1184 lines
29 KiB
C#
1184 lines
29 KiB
C#
using System;
|
|
using BizHawk.Common.NumberExtensions;
|
|
|
|
namespace BizHawk.Emulation.Cores.Nintendo.GBHawk
|
|
{
|
|
public class GB_PPU : PPU
|
|
{
|
|
public override byte ReadReg(int addr)
|
|
{
|
|
byte ret = 0;
|
|
|
|
switch (addr)
|
|
{
|
|
case 0xFF40: ret = LCDC; break; // LCDC
|
|
case 0xFF41: ret = STAT; break; // STAT
|
|
case 0xFF42: ret = scroll_y; break; // SCY
|
|
case 0xFF43: ret = scroll_x; break; // SCX
|
|
case 0xFF44: ret = LY; break; // LY
|
|
case 0xFF45: ret = LYC; break; // LYC
|
|
case 0xFF46: ret = DMA_addr; break; // DMA
|
|
case 0xFF47: ret = BGP; break; // BGP
|
|
case 0xFF48: ret = obj_pal_0; break; // OBP0
|
|
case 0xFF49: ret = obj_pal_1; break; // OBP1
|
|
case 0xFF4A: ret = window_y; break; // WY
|
|
case 0xFF4B: ret = window_x; break; // WX
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
public override void WriteReg(int addr, byte value)
|
|
{
|
|
switch (addr)
|
|
{
|
|
case 0xFF40: // LCDC
|
|
if (LCDC.Bit(7) && !value.Bit(7))
|
|
{
|
|
VRAM_access_read = true;
|
|
VRAM_access_write = true;
|
|
OAM_access_read = true;
|
|
OAM_access_write = true;
|
|
}
|
|
|
|
if (!LCDC.Bit(7) && value.Bit(7))
|
|
{
|
|
// don't draw for one frame after turning on
|
|
blank_frame = true;
|
|
}
|
|
|
|
LCDC = value;
|
|
break;
|
|
case 0xFF41: // STAT
|
|
// writing to STAT during mode 0 or 1 causes a STAT IRQ
|
|
// this appears to be a glitchy LYC compare
|
|
if (LCDC.Bit(7))
|
|
{
|
|
if (((STAT & 3) == 0) || ((STAT & 3) == 1))
|
|
{
|
|
LYC_INT = true;
|
|
//if (Core.REG_FFFF.Bit(1)) { Core.cpu.FlagI = true; }
|
|
//Core.REG_FF0F |= 0x02;
|
|
}
|
|
else
|
|
{
|
|
if (value.Bit(6))
|
|
{
|
|
if (LY == LYC) { LYC_INT = true; }
|
|
else { LYC_INT = false; }
|
|
}
|
|
}
|
|
}
|
|
STAT = (byte)((value & 0xF8) | (STAT & 7) | 0x80);
|
|
|
|
//if (!STAT.Bit(6)) { LYC_INT = false; }
|
|
if (!STAT.Bit(4)) { VBL_INT = false; }
|
|
break;
|
|
case 0xFF42: // SCY
|
|
scroll_y = value;
|
|
break;
|
|
case 0xFF43: // SCX
|
|
scroll_x = value;
|
|
break;
|
|
case 0xFF44: // LY
|
|
LY = 0; /*reset*/
|
|
break;
|
|
case 0xFF45: // LYC
|
|
LYC = value;
|
|
if (LCDC.Bit(7))
|
|
{
|
|
if (LY != LYC) { STAT &= 0xFB; LYC_INT = false; }
|
|
else { STAT |= 0x4; LYC_INT = true; }
|
|
|
|
// special case: the transition from 153 -> 0 acts strange
|
|
// the comparison to 153 expects to be true for longer then the value of LY expects to be 153
|
|
// this appears to be fixed in CGB
|
|
if ((LY_inc == 0) && cycle == 8)
|
|
{
|
|
if (153 != LYC) { STAT &= 0xFB; LYC_INT = false; }
|
|
else { STAT |= 0x4; LYC_INT = true; }
|
|
}
|
|
}
|
|
break;
|
|
case 0xFF46: // DMA
|
|
DMA_addr = value;
|
|
DMA_start = true;
|
|
DMA_OAM_access = true;
|
|
DMA_clock = 0;
|
|
DMA_inc = 0;
|
|
break;
|
|
case 0xFF47: // BGP
|
|
BGP = value;
|
|
break;
|
|
case 0xFF48: // OBP0
|
|
obj_pal_0 = value;
|
|
break;
|
|
case 0xFF49: // OBP1
|
|
obj_pal_1 = value;
|
|
break;
|
|
case 0xFF4A: // WY
|
|
window_y = value;
|
|
break;
|
|
case 0xFF4B: // WX
|
|
window_x = value;
|
|
break;
|
|
}
|
|
}
|
|
|
|
public override void tick()
|
|
{
|
|
// the ppu only does anything if it is turned on via bit 7 of LCDC
|
|
if (LCDC.Bit(7))
|
|
{
|
|
// start the next scanline
|
|
if (cycle == 456)
|
|
{
|
|
// scanline callback
|
|
if ((LY + LY_inc) == Core._scanlineCallbackLine)
|
|
{
|
|
if (Core._scanlineCallback != null)
|
|
{
|
|
Core.GetGPU();
|
|
Core._scanlineCallback(LCDC);
|
|
}
|
|
}
|
|
|
|
cycle = 0;
|
|
LY += LY_inc;
|
|
Core.cpu.LY = LY;
|
|
|
|
no_scan = false;
|
|
|
|
if (LY == 0 && LY_inc == 0)
|
|
{
|
|
LY_inc = 1;
|
|
Core.in_vblank = false;
|
|
|
|
STAT &= 0xFC;
|
|
|
|
// special note here, the y coordiate of the window is kept if the window is deactivated
|
|
// meaning it will pick up where it left off if re-enabled later
|
|
// so we don't reset it in the scanline loop
|
|
window_y_tile = 0;
|
|
window_y_tile_inc = 0;
|
|
window_started = false;
|
|
if (!LCDC.Bit(5)) { window_is_reset = true; }
|
|
}
|
|
|
|
// Automatically restore access to VRAM at this time (force end drawing)
|
|
// Who Framed Roger Rabbit seems to run into this.
