clang -cc1 -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name RISCVMatInt.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~++20210903100615+fd66b44ec19e/build-llvm/lib/Target/RISCV/MCTargetDesc -resource-dir /usr/lib/llvm-14/lib/clang/14.0.0 -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/build-llvm/lib/Target/RISCV/MCTargetDesc -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/lib/Target/RISCV/MCTargetDesc -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/lib/Target/RISCV -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/build-llvm/lib/Target/RISCV -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/build-llvm/include -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/include -D 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~++20210903100615+fd66b44ec19e/build-llvm/lib/Target/RISCV/MCTargetDesc -fdebug-prefix-map=/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e=. -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-2021-09-04-040900-46481-1 -x c++ /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/lib/Target/RISCV/MCTargetDesc/RISCVMatInt.cpp
1 | |
2 | |
3 | |
4 | |
5 | |
6 | |
7 | |
8 | |
9 | #include "RISCVMatInt.h" |
10 | #include "MCTargetDesc/RISCVMCTargetDesc.h" |
11 | #include "llvm/ADT/APInt.h" |
12 | #include "llvm/Support/MathExtras.h" |
13 | using namespace llvm; |
14 | |
15 | static int getInstSeqCost(RISCVMatInt::InstSeq &Res, bool HasRVC) { |
16 | if (!HasRVC) |
17 | return Res.size(); |
18 | |
19 | int Cost = 0; |
20 | for (auto Instr : Res) { |
21 | bool Compressed; |
22 | switch (Instr.Opc) { |
23 | default: llvm_unreachable("Unexpected opcode"); |
24 | case RISCV::SLLI: |
25 | case RISCV::SRLI: |
26 | Compressed = true; |
27 | break; |
28 | case RISCV::ADDI: |
29 | case RISCV::ADDIW: |
30 | case RISCV::LUI: |
31 | Compressed = isInt<6>(Instr.Imm); |
32 | break; |
33 | case RISCV::ADDUW: |
34 | Compressed = false; |
35 | break; |
36 | } |
37 | |
38 | |
39 | |
40 | |
41 | |
42 | if (!Compressed) |
43 | Cost += 100; |
44 | else |
45 | Cost += 70; |
46 | } |
47 | return Cost; |
48 | } |
49 | |
50 | |
51 | static void generateInstSeqImpl(int64_t Val, |
52 | const FeatureBitset &ActiveFeatures, |
53 | RISCVMatInt::InstSeq &Res) { |
54 | bool IsRV64 = ActiveFeatures[RISCV::Feature64Bit]; |
55 | |
56 | if (isInt<32>(Val)) { |
| |
| 18 | | Returning from 'isInt<32U>' | |
|
| |
57 | |
58 | |
59 | |
60 | |
61 | |
62 | |
63 | |
64 | int64_t Hi20 = ((Val + 0x800) >> 12) & 0xFFFFF; |
65 | int64_t Lo12 = SignExtend64<12>(Val); |
66 | |
67 | if (Hi20) |
68 | Res.push_back(RISCVMatInt::Inst(RISCV::LUI, Hi20)); |
69 | |
70 | if (Lo12 || Hi20 == 0) { |
71 | unsigned AddiOpc = (IsRV64 && Hi20) ? RISCV::ADDIW : RISCV::ADDI; |
72 | Res.push_back(RISCVMatInt::Inst(AddiOpc, Lo12)); |
73 | } |
74 | return; |
75 | } |
76 | |
77 | assert(IsRV64 && "Can't emit >32-bit imm for non-RV64 target"); |
78 | |
79 | |
80 | |
81 | |
82 | |
83 | |
84 | |
85 | |
86 | |
87 | |
88 | |
89 | |
90 | |
91 | |
92 | |
93 | |
94 | |
95 | |
96 | |
97 | |
98 | |
99 | |
100 | |
101 | |
102 | int64_t Lo12 = SignExtend64<12>(Val); |
103 | int64_t Hi52 = ((uint64_t)Val + 0x800ull) >> 12; |
104 | int ShiftAmount = 12 + findFirstSet((uint64_t)Hi52); |
105 | Hi52 = SignExtend64(Hi52 >> (ShiftAmount - 12), 64 - ShiftAmount); |
| 20 | | The result of the right shift is undefined because the right operand is negative |
