How to enable __int128 on Visual Studio?
I've found a treasure in my old Visual Studio 6.0 C++ from 1996 (32-bit) making use of MS's own assembler routines that enabled 64-bit math on a 32-bit CPU(__int64). Source-code unfortunately not available). However, doing a debug-session that calls these functions, copy/paste the disassembler-list, search-replace "dword ptr" -> "qword ptr", eax,ebx,... -> rax,rbx,... and a Little adjustment of registers used for parameter-passing (and a lot of coffee), I succeeded to make this code, that makes it possible to do _int128-math in x64-mode just as it is possible to do __int64-math with 32-bit. It's essential the same code, with a doubleup in bits/cycle. For the matter of copyrights, I've seen no licenses in the disassembler-list, and perhaps it's time for Microsoft to integrate this into their x64 C++ compiler (vers. 2015) The code goes here
// File:Int128.h
#pragma once
#include "PragmaLib.h" // contains #pragma comment(lib,"Yourlib.lib")
#ifndef _M_X64
#error Int128 is available only in x64 arhcitecture
#else
class _int128;
class _uint128;
extern "C" {
void int128sum( void *dst, const void *x, const void *y);
void int128dif( void *dst, const void *x, const void *y);
void int128mul( void *dst, const void *x, const void *y);
void int128div( void *dst, const void *x, const void *y);
void int128rem( void *dst, const void *x, const void *y);
void int128neg( void *dst, const void *x);
int int128cmp(const void *n1, const void *n2);
void uint128div( void *dst, const void *x, const void *y);
void uint128rem( void *dst, const void *x, const void *y);
int uint128cmp(const void *n1, const void *n2);
};
class _int128 {
private:
_int128(unsigned __int64 _lo, const unsigned __int64 _hi) : lo(_lo), hi(_hi) {
}
public:
unsigned __int64 lo;
unsigned __int64 hi;
inline _int128() {
}
inline _int128(unsigned __int64 n) : lo(n), hi(0) {
}
inline _int128(__int64 n) : lo(n), hi(n>=0)?0:-1) { // remember signextend hi if n < 0 (2-complement)
}
inline _int128(unsigned int n) : lo(n), hi(0) {
}
inline _int128(int n) : lo(n), hi(n>=0)?0:-1) {
}
inline _int128(unsigned short n) : lo(n), hi(0) {
}
inline _int128(short n) : lo(n), hi(n>=0)?0:-1) {
}
explicit _int128(const char *str);
operator unsigned __int64() const {
return lo;
}
operator __int64() const {
return lo;
}
operator unsigned int() const {
return (unsigned int)lo;
}
operator int() const {
return (int)lo;
}
inline _int128 operator+(const _int128 &rhs) const {
_int128 result;
int128sum(&result, this, &rhs);
return result;
}
inline _int128 operator-(const _int128 &rhs) const {
_int128 result;
int128dif(&result, this, &rhs);
return result;
}
inline _int128 operator-() const {
_int128 result;
int128neg(&result, this);
return result;
}
inline _int128 operator*(const _int128 &rhs) const {
_int128 result;
