gzdoom/game-music-emu/gme/Kss_Cpu.cpp

1707 lines
38 KiB
C++

// Game_Music_Emu 0.6.0. http://www.slack.net/~ant/
/*
Last validated with zexall 2006.11.14 2:19 PM
* Doesn't implement the R register or immediate interrupt after EI.
* Address wrap-around isn't completely correct, but is prevented from crashing emulator.
*/
#include "Kss_Cpu.h"
#include "blargg_endian.h"
#include <string.h>
//#include "z80_cpu_log.h"
/* Copyright (C) 2006 Shay Green. This module is free software; you
can redistribute it and/or modify it under the terms of the GNU Lesser
General Public License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version. This
module is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
details. You should have received a copy of the GNU Lesser General Public
License along with this module; if not, write to the Free Software Foundation,
Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */
#define SYNC_TIME() (void) (s.time = s_time)
#define RELOAD_TIME() (void) (s_time = s.time)
// Callbacks to emulator
#define CPU_OUT( cpu, addr, data, time )\
kss_cpu_out( this, time, addr, data )
#define CPU_IN( cpu, addr, time )\
kss_cpu_in( this, time, addr )
#define CPU_WRITE( cpu, addr, data, time )\
(SYNC_TIME(), kss_cpu_write( this, addr, data ))
#include "blargg_source.h"
// flags, named with hex value for clarity
int const S80 = 0x80;
int const Z40 = 0x40;
int const F20 = 0x20;
int const H10 = 0x10;
int const F08 = 0x08;
int const V04 = 0x04;
int const P04 = 0x04;
int const N02 = 0x02;
int const C01 = 0x01;
#define SZ28P( n ) szpc [n]
#define SZ28PC( n ) szpc [n]
#define SZ28C( n ) (szpc [n] & ~P04)
#define SZ28( n ) SZ28C( n )
#define SET_R( n ) (void) (r.r = n)
#define GET_R() (r.r)
Kss_Cpu::Kss_Cpu()
{
state = &state_;
for ( int i = 0x100; --i >= 0; )
{
int even = 1;
for ( int p = i; p; p >>= 1 )
even ^= p;
int n = (i & (S80 | F20 | F08)) | ((even & 1) * P04);
szpc [i] = n;
szpc [i + 0x100] = n | C01;
}
szpc [0x000] |= Z40;
szpc [0x100] |= Z40;
}
inline void Kss_Cpu::set_page( int i, void* write, void const* read )
{
blargg_long offset = KSS_CPU_PAGE_OFFSET( i * (blargg_long) page_size );
state->write [i] = (byte *) write - offset;
state->read [i] = (byte const*) read - offset;
}
void Kss_Cpu::reset( void* unmapped_write, void const* unmapped_read )
{
check( state == &state_ );
state = &state_;
state_.time = 0;
state_.base = 0;
end_time_ = 0;
for ( int i = 0; i < page_count + 1; i++ )
set_page( i, unmapped_write, unmapped_read );
memset( &r, 0, sizeof r );
}
void Kss_Cpu::map_mem( unsigned addr, blargg_ulong size, void* write, void const* read )
{
// address range must begin and end on page boundaries
require( addr % page_size == 0 );
require( size % page_size == 0 );
unsigned first_page = addr / page_size;
for ( unsigned i = size / page_size; i--; )
{
blargg_long offset = i * (blargg_long) page_size;
set_page( first_page + i, (byte*) write + offset, (byte const*) read + offset );
}
}
#define TIME (s_time + s.base)
#define RW_MEM( addr, rw ) (s.rw [(addr) >> page_shift] [KSS_CPU_PAGE_OFFSET( addr )])
#define READ_PROG( addr ) RW_MEM( addr, read )
#define READ( addr ) READ_PROG( addr )
//#define WRITE( addr, data ) (void) (RW_MEM( addr, write ) = data)
#define WRITE( addr, data ) CPU_WRITE( this, addr, data, TIME )
#define READ_WORD( addr ) GET_LE16( &READ( addr ) )
#define WRITE_WORD( addr, data ) SET_LE16( &RW_MEM( addr, write ), data )
#define IN( addr ) CPU_IN( this, addr, TIME )
#define OUT( addr, data ) CPU_OUT( this, addr, data, TIME )
#if BLARGG_BIG_ENDIAN
#define R8( n, offset ) ((r8_ - offset) [n])
#elif BLARGG_LITTLE_ENDIAN
#define R8( n, offset ) ((r8_ - offset) [(n) ^ 1])
#else
#error "Byte order of CPU must be known"
#endif
//#define R16( n, shift, offset ) (r16_ [((n) >> shift) - (offset >> shift)])
// help compiler see that it can just adjust stack offset, saving an extra instruction
#define R16( n, shift, offset )\
(*(uint16_t*) ((char*) r16_ - (offset >> (shift - 1)) + ((n) >> (shift - 1))))
#define CASE5( a, b, c, d, e ) case 0x##a:case 0x##b:case 0x##c:case 0x##d:case 0x##e
#define CASE6( a, b, c, d, e, f ) CASE5( a, b, c, d, e ): case 0x##f
#define CASE7( a, b, c, d, e, f, g ) CASE6( a, b, c, d, e, f ): case 0x##g
#define CASE8( a, b, c, d, e, f, g, h ) CASE7( a, b, c, d, e, f, g ): case 0x##h
// high four bits are $ED time - 8, low four bits are $DD/$FD time - 8
static byte const ed_dd_timing [0x100] = {
//0 1 2 3 4 5 6 7 8 9 A B C D E F
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x07,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x07,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x06,0x0C,0x02,0x00,0x00,0x03,0x00,0x00,0x07,0x0C,0x02,0x00,0x00,0x03,0x00,
0x00,0x00,0x00,0x00,0x0F,0x0F,0x0B,0x00,0x00,0x07,0x00,0x00,0x00,0x00,0x00,0x00,
0x40,0x40,0x70,0xC0,0x00,0x60,0x0B,0x10,0x40,0x40,0x70,0xC0,0x00,0x60,0x0B,0x10,
0x40,0x40,0x70,0xC0,0x00,0x60,0x0B,0x10,0x40,0x40,0x70,0xC0,0x00,0x60,0x0B,0x10,
0x40,0x40,0x70,0xC0,0x00,0x60,0x0B,0xA0,0x40,0x40,0x70,0xC0,0x00,0x60,0x0B,0xA0,
0x4B,0x4B,0x7B,0xCB,0x0B,0x6B,0x00,0x0B,0x40,0x40,0x70,0xC0,0x00,0x60,0x0B,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x0B,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x0B,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x0B,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x0B,0x00,
0x80,0x80,0x80,0x80,0x00,0x00,0x0B,0x00,0x80,0x80,0x80,0x80,0x00,0x00,0x0B,0x00,
0xD0,0xD0,0xD0,0xD0,0x00,0x00,0x0B,0x00,0xD0,0xD0,0xD0,0xD0,0x00,0x00,0x0B,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x0F,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x06,0x00,0x0F,0x00,0x07,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x02,0x00,0x00,0x00,0x00,0x00,0x00,
};
// even on x86, using short and unsigned char was slower
typedef int fint16;
typedef unsigned fuint16;
typedef unsigned fuint8;
bool Kss_Cpu::run( cpu_time_t end_time )
{
set_end_time( end_time );
state_t s = this->state_;
this->state = &s;
bool warning = false;
typedef BOOST::int8_t int8_t;
union {
regs_t rg;
pairs_t rp;
uint8_t r8_ [8]; // indexed
uint16_t r16_ [4];
};
rg = this->r.b;
cpu_time_t s_time = s.time;
fuint16 pc = r.pc;
fuint16 sp = r.sp;
fuint16 ix = r.ix; // TODO: keep in memory for direct access?
