From 65eb5d12d4aa8c2e4f0d04a46cd27dacc9601241 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Branimir=20Karad=C5=BEi=C4=87?= Date: Sat, 17 Jun 2017 14:50:29 -0700 Subject: [PATCH] Added fromString. --- include/bx/string.h | 3 + src/dtoa.cpp | 539 +++++++++++++++++++++++++++++++++++++++++- tests/string_test.cpp | 43 ++++ 3 files changed, 577 insertions(+), 8 deletions(-) diff --git a/include/bx/string.h b/include/bx/string.h index bcecc8b..447edf4 100644 --- a/include/bx/string.h +++ b/include/bx/string.h @@ -239,6 +239,9 @@ namespace bx /// int32_t toString(char* _out, int32_t _max, uint64_t _value, uint32_t _base = 10); + /// + double fromString(const char* _str); + } // namespace bx #include "inline/string.inl" diff --git a/src/dtoa.cpp b/src/dtoa.cpp index 57480ab..6b96c06 100644 --- a/src/dtoa.cpp +++ b/src/dtoa.cpp @@ -105,15 +105,15 @@ namespace bx void NormalizedBoundaries(DiyFp* minus, DiyFp* plus) const { - DiyFp pl = DiyFp((f << 1) + 1, e - 1).NormalizeBoundary(); - DiyFp mi = (f == kDpHiddenBit) ? DiyFp((f << 2) - 1, e - 2) : DiyFp((f << 1) - 1, e - 1); + DiyFp pl = DiyFp( (f << 1) + 1, e - 1).NormalizeBoundary(); + DiyFp mi = (f == kDpHiddenBit) ? DiyFp( (f << 2) - 1, e - 2) : DiyFp( (f << 1) - 1, e - 1); mi.f <<= mi.e - pl.e; mi.e = pl.e; *plus = pl; *minus = mi; } -#define UINT64_C2(h, l) ((static_cast(h) << 32) | static_cast(l)) +#define UINT64_C2(h, l) ( (static_cast(h) << 32) | static_cast(l) ) static const int32_t kDiySignificandSize = 64; static const int32_t kDpSignificandSize = 52; @@ -217,10 +217,10 @@ namespace bx k++; } - uint32_t index = static_cast((k >> 3) + 1); - *K = -(-348 + static_cast(index << 3)); // decimal exponent no need lookup table + uint32_t index = static_cast( (k >> 3) + 1); + *K = -(-348 + static_cast(index << 3) ); // decimal exponent no need lookup table - BX_CHECK(index < sizeof(s_kCachedPowers_F) / sizeof(s_kCachedPowers_F[0])); + BX_CHECK(index < sizeof(s_kCachedPowers_F) / sizeof(s_kCachedPowers_F[0]) ); return DiyFp(s_kCachedPowers_F[index], s_kCachedPowers_E[index]); } @@ -228,7 +228,7 @@ namespace bx { while (rest < wp_w && delta - rest >= ten_kappa - && (rest + ten_kappa < wp_w || wp_w - rest > rest + ten_kappa - wp_w)) + && (rest + ten_kappa < wp_w || wp_w - rest > rest + ten_kappa - wp_w) ) { buffer[len - 1]--; rest += ten_kappa; @@ -256,7 +256,7 @@ namespace bx const DiyFp wp_w = Mp - W; uint32_t p1 = static_cast(Mp.f >> -one.e); uint64_t p2 = Mp.f & (one.f - 1); - int32_t kappa = static_cast(CountDecimalDigit32(p1)); + int32_t kappa = static_cast(CountDecimalDigit32(p1) ); *len = 0; while (kappa > 0) @@ -557,4 +557,527 @@ namespace bx return toStringUnsigned(_dst, _max, _value, _base); } + /* + * MIT License + * + * Copyright (c) 2016 Grzegorz Kraszewski + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + + /* + * IMPORTANT + * + * The code works in "round towards zero" mode. This is different from + * GCC standard library strtod(), which uses "round half to even" rule. + * Therefore it cannot be used as a direct drop-in replacement, as in + * some cases results will be different on the least significant bit of + * mantissa. Read more in the README.md file. + */ + +#define DIGITS 18 + +#define DOUBLE_PLUS_ZERO 0x0000000000000000ull +#define DOUBLE_MINUS_ZERO 0x8000000000000000ull +#define DOUBLE_PLUS_INFINITY 