResNode != MachineNode) {
6614 LLVM_DEBUG(dbgs() << "CR Peephole replacing:\nOld: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "CR Peephole replacing:\nOld: "
; } } while (false)
;
6615 LLVM_DEBUG(MachineNode->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { MachineNode->dump(CurDAG); } } while (false
)
;
6616 LLVM_DEBUG(dbgs() << "\nNew: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\nNew: "; } } while (false
)
;
6617 LLVM_DEBUG(ResNode->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { ResNode->dump(CurDAG); } } while (false
)
;
6618 LLVM_DEBUG(dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\n"; } } while (false)
;
6619
6620 ReplaceUses(MachineNode, ResNode);
6621 IsModified = true;
6622 }
6623 }
6624 if (IsModified)
6625 CurDAG->RemoveDeadNodes();
6626 } while (IsModified);
6627}
6628
6629// Gather the set of 32-bit operations that are known to have their
6630// higher-order 32 bits zero, where ToPromote contains all such operations.
6631static bool PeepholePPC64ZExtGather(SDValue Op32,
6632 SmallPtrSetImpl<SDNode *> &ToPromote) {
6633 if (!Op32.isMachineOpcode())
6634 return false;
6635
6636 // First, check for the "frontier" instructions (those that will clear the
6637 // higher-order 32 bits.
6638
6639 // For RLWINM and RLWNM, we need to make sure that the mask does not wrap
6640 // around. If it does not, then these instructions will clear the
6641 // higher-order bits.
6642 if ((Op32.getMachineOpcode() == PPC::RLWINM ||
6643 Op32.getMachineOpcode() == PPC::RLWNM) &&
6644 Op32.getConstantOperandVal(2) <= Op32.getConstantOperandVal(3)) {
6645 ToPromote.insert(Op32.getNode());
6646 return true;
6647 }
6648
6649 // SLW and SRW always clear the higher-order bits.
6650 if (Op32.getMachineOpcode() == PPC::SLW ||
6651 Op32.getMachineOpcode() == PPC::SRW) {
6652 ToPromote.insert(Op32.getNode());
6653 return true;
6654 }
6655
6656 // For LI and LIS, we need the immediate to be positive (so that it is not
6657 // sign extended).
6658 if (Op32.getMachineOpcode() == PPC::LI ||
6659 Op32.getMachineOpcode() == PPC::LIS) {
6660 if (!isUInt<15>(Op32.getConstantOperandVal(0)))
6661 return false;
6662
6663 ToPromote.insert(Op32.getNode());
6664 return true;
6665 }
6666
6667 // LHBRX and LWBRX always clear the higher-order bits.
6668 if (Op32.getMachineOpcode() == PPC::LHBRX ||
6669 Op32.getMachineOpcode() == PPC::LWBRX) {
6670 ToPromote.insert(Op32.getNode());
6671 return true;
6672 }
6673
6674 // CNT[LT]ZW always produce a 64-bit value in [0,32], and so is zero extended.
6675 if (Op32.getMachineOpcode() == PPC::CNTLZW ||
6676 Op32.getMachineOpcode() == PPC::CNTTZW) {
6677 ToPromote.insert(Op32.getNode());
6678 return true;
6679 }
6680
6681 // Next, check for those instructions we can look through.
6682
6683 // Assuming the mask does not wrap around, then the higher-order bits are
6684 // taken directly from the first operand.
6685 if (Op32.getMachineOpcode() == PPC::RLWIMI &&
6686 Op32.getConstantOperandVal(3) <= Op32.getConstantOperandVal(4)) {
6687 SmallPtrSet<SDNode *, 16> ToPromote1;
6688 if (!PeepholePPC64ZExtGather(Op32.getOperand(0), ToPromote1))
6689 return false;
6690
6691 ToPromote.insert(Op32.getNode());
6692 ToPromote.insert(ToPromote1.begin(), ToPromote1.end());
6693 return true;
6694 }
6695
6696 // For OR, the higher-order bits are zero if that is true for both operands.
6697 // For SELECT_I4, the same is true (but the relevant operand numbers are
6698 // shifted by 1).
6699 if (Op32.getMachineOpcode() == PPC::OR ||
6700 Op32.getMachineOpcode() == PPC::SELECT_I4) {
6701 unsigned B = Op32.getMachineOpcode() == PPC::SELECT_I4 ? 1 : 0;
6702 SmallPtrSet<SDNode *, 16> ToPromote1;
6703 if (!PeepholePPC64ZExtGather(Op32.getOperand(B+0), ToPromote1))
6704 return false;
6705 if (!PeepholePPC64ZExtGather(Op32.getOperand(B+1), ToPromote1))
6706 return false;
6707
6708 ToPromote.insert(Op32.getNode());
6709 ToPromote.insert(ToPromote1.begin(), ToPromote1.end());
6710 return true;
6711 }
6712
6713 // For ORI and ORIS, we need the higher-order bits of the first operand to be
6714 // zero, and also for the constant to be positive (so that it is not sign
6715 // extended).
6716 if (Op32.getMachineOpcode() == PPC::ORI ||
6717 Op32.getMachineOpcode() == PPC::ORIS) {
6718 SmallPtrSet<SDNode *, 16> ToPromote1;
6719 if (!PeepholePPC64ZExtGather(Op32.getOperand(0), ToPromote1))
6720 return false;
6721 if (!isUInt<15>(Op32.getConstantOperandVal(1)))
6722 return false;
6723
6724 ToPromote.insert(Op32.getNode());
6725 ToPromote.insert(ToPromote1.begin(), ToPromote1.end());
6726 return true;
6727 }
6728
6729 // The higher-order bits of AND are zero if that is true for at least one of
6730 // the operands.
6731 if (Op32.getMachineOpcode() == PPC::AND) {
6732 SmallPtrSet<SDNode *, 16> ToPromote1, ToPromote2;
6733 bool Op0OK =
6734 PeepholePPC64ZExtGather(Op32.getOperand(0), ToPromote1);
6735 bool Op1OK =
6736 PeepholePPC64ZExtGather(Op32.getOperand(1), ToPromote2);
6737 if (!Op0OK && !Op1OK)
6738 return false;
6739
6740 ToPromote.insert(Op32.getNode());
6741
6742 if (Op0OK)
6743 ToPromote.insert(ToPromote1.begin(), ToPromote1.end());
6744
6745 if (Op1OK)
6746 ToPromote.insert(ToPromote2.begin(), ToPromote2.end());
6747
6748 return true;
6749 }
6750
6751 // For ANDI and ANDIS, the higher-order bits are zero if either that is true
6752 // of the first operand, or if the second operand is positive (so that it is
6753 // not sign extended).