fuint16 iy = r.iy;
int flags = r.b.flags;
goto loop;
jr_not_taken:
s_time -= 5;
goto loop;
call_not_taken:
s_time -= 7;
jp_not_taken:
pc += 2;
loop:
check( (unsigned long) pc < 0x10000 );
check( (unsigned long) sp < 0x10000 );
check( (unsigned) flags < 0x100 );
check( (unsigned) ix < 0x10000 );
check( (unsigned) iy < 0x10000 );
uint8_t const* instr = s.read [pc >> page_shift];
#define GET_ADDR() GET_LE16( instr )
fuint8 opcode;
// TODO: eliminate this special case
#if BLARGG_NONPORTABLE
opcode = instr [pc];
pc++;
instr += pc;
#else
instr += KSS_CPU_PAGE_OFFSET( pc );
opcode = *instr++;
pc++;
#endif
static byte const base_timing [0x100] = {
// 0 1 2 3 4 5 6 7 8 9 A B C D E F
4,10, 7, 6, 4, 4, 7, 4, 4,11, 7, 6, 4, 4, 7, 4, // 0
13,10, 7, 6, 4, 4, 7, 4,12,11, 7, 6, 4, 4, 7, 4, // 1
12,10,16, 6, 4, 4, 7, 4,12,11,16, 6, 4, 4, 7, 4, // 2
12,10,13, 6,11,11,10, 4,12,11,13, 6, 4, 4, 7, 4, // 3
4, 4, 4, 4, 4, 4, 7, 4, 4, 4, 4, 4, 4, 4, 7, 4, // 4
4, 4, 4, 4, 4, 4, 7, 4, 4, 4, 4, 4, 4, 4, 7, 4, // 5
4, 4, 4, 4, 4, 4, 7, 4, 4, 4, 4, 4, 4, 4, 7, 4, // 6
7, 7, 7, 7, 7, 7, 4, 7, 4, 4, 4, 4, 4, 4, 7, 4, // 7
4, 4, 4, 4, 4, 4, 7, 4, 4, 4, 4, 4, 4, 4, 7, 4, // 8
4, 4, 4, 4, 4, 4, 7, 4, 4, 4, 4, 4, 4, 4, 7, 4, // 9
4, 4, 4, 4, 4, 4, 7, 4, 4, 4, 4, 4, 4, 4, 7, 4, // A
4, 4, 4, 4, 4, 4, 7, 4, 4, 4, 4, 4, 4, 4, 7, 4, // B
11,10,10,10,17,11, 7,11,11,10,10, 8,17,17, 7,11, // C
11,10,10,11,17,11, 7,11,11, 4,10,11,17, 8, 7,11, // D
11,10,10,19,17,11, 7,11,11, 4,10, 4,17, 8, 7,11, // E
11,10,10, 4,17,11, 7,11,11, 6,10, 4,17, 8, 7,11, // F
};
fuint16 data;
data = base_timing [opcode];
if ( (s_time += data) >= 0 )
goto possibly_out_of_time;
almost_out_of_time:
data = READ_PROG( pc );
#ifdef Z80_CPU_LOG_H
//log_opcode( opcode, READ_PROG( pc ) );
z80_log_regs( rg.a, rp.bc, rp.de, rp.hl, sp, ix, iy );
z80_cpu_log( "new", pc - 1, opcode, READ_PROG( pc ),
READ_PROG( pc + 1 ), READ_PROG( pc + 2 ) );
#endif
switch ( opcode )
{
possibly_out_of_time:
if ( s_time < (int) data )
goto almost_out_of_time;
s_time -= data;
goto out_of_time;
// Common
case 0x00: // NOP
CASE7( 40, 49, 52, 5B, 64, 6D, 7F ): // LD B,B etc.
