more DSP unfucking: make Teakra directly use melonDS's NWRAM banks instead of trying to translate to a flat buffer, fixes bugs with the way the memory banks are ordered etc

This commit is contained in:
Arisotura 2022-10-14 00:02:54 +02:00
parent a8fba8cc34
commit cdd05c10b4
6 changed files with 87 additions and 108 deletions

View file

@ -991,23 +991,22 @@ void MapNWRAM_A(u32 num, u8 val)
MBK[1][mbkn] = MBK[0][mbkn];
// When we only update the mapping on the written MBK, we will
// have priority of the last witten MBK over the others
// However the hardware has a fixed order. Therefor
// have priority of the last written MBK over the others
// However the hardware has a fixed order. Therefore
// we need to iterate through them all in a fixed order and update
// the mapping, so the result is independend on the MBK write order
for (unsigned int part = 0; part < 4; part++)
// the mapping, so the result is independent on the MBK write order
for (int part = 0; part < 4; part++)
{
NWRAMMap_A[0][part] = NULL;
NWRAMMap_A[1][part] = NULL;
NWRAMMap_A[0][part] = nullptr;
NWRAMMap_A[1][part] = nullptr;
}
for (int part = 3; part >= 0; part--)
{
u8* ptr = &NWRAM_A[part << 16];
if ((MBK[0][0 + (part / 4)] >> ((part % 4) * 8)) & 0x80)
u8 mVal = (MBK[0][0 + (part >> 2)] >> ((part & 3) * 8)) & 0xFD;
if (mVal & 0x80)
{
u8 mVal = (MBK[0][0 + (part / 4)] >> ((part % 4) * 8)) & 0xfd;
NWRAMMap_A[mVal & 0x03][(mVal >> 2) & 0x3] = ptr;
}
}
@ -1039,30 +1038,24 @@ void MapNWRAM_B(u32 num, u8 val)
MBK[1][mbkn] = MBK[0][mbkn];
// When we only update the mapping on the written MBK, we will
// have priority of the last witten MBK over the others
// However the hardware has a fixed order. Therefor
// have priority of the last written MBK over the others
// However the hardware has a fixed order. Therefore
// we need to iterate through them all in a fixed order and update
// the mapping, so the result is independend on the MBK write order
for (unsigned int part = 0; part < 8; part++)
// the mapping, so the result is independent on the MBK write order
for (int part = 0; part < 8; part++)
{
NWRAMMap_B[0][part] = NULL;
NWRAMMap_B[1][part] = NULL;
NWRAMMap_B[2][part] = NULL;
NWRAMMap_B[0][part] = nullptr;
NWRAMMap_B[1][part] = nullptr;
NWRAMMap_B[2][part] = nullptr;
}
for (int part = 7; part >= 0; part--)
{
u8* ptr = &NWRAM_B[part << 15];
if (part == num)
u8 mVal = (MBK[0][1 + (part >> 2)] >> ((part & 3) * 8)) & 0xFF;
if (mVal & 0x80)
{
DSi_DSP::OnMBKCfg('B', num, oldval, val, ptr);
}
if ((MBK[0][1 + (part / 4)] >> ((part % 4) * 8)) & 0x80)
{
u8 mVal = (MBK[0][1 + (part / 4)] >> ((part % 4) * 8)) & 0xff;
if (mVal & 0x02) mVal &= 0xFE;
NWRAMMap_B[mVal & 0x03][(mVal >> 2) & 0x7] = ptr;
}
}
@ -1094,28 +1087,23 @@ void MapNWRAM_C(u32 num, u8 val)
MBK[1][mbkn] = MBK[0][mbkn];
// When we only update the mapping on the written MBK, we will
// have priority of the last witten MBK over the others
// However the hardware has a fixed order. Therefor
// have priority of the last written MBK over the others
// However the hardware has a fixed order. Therefore
// we need to iterate through them all in a fixed order and update
// the mapping, so the result is independend on the MBK write order
for (unsigned int part = 0; part < 8; part++)
// the mapping, so the result is independent on the MBK write order
for (int part = 0; part < 8; part++)
{
NWRAMMap_C[0][part] = NULL;
NWRAMMap_C[1][part] = NULL;
NWRAMMap_C[2][part] = NULL;
NWRAMMap_C[0][part] = nullptr;
NWRAMMap_C[1][part] = nullptr;
NWRAMMap_C[2][part] = nullptr;
}
for (int part = 7; part >= 0; part--)
{
u8* ptr = &NWRAM_C[part << 15];
if (part == num)
u8 mVal = MBK[0][3 + (part >> 2)] >> ((part & 3) * 8) & 0xFF;
if (mVal & 0x80)
{
DSi_DSP::OnMBKCfg('C', num, oldval, val, ptr);
}
if ((MBK[0][3 + (part / 4)] >> ((part % 4) * 8)) & 0x80)
{
u8 mVal = MBK[0][3 + (part / 4)] >> ((part % 4) *8) & 0xff;
if (mVal & 0x02) mVal &= 0xFE;
NWRAMMap_C[mVal & 0x03][(mVal >> 2) & 0x7] = ptr;
}