int128mul(&result, this, &rhs);
return result;
}
inline _int128 operator/(const _int128 &rhs) const {
_int128 result, copy(*this);
int128div(&result, ©, &rhs);
return result;
}
inline _int128 operator%(const _int128 &rhs) const {
_int128 result, copy(*this);
int128rem(&result, ©, &rhs);
return result;
};
inline _int128 &operator+=(const _int128 &rhs) {
const _int128 copy(*this);
int128sum(this, ©, &rhs);
return *this;
}
inline _int128 &operator-=(const _int128 &rhs) {
const _int128 copy(*this);
int128dif(this, ©, &rhs);
return *this;
}
inline _int128 &operator*=(const _int128 &rhs) {
const _int128 copy(*this);
int128mul(this, ©, &rhs);
return *this;
}
inline _int128 &operator/=(const _int128 &rhs) {
const _int128 copy(*this);
int128div(this, ©, &rhs);
return *this;
}
inline _int128 &operator%=(const _int128 &rhs) {
const _int128 copy(*this);
int128rem(this, ©, &rhs);
return *this;
}
inline _int128 operator&(const _int128 &rhs) const {
return _int128(lo&rhs.lo, hi&rhs.hi);
}
inline _int128 operator|(const _int128 &rhs) const {
return _int128(lo|rhs.lo, hi|rhs.hi);
}
inline _int128 operator^(const _int128 &rhs) const {
return _int128(lo^rhs.lo, hi^rhs.hi);
}
const char *parseDec(const char *str); // return pointer to char following the number
const char *parseHex(const char *str); // do
const char *parseOct(const char *str); // do
};
class _uint128 {
public:
unsigned __int64 lo;
unsigned __int64 hi;
inline _uint128() {
}
inline _uint128(const _int128 &n) : lo(n.lo), hi(n.hi) {
}
inline _uint128(unsigned __int64 n) : lo(n), hi(0) {
}
inline _uint128(__int64 n) : lo(n), hi(n>=0)?0:-1) {
}
inline _uint128(unsigned int n) : lo(n), hi(0) {
}
inline _uint128(int n) : lo(n), hi(n>=0)?0:-1) {
}
inline _uint128(unsigned short n) : lo(n), hi(0) {
}
inline _uint128(short n) : lo(n), hi(n>=0)?0:-1) {
}
explicit _uint128(const char *str);
inline operator _int128() const {
return *(_int128*)(void*)this;
}
inline operator unsigned __int64() const {
return lo;
}
inline operator __int64() const {
return lo;
}
inline operator unsigned int() const {
return (unsigned int)lo;
}
inline operator int() const {
return (int)lo;
}
inline _uint128 operator+(const _uint128 &rhs) const {
_uint128 result;
int128sum(&result, this, &rhs);
return result;
}
inline _uint128 operator-(const _uint128 &rhs) const {
_uint128 result;
int128dif(&result, this, &rhs);
return result;
}
inline _uint128 operator*(const _uint128 &rhs) const {
_uint128 result;
int128mul(&result, this, &rhs);
return result;
}
inline _uint128 operator/(const _uint128 &rhs) const {
_uint128 result, copy(*this);
uint128div(&result, ©, &rhs);
return result;
}
inline _uint128 operator%(const _uint128 &rhs) const {
_uint128 result, copy(*this);
uint128rem(&result, ©, &rhs);
return result;
};
inline _uint128 &operator+=(const _uint128 &rhs) {
