Microsoft MVP성태의 닷넷 이야기
닷넷: 2258. C# - double (배정도 실수) 저장소의 비트 구조 [링크 복사], [링크+제목 복사],
조회: 8644
글쓴 사람
정성태 (techsharer at outlook.com)
홈페이지
첨부 파일

(시리즈 글이 8개 있습니다.)
.NET Framework: 539. C# - 부동 소수 계산 왜 이렇게 나오죠? (1)
; https://www.sysnet.pe.kr/2/0/10872

.NET Framework: 540. C# - 부동 소수 계산 왜 이렇게 나오죠? (2)
; https://www.sysnet.pe.kr/2/0/10873

.NET Framework: 608. double 값을 구할 때는 반드시 피연산자를 double로 형변환!
; https://www.sysnet.pe.kr/2/0/11055

개발 환경 구성: 440. C#, C++ - double의 Infinity, NaN 표현 방식
; https://www.sysnet.pe.kr/2/0/11896

기타: 85. 단정도/배정도 부동 소수점의 정밀도(Precision)에 따른 형변환 손실
; https://www.sysnet.pe.kr/2/0/13212

닷넷: 2257. C# - float (단정도 실수) 저장소의 비트 구조
; https://www.sysnet.pe.kr/2/0/13617

닷넷: 2258. C# - double (배정도 실수) 저장소의 비트 구조
; https://www.sysnet.pe.kr/2/0/13618

닷넷: 2259. C# - decimal 저장소의 비트 구조
; https://www.sysnet.pe.kr/2/0/13619




C# - double (배정도 실수) 저장소의 비트 구조

double의 경우 지난 글에서 다룬 float 방식과 거의 같습니다.

C# - float (단정도 실수) 저장소의 비트 구조
; https://www.sysnet.pe.kr/2/0/13617

단지, 지수부와 가수부를 위한 비트 수만 좀 더 확장한 유형인데요, 다음의 그림에 따라,

[배정도 실수 - 그림 출처: https://en.wikipedia.org/wiki/Double-precision_floating-point_format]
single_float_2.png

비트 구조만 맞춰서 masking만 잘하면 됩니다.

static unsafe void Main(string[] args)
{
    double d = -118.625;
    Console.WriteLine($"{d} (decimal: {(decimal)d}): sizeof(double): {sizeof(double)}");
    Console.WriteLine();

    byte* pDouble = (byte*)&d;
    PrintDoubleFormat(pDouble);
}

private static unsafe void PrintDoubleFormat(byte* pDouble)
{
    ulong data = *(ulong*)pDouble;

    ulong signBitMask = 0b_1000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000;
    ulong exponentMask = 0b_0111_1111_1111_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000;
    ulong fractionMask = 0b_0000_0000_0000_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111;

    ulong signBit = (data & signBitMask);
    ulong exponentBits = (data & exponentMask);
    ulong fractionBits = (data & fractionMask);

    Console.WriteLine("signBit: \t" + Convert.ToString((long)signBit, 2).PadLeft(64, '0').Separator(4, '_'));
    Console.WriteLine("exponentBits: \t" + Convert.ToString((long)exponentBits, 2).PadLeft(64, '0').Separator(4, '_'));
    Console.WriteLine("fractionBits: \t" + Convert.ToString((long)fractionBits, 2).PadLeft(64, '0').Separator(4, '_'));
}

실행하면 다음과 같은 결과가 나오는데요,

-118.625 (decimal: -118.625): sizeof(double): 8

signBit:        1000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000
exponentBits:   0100_0000_0101_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000
fractionBits:   0000_0000_0000_1101_1010_1000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000

IEEE 754 표준에 따른 정규화 과정을 거쳐,

2진수 변환)
-118.625 ==> 1110110.101

지수 표현)
1110110.101 ==> 1.110110101 * 26
    지수부: 6
    가수부: 1.110110101

double의 경우 float와는 달리 exponentBits가 11비트로 늘었는데요, 따라서 0 ~ 2,047의 수를 표현할 수 있으니 bias 값이 1,023이 됩니다. 결국 지수부 6에 (float는 127을 더했지만) 1023을 더한 1029 값이 exponentBits(100 0000 0101)로 설정됩니다.

가수부의 처리도 비트만 52비트로 늘었을 뿐 float에서의 처리와 다르지 않습니다.




분해된 정보로부터 원래의 double 값을 복원하는 것도 float에서의 규칙과 같습니다. 가수부의 110110101에서 생략된 가장 상위의 1을 복원시키고,

1_1011_0101 ==> 11_1011_0101

그다음, 지수부의 1029를 원래의 지수로 만들어줍니다. (반대로 1023을 빼면 됩니다.)

6 = 1029 - 1023

이렇게 구한 값들을 통해 처음의 double 값으로 복원할 수 있습니다.

1.110110101 * 2E6

==> 1110110.101

10진수로 ==> 118.625

==> sign 비트 적용
-118.625

이 과정 역시 float의 코드를 조금만 바꾸면 double로 이렇게 만들 수 있고,

{
    bool minus = signBit != 0;
    ulong exponents = exponentBits >> 52;

    // 삭제된 1을 복원하고,
    ulong fractions = fractionBits | 0b_0000_0000_0001_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000;

    int shift = (int)exponents - 1023;

    // (삭제된 1로 인해 12비트가 아닌) 11비트만 shift 시키면 원래의 가수로 변환
    fractions = fractions << 11;
    string mantissa = Convert.ToString((long)fractions, 2).TrimEnd('0');
    Console.WriteLine($"{(minus ? "-" : "")}{mantissa} * 2E{shift}");

    mantissa = Convert.ToString((long)fractions, 2).TrimEnd('0');
    mantissa = MarkDecimalPoint(mantissa, shift);
    Console.WriteLine($"{(minus ? "-" : "")}{mantissa}");

    decimal value = Recomposite(mantissa) * (minus ? -1 : 1);
    Console.WriteLine($"{value}, (double: {(double)value})");
}

// ... 기타 코드는 float 예제와 동일 ...

