Bug Summary

File:llvm/lib/Object/ELFObjectFile.cpp
Warning:line 76, column 12
The result of the left shift is undefined due to shifting by '64', which is greater or equal to the width of type 'unsigned long long'

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 ELFObjectFile.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/Object -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/Object -I /build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Object -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/Object -fdebug-prefix-map=/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0=. -ferror-limit 19 -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/Object/ELFObjectFile.cpp

/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Object/ELFObjectFile.cpp

1//===- ELFObjectFile.cpp - ELF object file implementation -----------------===//
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// Part of the ELFObjectFile class implementation.
10//
11//===----------------------------------------------------------------------===//
12
13#include "llvm/Object/ELFObjectFile.h"
14#include "llvm/ADT/Triple.h"
15#include "llvm/BinaryFormat/ELF.h"
16#include "llvm/MC/MCInstrAnalysis.h"
17#include "llvm/MC/SubtargetFeature.h"
18#include "llvm/Object/ELF.h"
19#include "llvm/Object/ELFTypes.h"
20#include "llvm/Object/Error.h"
21#include "llvm/Support/ARMAttributeParser.h"
22#include "llvm/Support/ARMBuildAttributes.h"
23#include "llvm/Support/Endian.h"
24#include "llvm/Support/ErrorHandling.h"
25#include "llvm/Support/MathExtras.h"
26#include "llvm/Support/RISCVAttributeParser.h"
27#include "llvm/Support/RISCVAttributes.h"
28#include "llvm/Support/TargetRegistry.h"
29#include <algorithm>
30#include <cstddef>
31#include <cstdint>
32#include <memory>
33#include <string>
34#include <system_error>
35#include <utility>
36
37using namespace llvm;
38using namespace object;
39
40const EnumEntry<unsigned> llvm::object::ElfSymbolTypes[NumElfSymbolTypes] = {
41 {"None", "NOTYPE", ELF::STT_NOTYPE},
42 {"Object", "OBJECT", ELF::STT_OBJECT},
43 {"Function", "FUNC", ELF::STT_FUNC},
44 {"Section", "SECTION", ELF::STT_SECTION},
45 {"File", "FILE", ELF::STT_FILE},
46 {"Common", "COMMON", ELF::STT_COMMON},
47 {"TLS", "TLS", ELF::STT_TLS},
48 {"Unknown", "<unknown>: 7", 7},
49 {"Unknown", "<unknown>: 8", 8},
50 {"Unknown", "<unknown>: 9", 9},
51 {"GNU_IFunc", "IFUNC", ELF::STT_GNU_IFUNC},
52 {"OS Specific", "<OS specific>: 11", 11},
53 {"OS Specific", "<OS specific>: 12", 12},
54 {"Proc Specific", "<processor specific>: 13", 13},
55 {"Proc Specific", "<processor specific>: 14", 14},
56 {"Proc Specific", "<processor specific>: 15", 15}
57};
58
59ELFObjectFileBase::ELFObjectFileBase(unsigned int Type, MemoryBufferRef Source)
60 : ObjectFile(Type, Source) {}
61
62template <class ELFT>
63static Expected<std::unique_ptr<ELFObjectFile<ELFT>>>
64createPtr(MemoryBufferRef Object, bool InitContent) {
65 auto Ret = ELFObjectFile<ELFT>::create(Object, InitContent);
66 if (Error E = Ret.takeError())
67 return std::move(E);
68 return std::make_unique<ELFObjectFile<ELFT>>(std::move(*Ret));
69}
70
71Expected<std::unique_ptr<ObjectFile>>
72ObjectFile::createELFObjectFile(MemoryBufferRef Obj, bool InitContent) {
73 std::pair<unsigned char, unsigned char> Ident =
74 getElfArchType(Obj.getBuffer());
75 std::size_t MaxAlignment =
76 1ULL << countTrailingZeros(
