Bug Summary

File:llvm/lib/Target/Mips/MipsLegalizerInfo.cpp
Warning:line 359, column 25
The result of the left shift is undefined due to shifting by '4294967295', which is greater or equal to the width of type 'int'

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 MipsLegalizerInfo.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-14~++20210828111110+16086d47c0d0/build-llvm/lib/Target/Mips -resource-dir /usr/lib/llvm-14/lib/clang/14.0.0 -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/build-llvm/lib/Target/Mips -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/build-llvm/include -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/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-14/lib/clang/14.0.0/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-14~++20210828111110+16086d47c0d0/build-llvm/lib/Target/Mips -fdebug-prefix-map=/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0=. -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-2021-08-28-193554-24367-1 -x c++ /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp

/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp

→
1//===- MipsLegalizerInfo.cpp ------------------------------------*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8/// \file
9/// This file implements the targeting of the Machinelegalizer class for Mips.
10/// \todo This should be generated by TableGen.
11//===----------------------------------------------------------------------===//
12
13#include "MipsLegalizerInfo.h"
14#include "MipsTargetMachine.h"
15#include "llvm/CodeGen/GlobalISel/LegalizerHelper.h"
16#include "llvm/IR/IntrinsicsMips.h"
17
18using namespace llvm;
19
20struct TypesAndMemOps {
21 LLT ValTy;
22 LLT PtrTy;
23 unsigned MemSize;
24 bool SystemSupportsUnalignedAccess;
25};
26
27// Assumes power of 2 memory size. Subtargets that have only naturally-aligned
28// memory access need to perform additional legalization here.
29static bool isUnalignedMemmoryAccess(uint64_t MemSize, uint64_t AlignInBits) {
30 assert(isPowerOf2_64(MemSize) && "Expected power of 2 memory size")(static_cast <bool> (isPowerOf2_64(MemSize) && "Expected power of 2 memory size"
) ? void (0) : __assert_fail ("isPowerOf2_64(MemSize) && \"Expected power of 2 memory size\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 30, __extension__ __PRETTY_FUNCTION__));
31 assert(isPowerOf2_64(AlignInBits) && "Expected power of 2 align")(static_cast <bool> (isPowerOf2_64(AlignInBits) &&
"Expected power of 2 align") ? void (0) : __assert_fail ("isPowerOf2_64(AlignInBits) && \"Expected power of 2 align\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 31, __extension__ __PRETTY_FUNCTION__));
32 if (MemSize > AlignInBits)
33 return true;
34 return false;
35}
36
37static bool
38CheckTy0Ty1MemSizeAlign(const LegalityQuery &Query,
39 std::initializer_list<TypesAndMemOps> SupportedValues) {
40 unsigned QueryMemSize = Query.MMODescrs[0].MemoryTy.getSizeInBits();
41
42 // Non power of two memory access is never legal.
43 if (!isPowerOf2_64(QueryMemSize))
44 return false;
45
46 for (auto &Val : SupportedValues) {
47 if (Val.ValTy != Query.Types[0])
48 continue;
49 if (Val.PtrTy != Query.Types[1])
50 continue;
51 if (Val.MemSize != QueryMemSize)
52 continue;
53 if (!Val.SystemSupportsUnalignedAccess &&
54 isUnalignedMemmoryAccess(QueryMemSize, Query.MMODescrs[0].AlignInBits))
55 return false;
56 return true;
57 }
58 return false;
59}
60
61static bool CheckTyN(unsigned N, const LegalityQuery &Query,
62 std::initializer_list<LLT> SupportedValues) {
63 return llvm::is_contained(SupportedValues, Query.Types[N]);
64}
65
66MipsLegalizerInfo::MipsLegalizerInfo(const MipsSubtarget &ST) {
67 using namespace TargetOpcode;
68
69 const LLT s1 = LLT::scalar(1);
70 const LLT s8 = LLT::scalar(8);
71 const LLT s16 = LLT::scalar(16);
72 const LLT s32 = LLT::scalar(32);
73 const LLT s64 = LLT::scalar(64);
74 const LLT v16s8 = LLT::fixed_vector(16, 8);
75 const LLT v8s16 = LLT::fixed_vector(8, 16);
76 const LLT v4s32 = LLT::fixed_vector(4, 32);
77 const LLT v2s64 = LLT::fixed_vector(2, 64);
78 const LLT p0 = LLT::pointer(0, 32);
79
80 getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL})
81 .legalIf([=, &ST](const LegalityQuery &Query) {
82 if (CheckTyN(0, Query, {s32}))
83 return true;
84 if (ST.hasMSA() && CheckTyN(0, Query, {v16s8, v8s16, v4s32, v2s64}))
85 return true;
86 return false;
87 })
88 .clampScalar(0, s32, s32);
89
90 getActionDefinitionsBuilder({G_UADDO, G_UADDE, G_USUBO, G_USUBE, G_UMULO})
91 .lowerFor({{s32, s1}});
92
93 getActionDefinitionsBuilder(G_UMULH)
94 .legalFor({s32})
95 .maxScalar(0, s32);
96
97 // MIPS32r6 does not have alignment restrictions for memory access.
98 // For MIPS32r5 and older memory access must be naturally-aligned i.e. aligned
99 // to at least a multiple of its own size. There is however a two instruction
100 // combination that performs 4 byte unaligned access (lwr/lwl and swl/swr)
101 // therefore 4 byte load and store are legal and will use NoAlignRequirements.
102 bool NoAlignRequirements = true;
103
104 getActionDefinitionsBuilder({G_LOAD, G_STORE})
105 .legalIf([=, &ST](const LegalityQuery &Query) {
106 if (CheckTy0Ty1MemSizeAlign(
107 Query, {{s32, p0, 8, NoAlignRequirements},
108 {s32, p0, 16, ST.systemSupportsUnalignedAccess()},
109 {s32, p0, 32, NoAlignRequirements},
110 {p0, p0, 32, NoAlignRequirements},
111 {s64, p0, 64, ST.systemSupportsUnalignedAccess()}}))
112 return true;
113 if (ST.hasMSA() && CheckTy0Ty1MemSizeAlign(
114 Query, {{v16s8, p0, 128, NoAlignRequirements},
115 {v8s16, p0, 128, NoAlignRequirements},
116 {v4s32, p0, 128, NoAlignRequirements},
117 {v2s64, p0, 128, NoAlignRequirements}}))
118 return true;
119 return false;
120 })
121 // Custom lower scalar memory access, up to 8 bytes, for:
122 // - non-power-of-2 MemSizes
123 // - unaligned 2 or 8 byte MemSizes for MIPS32r5 and older
124 .customIf([=, &ST](const LegalityQuery &Query) {
125 if (!Query.Types[0].isScalar() || Query.Types[1] != p0 ||
126 Query.Types[0] == s1)
127 return false;
128
129 unsigned Size = Query.Types[0].getSizeInBits();
130 unsigned QueryMemSize = Query.MMODescrs[0].MemoryTy.getSizeInBits();
131 assert(QueryMemSize <= Size && "Scalar can't hold MemSize")(static_cast <bool> (QueryMemSize <= Size &&
"Scalar can't hold MemSize") ? void (0) : __assert_fail ("QueryMemSize <= Size && \"Scalar can't hold MemSize\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 131, __extension__ __PRETTY_FUNCTION__));
132
133 if (Size > 64 || QueryMemSize > 64)
134 return false;
135
136 if (!isPowerOf2_64(Query.MMODescrs[0].MemoryTy.getSizeInBits()))
137 return true;
138
139 if (!ST.systemSupportsUnalignedAccess() &&
140 isUnalignedMemmoryAccess(QueryMemSize,
141 Query.MMODescrs[0].AlignInBits)) {
142 assert(QueryMemSize != 32 && "4 byte load and store are legal")(static_cast <bool> (QueryMemSize != 32 && "4 byte load and store are legal"
) ? void (0) : __assert_fail ("QueryMemSize != 32 && \"4 byte load and store are legal\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 142, __extension__ __PRETTY_FUNCTION__));
143 return true;
144 }
145
146 return false;
147 })
148 .minScalar(0, s32)
149 .lower();
150
151 getActionDefinitionsBuilder(G_IMPLICIT_DEF)
152 .legalFor({s32, s64});
153
154 getActionDefinitionsBuilder(G_UNMERGE_VALUES)
155 .legalFor({{s32, s64}});
156
157 getActionDefinitionsBuilder(G_MERGE_VALUES)
158 .legalFor({{s64, s32}});
