xref: /freebsd/contrib/llvm-project/libcxx/src/include/ryu/d2s_intrinsics.h (revision 6580f5c38dd5b01aeeaed16b370f1a12423437f0)
1 //===----------------------------------------------------------------------===//
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 // Copyright (c) Microsoft Corporation.
10 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
11 
12 // Copyright 2018 Ulf Adams
13 // Copyright (c) Microsoft Corporation. All rights reserved.
14 
15 // Boost Software License - Version 1.0 - August 17th, 2003
16 
17 // Permission is hereby granted, free of charge, to any person or organization
18 // obtaining a copy of the software and accompanying documentation covered by
19 // this license (the "Software") to use, reproduce, display, distribute,
20 // execute, and transmit the Software, and to prepare derivative works of the
21 // Software, and to permit third-parties to whom the Software is furnished to
22 // do so, all subject to the following:
23 
24 // The copyright notices in the Software and this entire statement, including
25 // the above license grant, this restriction and the following disclaimer,
26 // must be included in all copies of the Software, in whole or in part, and
27 // all derivative works of the Software, unless such copies or derivative
28 // works are solely in the form of machine-executable object code generated by
29 // a source language processor.
30 
31 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
32 // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
33 // FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
34 // SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
35 // FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
36 // ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
37 // DEALINGS IN THE SOFTWARE.
38 
39 #ifndef _LIBCPP_SRC_INCLUDE_RYU_DS2_INTRINSICS_H
40 #define _LIBCPP_SRC_INCLUDE_RYU_DS2_INTRINSICS_H
41 
42 // Avoid formatting to keep the changes with the original code minimal.
43 // clang-format off
44 
45 #include <__assert>
46 #include <__config>
47 
48 #include "include/ryu/ryu.h"
49 
50 _LIBCPP_BEGIN_NAMESPACE_STD
51 
52 #if defined(_M_X64) && defined(_MSC_VER)
53 #define _LIBCPP_INTRINSIC128 1
54 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __ryu_umul128(const uint64_t __a, const uint64_t __b, uint64_t* const __productHi) {
55   return _umul128(__a, __b, __productHi);
56 }
57 
58 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __ryu_shiftright128(const uint64_t __lo, const uint64_t __hi, const uint32_t __dist) {
59   // For the __shiftright128 intrinsic, the shift value is always
60   // modulo 64.
61   // In the current implementation of the double-precision version
62   // of Ryu, the shift value is always < 64.
63   // (The shift value is in the range [49, 58].)
64   // Check this here in case a future change requires larger shift
65   // values. In this case this function needs to be adjusted.
66   _LIBCPP_ASSERT_INTERNAL(__dist < 64, "");
67   return __shiftright128(__lo, __hi, static_cast<unsigned char>(__dist));
68 }
69 
70 // ^^^ intrinsics available ^^^ / vvv __int128 available vvv
71 #elif defined(__SIZEOF_INT128__) && ( \
72     (defined(__clang__) && !defined(_MSC_VER)) || \
73     (defined(__GNUC__) && !defined(__clang__) && !defined(__CUDACC__)))
74 #define _LIBCPP_INTRINSIC128 1
75   // We have __uint128 support in clang or gcc
76 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __ryu_umul128(const uint64_t __a, const uint64_t __b, uint64_t* const __productHi) {
77   auto __temp = __a * (unsigned __int128)__b;
78   *__productHi = __temp >> 64;
79   return static_cast<uint64_t>(__temp);
80 }
81 
82 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __ryu_shiftright128(const uint64_t __lo, const uint64_t __hi, const uint32_t __dist) {
83   // In the current implementation of the double-precision version
84   // of Ryu, the shift value is always < 64.
85   // (The shift value is in the range [49, 58].)
86   // Check this here in case a future change requires larger shift
87   // values. In this case this function needs to be adjusted.
88   _LIBCPP_ASSERT_INTERNAL(__dist < 64, "");
89   auto __temp = __lo | ((unsigned __int128)__hi << 64);
90   // For x64 128-bit shfits using the `shrd` instruction and two 64-bit
91   // registers, the shift value is modulo 64.  Thus the `& 63` is free.