|
106 | |
107 | |
108 | |
109 | if (ShiftAmount > 12 && !isInt<12>(Hi52) && isInt<32>((uint64_t)Hi52 << 12)) { |
110 | |
111 | ShiftAmount -= 12; |
112 | Hi52 = (uint64_t)Hi52 << 12; |
113 | } |
114 | |
115 | generateInstSeqImpl(Hi52, ActiveFeatures, Res); |
116 | |
117 | Res.push_back(RISCVMatInt::Inst(RISCV::SLLI, ShiftAmount)); |
118 | if (Lo12) |
119 | Res.push_back(RISCVMatInt::Inst(RISCV::ADDI, Lo12)); |
120 | } |
121 | |
122 | namespace llvm { |
123 | namespace RISCVMatInt { |
124 | InstSeq generateInstSeq(int64_t Val, const FeatureBitset &ActiveFeatures) { |
125 | RISCVMatInt::InstSeq Res; |
126 | generateInstSeqImpl(Val, ActiveFeatures, Res); |
127 | |
128 | |
129 | |
130 | if (Val > 0 && Res.size() > 2) { |
| |
| 7 | | Assuming the condition is true | |
|
| |
131 | assert(ActiveFeatures[RISCV::Feature64Bit] && |
132 | "Expected RV32 to only need 2 instructions"); |
133 | unsigned LeadingZeros = countLeadingZeros((uint64_t)Val); |
134 | uint64_t ShiftedVal = (uint64_t)Val << LeadingZeros; |
135 | |
136 | |
137 | |
138 | ShiftedVal |= maskTrailingOnes<uint64_t>(LeadingZeros); |
139 | |
140 | RISCVMatInt::InstSeq TmpSeq; |
141 | generateInstSeqImpl(ShiftedVal, ActiveFeatures, TmpSeq); |
142 | TmpSeq.push_back(RISCVMatInt::Inst(RISCV::SRLI, LeadingZeros)); |
143 | |
144 | |
145 | if (TmpSeq.size() < Res.size()) { |
| 9 | | Assuming the condition is false | |
|
| |
146 | Res = TmpSeq; |
147 | |
148 | if (Res.size() <= 2) |
149 | return Res; |
150 | } |
151 | |
152 | |
153 | ShiftedVal &= maskTrailingZeros<uint64_t>(LeadingZeros); |
154 | TmpSeq.clear(); |
155 | generateInstSeqImpl(ShiftedVal, ActiveFeatures, TmpSeq); |
156 | TmpSeq.push_back(RISCVMatInt::Inst(RISCV::SRLI, LeadingZeros)); |
157 | |
158 | |
159 | if (TmpSeq.size() < Res.size()) { |
| 11 | | Assuming the condition is false | |
|
160 | Res = TmpSeq; |
161 | |
162 | if (Res.size() <= 2) |
163 | return Res; |
164 | } |
165 | |
166 | |
167 | |
168 | if (LeadingZeros == 32 && ActiveFeatures[RISCV::FeatureExtZba]) { |
| 12 | | Assuming 'LeadingZeros' is equal to 32 | |
|
| |
169 | |
170 | uint64_t LeadingOnesVal = Val | maskLeadingOnes<uint64_t>(LeadingZeros); |
171 | TmpSeq.clear(); |
172 | generateInstSeqImpl(LeadingOnesVal, ActiveFeatures, TmpSeq); |
| 14 | | Calling 'generateInstSeqImpl' | |
|
173 | TmpSeq.push_back(RISCVMatInt::Inst(RISCV::ADDUW, 0)); |
174 | |
175 | |
176 | if (TmpSeq.size() < Res.size()) { |
177 | Res = TmpSeq; |
178 | |
179 | if (Res.size() <= 2) |
180 | return Res; |
181 | } |
182 | } |
183 | } |
184 | |
185 | return Res; |
186 | } |
187 | |
188 | int getIntMatCost(const APInt &Val, unsigned Size, |
189 | const FeatureBitset &ActiveFeatures, |
190 | bool CompressionCost) { |
191 | bool IsRV64 = ActiveFeatures[RISCV::Feature64Bit]; |
192 | bool HasRVC = CompressionCost && ActiveFeatures[RISCV::FeatureStdExtC]; |
| 1 | Assuming 'CompressionCost' is false | |
|
193 | int PlatRegSize = IsRV64 ? 64 : 32; |
| |
194 | |
195 | |
196 | |
197 | int Cost = 0; |
198 | for (unsigned ShiftVal = 0; ShiftVal < Size; ShiftVal += PlatRegSize) { |
| 3 | | Assuming 'ShiftVal' is < 'Size' | |
|
| 4 | | Loop condition is true. Entering loop body | |
|
199 | APInt Chunk = Val.ashr(ShiftVal).sextOrTrunc(PlatRegSize); |
200 | InstSeq MatSeq = generateInstSeq(Chunk.getSExtValue(), ActiveFeatures); |