|
|
VRAM_access_write = true;
|
|
VRAM_access_read = true;
|
|
|
|
if (LY == 144)
|
|
{
|
|
Core.in_vblank = true;
|
|
}
|
|
}
|
|
|
|
// exit vblank if LCD went from off to on
|
|
if (LCD_was_off)
|
|
{
|
|
//VBL_INT = false;
|
|
Core.in_vblank = false;
|
|
LCD_was_off = false;
|
|
|
|
// we exit vblank into mode 0 for 4 cycles
|
|
// but no hblank interrupt, presumably this only happens transitioning from mode 3 to 0
|
|
STAT &= 0xFC;
|
|
|
|
// also the LCD doesn't turn on right away
|
|
|
|
// also, the LCD does not enter mode 2 on scanline 0 when first turned on
|
|
no_scan = true;
|
|
cycle = 8;
|
|
}
|
|
|
|
// the VBL stat is continuously asserted
|
|
if (LY >= 144)
|
|
{
|
|
if (STAT.Bit(4))
|
|
{
|
|
if ((cycle >= 4) && (LY == 144))
|
|
{
|
|
VBL_INT = true;
|
|
}
|
|
else if (LY > 144)
|
|
{
|
|
VBL_INT = true;
|
|
}
|
|
}
|
|
|
|
if ((cycle == 2) && (LY == 144))
|
|
{
|
|
// there is an edge case where a VBL INT is triggered if STAT bit 5 is set
|
|
if (STAT.Bit(5)) { VBL_INT = true; }
|
|
}
|
|
|
|
if ((cycle == 4) && (LY == 144))
|
|
{
|
|
HBL_INT = false;
|
|
|
|
// set STAT mode to 1 (VBlank) and interrupt flag if it is enabled
|
|
STAT &= 0xFC;
|
|
STAT |= 0x01;
|
|
|
|
if (Core.REG_FFFF.Bit(0)) { Core.cpu.FlagI = true; }
|
|
Core.REG_FF0F |= 0x01;
|
|
}
|
|
|
|
if ((cycle == 4) && (LY == 144))
|
|
{
|
|
if (STAT.Bit(5)) { VBL_INT = false; }
|
|
}
|
|
|
|
if ((cycle == 6) && (LY == 153))
|
|
{
|
|
LY = 0;
|
|
LY_inc = 0;
|
|
Core.cpu.LY = LY;
|
|
}
|
|
}
|
|
|
|
if (!Core.in_vblank)
|
|
{
|
|
if (no_scan)
|
|
{
|
|
// timings are slightly different if we just turned on the LCD
|
|
// there is no mode 2 (presumably it missed the trigger)
|
|
// mode 3 is very short, probably in some self test mode before turning on?
|
|
|
|
if (cycle == 8)
|
|
{
|
|
if (LY != LYC)
|
|
{
|
|
LYC_INT = false;
|
|
STAT &= 0xFB;
|
|
}
|
|
|
|
if ((LY == LYC) && !STAT.Bit(2))
|
|
{
|
|
// set STAT coincidence FLAG and interrupt flag if it is enabled
|
|
STAT |= 0x04;
|
|
if (STAT.Bit(6)) { LYC_INT = true; }
|
|
}
|
|
}
|
|
|
|
if (cycle == 84)
|
|
{
|
|
STAT &= 0xFC;
|
|
STAT |= 0x03;
|
|
OAM_INT = false;
|
|
|
|
OAM_access_read = false;
|
|
OAM_access_write = false;
|
|
VRAM_access_read = false;
|
|
VRAM_access_write = false;
|
|
}
|
|
|
|
if (cycle == 256)
|
|
{
|
|
STAT &= 0xFC;
|
|
OAM_INT = false;
|
|
|
|
if (STAT.Bit(3)) { HBL_INT = true; }
|
|
|
|
OAM_access_read = true;
|
|
OAM_access_write = true;
|
|
VRAM_access_read = true;
|
|
VRAM_access_write = true;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
if (cycle < 80)
|
|
{
|
|
if (cycle == 2)
|
|
{
|
|
if (LY != 0)
|
|
{
|
|
HBL_INT = false;
|
|
if (STAT.Bit(5)) { OAM_INT = true; }
|
|
}
|
|
}
|
|
else if (cycle == 4)
|
|
{
|
|
// apparently, writes can make it to OAM one cycle longer then reads
|
|
OAM_access_write = false;
|
|
|
|
// here mode 2 will be set to true and interrupts fired if enabled
|
|
STAT &= 0xFC;
|
|
STAT |= 0x2;
|
|
|
|
if (LY == 0)
|
|
{
|
|
VBL_INT = false;
|
|
if (STAT.Bit(5)) { OAM_INT = true; }
|
|
}
|
|
}
|
|
|
|
// here OAM scanning is performed
|
|
OAM_scan(cycle);
|
|
}
|
|
else if ((cycle >= 80) && (LY < 144))
|
|
{
|
|
if (cycle >= 83)
|
|
{
|