159
160 getActionDefinitionsBuilder({G_ZEXTLOAD, G_SEXTLOAD})
161 .legalForTypesWithMemDesc({{s32, p0, s8, 8},
162 {s32, p0, s16, 8}})
163 .clampScalar(0, s32, s32);
164
165 getActionDefinitionsBuilder({G_ZEXT, G_SEXT, G_ANYEXT})
166 .legalIf([](const LegalityQuery &Query) { return false; })
167 .maxScalar(0, s32);
168
169 getActionDefinitionsBuilder(G_TRUNC)
170 .legalIf([](const LegalityQuery &Query) { return false; })
171 .maxScalar(1, s32);
172
173 getActionDefinitionsBuilder(G_SELECT)
174 .legalForCartesianProduct({p0, s32, s64}, {s32})
175 .minScalar(0, s32)
176 .minScalar(1, s32);
177
178 getActionDefinitionsBuilder(G_BRCOND)
179 .legalFor({s32})
180 .minScalar(0, s32);
181
182 getActionDefinitionsBuilder(G_BRJT)
183 .legalFor({{p0, s32}});
184
185 getActionDefinitionsBuilder(G_BRINDIRECT)
186 .legalFor({p0});
187
188 getActionDefinitionsBuilder(G_PHI)
189 .legalFor({p0, s32, s64})
190 .minScalar(0, s32);
191
192 getActionDefinitionsBuilder({G_AND, G_OR, G_XOR})
193 .legalFor({s32})
194 .clampScalar(0, s32, s32);
195
196 getActionDefinitionsBuilder({G_SDIV, G_SREM, G_UDIV, G_UREM})
197 .legalIf([=, &ST](const LegalityQuery &Query) {
198 if (CheckTyN(0, Query, {s32}))
199 return true;
200 if (ST.hasMSA() && CheckTyN(0, Query, {v16s8, v8s16, v4s32, v2s64}))
201 return true;
202 return false;
203 })
204 .minScalar(0, s32)
205 .libcallFor({s64});
206
207 getActionDefinitionsBuilder({G_SHL, G_ASHR, G_LSHR})
208 .legalFor({{s32, s32}})
209 .clampScalar(1, s32, s32)
210 .clampScalar(0, s32, s32);
211
212 getActionDefinitionsBuilder(G_ICMP)
213 .legalForCartesianProduct({s32}, {s32, p0})
214 .clampScalar(1, s32, s32)
215 .minScalar(0, s32);
216
217 getActionDefinitionsBuilder(G_CONSTANT)
218 .legalFor({s32})
219 .clampScalar(0, s32, s32);
220
221 getActionDefinitionsBuilder({G_PTR_ADD, G_INTTOPTR})
222 .legalFor({{p0, s32}});
223
224 getActionDefinitionsBuilder(G_PTRTOINT)
225 .legalFor({{s32, p0}});
226
227 getActionDefinitionsBuilder(G_FRAME_INDEX)
228 .legalFor({p0});
229
230 getActionDefinitionsBuilder({G_GLOBAL_VALUE, G_JUMP_TABLE})
231 .legalFor({p0});
232
233 getActionDefinitionsBuilder(G_DYN_STACKALLOC)
234 .lowerFor({{p0, s32}});
235
236 getActionDefinitionsBuilder(G_VASTART)
237 .legalFor({p0});
238
239 getActionDefinitionsBuilder(G_BSWAP)
240 .legalIf([=, &ST](const LegalityQuery &Query) {
241 if (ST.hasMips32r2() && CheckTyN(0, Query, {s32}))
242 return true;
243 return false;
244 })
245 .lowerIf([=, &ST](const LegalityQuery &Query) {
246 if (!ST.hasMips32r2() && CheckTyN(0, Query, {s32}))
247 return true;
248 return false;
249 })
250 .maxScalar(0, s32);
251
252 getActionDefinitionsBuilder(G_BITREVERSE)
253 .lowerFor({s32})
254 .maxScalar(0, s32);
255
256 getActionDefinitionsBuilder(G_CTLZ)
257 .legalFor({{s32, s32}})
258 .maxScalar(0, s32)
259 .maxScalar(1, s32);
260 getActionDefinitionsBuilder(G_CTLZ_ZERO_UNDEF)
261 .lowerFor({{s32, s32}});
262
263 getActionDefinitionsBuilder(G_CTTZ)
264 .lowerFor({{s32, s32}})
265 .maxScalar(0, s32)
266 .maxScalar(1, s32);
267 getActionDefinitionsBuilder(G_CTTZ_ZERO_UNDEF)
268 .lowerFor({{s32, s32}, {s64, s64}});
269
270 getActionDefinitionsBuilder(G_CTPOP)
271 .lowerFor({{s32, s32}})
272 .clampScalar(0, s32, s32)
273 .clampScalar(1, s32, s32);
274
275 // FP instructions
276 getActionDefinitionsBuilder(G_FCONSTANT)
277 .legalFor({s32, s64});
278
279 getActionDefinitionsBuilder({G_FADD, G_FSUB, G_FMUL, G_FDIV, G_FABS, G_FSQRT})
280 .legalIf([=, &ST](const LegalityQuery &Query) {
281 if (CheckTyN(0, Query, {s32, s64}))
282 return true;
283 if (ST.hasMSA() && CheckTyN(0, Query, {v16s8, v8s16, v4s32, v2s64}))
284 return true;
285 return false;
286 });
287
288 getActionDefinitionsBuilder(G_FCMP)
289 .legalFor({{s32, s32}, {s32, s64}})
290 .minScalar(0, s32);
291
292 getActionDefinitionsBuilder({G_FCEIL, G_FFLOOR})
293 .libcallFor({s32, s64});
294
295 getActionDefinitionsBuilder(G_FPEXT)
296 .legalFor({{s64, s32}});
297
298 getActionDefinitionsBuilder(G_FPTRUNC)
299 .legalFor({{s32, s64}});
300
301 // FP to int conversion instructions
302 getActionDefinitionsBuilder(G_FPTOSI)
303 .legalForCartesianProduct({s32}, {s64, s32})
304 .libcallForCartesianProduct({s64}, {s64, s32})
305 .minScalar(0, s32);
306
307 getActionDefinitionsBuilder(G_FPTOUI)
308 .libcallForCartesianProduct({s64}, {s64, s32})
309 .lowerForCartesianProduct({s32}, {s64, s32})
310 .minScalar(0, s32);
311
312 // Int to FP conversion instructions
313 getActionDefinitionsBuilder(G_SITOFP)
314 .legalForCartesianProduct({s64, s32}, {s32})
315 .libcallForCartesianProduct({s64, s32}, {s64})
316 .minScalar(1, s32);
317
318 getActionDefinitionsBuilder(G_UITOFP)
319 .libcallForCartesianProduct({s64, s32}, {s64})
320 .customForCartesianProduct({s64, s32}, {s32})
321 .minScalar(1, s32);
322
323 getActionDefinitionsBuilder(G_SEXT_INREG).lower();
324
325 getActionDefinitionsBuilder({G_MEMCPY, G_MEMMOVE, G_MEMSET}).libcall();
326
327 getLegacyLegalizerInfo().computeTables();
328 verify(*ST.getInstrInfo());
329}
330
331bool MipsLegalizerInfo::legalizeCustom(LegalizerHelper &Helper,
332 MachineInstr &MI) const {
333 using namespace TargetOpcode;
334
335 MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
336 MachineRegisterInfo &MRI = *MIRBuilder.getMRI();
337
338 const LLT s32 = LLT::scalar(32);
339 const LLT s64 = LLT::scalar(64);
340
341 switch (MI.getOpcode()) {
1
Control jumps to 'case G_STORE:' at line 343
→
342 case G_LOAD:
343 case G_STORE: {
344 unsigned MemSize = (**MI.memoperands_begin()).getSize();
345 Register Val = MI.getOperand(0).getReg();
346 unsigned Size = MRI.getType(Val).getSizeInBits();
347
348 MachineMemOperand *MMOBase = *MI.memoperands_begin();
349
350 assert(MemSize <= 8 && "MemSize is too large")(static_cast <bool> (MemSize <= 8 && "MemSize is too large"
) ? void (0) : __assert_fail ("MemSize <= 8 && \"MemSize is too large\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 350, __extension__ __PRETTY_FUNCTION__));
2
←
Assuming 'MemSize' is <= 8
→
3
←
'?' condition is true
→
351 assert(Size <= 64 && "Scalar size is too large")(static_cast <bool> (Size <= 64 && "Scalar size is too large"
) ? void (0) : __assert_fail ("Size <= 64 && \"Scalar size is too large\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 351, __extension__ __PRETTY_FUNCTION__));
4
←
Assuming 'Size' is <= 64
→
5
←
'?' condition is true
→
352
353 // Split MemSize into two, P2HalfMemSize is largest power of two smaller
354 // then MemSize. e.g. 8 = 4 + 4 , 6 = 4 + 2, 3 = 2 + 1.
355 unsigned P2HalfMemSize, RemMemSize;
356 if (isPowerOf2_64(MemSize)) {
6
←
Calling 'isPowerOf2_64'
→
9
←
Returning from 'isPowerOf2_64'
→
10
←
Taking false branch
→
357 P2HalfMemSize = RemMemSize = MemSize / 2;
358 } else {
359 P2HalfMemSize = 1 << Log2_32(MemSize);
11
←
Calling 'Log2_32'
→
13
←
Returning from 'Log2_32'
→
14
←
The result of the left shift is undefined due to shifting by '4294967295', which is greater or equal to the width of type 'int'
360 RemMemSize = MemSize - P2HalfMemSize;
361 }
362
363 Register BaseAddr = MI.getOperand(1).getReg();
364 LLT PtrTy = MRI.getType(BaseAddr);
365 MachineFunction &MF = MIRBuilder.getMF();
366
367 auto P2HalfMemOp = MF.getMachineMemOperand(MMOBase, 0, P2HalfMemSize);
368 auto RemMemOp = MF.getMachineMemOperand(MMOBase, P2HalfMemSize, RemMemSize);
369
370 if (MI.getOpcode() == G_STORE) {
371 // Widen Val to s32 or s64 in order to create legal G_LSHR or G_UNMERGE.