goto loop;
case 0x08:{// EX AF,AF'
int temp = r.alt.b.a;
r.alt.b.a = rg.a;
rg.a = temp;
temp = r.alt.b.flags;
r.alt.b.flags = flags;
flags = temp;
goto loop;
}
case 0xD3: // OUT (imm),A
pc++;
OUT( data + rg.a * 0x100, rg.a );
goto loop;
case 0x2E: // LD L,imm
pc++;
rg.l = data;
goto loop;
case 0x3E: // LD A,imm
pc++;
rg.a = data;
goto loop;
case 0x3A:{// LD A,(addr)
fuint16 addr = GET_ADDR();
pc += 2;
rg.a = READ( addr );
goto loop;
}
// Conditional
#define ZERO (flags & Z40)
#define CARRY (flags & C01)
#define EVEN (flags & P04)
#define MINUS (flags & S80)
// JR
// TODO: more efficient way to handle negative branch that wraps PC around
#define JR( cond ) {\
int offset = (BOOST::int8_t) data;\
pc++;\
if ( !(cond) )\
goto jr_not_taken;\
pc = uint16_t (pc + offset);\
goto loop;\
}
case 0x20: JR( !ZERO ) // JR NZ,disp
case 0x28: JR( ZERO ) // JR Z,disp
case 0x30: JR( !CARRY ) // JR NC,disp
case 0x38: JR( CARRY ) // JR C,disp
case 0x18: JR( true ) // JR disp
case 0x10:{// DJNZ disp
int temp = rg.b - 1;
rg.b = temp;
JR( temp )
}
// JP
#define JP( cond ) if ( !(cond) ) goto jp_not_taken; pc = GET_ADDR(); goto loop;
case 0xC2: JP( !ZERO ) // JP NZ,addr
case 0xCA: JP( ZERO ) // JP Z,addr
case 0xD2: JP( !CARRY ) // JP NC,addr
case 0xDA: JP( CARRY ) // JP C,addr
case 0xE2: JP( !EVEN ) // JP PO,addr
case 0xEA: JP( EVEN ) // JP PE,addr
case 0xF2: JP( !MINUS ) // JP P,addr
case 0xFA: JP( MINUS ) // JP M,addr
case 0xC3: // JP addr
pc = GET_ADDR();
goto loop;
case 0xE9: // JP HL
pc = rp.hl;
goto loop;
// RET
#define RET( cond ) if ( cond ) goto ret_taken; s_time -= 6; goto loop;
case 0xC0: RET( !ZERO ) // RET NZ
case 0xC8: RET( ZERO ) // RET Z
case 0xD0: RET( !CARRY ) // RET NC
case 0xD8: RET( CARRY ) // RET C
case 0xE0: RET( !EVEN ) // RET PO
case 0xE8: RET( EVEN ) // RET PE
case 0xF0: RET( !MINUS ) // RET P
case 0xF8: RET( MINUS ) // RET M
case 0xC9: // RET
ret_taken:
pc = READ_WORD( sp );
sp = uint16_t (sp + 2);
goto loop;
// CALL
#define CALL( cond ) if ( cond ) goto call_taken; goto call_not_taken;
case 0xC4: CALL( !ZERO ) // CALL NZ,addr
case 0xCC: CALL( ZERO ) // CALL Z,addr
case 0xD4: CALL( !CARRY ) // CALL NC,addr
case 0xDC: CALL( CARRY ) // CALL C,addr
case 0xE4: CALL( !EVEN ) // CALL PO,addr
case 0xEC: CALL( EVEN ) // CALL PE,addr
case 0xF4: CALL( !MINUS ) // CALL P,addr
case 0xFC: CALL( MINUS ) // CALL M,addr
case 0xCD:{// CALL addr
call_taken:
fuint16 addr = pc + 2;
pc = GET_ADDR();
sp = uint16_t (sp - 2);
WRITE_WORD( sp, addr );
goto loop;
}
case 0xFF: // RST
if ( pc > idle_addr )
goto hit_idle_addr;
CASE7( C7, CF, D7, DF, E7, EF, F7 ):
data = pc;
pc = opcode & 0x38;
goto push_data;
// PUSH/POP
case 0xF5: // PUSH AF
data = rg.a * 0x100u + flags;
goto push_data;
case 0xC5: // PUSH BC
case 0xD5: // PUSH DE
case 0xE5: // PUSH HL
data = R16( opcode, 4, 0xC5 );
push_data:
sp = uint16_t (sp - 2);
WRITE_WORD( sp, data );
goto loop;
case 0xF1: // POP AF
flags = READ( sp );
rg.a = READ( sp + 1 );
sp = uint16_t (sp + 2);
goto loop;
case 0xC1: // POP BC
case 0xD1: // POP DE
case 0xE1: // POP HL
R16( opcode, 4, 0xC1 ) = READ_WORD( sp );
sp = uint16_t (sp + 2);
goto loop;
// ADC/ADD/SBC/SUB
case 0x96: // SUB (HL)
case 0x86: // ADD (HL)
flags &= ~C01;
case 0x9E: // SBC (HL)
case 0x8E: // ADC (HL)
data = READ( rp.hl );
goto adc_data;
case 0xD6: // SUB A,imm
case 0xC6: // ADD imm
flags &= ~C01;
case 0xDE: // SBC A,imm
case 0xCE: // ADC imm
pc++;
goto adc_data;
CASE7( 90, 91, 92, 93, 94, 95, 97 ): // SUB r
CASE7( 80, 81, 82, 83, 84, 85, 87 ): // ADD r
flags &= ~C01;
CASE7( 98, 99, 9A, 9B, 9C, 9D, 9F ): // SBC r
CASE7( 88, 89, 8A, 8B, 8C, 8D, 8F ): // ADC r
data = R8( opcode & 7, 0 );
adc_data: {
int result = data + (flags & C01);
data ^= rg.a;
flags = opcode >> 3 & N02; // bit 4 is set in subtract opcodes
if ( flags )
result = -result;
result += rg.a;
data ^= result;
flags |=(data & H10) |
((data - -0x80) >> 6 & V04) |
SZ28C( result & 0x1FF );
rg.a = result;
goto loop;
}
// CP
case 0xBE: // CP (HL)
data = READ( rp.hl );
goto cp_data;
case 0xFE: // CP imm
pc++;
goto cp_data;
CASE7( B8, B9, BA, BB, BC, BD, BF ): // CP r
data = R8( opcode, 0xB8 );
cp_data: {
int result = rg.a - data;
flags = N02 | (data & (F20 | F08)) | (result >> 8 & C01);
data ^= rg.a;
flags |=(((result ^ rg.a) & data) >> 5 & V04) |
(((data & H10) ^ result) & (S80 | H10));
if ( (uint8_t) result )
goto loop;
flags |= Z40;
goto loop;
}
// ADD HL,rp
case 0x39: // ADD HL,SP
data = sp;
goto add_hl_data;
case 0x09: // ADD HL,BC
case 0x19: // ADD HL,DE
case 0x29: // ADD HL,HL
data = R16( opcode, 4, 0x09 );
add_hl_data: {
blargg_ulong sum = rp.hl + data;
data ^= rp.hl;
rp.hl = (uint16_t)sum;
flags = (flags & (S80 | Z40 | V04)) |
(sum >> 16) |
(sum >> 8 & (F20 | F08)) |
((data ^ sum) >> 8 & H10);
goto loop;
}
case 0x27:{// DAA
int a = rg.a;
if ( a > 0x99 )
flags |= C01;
int adjust = 0x60 & -(flags & C01);
if ( flags & H10 || (a & 0x0F) > 9 )
adjust |= 0x06;
if ( flags & N02 )
adjust = -adjust;
a += adjust;
flags = (flags & (C01 | N02)) |
((rg.a ^ a) & H10) |
SZ28P( (uint8_t) a );
rg.a = a;
goto loop;
}
/*
case 0x27:{// DAA
// more optimized, but probably not worth the obscurity
int f = (rg.a + (0xFF - 0x99)) >> 8 | flags; // (a > 0x99 ? C01 : 0) | flags
int adjust = 0x60 & -(f & C01); // f & C01 ? 0x60 : 0