|
if (cycle == 84)
|
|
{
|
|
STAT &= 0xFC;
|
|
STAT |= 0x03;
|
|
OAM_INT = false;
|
|
OAM_access_write = false;
|
|
VRAM_access_write = false;
|
|
|
|
// x-scroll is expected to be latched one cycle later
|
|
// this is fine since nothing has started in the rendering until the second cycle
|
|
// calculate the column number of the tile to start with
|
|
x_tile = (int)Math.Floor((float)(scroll_x) / 8);
|
|
render_offset = scroll_x % 8;
|
|
}
|
|
|
|
// render the screen and handle hblank
|
|
render(cycle - 83);
|
|
}
|
|
else if (cycle == 80)
|
|
{
|
|
OAM_access_read = false;
|
|
OAM_access_write = true;
|
|
VRAM_access_read = false;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (LY_inc == 0)
|
|
{
|
|
if (cycle == 10)
|
|
{
|
|
LYC_INT = false;
|
|
STAT &= 0xFB;
|
|
}
|
|
else if (cycle == 12)
|
|
{
|
|
// Special case of LY = LYC
|
|
if ((LY == LYC) && !STAT.Bit(2))
|
|
{
|
|
// set STAT coincidence FLAG and interrupt flag if it is enabled
|
|
STAT |= 0x04;
|
|
if (STAT.Bit(6)) { LYC_INT = true; }
|
|
}
|
|
}
|
|
}
|
|
|
|
// here LY=LYC will be asserted or cleared (but only if LY isnt 0 as that's a special case)
|
|
if ((cycle == 2) && (LY != 0))
|
|
{
|
|
if (LY_inc == 1)
|
|
{
|
|
LYC_INT = false;
|
|
STAT &= 0xFB;
|
|
}
|
|
}
|
|
else if ((cycle == 4) && (LY != 0))
|
|
{
|
|
if ((LY == LYC) && !STAT.Bit(2))
|
|
{
|
|
// set STAT coincidence FLAG and interrupt flag if it is enabled
|
|
STAT |= 0x04;
|
|
if (STAT.Bit(6)) { LYC_INT = true; }
|
|
}
|
|
}
|
|
|
|
cycle++;
|
|
}
|
|
else
|
|
{
|
|
STAT &= 0xFC;
|
|
|
|
VBL_INT = LYC_INT = HBL_INT = OAM_INT = false;
|
|
|
|
Core.in_vblank = true;
|
|
|
|
LCD_was_off = true;
|
|
|
|
LY = 0;
|
|
Core.cpu.LY = LY;
|
|
|
|
cycle = 0;
|
|
}
|
|
|
|
// assert the STAT IRQ line if the line went from zero to 1
|
|
stat_line = VBL_INT | LYC_INT | HBL_INT | OAM_INT;
|
|
|
|
if (stat_line && !stat_line_old)
|
|
{
|
|
if (Core.REG_FFFF.Bit(1)) { Core.cpu.FlagI = true; }
|
|
Core.REG_FF0F |= 0x02;
|
|
}
|
|
|
|
stat_line_old = stat_line;
|
|
|
|
// process latch delays
|
|
//latch_delay();
|
|
}
|
|
|
|
// might be needed, not sure yet
|
|
public override void latch_delay()
|
|
{
|
|
|
|
}
|
|
|
|
public override void render(int render_cycle)
|
|
{
|
|
// we are now in STAT mode 3
|
|
// NOTE: presumably the first necessary sprite is fetched at sprite evaulation
|
|
// i.e. just keeping track of the lowest x-value sprite
|
|
if (render_cycle == 0)
|
|
{
|
|
/*
|
|
OAM_access_read = false;
|
|
OAM_access_write = true;
|
|
VRAM_access_read = false;
|
|
*/
|
|
// window X is latched for the scanline, mid-line changes have no effect
|
|
window_x_latch = window_x;
|
|
|
|
OAM_scan_index = 0;
|
|
read_case = 0;
|
|
internal_cycle = 0;
|
|
pre_render = true;
|
|
pre_render_2 = true;
|
|
tile_inc = 0;
|
|
pixel_counter = -8;
|
|
sl_use_index = 0;
|
|
fetch_sprite = false;
|
|
going_to_fetch = false;
|
|
first_fetch = true;
|
|
consecutive_sprite = -render_offset + 8;
|
|
no_sprites = false;
|
|
evaled_sprites = 0;
|
|
window_pre_render = false;
|
|
window_latch = LCDC.Bit(5);
|
|
|
|
total_counter = 0;
|
|
|
|
// TODO: If Window is turned on midscanline what happens? When is this check done exactly?
