Sprinter/mame
2
0
Fork 0
mirror of https://github.com/holub/mame synced 2025-04-19 23:12:11 +03:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

(MESS) gba: vblank at line 160, don't do HIRQ or HDMA during Vblank. This fixes raster effects and glitches in dozens if not hundreds of games. [R. Belmont, endrift]

Browse Source
This commit is contained in:
arbee 2015-03-25 22:55:51 -04:00
parent d6e2bdbb6f
commit 8ba6abb1c0
1 changed files with 39 additions and 59 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

98
src/mess/drivers/gba.c
View File

@ -161,7 +161,7 @@ void gba_state::dma_exec(FPTR ch)
else
{
// if (dst >= 0x6000000 && dst <= 0x6017fff)
// printf("DMA exec: ch %d from %08x to %08x, mode %04x, count %04x (PC %x) (%s)\n", (int)ch, src, dst, ctrl, cnt, activecpu_get_pc(), ((ctrl>>10) & 1) ? "32" : "16");
// printf("DMA exec: ch %d from %08x to %08x, mode %04x, count %04x (%s)\n", (int)ch, src, dst, ctrl, cnt, ((ctrl>>10) & 1) ? "32" : "16");
}
for (int i = 0; i < cnt; i++)
@ -1559,7 +1559,7 @@ WRITE32_MEMBER(gba_state::gba_io_w)
ch = offset / 3;
// printf("%08x: DMA(%d): %x to reg %d (mask %08x)\n", activecpu_get_pc(), ch, data, offset%3, ~mem_mask);
// printf("%08x: DMA(%d): %x to reg %d (mask %08x)\n", space.device().safe_pc(), ch, data, offset%3, ~mem_mask);
if (((offset % 3) == 2) && ((~mem_mask & 0xffff0000) == 0))
{
@ -1961,24 +1961,26 @@ TIMER_CALLBACK_MEMBER(gba_state::perform_hbl)
if (scanline < 160)
{
draw_scanline(scanline);
}
m_DISPSTAT |= DISPSTAT_HBL;
if ((m_DISPSTAT & DISPSTAT_HBL_IRQ_EN ) != 0)
{
request_irq(INT_HBL);
}
for (ch = 0; ch < 4; ch++)
{
ctrl = m_dma_regs[(ch*3)+2]>>16;
// HBL-triggered DMA?
if ((ctrl & 0x8000) && ((ctrl & 0x3000) == 0x2000))
if ((m_DISPSTAT & DISPSTAT_HBL_IRQ_EN ) != 0)
{
dma_exec(ch);
request_irq(INT_HBL);
}
for (ch = 0; ch < 4; ch++)
{
ctrl = m_dma_regs[(ch*3)+2]>>16;
// HBL-triggered DMA?
if ((ctrl & 0x8000) && ((ctrl & 0x3000) == 0x2000))
{
dma_exec(ch);
}
}
}
m_DISPSTAT |= DISPSTAT_HBL;
m_hbl_timer->adjust(attotime::never);
}
@ -1995,6 +1997,28 @@ TIMER_CALLBACK_MEMBER(gba_state::perform_scan)
if (scanline >= 160 && scanline < 227)
{
m_DISPSTAT |= DISPSTAT_VBL;
// VBL IRQ and DMA on line 160
if (scanline == 160)
{
int ch, ctrl;
if (m_DISPSTAT & DISPSTAT_VBL_IRQ_EN)
{
request_irq(INT_VBL);
}
for (ch = 0; ch < 4; ch++)
{
ctrl = m_dma_regs[(ch*3)+2]>>16;
// VBL-triggered DMA?
if ((ctrl & 0x8000) && ((ctrl & 0x3000) == 0x1000))
{
dma_exec(ch);
}
}
}
}
else
{
@ -2011,50 +2035,6 @@ TIMER_CALLBACK_MEMBER(gba_state::perform_scan)
}
}
// exiting VBL, handle interrupts and DMA triggers
if (scanline == 224)
{
// FIXME: some games are very picky with this trigger!
// * Mario & Luigi SuperStar Saga loses pieces of gfx for 225-227.
// * Driver 2 does not work with values > 217.
// * Prince of Persia Sands of Time, Rayman Hoodlum's Revenge, Rayman 3 breaks for large
// values (say > 200, but exact threshold varies).
// * Scooby-Doo Unmasked and Mystery Mayhem have problems with large values (missing dialogue
// text).
// * Nicktoons Racign does not start with 227; and it resets before going to the race with
// values > 206.
// * Phil of Future does not start for values > 221.
// * Sabrina Teenage Witch does not even reach the Ubi Soft logo if we use the VBL exit value
// 227; it does not display title screen graphics when using 225-226; the intro is broken
// with anything between 207-224.
// * Anstoss Action and Ueki no Housoku have broken graphics for values > 223.
// However, taking smaller values breaks raster effects in a LOT of games (e.g. Castlevania
// series, Final Fantasy series, Tales of Phantasia, Banjo Pilot, NES 'collections' by Hudson,
// Jaleco and Technos, plus tons of racing games, which show garbage in the lower half of the
// screen with smaller values).
// Already choosing 224 instead of 227 makes some glitches to appear in the bottom scanlines.
// Other test cases are EA Sport games (like FIFA or Madden) which have various degrees of
// glitchness depending on the value used here.
// More work on IRQs is definitely necessary!
int ch, ctrl;
if (m_DISPSTAT & DISPSTAT_VBL_IRQ_EN)
{
request_irq(INT_VBL);
}
for (ch = 0; ch < 4; ch++)
{
ctrl = m_dma_regs[(ch*3)+2]>>16;
// VBL-triggered DMA?
if ((ctrl & 0x8000) && ((ctrl & 0x3000) == 0x1000))
{
dma_exec(ch);
}
}
}
m_hbl_timer->adjust(machine().first_screen()->time_until_pos(scanline, 240));
m_scan_timer->adjust(machine().first_screen()->time_until_pos(( scanline + 1 ) % 228, 0));
}