372 if (Size < 32)
373 Val = MIRBuilder.buildAnyExt(s32, Val).getReg(0);
374 if (Size > 32 && Size < 64)
375 Val = MIRBuilder.buildAnyExt(s64, Val).getReg(0);
376
377 auto C_P2HalfMemSize = MIRBuilder.buildConstant(s32, P2HalfMemSize);
378 auto Addr = MIRBuilder.buildPtrAdd(PtrTy, BaseAddr, C_P2HalfMemSize);
379
380 if (MI.getOpcode() == G_STORE && MemSize <= 4) {
381 MIRBuilder.buildStore(Val, BaseAddr, *P2HalfMemOp);
382 auto C_P2Half_InBits = MIRBuilder.buildConstant(s32, P2HalfMemSize * 8);
383 auto Shift = MIRBuilder.buildLShr(s32, Val, C_P2Half_InBits);
384 MIRBuilder.buildStore(Shift, Addr, *RemMemOp);
385 } else {
386 auto Unmerge = MIRBuilder.buildUnmerge(s32, Val);
387 MIRBuilder.buildStore(Unmerge.getReg(0), BaseAddr, *P2HalfMemOp);
388 MIRBuilder.buildStore(Unmerge.getReg(1), Addr, *RemMemOp);
389 }
390 }
391
392 if (MI.getOpcode() == G_LOAD) {
393
394 if (MemSize <= 4) {
395 // This is anyextending load, use 4 byte lwr/lwl.
396 auto *Load4MMO = MF.getMachineMemOperand(MMOBase, 0, 4);
397
398 if (Size == 32)
399 MIRBuilder.buildLoad(Val, BaseAddr, *Load4MMO);
400 else {
401 auto Load = MIRBuilder.buildLoad(s32, BaseAddr, *Load4MMO);
402 MIRBuilder.buildTrunc(Val, Load.getReg(0));
403 }
404
405 } else {
406 auto C_P2HalfMemSize = MIRBuilder.buildConstant(s32, P2HalfMemSize);
407 auto Addr = MIRBuilder.buildPtrAdd(PtrTy, BaseAddr, C_P2HalfMemSize);
408
409 auto Load_P2Half = MIRBuilder.buildLoad(s32, BaseAddr, *P2HalfMemOp);
410 auto Load_Rem = MIRBuilder.buildLoad(s32, Addr, *RemMemOp);
411
412 if (Size == 64)
413 MIRBuilder.buildMerge(Val, {Load_P2Half, Load_Rem});
414 else {
415 auto Merge = MIRBuilder.buildMerge(s64, {Load_P2Half, Load_Rem});
416 MIRBuilder.buildTrunc(Val, Merge);
417 }
418 }
419 }
420 MI.eraseFromParent();
421 break;
422 }
423 case G_UITOFP: {
424 Register Dst = MI.getOperand(0).getReg();
425 Register Src = MI.getOperand(1).getReg();
426 LLT DstTy = MRI.getType(Dst);
427 LLT SrcTy = MRI.getType(Src);
428
429 if (SrcTy != s32)
430 return false;
431 if (DstTy != s32 && DstTy != s64)
432 return false;
433
434 // Let 0xABCDEFGH be given unsigned in MI.getOperand(1). First let's convert
435 // unsigned to double. Mantissa has 52 bits so we use following trick:
436 // First make floating point bit mask 0x43300000ABCDEFGH.
437 // Mask represents 2^52 * 0x1.00000ABCDEFGH i.e. 0x100000ABCDEFGH.0 .
438 // Next, subtract 2^52 * 0x1.0000000000000 i.e. 0x10000000000000.0 from it.
439 // Done. Trunc double to float if needed.
440
441 auto C_HiMask = MIRBuilder.buildConstant(s32, UINT32_C(0x43300000)0x43300000U);
442 auto Bitcast = MIRBuilder.buildMerge(s64, {Src, C_HiMask.getReg(0)});
443
444 MachineInstrBuilder TwoP52FP = MIRBuilder.buildFConstant(
445 s64, BitsToDouble(UINT64_C(0x4330000000000000)0x4330000000000000UL));
446
447 if (DstTy == s64)
448 MIRBuilder.buildFSub(Dst, Bitcast, TwoP52FP);
449 else {
450 MachineInstrBuilder ResF64 = MIRBuilder.buildFSub(s64, Bitcast, TwoP52FP);
451 MIRBuilder.buildFPTrunc(Dst, ResF64);
452 }
453
454 MI.eraseFromParent();
455 break;
456 }
457 default:
458 return false;
459 }
460
461 return true;
462}
463
464static bool SelectMSA3OpIntrinsic(MachineInstr &MI, unsigned Opcode,
465 MachineIRBuilder &MIRBuilder,
466 const MipsSubtarget &ST) {
467 assert(ST.hasMSA() && "MSA intrinsic not supported on target without MSA.")(static_cast <bool> (ST.hasMSA() && "MSA intrinsic not supported on target without MSA."
) ? void (0) : __assert_fail ("ST.hasMSA() && \"MSA intrinsic not supported on target without MSA.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 467, __extension__ __PRETTY_FUNCTION__));
468 if (!MIRBuilder.buildInstr(Opcode)
469 .add(MI.getOperand(0))
470 .add(MI.getOperand(2))
471 .add(MI.getOperand(3))
472 .constrainAllUses(MIRBuilder.getTII(), *ST.getRegisterInfo(),
473 *ST.getRegBankInfo()))
474 return false;
475 MI.eraseFromParent();
476 return true;
477}
478
479static bool MSA3OpIntrinsicToGeneric(MachineInstr &MI, unsigned Opcode,
480 MachineIRBuilder &MIRBuilder,
481 const MipsSubtarget &ST) {
482 assert(ST.hasMSA() && "MSA intrinsic not supported on target without MSA.")(static_cast <bool> (ST.hasMSA() && "MSA intrinsic not supported on target without MSA."
) ? void (0) : __assert_fail ("ST.hasMSA() && \"MSA intrinsic not supported on target without MSA.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 482, __extension__ __PRETTY_FUNCTION__));
483 MIRBuilder.buildInstr(Opcode)
484 .add(MI.getOperand(0))
485 .add(MI.getOperand(2))
486 .add(MI.getOperand(3));
487 MI.eraseFromParent();
488 return true;
489}
490
491static bool MSA2OpIntrinsicToGeneric(MachineInstr &MI, unsigned Opcode,
492 MachineIRBuilder &MIRBuilder,
493 const MipsSubtarget &ST) {
494 assert(ST.hasMSA() && "MSA intrinsic not supported on target without MSA.")(static_cast <bool> (ST.hasMSA() && "MSA intrinsic not supported on target without MSA."
