/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/rapidjson/internal/dtoa.h Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/rapidjson/internal/dtoa.h Source File#

Composable Kernel: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/rapidjson/internal/dtoa.h Source File
Go to the documentation of this file.
 // Tencent is pleased to support the open source community by making RapidJSON available.
 // 
 // Copyright (C) 2015 THL A29 Limited, a Tencent company, and Milo Yip.
 //
 // Licensed under the MIT License (the "License"); you may not use this file except
 // in compliance with the License. You may obtain a copy of the License at
 //
 // http://opensource.org/licenses/MIT
 //
 // Unless required by applicable law or agreed to in writing, software distributed 
 // under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR 
 // CONDITIONS OF ANY KIND, either express or implied. See the License for the 
 // specific language governing permissions and limitations under the License.
  
 // This is a C++ header-only implementation of Grisu2 algorithm from the publication:
 // Loitsch, Florian. "Printing floating-point numbers quickly and accurately with
 // integers." ACM Sigplan Notices 45.6 (2010): 233-243.
  
 #ifndef RAPIDJSON_DTOA_
 #define RAPIDJSON_DTOA_
  
 #include "itoa.h" // GetDigitsLut()
 #include "diyfp.h"
 #include "ieee754.h"
  
 RAPIDJSON_NAMESPACE_BEGIN
 namespace internal {
  
 #ifdef __GNUC__
 RAPIDJSON_DIAG_PUSH
 RAPIDJSON_DIAG_OFF(effc++)
 RAPIDJSON_DIAG_OFF(array-bounds) // some gcc versions generate wrong warnings https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59124
 #endif
  
 inline void GrisuRound(char* buffer, int len, uint64_t delta, uint64_t rest, uint64_t ten_kappa, uint64_t wp_w) {
     while (rest < wp_w && delta - rest >= ten_kappa &&
            (rest + ten_kappa < wp_w ||  
             wp_w - rest > rest + ten_kappa - wp_w)) {
         buffer[len - 1]--;
         rest += ten_kappa;
     }
 }
  
 inline int CountDecimalDigit32(uint32_t n) {
     // Simple pure C++ implementation was faster than __builtin_clz version in this situation.
     if (n < 10) return 1;
     if (n < 100) return 2;
     if (n < 1000) return 3;
     if (n < 10000) return 4;
     if (n < 100000) return 5;
     if (n < 1000000) return 6;
     if (n < 10000000) return 7;
     if (n < 100000000) return 8;
     // Will not reach 10 digits in DigitGen()
     //if (n < 1000000000) return 9;
     //return 10;
     return 9;
 }
  
 inline void DigitGen(const DiyFp& W, const DiyFp& Mp, uint64_t delta, char* buffer, int* len, int* K) {
     static const uint64_t kPow10[] = { 1ULL, 10ULL, 100ULL, 1000ULL, 10000ULL, 100000ULL, 1000000ULL, 10000000ULL, 100000000ULL,
                                        1000000000ULL, 10000000000ULL, 100000000000ULL, 1000000000000ULL,
                                        10000000000000ULL, 100000000000000ULL, 1000000000000000ULL,
                                        10000000000000000ULL, 100000000000000000ULL, 1000000000000000000ULL,
                                        10000000000000000000ULL };
     const DiyFp one(uint64_t(1) << -Mp.e, Mp.e);
     const DiyFp wp_w = Mp - W;
     uint32_t p1 = static_cast<uint32_t>(Mp.f >> -one.e);
     uint64_t p2 = Mp.f & (one.f - 1);
     int kappa = CountDecimalDigit32(p1); // kappa in [0, 9]
     *len = 0;
  
     while (kappa > 0) {
         uint32_t d = 0;
         switch (kappa) {
             case  9: d = p1 /  100000000; p1 %=  100000000; break;
             case  8: d = p1 /   10000000; p1 %=   10000000; break;
             case  7: d = p1 /    1000000; p1 %=    1000000; break;
             case  6: d = p1 /     100000; p1 %=     100000; break;
             case  5: d = p1 /      10000; p1 %=      10000; break;
             case  4: d = p1 /       1000; p1 %=       1000; break;
             case  3: d = p1 /        100; p1 %=        100; break;
             case  2: d = p1 /         10; p1 %=         10; break;
             case  1: d = p1;              p1 =           0; break;
             default:;
         }
         if (d || *len)
             buffer[(*len)++] = static_cast<char>('0' + static_cast<char>(d));
         kappa--;
         uint64_t tmp = (static_cast<uint64_t>(p1) << -one.e) + p2;
         if (tmp <= delta) {
             *K += kappa;
             GrisuRound(buffer, *len, delta, tmp, kPow10[kappa] << -one.e, wp_w.f);
             return;
         }
     }
  