const _uint128 copy(*this);
int128sum(this, ©, &rhs);
return *this;
}
inline _uint128 &operator-=(const _uint128 &rhs) {
const _uint128 copy(*this);
int128dif(this, ©, &rhs);
return *this;
}
inline _uint128 &operator*=(const _uint128 &rhs) {
const _uint128 copy(*this);
int128mul(this, ©, &rhs);
return *this;
}
inline _uint128 &operator/=(const _uint128 &rhs) {
const _uint128 copy(*this);
uint128div(this, ©, &rhs);
return *this;
}
inline _uint128 &operator%=(const _uint128 &rhs) {
const _uint128 copy(*this);
uint128rem(this, ©, &rhs);
return *this;
}
const char *parseDec(const char *str); // return pointer to char following the number
const char *parseHex(const char *str); // do
const char *parseOct(const char *str); // do
};
inline bool operator==(const _int128 &lft, const _int128 &rhs) {
return (lft.lo == rhs.lo) && (lft.hi == rhs.hi);
}
inline bool operator==(const _int128 &lft, const _uint128 &rhs) {
return (lft.lo == rhs.lo) && (lft.hi == rhs.hi);
}
inline bool operator==(const _uint128 &lft, const _int128 &rhs) {
return (lft.lo == rhs.lo) && (lft.hi == rhs.hi);
}
inline bool operator==(const _uint128 &lft, const _uint128 &rhs) {
return (lft.lo == rhs.lo) && (lft.hi == rhs.hi);
}
inline bool operator!=(const _int128 &lft, const _int128 &rhs) {
return (lft.lo != rhs.lo) || (lft.hi != rhs.hi);
}
inline bool operator!=(const _int128 &lft, const _uint128 &rhs) {
return (lft.lo != rhs.lo) || (lft.hi != rhs.hi);
}
inline bool operator!=(const _uint128 &lft, const _int128 &rhs) {
return (lft.lo != rhs.lo) || (lft.hi != rhs.hi);
}
inline bool operator!=(const _uint128 &lft, const _uint128 &rhs) {
return (lft.lo != rhs.lo) || (lft.hi != rhs.hi);
}
inline bool operator>(const _int128 &lft, const _int128 &rhs) {
return int128cmp(&lft, &rhs) > 0;
}
inline bool operator>(const _int128 &lft, const _uint128 &rhs) {
return uint128cmp(&lft, &rhs) > 0;
}
inline bool operator>(const _uint128 &lft, const _int128 &rhs) {
return uint128cmp(&lft, &rhs) > 0;
}
inline bool operator>(const _uint128 &lft, const _uint128 &rhs) {
return uint128cmp(&lft, &rhs) > 0;
}
inline bool operator>=(const _int128 &lft, const _int128 &rhs) {
return int128cmp(&lft, &rhs) >= 0;
}
inline bool operator>=(const _int128 &lft, const _uint128 &rhs) {
return uint128cmp(&lft, &rhs) >= 0;
}
inline bool operator>=(const _uint128 &lft, const _int128 &rhs) {
return uint128cmp(&lft, &rhs) >= 0;
}
inline bool operator>=(const _uint128 &lft, const _uint128 &rhs) {
return uint128cmp(&lft, &rhs) >= 0;
}
inline bool operator<(const _int128 &lft, const _int128 &rhs) {
return int128cmp(&lft, &rhs) < 0;
}
inline bool operator<(const _int128 &lft, const _uint128 &rhs) {
return uint128cmp(&lft, &rhs) < 0;
}
inline bool operator<(const _uint128 &lft, const _int128 &rhs) {
return uint128cmp(&lft, &rhs) < 0;
}