6754 if (Op32.getMachineOpcode() == PPC::ANDI_rec ||
6755 Op32.getMachineOpcode() == PPC::ANDIS_rec) {
6756 SmallPtrSet<SDNode *, 16> ToPromote1;
6757 bool Op0OK =
6758 PeepholePPC64ZExtGather(Op32.getOperand(0), ToPromote1);
6759 bool Op1OK = isUInt<15>(Op32.getConstantOperandVal(1));
6760 if (!Op0OK && !Op1OK)
6761 return false;
6762
6763 ToPromote.insert(Op32.getNode());
6764
6765 if (Op0OK)
6766 ToPromote.insert(ToPromote1.begin(), ToPromote1.end());
6767
6768 return true;
6769 }
6770
6771 return false;
6772}
6773
6774void PPCDAGToDAGISel::PeepholePPC64ZExt() {
6775 if (!Subtarget->isPPC64())
6776 return;
6777
6778 // When we zero-extend from i32 to i64, we use a pattern like this:
6779 // def : Pat<(i64 (zext i32:$in)),
6780 // (RLDICL (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $in, sub_32),
6781 // 0, 32)>;
6782 // There are several 32-bit shift/rotate instructions, however, that will
6783 // clear the higher-order bits of their output, rendering the RLDICL
6784 // unnecessary. When that happens, we remove it here, and redefine the
6785 // relevant 32-bit operation to be a 64-bit operation.
6786
6787 SelectionDAG::allnodes_iterator Position = CurDAG->allnodes_end();
6788
6789 bool MadeChange = false;
6790 while (Position != CurDAG->allnodes_begin()) {
6791 SDNode *N = &*--Position;
6792 // Skip dead nodes and any non-machine opcodes.
6793 if (N->use_empty() || !N->isMachineOpcode())
6794 continue;
6795
6796 if (N->getMachineOpcode() != PPC::RLDICL)
6797 continue;
6798
6799 if (N->getConstantOperandVal(1) != 0 ||
6800 N->getConstantOperandVal(2) != 32)
6801 continue;
6802
6803 SDValue ISR = N->getOperand(0);
6804 if (!ISR.isMachineOpcode() ||
6805 ISR.getMachineOpcode() != TargetOpcode::INSERT_SUBREG)
6806 continue;
6807
6808 if (!ISR.hasOneUse())
6809 continue;
6810
6811 if (ISR.getConstantOperandVal(2) != PPC::sub_32)
6812 continue;
6813
6814 SDValue IDef = ISR.getOperand(0);
6815 if (!IDef.isMachineOpcode() ||
6816 IDef.getMachineOpcode() != TargetOpcode::IMPLICIT_DEF)
6817 continue;
6818
6819 // We now know that we're looking at a canonical i32 -> i64 zext. See if we
6820 // can get rid of it.
6821
6822 SDValue Op32 = ISR->getOperand(1);
6823 if (!Op32.isMachineOpcode())
6824 continue;
6825
6826 // There are some 32-bit instructions that always clear the high-order 32
6827 // bits, there are also some instructions (like AND) that we can look
6828 // through.
6829 SmallPtrSet<SDNode *, 16> ToPromote;
6830 if (!PeepholePPC64ZExtGather(Op32, ToPromote))
6831 continue;
6832
6833 // If the ToPromote set contains nodes that have uses outside of the set
6834 // (except for the original INSERT_SUBREG), then abort the transformation.
6835 bool OutsideUse = false;
6836 for (SDNode *PN : ToPromote) {
6837 for (SDNode *UN : PN->uses()) {
6838 if (!ToPromote.count(UN) && UN != ISR.getNode()) {
6839 OutsideUse = true;
6840 break;
6841 }
6842 }
6843
6844 if (OutsideUse)
6845 break;
6846 }
6847 if (OutsideUse)
6848 continue;
6849
6850 MadeChange = true;
6851
6852 // We now know that this zero extension can be removed by promoting to
6853 // nodes in ToPromote to 64-bit operations, where for operations in the
6854 // frontier of the set, we need to insert INSERT_SUBREGs for their
6855 // operands.
6856 for (SDNode *PN : ToPromote) {
6857 unsigned NewOpcode;
6858 switch (PN->getMachineOpcode()) {
6859 default:
6860 llvm_unreachable("Don't know the 64-bit variant of this instruction")::llvm::llvm_unreachable_internal("Don't know the 64-bit variant of this instruction"
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp"
, 6860)
;
6861 case PPC::RLWINM: NewOpcode = PPC::RLWINM8; break;
6862 case PPC::RLWNM: NewOpcode = PPC::RLWNM8; break;
6863 case PPC::SLW: NewOpcode = PPC::SLW8; break;
6864 case PPC::SRW: NewOpcode = PPC::SRW8; break;
6865 case PPC::LI: NewOpcode = PPC::LI8; break;
6866 case PPC::LIS: NewOpcode = PPC::LIS8; break;
6867 case PPC::LHBRX: NewOpcode = PPC::LHBRX8; break;
6868 case PPC::LWBRX: NewOpcode = PPC::LWBRX8; break;
6869 case PPC::CNTLZW: NewOpcode = PPC::CNTLZW8; break;
6870 case PPC::CNTTZW: NewOpcode = PPC::CNTTZW8; break;
6871 case PPC::RLWIMI: NewOpcode = PPC::RLWIMI8; break;
6872 case PPC::OR: NewOpcode = PPC::OR8; break;
6873 case PPC::SELECT_I4: NewOpcode = PPC::SELECT_I8; break;
6874 case PPC::ORI: NewOpcode = PPC::ORI8; break;
6875 case PPC::ORIS: NewOpcode = PPC::ORIS8; break;
6876 case PPC::AND: NewOpcode = PPC::AND8; break;
6877 case PPC::ANDI_rec:
6878 NewOpcode = PPC::ANDI8_rec;
6879 break;
6880 case PPC::ANDIS_rec:
6881 NewOpcode = PPC::ANDIS8_rec;
6882 break;
6883 }
6884
6885 // Note: During the replacement process, the nodes will be in an
6886 // inconsistent state (some instructions will have operands with values
6887 // of the wrong type). Once done, however, everything should be right
6888 // again.