1
Calling 'countTrailingZeros<unsigned long>'
8
Returning from 'countTrailingZeros<unsigned long>'
9
The result of the left shift is undefined due to shifting by '64', which is greater or equal to the width of type 'unsigned long long'
77 reinterpret_cast<uintptr_t>(Obj.getBufferStart()));
78
79 if (MaxAlignment < 2)
80 return createError("Insufficient alignment");
81
82 if (Ident.first == ELF::ELFCLASS32) {
83 if (Ident.second == ELF::ELFDATA2LSB)
84 return createPtr<ELF32LE>(Obj, InitContent);
85 else if (Ident.second == ELF::ELFDATA2MSB)
86 return createPtr<ELF32BE>(Obj, InitContent);
87 else
88 return createError("Invalid ELF data");
89 } else if (Ident.first == ELF::ELFCLASS64) {
90 if (Ident.second == ELF::ELFDATA2LSB)
91 return createPtr<ELF64LE>(Obj, InitContent);
92 else if (Ident.second == ELF::ELFDATA2MSB)
93 return createPtr<ELF64BE>(Obj, InitContent);
94 else
95 return createError("Invalid ELF data");
96 }
97 return createError("Invalid ELF class");
98}
99
100SubtargetFeatures ELFObjectFileBase::getMIPSFeatures() const {
101 SubtargetFeatures Features;
102 unsigned PlatformFlags = getPlatformFlags();
103
104 switch (PlatformFlags & ELF::EF_MIPS_ARCH) {
105 case ELF::EF_MIPS_ARCH_1:
106 break;
107 case ELF::EF_MIPS_ARCH_2:
108 Features.AddFeature("mips2");
109 break;
110 case ELF::EF_MIPS_ARCH_3:
111 Features.AddFeature("mips3");
112 break;
113 case ELF::EF_MIPS_ARCH_4:
114 Features.AddFeature("mips4");
115 break;
116 case ELF::EF_MIPS_ARCH_5:
117 Features.AddFeature("mips5");
118 break;
119 case ELF::EF_MIPS_ARCH_32:
120 Features.AddFeature("mips32");
121 break;
122 case ELF::EF_MIPS_ARCH_64:
123 Features.AddFeature("mips64");
124 break;
125 case ELF::EF_MIPS_ARCH_32R2:
126 Features.AddFeature("mips32r2");
127 break;
128 case ELF::EF_MIPS_ARCH_64R2:
129 Features.AddFeature("mips64r2");
130 break;
131 case ELF::EF_MIPS_ARCH_32R6:
132 Features.AddFeature("mips32r6");
133 break;
134 case ELF::EF_MIPS_ARCH_64R6:
135 Features.AddFeature("mips64r6");
136 break;
137 default:
138 llvm_unreachable("Unknown EF_MIPS_ARCH value")::llvm::llvm_unreachable_internal("Unknown EF_MIPS_ARCH value"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Object/ELFObjectFile.cpp"
, 138)
;
139 }
140
141 switch (PlatformFlags & ELF::EF_MIPS_MACH) {
142 case ELF::EF_MIPS_MACH_NONE:
143 // No feature associated with this value.
144 break;
145 case ELF::EF_MIPS_MACH_OCTEON:
146 Features.AddFeature("cnmips");
147 break;
148 default:
149 llvm_unreachable("Unknown EF_MIPS_ARCH value")::llvm::llvm_unreachable_internal("Unknown EF_MIPS_ARCH value"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Object/ELFObjectFile.cpp"
, 149)
;
150 }
151
152 if (PlatformFlags & ELF::EF_MIPS_ARCH_ASE_M16)
153 Features.AddFeature("mips16");
154 if (PlatformFlags & ELF::EF_MIPS_MICROMIPS)
155 Features.AddFeature("micromips");
156
157 return Features;
158}
159
160SubtargetFeatures ELFObjectFileBase::getARMFeatures() const {
161 SubtargetFeatures Features;
162 ARMAttributeParser Attributes;
163 if (Error E = getBuildAttributes(Attributes)) {
164 consumeError(std::move(E));
165 return SubtargetFeatures();
166 }
167
168 // both ARMv7-M and R have to support thumb hardware div
169 bool isV7 = false;
170 Optional<unsigned> Attr =
171 Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch);
172 if (Attr.hasValue())
173 isV7 = Attr.getValue() == ARMBuildAttrs::v7;