|
|
if ((window_started && window_latch) || (window_is_reset && !window_latch && (LY >= window_y)))
|
|
{
|
|
window_y_tile_inc++;
|
|
if (window_y_tile_inc==8)
|
|
{
|
|
window_y_tile_inc = 0;
|
|
window_y_tile++;
|
|
window_y_tile %= 32;
|
|
}
|
|
}
|
|
window_started = false;
|
|
|
|
if (SL_sprites_index == 0) { no_sprites = true; }
|
|
// it is much easier to process sprites if we order them according to the rules of sprite priority first
|
|
if (!no_sprites) { reorder_and_assemble_sprites(); }
|
|
}
|
|
|
|
// before anything else, we have to check if windowing is in effect
|
|
if (window_latch && !window_started && (LY >= window_y) && (pixel_counter >= (window_x_latch - 7)) && (window_x_latch < 167))
|
|
{
|
|
/*
|
|
Console.Write(LY);
|
|
Console.Write(" ");
|
|
Console.Write(cycle);
|
|
Console.Write(" ");
|
|
Console.Write(window_y_tile_inc);
|
|
Console.Write(" ");
|
|
Console.Write(window_x_latch);
|
|
Console.Write(" ");
|
|
Console.WriteLine(pixel_counter);
|
|
*/
|
|
if (window_x_latch <= 7)
|
|
{
|
|
// if the window starts at zero, we still do the first access to the BG
|
|
// but then restart all over again at the window
|
|
read_case = 9;
|
|
}
|
|
else
|
|
{
|
|
// otherwise, just restart the whole process as if starting BG again
|
|
read_case = 4;
|
|
}
|
|
window_pre_render = true;
|
|
|
|
window_counter = 0;
|
|
render_counter = 0;
|
|
|
|
window_x_tile = (int)Math.Floor((float)(pixel_counter - (window_x_latch - 7)) / 8);
|
|
|
|
window_tile_inc = 0;
|
|
window_started = true;
|
|
window_is_reset = false;
|
|
}
|
|
|
|
if (!pre_render && !fetch_sprite)
|
|
{
|
|
// start shifting data into the LCD
|
|
if (render_counter >= (render_offset + 8))
|
|
{
|
|
pixel = tile_data_latch[0].Bit(7 - (render_counter % 8)) ? 1 : 0;
|
|
pixel |= tile_data_latch[1].Bit(7 - (render_counter % 8)) ? 2 : 0;
|
|
|
|
int ref_pixel = pixel;
|
|
if (LCDC.Bit(0))
|
|
{
|
|
pixel = (BGP >> (pixel * 2)) & 3;
|
|
}
|
|
else
|
|
{
|
|
pixel = 0;
|
|
}
|
|
|
|
// now we have the BG pixel, we next need the sprite pixel
|
|
if (!no_sprites)
|
|
{
|
|
bool have_sprite = false;
|
|
int s_pixel = 0;
|
|
int sprite_attr = 0;
|
|
|
|
if (sprite_present_list[pixel_counter] == 1)
|
|
{
|
|
have_sprite = true;
|
|
s_pixel = sprite_pixel_list[pixel_counter];
|
|
sprite_attr = sprite_attr_list[pixel_counter];
|
|
}
|
|
|
|
if (have_sprite)
|
|
{
|
|
bool use_sprite = false;
|
|
if (LCDC.Bit(1))
|
|
{
|
|
if (!sprite_attr.Bit(7))
|
|
{
|
|
use_sprite = true;
|
|
}
|
|
else if (ref_pixel == 0)
|
|
{
|
|
use_sprite = true;
|
|
}
|
|
|
|
if (!LCDC.Bit(0))
|
|
{
|
|
use_sprite = true;
|
|
}
|
|
}
|
|
|
|
if (use_sprite)
|
|
{
|
|
if (sprite_attr.Bit(4))
|
|
{
|
|
pixel = (obj_pal_1 >> (s_pixel * 2)) & 3;
|
|
}
|
|
else
|
|
{
|
|
pixel = (obj_pal_0 >> (s_pixel * 2)) & 3;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// based on sprite priority and pixel values, pick a final pixel color
|
|
Core._vidbuffer[LY * 160 + pixel_counter] = (int)Core.color_palette[pixel];
|
|
pixel_counter++;
|
|
|
|
if (pixel_counter == 160)
|
|
{
|
|
read_case = 8;
|
|
}
|
|
}
|
|
else if (pixel_counter < 0)
|
|
{
|
|
pixel_counter++;
|
|
}
|
|
render_counter++;
|
|
}
|
|
|
|
if (!fetch_sprite)
|
|
{
|
|
if (!pre_render_2)
|
|
{
|
|
// before we go on to read case 3, we need to know if we stall there or not
|
|
// Gekkio's tests show that if sprites are at position 0 or 1 (mod 8)
|
|
// then it takes an extra cycle (1 or 2 more t-states) to process them
|
|
// Also, on DMG only, this process only runs if sprites are on in the LCDC (on GBC it always runs)
|
|