92   return static_cast<uint64_t>(__temp >> (__dist & 63));
93 }
94 #else // ^^^ __int128 available ^^^ / vvv intrinsics unavailable vvv
95 
96 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline _LIBCPP_ALWAYS_INLINE uint64_t __ryu_umul128(const uint64_t __a, const uint64_t __b, uint64_t* const __productHi) {
97   // TRANSITION, VSO-634761
98   // The casts here help MSVC to avoid calls to the __allmul library function.
99   const uint32_t __aLo = static_cast<uint32_t>(__a);
100   const uint32_t __aHi = static_cast<uint32_t>(__a >> 32);
101   const uint32_t __bLo = static_cast<uint32_t>(__b);
102   const uint32_t __bHi = static_cast<uint32_t>(__b >> 32);
103 
104   const uint64_t __b00 = static_cast<uint64_t>(__aLo) * __bLo;
105   const uint64_t __b01 = static_cast<uint64_t>(__aLo) * __bHi;
106   const uint64_t __b10 = static_cast<uint64_t>(__aHi) * __bLo;
107   const uint64_t __b11 = static_cast<uint64_t>(__aHi) * __bHi;
108 
109   const uint32_t __b00Lo = static_cast<uint32_t>(__b00);
110   const uint32_t __b00Hi = static_cast<uint32_t>(__b00 >> 32);
111 
112   const uint64_t __mid1 = __b10 + __b00Hi;
113   const uint32_t __mid1Lo = static_cast<uint32_t>(__mid1);
114   const uint32_t __mid1Hi = static_cast<uint32_t>(__mid1 >> 32);
115 
116   const uint64_t __mid2 = __b01 + __mid1Lo;
117   const uint32_t __mid2Lo = static_cast<uint32_t>(__mid2);
118   const uint32_t __mid2Hi = static_cast<uint32_t>(__mid2 >> 32);
119 
120   const uint64_t __pHi = __b11 + __mid1Hi + __mid2Hi;
121   const uint64_t __pLo = (static_cast<uint64_t>(__mid2Lo) << 32) | __b00Lo;
122 
123   *__productHi = __pHi;
124   return __pLo;
125 }
126 
127 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __ryu_shiftright128(const uint64_t __lo, const uint64_t __hi, const uint32_t __dist) {
128   // We don't need to handle the case __dist >= 64 here (see above).
129   _LIBCPP_ASSERT_INTERNAL(__dist < 64, "");
130 #ifdef _LIBCPP_64_BIT
131   _LIBCPP_ASSERT_INTERNAL(__dist > 0, "");
132   return (__hi << (64 - __dist)) | (__lo >> __dist);
133 #else // ^^^ 64-bit ^^^ / vvv 32-bit vvv
134   // Avoid a 64-bit shift by taking advantage of the range of shift values.
135   _LIBCPP_ASSERT_INTERNAL(__dist >= 32, "");
136   return (__hi << (64 - __dist)) | (static_cast<uint32_t>(__lo >> 32) >> (__dist - 32));
137 #endif // ^^^ 32-bit ^^^
138 }
139 
140 #endif // ^^^ intrinsics unavailable ^^^
141 
142 #ifndef _LIBCPP_64_BIT
143 
144 // Returns the high 64 bits of the 128-bit product of __a and __b.
145 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __umulh(const uint64_t __a, const uint64_t __b) {
146   // Reuse the __ryu_umul128 implementation.
147   // Optimizers will likely eliminate the instructions used to compute the
148   // low part of the product.
149   uint64_t __hi;
150   (void) __ryu_umul128(__a, __b, &__hi);
151   return __hi;
152 }
153 
154 // On 32-bit platforms, compilers typically generate calls to library
155 // functions for 64-bit divisions, even if the divisor is a constant.
156 //
157 // TRANSITION, LLVM-37932
158 //
159 // The functions here perform division-by-constant using multiplications
160 // in the same way as 64-bit compilers would do.