if ( (((rg.a + (0x0F - 9)) ^ rg.a) | f) & H10 ) // flags & H10 || (rg.a & 0x0F) > 9
adjust |= 0x06;
if ( f & N02 )
adjust = -adjust;
int a = rg.a + adjust;
flags = (f & (N02 | C01)) | ((rg.a ^ a) & H10) | SZ28P( (uint8_t) a );
rg.a = a;
goto loop;
}
*/
// INC/DEC
case 0x34: // INC (HL)
data = READ( rp.hl ) + 1;
WRITE( rp.hl, data );
goto inc_set_flags;
CASE7( 04, 0C, 14, 1C, 24, 2C, 3C ): // INC r
data = ++R8( opcode >> 3, 0 );
inc_set_flags:
flags = (flags & C01) |
(((data & 0x0F) - 1) & H10) |
SZ28( (uint8_t) data );
if ( data != 0x80 )
goto loop;
flags |= V04;
goto loop;
case 0x35: // DEC (HL)
data = READ( rp.hl ) - 1;
WRITE( rp.hl, data );
goto dec_set_flags;
CASE7( 05, 0D, 15, 1D, 25, 2D, 3D ): // DEC r
data = --R8( opcode >> 3, 0 );
dec_set_flags:
flags = (flags & C01) | N02 |
(((data & 0x0F) + 1) & H10) |
SZ28( (uint8_t) data );
if ( data != 0x7F )
goto loop;
flags |= V04;
goto loop;
case 0x03: // INC BC
case 0x13: // INC DE
case 0x23: // INC HL
R16( opcode, 4, 0x03 )++;
goto loop;
case 0x33: // INC SP
sp = uint16_t (sp + 1);
goto loop;
case 0x0B: // DEC BC
case 0x1B: // DEC DE
case 0x2B: // DEC HL
R16( opcode, 4, 0x0B )--;
goto loop;
case 0x3B: // DEC SP
sp = uint16_t (sp - 1);
goto loop;
// AND
case 0xA6: // AND (HL)
data = READ( rp.hl );
goto and_data;
case 0xE6: // AND imm
pc++;
goto and_data;
CASE7( A0, A1, A2, A3, A4, A5, A7 ): // AND r
data = R8( opcode, 0xA0 );
and_data:
rg.a &= data;
flags = SZ28P( rg.a ) | H10;
goto loop;
// OR
case 0xB6: // OR (HL)
data = READ( rp.hl );
goto or_data;
case 0xF6: // OR imm
pc++;
goto or_data;
CASE7( B0, B1, B2, B3, B4, B5, B7 ): // OR r
data = R8( opcode, 0xB0 );
or_data:
rg.a |= data;
flags = SZ28P( rg.a );
goto loop;
// XOR
case 0xAE: // XOR (HL)
data = READ( rp.hl );
goto xor_data;
case 0xEE: // XOR imm
pc++;
goto xor_data;
CASE7( A8, A9, AA, AB, AC, AD, AF ): // XOR r
data = R8( opcode, 0xA8 );
xor_data:
rg.a ^= data;
flags = SZ28P( rg.a );
goto loop;
// LD
CASE7( 70, 71, 72, 73, 74, 75, 77 ): // LD (HL),r
WRITE( rp.hl, R8( opcode, 0x70 ) );
goto loop;
CASE6( 41, 42, 43, 44, 45, 47 ): // LD B,r
CASE6( 48, 4A, 4B, 4C, 4D, 4F ): // LD C,r
CASE6( 50, 51, 53, 54, 55, 57 ): // LD D,r
CASE6( 58, 59, 5A, 5C, 5D, 5F ): // LD E,r
CASE6( 60, 61, 62, 63, 65, 67 ): // LD H,r
CASE6( 68, 69, 6A, 6B, 6C, 6F ): // LD L,r
CASE6( 78, 79, 7A, 7B, 7C, 7D ): // LD A,r
R8( opcode >> 3 & 7, 0 ) = R8( opcode & 7, 0 );
goto loop;
CASE5( 06, 0E, 16, 1E, 26 ): // LD r,imm
R8( opcode >> 3, 0 ) = data;
pc++;
goto loop;
case 0x36: // LD (HL),imm
pc++;
WRITE( rp.hl, data );
goto loop;
CASE7( 46, 4E, 56, 5E, 66, 6E, 7E ): // LD r,(HL)
R8( opcode >> 3, 8 ) = READ( rp.hl );
goto loop;
case 0x01: // LD rp,imm
case 0x11:
case 0x21:
R16( opcode, 4, 0x01 ) = GET_ADDR();
pc += 2;
goto loop;
case 0x31: // LD sp,imm
sp = GET_ADDR();
pc += 2;
goto loop;
case 0x2A:{// LD HL,(addr)
fuint16 addr = GET_ADDR();
pc += 2;
rp.hl = READ_WORD( addr );
goto loop;
}
case 0x32:{// LD (addr),A
fuint16 addr = GET_ADDR();
pc += 2;
WRITE( addr, rg.a );
goto loop;
}
case 0x22:{// LD (addr),HL
fuint16 addr = GET_ADDR();
pc += 2;
WRITE_WORD( addr, rp.hl );
goto loop;
}
case 0x02: // LD (BC),A
case 0x12: // LD (DE),A
WRITE( R16( opcode, 4, 0x02 ), rg.a );
goto loop;
case 0x0A: // LD A,(BC)
case 0x1A: // LD A,(DE)
rg.a = READ( R16( opcode, 4, 0x0A ) );
goto loop;
case 0xF9: // LD SP,HL
sp = rp.hl;
goto loop;
// Rotate
case 0x07:{// RLCA
fuint16 temp = rg.a;
temp = (temp << 1) | (temp >> 7);
flags = (flags & (S80 | Z40 | P04)) |
(temp & (F20 | F08 | C01));
rg.a = temp;
goto loop;
}
case 0x0F:{// RRCA
fuint16 temp = rg.a;
flags = (flags & (S80 | Z40 | P04)) |
(temp & C01);
temp = (temp << 7) | (temp >> 1);
flags |= temp & (F20 | F08);
rg.a = temp;
goto loop;
}
case 0x17:{// RLA
blargg_ulong temp = (rg.a << 1) | (flags & C01);
flags = (flags & (S80 | Z40 | P04)) |
(temp & (F20 | F08)) |
(temp >> 8);
rg.a = (uint8_t)temp;
goto loop;
}
case 0x1F:{// RRA
fuint16 temp = (flags << 7) | (rg.a >> 1);
flags = (flags & (S80 | Z40 | P04)) |
(temp & (F20 | F08)) |
(rg.a & C01);
rg.a = temp;
goto loop;
}
// Misc
case 0x2F:{// CPL
fuint16 temp = ~rg.a;
flags = (flags & (S80 | Z40 | P04 | C01)) |
(temp & (F20 | F08)) |
(H10 | N02);
rg.a = temp;
goto loop;
}
case 0x3F:{// CCF
flags = ((flags & (S80 | Z40 | P04 | C01)) ^ C01) |
(flags << 4 & H10) |
(rg.a & (F20 | F08));
goto loop;
}
case 0x37: // SCF
flags = (flags & (S80 | Z40 | P04)) | C01 |
(rg.a & (F20 | F08));
goto loop;
case 0xDB: // IN A,(imm)
pc++;
rg.a = IN( data + rg.a * 0x100 );
goto loop;
case 0xE3:{// EX (SP),HL
fuint16 temp = READ_WORD( sp );
WRITE_WORD( sp, rp.hl );
rp.hl = temp;
goto loop;
}
case 0xEB:{// EX DE,HL
fuint16 temp = rp.hl;
rp.hl = rp.de;
rp.de = temp;
goto loop;
}
case 0xD9:{// EXX DE,HL
fuint16 temp = r.alt.w.bc;
r.alt.w.bc = rp.bc;
rp.bc = temp;
temp = r.alt.w.de;
r.alt.w.de = rp.de;
rp.de = temp;
temp = r.alt.w.hl;
r.alt.w.hl = rp.hl;
rp.hl = temp;
goto loop;
}
case 0xF3: // DI
r.iff1 = 0;
r.iff2 = 0;
goto loop;
case 0xFB: // EI
r.iff1 = 1;
r.iff2 = 1;
// TODO: delayed effect
goto loop;
case 0x76: // HALT
goto halt;
//////////////////////////////////////// CB prefix
{
case 0xCB:
unsigned data2;
data2 = instr [1];
(void) data2; // TODO is this the same as data in all cases?