if (!no_sprites && (pixel_counter < 160) && LCDC.Bit(1))
|
|
{
|
|
for (int i = 0; i < SL_sprites_index; i++)
|
|
{
|
|
if ((pixel_counter >= (SL_sprites[i * 4 + 1] - 8)) &&
|
|
(pixel_counter < (SL_sprites[i * 4 + 1])) &&
|
|
!evaled_sprites.Bit(i))
|
|
{
|
|
going_to_fetch = true;
|
|
fetch_sprite = true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
switch (read_case)
|
|
{
|
|
case 0: // read a background tile
|
|
if ((internal_cycle % 2) == 1)
|
|
{
|
|
// calculate the row number of the tiles to be fetched
|
|
y_tile = ((int)Math.Floor((float)(scroll_y + LY) / 8)) % 32;
|
|
|
|
temp_fetch = y_tile * 32 + (x_tile + tile_inc) % 32;
|
|
tile_byte = Core.VRAM[0x1800 + (LCDC.Bit(3) ? 1 : 0) * 0x400 + temp_fetch];
|
|
|
|
read_case = 1;
|
|
if (!pre_render)
|
|
{
|
|
tile_inc++;
|
|
}
|
|
}
|
|
break;
|
|
|
|
case 1: // read from tile graphics (0)
|
|
if ((internal_cycle % 2) == 1)
|
|
{
|
|
y_scroll_offset = (scroll_y + LY) % 8;
|
|
|
|
if (LCDC.Bit(4))
|
|
{
|
|
tile_data[0] = Core.VRAM[tile_byte * 16 + y_scroll_offset * 2];
|
|
}
|
|
else
|
|
{
|
|
// same as before except now tile byte represents a signed byte
|
|
if (tile_byte.Bit(7))
|
|
{
|
|
tile_byte -= 256;
|
|
}
|
|
tile_data[0] = Core.VRAM[0x1000 + tile_byte * 16 + y_scroll_offset * 2];
|
|
}
|
|
|
|
read_case = 2;
|
|
}
|
|
break;
|
|
|
|
case 2: // read from tile graphics (1)
|
|
if ((internal_cycle % 2) == 0)
|
|
{
|
|
pre_render_2 = false;
|
|
}
|
|
else
|
|
{
|
|
y_scroll_offset = (scroll_y + LY) % 8;
|
|
|
|
if (LCDC.Bit(4))
|
|
{
|
|
// if LCDC somehow changed between the two reads, make sure we have a positive number
|
|
if (tile_byte < 0)
|
|
{
|
|
tile_byte += 256;
|
|
}
|
|
|
|
tile_data[1] = Core.VRAM[tile_byte * 16 + y_scroll_offset * 2 + 1];
|
|
}
|
|
else
|
|
{
|
|
// same as before except now tile byte represents a signed byte
|
|
if (tile_byte.Bit(7) && tile_byte > 0)
|
|
{
|
|
tile_byte -= 256;
|
|
}
|
|
|
|
tile_data[1] = Core.VRAM[0x1000 + tile_byte * 16 + y_scroll_offset * 2 + 1];
|
|
}
|
|
|
|
if (pre_render)
|
|
{
|
|
// here we set up rendering
|
|
pre_render = false;
|
|
|
|
render_counter = 0;
|
|
latch_counter = 0;
|
|
read_case = 0;
|
|
}
|
|
else
|
|
{
|
|
read_case = 3;
|
|
}
|
|
}
|
|
break;
|
|
|
|
case 3: // read from sprite data
|
|
if ((internal_cycle % 2) == 1)
|
|
{
|
|
read_case = 0;
|
|
latch_new_data = true;
|
|
}
|
|
break;
|
|
|
|
case 4: // read from window data
|
|
if ((window_counter % 2) == 1)
|
|
{
|
|
temp_fetch = window_y_tile * 32 + (window_x_tile + window_tile_inc) % 32;
|
|
tile_byte = Core.VRAM[0x1800 + (LCDC.Bit(6) ? 1 : 0) * 0x400 + temp_fetch];
|
|
|
|
window_tile_inc++;
|
|
read_case = 5;
|
|
}
|
|
window_counter++;
|
|
break;
|
|
|
|
case 5: // read from tile graphics (for the window)
|
|
if ((window_counter % 2) == 1)
|
|
{
|
|
y_scroll_offset = (window_y_tile_inc) % 8;
|
|
|
|
if (LCDC.Bit(4))
|
|
{
|
|
|
|
tile_data[0] = Core.VRAM[tile_byte * 16 + y_scroll_offset * 2];
|
|
|
|
}
|
|
else
|
|
{
|
|
// same as before except now tile byte represents a signed byte
|
|
if (tile_byte.Bit(7))
|
|
{
|
|
tile_byte -= 256;
|
|
}
|
|
|
|
tile_data[0] = Core.VRAM[0x1000 + tile_byte * 16 + y_scroll_offset * 2];
|
|
}
|
|
|
|
read_case = 6;
|
|
}
|
|
window_counter++;
|
|
break;
|
|
|
|
case 6: // read from tile graphics (for the window)
|
|
if ((window_counter % 2) == 1)
|
|
{
|
|
y_scroll_offset = (window_y_tile_inc) % 8;
|
|
if (LCDC.Bit(4))
|
|
{
|
|
// if LCDC somehow changed between the two reads, make sure we have a positive number
|
|
if (tile_byte < 0)
|
|
{
|
|
tile_byte += 256;
|
|
}
|
|
|
|