161 //
162 // NB:
163 // The multipliers and shift values are the ones generated by clang x64
164 // for expressions like x/5, x/10, etc.
165 
166 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div5(const uint64_t __x) {
167   return __umulh(__x, 0xCCCCCCCCCCCCCCCDu) >> 2;
168 }
169 
170 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div10(const uint64_t __x) {
171   return __umulh(__x, 0xCCCCCCCCCCCCCCCDu) >> 3;
172 }
173 
174 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div100(const uint64_t __x) {
175   return __umulh(__x >> 2, 0x28F5C28F5C28F5C3u) >> 2;
176 }
177 
178 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div1e8(const uint64_t __x) {
179   return __umulh(__x, 0xABCC77118461CEFDu) >> 26;
180 }
181 
182 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div1e9(const uint64_t __x) {
183   return __umulh(__x >> 9, 0x44B82FA09B5A53u) >> 11;
184 }
185 
186 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint32_t __mod1e9(const uint64_t __x) {
187   // Avoid 64-bit math as much as possible.
188   // Returning static_cast<uint32_t>(__x - 1000000000 * __div1e9(__x)) would
189   // perform 32x64-bit multiplication and 64-bit subtraction.
190   // __x and 1000000000 * __div1e9(__x) are guaranteed to differ by
191   // less than 10^9, so their highest 32 bits must be identical,
192   // so we can truncate both sides to uint32_t before subtracting.
193   // We can also simplify static_cast<uint32_t>(1000000000 * __div1e9(__x)).
194   // We can truncate before multiplying instead of after, as multiplying
195   // the highest 32 bits of __div1e9(__x) can't affect the lowest 32 bits.
196   return static_cast<uint32_t>(__x) - 1000000000 * static_cast<uint32_t>(__div1e9(__x));
197 }
198 
199 #else // ^^^ 32-bit ^^^ / vvv 64-bit vvv
200 
201 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div5(const uint64_t __x) {
202   return __x / 5;
203 }
204 
205 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div10(const uint64_t __x) {
206   return __x / 10;
207 }
208 
209 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div100(const uint64_t __x) {
210   return __x / 100;
211 }
212 
213 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div1e8(const uint64_t __x) {
214   return __x / 100000000;
215 }
216 
217 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint64_t __div1e9(const uint64_t __x) {
218   return __x / 1000000000;
219 }
220 
221 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint32_t __mod1e9(const uint64_t __x) {
222   return static_cast<uint32_t>(__x - 1000000000 * __div1e9(__x));
223 }
224 
225 #endif // ^^^ 64-bit ^^^
226 
227 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline uint32_t __pow5Factor(uint64_t __value) {
228   uint32_t __count = 0;
229   for (;;) {
230     _LIBCPP_ASSERT_INTERNAL(__value != 0, "");
231     const uint64_t __q = __div5(__value);
232     const uint32_t __r = static_cast<uint32_t>(__value) - 5 * static_cast<uint32_t>(__q);
233     if (__r != 0) {
234       break;
235     }
236     __value = __q;
237     ++__count;
238   }
239   return __count;
240 }
241 
242 // Returns true if __value is divisible by 5^__p.
243 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline bool __multipleOfPowerOf5(const uint64_t __value, const uint32_t __p) {
244   // I tried a case distinction on __p, but there was no performance difference.
245   return __pow5Factor(__value) >= __p;
246 }
247 
248 // Returns true if __value is divisible by 2^__p.
249 [[nodiscard]] _LIBCPP_HIDE_FROM_ABI inline bool __multipleOfPowerOf2(const uint64_t __value, const uint32_t __p) {
250   _LIBCPP_ASSERT_INTERNAL(__value != 0, "");
251   _LIBCPP_ASSERT_INTERNAL(__p < 64, "");
252   // __builtin_ctzll doesn't appear to be faster here.
253   return (__value & ((1ull << __p) - 1)) == 0;
254 }
255 
256 _LIBCPP_END_NAMESPACE_STD
257 
258 // clang-format on
259 
260 #endif // _LIBCPP_SRC_INCLUDE_RYU_DS2_INTRINSICS_H
261