inline bool operator<(const _uint128 &lft, const _uint128 &rhs) {
return uint128cmp(&lft, &rhs) < 0;
}
inline bool operator<=(const _int128 &lft, const _int128 &rhs) {
return int128cmp(&lft, &rhs) <= 0;
}
inline bool operator<=(const _int128 &lft, const _uint128 &rhs) {
return uint128cmp(&lft, &rhs) <= 0;
}
inline bool operator<=(const _uint128 &lft, const _int128 &rhs) {
return uint128cmp(&lft, &rhs) <= 0;
}
inline bool operator<=(const _uint128 &lft, const _uint128 &rhs) {
return uint128cmp(&lft, &rhs) <= 0;
}
char * _i128toa(_int128 value, char *str, int radix);
char * _ui128toa(_uint128 value, char *str, int radix);
wchar_t * _i128tow(_int128 value, wchar_t *str, int radix);
wchar_t * _ui128tow(_uint128 value, wchar_t *str, int radix);
inline char radixLetter(unsigned int c) {
return (c < 10) ? ('0' + c) : ('a' + (c-10));
}
inline wchar_t wradixLetter(unsigned int c) {
return (c < 10) ? ('0' + c) : ('a' + (c-10));
}
inline bool isodigit(unsigned char ch) {
return ('0' <= ch) && (ch < '8');
}
unsigned int convertNumberChar(char digit);
#endif // _M_X64
; File: Int128x64.asm
; build obj-file with
; ml64 /nologo /c /Zf /Fo$(IntDir)Int128x64.obj Int128x64.asm
.CODE
;void int128sum(_int128 &dst, cnost _int128 &x, const _int128 &y);
int128sum PROC
push rbx
mov rax, qword ptr[rdx]
add rax, qword ptr[r8]
mov rbx, qword ptr[rdx+8]
adc rbx, qword ptr[r8+8]
mov qword ptr[rcx], rax
mov qword ptr[rcx+8], rbx
pop rbx
ret
int128sum ENDP
;void int128dif( _int128 &dst, const _int128 &x, const _int128 &y);
int128dif PROC
push rbx
mov rax, qword ptr[rdx]
sub rax, qword ptr[r8]
mov rbx, qword ptr[rdx+8]
sbb rbx, qword ptr[r8+8]
mov qword ptr[rcx] , rax
mov qword ptr[rcx+8], rbx
pop rbx
ret
int128dif ENDP
;void int128mul(_int128 &dst, const _int128 &x, const _int128 &y);
int128mul PROC
push rbx
mov rax, qword ptr[rdx+8] ; rax = x.hi
mov rbx, qword ptr[r8+8] ; rbx = y.hi
or rbx, rax ; rbx = x.hi | y.hi
mov rbx, qword ptr[r8] ; rbx = y.lo
jne Hard ; if(x.hi|y.hi) goto Hard
; simple int64 multiplication
mov rax, qword ptr[rdx] ; rax = x.lo
mul rbx ; rdx:rax = rax * rbx
mov qword ptr[rcx] , rax ; dst.lo = rax
mov qword ptr[rcx+8], rdx ; dst.hi = rdx
pop rbx
ret
Hard: ; assume rax = x.hi, rbx = y.lo
push rsi
mov rsi, rdx ; need rdx for highend of mul, so rsi=&x
mul rbx ; rdx:rax = x.hi * y.lo
mov r9 , rax ;
mov rax, qword ptr[rsi] ; rax = x.lo
mul qword ptr[r8+8] ; rdx:rax = x.lo * y.hi
add r9, rax ; r9 = lo(x.hi*y.lo+x.lo*y.hi);
mov rax, qword ptr[rsi] ; rax = x.lo
mul rbx ; rdx:rax = x.lo * y.lo
add rdx, r9
mov qword ptr[rcx] , rax
mov qword ptr[rcx+8], rdx
pop rsi
pop rbx
ret
int128mul ENDP
;void int128div(_int128 &dst, const _int128 &x, const _int128 &y);
int128div PROC
push rdi
push rsi
push rbx
push rcx
mov r9, rdx
xor rdi, rdi
mov rax, qword ptr[r9+8]
or rax, rax
jge L1
inc rdi
mov rdx, qword ptr[r9]