tile_data[1] = Core.VRAM[tile_byte * 16 + y_scroll_offset * 2 + 1];
|
|
}
|
|
else
|
|
{
|
|
// same as before except now tile byte represents a signed byte
|
|
if (tile_byte.Bit(7) && tile_byte > 0)
|
|
{
|
|
tile_byte -= 256;
|
|
}
|
|
|
|
tile_data[1] = Core.VRAM[0x1000 + tile_byte * 16 + y_scroll_offset * 2 + 1];
|
|
}
|
|
|
|
if (window_pre_render)
|
|
{
|
|
// here we set up rendering
|
|
// unlike for the normal background case, there is no pre-render period for the window
|
|
// so start shifting in data to the screen right away
|
|
render_offset = 0;
|
|
render_counter = 8;
|
|
latch_counter = 0;
|
|
latch_new_data = true;
|
|
|
|
window_pre_render = false;
|
|
read_case = 4;
|
|
}
|
|
else
|
|
{
|
|
read_case = 7;
|
|
}
|
|
}
|
|
window_counter++;
|
|
break;
|
|
|
|
case 7: // read from sprite data
|
|
if ((window_counter % 2) == 1)
|
|
{
|
|
read_case = 4;
|
|
latch_new_data = true;
|
|
}
|
|
window_counter++;
|
|
break;
|
|
|
|
case 8: // done reading, we are now in phase 0
|
|
pre_render = true;
|
|
|
|
STAT &= 0xFC;
|
|
STAT |= 0x00;
|
|
|
|
if (STAT.Bit(3)) { HBL_INT = true; }
|
|
|
|
OAM_access_read = true;
|
|
OAM_access_write = true;
|
|
VRAM_access_read = true;
|
|
VRAM_access_write = true;
|
|
break;
|
|
|
|
case 9:
|
|
// this is a degenerate case for starting the window at 0
|
|
// kevtris' timing doc indicates an additional normal BG access
|
|
// but this information is thrown away, so it's faster to do this then constantly check
|
|
// for it in read case 0
|
|
read_case = 4;
|
|
break;
|
|
}
|
|
internal_cycle++;
|
|
|
|
if (latch_new_data)
|
|
{
|
|
latch_new_data = false;
|
|
tile_data_latch[0] = tile_data[0];
|
|
tile_data_latch[1] = tile_data[1];
|
|
}
|
|
}
|
|
|
|
// every in range sprite takes 6 cycles to process
|
|
// sprites located at x=0 still take 6 cycles to process even though they don't appear on screen
|
|
// sprites above x=168 do not take any cycles to process however
|
|
if (fetch_sprite)
|
|
{
|
|
if (going_to_fetch)
|
|
{
|
|
going_to_fetch = false;
|
|
|
|
last_eval = 0;
|
|
|
|
// at this time it is unknown what each cycle does, but we only need to accurately keep track of cycles
|
|
for (int i = 0; i < SL_sprites_index; i++)
|
|
{
|
|
if ((pixel_counter >= (SL_sprites[i * 4 + 1] - 8)) &&
|
|
(pixel_counter < (SL_sprites[i * 4 + 1])) &&
|
|
!evaled_sprites.Bit(i))
|
|
{
|
|
sprite_fetch_counter += 6;
|
|
evaled_sprites |= (1 << i);
|
|
last_eval = SL_sprites[i * 4 + 1];
|
|
}
|
|
}
|
|
|
|
// x scroll offsets the penalty table
|
|
// there is no penalty if the next sprites to be fetched are within the currentfetch block (8 pixels)
|
|
if (first_fetch || (last_eval >= consecutive_sprite))
|
|
{
|
|
if (((last_eval + render_offset) % 8) == 0) { sprite_fetch_counter += 5; }
|
|
else if (((last_eval + render_offset) % 8) == 1) { sprite_fetch_counter += 4; }
|
|
else if (((last_eval + render_offset) % 8) == 2) { sprite_fetch_counter += 3; }
|
|
else if (((last_eval + render_offset) % 8) == 3) { sprite_fetch_counter += 2; }
|
|
else if (((last_eval + render_offset) % 8) == 4) { sprite_fetch_counter += 1; }
|
|
else if (((last_eval + render_offset) % 8) == 5) { sprite_fetch_counter += 0; }
|
|
else if (((last_eval + render_offset) % 8) == 6) { sprite_fetch_counter += 0; }
|
|
else if (((last_eval + render_offset) % 8) == 7) { sprite_fetch_counter += 0; }
|
|
|
|
consecutive_sprite = (int)Math.Floor((double)(last_eval + render_offset) / 8) * 8 + 8 - render_offset;
|
|
|
|
// special case exists here for sprites at zero with non-zero x-scroll. Not sure exactly the reason for it.