| 5 | | Calling 'generateInstSeq' | |
|
201 | Cost += getInstSeqCost(MatSeq, HasRVC); |
202 | } |
203 | return std::max(1, Cost); |
204 | } |
205 | } |
206 | } |
1 | |
2 | |
3 | |
4 | |
5 | |
6 | |
7 | |
8 | |
9 | |
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 | |
31 | |
32 | |
33 | extern "C" { |
34 | unsigned char _BitScanForward(unsigned long *_Index, unsigned long _Mask); |
35 | unsigned char _BitScanForward64(unsigned long *_Index, unsigned __int64 _Mask); |
36 | unsigned char _BitScanReverse(unsigned long *_Index, unsigned long _Mask); |
37 | unsigned char _BitScanReverse64(unsigned long *_Index, unsigned __int64 _Mask); |
38 | } |
39 | #endif |
40 | |
41 | namespace llvm { |
42 | |
43 | |
44 | enum ZeroBehavior { |
45 | |
46 | ZB_Undefined, |
47 | |
48 | ZB_Max, |
49 | |
50 | ZB_Width |
51 | }; |
52 | |
53 | |
54 | namespace numbers { |
55 | |
56 | |
57 | constexpr double e = 2.7182818284590452354, |
58 | egamma = .57721566490153286061, |
59 | ln2 = .69314718055994530942, |
60 | ln10 = 2.3025850929940456840, |
61 | log2e = 1.4426950408889634074, |
62 | log10e = .43429448190325182765, |
63 | pi = 3.1415926535897932385, |
64 | inv_pi = .31830988618379067154, |
65 | sqrtpi = 1.7724538509055160273, |
66 | inv_sqrtpi = .56418958354775628695, |
67 | sqrt2 = 1.4142135623730950488, |
68 | inv_sqrt2 = .70710678118654752440, |
69 | sqrt3 = 1.7320508075688772935, |
70 | inv_sqrt3 = .57735026918962576451, |
71 | phi = 1.6180339887498948482; |
72 | constexpr float ef = 2.71828183F, |
73 | egammaf = .577215665F, |
74 | ln2f = .693147181F, |
75 | ln10f = 2.30258509F, |
76 | log2ef = 1.44269504F, |
77 | log10ef = .434294482F, |
78 | pif = 3.14159265F, |
79 | inv_pif = .318309886F, |
80 | sqrtpif = 1.77245385F, |
81 | inv_sqrtpif = .564189584F, |
82 | sqrt2f = 1.41421356F, |
83 | inv_sqrt2f = .707106781F, |
84 | sqrt3f = 1.73205081F, |
85 | inv_sqrt3f = .577350269F, |
86 | phif = 1.61803399F; |
87 | } |
88 | |
89 | namespace detail { |
90 | template <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 | |
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__) || defined(_MSC_VER) |
114 | template <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) || defined(__GNUC__) |
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) |
130 | template <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) || defined(__GNUC__) |
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 | } |
147 | |
148 | |
149 | |
150 | |
151 | |
152 | |
153 | |
154 | |
155 | template <typename T> |
156 | unsigned 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 | |
163 | namespace detail { |
164 | template <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 | |
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__) || defined(_MSC_VER) |
183 | template <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) || defined(__GNUC__) |
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) |
199 | template <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) || defined(__GNUC__) |
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 | } |
216 | |
217 | |
218 | |
219 | |
220 | |
221 | |
222 | |
223 | |
224 | template <typename T> |
225 | unsigned 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 | |
233 | |
234 | |
235 | |
236 | |
237 | |
238 | |
239 | template <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 | |