6889
6890 SmallVector<SDValue, 4> Ops;
6891 for (const SDValue &V : PN->ops()) {
6892 if (!ToPromote.count(V.getNode()) && V.getValueType() == MVT::i32 &&
6893 !isa<ConstantSDNode>(V)) {
6894 SDValue ReplOpOps[] = { ISR.getOperand(0), V, ISR.getOperand(2) };
6895 SDNode *ReplOp =
6896 CurDAG->getMachineNode(TargetOpcode::INSERT_SUBREG, SDLoc(V),
6897 ISR.getNode()->getVTList(), ReplOpOps);
6898 Ops.push_back(SDValue(ReplOp, 0));
6899 } else {
6900 Ops.push_back(V);
6901 }
6902 }
6903
6904 // Because all to-be-promoted nodes only have users that are other
6905 // promoted nodes (or the original INSERT_SUBREG), we can safely replace
6906 // the i32 result value type with i64.
6907
6908 SmallVector<EVT, 2> NewVTs;
6909 SDVTList VTs = PN->getVTList();
6910 for (unsigned i = 0, ie = VTs.NumVTs; i != ie; ++i)
6911 if (VTs.VTs[i] == MVT::i32)
6912 NewVTs.push_back(MVT::i64);
6913 else
6914 NewVTs.push_back(VTs.VTs[i]);
6915
6916 LLVM_DEBUG(dbgs() << "PPC64 ZExt Peephole morphing:\nOld: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "PPC64 ZExt Peephole morphing:\nOld: "
; } } while (false)
;
6917 LLVM_DEBUG(PN->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { PN->dump(CurDAG); } } while (false)
;
6918
6919 CurDAG->SelectNodeTo(PN, NewOpcode, CurDAG->getVTList(NewVTs), Ops);
6920
6921 LLVM_DEBUG(dbgs() << "\nNew: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\nNew: "; } } while (false
)
;
6922 LLVM_DEBUG(PN->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { PN->dump(CurDAG); } } while (false)
;
6923 LLVM_DEBUG(dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\n"; } } while (false)
;
6924 }
6925
6926 // Now we replace the original zero extend and its associated INSERT_SUBREG
6927 // with the value feeding the INSERT_SUBREG (which has now been promoted to
6928 // return an i64).
6929
6930 LLVM_DEBUG(dbgs() << "PPC64 ZExt Peephole replacing:\nOld: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "PPC64 ZExt Peephole replacing:\nOld: "
; } } while (false)
;
6931 LLVM_DEBUG(N->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { N->dump(CurDAG); } } while (false)
;
6932 LLVM_DEBUG(dbgs() << "\nNew: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\nNew: "; } } while (false
)
;
6933 LLVM_DEBUG(Op32.getNode()->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { Op32.getNode()->dump(CurDAG); } } while
(false)
;
6934 LLVM_DEBUG(dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\n"; } } while (false)
;
6935
6936 ReplaceUses(N, Op32.getNode());
6937 }
6938
6939 if (MadeChange)
6940 CurDAG->RemoveDeadNodes();
6941}
6942
6943static bool isVSXSwap(SDValue N) {
6944 if (!N->isMachineOpcode())
6945 return false;
6946 unsigned Opc = N->getMachineOpcode();
6947
6948 // Single-operand XXPERMDI or the regular XXPERMDI/XXSLDWI where the immediate
6949 // operand is 2.
6950 if (Opc == PPC::XXPERMDIs) {
6951 return isa<ConstantSDNode>(N->getOperand(1)) &&
6952 N->getConstantOperandVal(1) == 2;
6953 } else if (Opc == PPC::XXPERMDI || Opc == PPC::XXSLDWI) {
6954 return N->getOperand(0) == N->getOperand(1) &&
6955 isa<ConstantSDNode>(N->getOperand(2)) &&
6956 N->getConstantOperandVal(2) == 2;
6957 }
6958
6959 return false;
6960}
6961
6962// TODO: Make this complete and replace with a table-gen bit.
6963static bool isLaneInsensitive(SDValue N) {
6964 if (!N->isMachineOpcode())
6965 return false;
6966 unsigned Opc = N->getMachineOpcode();
6967
6968 switch (Opc) {
6969 default:
6970 return false;
6971 case PPC::VAVGSB:
6972 case PPC::VAVGUB:
6973 case PPC::VAVGSH:
6974 case PPC::VAVGUH:
6975 case PPC::VAVGSW:
6976 case PPC::VAVGUW:
6977 case PPC::VMAXFP:
6978 case PPC::VMAXSB:
6979 case PPC::VMAXUB:
6980 case PPC::VMAXSH:
6981 case PPC::VMAXUH:
6982 case PPC::VMAXSW:
6983 case PPC::VMAXUW:
6984 case PPC::VMINFP:
6985 case PPC::VMINSB:
6986 case PPC::VMINUB:
6987 case PPC::VMINSH:
6988 case PPC::VMINUH:
6989 case PPC::VMINSW:
6990 case PPC::VMINUW:
6991 case PPC::VADDFP:
6992 case PPC::VADDUBM:
6993 case PPC::VADDUHM:
6994 case PPC::VADDUWM:
6995 case PPC::VSUBFP:
6996 case PPC::VSUBUBM:
6997 case PPC::VSUBUHM:
6998 case PPC::VSUBUWM:
6999 case PPC::VAND:
7000 case PPC::VANDC:
7001 case PPC::VOR:
7002 case PPC::VORC:
7003 case PPC::VXOR:
7004 case PPC::VNOR:
7005 case PPC::VMULUWM:
7006 return true;
7007 }
7008}
7009
7010// Try to simplify (xxswap (vec-op (xxswap) (xxswap))) where vec-op is
7011// lane-insensitive.
7012static void reduceVSXSwap(SDNode *N, SelectionDAG *DAG) {
7013 // Our desired xxswap might be source of COPY_TO_REGCLASS.
7014 // TODO: Can we put this a common method for DAG?