0x7ff0000000000000ull +#define DOUBLE_MINUS_INFINITY 0xfff0000000000000ull + + union HexDouble + { + double d; + uint64_t u; + }; + +#define lsr96(s2, s1, s0, d2, d1, d0) \ + d0 = ( (s0) >> 1) | ( ( (s1) & 1) << 31); \ + d1 = ( (s1) >> 1) | ( ( (s2) & 1) << 31); \ + d2 = (s2) >> 1; + +#define lsl96(s2, s1, s0, d2, d1, d0) \ + d2 = ( (s2) << 1) | ( ( (s1) & (1 << 31) ) >> 31); \ + d1 = ( (s1) << 1) | ( ( (s0) & (1 << 31) ) >> 31); \ + d0 = (s0) << 1; + + /* + * Undefine the below constant if your processor or compiler is slow + * at 64-bit arithmetic. This is a rare case however. 64-bit macros are + * better for deeply pipelined CPUs (no conditional execution), are + * very efficient for 64-bit processors and also fast on 32-bit processors + * featuring extended precision arithmetic (x86, PowerPC_32, M68k and probably + * more). + */ + +#define USE_64BIT_FOR_ADDSUB_MACROS 1 + +#if USE_64BIT_FOR_ADDSUB_MACROS + +#define add96(s2, s1, s0, d2, d1, d0) { \ + uint64_t w; \ + w = (uint64_t)(s0) + (uint64_t)(d0); \ + (s0) = w; \ + w >>= 32; \ + w += (uint64_t)(s1) + (uint64_t)(d1); \ + (s1) = w; \ + w >>= 32; \ + w += (uint64_t)(s2) + (uint64_t)(d2); \ + (s2) = w; } + +#define sub96(s2, s1, s0, d2, d1, d0) { \ + uint64_t w; \ + w = (uint64_t)(s0) - (uint64_t)(d0); \ + (s0) = w; \ + w >>= 32; \ + w += (uint64_t)(s1) - (uint64_t)(d1); \ + (s1) = w; \ + w >>= 32; \ + w += (uint64_t)(s2) - (uint64_t)(d2); \ + (s2) = w; } + +#else + +#define add96(s2, s1, s0, d2, d1, d0) { \ + uint32_t _x, _c; \ + _x = (s0); (s0) += (d0); \ + if ( (s0) < _x) _c = 1; else _c = 0; \ + _x = (s1); (s1) += (d1) + _c; \ + if ( ( (s1) < _x) || ( ( (s1) == _x) && _c) ) _c = 1; else _c = 0; \ + (s2) += (d2) + _c; } + +#define sub96(s2, s1, s0, d2, d1, d0) { \ + uint32_t _x, _c; \ + _x = (s0); (s0) -= (d0); \ + if ( (s0) > _x) _c = 1; else _c = 0; \ + _x = (s1); (s1) -= (d1) + _c; \ + if ( ( (s1) > _x) || ( ( (s1) == _x) && _c) ) _c = 1; else _c = 0; \ + (s2) -= (d2) + _c; } + +#endif /* USE_64BIT_FOR_ADDSUB_MACROS */ + + /* parser state machine states */ + +#define FSM_A 0 +#define FSM_B 1 +#define FSM_C 2 +#define FSM_D 3 +#define FSM_E 4 +#define FSM_F 5 +#define FSM_G 6 +#define FSM_H 7 +#define FSM_I 8 +#define FSM_STOP 9 + + /* Modify these if working with non-ASCII encoding */ + +#define DPOINT '.' +#define ISEXP(x) ( ( (x) == 'E') || ( (x) == 'e') ) + + /* The structure is filled by parser, then given to converter. */ + struct PrepNumber + { + int negative; /* 0 if positive number, 1 if negative */ + int32_t exponent; /* power of 10 exponent */ + uint64_t mantissa; /* integer mantissa */ + }; + + /* Possible parser return values. */ + +#define PARSER_OK 0 // parser finished OK +#define PARSER_PZERO 1 // no digits or number is smaller than +-2^-1022 +#define PARSER_MZERO 2 // number is negative, module smaller +#define PARSER_PINF 3 // number is higher than +HUGE_VAL +#define PARSER_MINF 4 // number is lower than -HUGE_VAL + + /* GETC() macro gets next character from processed string. */ + +#define GETC(s) *s++ + + static int parser(const char *s, struct PrepNumber *pn) + { + int state = FSM_A; + int digx = 0, c = ' '; /* initial value for kicking off the state machine */ + int result = PARSER_OK; + int expneg = 0; + int32_t expexp = 0; + + while (state != FSM_STOP) + { + switch (state) + { + case FSM_A: + if (isSpace(c) ) + { + c = GETC(s); + } + else + { + state = FSM_B; + } + break; + + case FSM_B: + state = FSM_C; + + if (c == '+') + { + c = GETC(s); + } + else if (c == '-') + { + pn->negative = 1; + c = GETC(s); + } + else if (isNumeric(c) ) + { + } + else if (c == DPOINT) + { + } + else + { + state = FSM_STOP; + } + break; + + case FSM_C: + if (c == '0') + { + c = GETC(s); + } + else if (c == DPOINT) + { + c = GETC(s); + state = FSM_D; + } + else + { + state = FSM_E; + } + break; + + case FSM_D: + if (c == '0') + { + c = GETC(s); + if (pn->exponent > -2147483647) pn->exponent--; + } + else + { + state = FSM_F; + } + break; + + case FSM_E: + if (isNumeric(c) ) + { + if (digx < DIGITS) + { + pn->mantissa *= 10; + pn->mantissa += c - '0'; + digx++; + } + else if (pn->exponent < 2147483647) + { + pn->exponent++; + } + + c = GETC(s); + } + else if (c == DPOINT) + { + c = GETC(s); + state = FSM_F; + } + else + { + state = FSM_F; + } + break; + + case FSM_F: + if (isNumeric(c) ) + { + if (digx < DIGITS) + { + pn->mantissa *= 10; + pn->mantissa += c - '0'; + pn->exponent--; + digx++; + } + + c = GETC(s); + } + else if (ISEXP(c) ) + { + c = GETC(s); + state = FSM_G; + } + else + { + state = FSM_G; + } + break; + + case FSM_G: + if (c == '+') + { + c = GETC(s); + } + else if (c == '-') + { + expneg = 1; + c = GETC(s); + } + + state = FSM_H; + break; + + case FSM_H: + if (c == '0') + { + c = GETC(s); + } + else + { + state = FSM_I; + } + break; + + case FSM_I: + if (isNumeric(c) ) + { + if (expexp < 214748364) + { + expexp *= 10; + expexp += c - '0'; + } + + c = GETC(s); + } + else + { + state = FSM_STOP; + } + break; + } + } + + if (expneg) + { + expexp = -expexp; + } + + pn->exponent += expexp; + + if (pn->mantissa == 0) + { + if (pn->negative) + { + result = PARSER_MZERO; + } + else + { + result = PARSER_PZERO; + } + } + else if (pn->exponent > 309) + { + if (pn->negative) + { + result = PARSER_MINF; + } + else + { + result = PARSER_PINF; + } + } + else if (pn->exponent < -328) + { + if (pn->negative) + { + result = PARSER_MZERO; + } + else + { + result = PARSER_PZERO; + } + } + + return result; + } + + static double converter(struct PrepNumber *pn) + { + int binexp = 92; + union HexDouble hd; + uint32_t s2, s1, s0; /* 96-bit precision integer */ + uint32_t q2, q1, q0; /* 96-bit precision integer */ + uint32_t r2, r1, r0; /* 96-bit precision integer */ + uint32_t mask28 = 0xF << 28; + + hd.u = 0; + + s0 = (uint32_t)(pn->mantissa & 0xFFFFFFFF); + s1 = (uint32_t)(pn->mantissa >> 32); + s2 = 0; + + while (pn->exponent > 0) + { + lsl96(s2, s1, s0, q2, q1, q0); // q = p << 1 + lsl96(q2, q1, q0, r2, r1, r0); // r = p << 2 + lsl96(r2, r1, r0, s2, s1, s0); // p = p << 3 + add96(s2, s1, s0, q2, q1, q0); // p = (p << 3) + (p << 1) + + pn->exponent--; + + while (s2 & mask28) + { + lsr96(s2, s1, s0, q2, q1, q0); + binexp++; + s2 = q2; + s1 = q1; + s0 = q0; + } + } + + while (pn->exponent < 0) + { + while (!(s2 & (1 << 31) ) ) + { + lsl96(s2, s1, s0, q2, q1, q0); + binexp--; + s2 = q2; + s1 = q1; + s0 = q0; + } + + q2 = s2 / 10; + r1 = s2 % 10; + r2 = (s1 >> 8) | (r1 << 24); + q1 = r2 / 10; + r1 = r2 % 10; + r2 = ( (s1 & 0xFF) << 16) | (s0 >> 16) | (r1 << 24); + r0 = r2 / 10; + r1 = r2 % 10; + q1 = (q1 << 8) | ( (r0 & 0x00FF0000) >> 16); + q0 = r0 << 16; + r2 = (s0 & 0xFFFF) | (r1 << 16); + q0 |= r2 / 10; + s2 = q2; + s1 = q1; + s0 = q0; + + pn->exponent++; + } + + if (s2 || s1 || s0) + { + while (!