|
|
if (last_eval == 0 && render_offset != 0)
|
|
{
|
|
sprite_fetch_counter += render_offset;
|
|
}
|
|
}
|
|
|
|
total_counter += sprite_fetch_counter;
|
|
|
|
first_fetch = false;
|
|
}
|
|
else
|
|
{
|
|
sprite_fetch_counter--;
|
|
if (sprite_fetch_counter == 0)
|
|
{
|
|
fetch_sprite = false;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
public override void process_sprite()
|
|
{
|
|
int y;
|
|
|
|
if (SL_sprites[sl_use_index * 4 + 3].Bit(6))
|
|
{
|
|
if (LCDC.Bit(2))
|
|
{
|
|
y = LY - (SL_sprites[sl_use_index * 4] - 16);
|
|
y = 15 - y;
|
|
sprite_sel[0] = Core.VRAM[(SL_sprites[sl_use_index * 4 + 2] & 0xFE) * 16 + y * 2];
|
|
sprite_sel[1] = Core.VRAM[(SL_sprites[sl_use_index * 4 + 2] & 0xFE) * 16 + y * 2 + 1];
|
|
}
|
|
else
|
|
{
|
|
y = LY - (SL_sprites[sl_use_index * 4] - 16);
|
|
y = 7 - y;
|
|
sprite_sel[0] = Core.VRAM[SL_sprites[sl_use_index * 4 + 2] * 16 + y * 2];
|
|
sprite_sel[1] = Core.VRAM[SL_sprites[sl_use_index * 4 + 2] * 16 + y * 2 + 1];
|
|
}
|
|
}
|
|
else
|
|
{
|
|
if (LCDC.Bit(2))
|
|
{
|
|
y = LY - (SL_sprites[sl_use_index * 4] - 16);
|
|
sprite_sel[0] = Core.VRAM[(SL_sprites[sl_use_index * 4 + 2] & 0xFE) * 16 + y * 2];
|
|
sprite_sel[1] = Core.VRAM[(SL_sprites[sl_use_index * 4 + 2] & 0xFE) * 16 + y * 2 + 1];
|
|
}
|
|
else
|
|
{
|
|
y = LY - (SL_sprites[sl_use_index * 4] - 16);
|
|
sprite_sel[0] = Core.VRAM[SL_sprites[sl_use_index * 4 + 2] * 16 + y * 2];
|
|
sprite_sel[1] = Core.VRAM[SL_sprites[sl_use_index * 4 + 2] * 16 + y * 2 + 1];
|
|
}
|
|
}
|
|
|
|
if (SL_sprites[sl_use_index * 4 + 3].Bit(5))
|
|
{
|
|
int b0, b1, b2, b3, b4, b5, b6, b7 = 0;
|
|
for (int i = 0; i < 2; i++)
|
|
{
|
|
b0 = (sprite_sel[i] & 0x01) << 7;
|
|
b1 = (sprite_sel[i] & 0x02) << 5;
|
|
b2 = (sprite_sel[i] & 0x04) << 3;
|
|
b3 = (sprite_sel[i] & 0x08) << 1;
|
|
b4 = (sprite_sel[i] & 0x10) >> 1;
|
|
b5 = (sprite_sel[i] & 0x20) >> 3;
|
|
b6 = (sprite_sel[i] & 0x40) >> 5;
|
|
b7 = (sprite_sel[i] & 0x80) >> 7;
|
|
|
|
sprite_sel[i] = (byte)(b0 | b1 | b2 | b3 | b4 | b5 | b6 | b7);
|
|
}
|
|
}
|
|
}
|
|
|
|
// normal DMA moves twice as fast in double speed mode on GBC
|
|
// So give it it's own function so we can seperate it from PPU tick
|
|
public override void DMA_tick()
|
|
{
|
|
// Note that DMA is halted when the CPU is halted
|
|
if (DMA_start && !Core.cpu.halted)
|
|
{
|
|
if (DMA_clock >= 4)
|
|
{
|
|
DMA_OAM_access = false;
|
|
if ((DMA_clock % 4) == 1)
|
|
{
|
|
// the cpu can't access memory during this time, but we still need the ppu to be able to.
|
|
DMA_start = false;
|
|
// Gekkio reports that A14 being high on DMA transfers always represent WRAM accesses
|
|
// So transfers nominally from higher memory areas are actually still from there (i.e. FF -> DF)
|
|
byte DMA_actual = DMA_addr;
|
|
if (DMA_addr > 0xDF) { DMA_actual &= 0xDF; }
|
|
DMA_byte = Core.ReadMemory((ushort)((DMA_actual << 8) + DMA_inc));
|
|
DMA_start = true;
|
|
}
|
|
else if ((DMA_clock % 4) == 3)
|
|
{
|
|
Core.OAM[DMA_inc] = DMA_byte;
|
|
|
|
if (DMA_inc < (0xA0 - 1)) { DMA_inc++; }
|
|
}
|
|
}
|
|
|
|
DMA_clock++;
|
|
|
|
if (DMA_clock == 648)
|
|
{
|
|
DMA_start = false;
|
|
DMA_OAM_access = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
// order sprites according to x coordinate
|
|
// note that for sprites of equal x coordinate, priority goes to first on the list
|
|
public override void reorder_and_assemble_sprites()
|
|
{
|
|
sprite_ordered_index = 0;
|
|
|
|
for (int i = 0; i < 256; i++)
|
|
{
|
|
for (int j = 0; j < SL_sprites_index; j++)
|
|
{
|
|
if (SL_sprites[j * 4 + 1] == i)
|
|
{
|
|
sl_use_index = j;
|
|
process_sprite();
|
|
SL_sprites_ordered[sprite_ordered_index * 4] = SL_sprites[j * 4 + 1];
|
|
SL_sprites_ordered[sprite_ordered_index * 4 + 1] = sprite_sel[0];
|
|
SL_sprites_ordered[sprite_ordered_index * 4 + 2] = sprite_sel[1];
|
|
SL_sprites_ordered[sprite_ordered_index * 4 + 3] = SL_sprites[j * 4 + 3];