) ? void (0) : __assert_fail ("ST.hasMSA() && \"MSA intrinsic not supported on target without MSA.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Target/Mips/MipsLegalizerInfo.cpp"
, 494, __extension__ __PRETTY_FUNCTION__));
495 MIRBuilder.buildInstr(Opcode)
496 .add(MI.getOperand(0))
497 .add(MI.getOperand(2));
498 MI.eraseFromParent();
499 return true;
500}
501
502bool MipsLegalizerInfo::legalizeIntrinsic(LegalizerHelper &Helper,
503 MachineInstr &MI) const {
504 MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
505 const MipsSubtarget &ST =
506 static_cast<const MipsSubtarget &>(MI.getMF()->getSubtarget());
507 const MipsInstrInfo &TII = *ST.getInstrInfo();
508 const MipsRegisterInfo &TRI = *ST.getRegisterInfo();
509 const RegisterBankInfo &RBI = *ST.getRegBankInfo();
510
511 switch (MI.getIntrinsicID()) {
512 case Intrinsic::trap: {
513 MachineInstr *Trap = MIRBuilder.buildInstr(Mips::TRAP);
514 MI.eraseFromParent();
515 return constrainSelectedInstRegOperands(*Trap, TII, TRI, RBI);
516 }
517 case Intrinsic::vacopy: {
518 MachinePointerInfo MPO;
519 LLT PtrTy = LLT::pointer(0, 32);
520 auto Tmp =
521 MIRBuilder.buildLoad(PtrTy, MI.getOperand(2),
522 *MI.getMF()->getMachineMemOperand(
523 MPO, MachineMemOperand::MOLoad, PtrTy, Align(4)));
524 MIRBuilder.buildStore(Tmp, MI.getOperand(1),
525 *MI.getMF()->getMachineMemOperand(
526 MPO, MachineMemOperand::MOStore, PtrTy, Align(4)));
527 MI.eraseFromParent();
528 return true;
529 }
530 case Intrinsic::mips_addv_b:
531 case Intrinsic::mips_addv_h:
532 case Intrinsic::mips_addv_w:
533 case Intrinsic::mips_addv_d:
534 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_ADD, MIRBuilder, ST);
535 case Intrinsic::mips_addvi_b:
536 return SelectMSA3OpIntrinsic(MI, Mips::ADDVI_B, MIRBuilder, ST);
537 case Intrinsic::mips_addvi_h:
538 return SelectMSA3OpIntrinsic(MI, Mips::ADDVI_H, MIRBuilder, ST);
539 case Intrinsic::mips_addvi_w:
540 return SelectMSA3OpIntrinsic(MI, Mips::ADDVI_W, MIRBuilder, ST);
541 case Intrinsic::mips_addvi_d:
542 return SelectMSA3OpIntrinsic(MI, Mips::ADDVI_D, MIRBuilder, ST);
543 case Intrinsic::mips_subv_b:
544 case Intrinsic::mips_subv_h:
545 case Intrinsic::mips_subv_w:
546 case Intrinsic::mips_subv_d:
547 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_SUB, MIRBuilder, ST);
548 case Intrinsic::mips_subvi_b:
549 return SelectMSA3OpIntrinsic(MI, Mips::SUBVI_B, MIRBuilder, ST);
550 case Intrinsic::mips_subvi_h:
551 return SelectMSA3OpIntrinsic(MI, Mips::SUBVI_H, MIRBuilder, ST);
552 case Intrinsic::mips_subvi_w:
553 return SelectMSA3OpIntrinsic(MI, Mips::SUBVI_W, MIRBuilder, ST);
554 case Intrinsic::mips_subvi_d:
555 return SelectMSA3OpIntrinsic(MI, Mips::SUBVI_D, MIRBuilder, ST);
556 case Intrinsic::mips_mulv_b:
557 case Intrinsic::mips_mulv_h:
558 case Intrinsic::mips_mulv_w:
559 case Intrinsic::mips_mulv_d:
560 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_MUL, MIRBuilder, ST);
561 case Intrinsic::mips_div_s_b:
562 case Intrinsic::mips_div_s_h:
563 case Intrinsic::mips_div_s_w:
564 case Intrinsic::mips_div_s_d:
565 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_SDIV, MIRBuilder, ST);
566 case Intrinsic::mips_mod_s_b:
567 case Intrinsic::mips_mod_s_h:
568 case Intrinsic::mips_mod_s_w:
569 case Intrinsic::mips_mod_s_d:
570 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_SREM, MIRBuilder, ST);
571 case Intrinsic::mips_div_u_b:
572 case Intrinsic::mips_div_u_h:
573 case Intrinsic::mips_div_u_w:
574 case Intrinsic::mips_div_u_d:
575 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_UDIV, MIRBuilder, ST);
576 case Intrinsic::mips_mod_u_b:
577 case Intrinsic::mips_mod_u_h:
578 case Intrinsic::mips_mod_u_w:
579 case Intrinsic::mips_mod_u_d:
580 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_UREM, MIRBuilder, ST);
581 case Intrinsic::mips_fadd_w:
582 case Intrinsic::mips_fadd_d:
583 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_FADD, MIRBuilder, ST);
584 case Intrinsic::mips_fsub_w:
585 case Intrinsic::mips_fsub_d:
586 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_FSUB, MIRBuilder, ST);
587 case Intrinsic::mips_fmul_w:
588 case Intrinsic::mips_fmul_d:
589 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_FMUL, MIRBuilder, ST);
590 case Intrinsic::mips_fdiv_w:
591 case Intrinsic::mips_fdiv_d:
592 return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_FDIV, MIRBuilder, ST);
593 case Intrinsic::mips_fmax_a_w:
594 return SelectMSA3OpIntrinsic(MI, Mips::FMAX_A_W, MIRBuilder, ST);
595 case Intrinsic::mips_fmax_a_d:
596 return SelectMSA3OpIntrinsic(MI, Mips::FMAX_A_D, MIRBuilder, ST);
597 case Intrinsic::mips_fsqrt_w:
598 return MSA2OpIntrinsicToGeneric(MI, TargetOpcode::G_FSQRT, MIRBuilder, ST);
599 case Intrinsic::mips_fsqrt_d:
600 return MSA2OpIntrinsicToGeneric(MI, TargetOpcode::G_FSQRT, MIRBuilder, ST);
601 default:
602 break;
603 }
604 return true;
605}
←

/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h

1//===-- llvm/Support/MathExtras.h - Useful math functions -------*- 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 contains some functions that are useful for math stuff.
10//
11//===----------------------------------------------------------------------===//
12
13#ifndef LLVM_SUPPORT_MATHEXTRAS_H
14#define LLVM_SUPPORT_MATHEXTRAS_H
15
16#include "llvm/Support/Compiler.h"
17#include <cassert>
18#include <climits>
19#include <cmath>
20#include <cstdint>
21#include <cstring>
22#include <limits>
23#include <type_traits>
24
25#ifdef __ANDROID_NDK__
26#include <android/api-level.h>
27#endif
28
29#ifdef _MSC_VER
30// Declare these intrinsics manually rather including intrin.h. It's very
31// expensive, and MathExtras.h is popular.
32// #include <intrin.h>
33extern "C" {
34unsigned char _BitScanForward(unsigned long *_Index, unsigned long _Mask);
35unsigned char _BitScanForward64(unsigned long *_Index, unsigned __int64 _Mask);
36unsigned char _BitScanReverse(unsigned long *_Index, unsigned long _Mask);
37unsigned char _BitScanReverse64(unsigned long *_Index, unsigned __int64 _Mask);
38}
39#endif
40
41namespace llvm {
42
43/// The behavior an operation has on an input of 0.
44enum ZeroBehavior {
45 /// The returned value is undefined.
46 ZB_Undefined,
47 /// The returned value is numeric_limits<T>::max()
48 ZB_Max,
49 /// The returned value is numeric_limits<T>::digits
50 ZB_Width
51};
52
53/// Mathematical constants.
54namespace numbers {
55// TODO: Track C++20 std::numbers.
56// TODO: Favor using the hexadecimal FP constants (requires C++17).
57constexpr double e = 2.7182818284590452354, // (0x1.5bf0a8b145749P+1) https://oeis.org/A001113
58 egamma = .57721566490153286061, // (0x1.2788cfc6fb619P-1) https://oeis.org/A001620
59 ln2 = .69314718055994530942, // (0x1.62e42fefa39efP-1) https://oeis.org/A002162
60 ln10 = 2.3025850929940456840, // (0x1.24bb1bbb55516P+1) https://oeis.org/A002392
61 log2e = 1.4426950408889634074, // (0x1.71547652b82feP+0)
62 log10e = .43429448190325182765, // (0x1.bcb7b1526e50eP-2)
63 pi = 3.1415926535897932385, // (0x1.921fb54442d18P+1) https://oeis.org/A000796
64 inv_pi = .31830988618379067154, // (0x1.45f306bc9c883P-2) https://oeis.org/A049541
65 sqrtpi = 1.7724538509055160273, // (0x1.c5bf891b4ef6bP+0) https://oeis.org/A002161
66 inv_sqrtpi = .56418958354775628695, // (0x1.20dd750429b6dP-1) https://oeis.org/A087197
67 sqrt2 = 1.4142135623730950488, // (0x1.6a09e667f3bcdP+0) https://oeis.org/A00219
68 inv_sqrt2 = .70710678118654752440, // (0x1.6a09e667f3bcdP-1)
69 sqrt3 = 1.7320508075688772935, // (0x1.bb67ae8584caaP+0) https://oeis.org/A002194
70 inv_sqrt3 = .57735026918962576451, // (0x1.279a74590331cP-1)
71 phi = 1.6180339887498948482; // (0x1.9e3779b97f4a8P+0) https://oeis.org/A001622
72constexpr float ef = 2.71828183F, // (0x1.5bf0a8P+1) https://oeis.org/A001113
73 egammaf = .577215665F, // (0x1.2788d0P-1) https://oeis.org/A001620
74 ln2f = .693147181F, // (0x1.62e430P-1) https://oeis.org/A002162
75 ln10f = 2.30258509F, // (0x1.26bb1cP+1) https://oeis.org/A002392
76 log2ef = 1.44269504F, // (0x1.715476P+0)
77 log10ef = .434294482F, // (0x1.bcb7b2P-2)
78 pif = 3.14159265F, // (0x1.921fb6P+1) https://oeis.org/A000796
79 inv_pif = .318309886F, // (0x1.45f306P-2) https://oeis.org/A049541
80 sqrtpif = 1.77245385F, // (0x1.c5bf8aP+0) https://oeis.org/A002161
81 inv_sqrtpif = .564189584F, // (0x1.20dd76P-1) https://oeis.org/A087197
82 sqrt2f = 1.41421356F, // (0x1.6a09e6P+0) https://oeis.org/A002193
83 inv_sqrt2f = .707106781F, // (0x1.6a09e6P-1)
84 sqrt3f = 1.73205081F, // (0x1.bb67aeP+0) https://oeis.org/A002194
85 inv_sqrt3f = .577350269F, // (0x1.279a74P-1)
86 phif = 1.61803399F; // (0x1.9e377aP+0) https://oeis.org/A001622
87} // namespace numbers
88
89namespace detail {
90template <typename T, std::size_t SizeOfT> struct TrailingZerosCounter {
91 static unsigned count(T Val, ZeroBehavior) {
92 if (!Val)
93 return std::numeric_limits<T>::digits;
94 if (Val & 0x1)
95 return 0;
96
97 // Bisection method.