7015 auto SkipRCCopy = [](SDValue V) {
7016 while (V->isMachineOpcode() &&
7017 V->getMachineOpcode() == TargetOpcode::COPY_TO_REGCLASS) {
7018 // All values in the chain should have single use.
7019 if (V->use_empty() || !V->use_begin()->isOnlyUserOf(V.getNode()))
7020 return SDValue();
7021 V = V->getOperand(0);
7022 }
7023 return V.hasOneUse() ? V : SDValue();
7024 };
7025
7026 SDValue VecOp = SkipRCCopy(N->getOperand(0));
7027 if (!VecOp || !isLaneInsensitive(VecOp))
7028 return;
7029
7030 SDValue LHS = SkipRCCopy(VecOp.getOperand(0)),
7031 RHS = SkipRCCopy(VecOp.getOperand(1));
7032 if (!LHS || !RHS || !isVSXSwap(LHS) || !isVSXSwap(RHS))
7033 return;
7034
7035 // These swaps may still have chain-uses here, count on dead code elimination
7036 // in following passes to remove them.
7037 DAG->ReplaceAllUsesOfValueWith(LHS, LHS.getOperand(0));
7038 DAG->ReplaceAllUsesOfValueWith(RHS, RHS.getOperand(0));
7039 DAG->ReplaceAllUsesOfValueWith(SDValue(N, 0), N->getOperand(0));
7040}
7041
7042void PPCDAGToDAGISel::PeepholePPC64() {
7043 SelectionDAG::allnodes_iterator Position = CurDAG->allnodes_end();
7044
7045 while (Position != CurDAG->allnodes_begin()) {
7046 SDNode *N = &*--Position;
7047 // Skip dead nodes and any non-machine opcodes.
7048 if (N->use_empty() || !N->isMachineOpcode())
7049 continue;
7050
7051 if (isVSXSwap(SDValue(N, 0)))
7052 reduceVSXSwap(N, CurDAG);
7053
7054 unsigned FirstOp;
7055 unsigned StorageOpcode = N->getMachineOpcode();
7056 bool RequiresMod4Offset = false;
7057
7058 switch (StorageOpcode) {
7059 default: continue;
7060
7061 case PPC::LWA:
7062 case PPC::LD:
7063 case PPC::DFLOADf64:
7064 case PPC::DFLOADf32:
7065 RequiresMod4Offset = true;
7066 LLVM_FALLTHROUGH[[gnu::fallthrough]];
7067 case PPC::LBZ:
7068 case PPC::LBZ8:
7069 case PPC::LFD:
7070 case PPC::LFS:
7071 case PPC::LHA:
7072 case PPC::LHA8:
7073 case PPC::LHZ:
7074 case PPC::LHZ8:
7075 case PPC::LWZ:
7076 case PPC::LWZ8:
7077 FirstOp = 0;
7078 break;
7079
7080 case PPC::STD:
7081 case PPC::DFSTOREf64:
7082 case PPC::DFSTOREf32:
7083 RequiresMod4Offset = true;
7084 LLVM_FALLTHROUGH[[gnu::fallthrough]];
7085 case PPC::STB:
7086 case PPC::STB8:
7087 case PPC::STFD:
7088 case PPC::STFS:
7089 case PPC::STH:
7090 case PPC::STH8:
7091 case PPC::STW:
7092 case PPC::STW8:
7093 FirstOp = 1;
7094 break;
7095 }
7096
7097 // If this is a load or store with a zero offset, or within the alignment,
7098 // we may be able to fold an add-immediate into the memory operation.
7099 // The check against alignment is below, as it can't occur until we check
7100 // the arguments to N
7101 if (!isa<ConstantSDNode>(N->getOperand(FirstOp)))
7102 continue;
7103
7104 SDValue Base = N->getOperand(FirstOp + 1);
7105 if (!Base.isMachineOpcode())
7106 continue;
7107
7108 unsigned Flags = 0;
7109 bool ReplaceFlags = true;
7110
7111 // When the feeding operation is an add-immediate of some sort,
7112 // determine whether we need to add relocation information to the
7113 // target flags on the immediate operand when we fold it into the
7114 // load instruction.
7115 //
7116 // For something like ADDItocL, the relocation information is
7117 // inferred from the opcode; when we process it in the AsmPrinter,
7118 // we add the necessary relocation there. A load, though, can receive
7119 // relocation from various flavors of ADDIxxx, so we need to carry
7120 // the relocation information in the target flags.
7121 switch (Base.getMachineOpcode()) {
7122 default: continue;
7123
7124 case PPC::ADDI8:
7125 case PPC::ADDI:
7126 // In some cases (such as TLS) the relocation information
7127 // is already in place on the operand, so copying the operand
7128 // is sufficient.
7129 ReplaceFlags = false;
7130 // For these cases, the immediate may not be divisible by 4, in
7131 // which case the fold is illegal for DS-form instructions. (The
7132 // other cases provide aligned addresses and are always safe.)
7133 if (RequiresMod4Offset &&
7134 (!isa<ConstantSDNode>(Base.getOperand(1)) ||
7135 Base.getConstantOperandVal(1) % 4 != 0))
7136 continue;
7137 break;
7138 case PPC::ADDIdtprelL:
7139 Flags = PPCII::MO_DTPREL_LO;
7140 break;
7141 case PPC::ADDItlsldL:
7142 Flags = PPCII::MO_TLSLD_LO;
7143 break;
7144 case PPC::ADDItocL:
7145 Flags = PPCII::MO_TOC_LO;
7146 break;
7147 }
7148
7149 SDValue ImmOpnd = Base.getOperand(1);
7150
7151 // On PPC64, the TOC base pointer is guaranteed by the ABI only to have
7152 // 8-byte alignment, and so we can only use offsets less than 8 (otherwise,
7153 // we might have needed different @ha relocation values for the offset
7154 // pointers).
7155 int MaxDisplacement = 7;
7156 if (GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(ImmOpnd)) {
7157 const GlobalValue *GV = GA->getGlobal();
7158 Align Alignment = GV->getPointerAlignment(CurDAG->getDataLayout());
7159 MaxDisplacement = std::min((int)Alignment.value() - 1, MaxDisplacement);
7160 }
7161
7162 bool UpdateHBase = false;
7163 SDValue HBase = Base.getOperand(0);
7164
7165 int Offset = N->getConstantOperandVal(FirstOp);
7166 if (ReplaceFlags) {
7167 if (Offset < 0 || Offset > MaxDisplacement) {
7168 // If we have a addi(toc@l)/addis(toc@ha) pair, and the addis has only
7169 // one use, then we can do this for any offset, we just need to also
7170 // update the offset (i.e. the symbol addend) on the addis also.