174
175 Attr = Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch_profile);
176 if (Attr.hasValue()) {
177 switch (Attr.getValue()) {
178 case ARMBuildAttrs::ApplicationProfile:
179 Features.AddFeature("aclass");
180 break;
181 case ARMBuildAttrs::RealTimeProfile:
182 Features.AddFeature("rclass");
183 if (isV7)
184 Features.AddFeature("hwdiv");
185 break;
186 case ARMBuildAttrs::MicroControllerProfile:
187 Features.AddFeature("mclass");
188 if (isV7)
189 Features.AddFeature("hwdiv");
190 break;
191 }
192 }
193
194 Attr = Attributes.getAttributeValue(ARMBuildAttrs::THUMB_ISA_use);
195 if (Attr.hasValue()) {
196 switch (Attr.getValue()) {
197 default:
198 break;
199 case ARMBuildAttrs::Not_Allowed:
200 Features.AddFeature("thumb", false);
201 Features.AddFeature("thumb2", false);
202 break;
203 case ARMBuildAttrs::AllowThumb32:
204 Features.AddFeature("thumb2");
205 break;
206 }
207 }
208
209 Attr = Attributes.getAttributeValue(ARMBuildAttrs::FP_arch);
210 if (Attr.hasValue()) {
211 switch (Attr.getValue()) {
212 default:
213 break;
214 case ARMBuildAttrs::Not_Allowed:
215 Features.AddFeature("vfp2sp", false);
216 Features.AddFeature("vfp3d16sp", false);
217 Features.AddFeature("vfp4d16sp", false);
218 break;
219 case ARMBuildAttrs::AllowFPv2:
220 Features.AddFeature("vfp2");
221 break;
222 case ARMBuildAttrs::AllowFPv3A:
223 case ARMBuildAttrs::AllowFPv3B:
224 Features.AddFeature("vfp3");
225 break;
226 case ARMBuildAttrs::AllowFPv4A:
227 case ARMBuildAttrs::AllowFPv4B:
228 Features.AddFeature("vfp4");
229 break;
230 }
231 }
232
233 Attr = Attributes.getAttributeValue(ARMBuildAttrs::Advanced_SIMD_arch);
234 if (Attr.hasValue()) {
235 switch (Attr.getValue()) {
236 default:
237 break;
238 case ARMBuildAttrs::Not_Allowed:
239 Features.AddFeature("neon", false);
240 Features.AddFeature("fp16", false);
241 break;
242 case ARMBuildAttrs::AllowNeon:
243 Features.AddFeature("neon");
244 break;
245 case ARMBuildAttrs::AllowNeon2:
246 Features.AddFeature("neon");
247 Features.AddFeature("fp16");
248 break;
249 }
250 }
251
252 Attr = Attributes.getAttributeValue(ARMBuildAttrs::MVE_arch);
253 if (Attr.hasValue()) {
254 switch (Attr.getValue()) {
255 default:
256 break;
257 case ARMBuildAttrs::Not_Allowed:
258 Features.AddFeature("mve", false);
259 Features.AddFeature("mve.fp", false);
260 break;
261 case ARMBuildAttrs::AllowMVEInteger:
262 Features.AddFeature("mve.fp", false);
263 Features.AddFeature("mve");
264 break;
265 case ARMBuildAttrs::AllowMVEIntegerAndFloat:
266 Features.AddFeature("mve.fp");
267 break;
268 }
269 }
270
271 Attr = Attributes.getAttributeValue(ARMBuildAttrs::DIV_use);
272 if (Attr.hasValue()) {
273 switch (Attr.getValue()) {
274 default:
275 break;
276 case ARMBuildAttrs::DisallowDIV:
277 Features.AddFeature("hwdiv", false);
278 Features.AddFeature("hwdiv-arm", false);
279 break;
280 case ARMBuildAttrs::AllowDIVExt:
281 Features.AddFeature("hwdiv");
282 Features.AddFeature("hwdiv-arm");
283 break;
284 }
285 }
286
287 return Features;
288}
289
290SubtargetFeatures ELFObjectFileBase::getRISCVFeatures() const {
291 SubtargetFeatures Features;
292 unsigned PlatformFlags = getPlatformFlags();
293
294 if (PlatformFlags & ELF::EF_RISCV_RVC) {
295 Features.AddFeature("c");
296 }
297
298 // Add features according to the ELF attribute section.
299 // If there are any unrecognized features, ignore them.