98 unsigned ZeroBits = 0;
99 T Shift = std::numeric_limits<T>::digits >> 1;
100 T Mask = std::numeric_limits<T>::max() >> Shift;
101 while (Shift) {
102 if ((Val & Mask) == 0) {
103 Val >>= Shift;
104 ZeroBits |= Shift;
105 }
106 Shift >>= 1;
107 Mask >>= Shift;
108 }
109 return ZeroBits;
110 }
111};
112
113#if defined(__GNUC__4) || defined(_MSC_VER)
114template <typename T> struct TrailingZerosCounter<T, 4> {
115 static unsigned count(T Val, ZeroBehavior ZB) {
116 if (ZB != ZB_Undefined && Val == 0)
117 return 32;
118
119#if __has_builtin(__builtin_ctz)1 || defined(__GNUC__4)
120 return __builtin_ctz(Val);
121#elif defined(_MSC_VER)
122 unsigned long Index;
123 _BitScanForward(&Index, Val);
124 return Index;
125#endif
126 }
127};
128
129#if !defined(_MSC_VER) || defined(_M_X64)
130template <typename T> struct TrailingZerosCounter<T, 8> {
131 static unsigned count(T Val, ZeroBehavior ZB) {
132 if (ZB != ZB_Undefined && Val == 0)
133 return 64;
134
135#if __has_builtin(__builtin_ctzll)1 || defined(__GNUC__4)
136 return __builtin_ctzll(Val);
137#elif defined(_MSC_VER)
138 unsigned long Index;
139 _BitScanForward64(&Index, Val);
140 return Index;
141#endif
142 }
143};
144#endif
145#endif
146} // namespace detail
147
148/// Count number of 0's from the least significant bit to the most
149/// stopping at the first 1.
150///
151/// Only unsigned integral types are allowed.
152///
153/// \param ZB the behavior on an input of 0. Only ZB_Width and ZB_Undefined are
154/// valid arguments.
155template <typename T>
156unsigned countTrailingZeros(T Val, ZeroBehavior ZB = ZB_Width) {
157 static_assert(std::numeric_limits<T>::is_integer &&
158 !std::numeric_limits<T>::is_signed,
159 "Only unsigned integral types are allowed.");
160 return llvm::detail::TrailingZerosCounter<T, sizeof(T)>::count(Val, ZB);
161}
162
163namespace detail {
164template <typename T, std::size_t SizeOfT> struct LeadingZerosCounter {
165 static unsigned count(T Val, ZeroBehavior) {
166 if (!Val)
167 return std::numeric_limits<T>::digits;
168
169 // Bisection method.
170 unsigned ZeroBits = 0;
171 for (T Shift = std::numeric_limits<T>::digits >> 1; Shift; Shift >>= 1) {
172 T Tmp = Val >> Shift;
173 if (Tmp)
174 Val = Tmp;
175 else
176 ZeroBits |= Shift;
177 }
178 return ZeroBits;
179 }
180};
181
182#if defined(__GNUC__4) || defined(_MSC_VER)
183template <typename T> struct LeadingZerosCounter<T, 4> {
184 static unsigned count(T Val, ZeroBehavior ZB) {
185 if (ZB != ZB_Undefined && Val == 0)
186 return 32;
187
188#if __has_builtin(__builtin_clz)1 || defined(__GNUC__4)
189 return __builtin_clz(Val);
190#elif defined(_MSC_VER)
191 unsigned long Index;
192 _BitScanReverse(&Index, Val);
193 return Index ^ 31;
194#endif
195 }
196};
197
198#if !defined(_MSC_VER) || defined(_M_X64)
199template <typename T> struct LeadingZerosCounter<T, 8> {
200 static unsigned count(T Val, ZeroBehavior ZB) {
201 if (ZB != ZB_Undefined && Val == 0)
202 return 64;
203
204#if __has_builtin(__builtin_clzll)1 || defined(__GNUC__4)
205 return __builtin_clzll(Val);
206#elif defined(_MSC_VER)
207 unsigned long Index;
208 _BitScanReverse64(&Index, Val);
209 return Index ^ 63;
210#endif
211 }
212};
213#endif
214#endif
215} // namespace detail
216
217/// Count number of 0's from the most significant bit to the least
218/// stopping at the first 1.
219///
220/// Only unsigned integral types are allowed.
221///
222/// \param ZB the behavior on an input of 0. Only ZB_Width and ZB_Undefined are
223/// valid arguments.
224template <typename T>
225unsigned countLeadingZeros(T Val, ZeroBehavior ZB = ZB_Width) {
226 static_assert(std::numeric_limits<T>::is_integer &&
227 !std::numeric_limits<T>::is_signed,
228 "Only unsigned integral types are allowed.");
229 return llvm::detail::LeadingZerosCounter<T, sizeof(T)>::count(Val, ZB);
230}
231
232/// Get the index of the first set bit starting from the least
233/// significant bit.
234///
235/// Only unsigned integral types are allowed.
236///
237/// \param ZB the behavior on an input of 0. Only ZB_Max and ZB_Undefined are
238/// valid arguments.
239template <typename T> T findFirstSet(T Val, ZeroBehavior ZB = ZB_Max) {
240 if (ZB == ZB_Max && Val == 0)
241 return std::numeric_limits<T>::max();
242
243 return countTrailingZeros(Val, ZB_Undefined);
244}
245
246/// Create a bitmask with the N right-most bits set to 1, and all other
247/// bits set to 0. Only unsigned types are allowed.
248template <typename T> T maskTrailingOnes(unsigned N) {
249 static_assert(std::is_unsigned<T>::value, "Invalid type!");
250 const unsigned Bits = CHAR_BIT8 * sizeof(T);
251 assert(N <= Bits && "Invalid bit index")(static_cast <bool> (N <= Bits && "Invalid bit index"
) ? void (0) : __assert_fail ("N <= Bits && \"Invalid bit index\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 251, __extension__ __PRETTY_FUNCTION__));
252 return N == 0 ? 0 : (T(-1) >> (Bits - N));
253}
254
255/// Create a bitmask with the N left-most bits set to 1, and all other
256/// bits set to 0. Only unsigned types are allowed.
257template <typename T> T maskLeadingOnes(unsigned N) {
258 return ~maskTrailingOnes<T>(CHAR_BIT8 * sizeof(T) - N);
259}
260
261/// Create a bitmask with the N right-most bits set to 0, and all other
262/// bits set to 1. Only unsigned types are allowed.
263template <typename T> T maskTrailingZeros(unsigned N) {
264 return maskLeadingOnes<T>(CHAR_BIT8 * sizeof(T) - N);
265}
266
267/// Create a bitmask with the N left-most bits set to 0, and all other
268/// bits set to 1. Only unsigned types are allowed.
269template <typename T> T maskLeadingZeros(unsigned N) {
270 return maskTrailingOnes<T>(CHAR_BIT8 * sizeof(T) - N);
271}
272
273/// Get the index of the last set bit starting from the least
274/// significant bit.
275///
276/// Only unsigned integral types are allowed.
277///
278/// \param ZB the behavior on an input of 0. Only ZB_Max and ZB_Undefined are
279/// valid arguments.
280template <typename T> T findLastSet(T Val, ZeroBehavior ZB = ZB_Max) {
281 if (ZB == ZB_Max && Val == 0)
282 return std::numeric_limits<T>::max();
283
284 // Use ^ instead of - because both gcc and llvm can remove the associated ^
285 // in the __builtin_clz intrinsic on x86.
286 return countLeadingZeros(Val, ZB_Undefined) ^
287 (std::numeric_limits<T>::digits - 1);
288}
289
290/// Macro compressed bit reversal table for 256 bits.
291///
292/// http://graphics.stanford.edu/~seander/bithacks.html#BitReverseTable
293static const unsigned char BitReverseTable256[256] = {
294#define R2(n) n, n + 2 * 64, n + 1 * 64, n + 3 * 64
295#define R4(n) R2(n), R2(n + 2 * 16), R2(n + 1 * 16), R2(n + 3 * 16)
296#define R6(n) R4(n), R4(n + 2 * 4), R4(n + 1 * 4), R4(n + 3 * 4)
297 R6(0), R6(2), R6(1), R6(3)
298#undef R2
299#undef R4
300#undef R6
301};
302
303/// Reverse the bits in \p Val.
304template <typename T>
305T reverseBits(T Val) {
306 unsigned char in[sizeof(Val)];
307 unsigned char out[sizeof(Val)];
308 std::memcpy(in, &Val, sizeof(Val));
309 for (unsigned i = 0; i < sizeof(Val); ++i)
310 out[(sizeof(Val) - i) - 1] = BitReverseTable256[in[i]];
311 std::memcpy(&Val, out, sizeof(Val));
312 return Val;
313}
314
315#if __has_builtin(__builtin_bitreverse8)1
316template<>
317inline uint8_t reverseBits<uint8_t>(uint8_t Val) {
318 return __builtin_bitreverse8(Val);
319}
320#endif
321
322#if __has_builtin(__builtin_bitreverse16)1
323template<>
324inline uint16_t reverseBits<uint16_t>(uint16_t Val) {
325 return __builtin_bitreverse16(Val);
326}
327#endif
328
329#if __has_builtin(__builtin_bitreverse32)1
330template<>
331inline uint32_t reverseBits<uint32_t>(uint32_t Val) {
332 return __builtin_bitreverse32(Val);
333}
334#endif
335
336#if __has_builtin(__builtin_bitreverse64)1
337template<>
338inline uint64_t reverseBits<uint64_t>(uint64_t Val) {
339 return __builtin_bitreverse64(Val);
340}
341#endif
342
343// NOTE: The following support functions use the _32/_64 extensions instead of
344// type overloading so that signed and unsigned integers can be used without
345// ambiguity.
346
347/// Return the high 32 bits of a 64 bit value.
348constexpr inline uint32_t Hi_32(uint64_t Value) {
349 return static_cast<uint32_t>(Value >> 32);
350}
351
352/// Return the low 32 bits of a 64 bit value.