(s2 & mask28) ) + { + lsl96(s2, s1, s0, q2, q1, q0); + binexp--; + s2 = q2; + s1 = q1; + s0 = q0; + } + } + + binexp += 1023; + + if (binexp > 2046) + { + if (pn->negative) + { + hd.u = DOUBLE_MINUS_INFINITY; + } + else + { + hd.u = DOUBLE_PLUS_INFINITY; + } + } + else if (binexp < 1) + { + if (pn->negative) + { + hd.u = DOUBLE_MINUS_ZERO; + } + } + else if (s2) + { + uint64_t q; + uint64_t binexs2 = (uint64_t)binexp; + + binexs2 <<= 52; + q = ( (uint64_t)(s2 & ~mask28) << 24) + | ( ( (uint64_t)s1 + 128) >> 8) | binexs2; + + if (pn->negative) + { + q |= (1ULL << 63); + } + + hd.u = q; + } + + return hd.d; + } + + double fromString(const char* _str) + { + struct PrepNumber pn; + union HexDouble hd; + int i; + double result; + + pn.mantissa = 0; + pn.negative = 0; + pn.exponent = 0; + hd.u = DOUBLE_PLUS_ZERO; + + i = parser(_str, &pn); + + switch (i) + { + case PARSER_OK: + result = converter(&pn); + break; + + case PARSER_PZERO: + result = hd.d; + break; + + case PARSER_MZERO: + hd.u = DOUBLE_MINUS_ZERO; + result = hd.d; + break; + + case PARSER_PINF: + hd.u = DOUBLE_PLUS_INFINITY; + result = hd.d; + break; + + case PARSER_MINF: + hd.u = DOUBLE_MINUS_INFINITY; + result = hd.d; + break; + } + + return result; + } + } // namespace bx diff --git a/tests/string_test.cpp b/tests/string_test.cpp index 59f1380..4f5d374 100644 --- a/tests/string_test.cpp +++ b/tests/string_test.cpp @@ -182,6 +182,49 @@ TEST_CASE("toString double", "") REQUIRE(testToString(0.0000000001, "1e-10") ); } +static bool testFromString(double _value, const char* _input) +{ + char tmp[1024]; + int32_t num = bx::toString(tmp, BX_COUNTOF(tmp), _value); + const double lhs = bx::fromString(tmp); + const double rhs = bx::fromString(_input); + + if (lhs == rhs) + { + return true; + } + + printf("result '%f', input '%s'\n", _value, _input); + return false; +} + +TEST_CASE("fromString double", "") +{ + REQUIRE(testFromString(0.0, "0.0") ); + REQUIRE(testFromString(-0.0, "-0.0") ); + REQUIRE(testFromString(1.0, "1.0") ); + REQUIRE(testFromString(-1.0, "-1.0") ); + REQUIRE(testFromString(1.2345, "1.2345") ); + REQUIRE(testFromString(1.2345678, "1.2345678") ); + REQUIRE(testFromString(0.123456789012, "0.123456789012") ); + REQUIRE(testFromString(1234567.8, "1234567.8") ); + REQUIRE(testFromString(-79.39773355813419, "-79.39773355813419") ); + REQUIRE(testFromString(0.000001, "0.000001") ); + REQUIRE(testFromString(0.0000001, "1e-7") ); + REQUIRE(testFromString(1e30, "1e30") ); + REQUIRE(testFromString(1.234567890123456e30, "1.234567890123456e30") ); + REQUIRE(testFromString(-5e-324, "-5e-324") ); + REQUIRE(testFromString(2.225073858507201e-308, "2.225073858507201e-308") ); + REQUIRE(testFromString(2.2250738585072014e-308, "2.2250738585072014e-308") ); + REQUIRE(testFromString(1.7976931348623157e308, "1.7976931348623157e308") ); + REQUIRE(testFromString(0.00000123123123, "0.00000123123123") ); + REQUIRE(testFromString(0.000000123123123, "1.23123123e-7") ); + REQUIRE(testFromString(123123.123, "123123.123") ); + REQUIRE(testFromString(1231231.23, "1231231.23") ); + REQUIRE(testFromString(0.000000000123123, "1.23123e-10") ); + REQUIRE(testFromString(0.0000000001, "1e-10") ); +} + TEST_CASE("StringView", "") { bx::StringView sv("test");