247 | |
248 | template <typename T> T maskTrailingOnes(unsigned N) { |
249 | static_assert(std::is_unsigned<T>::value, "Invalid type!"); |
250 | const unsigned Bits = CHAR_BIT * sizeof(T); |
251 | assert(N <= Bits && "Invalid bit index"); |
252 | return N == 0 ? 0 : (T(-1) >> (Bits - N)); |
253 | } |
254 | |
255 | |
256 | |
257 | template <typename T> T maskLeadingOnes(unsigned N) { |
258 | return ~maskTrailingOnes<T>(CHAR_BIT * sizeof(T) - N); |
259 | } |
260 | |
261 | |
262 | |
263 | template <typename T> T maskTrailingZeros(unsigned N) { |
264 | return maskLeadingOnes<T>(CHAR_BIT * sizeof(T) - N); |
265 | } |
266 | |
267 | |
268 | |
269 | template <typename T> T maskLeadingZeros(unsigned N) { |
270 | return maskTrailingOnes<T>(CHAR_BIT * sizeof(T) - N); |
271 | } |
272 | |
273 | |
274 | |
275 | |
276 | |
277 | |
278 | |
279 | |
280 | template <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 | |
285 | |
286 | return countLeadingZeros(Val, ZB_Undefined) ^ |
287 | (std::numeric_limits<T>::digits - 1); |
288 | } |
289 | |
290 | |
291 | |
292 | |
293 | static 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 | |
304 | template <typename T> |
305 | T 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) |
316 | template<> |
317 | inline uint8_t reverseBits<uint8_t>(uint8_t Val) { |
318 | return __builtin_bitreverse8(Val); |
319 | } |
320 | #endif |
321 | |
322 | #if __has_builtin(__builtin_bitreverse16) |
323 | template<> |
324 | inline uint16_t reverseBits<uint16_t>(uint16_t Val) { |
325 | return __builtin_bitreverse16(Val); |
326 | } |
327 | #endif |
328 | |
329 | #if __has_builtin(__builtin_bitreverse32) |
330 | template<> |
331 | inline uint32_t reverseBits<uint32_t>(uint32_t Val) { |
332 | return __builtin_bitreverse32(Val); |
333 | } |
334 | #endif |
335 | |
336 | #if __has_builtin(__builtin_bitreverse64) |
337 | template<> |
338 | inline uint64_t reverseBits<uint64_t>(uint64_t Val) { |
339 | return __builtin_bitreverse64(Val); |
340 | } |
341 | #endif |
342 | |
343 | |
344 | |
345 | |
346 | |
347 | |
348 | constexpr inline uint32_t Hi_32(uint64_t Value) { |
349 | return static_cast<uint32_t>(Value >> 32); |
350 | } |
351 | |
352 | |
353 | constexpr inline uint32_t Lo_32(uint64_t Value) { |
354 | return static_cast<uint32_t>(Value); |
355 | } |
356 | |
357 | |
358 | constexpr inline uint64_t Make_64(uint32_t High, uint32_t Low) { |
359 | return ((uint64_t)High << 32) | (uint64_t)Low; |
360 | } |
361 | |
362 | |
363 | template <unsigned N> constexpr inline bool isInt(int64_t x) { |
364 | return N >= 64 || (-(INT64_C(1)<<(N-1)) <= x && x < (INT64_C(1)<<(N-1))); |
365 | } |
366 | |
367 | template <> constexpr inline bool isInt<8>(int64_t x) { |
368 | return static_cast<int8_t>(x) == x; |
369 | } |
370 | template <> constexpr inline bool isInt<16>(int64_t x) { |
371 | return static_cast<int16_t>(x) == x; |
372 | } |
373 | template <> constexpr inline bool isInt<32>(int64_t x) { |
374 | return static_cast<int32_t>(x) == x; |
| 16 | | Assuming 'x' is not equal to 'x' | |
|
| 17 | | Returning zero, which participates in a condition later | |
|
375 | } |
376 | |
377 | |
378 | template <unsigned N, unsigned S> |
379 | constexpr 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) << S) == 0); |
384 | } |
385 | |
386 | |
387 | |
388 | |
389 | |
390 | |
391 | |
392 | |
393 | |