353constexpr inline uint32_t Lo_32(uint64_t Value) {
354 return static_cast<uint32_t>(Value);
355}
356
357/// Make a 64-bit integer from a high / low pair of 32-bit integers.
358constexpr inline uint64_t Make_64(uint32_t High, uint32_t Low) {
359 return ((uint64_t)High << 32) | (uint64_t)Low;
360}
361
362/// Checks if an integer fits into the given bit width.
363template <unsigned N> constexpr inline bool isInt(int64_t x) {
364 return N >= 64 || (-(INT64_C(1)1L<<(N-1)) <= x && x < (INT64_C(1)1L<<(N-1)));
365}
366// Template specializations to get better code for common cases.
367template <> constexpr inline bool isInt<8>(int64_t x) {
368 return static_cast<int8_t>(x) == x;
369}
370template <> constexpr inline bool isInt<16>(int64_t x) {
371 return static_cast<int16_t>(x) == x;
372}
373template <> constexpr inline bool isInt<32>(int64_t x) {
374 return static_cast<int32_t>(x) == x;
375}
376
377/// Checks if a signed integer is an N bit number shifted left by S.
378template <unsigned N, unsigned S>
379constexpr inline bool isShiftedInt(int64_t x) {
380 static_assert(
381 N > 0, "isShiftedInt<0> doesn't make sense (refers to a 0-bit number.");
382 static_assert(N + S <= 64, "isShiftedInt<N, S> with N + S > 64 is too wide.");
383 return isInt<N + S>(x) && (x % (UINT64_C(1)1UL << S) == 0);
384}
385
386/// Checks if an unsigned integer fits into the given bit width.
387///
388/// This is written as two functions rather than as simply
389///
390/// return N >= 64 || X < (UINT64_C(1) << N);
391///
392/// to keep MSVC from (incorrectly) warning on isUInt<64> that we're shifting
393/// left too many places.
394template <unsigned N>
395constexpr inline std::enable_if_t<(N < 64), bool> isUInt(uint64_t X) {
396 static_assert(N > 0, "isUInt<0> doesn't make sense");
397 return X < (UINT64_C(1)1UL << (N));
398}
399template <unsigned N>
400constexpr inline std::enable_if_t<N >= 64, bool> isUInt(uint64_t) {
401 return true;
402}
403
404// Template specializations to get better code for common cases.
405template <> constexpr inline bool isUInt<8>(uint64_t x) {
406 return static_cast<uint8_t>(x) == x;
407}
408template <> constexpr inline bool isUInt<16>(uint64_t x) {
409 return static_cast<uint16_t>(x) == x;
410}
411template <> constexpr inline bool isUInt<32>(uint64_t x) {
412 return static_cast<uint32_t>(x) == x;
413}
414
415/// Checks if a unsigned integer is an N bit number shifted left by S.
416template <unsigned N, unsigned S>
417constexpr inline bool isShiftedUInt(uint64_t x) {
418 static_assert(
419 N > 0, "isShiftedUInt<0> doesn't make sense (refers to a 0-bit number)");
420 static_assert(N + S <= 64,
421 "isShiftedUInt<N, S> with N + S > 64 is too wide.");
422 // Per the two static_asserts above, S must be strictly less than 64. So
423 // 1 << S is not undefined behavior.
424 return isUInt<N + S>(x) && (x % (UINT64_C(1)1UL << S) == 0);
425}
426
427/// Gets the maximum value for a N-bit unsigned integer.
428inline uint64_t maxUIntN(uint64_t N) {
429 assert(N > 0 && N <= 64 && "integer width out of range")(static_cast <bool> (N > 0 && N <= 64 &&
"integer width out of range") ? void (0) : __assert_fail ("N > 0 && N <= 64 && \"integer width out of range\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 429, __extension__ __PRETTY_FUNCTION__));
430
431 // uint64_t(1) << 64 is undefined behavior, so we can't do
432 // (uint64_t(1) << N) - 1
433 // without checking first that N != 64. But this works and doesn't have a
434 // branch.
435 return UINT64_MAX(18446744073709551615UL) >> (64 - N);
436}
437
438/// Gets the minimum value for a N-bit signed integer.
439inline int64_t minIntN(int64_t N) {
440 assert(N > 0 && N <= 64 && "integer width out of range")(static_cast <bool> (N > 0 && N <= 64 &&
"integer width out of range") ? void (0) : __assert_fail ("N > 0 && N <= 64 && \"integer width out of range\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 440, __extension__ __PRETTY_FUNCTION__));
441
442 return UINT64_C(1)1UL + ~(UINT64_C(1)1UL << (N - 1));
443}
444
445/// Gets the maximum value for a N-bit signed integer.
446inline int64_t maxIntN(int64_t N) {
447 assert(N > 0 && N <= 64 && "integer width out of range")(static_cast <bool> (N > 0 && N <= 64 &&
"integer width out of range") ? void (0) : __assert_fail ("N > 0 && N <= 64 && \"integer width out of range\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 447, __extension__ __PRETTY_FUNCTION__));
448
449 // This relies on two's complement wraparound when N == 64, so we convert to
450 // int64_t only at the very end to avoid UB.
451 return (UINT64_C(1)1UL << (N - 1)) - 1;
452}
453
454/// Checks if an unsigned integer fits into the given (dynamic) bit width.
455inline bool isUIntN(unsigned N, uint64_t x) {
456 return N >= 64 || x <= maxUIntN(N);
457}
458
459/// Checks if an signed integer fits into the given (dynamic) bit width.
460inline bool isIntN(unsigned N, int64_t x) {
461 return N >= 64 || (minIntN(N) <= x && x <= maxIntN(N));
462}
463
464/// Return true if the argument is a non-empty sequence of ones starting at the
465/// least significant bit with the remainder zero (32 bit version).
466/// Ex. isMask_32(0x0000FFFFU) == true.
467constexpr inline bool isMask_32(uint32_t Value) {
468 return Value && ((Value + 1) & Value) == 0;
469}
470
471/// Return true if the argument is a non-empty sequence of ones starting at the
472/// least significant bit with the remainder zero (64 bit version).
473constexpr inline bool isMask_64(uint64_t Value) {
474 return Value && ((Value + 1) & Value) == 0;
475}
476
477/// Return true if the argument contains a non-empty sequence of ones with the
478/// remainder zero (32 bit version.) Ex. isShiftedMask_32(0x0000FF00U) == true.
479constexpr inline bool isShiftedMask_32(uint32_t Value) {
480 return Value && isMask_32((Value - 1) | Value);
481}
482
483/// Return true if the argument contains a non-empty sequence of ones with the
484/// remainder zero (64 bit version.)
485constexpr inline bool isShiftedMask_64(uint64_t Value) {
486 return Value && isMask_64((Value - 1) | Value);
487}
488
489/// Return true if the argument is a power of two > 0.
490/// Ex. isPowerOf2_32(0x00100000U) == true (32 bit edition.)
491constexpr inline bool isPowerOf2_32(uint32_t Value) {
492 return Value && !(Value & (Value - 1));
493}
494
495/// Return true if the argument is a power of two > 0 (64 bit edition.)
496constexpr inline bool isPowerOf2_64(uint64_t Value) {
497 return Value && !(Value & (Value - 1));
7
←
Assuming 'Value' is 0
→
8
←
Returning zero, which participates in a condition later
→
498}
499
500/// Count the number of ones from the most significant bit to the first
501/// zero bit.
502///
503/// Ex. countLeadingOnes(0xFF0FFF00) == 8.
504/// Only unsigned integral types are allowed.
505///
506/// \param ZB the behavior on an input of all ones. Only ZB_Width and
507/// ZB_Undefined are valid arguments.
508template <typename T>
509unsigned countLeadingOnes(T Value, ZeroBehavior ZB = ZB_Width) {
510 static_assert(std::numeric_limits<T>::is_integer &&
511 !std::numeric_limits<T>::is_signed,
512 "Only unsigned integral types are allowed.");
513 return countLeadingZeros<T>(~Value, ZB);
514}
515
516/// Count the number of ones from the least significant bit to the first
517/// zero bit.
518///
519/// Ex. countTrailingOnes(0x00FF00FF) == 8.
520/// Only unsigned integral types are allowed.
521///
522/// \param ZB the behavior on an input of all ones. Only ZB_Width and
523/// ZB_Undefined are valid arguments.
524template <typename T>
525unsigned countTrailingOnes(T Value, ZeroBehavior ZB = ZB_Width) {
526 static_assert(std::numeric_limits<T>::is_integer &&
527 !std::numeric_limits<T>::is_signed,
528 "Only unsigned integral types are allowed.");
529 return countTrailingZeros<T>(~Value, ZB);
530}
531
532namespace detail {
533template <typename T, std::size_t SizeOfT> struct PopulationCounter {
534 static unsigned count(T Value) {
535 // Generic version, forward to 32 bits.