7171 if (Base.getMachineOpcode() != PPC::ADDItocL)
7172 continue;
7173
7174 if (!HBase.isMachineOpcode() ||
7175 HBase.getMachineOpcode() != PPC::ADDIStocHA8)
7176 continue;
7177
7178 if (!Base.hasOneUse() || !HBase.hasOneUse())
7179 continue;
7180
7181 SDValue HImmOpnd = HBase.getOperand(1);
7182 if (HImmOpnd != ImmOpnd)
7183 continue;
7184
7185 UpdateHBase = true;
7186 }
7187 } else {
7188 // If we're directly folding the addend from an addi instruction, then:
7189 // 1. In general, the offset on the memory access must be zero.
7190 // 2. If the addend is a constant, then it can be combined with a
7191 // non-zero offset, but only if the result meets the encoding
7192 // requirements.
7193 if (auto *C = dyn_cast<ConstantSDNode>(ImmOpnd)) {
7194 Offset += C->getSExtValue();
7195
7196 if (RequiresMod4Offset && (Offset % 4) != 0)
7197 continue;
7198
7199 if (!isInt<16>(Offset))
7200 continue;
7201
7202 ImmOpnd = CurDAG->getTargetConstant(Offset, SDLoc(ImmOpnd),
7203 ImmOpnd.getValueType());
7204 } else if (Offset != 0) {
7205 continue;
7206 }
7207 }
7208
7209 // We found an opportunity. Reverse the operands from the add
7210 // immediate and substitute them into the load or store. If
7211 // needed, update the target flags for the immediate operand to
7212 // reflect the necessary relocation information.
7213 LLVM_DEBUG(dbgs() << "Folding add-immediate into mem-op:\nBase: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Folding add-immediate into mem-op:\nBase: "
; } } while (false)
;
7214 LLVM_DEBUG(Base->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { Base->dump(CurDAG); } } while (false)
;
7215 LLVM_DEBUG(dbgs() << "\nN: ")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\nN: "; } } while (false)
;
7216 LLVM_DEBUG(N->dump(CurDAG))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { N->dump(CurDAG); } } while (false)
;
7217 LLVM_DEBUG(dbgs() << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "\n"; } } while (false)
;
7218
7219 // If the relocation information isn't already present on the
7220 // immediate operand, add it now.
7221 if (ReplaceFlags) {
7222 if (GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(ImmOpnd)) {
7223 SDLoc dl(GA);
7224 const GlobalValue *GV = GA->getGlobal();
7225 Align Alignment = GV->getPointerAlignment(CurDAG->getDataLayout());
7226 // We can't perform this optimization for data whose alignment
7227 // is insufficient for the instruction encoding.
7228 if (Alignment < 4 && (RequiresMod4Offset || (Offset % 4) != 0)) {
7229 LLVM_DEBUG(dbgs() << "Rejected this candidate for alignment.\n\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("ppc-codegen")) { dbgs() << "Rejected this candidate for alignment.\n\n"
; } } while (false)
;
7230 continue;
7231 }
7232 ImmOpnd = CurDAG->getTargetGlobalAddress(GV, dl, MVT::i64, Offset, Flags);
7233 } else if (ConstantPoolSDNode *CP =
7234 dyn_cast<ConstantPoolSDNode>(ImmOpnd)) {
7235 const Constant *C = CP->getConstVal();
7236 ImmOpnd = CurDAG->getTargetConstantPool(C, MVT::i64, CP->getAlign(),
7237 Offset, Flags);
7238 }
7239 }
7240
7241 if (FirstOp == 1) // Store
7242 (void)CurDAG->UpdateNodeOperands(N, N->getOperand(0), ImmOpnd,
7243 Base.getOperand(0), N->getOperand(3));
7244 else // Load
7245 (void)CurDAG->UpdateNodeOperands(N, ImmOpnd, Base.getOperand(0),
7246 N->getOperand(2));
7247
7248 if (UpdateHBase)
7249 (void)CurDAG->UpdateNodeOperands(HBase.getNode(), HBase.getOperand(0),
7250 ImmOpnd);
7251
7252 // The add-immediate may now be dead, in which case remove it.
7253 if (Base.getNode()->use_empty())
7254 CurDAG->RemoveDeadNode(Base.getNode());
7255 }
7256}
7257
7258/// createPPCISelDag - This pass converts a legalized DAG into a
7259/// PowerPC-specific DAG, ready for instruction scheduling.
7260///
7261FunctionPass *llvm::createPPCISelDag(PPCTargetMachine &TM,
7262 CodeGenOpt::Level OptLevel) {
7263 return new PPCDAGToDAGISel(TM, OptLevel);
7264}

/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h

1//===- llvm/Support/Casting.h - Allow flexible, checked, casts --*- 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//
9// This file defines the isa<X>(), cast<X>(), dyn_cast<X>(), cast_or_null<X>(),
10// and dyn_cast_or_null<X>() templates.
11//
12//===----------------------------------------------------------------------===//
13
14#ifndef LLVM_SUPPORT_CASTING_H
15#define LLVM_SUPPORT_CASTING_H
16
17#include "llvm/Support/Compiler.h"
18#include "llvm/Support/type_traits.h"
19#include <cassert>
20#include <memory>
21#include <type_traits>
22
23namespace llvm {
24
25//===----------------------------------------------------------------------===//
26// isa<x> Support Templates
27//===----------------------------------------------------------------------===//
28
29// Define a template that can be specialized by smart pointers to reflect the
30// fact that they are automatically dereferenced, and are not involved with the
31// template selection process... the default implementation is a noop.
32//
33template<typename From> struct simplify_type {
34 using SimpleType = From; // The real type this represents...
35
36 // An accessor to get the real value...
