MULDIV32.H

//==========================================================================; 
//
// THIS CODE AND INFORMATION IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY
// KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND/OR FITNESS FOR A PARTICULAR
// PURPOSE.
//
// Copyright 1992 - 1998 Microsoft Corporation. All Rights Reserved.
//
//--------------------------------------------------------------------------;
//
// muldiv32.h
//
// Description:
// math routines for 32 bit signed and unsiged numbers.
//
// MulDiv32(a,b,c) = (a * b) / c (round down, signed)
//
// MulDivRD(a,b,c) = (a * b) / c (round down, unsigned)
// MulDivRN(a,b,c) = (a * b + c/2) / c (round nearest, unsigned)
// MulDivRU(a,b,c) = (a * b + c-1) / c (round up, unsigned)
//
//==========================================================================;

#ifndef _INC_MULDIV32
#define _INC_MULDIV32


#ifndef INLINE
#define INLINE __inline
#endif


#ifdef WIN32

//----------------------------------------------------------------------;
//
// Win 32
//
//----------------------------------------------------------------------;

#ifdef _X86_

//
// Use 32-bit x86 assembly.
//

#pragma warning(disable:4035 4704)

INLINE LONG MulDiv32(LONG a,LONG b,LONG c)
{
_asm mov eax,dword ptr a // mov eax, a
_asm mov ebx,dword ptr b // mov ebx, b
_asm mov ecx,dword ptr c // mov ecx, c
_asm imul ebx // imul ebx
_asm idiv ecx // idiv ecx
_asm shld edx, eax, 16 // shld edx, eax, 16

} // MulDiv32()

INLINE DWORD MulDivRN(DWORD a,DWORD b,DWORD c)
{
_asm mov eax,dword ptr a // mov eax, a
_asm mov ebx,dword ptr b // mov ebx, b
_asm mov ecx,dword ptr c // mov ecx, c
_asm mul ebx // mul ebx
_asm mov ebx,ecx // mov ebx,ecx
_asm shr ebx,1 // sar ebx,1
_asm add eax,ebx // add eax,ebx
_asm adc edx,0 // adc edx,0
_asm div ecx // div ecx
_asm shld edx, eax, 16 // shld edx, eax, 16

} // MulDiv32()

INLINE DWORD MulDivRU(DWORD a,DWORD b,DWORD c)
{
_asm mov eax,dword ptr a // mov eax, a
_asm mov ebx,dword ptr b // mov ebx, b
_asm mov ecx,dword ptr c // mov ecx, c
_asm mul ebx // mul ebx
_asm mov ebx,ecx // mov ebx,ecx
_asm dec ebx // dec ebx
_asm add eax,ebx // add eax,ebx
_asm adc edx,0 // adc edx,0
_asm div ecx // div ecx
_asm shld edx, eax, 16 // shld edx, eax, 16

} // MulDivRU32()

INLINE DWORD MulDivRD(DWORD a,DWORD b,DWORD c)
{
_asm mov eax,dword ptr a // mov eax, a
_asm mov ebx,dword ptr b // mov ebx, b
_asm mov ecx,dword ptr c // mov ecx, c
_asm mul ebx // mul ebx
_asm div ecx // div ecx
_asm shld edx, eax, 16 // shld edx, eax, 16

} // MulDivRD32()

#pragma warning(default:4035 4704)


#else

//
// Use C9 __int64 support for Daytona RISC platforms.
//

INLINE LONG MulDiv32( LONG a, LONG b, LONG c )
{
return (LONG)( Int32x32To64(a,b) / c );
}


INLINE DWORD MulDivRD( DWORD a, DWORD b, DWORD c )
{
return (DWORD)( UInt32x32To64(a,b) / c );
}


INLINE DWORD MulDivRN( DWORD a, DWORD b, DWORD c )
{
return (DWORD)( (UInt32x32To64(a,b)+c/2) / c );
}