     // kappa = 0
     for (;;) {
         p2 *= 10;
         delta *= 10;
         char d = static_cast<char>(p2 >> -one.e);
         if (d || *len)
             buffer[(*len)++] = static_cast<char>('0' + d);
         p2 &= one.f - 1;
         kappa--;
         if (p2 < delta) {
             *K += kappa;
             int index = -kappa;
             GrisuRound(buffer, *len, delta, p2, one.f, wp_w.f * (index < 20 ? kPow10[index] : 0));
             return;
         }
     }
 }
  
 inline void Grisu2(double value, char* buffer, int* length, int* K) {
     const DiyFp v(value);
     DiyFp w_m, w_p;
     v.NormalizedBoundaries(&w_m, &w_p);
  
     const DiyFp c_mk = GetCachedPower(w_p.e, K);
     const DiyFp W = v.Normalize() * c_mk;
     DiyFp Wp = w_p * c_mk;
     DiyFp Wm = w_m * c_mk;
     Wm.f++;
     Wp.f--;
     DigitGen(W, Wp, Wp.f - Wm.f, buffer, length, K);
 }
  
 inline char* WriteExponent(int K, char* buffer) {
     if (K < 0) {
         *buffer++ = '-';
         K = -K;
     }
  
     if (K >= 100) {
         *buffer++ = static_cast<char>('0' + static_cast<char>(K / 100));
         K %= 100;
         const char* d = GetDigitsLut() + K * 2;
         *buffer++ = d[0];
         *buffer++ = d[1];
     }
     else if (K >= 10) {
         const char* d = GetDigitsLut() + K * 2;
         *buffer++ = d[0];
         *buffer++ = d[1];
     }
     else
         *buffer++ = static_cast<char>('0' + static_cast<char>(K));
  
     return buffer;
 }
  
 inline char* Prettify(char* buffer, int length, int k, int maxDecimalPlaces) {
     const int kk = length + k;  // 10^(kk-1) <= v < 10^kk
  
     if (0 <= k && kk <= 21) {
         // 1234e7 -> 12340000000
         for (int i = length; i < kk; i++)
             buffer[i] = '0';
         buffer[kk] = '.';
         buffer[kk + 1] = '0';
         return &buffer[kk + 2];
     }
     else if (0 < kk && kk <= 21) {
         // 1234e-2 -> 12.34
         std::memmove(&buffer[kk + 1], &buffer[kk], static_cast<size_t>(length - kk));
         buffer[kk] = '.';
         if (0 > k + maxDecimalPlaces) {
             // When maxDecimalPlaces = 2, 1.2345 -> 1.23, 1.102 -> 1.1
             // Remove extra trailing zeros (at least one) after truncation.
             for (int i = kk + maxDecimalPlaces; i > kk + 1; i--)
                 if (buffer[i] != '0')
                     return &buffer[i + 1];
             return &buffer[kk + 2]; // Reserve one zero
         }
         else
             return &buffer[length + 1];
     }
     else if (-6 < kk && kk <= 0) {
         // 1234e-6 -> 0.001234
         const int offset = 2 - kk;
         std::memmove(&buffer[offset], &buffer[0], static_cast<size_t>(length));
         buffer[0] = '0';
         buffer[1] = '.';
         for (int i = 2; i < offset; i++)
             buffer[i] = '0';
         if (length - kk > maxDecimalPlaces) {
             // When maxDecimalPlaces = 2, 0.123 -> 0.12, 0.102 -> 0.1
             // Remove extra trailing zeros (at least one) after truncation.
             for (int i = maxDecimalPlaces + 1; i > 2; i--)
                 if (buffer[i] != '0')
                     return &buffer[i + 1];
             return &buffer[3]; // Reserve one zero
         }
         else
             return &buffer[length + offset];
     }
     else if (kk < -maxDecimalPlaces) {
         // Truncate to zero
         buffer[0] = '0';
         buffer[1] = '.';
         buffer[2] = '0';
         return &buffer[3];
     }
     else if (length == 1) {
         // 1e30
         buffer[1] = 'e';
         return WriteExponent(kk - 1, &buffer[2]);
     }
     else {
         // 1234e30 -> 1.234e33
         std::memmove(&buffer[2], &buffer[1], static_cast<size_t>(length - 1));
         buffer[1] = '.';
         buffer[length + 1] = 'e';
         return WriteExponent(kk - 1, &buffer[0 + length + 2]);
     }
 }
  
 inline char* dtoa(double value, char* buffer, int maxDecimalPlaces = 324) {
     RAPIDJSON_ASSERT(maxDecimalPlaces >= 1);
     Double d(value);
     if (d.IsZero()) {
         if (d.Sign())
             *buffer++ = '-';     // -0.0, Issue #289
         buffer[0] = '0';
         buffer[1] = '.';
         buffer[2] = '0';
         return &buffer[3];
     }
     else {
         if (value < 0) {
             *buffer++ = '-';
             value = -value;
         }
         int length, K;
         Grisu2(value, buffer, &length, &K);
         return Prettify(buffer, length, K, maxDecimalPlaces);
     }
 }
  
 #ifdef __GNUC__
 RAPIDJSON_DIAG_POP
 #endif
  
 } // namespace internal
 RAPIDJSON_NAMESPACE_END
  
 #endif // RAPIDJSON_DTOA_