pc++;
switch ( data )
{
// Rotate left
#define RLC( read, write ) {\
fuint8 result = read;\
result = uint8_t (result << 1) | (result >> 7);\
flags = SZ28P( result ) | (result & C01);\
write;\
goto loop;\
}
case 0x06: // RLC (HL)
s_time += 7;
data = rp.hl;
rlc_data_addr:
RLC( READ( data ), WRITE( data, result ) )
CASE7( 00, 01, 02, 03, 04, 05, 07 ):{// RLC r
uint8_t& reg = R8( data, 0 );
RLC( reg, reg = result )
}
#define RL( read, write ) {\
fuint16 result = (read << 1) | (flags & C01);\
flags = SZ28PC( result );\
write;\
goto loop;\
}
case 0x16: // RL (HL)
s_time += 7;
data = rp.hl;
rl_data_addr:
RL( READ( data ), WRITE( data, result ) )
CASE7( 10, 11, 12, 13, 14, 15, 17 ):{// RL r
uint8_t& reg = R8( data, 0x10 );
RL( reg, reg = result )
}
#define SLA( read, add, write ) {\
fuint16 result = (read << 1) | add;\
flags = SZ28PC( result );\
write;\
goto loop;\
}
case 0x26: // SLA (HL)
s_time += 7;
data = rp.hl;
sla_data_addr:
SLA( READ( data ), 0, WRITE( data, result ) )
CASE7( 20, 21, 22, 23, 24, 25, 27 ):{// SLA r
uint8_t& reg = R8( data, 0x20 );
SLA( reg, 0, reg = result )
}
case 0x36: // SLL (HL)
s_time += 7;
data = rp.hl;
sll_data_addr:
SLA( READ( data ), 1, WRITE( data, result ) )
CASE7( 30, 31, 32, 33, 34, 35, 37 ):{// SLL r
uint8_t& reg = R8( data, 0x30 );
SLA( reg, 1, reg = result )
}
// Rotate right
#define RRC( read, write ) {\
fuint8 result = read;\
flags = result & C01;\
result = uint8_t (result << 7) | (result >> 1);\
flags |= SZ28P( result );\
write;\
goto loop;\
}
case 0x0E: // RRC (HL)
s_time += 7;
data = rp.hl;
rrc_data_addr:
RRC( READ( data ), WRITE( data, result ) )
CASE7( 08, 09, 0A, 0B, 0C, 0D, 0F ):{// RRC r
uint8_t& reg = R8( data, 0x08 );
RRC( reg, reg = result )
}
#define RR( read, write ) {\
fuint8 result = read;\
fuint8 temp = result & C01;\
result = uint8_t (flags << 7) | (result >> 1);\
flags = SZ28P( result ) | temp;\
write;\
goto loop;\
}
case 0x1E: // RR (HL)
s_time += 7;
data = rp.hl;
rr_data_addr:
RR( READ( data ), WRITE( data, result ) )
CASE7( 18, 19, 1A, 1B, 1C, 1D, 1F ):{// RR r
uint8_t& reg = R8( data, 0x18 );
RR( reg, reg = result )
}
#define SRA( read, write ) {\
fuint8 result = read;\
flags = result & C01;\
result = (result & 0x80) | (result >> 1);\
flags |= SZ28P( result );\
write;\
goto loop;\
}
case 0x2E: // SRA (HL)
data = rp.hl;
s_time += 7;
sra_data_addr:
SRA( READ( data ), WRITE( data, result ) )
CASE7( 28, 29, 2A, 2B, 2C, 2D, 2F ):{// SRA r
uint8_t& reg = R8( data, 0x28 );
SRA( reg, reg = result )
}
#define SRL( read, write ) {\
fuint8 result = read;\
flags = result & C01;\
result >>= 1;\
flags |= SZ28P( result );\
write;\
goto loop;\
}
case 0x3E: // SRL (HL)
s_time += 7;
data = rp.hl;
srl_data_addr:
SRL( READ( data ), WRITE( data, result ) )
CASE7( 38, 39, 3A, 3B, 3C, 3D, 3F ):{// SRL r
uint8_t& reg = R8( data, 0x38 );
SRL( reg, reg = result )
}
// BIT
{
unsigned temp;
CASE8( 46, 4E, 56, 5E, 66, 6E, 76, 7E ): // BIT b,(HL)
s_time += 4;
temp = READ( rp.hl );
flags &= C01;
goto bit_temp;
CASE7( 40, 41, 42, 43, 44, 45, 47 ): // BIT 0,r
CASE7( 48, 49, 4A, 4B, 4C, 4D, 4F ): // BIT 1,r
CASE7( 50, 51, 52, 53, 54, 55, 57 ): // BIT 2,r
CASE7( 58, 59, 5A, 5B, 5C, 5D, 5F ): // BIT 3,r
CASE7( 60, 61, 62, 63, 64, 65, 67 ): // BIT 4,r
CASE7( 68, 69, 6A, 6B, 6C, 6D, 6F ): // BIT 5,r
CASE7( 70, 71, 72, 73, 74, 75, 77 ): // BIT 6,r
CASE7( 78, 79, 7A, 7B, 7C, 7D, 7F ): // BIT 7,r
temp = R8( data & 7, 0 );
flags = (flags & C01) | (temp & (F20 | F08));
bit_temp:
int masked = temp & 1 << (data >> 3 & 7);
flags |=(masked & S80) | H10 |
((masked - 1) >> 8 & (Z40 | P04));
goto loop;
}
// SET/RES
CASE8( 86, 8E, 96, 9E, A6, AE, B6, BE ): // RES b,(HL)
CASE8( C6, CE, D6, DE, E6, EE, F6, FE ):{// SET b,(HL)
s_time += 7;
int temp = READ( rp.hl );
int bit = 1 << (data >> 3 & 7);
temp |= bit; // SET
if ( !(data & 0x40) )
temp ^= bit; // RES
WRITE( rp.hl, temp );
goto loop;
}
CASE7( C0, C1, C2, C3, C4, C5, C7 ): // SET 0,r
CASE7( C8, C9, CA, CB, CC, CD, CF ): // SET 1,r