394 | template <unsigned N> |
395 | constexpr 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) << (N)); |
398 | } |
399 | template <unsigned N> |
400 | constexpr inline std::enable_if_t<N >= 64, bool> isUInt(uint64_t) { |
401 | return true; |
402 | } |
403 | |
404 | |
405 | template <> constexpr inline bool isUInt<8>(uint64_t x) { |
406 | return static_cast<uint8_t>(x) == x; |
407 | } |
408 | template <> constexpr inline bool isUInt<16>(uint64_t x) { |
409 | return static_cast<uint16_t>(x) == x; |
410 | } |
411 | template <> constexpr inline bool isUInt<32>(uint64_t x) { |
412 | return static_cast<uint32_t>(x) == x; |
413 | } |
414 | |
415 | |
416 | template <unsigned N, unsigned S> |
417 | constexpr 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 | |
423 | |
424 | return isUInt<N + S>(x) && (x % (UINT64_C(1) << S) == 0); |
425 | } |
426 | |
427 | |
428 | inline uint64_t maxUIntN(uint64_t N) { |
429 | assert(N > 0 && N <= 64 && "integer width out of range"); |
430 | |
431 | |
432 | |
433 | |
434 | |
435 | return UINT64_MAX >> (64 - N); |
436 | } |
437 | |
438 | |
439 | inline int64_t minIntN(int64_t N) { |
440 | assert(N > 0 && N <= 64 && "integer width out of range"); |
441 | |
442 | return UINT64_C(1) + ~(UINT64_C(1) << (N - 1)); |
443 | } |
444 | |
445 | |
446 | inline int64_t maxIntN(int64_t N) { |
447 | assert(N > 0 && N <= 64 && "integer width out of range"); |
448 | |
449 | |
450 | |
451 | return (UINT64_C(1) << (N - 1)) - 1; |
452 | } |
453 | |
454 | |
455 | inline bool isUIntN(unsigned N, uint64_t x) { |
456 | return N >= 64 || x <= maxUIntN(N); |
457 | } |
458 | |
459 | |
460 | inline bool isIntN(unsigned N, int64_t x) { |
461 | return N >= 64 || (minIntN(N) <= x && x <= maxIntN(N)); |
462 | } |
463 | |
464 | |
465 | |
466 | |
467 | constexpr inline bool isMask_32(uint32_t Value) { |
468 | return Value && ((Value + 1) & Value) == 0; |
469 | } |
470 | |
471 | |
472 | |
473 | constexpr inline bool isMask_64(uint64_t Value) { |
474 | return Value && ((Value + 1) & Value) == 0; |
475 | } |
476 | |
477 | |
478 | |
479 | constexpr inline bool isShiftedMask_32(uint32_t Value) { |
480 | return Value && isMask_32((Value - 1) | Value); |
481 | } |
482 | |
483 | |
484 | |
485 | constexpr inline bool isShiftedMask_64(uint64_t Value) { |
486 | return Value && isMask_64((Value - 1) | Value); |
487 | } |
488 | |
489 | |
490 | |
491 | constexpr inline bool isPowerOf2_32(uint32_t Value) { |
492 | return Value && !(Value & (Value - 1)); |
493 | } |
494 | |
495 | |
496 | constexpr inline bool isPowerOf2_64(uint64_t Value) { |
497 | return Value && !(Value & (Value - 1)); |
498 | } |
499 | |
500 | |
501 | |
502 | |
503 | |
504 | |
505 | |
506 | |
507 | |
508 | template <typename T> |
509 | unsigned 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 | |
517 | |
518 | |
519 | |
520 | |
521 | |
522 | |
523 | |
524 | template <typename T> |
525 | unsigned 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 | |
532 | namespace detail { |
533 | template <typename T, std::size_t SizeOfT> struct PopulationCounter { |
534 | static unsigned count(T Value) { |
535 | |
536 | static_assert(SizeOfT <= 4, "Not implemented!"); |
537 | #if defined(__GNUC__) |
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 | |
548 | template <typename T> struct PopulationCounter<T, 8> { |
549 | static unsigned count(T Value) { |