536 static_assert(SizeOfT <= 4, "Not implemented!");
537#if defined(__GNUC__4)
538 return __builtin_popcount(Value);
539#else
540 uint32_t v = Value;
541 v = v - ((v >> 1) & 0x55555555);
542 v = (v & 0x33333333) + ((v >> 2) & 0x33333333);
543 return ((v + (v >> 4) & 0xF0F0F0F) * 0x1010101) >> 24;
544#endif
545 }
546};
547
548template <typename T> struct PopulationCounter<T, 8> {
549 static unsigned count(T Value) {
550#if defined(__GNUC__4)
551 return __builtin_popcountll(Value);
552#else
553 uint64_t v = Value;
554 v = v - ((v >> 1) & 0x5555555555555555ULL);
555 v = (v & 0x3333333333333333ULL) + ((v >> 2) & 0x3333333333333333ULL);
556 v = (v + (v >> 4)) & 0x0F0F0F0F0F0F0F0FULL;
557 return unsigned((uint64_t)(v * 0x0101010101010101ULL) >> 56);
558#endif
559 }
560};
561} // namespace detail
562
563/// Count the number of set bits in a value.
564/// Ex. countPopulation(0xF000F000) = 8
565/// Returns 0 if the word is zero.
566template <typename T>
567inline unsigned countPopulation(T Value) {
568 static_assert(std::numeric_limits<T>::is_integer &&
569 !std::numeric_limits<T>::is_signed,
570 "Only unsigned integral types are allowed.");
571 return detail::PopulationCounter<T, sizeof(T)>::count(Value);
572}
573
574/// Compile time Log2.
575/// Valid only for positive powers of two.
576template <size_t kValue> constexpr inline size_t CTLog2() {
577 static_assert(kValue > 0 && llvm::isPowerOf2_64(kValue),
578 "Value is not a valid power of 2");
579 return 1 + CTLog2<kValue / 2>();
580}
581
582template <> constexpr inline size_t CTLog2<1>() { return 0; }
583
584/// Return the log base 2 of the specified value.
585inline double Log2(double Value) {
586#if defined(__ANDROID_API__) && __ANDROID_API__ < 18
587 return __builtin_log(Value) / __builtin_log(2.0);
588#else
589 return log2(Value);
590#endif
591}
592
593/// Return the floor log base 2 of the specified value, -1 if the value is zero.
594/// (32 bit edition.)
595/// Ex. Log2_32(32) == 5, Log2_32(1) == 0, Log2_32(0) == -1, Log2_32(6) == 2
596inline unsigned Log2_32(uint32_t Value) {
597 return 31 - countLeadingZeros(Value);
12
←
Returning the value 4294967295
→
598}
599
600/// Return the floor log base 2 of the specified value, -1 if the value is zero.
601/// (64 bit edition.)
602inline unsigned Log2_64(uint64_t Value) {
603 return 63 - countLeadingZeros(Value);
604}
605
606/// Return the ceil log base 2 of the specified value, 32 if the value is zero.
607/// (32 bit edition).
608/// Ex. Log2_32_Ceil(32) == 5, Log2_32_Ceil(1) == 0, Log2_32_Ceil(6) == 3
609inline unsigned Log2_32_Ceil(uint32_t Value) {
610 return 32 - countLeadingZeros(Value - 1);
611}
612
613/// Return the ceil log base 2 of the specified value, 64 if the value is zero.
614/// (64 bit edition.)
615inline unsigned Log2_64_Ceil(uint64_t Value) {
616 return 64 - countLeadingZeros(Value - 1);
617}
618
619/// Return the greatest common divisor of the values using Euclid's algorithm.
620template <typename T>
621inline T greatestCommonDivisor(T A, T B) {
622 while (B) {
623 T Tmp = B;
624 B = A % B;
625 A = Tmp;
626 }
627 return A;
628}
629
630inline uint64_t GreatestCommonDivisor64(uint64_t A, uint64_t B) {
631 return greatestCommonDivisor<uint64_t>(A, B);
632}
633
634/// This function takes a 64-bit integer and returns the bit equivalent double.
635inline double BitsToDouble(uint64_t Bits) {
636 double D;
637 static_assert(sizeof(uint64_t) == sizeof(double), "Unexpected type sizes");
638 memcpy(&D, &Bits, sizeof(Bits));
639 return D;
640}
641
642/// This function takes a 32-bit integer and returns the bit equivalent float.
643inline float BitsToFloat(uint32_t Bits) {
644 float F;
645 static_assert(sizeof(uint32_t) == sizeof(float), "Unexpected type sizes");
646 memcpy(&F, &Bits, sizeof(Bits));
647 return F;
648}
649
650/// This function takes a double and returns the bit equivalent 64-bit integer.
651/// Note that copying doubles around changes the bits of NaNs on some hosts,
652/// notably x86, so this routine cannot be used if these bits are needed.
653inline uint64_t DoubleToBits(double Double) {
654 uint64_t Bits;
655 static_assert(sizeof(uint64_t) == sizeof(double), "Unexpected type sizes");
656 memcpy(&Bits, &Double, sizeof(Double));
657 return Bits;
658}
659
660/// This function takes a float and returns the bit equivalent 32-bit integer.
661/// Note that copying floats around changes the bits of NaNs on some hosts,
662/// notably x86, so this routine cannot be used if these bits are needed.
663inline uint32_t FloatToBits(float Float) {
664 uint32_t Bits;
665 static_assert(sizeof(uint32_t) == sizeof(float), "Unexpected type sizes");
666 memcpy(&Bits, &Float, sizeof(Float));
667 return Bits;
668}
669
670/// A and B are either alignments or offsets. Return the minimum alignment that
671/// may be assumed after adding the two together.
672constexpr inline uint64_t MinAlign(uint64_t A, uint64_t B) {
673 // The largest power of 2 that divides both A and B.
674 //
675 // Replace "-Value" by "1+~Value" in the following commented code to avoid
676 // MSVC warning C4146
677 // return (A | B) & -(A | B);
678 return (A | B) & (1 + ~(A | B));
679}
680
681/// Returns the next power of two (in 64-bits) that is strictly greater than A.
682/// Returns zero on overflow.
683inline uint64_t NextPowerOf2(uint64_t A) {
684 A |= (A >> 1);
685 A |= (A >> 2);
686 A |= (A >> 4);
687 A |= (A >> 8);
688 A |= (A >> 16);
689 A |= (A >> 32);
690 return A + 1;
691}
692
693/// Returns the power of two which is less than or equal to the given value.
694/// Essentially, it is a floor operation across the domain of powers of two.
695inline uint64_t PowerOf2Floor(uint64_t A) {
696 if (!A) return 0;
697 return 1ull << (63 - countLeadingZeros(A, ZB_Undefined));
698}
699
700/// Returns the power of two which is greater than or equal to the given value.
701/// Essentially, it is a ceil operation across the domain of powers of two.
702inline uint64_t PowerOf2Ceil(uint64_t A) {
703 if (!A)
704 return 0;
705 return NextPowerOf2(A - 1);
706}
707
708/// Returns the next integer (mod 2**64) that is greater than or equal to
709/// \p Value and is a multiple of \p Align. \p Align must be non-zero.
710///
711/// If non-zero \p Skew is specified, the return value will be a minimal
712/// integer that is greater than or equal to \p Value and equal to
713/// \p Align * N + \p Skew for some integer N. If \p Skew is larger than
714/// \p Align, its value is adjusted to '\p Skew mod \p Align'.
715///
716/// Examples:
717/// \code
718/// alignTo(5, 8) = 8
719/// alignTo(17, 8) = 24
720/// alignTo(~0LL, 8) = 0
721/// alignTo(321, 255) = 510
722///
723/// alignTo(5, 8, 7) = 7
724/// alignTo(17, 8, 1) = 17
725/// alignTo(~0LL, 8, 3) = 3
726/// alignTo(321, 255, 42) = 552
727/// \endcode
728inline uint64_t alignTo(uint64_t Value, uint64_t Align, uint64_t Skew = 0) {
729 assert(Align != 0u && "Align can't be 0.")(static_cast <bool> (Align != 0u && "Align can't be 0."
) ? void (0) : __assert_fail ("Align != 0u && \"Align can't be 0.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 729, __extension__ __PRETTY_FUNCTION__));
730 Skew %= Align;
731 return (Value + Align - 1 - Skew) / Align * Align + Skew;
732}
733
734/// Returns the next integer (mod 2**64) that is greater than or equal to
735/// \p Value and is a multiple of \c Align. \c Align must be non-zero.
736template <uint64_t Align> constexpr inline uint64_t alignTo(uint64_t Value) {
737 static_assert(Align != 0u, "Align must be non-zero");
738 return (Value + Align - 1) / Align * Align;
739}
740
741/// Returns the integer ceil(Numerator / Denominator).
742inline uint64_t divideCeil(uint64_t Numerator, uint64_t Denominator) {
743 return alignTo(Numerator, Denominator) / Denominator;
744}
745
746/// Returns the integer nearest(Numerator / Denominator).
747inline uint64_t divideNearest(uint64_t Numerator, uint64_t Denominator) {
748 return (Numerator + (Denominator / 2)) / Denominator;
749}
750
751/// Returns the largest uint64_t less than or equal to \p Value and is
752/// \p Skew mod \p Align. \p Align must be non-zero
753inline uint64_t alignDown(uint64_t Value, uint64_t Align, uint64_t Skew = 0) {
754 assert(Align != 0u && "Align can't be 0.")(static_cast <bool> (Align != 0u && "Align can't be 0."
) ? void (0) : __assert_fail ("Align != 0u && \"Align can't be 0.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 754, __extension__ __PRETTY_FUNCTION__));
755 Skew %= Align;
756 return (Value - Skew) / Align * Align + Skew;
757}
758
759/// Sign-extend the number in the bottom B bits of X to a 32-bit integer.