CASE7( D0, D1, D2, D3, D4, D5, D7 ): // SET 2,r
CASE7( D8, D9, DA, DB, DC, DD, DF ): // SET 3,r
CASE7( E0, E1, E2, E3, E4, E5, E7 ): // SET 4,r
CASE7( E8, E9, EA, EB, EC, ED, EF ): // SET 5,r
CASE7( F0, F1, F2, F3, F4, F5, F7 ): // SET 6,r
CASE7( F8, F9, FA, FB, FC, FD, FF ): // SET 7,r
R8( data & 7, 0 ) |= 1 << (data >> 3 & 7);
goto loop;
CASE7( 80, 81, 82, 83, 84, 85, 87 ): // RES 0,r
CASE7( 88, 89, 8A, 8B, 8C, 8D, 8F ): // RES 1,r
CASE7( 90, 91, 92, 93, 94, 95, 97 ): // RES 2,r
CASE7( 98, 99, 9A, 9B, 9C, 9D, 9F ): // RES 3,r
CASE7( A0, A1, A2, A3, A4, A5, A7 ): // RES 4,r
CASE7( A8, A9, AA, AB, AC, AD, AF ): // RES 5,r
CASE7( B0, B1, B2, B3, B4, B5, B7 ): // RES 6,r
CASE7( B8, B9, BA, BB, BC, BD, BF ): // RES 7,r
R8( data & 7, 0 ) &= ~(1 << (data >> 3 & 7));
goto loop;
}
assert( false );
}
#undef GET_ADDR
#define GET_ADDR() GET_LE16( instr + 1 )
//////////////////////////////////////// ED prefix
{
case 0xED:
pc++;
s_time += ed_dd_timing [data] >> 4;
switch ( data )
{
{
blargg_ulong temp;
case 0x72: // SBC HL,SP
case 0x7A: // ADC HL,SP
temp = sp;
if ( 0 )
case 0x42: // SBC HL,BC
case 0x52: // SBC HL,DE
case 0x62: // SBC HL,HL
case 0x4A: // ADC HL,BC
case 0x5A: // ADC HL,DE
case 0x6A: // ADC HL,HL
temp = R16( data >> 3 & 6, 1, 0 );
blargg_ulong sum = temp + (flags & C01);
flags = ~data >> 2 & N02;
if ( flags )
sum = (blargg_ulong)-(blargg_long)sum;
sum += rp.hl;
temp ^= rp.hl;
temp ^= sum;
flags |=(sum >> 16 & C01) |
(temp >> 8 & H10) |
(sum >> 8 & (S80 | F20 | F08)) |
((temp - -0x8000) >> 14 & V04);
rp.hl = (uint16_t)sum;
if ( (uint16_t) sum )
goto loop;
flags |= Z40;
goto loop;
}
CASE8( 40, 48, 50, 58, 60, 68, 70, 78 ):{// IN r,(C)
int temp = IN( rp.bc );
R8( data >> 3, 8 ) = temp;
flags = (flags & C01) | SZ28P( temp );
goto loop;
}
case 0x71: // OUT (C),0
rg.flags = 0;
CASE7( 41, 49, 51, 59, 61, 69, 79 ): // OUT (C),r
OUT( rp.bc, R8( data >> 3, 8 ) );
goto loop;
{
unsigned temp;
case 0x73: // LD (ADDR),SP
temp = sp;
if ( 0 )
case 0x43: // LD (ADDR),BC
case 0x53: // LD (ADDR),DE
temp = R16( data, 4, 0x43 );
fuint16 addr = GET_ADDR();
pc += 2;
WRITE_WORD( addr, temp );
goto loop;
}
case 0x4B: // LD BC,(ADDR)
case 0x5B:{// LD DE,(ADDR)
fuint16 addr = GET_ADDR();
pc += 2;
R16( data, 4, 0x4B ) = READ_WORD( addr );
goto loop;
}
case 0x7B:{// LD SP,(ADDR)
fuint16 addr = GET_ADDR();
pc += 2;
sp = READ_WORD( addr );
goto loop;
}
case 0x67:{// RRD
fuint8 temp = READ( rp.hl );
WRITE( rp.hl, (rg.a << 4) | (temp >> 4) );
temp = (rg.a & 0xF0) | (temp & 0x0F);
flags = (flags & C01) | SZ28P( temp );
rg.a = temp;
goto loop;
}
case 0x6F:{// RLD
fuint8 temp = READ( rp.hl );
WRITE( rp.hl, (temp << 4) | (rg.a & 0x0F) );
temp = (rg.a & 0xF0) | (temp >> 4);
flags = (flags & C01) | SZ28P( temp );
rg.a = temp;
goto loop;
}
CASE8( 44, 4C, 54, 5C, 64, 6C, 74, 7C ): // NEG
opcode = 0x10; // flag to do SBC instead of ADC
flags &= ~C01;
data = rg.a;
rg.a = 0;
goto adc_data;
{
int inc;
case 0xA9: // CPD
case 0xB9: // CPDR
inc = -1;
if ( 0 )
case 0xA1: // CPI
case 0xB1: // CPIR
inc = +1;
fuint16 addr = rp.hl;
rp.hl = addr + inc;
int temp = READ( addr );
int result = rg.a - temp;
flags = (flags & C01) | N02 |
((((temp ^ rg.a) & H10) ^ result) & (S80 | H10));
if ( !(uint8_t) result ) flags |= Z40;
result -= (flags & H10) >> 4;
flags |= result & F08;
flags |= result << 4 & F20;
if ( !--rp.bc )
goto loop;
flags |= V04;
if ( flags & Z40 || data < 0xB0 )
goto loop;
pc -= 2;
s_time += 5;
goto loop;
}
{
int inc;
case 0xA8: // LDD
case 0xB8: // LDDR
inc = -1;
if ( 0 )
case 0xA0: // LDI
case 0xB0: // LDIR
inc = +1;
fuint16 addr = rp.hl;
rp.hl = addr + inc;
int temp = READ( addr );
addr = rp.de;
rp.de = addr + inc;
WRITE( addr, temp );
temp += rg.a;
flags = (flags & (S80 | Z40 | C01)) |
(temp & F08) | (temp << 4 & F20);
if ( !--rp.bc )
goto loop;
flags |= V04;
if ( data < 0xB0 )
goto loop;
pc -= 2;
s_time += 5;
goto loop;
}
{
int inc;
case 0xAB: // OUTD
case 0xBB: // OTDR
inc = -1;
if ( 0 )
case 0xA3: // OUTI