37 static SimpleType &getSimplifiedValue(From &Val) { return Val; }
38};
39
40template<typename From> struct simplify_type<const From> {
41 using NonConstSimpleType = typename simplify_type<From>::SimpleType;
42 using SimpleType =
43 typename add_const_past_pointer<NonConstSimpleType>::type;
44 using RetType =
45 typename add_lvalue_reference_if_not_pointer<SimpleType>::type;
46
47 static RetType getSimplifiedValue(const From& Val) {
48 return simplify_type<From>::getSimplifiedValue(const_cast<From&>(Val));
49 }
50};
51
52// The core of the implementation of isa<X> is here; To and From should be
53// the names of classes. This template can be specialized to customize the
54// implementation of isa<> without rewriting it from scratch.
55template <typename To, typename From, typename Enabler = void>
56struct isa_impl {
57 static inline bool doit(const From &Val) {
58 return To::classof(&Val);
59 }
60};
61
62/// Always allow upcasts, and perform no dynamic check for them.
63template <typename To, typename From>
64struct isa_impl<To, From, std::enable_if_t<std::is_base_of<To, From>::value>> {
65 static inline bool doit(const From &) { return true; }
66};
67
68template <typename To, typename From> struct isa_impl_cl {
69 static inline bool doit(const From &Val) {
70 return isa_impl<To, From>::doit(Val);
71 }
72};
73
74template <typename To, typename From> struct isa_impl_cl<To, const From> {
75 static inline bool doit(const From &Val) {
76 return isa_impl<To, From>::doit(Val);
77 }
78};
79
80template <typename To, typename From>
81struct isa_impl_cl<To, const std::unique_ptr<From>> {
82 static inline bool doit(const std::unique_ptr<From> &Val) {
83 assert(Val && "isa<> used on a null pointer")(static_cast <bool> (Val && "isa<> used on a null pointer"
) ? void (0) : __assert_fail ("Val && \"isa<> used on a null pointer\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 83, __extension__ __PRETTY_FUNCTION__))
;
84 return isa_impl_cl<To, From>::doit(*Val);
85 }
86};
87
88template <typename To, typename From> struct isa_impl_cl<To, From*> {
89 static inline bool doit(const From *Val) {
90 assert(Val && "isa<> used on a null pointer")(static_cast <bool> (Val && "isa<> used on a null pointer"
) ? void (0) : __assert_fail ("Val && \"isa<> used on a null pointer\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 90, __extension__ __PRETTY_FUNCTION__))
;
91 return isa_impl<To, From>::doit(*Val);
92 }
93};
94
95template <typename To, typename From> struct isa_impl_cl<To, From*const> {
96 static inline bool doit(const From *Val) {
97 assert(Val && "isa<> used on a null pointer")(static_cast <bool> (Val && "isa<> used on a null pointer"
) ? void (0) : __assert_fail ("Val && \"isa<> used on a null pointer\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 97, __extension__ __PRETTY_FUNCTION__))
;
98 return isa_impl<To, From>::doit(*Val);
99 }
100};
101
102template <typename To, typename From> struct isa_impl_cl<To, const From*> {
103 static inline bool doit(const From *Val) {
104 assert(Val && "isa<> used on a null pointer")(static_cast <bool> (Val && "isa<> used on a null pointer"
) ? void (0) : __assert_fail ("Val && \"isa<> used on a null pointer\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 104, __extension__ __PRETTY_FUNCTION__))
;
105 return isa_impl<To, From>::doit(*Val);
106 }
107};
108
109template <typename To, typename From> struct isa_impl_cl<To, const From*const> {
110 static inline bool doit(const From *Val) {
111 assert(Val && "isa<> used on a null pointer")(static_cast <bool> (Val && "isa<> used on a null pointer"
) ? void (0) : __assert_fail ("Val && \"isa<> used on a null pointer\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 111, __extension__ __PRETTY_FUNCTION__))
;
112 return isa_impl<To, From>::doit(*Val);
113 }
114};
115
116template<typename To, typename From, typename SimpleFrom>
117struct isa_impl_wrap {
118 // When From != SimplifiedType, we can simplify the type some more by using
119 // the simplify_type template.
120 static bool doit(const From &Val) {
121 return isa_impl_wrap<To, SimpleFrom,
122 typename simplify_type<SimpleFrom>::SimpleType>::doit(
123 simplify_type<const From>::getSimplifiedValue(Val));
124 }
125};
126
127template<typename To, typename FromTy>
128struct isa_impl_wrap<To, FromTy, FromTy> {
129 // When From == SimpleType, we are as simple as we are going to get.
130 static bool doit(const FromTy &Val) {
131 return isa_impl_cl<To,FromTy>::doit(Val);
132 }
133};
134
135// isa<X> - Return true if the parameter to the template is an instance of one
136// of the template type arguments. Used like this:
137//
138// if (isa<Type>(myVal)) { ... }
139// if (isa<Type0, Type1, Type2>(myVal)) { ... }
140//
141template <class X, class Y> LLVM_NODISCARD[[clang::warn_unused_result]] inline bool isa(const Y &Val) {
142 return isa_impl_wrap<X, const Y,
143 typename simplify_type<const Y>::SimpleType>::doit(Val);
144}
145
146template <typename First, typename Second, typename... Rest, typename Y>
147LLVM_NODISCARD[[clang::warn_unused_result]] inline bool isa(const Y &Val) {
148 return isa<First>(Val) || isa<Second, Rest...>(Val);
149}
150
151// isa_and_nonnull<X> - Functionally identical to isa, except that a null value
152// is accepted.
153//
154template <typename... X, class Y>
155LLVM_NODISCARD[[clang::warn_unused_result]] inline bool isa_and_nonnull(const Y &Val) {
156 if (!Val)
157 return false;
158 return isa<X...>(Val);
159}
160
161//===----------------------------------------------------------------------===//
162// cast<x> Support Templates
163//===----------------------------------------------------------------------===//
164
165template<class To, class From> struct cast_retty;
166
167// Calculate what type the 'cast' function should return, based on a requested
168// type of To and a source type of From.