300 RISCVAttributeParser Attributes;
301 if (Error E = getBuildAttributes(Attributes)) {
302 // TODO Propagate Error.
303 consumeError(std::move(E));
304 return Features; // Keep "c" feature if there is one in PlatformFlags.
305 }
306
307 Optional<StringRef> Attr = Attributes.getAttributeString(RISCVAttrs::ARCH);
308 if (Attr.hasValue()) {
309 // The Arch pattern is [rv32|rv64][i|e]version(_[m|a|f|d|c]version)*
310 // Version string pattern is (major)p(minor). Major and minor are optional.
311 // For example, a version number could be 2p0, 2, or p92.
312 StringRef Arch = Attr.getValue();
313 if (Arch.consume_front("rv32"))
314 Features.AddFeature("64bit", false);
315 else if (Arch.consume_front("rv64"))
316 Features.AddFeature("64bit");
317
318 while (!Arch.empty()) {
319 switch (Arch[0]) {
320 default:
321 break; // Ignore unexpected features.
322 case 'i':
323 Features.AddFeature("e", false);
324 break;
325 case 'd':
326 Features.AddFeature("f"); // D-ext will imply F-ext.
327 LLVM_FALLTHROUGH[[gnu::fallthrough]];
328 case 'e':
329 case 'm':
330 case 'a':
331 case 'f':
332 case 'c':
333 Features.AddFeature(Arch.take_front());
334 break;
335 }
336
337 // FIXME: Handle version numbers.
338 Arch = Arch.drop_until([](char c) { return c == '_' || c == '\0'; });
339 Arch = Arch.drop_while([](char c) { return c == '_'; });
340 }
341 }
342
343 return Features;
344}
345
346SubtargetFeatures ELFObjectFileBase::getFeatures() const {
347 switch (getEMachine()) {
348 case ELF::EM_MIPS:
349 return getMIPSFeatures();
350 case ELF::EM_ARM:
351 return getARMFeatures();
352 case ELF::EM_RISCV:
353 return getRISCVFeatures();
354 default:
355 return SubtargetFeatures();
356 }
357}
358
359Optional<StringRef> ELFObjectFileBase::tryGetCPUName() const {
360 switch (getEMachine()) {
361 case ELF::EM_AMDGPU:
362 return getAMDGPUCPUName();
363 default:
364 return None;
365 }
366}
367
368StringRef ELFObjectFileBase::getAMDGPUCPUName() const {
369 assert(getEMachine() == ELF::EM_AMDGPU)(static_cast <bool> (getEMachine() == ELF::EM_AMDGPU) ?
void (0) : __assert_fail ("getEMachine() == ELF::EM_AMDGPU",
"/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Object/ELFObjectFile.cpp"
, 369, __extension__ __PRETTY_FUNCTION__))
;
370 unsigned CPU = getPlatformFlags() & ELF::EF_AMDGPU_MACH;
371
372 switch (CPU) {
373 // Radeon HD 2000/3000 Series (R600).
374 case ELF::EF_AMDGPU_MACH_R600_R600:
375 return "r600";
376 case ELF::EF_AMDGPU_MACH_R600_R630:
377 return "r630";
378 case ELF::EF_AMDGPU_MACH_R600_RS880:
379 return "rs880";
380 case ELF::EF_AMDGPU_MACH_R600_RV670:
381 return "rv670";
382
383 // Radeon HD 4000 Series (R700).
384 case ELF::EF_AMDGPU_MACH_R600_RV710:
385 return "rv710";
386 case ELF::EF_AMDGPU_MACH_R600_RV730:
387 return "rv730";
388 case ELF::EF_AMDGPU_MACH_R600_RV770:
389 return "rv770";
390
391 // Radeon HD 5000 Series (Evergreen).
392 case ELF::EF_AMDGPU_MACH_R600_CEDAR:
393 return "cedar";
394 case ELF::EF_AMDGPU_MACH_R600_CYPRESS:
395 return "cypress";
396 case ELF::EF_AMDGPU_MACH_R600_JUNIPER:
397 return "juniper";
398 case ELF::EF_AMDGPU_MACH_R600_REDWOOD:
399 return "redwood";
400 case ELF::EF_AMDGPU_MACH_R600_SUMO:
401 return "sumo";
402
403 // Radeon HD 6000 Series (Northern Islands).