neg rax
neg rdx
sbb rax, 0
mov qword ptr[r9+8], rax
mov qword ptr[r9], rdx
L1:
mov rax, qword ptr[r8+8]
or rax, rax
jge L2
inc rdi
mov rdx, qword ptr[r8]
neg rax
neg rdx
sbb rax,0
mov qword ptr[r8+8], rax
mov qword ptr[r8], rdx
L2:
or rax, rax
jne L3
mov rcx, qword ptr[r8]
mov rax, qword ptr[r9+8]
xor rdx, rdx
div rcx
mov rbx, rax
mov rax, qword ptr[r9]
div rcx
mov rdx, rbx
jmp L4
L3:
mov rbx,rax
mov rcx,qword ptr[r8]
mov rdx,qword ptr[r9+8]
mov rax,qword ptr[r9]
L5:
shr rbx, 1
rcr rcx, 1
shr rdx, 1
rcr rax, 1
or rbx, rbx
jne L5
div rcx
mov rsi, rax
mul qword ptr[r8+8]
mov rcx, rax
mov rax, qword ptr[r8]
mul rsi
add rdx, rcx
jb L6
cmp rdx, qword ptr[r9+8]
ja L6
jb L7
cmp rax, qword ptr[rdx]
jbe L7
L6:
dec rsi
L7:
xor rdx, rdx
mov rax, rsi
L4:
dec rdi
jne L8
neg rdx
neg rax
sbb rdx, 0
L8:
pop rcx
pop rbx
pop rsi
pop rdi
mov qword ptr[rcx], rax
mov qword ptr[rcx+8], rdx
ret
int128div ENDP
;void int128rem( _int128 &dst, const _int128 &x, const _int128 &y);
int128rem PROC
push rbx
push rdi
push rcx
mov r9, rdx
xor rdi, rdi
mov rax, qword ptr[r9+8]
or rax, rax
jge L1
inc rdi
mov rdx, qword ptr[r9]
neg rax
neg rdx
sbb rax, 0
mov qword ptr[r9+8], rax
mov qword ptr[r9], rdx
L1:
mov rax, qword ptr[r8+8]
or rax, rax
jge L2
mov rdx, qword ptr[r8]
neg rax
neg rdx
sbb rax, 0
mov qword ptr[r8+8], rax
mov qword ptr[r8], rdx
L2:
or rax, rax
jne L3
mov rcx, qword ptr[r8]
mov rax, qword ptr[r9+8]
xor rdx, rdx
div rcx
mov rax, qword ptr[r9]
div rcx
mov rax, rdx
xor rdx, rdx
dec rdi
jns L4
jmp L8
L3:
mov rbx, rax
mov rcx, qword ptr[r8]
mov rdx, qword ptr[r9+8]
mov rax, qword ptr[r9]
L5:
shr rbx, 1
rcr rcx, 1
shr rdx, 1
rcr rax, 1
or rbx, rbx
jne L5
div rcx
mov rcx, rax
mul qword ptr[r8+8]
xchg rax, rcx
mul qword ptr[r8]
add rdx, rcx
jb L6
cmp rdx, qword ptr[r9+8]
ja L6
jb L7
cmp rax, qword ptr[r9]
jbe L7
L6:
sub rax, qword ptr[r8]
sbb rdx, qword ptr[r8+8]
L7:
sub rax, qword ptr[r9]
sbb rdx, qword ptr[r9+8]
dec rdi
jns L8
L4:
neg rdx
neg rax
sbb rdx, 0
L8:
pop rcx
pop rdi
pop rbx
mov qword ptr[rcx], rax
mov qword ptr[rcx+8], rdx
ret
int128rem ENDP
;void int128neg( _int128 &dst, const _int128 &x);
int128neg PROC
mov rax,qword ptr[rdx]
neg rax
mov r8, qword ptr[rdx+8]
adc r8, 0
neg r8
mov qword ptr[rcx], rax
mov qword ptr[rcx+8], r8
ret
int128neg ENDP
;int int128cmp(const _int128 &n1, const _int128 &n2);
int128cmp PROC
mov rax, qword ptr[rcx+8] ; n1.hi
cmp rax, qword ptr[rdx+8] ; n2.hi
jl lessthan ; signed compare of n1.hi and n2.hi
jg greaterthan
mov rax, qword ptr[rcx] ; n2.lo
cmp rax, qword ptr[rdx] ; n2.lo
jb lessthan ; unsigned compare of n1.lo and n2.lo
ja greaterthan
mov rax, 0 ; they are equal
ret
greaterthan:
mov rax, 1
ret
lessthan:
mov rax, -1
ret
int128cmp ENDP
END
; File:UInt128x64.asm
; build obj-file with
; ml64 /nologo /c /Zf /Fo$(IntDir)UInt128x64.obj UInt128x64.asm