550 | #if defined(__GNUC__) |
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 | } |
562 | |
563 | |
564 | |
565 | |
566 | template <typename T> |
567 | inline 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 | |
575 | |
576 | template <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 | |
582 | template <> constexpr inline size_t CTLog2<1>() { return 0; } |
583 | |
584 | |
585 | inline 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 | |
594 | |
595 | |
596 | inline unsigned Log2_32(uint32_t Value) { |
597 | return 31 - countLeadingZeros(Value); |
598 | } |
599 | |
600 | |
601 | |
602 | inline unsigned Log2_64(uint64_t Value) { |
603 | return 63 - countLeadingZeros(Value); |
604 | } |
605 | |
606 | |
607 | |
608 | |
609 | inline unsigned Log2_32_Ceil(uint32_t Value) { |
610 | return 32 - countLeadingZeros(Value - 1); |
611 | } |
612 | |
613 | |
614 | |
615 | inline unsigned Log2_64_Ceil(uint64_t Value) { |
616 | return 64 - countLeadingZeros(Value - 1); |
617 | } |
618 | |
619 | |
620 | template <typename T> |
621 | inline 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 | |
630 | inline uint64_t GreatestCommonDivisor64(uint64_t A, uint64_t B) { |
631 | return greatestCommonDivisor<uint64_t>(A, B); |
632 | } |
633 | |
634 | |
635 | inline 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 | |
643 | inline 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 | |
651 | |
652 | |
653 | inline 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 | |
661 | |
662 | |
663 | inline 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 | |
671 | |
672 | constexpr inline uint64_t MinAlign(uint64_t A, uint64_t B) { |
673 | |
674 | |
675 | |
676 | |
677 | |
678 | return (A | B) & (1 + ~(A | B)); |
679 | } |
680 | |
681 | |
682 | |
683 | inline 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 | |
694 | |
695 | inline uint64_t PowerOf2Floor(uint64_t A) { |
696 | if (!A) return 0; |
697 | return 1ull << (63 - countLeadingZeros(A, ZB_Undefined)); |
698 | } |
699 | |
700 | |
701 | |
702 | inline uint64_t PowerOf2Ceil(uint64_t A) { |
703 | if (!A) |
704 | return 0; |
705 | return NextPowerOf2(A - 1); |
706 | } |
707 | |
708 | |
709 | |
710 | |
711 | |
712 | |
713 | |
714 | |
715 | |
716 | |
717 | |
718 | |
719 | |
720 | |
721 | |
722 | |
723 | |
724 | |
725 | |
726 | |
727 | |
728 | inline uint64_t alignTo(uint64_t Value, uint64_t Align, uint64_t Skew = 0) { |
729 | assert(Align != 0u && "Align can't be 0."); |
730 | Skew %= Align; |
731 | return (Value + Align - 1 - Skew) / Align * Align + Skew; |
732 | } |
733 | |
734 | |
735 | |
736 | template <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 | |
742 | inline uint64_t divideCeil(uint64_t Numerator, uint64_t Denominator) { |
743 | return alignTo(Numerator, Denominator) / Denominator; |
744 | } |
745 | |
746 | |
747 | inline uint64_t divideNearest(uint64_t Numerator, uint64_t Denominator) { |
748 | return (Numerator + (Denominator / 2)) / Denominator; |
749 | } |
750 | |
751 | |
752 | |
753 | inline uint64_t alignDown(uint64_t Value, uint64_t Align, uint64_t Skew = 0) { |
754 | assert(Align != 0u && "Align can't be 0."); |
755 | Skew %= Align; |
756 | return (Value - Skew) / Align * Align + Skew; |
757 | } |
758 | |
759 | |
760 | |
761 | template <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 | |
768 | |