404 case ELF::EF_AMDGPU_MACH_R600_BARTS:
405 return "barts";
406 case ELF::EF_AMDGPU_MACH_R600_CAICOS:
407 return "caicos";
408 case ELF::EF_AMDGPU_MACH_R600_CAYMAN:
409 return "cayman";
410 case ELF::EF_AMDGPU_MACH_R600_TURKS:
411 return "turks";
412
413 // AMDGCN GFX6.
414 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX600:
415 return "gfx600";
416 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX601:
417 return "gfx601";
418 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX602:
419 return "gfx602";
420
421 // AMDGCN GFX7.
422 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX700:
423 return "gfx700";
424 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX701:
425 return "gfx701";
426 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX702:
427 return "gfx702";
428 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX703:
429 return "gfx703";
430 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX704:
431 return "gfx704";
432 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX705:
433 return "gfx705";
434
435 // AMDGCN GFX8.
436 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX801:
437 return "gfx801";
438 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX802:
439 return "gfx802";
440 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX803:
441 return "gfx803";
442 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX805:
443 return "gfx805";
444 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX810:
445 return "gfx810";
446
447 // AMDGCN GFX9.
448 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX900:
449 return "gfx900";
450 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX902:
451 return "gfx902";
452 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX904:
453 return "gfx904";
454 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX906:
455 return "gfx906";
456 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX908:
457 return "gfx908";
458 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX909:
459 return "gfx909";
460 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX90A:
461 return "gfx90a";
462 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX90C:
463 return "gfx90c";
464
465 // AMDGCN GFX10.
466 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1010:
467 return "gfx1010";
468 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1011:
469 return "gfx1011";
470 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1012:
471 return "gfx1012";
472 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1013:
473 return "gfx1013";
474 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1030:
475 return "gfx1030";
476 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1031:
477 return "gfx1031";
478 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1032:
479 return "gfx1032";
480 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1033:
481 return "gfx1033";
482 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1034:
483 return "gfx1034";
484 case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1035:
485 return "gfx1035";
486 default:
487 llvm_unreachable("Unknown EF_AMDGPU_MACH value")::llvm::llvm_unreachable_internal("Unknown EF_AMDGPU_MACH value"
, "/build/llvm-toolchain-snapshot-14~++20210828111110+16086d47c0d0/llvm/lib/Object/ELFObjectFile.cpp"
, 487)
;
488 }
489}
490
491// FIXME Encode from a tablegen description or target parser.
492void ELFObjectFileBase::setARMSubArch(Triple &TheTriple) const {
493 if (TheTriple.getSubArch() != Triple::NoSubArch)
494 return;
495
496 ARMAttributeParser Attributes;
497 if (Error E = getBuildAttributes(Attributes)) {
498 // TODO Propagate Error.
499 consumeError(std::move(E));
500 return;
501 }
502
503 std::string Triple;
504 // Default to ARM, but use the triple if it's been set.