.CODE
;void uint128div(_uint128 &dst, const _uint128 &x, const _uint128 &y);
uint128div PROC
push rbx
push rsi
push rcx
mov r9, rdx
mov rax, qword ptr[r8+8]
or rax, rax
jne L1
mov rcx, qword ptr[r8]
mov rax, qword ptr[r9+8]
xor rdx, rdx
div rcx
mov rbx, rax
mov rax, qword ptr[r9]
div rcx
mov rdx, rbx
jmp L2
L1:
mov rcx, rax
mov rbx, qword ptr[r8]
mov rdx, qword ptr[r9+8]
mov rax, qword ptr[r9]
L3:
shr rcx, 1
rcr rbx, 1
shr rdx, 1
rcr rax, 1
or rcx, rcx
jne L3
div rbx
mov rsi, rax
mul qword ptr[r8+8]
mov rcx, rax
mov rax, qword ptr[r8]
mul rsi
add rdx, rcx
jb L4
cmp rdx, qword ptr[r9+8]
ja L4
jb L5
cmp rax, qword ptr[r9]
jbe L5
L4:
dec rsi
L5:
xor rdx, rdx
mov rax, rsi
L2:
pop rcx
pop rsi
pop rbx
mov qword ptr[rcx], rax
mov qword ptr[rcx+8], rdx
ret
uint128div ENDP
;void uint128rem(_uint128 &dst, const _uint128 &x, const _uint128 &y);
uint128rem PROC
push rbx
push rcx
mov r9, rdx
mov rax, qword ptr[r8+8]
or rax, rax
jne L1
mov rcx, qword ptr[r8]
mov rax, qword ptr[r9+8]
xor rdx, rdx
div rcx
mov rax, qword ptr[r9]
div rcx
mov rax, rdx
xor rdx, rdx
jmp L2
L1:
mov rcx, rax
mov rbx, qword ptr[r8]
mov rdx, qword ptr[r9+8]
mov rax, qword ptr[r9]
L3:
shr rcx, 1
rcr rbx, 1
shr rdx, 1
rcr rax, 1
or rcx, rcx
jne L3
div rbx
mov rcx, rax
mul qword ptr[r8+8]
xchg rax, rcx
mul qword ptr[r8]
add rdx, rcx
jb L4
cmp rdx, qword ptr[r9+8]
ja L4
jb L5
cmp rax, qword ptr[r9]
jbe L5
L4:
sub rax, qword ptr[r8]
sbb rdx, qword ptr[r8+8]
L5:
sub rax, qword ptr[r9]
sbb rdx, qword ptr[r9+8]
neg rdx
neg rax
sbb rdx, 0
L2:
pop rcx
pop rbx
mov qword ptr[rcx], rax
mov qword ptr[rcx+8], rdx
ret
uint128rem ENDP
;int uint128cmp(const _uint128 &n1, const _uint128 &n2);
uint128cmp PROC
mov rax, qword ptr[rcx+8] ; n1.hi
cmp rax, qword ptr[rdx+8] ; n2.hi
jb lessthan ; usigned compare of n1.hi and n2.hi
ja greaterthan
mov rax, qword ptr[rcx] ; n2.lo
cmp rax, qword ptr[rdx] ; n2.lo
jb lessthan ; unsigned compare of n1.lo and n2.lo
ja greaterthan
mov rax, 0 ; they are equal
ret
greaterthan:
mov rax, 1
ret
lessthan:
mov rax, -1
ret
uint128cmp ENDP
END
There will be 3 more files. not enough Space here...
MSDN doesn't list it as being available, and this recent response agrees, so officially, no, there is no type called __int128 and it can't be enabled.
Additionally, never trust the syntax hilighter; it is user editable, and thus likely to either have bogus or 'future' types in it. (it is probably a reserved word however, due to the error, thus you should avoid naming any types __int128, this follows the convention that anything prefixed with a double underscore should reserved for compiler use).
One would think the __int128 might be available on x64/IPF machines via register spanning, like __in64 is on 32bit targets, but right now the only 128 bit types stem from SIMD types (__m128 and its various typed forms).