769 | inline int32_t SignExtend32(uint32_t X, unsigned B) { |
770 | assert(B > 0 && "Bit width can't be 0."); |
771 | assert(B <= 32 && "Bit width out of range."); |
772 | return int32_t(X << (32 - B)) >> (32 - B); |
773 | } |
774 | |
775 | |
776 | |
777 | template <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 | |
784 | |
785 | inline int64_t SignExtend64(uint64_t X, unsigned B) { |
786 | assert(B > 0 && "Bit width can't be 0."); |
787 | assert(B <= 64 && "Bit width out of range."); |
788 | return int64_t(X << (64 - B)) >> (64 - B); |
789 | } |
790 | |
791 | |
792 | |
793 | template <typename T> |
794 | std::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 | |
799 | |
800 | |
801 | template <typename T> |
802 | std::enable_if_t<std::is_unsigned<T>::value, T> |
803 | SaturatingAdd(T X, T Y, bool *ResultOverflowed = nullptr) { |
804 | bool Dummy; |
805 | bool &Overflowed = ResultOverflowed ? *ResultOverflowed : Dummy; |
806 | |
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 | |
816 | |
817 | |
818 | template <typename T> |
819 | std::enable_if_t<std::is_unsigned<T>::value, T> |
820 | SaturatingMultiply(T X, T Y, bool *ResultOverflowed = nullptr) { |
821 | bool Dummy; |
822 | bool &Overflowed = ResultOverflowed ? *ResultOverflowed : Dummy; |
823 | |
824 | |
825 | |
826 | |
827 | |
828 | |
829 | Overflowed = false; |
830 | |
831 | |
832 | |
833 | |
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 | |
846 | |
847 | |
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 | |
861 | |
862 | |
863 | |
864 | template <typename T> |
865 | std::enable_if_t<std::is_unsigned<T>::value, T> |
866 | SaturatingMultiplyAdd(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 | |
878 | extern const float huge_valf; |
879 | |
880 | |
881 | |
882 | |
883 | template <typename T> |
884 | std::enable_if_t<std::is_signed<T>::value, T> AddOverflow(T X, T Y, T &Result) { |
885 | #if __has_builtin(__builtin_add_overflow) |
886 | return __builtin_add_overflow(X, Y, &Result); |
887 | #else |
888 | |
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 | |
895 | Result = static_cast<T>(UResult); |
896 | |
897 | |
898 | if (X > 0 && Y > 0) |
899 | return Result <= 0; |
900 | |
901 | if (X < 0 && Y < 0) |
902 | return Result >= 0; |
903 | return false; |
904 | #endif |
905 | } |
906 | |
907 | |
908 | |
909 | template <typename T> |
910 | std::enable_if_t<std::is_signed<T>::value, T> SubOverflow(T X, T Y, T &Result) { |
911 | #if __has_builtin(__builtin_sub_overflow) |
912 | return __builtin_sub_overflow(X, Y, &Result); |
913 | #else |
914 | |
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 | |
921 | Result = static_cast<T>(UResult); |
922 | |
923 | |
924 | if (X <= 0 && Y > 0) |
925 | return Result >= 0; |
926 | |
927 | if (X >= 0 && Y < 0) |
928 | return Result <= 0; |
929 | return false; |
930 | #endif |
931 | } |
932 | |
933 | |
934 | |
935 | template <typename T> |
936 | std::enable_if_t<std::is_signed<T>::value, T> MulOverflow(T X, T Y, T &Result) { |
937 | |
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 | |
944 | const bool IsNegative = (X < 0) ^ (Y < 0); |
945 | Result = IsNegative ? (0 - UResult) : UResult; |
946 | |
947 | |
948 | if (UX == 0 || UY == 0) |
949 | return false; |
950 | |
951 | |
952 | |
953 | |
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 | } |
961 | |
962 | #endif |