505 if (TheTriple.isThumb())
506 Triple = "thumb";
507 else
508 Triple = "arm";
509
510 Optional<unsigned> Attr =
511 Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch);
512 if (Attr.hasValue()) {
513 switch (Attr.getValue()) {
514 case ARMBuildAttrs::v4:
515 Triple += "v4";
516 break;
517 case ARMBuildAttrs::v4T:
518 Triple += "v4t";
519 break;
520 case ARMBuildAttrs::v5T:
521 Triple += "v5t";
522 break;
523 case ARMBuildAttrs::v5TE:
524 Triple += "v5te";
525 break;
526 case ARMBuildAttrs::v5TEJ:
527 Triple += "v5tej";
528 break;
529 case ARMBuildAttrs::v6:
530 Triple += "v6";
531 break;
532 case ARMBuildAttrs::v6KZ:
533 Triple += "v6kz";
534 break;
535 case ARMBuildAttrs::v6T2:
536 Triple += "v6t2";
537 break;
538 case ARMBuildAttrs::v6K:
539 Triple += "v6k";
540 break;
541 case ARMBuildAttrs::v7: {
542 Optional<unsigned> ArchProfileAttr =
543 Attributes.getAttributeValue(ARMBuildAttrs::CPU_arch_profile);
544 if (ArchProfileAttr.hasValue() &&
545 ArchProfileAttr.getValue() == ARMBuildAttrs::MicroControllerProfile)
546 Triple += "v7m";
547 else
548 Triple += "v7";
549 break;
550 }
551 case ARMBuildAttrs::v6_M:
552 Triple += "v6m";
553 break;
554 case ARMBuildAttrs::v6S_M:
555 Triple += "v6sm";
556 break;
557 case ARMBuildAttrs::v7E_M:
558 Triple += "v7em";
559 break;
560 case ARMBuildAttrs::v8_A:
561 Triple += "v8a";
562 break;
563 case ARMBuildAttrs::v8_R:
564 Triple += "v8r";
565 break;
566 case ARMBuildAttrs::v8_M_Base:
567 Triple += "v8m.base";
568 break;
569 case ARMBuildAttrs::v8_M_Main:
570 Triple += "v8m.main";
571 break;
572 case ARMBuildAttrs::v8_1_M_Main:
573 Triple += "v8.1m.main";
574 break;
575 }
576 }
577 if (!isLittleEndian())
578 Triple += "eb";
579
580 TheTriple.setArchName(Triple);
581}
582
583std::vector<std::pair<Optional<DataRefImpl>, uint64_t>>
584ELFObjectFileBase::getPltAddresses() const {
585 std::string Err;
586 const auto Triple = makeTriple();
587 const auto *T = TargetRegistry::lookupTarget(Triple.str(), Err);
588 if (!T)
589 return {};
590 uint64_t JumpSlotReloc = 0;
591 switch (Triple.getArch()) {
592 case Triple::x86:
593 JumpSlotReloc = ELF::R_386_JUMP_SLOT;
594 break;
595 case Triple::x86_64:
596 JumpSlotReloc = ELF::R_X86_64_JUMP_SLOT;
597 break;
598 case Triple::aarch64:
599 case Triple::aarch64_be:
600 JumpSlotReloc = ELF::R_AARCH64_JUMP_SLOT;
601 break;
602 default:
603 return {};
604 }
605 std::unique_ptr<const MCInstrInfo> MII(T->createMCInstrInfo());
606 std::unique_ptr<const MCInstrAnalysis> MIA(
607 T->createMCInstrAnalysis(MII.get()));
608 if (!MIA)
609 return {};
610 Optional<SectionRef> Plt = None, RelaPlt = None, GotPlt = None;
611 for (const SectionRef &Section : sections()) {
612 Expected<StringRef> NameOrErr = Section.getName();
613 if (!NameOrErr) {
614 consumeError(NameOrErr.takeError());
615 continue;
616 }
617 StringRef Name = *NameOrErr;
618
619 if (Name == ".plt")
620 Plt = Section;
621 else if (Name == ".rela.plt" || Name == ".rel.plt")
622 RelaPlt = Section;
623 else if (Name == ".got.plt")
624 GotPlt = Section;
625 }
626 if (!Plt || !RelaPlt || !GotPlt)
627 return {};
628 Expected<StringRef> PltContents = Plt->getContents();
629 if (!PltContents) {
630 consumeError(PltContents.takeError());
631 return {};
632 }
633 auto PltEntries = MIA->findPltEntries(Plt->getAddress(),
634 arrayRefFromStringRef(*PltContents),
635 GotPlt->getAddress(), Triple);
636 // Build a map from GOT entry virtual address to PLT entry virtual address.
637 DenseMap<uint64_t, uint64_t> GotToPlt;
638 for (const auto &Entry : PltEntries)
639 GotToPlt.insert(std::make_pair(Entry.second, Entry.first));
640 // Find the relocations in the dynamic relocation table that point to
641 // locations in the GOT for which we know the corresponding PLT entry.