169template<class To, class From> struct cast_retty_impl {
170 using ret_type = To &; // Normal case, return Ty&
171};
172template<class To, class From> struct cast_retty_impl<To, const From> {
173 using ret_type = const To &; // Normal case, return Ty&
174};
175
176template<class To, class From> struct cast_retty_impl<To, From*> {
177 using ret_type = To *; // Pointer arg case, return Ty*
178};
179
180template<class To, class From> struct cast_retty_impl<To, const From*> {
181 using ret_type = const To *; // Constant pointer arg case, return const Ty*
182};
183
184template<class To, class From> struct cast_retty_impl<To, const From*const> {
185 using ret_type = const To *; // Constant pointer arg case, return const Ty*
186};
187
188template <class To, class From>
189struct cast_retty_impl<To, std::unique_ptr<From>> {
190private:
191 using PointerType = typename cast_retty_impl<To, From *>::ret_type;
192 using ResultType = std::remove_pointer_t<PointerType>;
193
194public:
195 using ret_type = std::unique_ptr<ResultType>;
196};
197
198template<class To, class From, class SimpleFrom>
199struct cast_retty_wrap {
200 // When the simplified type and the from type are not the same, use the type
201 // simplifier to reduce the type, then reuse cast_retty_impl to get the
202 // resultant type.
203 using ret_type = typename cast_retty<To, SimpleFrom>::ret_type;
204};
205
206template<class To, class FromTy>
207struct cast_retty_wrap<To, FromTy, FromTy> {
208 // When the simplified type is equal to the from type, use it directly.
209 using ret_type = typename cast_retty_impl<To,FromTy>::ret_type;
210};
211
212template<class To, class From>
213struct cast_retty {
214 using ret_type = typename cast_retty_wrap<
215 To, From, typename simplify_type<From>::SimpleType>::ret_type;
216};
217
218// Ensure the non-simple values are converted using the simplify_type template
219// that may be specialized by smart pointers...
220//
221template<class To, class From, class SimpleFrom> struct cast_convert_val {
222 // This is not a simple type, use the template to simplify it...
223 static typename cast_retty<To, From>::ret_type doit(From &Val) {
224 return cast_convert_val<To, SimpleFrom,
225 typename simplify_type<SimpleFrom>::SimpleType>::doit(
226 simplify_type<From>::getSimplifiedValue(Val));
227 }
228};
229
230template<class To, class FromTy> struct cast_convert_val<To,FromTy,FromTy> {
231 // This _is_ a simple type, just cast it.
232 static typename cast_retty<To, FromTy>::ret_type doit(const FromTy &Val) {
233 typename cast_retty<To, FromTy>::ret_type Res2
234 = (typename cast_retty<To, FromTy>::ret_type)const_cast<FromTy&>(Val);
235 return Res2;
236 }
237};
238
239template <class X> struct is_simple_type {
240 static const bool value =
241 std::is_same<X, typename simplify_type<X>::SimpleType>::value;
242};
243
244// cast<X> - Return the argument parameter cast to the specified type. This
245// casting operator asserts that the type is correct, so it does not return null
246// on failure. It does not allow a null argument (use cast_or_null for that).
247// It is typically used like this:
248//
249// cast<Instruction>(myVal)->getParent()
250//
251template <class X, class Y>
252inline std::enable_if_t<!is_simple_type<Y>::value,
253 typename cast_retty<X, const Y>::ret_type>
254cast(const Y &Val) {
255 assert(isa<X>(Val) && "cast<Ty>() argument of incompatible type!")(static_cast <bool> (isa<X>(Val) && "cast<Ty>() argument of incompatible type!"
) ? void (0) : __assert_fail ("isa<X>(Val) && \"cast<Ty>() argument of incompatible type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 255, __extension__ __PRETTY_FUNCTION__))
;
256 return cast_convert_val<
257 X, const Y, typename simplify_type<const Y>::SimpleType>::doit(Val);
258}
259
260template <class X, class Y>
261inline typename cast_retty<X, Y>::ret_type cast(Y &Val) {
262 assert(isa<X>(Val) && "cast<Ty>() argument of incompatible type!")(static_cast <bool> (isa<X>(Val) && "cast<Ty>() argument of incompatible type!"
) ? void (0) : __assert_fail ("isa<X>(Val) && \"cast<Ty>() argument of incompatible type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 262, __extension__ __PRETTY_FUNCTION__))
;
5
Assuming 'Val' is a 'GlobalAddressSDNode'
6
'?' condition is true
263 return cast_convert_val<X, Y,
7
Returning pointer, which participates in a condition later
264 typename simplify_type<Y>::SimpleType>::doit(Val);
265}
266
267template <class X, class Y>
268inline typename cast_retty<X, Y *>::ret_type cast(Y *Val) {
269 assert(isa<X>(Val) && "cast<Ty>() argument of incompatible type!")(static_cast <bool> (isa<X>(Val) && "cast<Ty>() argument of incompatible type!"
) ? void (0) : __assert_fail ("isa<X>(Val) && \"cast<Ty>() argument of incompatible type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 269, __extension__ __PRETTY_FUNCTION__))
;
270 return cast_convert_val<X, Y*,
271 typename simplify_type<Y*>::SimpleType>::doit(Val);
272}
273
274template <class X, class Y>
275inline typename cast_retty<X, std::unique_ptr<Y>>::ret_type
276cast(std::unique_ptr<Y> &&Val) {
277 assert(isa<X>(Val.get()) && "cast<Ty>() argument of incompatible type!")(static_cast <bool> (isa<X>(Val.get()) &&
"cast<Ty>() argument of incompatible type!") ? void (0
) : __assert_fail ("isa<X>(Val.get()) && \"cast<Ty>() argument of incompatible type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 277, __extension__ __PRETTY_FUNCTION__))
;
278 using ret_type = typename cast_retty<X, std::unique_ptr<Y>>::ret_type;
279 return ret_type(
280 cast_convert_val<X, Y *, typename simplify_type<Y *>::SimpleType>::doit(
281 Val.release()));
282}
283
284// cast_or_null<X> - Functionally identical to cast, except that a null value is
285// accepted.
286//
287template <class X, class Y>
288LLVM_NODISCARD[[clang::warn_unused_result]] inline std::enable_if_t<
289 !is_simple_type<Y>::value, typename cast_retty<X, const Y>::ret_type>
290cast_or_null(const Y &Val) {
291 if (!Val)
292 return nullptr;
293 assert(isa<X>(Val) && "cast_or_null<Ty>() argument of incompatible type!")(static_cast <bool> (isa<X>(Val) && "cast_or_null<Ty>() argument of incompatible type!"