INLINE DWORD MulDivRU( DWORD a, DWORD b, DWORD c )
{
return (DWORD)( (UInt32x32To64(a,b)+c-1) / c );
}

#endif


#else

//----------------------------------------------------------------------;
//
// Win 16
//
//----------------------------------------------------------------------;

#pragma warning(disable:4035 4704)

//
// Compile for 16-bit - we can use x86 with proper opcode prefixes
// to get 32-bit instructions.
//

INLINE LONG MulDiv32(LONG a,LONG b,LONG c)
{
_asm _emit 0x66 _asm mov ax,word ptr a // mov eax, a
_asm _emit 0x66 _asm mov bx,word ptr b // mov ebx, b
_asm _emit 0x66 _asm mov cx,word ptr c // mov ecx, c
_asm _emit 0x66 _asm imul bx // imul ebx
_asm _emit 0x66 _asm idiv cx // idiv ecx
_asm _emit 0x66 // shld edx, eax, 16
_asm _emit 0x0F
_asm _emit 0xA4
_asm _emit 0xC2
_asm _emit 0x10

} // MulDiv32()

INLINE DWORD MulDivRN(DWORD a,DWORD b,DWORD c)
{
_asm _emit 0x66 _asm mov ax,word ptr a // mov eax, a
_asm _emit 0x66 _asm mov bx,word ptr b // mov ebx, b
_asm _emit 0x66 _asm mov cx,word ptr c // mov ecx, c
_asm _emit 0x66 _asm mul bx // mul ebx
_asm _emit 0x66 _asm mov bx,cx // mov ebx,ecx
_asm _emit 0x66 _asm shr bx,1 // sar ebx,1
_asm _emit 0x66 _asm add ax,bx // add eax,ebx
_asm _emit 0x66 _asm adc dx,0 // adc edx,0
_asm _emit 0x66 _asm div cx // div ecx
_asm _emit 0x66 // shld edx, eax, 16
_asm _emit 0x0F
_asm _emit 0xA4
_asm _emit 0xC2
_asm _emit 0x10

} // MulDiv32()

INLINE DWORD MulDivRU(DWORD a,DWORD b,DWORD c)
{
_asm _emit 0x66 _asm mov ax,word ptr a // mov eax, a
_asm _emit 0x66 _asm mov bx,word ptr b // mov ebx, b
_asm _emit 0x66 _asm mov cx,word ptr c // mov ecx, c
_asm _emit 0x66 _asm mul bx // mul ebx
_asm _emit 0x66 _asm mov bx,cx // mov ebx,ecx
_asm _emit 0x66 _asm dec bx // dec ebx
_asm _emit 0x66 _asm add ax,bx // add eax,ebx
_asm _emit 0x66 _asm adc dx,0 // adc edx,0
_asm _emit 0x66 _asm div cx // div ecx
_asm _emit 0x66 // shld edx, eax, 16
_asm _emit 0x0F
_asm _emit 0xA4
_asm _emit 0xC2
_asm _emit 0x10

} // MulDivRU32()


INLINE DWORD MulDivRD(DWORD a,DWORD b,DWORD c)
{
_asm _emit 0x66 _asm mov ax,word ptr a // mov eax, a
_asm _emit 0x66 _asm mov bx,word ptr b // mov ebx, b
_asm _emit 0x66 _asm mov cx,word ptr c // mov ecx, c
_asm _emit 0x66 _asm mul bx // mul ebx
_asm _emit 0x66 _asm div cx // div ecx
_asm _emit 0x66 // shld edx, eax, 16
_asm _emit 0x0F
_asm _emit 0xA4
_asm _emit 0xC2
_asm _emit 0x10

} // MulDivRD32()

#pragma warning(default:4035 4704)

#endif


//
// some code references these by other names.
//
#define muldiv32 MulDivRN
#define muldivrd32 MulDivRD
#define muldivru32 MulDivRU

#endif // _INC_MULDIV32