|
|
sprite_ordered_index++;
|
|
}
|
|
}
|
|
}
|
|
|
|
bool have_pixel = false;
|
|
byte s_pixel = 0;
|
|
byte sprite_attr = 0;
|
|
|
|
for (int i = 0; i < 160; i++)
|
|
{
|
|
have_pixel = false;
|
|
for (int j = 0; j < SL_sprites_index; j++)
|
|
{
|
|
if ((i >= (SL_sprites_ordered[j * 4] - 8)) &&
|
|
(i < SL_sprites_ordered[j * 4]) &&
|
|
!have_pixel)
|
|
{
|
|
// we can use the current sprite, so pick out a pixel for it
|
|
int t_index = i - (SL_sprites_ordered[j * 4] - 8);
|
|
|
|
t_index = 7 - t_index;
|
|
|
|
sprite_data[0] = (byte)((SL_sprites_ordered[j * 4 + 1] >> t_index) & 1);
|
|
sprite_data[1] = (byte)(((SL_sprites_ordered[j * 4 + 2] >> t_index) & 1) << 1);
|
|
|
|
s_pixel = (byte)(sprite_data[0] + sprite_data[1]);
|
|
sprite_attr = (byte)SL_sprites_ordered[j * 4 + 3];
|
|
|
|
// pixel color of 0 is transparent, so if this is the case we dont have a pixel
|
|
if (s_pixel != 0)
|
|
{
|
|
have_pixel = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (have_pixel)
|
|
{
|
|
sprite_present_list[i] = 1;
|
|
sprite_pixel_list[i] = s_pixel;
|
|
sprite_attr_list[i] = sprite_attr;
|
|
}
|
|
else
|
|
{
|
|
sprite_present_list[i] = 0;
|
|
}
|
|
}
|
|
}
|
|
|
|
public override void OAM_scan(int OAM_cycle)
|
|
{
|
|
// we are now in STAT mode 2
|
|
// TODO: maybe stat mode 2 flags are set at cycle 0 on visible scanlines?
|
|
if (OAM_cycle == 0)
|
|
{
|
|
OAM_access_read = false;
|
|
|
|
OAM_scan_index = 0;
|
|
SL_sprites_index = 0;
|
|
write_sprite = 0;
|
|
}
|
|
|
|
// the gameboy has 80 cycles to scan through 40 sprites, picking out the first 10 it finds to draw
|
|
// the following is a guessed at implmenentation based on how NES does it, it's probably pretty close
|
|
if (OAM_cycle < 10)
|
|
{
|
|
// start by clearing the sprite table (probably just clears X on hardware, but let's be safe here.)
|
|
SL_sprites[OAM_cycle * 4] = 0;
|
|
SL_sprites[OAM_cycle * 4 + 1] = 0;
|
|
SL_sprites[OAM_cycle * 4 + 2] = 0;
|
|
SL_sprites[OAM_cycle * 4 + 3] = 0;
|
|
}
|
|
else
|
|
{
|
|
if (write_sprite == 0)
|
|
{
|
|
if (OAM_scan_index < 40)
|
|
{
|
|
ushort temp = DMA_OAM_access ? Core.OAM[OAM_scan_index * 4] : (ushort)0xFF;
|
|
// (sprite Y - 16) equals LY, we have a sprite
|
|
if ((temp - 16) <= LY &&
|
|
((temp - 16) + 8 + (LCDC.Bit(2) ? 8 : 0)) > LY)
|
|
{
|
|
// always pick the first 10 in range sprites
|
|
if (SL_sprites_index < 10)
|
|
{
|
|
SL_sprites[SL_sprites_index * 4] = temp;
|
|
|
|
write_sprite = 1;
|
|
}
|
|
else
|
|
{
|
|
// if we already have 10 sprites, there's nothing to do, increment the index
|
|
OAM_scan_index++;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
OAM_scan_index++;
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
ushort temp2 = DMA_OAM_access ? Core.OAM[OAM_scan_index * 4 + write_sprite] : (ushort)0xFF;
|
|
SL_sprites[SL_sprites_index * 4 + write_sprite] = temp2;
|
|
write_sprite++;
|
|
|
|
if (write_sprite == 4)
|
|
{
|
|
write_sprite = 0;
|
|
SL_sprites_index++;
|
|
OAM_scan_index++;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
public override void Reset()
|
|
{
|
|
LCDC = 0;
|
|
STAT = 0x80;
|
|
scroll_y = 0;
|
|
scroll_x = 0;
|
|
LY = 0;
|
|
LYC = 0;
|
|
DMA_addr = 0xFF;
|
|
BGP = 0xFF;
|
|
obj_pal_0 = 0xFF;
|
|
obj_pal_1 = 0xFF;
|
|
window_y = 0x0;
|
|
window_x = 0x0;
|
|
window_x_latch = 0xFF;
|
|
LY_inc = 1;
|
|
no_scan = false;
|
|
OAM_access_read = true;
|
|
VRAM_access_read = true;
|
|
OAM_access_write = true;
|
|
VRAM_access_write = true;
|
|
DMA_OAM_access = true;
|
|
|
|
cycle = 0;
|
|
LYC_INT = false;
|
|
HBL_INT = false;
|
|
VBL_INT = false;
|
|
OAM_INT = false;
|
|
|
|
stat_line = false;
|
|
stat_line_old = false;
|
|
|
|
window_counter = 0;
|
|
window_pre_render = false;
|
|
window_started = false;
|
|
window_is_reset = true;
|
|
window_tile_inc = 0;
|
|
window_y_tile = 0;
|
|
window_x_tile = 0;
|
|
window_y_tile_inc = 0;
|
|
}
|
|
}
|
|
}
|