) ? void (0) : __assert_fail ("isa<X>(Val) && \"cast_or_null<Ty>() argument of incompatible type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 293, __extension__ __PRETTY_FUNCTION__))
;
294 return cast<X>(Val);
295}
296
297template <class X, class Y>
298LLVM_NODISCARD[[clang::warn_unused_result]] inline std::enable_if_t<!is_simple_type<Y>::value,
299 typename cast_retty<X, Y>::ret_type>
300cast_or_null(Y &Val) {
301 if (!Val)
302 return nullptr;
303 assert(isa<X>(Val) && "cast_or_null<Ty>() argument of incompatible type!")(static_cast <bool> (isa<X>(Val) && "cast_or_null<Ty>() argument of incompatible type!"
) ? void (0) : __assert_fail ("isa<X>(Val) && \"cast_or_null<Ty>() argument of incompatible type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 303, __extension__ __PRETTY_FUNCTION__))
;
304 return cast<X>(Val);
305}
306
307template <class X, class Y>
308LLVM_NODISCARD[[clang::warn_unused_result]] inline typename cast_retty<X, Y *>::ret_type
309cast_or_null(Y *Val) {
310 if (!Val) return nullptr;
311 assert(isa<X>(Val) && "cast_or_null<Ty>() argument of incompatible type!")(static_cast <bool> (isa<X>(Val) && "cast_or_null<Ty>() argument of incompatible type!"
) ? void (0) : __assert_fail ("isa<X>(Val) && \"cast_or_null<Ty>() argument of incompatible type!\""
, "/build/llvm-toolchain-snapshot-13~++20210711100610+f0393deb3367/llvm/include/llvm/Support/Casting.h"
, 311, __extension__ __PRETTY_FUNCTION__))
;
312 return cast<X>(Val);
313}
314
315template <class X, class Y>
316inline typename cast_retty<X, std::unique_ptr<Y>>::ret_type
317cast_or_null(std::unique_ptr<Y> &&Val) {
318 if (!Val)
319 return nullptr;
320 return cast<X>(std::move(Val));
321}
322
323// dyn_cast<X> - Return the argument parameter cast to the specified type. This
324// casting operator returns null if the argument is of the wrong type, so it can
325// be used to test for a type as well as cast if successful. This should be
326// used in the context of an if statement like this:
327//
328// if (const Instruction *I = dyn_cast<Instruction>(myVal)) { ... }
329//
330
331template <class X, class Y>
332LLVM_NODISCARD[[clang::warn_unused_result]] inline std::enable_if_t<
333 !is_simple_type<Y>::value, typename cast_retty<X, const Y>::ret_type>
334dyn_cast(const Y &Val) {
335 return isa<X>(Val) ? cast<X>(Val) : nullptr;
336}
337
338template <class X, class Y>
339LLVM_NODISCARD[[clang::warn_unused_result]] inline typename cast_retty<X, Y>::ret_type dyn_cast(Y &Val) {
340 return isa<X>(Val) ? cast<X>(Val) : nullptr;
2
Assuming 'Val' is a 'GlobalAddressSDNode'
3
'?' condition is true
4
Calling 'cast<llvm::GlobalAddressSDNode, llvm::SDValue>'
8
Returning from 'cast<llvm::GlobalAddressSDNode, llvm::SDValue>'
9
Returning pointer, which participates in a condition later
341}
342
343template <class X, class Y>
344LLVM_NODISCARD[[clang::warn_unused_result]] inline typename cast_retty<X, Y *>::ret_type dyn_cast(Y *Val) {
345 return isa<X>(Val) ? cast<X>(Val) : nullptr;
346}
347
348// dyn_cast_or_null<X> - Functionally identical to dyn_cast, except that a null
349// value is accepted.
350//
351template <class X, class Y>
352LLVM_NODISCARD[[clang::warn_unused_result]] inline std::enable_if_t<
353 !is_simple_type<Y>::value, typename cast_retty<X, const Y>::ret_type>
354dyn_cast_or_null(const Y &Val) {
355 return (Val && isa<X>(Val)) ? cast<X>(Val) : nullptr;
356}
357
358template <class X, class Y>
359LLVM_NODISCARD[[clang::warn_unused_result]] inline std::enable_if_t<!is_simple_type<Y>::value,
360 typename cast_retty<X, Y>::ret_type>
361dyn_cast_or_null(Y &Val) {
362 return (Val && isa<X>(Val)) ? cast<X>(Val) : nullptr;
363}
364
365template <class X, class Y>
366LLVM_NODISCARD[[clang::warn_unused_result]] inline typename cast_retty<X, Y *>::ret_type
367dyn_cast_or_null(Y *Val) {
368 return (Val && isa<X>(Val)) ? cast<X>(Val) : nullptr;
369}
370
371// unique_dyn_cast<X> - Given a unique_ptr<Y>, try to return a unique_ptr<X>,
372// taking ownership of the input pointer iff isa<X>(Val) is true. If the
373// cast is successful, From refers to nullptr on exit and the casted value
374// is returned. If the cast is unsuccessful, the function returns nullptr
375// and From is unchanged.
376template <class X, class Y>
377LLVM_NODISCARD[[clang::warn_unused_result]] inline auto unique_dyn_cast(std::unique_ptr<Y> &Val)
378 -> decltype(cast<X>(Val)) {
379 if (!isa<X>(Val))
380 return nullptr;
381 return cast<X>(std::move(Val));
382}
383
384template <class X, class Y>
385LLVM_NODISCARD[[clang::warn_unused_result]] inline auto unique_dyn_cast(std::unique_ptr<Y> &&Val) {
386 return unique_dyn_cast<X, Y>(Val);
387}
388
389// dyn_cast_or_null<X> - Functionally identical to unique_dyn_cast, except that
390// a null value is accepted.
391template <class X, class Y>
392LLVM_NODISCARD[[clang::warn_unused_result]] inline auto unique_dyn_cast_or_null(std::unique_ptr<Y> &Val)
393 -> decltype(cast<X>(Val)) {
394 if (!Val)
395 return nullptr;
396 return unique_dyn_cast<X, Y>(Val);
397}
398
399template <class X, class Y>
400LLVM_NODISCARD[[clang::warn_unused_result]] inline auto unique_dyn_cast_or_null(std::unique_ptr<Y> &&Val) {
401 return unique_dyn_cast_or_null<X, Y>(Val);
402}
403
404} // end namespace llvm
405
406#endif // LLVM_SUPPORT_CASTING_H