case 0xB3: // OTIR
inc = +1;
fuint16 addr = rp.hl;
rp.hl = addr + inc;
int temp = READ( addr );
int b = --rg.b;
flags = (temp >> 6 & N02) | SZ28( b );
if ( b && data >= 0xB0 )
{
pc -= 2;
s_time += 5;
}
OUT( rp.bc, temp );
goto loop;
}
{
int inc;
case 0xAA: // IND
case 0xBA: // INDR
inc = -1;
if ( 0 )
case 0xA2: // INI
case 0xB2: // INIR
inc = +1;
fuint16 addr = rp.hl;
rp.hl = addr + inc;
int temp = IN( rp.bc );
int b = --rg.b;
flags = (temp >> 6 & N02) | SZ28( b );
if ( b && data >= 0xB0 )
{
pc -= 2;
s_time += 5;
}
WRITE( addr, temp );
goto loop;
}
case 0x47: // LD I,A
r.i = rg.a;
goto loop;
case 0x4F: // LD R,A
SET_R( rg.a );
debug_printf( "LD R,A not supported\n" );
warning = true;
goto loop;
case 0x57: // LD A,I
rg.a = r.i;
goto ld_ai_common;
case 0x5F: // LD A,R
rg.a = GET_R();
debug_printf( "LD A,R not supported\n" );
warning = true;
ld_ai_common:
flags = (flags & C01) | SZ28( rg.a ) | (r.iff2 << 2 & V04);
goto loop;
CASE8( 45, 4D, 55, 5D, 65, 6D, 75, 7D ): // RETI/RETN
r.iff1 = r.iff2;
goto ret_taken;
case 0x46: case 0x4E: case 0x66: case 0x6E: // IM 0
r.im = 0;
goto loop;
case 0x56: case 0x76: // IM 1
r.im = 1;
goto loop;
case 0x5E: case 0x7E: // IM 2
r.im = 2;
goto loop;
default:
debug_printf( "Opcode $ED $%02X not supported\n", data );
warning = true;
goto loop;
}
assert( false );
}
//////////////////////////////////////// DD/FD prefix
{
fuint16 ixy;
case 0xDD:
ixy = ix;
goto ix_prefix;
case 0xFD:
ixy = iy;
ix_prefix:
pc++;
unsigned data2 = READ_PROG( pc );
s_time += ed_dd_timing [data] & 0x0F;
switch ( data )
{
// TODO: more efficient way of avoid negative address
// TODO: avoid using this as argument to READ() since it is evaluated twice
#define IXY_DISP( ixy, disp ) uint16_t ((ixy) + (disp))
#define SET_IXY( in ) if ( opcode == 0xDD ) ix = in; else iy = in;
// ADD/ADC/SUB/SBC
case 0x96: // SUB (IXY+disp)
case 0x86: // ADD (IXY+disp)
flags &= ~C01;
case 0x9E: // SBC (IXY+disp)
case 0x8E: // ADC (IXY+disp)
pc++;
opcode = data;
data = READ( IXY_DISP( ixy, (int8_t) data2 ) );
goto adc_data;
case 0x94: // SUB HXY
case 0x84: // ADD HXY
flags &= ~C01;
case 0x9C: // SBC HXY
case 0x8C: // ADC HXY
opcode = data;
data = ixy >> 8;
goto adc_data;
case 0x95: // SUB LXY
case 0x85: // ADD LXY
flags &= ~C01;
case 0x9D: // SBC LXY
case 0x8D: // ADC LXY
opcode = data;
data = (uint8_t) ixy;
goto adc_data;
{
unsigned temp;
case 0x39: // ADD IXY,SP
temp = sp;
goto add_ixy_data;
case 0x29: // ADD IXY,HL
temp = ixy;
goto add_ixy_data;
case 0x09: // ADD IXY,BC
case 0x19: // ADD IXY,DE
temp = R16( data, 4, 0x09 );
add_ixy_data: {
blargg_ulong sum = ixy + temp;
temp ^= ixy;
ixy = (uint16_t) sum;
flags = (flags & (S80 | Z40 | V04)) |
(sum >> 16) |
(sum >> 8 & (F20 | F08)) |
((temp ^ sum) >> 8 & H10);
goto set_ixy;
}
}
// AND
case 0xA6: // AND (IXY+disp)
pc++;
data = READ( IXY_DISP( ixy, (int8_t) data2 ) );
goto and_data;
case 0xA4: // AND HXY
data = ixy >> 8;
goto and_data;
case 0xA5: // AND LXY
data = (uint8_t) ixy;
goto and_data;
// OR
case 0xB6: // OR (IXY+disp)
pc++;
data = READ( IXY_DISP( ixy, (int8_t) data2 ) );
goto or_data;
case 0xB4: // OR HXY
data = ixy >> 8;
goto or_data;
case 0xB5: // OR LXY
data = (uint8_t) ixy;
goto or_data;
// XOR
case 0xAE: // XOR (IXY+disp)
pc++;
data = READ( IXY_DISP( ixy, (int8_t) data2 ) );
goto xor_data;
case 0xAC: // XOR HXY
data = ixy >> 8;
goto xor_data;
case 0xAD: // XOR LXY
data = (uint8_t) ixy;
goto xor_data;
// CP
case 0xBE: // CP (IXY+disp)
pc++;
data = READ( IXY_DISP( ixy, (int8_t) data2 ) );
goto cp_data;
case 0xBC: // CP HXY
data = ixy >> 8;
goto cp_data;
case 0xBD: // CP LXY
data = (uint8_t) ixy;
goto cp_data;
// LD
CASE7( 70, 71, 72, 73, 74, 75, 77 ): // LD (IXY+disp),r
data = R8( data, 0x70 );
if ( 0 )
case 0x36: // LD (IXY+disp),imm
pc++, data = READ_PROG( pc );
pc++;
WRITE( IXY_DISP( ixy, (int8_t) data2 ), data );
goto loop;
CASE5( 44, 4C, 54, 5C, 7C ): // LD r,HXY
R8( data >> 3, 8 ) = ixy >> 8;
goto loop;
case 0x64: // LD HXY,HXY