642 std::vector<std::pair<Optional<DataRefImpl>, uint64_t>> Result;
643 for (const auto &Relocation : RelaPlt->relocations()) {
644 if (Relocation.getType() != JumpSlotReloc)
645 continue;
646 auto PltEntryIter = GotToPlt.find(Relocation.getOffset());
647 if (PltEntryIter != GotToPlt.end()) {
648 symbol_iterator Sym = Relocation.getSymbol();
649 if (Sym == symbol_end())
650 Result.emplace_back(None, PltEntryIter->second);
651 else
652 Result.emplace_back(Sym->getRawDataRefImpl(), PltEntryIter->second);
653 }
654 }
655 return Result;
656}
657
658template <class ELFT>
659static Expected<std::vector<VersionEntry>>
660readDynsymVersionsImpl(const ELFFile<ELFT> &EF,
661 ELFObjectFileBase::elf_symbol_iterator_range Symbols) {
662 using Elf_Shdr = typename ELFT::Shdr;
663 const Elf_Shdr *VerSec = nullptr;
664 const Elf_Shdr *VerNeedSec = nullptr;
665 const Elf_Shdr *VerDefSec = nullptr;
666 // The user should ensure sections() can't fail here.
667 for (const Elf_Shdr &Sec : cantFail(EF.sections())) {
668 if (Sec.sh_type == ELF::SHT_GNU_versym)
669 VerSec = &Sec;
670 else if (Sec.sh_type == ELF::SHT_GNU_verdef)
671 VerDefSec = &Sec;
672 else if (Sec.sh_type == ELF::SHT_GNU_verneed)
673 VerNeedSec = &Sec;
674 }
675 if (!VerSec)
676 return std::vector<VersionEntry>();
677
678 Expected<SmallVector<Optional<VersionEntry>, 0>> MapOrErr =
679 EF.loadVersionMap(VerNeedSec, VerDefSec);
680 if (!MapOrErr)
681 return MapOrErr.takeError();
682
683 std::vector<VersionEntry> Ret;
684 size_t I = 0;
685 for (auto It = Symbols.begin(), E = Symbols.end(); It != E; ++It) {
686 ++I;
687 Expected<const typename ELFT::Versym *> VerEntryOrErr =
688 EF.template getEntry<typename ELFT::Versym>(*VerSec, I);
689 if (!VerEntryOrErr)
690 return createError("unable to read an entry with index " + Twine(I) +
691 " from " + describe(EF, *VerSec) + ": " +
692 toString(VerEntryOrErr.takeError()));
693
694 Expected<uint32_t> FlagsOrErr = It->getFlags();
695 if (!FlagsOrErr)
696 return createError("unable to read flags for symbol with index " +
697 Twine(I) + ": " + toString(FlagsOrErr.takeError()));
698
699 bool IsDefault;
700 Expected<StringRef> VerOrErr = EF.getSymbolVersionByIndex(
701 (*VerEntryOrErr)->vs_index, IsDefault, *MapOrErr,
702 (*FlagsOrErr) & SymbolRef::SF_Undefined);
703 if (!VerOrErr)
704 return createError("unable to get a version for entry " + Twine(I) +
705 " of " + describe(EF, *VerSec) + ": " +
706 toString(VerOrErr.takeError()));
707
708 Ret.push_back({(*VerOrErr).str(), IsDefault});
709 }
710
711 return Ret;
712}
713
714Expected<std::vector<VersionEntry>>
715ELFObjectFileBase::readDynsymVersions() const {
716 elf_symbol_iterator_range Symbols = getDynamicSymbolIterators();
717 if (const auto *Obj = dyn_cast<ELF32LEObjectFile>(this))
718 return readDynsymVersionsImpl(Obj->getELFFile(), Symbols);
719 if (const auto *Obj = dyn_cast<ELF32BEObjectFile>(this))
720 return readDynsymVersionsImpl(Obj->getELFFile(), Symbols);
721 if (const auto *Obj = dyn_cast<ELF64LEObjectFile>(this))
722 return readDynsymVersionsImpl(Obj->getELFFile(), Symbols);
723 return readDynsymVersionsImpl(cast<ELF64BEObjectFile>(this)->getELFFile(),
724 Symbols);
725}

/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
2.1
'ZB' is not equal to ZB_Undefined
2.1
'ZB' is not equal to ZB_Undefined
!= ZB_Undefined && Val == 0)
3
Assuming 'Val' is equal to 0
4
Taking true branch
133 return 64;
5
Returning the value 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);
2
Calling 'TrailingZerosCounter::count'
6
Returning from 'TrailingZerosCounter::count'
7
Returning the value 64
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));
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);
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