이전 코드와 합쳐서 실행해 보면 이런 결과를 얻을 수 있습니다.

-118.625 (decimal: -118.625): sizeof(double): 8

signBit:        1000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000
exponentBits:   0100_0000_0101_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000
fractionBits:   0000_0000_0000_1101_1010_1000_0000_0000_0000_0000_0000_0000_0000_0000_0000_0000

-1110110101 * 2E6
-1110110.101
-118.625, (double: -118.625)

(첨부 파일은 이 글의 예제 코드를 포함합니다.)




참고로, NaN과 infinity 처리도 있는데 이에 대해서는 전에 설명한 적이 있습니다.

C#, C++ - double의 Infinity, NaN 표현 방식
; https://www.sysnet.pe.kr/2/0/11896

그런데 저렇게 장황하게 설명한 것보다 아래의 글에 도표로 잘 정리된 것이 있군요. ^^

Binary floating point and .NET
; https://csharpindepth.com/Articles/FloatingPoint

float_nan_infinity_1.png




[이 글에 대해서 여러분들과 의견을 공유하고 싶습니다. 틀리거나 미흡한 부분 또는 의문 사항이 있으시면 언제든 댓글 남겨주십시오.]







[최초 등록일: ]
[최종 수정일: 5/10/2024]

Creative Commons License
이 저작물은 크리에이티브 커먼즈 코리아 저작자표시-비영리-변경금지 2.0 대한민국 라이센스에 따라 이용하실 수 있습니다.
by SeongTae Jeong, mailto:techsharer at outlook.com

비밀번호

댓글 작성자
 




1  2  3  4  [5]  6  7  8  9  10  11  12  13  14  15  ...
NoWriterDateCnt.TitleFile(s)
13818정성태11/15/20245256Windows: 272. Windows 11 24H2 - sudo 추가
13817정성태11/14/20244918Linux: 106. eBPF / bpf2go - (BPF_MAP_TYPE_HASH) Map을 이용한 전역 변수 구현
13816정성태11/14/20245382닷넷: 2312. C#, C++ - Windows / Linux 환경의 Thread Name 설정파일 다운로드1
13815정성태11/13/20244803Linux: 105. eBPF - bpf2go에서 전역 변수 설정 방법
13814정성태11/13/20245277닷넷: 2311. C# - Windows / Linux 환경에서 Native Thread ID 가져오기파일 다운로드1
13813정성태11/12/20245014닷넷: 2310. .NET의 Rune 타입과 emoji 표현파일 다운로드1
13812정성태11/11/20245245오류 유형: 933. Active Directory - The forest functional level is not supported.
13811정성태11/11/20244846Linux: 104. Linux - COLUMNS 환경변수가 언제나 80으로 설정되는 환경
13810정성태11/10/20245374Linux: 103. eBPF (bpf2go) - Tracepoint를 이용한 트레이스 (BPF_PROG_TYPE_TRACEPOINT)
13809정성태11/10/20245250Windows: 271. 윈도우 서버 2025 마이그레이션
13808정성태11/9/20245248오류 유형: 932. Linux - 커널 업그레이드 후 "error: bad shim signature" 오류 발생
13807정성태11/9/20244983Linux: 102. Linux - 커널 이미지 파일 서명 (Ubuntu 환경)
13806정성태11/8/20244893Windows: 270. 어댑터 상세 정보(Network Connection Details) 창의 내용이 비어 있는 경우
13805정성태11/8/20244733오류 유형: 931. Active Directory의 adprep 또는 복제가 안 되는 경우
13804정성태11/7/20245362Linux: 101. eBPF 함수의 인자를 다루는 방법
13803정성태11/7/20245309닷넷: 2309. C# - .NET Core에서 바뀐 DateTime.Ticks의 정밀도
13802정성태11/6/20245691Windows: 269. GetSystemTimeAsFileTime과 GetSystemTimePreciseAsFileTime의 차이점파일 다운로드1
13801정성태11/5/20245480Linux: 100. eBPF의 2가지 방식 - libbcc와 libbpf(CO-RE)
13800정성태11/3/20246315닷넷: 2308. C# - ICU 라이브러리를 활용한 문자열의 대소문자 변환 [2]파일 다운로드1
13799정성태11/2/20244891개발 환경 구성: 732. 모바일 웹 브라우저에서 유니코드 문자가 표시되지 않는 경우
13798정성태11/2/20245490개발 환경 구성: 731. 유니코드 - 출력 예시 및 폰트 찾기
13797정성태11/1/20245484C/C++: 185. C++ - 문자열의 대소문자를 변환하는 transform + std::tolower/toupper 방식의 문제점파일 다운로드1
13796정성태10/31/20245368C/C++: 184. C++ - ICU dll을 이용하는 예제 코드 (Windows)파일 다운로드1
13795정성태10/31/20245163Windows: 268. Windows - 리눅스 환경처럼 공백으로 끝나는 프롬프트 만들기
13794정성태10/30/20245261닷넷: 2307. C# - 윈도우에서 한글(및 유니코드)을 포함한 콘솔 프로그램을 컴파일 및 실행하는 방법
13793정성태10/28/20245125C/C++: 183. C++ - 윈도우에서 한글(및 유니코드)을 포함한 콘솔 프로그램을 컴파일 및 실행하는 방법
1  2  3  4  [5]  6  7  8  9  10  11  12  13  14  15  ...