case 0x6D: // LD LXY,LXY
goto loop;
CASE5( 45, 4D, 55, 5D, 7D ): // LD r,LXY
R8( data >> 3, 8 ) = ixy;
goto loop;
CASE7( 46, 4E, 56, 5E, 66, 6E, 7E ): // LD r,(IXY+disp)
pc++;
R8( data >> 3, 8 ) = READ( IXY_DISP( ixy, (int8_t) data2 ) );
goto loop;
case 0x26: // LD HXY,imm
pc++;
goto ld_hxy_data;
case 0x65: // LD HXY,LXY
data2 = (uint8_t) ixy;
goto ld_hxy_data;
CASE5( 60, 61, 62, 63, 67 ): // LD HXY,r
data2 = R8( data, 0x60 );
ld_hxy_data:
ixy = (uint8_t) ixy | (data2 << 8);
goto set_ixy;
case 0x2E: // LD LXY,imm
pc++;
goto ld_lxy_data;
case 0x6C: // LD LXY,HXY
data2 = ixy >> 8;
goto ld_lxy_data;
CASE5( 68, 69, 6A, 6B, 6F ): // LD LXY,r
data2 = R8( data, 0x68 );
ld_lxy_data:
ixy = (ixy & 0xFF00) | data2;
set_ixy:
if ( opcode == 0xDD )
{
ix = ixy;
goto loop;
}
iy = ixy;
goto loop;
case 0xF9: // LD SP,IXY
sp = ixy;
goto loop;
case 0x22:{// LD (ADDR),IXY
fuint16 addr = GET_ADDR();
pc += 2;
WRITE_WORD( addr, ixy );
goto loop;
}
case 0x21: // LD IXY,imm
ixy = GET_ADDR();
pc += 2;
goto set_ixy;
case 0x2A:{// LD IXY,(addr)
fuint16 addr = GET_ADDR();
ixy = READ_WORD( addr );
pc += 2;
goto set_ixy;
}
// DD/FD CB prefix
case 0xCB: {
data = IXY_DISP( ixy, (int8_t) data2 );
pc++;
data2 = READ_PROG( pc );
pc++;
switch ( data2 )
{
case 0x06: goto rlc_data_addr; // RLC (IXY)
case 0x16: goto rl_data_addr; // RL (IXY)
case 0x26: goto sla_data_addr; // SLA (IXY)
case 0x36: goto sll_data_addr; // SLL (IXY)
case 0x0E: goto rrc_data_addr; // RRC (IXY)
case 0x1E: goto rr_data_addr; // RR (IXY)
case 0x2E: goto sra_data_addr; // SRA (IXY)
case 0x3E: goto srl_data_addr; // SRL (IXY)
CASE8( 46, 4E, 56, 5E, 66, 6E, 76, 7E ):{// BIT b,(IXY+disp)
fuint8 temp = READ( data );
int masked = temp & 1 << (data2 >> 3 & 7);
flags = (flags & C01) | H10 |
(masked & S80) |
((masked - 1) >> 8 & (Z40 | P04));
goto loop;
}
CASE8( 86, 8E, 96, 9E, A6, AE, B6, BE ): // RES b,(IXY+disp)
CASE8( C6, CE, D6, DE, E6, EE, F6, FE ):{// SET b,(IXY+disp)
int temp = READ( data );
int bit = 1 << (data2 >> 3 & 7);
temp |= bit; // SET
if ( !(data2 & 0x40) )
temp ^= bit; // RES
WRITE( data, temp );
goto loop;
}
default:
debug_printf( "Opcode $%02X $CB $%02X not supported\n", opcode, data2 );
warning = true;
goto loop;
}
assert( false );
}
// INC/DEC
case 0x23: // INC IXY
ixy = uint16_t (ixy + 1);
goto set_ixy;
case 0x2B: // DEC IXY
ixy = uint16_t (ixy - 1);
goto set_ixy;
case 0x34: // INC (IXY+disp)
ixy = IXY_DISP( ixy, (int8_t) data2 );
pc++;
data = READ( ixy ) + 1;
WRITE( ixy, data );
goto inc_set_flags;
case 0x35: // DEC (IXY+disp)
ixy = IXY_DISP( ixy, (int8_t) data2 );
pc++;
data = READ( ixy ) - 1;
WRITE( ixy, data );
goto dec_set_flags;
case 0x24: // INC HXY
ixy = uint16_t (ixy + 0x100);
data = ixy >> 8;
goto inc_xy_common;
case 0x2C: // INC LXY
data = uint8_t (ixy + 1);
ixy = (ixy & 0xFF00) | data;
inc_xy_common:
if ( opcode == 0xDD )
{
ix = ixy;
goto inc_set_flags;
}
iy = ixy;
goto inc_set_flags;
case 0x25: // DEC HXY
ixy = uint16_t (ixy - 0x100);
data = ixy >> 8;
goto dec_xy_common;
case 0x2D: // DEC LXY
data = uint8_t (ixy - 1);
ixy = (ixy & 0xFF00) | data;
dec_xy_common:
if ( opcode == 0xDD )
{
ix = ixy;
goto dec_set_flags;
}
iy = ixy;
goto dec_set_flags;
// PUSH/POP
case 0xE5: // PUSH IXY
data = ixy;
goto push_data;
case 0xE1:{// POP IXY
ixy = READ_WORD( sp );
sp = uint16_t (sp + 2);
goto set_ixy;
}
// Misc
case 0xE9: // JP (IXY)
pc = ixy;
goto loop;
case 0xE3:{// EX (SP),IXY
fuint16 temp = READ_WORD( sp );
WRITE_WORD( sp, ixy );
ixy = temp;
goto set_ixy;
}
default:
debug_printf( "Unnecessary DD/FD prefix encountered\n" );
warning = true;
pc--;
goto loop;
}
assert( false );
}
}
debug_printf( "Unhandled main opcode: $%02X\n", opcode );
assert( false );
hit_idle_addr:
s_time -= 11;
goto out_of_time;
halt:
s_time &= 3; // increment by multiple of 4
out_of_time:
pc--;
s.time = s_time;
rg.flags = flags;
r.ix = ix;
r.iy = iy;
r.sp = sp;
r.pc = pc;
this->r.b = rg;
this->state_ = s;
this->state = &this->state_;
return warning;
}