mirror of https://github.com/PCSX2/pcsx2.git
dmac: implement memcpy wrapping for SPR
Partially fix the dma test
This commit is contained in:
parent
e983be4060
commit
b2fbbddfa0
|
@ -66,6 +66,36 @@ static void TestClearVUs(u32 madr, u32 qwc, bool isWrite)
|
|||
}
|
||||
}
|
||||
|
||||
static void memcpy_to_spr(u32 dst, u8* src, size_t size)
|
||||
{
|
||||
dst &= _16kb - 1;
|
||||
|
||||
if (dst + size >= _16kb) {
|
||||
size_t end = _16kb - dst;
|
||||
memcpy(&psSu128(dst), src, end);
|
||||
|
||||
src += end;
|
||||
memcpy(&psSu128(0) , src, size - end);
|
||||
} else {
|
||||
memcpy(&psSu128(dst), src, size);
|
||||
}
|
||||
}
|
||||
|
||||
static void memcpy_from_spr(u8* dst, u32 src, size_t size)
|
||||
{
|
||||
src &= _16kb - 1;
|
||||
|
||||
if (src + size >= _16kb) {
|
||||
size_t end = _16kb - src;
|
||||
memcpy(dst, &psSu128(src), end);
|
||||
|
||||
dst += end;
|
||||
memcpy(dst, &psSu128(0) , size - end);
|
||||
} else {
|
||||
memcpy(dst, &psSu128(src), size);
|
||||
}
|
||||
}
|
||||
|
||||
int _SPR0chain()
|
||||
{
|
||||
tDMA_TAG *pMem;
|
||||
|
@ -98,8 +128,7 @@ int _SPR0chain()
|
|||
//Taking an arbitary small value for games which like to check the QWC/MADR instead of STR, so get most of
|
||||
//the cycle delay out of the way before the end.
|
||||
partialqwc = spr0ch.qwc;
|
||||
pxAssertMsg((spr0ch.sadr + partialqwc*16) < 0x4000, "TODO: Copy must be wrapped");
|
||||
memcpy(pMem, &psSu128(spr0ch.sadr), partialqwc*16);
|
||||
memcpy_from_spr((u8*)pMem, spr0ch.sadr, partialqwc*16);
|
||||
|
||||
// clear VU mem also!
|
||||
TestClearVUs(spr0ch.madr, partialqwc, true);
|
||||
|
@ -154,8 +183,7 @@ void _SPR0interleave()
|
|||
case MFD_RESERVED:
|
||||
// clear VU mem also!
|
||||
TestClearVUs(spr0ch.madr, spr0ch.qwc, true);
|
||||
pxAssertMsg((spr0ch.sadr + spr0ch.qwc*16) < 0x4000, "TODO: Copy must be wrapped");
|
||||
memcpy(pMem, &psSu128(spr0ch.sadr), spr0ch.qwc*16);
|
||||
memcpy_from_spr((u8*)pMem, spr0ch.sadr, spr0ch.qwc*16);
|
||||
break;
|
||||
}
|
||||
spr0ch.sadr += spr0ch.qwc * 16;
|
||||
|
@ -328,8 +356,7 @@ __fi static void SPR1transfer(const void* data, int qwc)
|
|||
TestClearVUs(spr1ch.madr, spr1ch.qwc, false);
|
||||
}
|
||||
|
||||
pxAssertMsg((spr1ch.sadr + qwc*16) < 0x4000, "TODO: Copy must be wrapped");
|
||||
memcpy(&psSu128(spr1ch.sadr), data, qwc*16);
|
||||
memcpy_to_spr(spr1ch.sadr, (u8*)data, qwc*16);
|
||||
spr1ch.sadr += qwc * 16;
|
||||
spr1ch.sadr &= 0x3FFF; // Limited to 16K
|
||||
}
|
||||
|
@ -389,8 +416,7 @@ void _SPR1interleave()
|
|||
spr1ch.qwc = std::min(tqwc, qwc);
|
||||
qwc -= spr1ch.qwc;
|
||||
pMem = SPRdmaGetAddr(spr1ch.madr, false);
|
||||
pxAssertMsg((spr1ch.sadr + spr1ch.qwc*16) < 0x4000, "TODO: Copy must be wrapped");
|
||||
memcpy(&psSu128(spr1ch.sadr), pMem, spr1ch.qwc*16);
|
||||
memcpy_to_spr(spr1ch.sadr, (u8*)pMem, spr1ch.qwc*16);
|
||||
spr1ch.sadr += spr1ch.qwc * 16;
|
||||
spr1ch.sadr &= 0x3FFF; // Limited to 16K
|
||||
spr1ch.madr += (sqwc + spr1ch.qwc) * 16;
|
||||
|
|
Loading…
Reference in New Issue