760/// Requires 0 < B <= 32.
761template <unsigned B> constexpr inline int32_t SignExtend32(uint32_t X) {
762 static_assert(B > 0, "Bit width can't be 0.");
763 static_assert(B <= 32, "Bit width out of range.");
764 return int32_t(X << (32 - B)) >> (32 - B);
765}
766
767/// Sign-extend the number in the bottom B bits of X to a 32-bit integer.
768/// Requires 0 < B <= 32.
769inline int32_t SignExtend32(uint32_t X, unsigned B) {
770 assert(B > 0 && "Bit width can't be 0.")(static_cast <bool> (B > 0 && "Bit width can't be 0."
) ? void (0) : __assert_fail ("B > 0 && \"Bit width can't be 0.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 770, __extension__ __PRETTY_FUNCTION__));
771 assert(B <= 32 && "Bit width out of range.")(static_cast <bool> (B <= 32 && "Bit width out of range."
) ? void (0) : __assert_fail ("B <= 32 && \"Bit width out of range.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 771, __extension__ __PRETTY_FUNCTION__));
772 return int32_t(X << (32 - B)) >> (32 - B);
773}
774
775/// Sign-extend the number in the bottom B bits of X to a 64-bit integer.
776/// Requires 0 < B <= 64.
777template <unsigned B> constexpr inline int64_t SignExtend64(uint64_t x) {
778 static_assert(B > 0, "Bit width can't be 0.");
779 static_assert(B <= 64, "Bit width out of range.");
780 return int64_t(x << (64 - B)) >> (64 - B);
781}
782
783/// Sign-extend the number in the bottom B bits of X to a 64-bit integer.
784/// Requires 0 < B <= 64.
785inline int64_t SignExtend64(uint64_t X, unsigned B) {
786 assert(B > 0 && "Bit width can't be 0.")(static_cast <bool> (B > 0 && "Bit width can't be 0."
) ? void (0) : __assert_fail ("B > 0 && \"Bit width can't be 0.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 786, __extension__ __PRETTY_FUNCTION__));
787 assert(B <= 64 && "Bit width out of range.")(static_cast <bool> (B <= 64 && "Bit width out of range."
) ? void (0) : __assert_fail ("B <= 64 && \"Bit width out of range.\""
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/include/llvm/Support/MathExtras.h"
, 787, __extension__ __PRETTY_FUNCTION__));
788 return int64_t(X << (64 - B)) >> (64 - B);
789}
790
791/// Subtract two unsigned integers, X and Y, of type T and return the absolute
792/// value of the result.
793template <typename T>
794std::enable_if_t<std::is_unsigned<T>::value, T> AbsoluteDifference(T X, T Y) {
795 return X > Y ? (X - Y) : (Y - X);
796}
797
798/// Add two unsigned integers, X and Y, of type T. Clamp the result to the
799/// maximum representable value of T on overflow. ResultOverflowed indicates if
800/// the result is larger than the maximum representable value of type T.
801template <typename T>
802std::enable_if_t<std::is_unsigned<T>::value, T>
803SaturatingAdd(T X, T Y, bool *ResultOverflowed = nullptr) {
804 bool Dummy;
805 bool &Overflowed = ResultOverflowed ? *ResultOverflowed : Dummy;
806 // Hacker's Delight, p. 29
807 T Z = X + Y;
808 Overflowed = (Z < X || Z < Y);
809 if (Overflowed)
810 return std::numeric_limits<T>::max();
811 else
812 return Z;
813}
814
815/// Multiply two unsigned integers, X and Y, of type T. Clamp the result to the
816/// maximum representable value of T on overflow. ResultOverflowed indicates if
817/// the result is larger than the maximum representable value of type T.
818template <typename T>
819std::enable_if_t<std::is_unsigned<T>::value, T>
820SaturatingMultiply(T X, T Y, bool *ResultOverflowed = nullptr) {
821 bool Dummy;
822 bool &Overflowed = ResultOverflowed ? *ResultOverflowed : Dummy;
823
824 // Hacker's Delight, p. 30 has a different algorithm, but we don't use that
825 // because it fails for uint16_t (where multiplication can have undefined
826 // behavior due to promotion to int), and requires a division in addition
827 // to the multiplication.
828
829 Overflowed = false;
830
831 // Log2(Z) would be either Log2Z or Log2Z + 1.
832 // Special case: if X or Y is 0, Log2_64 gives -1, and Log2Z
833 // will necessarily be less than Log2Max as desired.
834 int Log2Z = Log2_64(X) + Log2_64(Y);
835 const T Max = std::numeric_limits<T>::max();
836 int Log2Max = Log2_64(Max);
837 if (Log2Z < Log2Max) {
838 return X * Y;
839 }
840 if (Log2Z > Log2Max) {
841 Overflowed = true;
842 return Max;
843 }
844
845 // We're going to use the top bit, and maybe overflow one
846 // bit past it. Multiply all but the bottom bit then add
847 // that on at the end.
848 T Z = (X >> 1) * Y;
849 if (Z & ~(Max >> 1)) {
850 Overflowed = true;
851 return Max;
852 }
853 Z <<= 1;
854 if (X & 1)
855 return SaturatingAdd(Z, Y, ResultOverflowed);
856
857 return Z;
858}
859
860/// Multiply two unsigned integers, X and Y, and add the unsigned integer, A to
861/// the product. Clamp the result to the maximum representable value of T on
862/// overflow. ResultOverflowed indicates if the result is larger than the
863/// maximum representable value of type T.
864template <typename T>
865std::enable_if_t<std::is_unsigned<T>::value, T>
866SaturatingMultiplyAdd(T X, T Y, T A, bool *ResultOverflowed = nullptr) {
867 bool Dummy;
868 bool &Overflowed = ResultOverflowed ? *ResultOverflowed : Dummy;
869
870 T Product = SaturatingMultiply(X, Y, &Overflowed);
871 if (Overflowed)
872 return Product;
873
874 return SaturatingAdd(A, Product, &Overflowed);
875}
876
877/// Use this rather than HUGE_VALF; the latter causes warnings on MSVC.
878extern const float huge_valf;
879
880
881/// Add two signed integers, computing the two's complement truncated result,
882/// returning true if overflow occured.
883template <typename T>
884std::enable_if_t<std::is_signed<T>::value, T> AddOverflow(T X, T Y, T &Result) {
885#if __has_builtin(__builtin_add_overflow)1
886 return __builtin_add_overflow(X, Y, &Result);
887#else
888 // Perform the unsigned addition.
889 using U = std::make_unsigned_t<T>;
890 const U UX = static_cast<U>(X);
891 const U UY = static_cast<U>(Y);
892 const U UResult = UX + UY;
893
894 // Convert to signed.
895 Result = static_cast<T>(UResult);
896
897 // Adding two positive numbers should result in a positive number.
898 if (X > 0 && Y > 0)
899 return Result <= 0;
900 // Adding two negatives should result in a negative number.
901 if (X < 0 && Y < 0)
902 return Result >= 0;
903 return false;
904#endif
905}
906
907/// Subtract two signed integers, computing the two's complement truncated
908/// result, returning true if an overflow ocurred.
909template <typename T>
910std::enable_if_t<std::is_signed<T>::value, T> SubOverflow(T X, T Y, T &Result) {
911#if __has_builtin(__builtin_sub_overflow)1
912 return __builtin_sub_overflow(X, Y, &Result);
913#else
914 // Perform the unsigned addition.
915 using U = std::make_unsigned_t<T>;
916 const U UX = static_cast<U>(X);
917 const U UY = static_cast<U>(Y);
918 const U UResult = UX - UY;
919
920 // Convert to signed.
921 Result = static_cast<T>(UResult);
922
923 // Subtracting a positive number from a negative results in a negative number.
924 if (X <= 0 && Y > 0)
925 return Result >= 0;
926 // Subtracting a negative number from a positive results in a positive number.
927 if (X >= 0 && Y < 0)
928 return Result <= 0;
929 return false;
930#endif
931}
932
933/// Multiply two signed integers, computing the two's complement truncated
934/// result, returning true if an overflow ocurred.
935template <typename T>
936std::enable_if_t<std::is_signed<T>::value, T> MulOverflow(T X, T Y, T &Result) {
937 // Perform the unsigned multiplication on absolute values.
938 using U = std::make_unsigned_t<T>;
939 const U UX = X < 0 ? (0 - static_cast<U>(X)) : static_cast<U>(X);
940 const U UY = Y < 0 ? (0 - static_cast<U>(Y)) : static_cast<U>(Y);
941 const U UResult = UX * UY;
942
943 // Convert to signed.
944 const bool IsNegative = (X < 0) ^ (Y < 0);
945 Result = IsNegative ? (0 - UResult) : UResult;
946
947 // If any of the args was 0, result is 0 and no overflow occurs.
948 if (UX == 0 || UY == 0)
949 return false;
950
951 // UX and UY are in [1, 2^n], where n is the number of digits.
952 // Check how the max allowed absolute value (2^n for negative, 2^(n-1) for
953 // positive) divided by an argument compares to the other.
954 if (IsNegative)
955 return UX > (static_cast<U>(std::numeric_limits<T>::max()) + U(1)) / UY;
956 else
957 return UX > (static_cast<U>(std::numeric_limits<T>::max())) / UY;
958}
959
960} // End llvm namespace
961
962#endif