View file

@ -89,6 +89,36 @@ void IrqSem()
NDS::SetIRQ(0, NDS::IRQ_DSi_DSP);
}
u16 DSPRead16(u32 addr)
{
if (!(addr & 0x40000))
{
u8* ptr = DSi::NWRAMMap_B[2][(addr >> 15) & DSi::NWRAMMask[0][1]];
return ptr ? *(u16*)&ptr[addr & 0x7FFF] : 0;
}
else
{
u8* ptr = DSi::NWRAMMap_C[2][(addr >> 15) & DSi::NWRAMMask[0][2]];
return ptr ? *(u16*)&ptr[addr & 0x7FFF] : 0;
}
}
void DSPWrite16(u32 addr, u16 val)
{
// TODO: does the rule for overlapping NWRAM slots also apply to the DSP side?
if (!(addr & 0x40000))
{
u8* ptr = DSi::NWRAMMap_B[2][(addr >> 15) & DSi::NWRAMMask[0][1]];
if (ptr) *(u16*)&ptr[addr & 0x7FFF] = val;
}
else
{
u8* ptr = DSi::NWRAMMap_C[2][(addr >> 15) & DSi::NWRAMMask[0][2]];
if (ptr) *(u16*)&ptr[addr & 0x7FFF] = val;
}
}
void AudioCb(std::array<s16, 2> frame)
{
// TODO
@ -107,6 +137,11 @@ bool Init()
TeakraCore->SetSemaphoreHandler(IrqSem);
Teakra::SharedMemoryCallback smcb;
smcb.read16 = DSPRead16;
smcb.write16 = DSPWrite16;
TeakraCore->SetSharedMemoryCallback(smcb);
// these happen instantaneously and without too much regard for bus aribtration
// rules, so, this might have to be changed later on
Teakra::AHBMCallback cb;
@ -169,46 +204,6 @@ void SetRstLine(bool release)
DSPTimestamp = NDS::ARM9Timestamp; // only start now!
}
void OnMBKCfg(char bank, u32 slot, u8 oldcfg, u8 newcfg, u8* nwrambacking)
{
if (bank != 'B' && bank != 'C')
{
printf("WTF?? (DSP MBK recfg, nonsense bank '%c')\n", bank);
return;
}
bool olddsp = (oldcfg & 3) >= 2, // was mapped to the DSP
newdsp = (newcfg & 3) >= 2; // will be mapped to the DSP
// nothing changes
if (olddsp == newdsp)
return;
const u8* src;
u8* dst;
if (newdsp)
{
// need to map stuff to DSP memory (== Teakra-owned memory) from NWRAM
src = nwrambacking;
dst = &TeakraCore->GetDspMemory()[((newcfg >> 2) & 7) << 15];
if (bank == 'C') // C: DSP data (B: DSP code)
dst += DataMemoryOffset*2;
}
else //if (olddsp)
{
// it was mapped to the DSP, but now we have to copy it out, back to NWRAM
src = &TeakraCore->GetDspMemory()[((oldcfg >> 2) & 7) << 15];
dst = nwrambacking;
if (bank == 'C') // C: DSP data (B: DSP code)
src += DataMemoryOffset*2;
}
memcpy(dst, src, 1<<15); // 1 full slot
}
inline bool IsDSPCoreEnabled()
{
return (DSi::SCFG_Clock9 & (1<<1)) && SCFG_RST && (!(DSP_PCFG & (1<<0)));
@ -395,7 +390,7 @@ u16 PDataDMAReadMMIO()
u8 Read8(u32 addr)
{
if (!IsDSPIOEnabled()) return 0;
//if (!IsDSPIOEnabled()) return 0;
DSPCatchUp();
addr &= 0x3F; // mirroring wheee
@ -423,7 +418,7 @@ u8 Read8(u32 addr)
u16 Read16(u32 addr)
{
//printf("DSP READ16 %d %08X %08X\n", IsDSPCoreEnabled(), addr, NDS::GetPC(0));
if (!IsDSPIOEnabled()) return 0;
//if (!IsDSPIOEnabled()) return 0;
DSPCatchUp();
addr &= 0x3E; // mirroring wheee
@ -472,7 +467,7 @@ u32 Read32(u32 addr)
void Write8(u32 addr, u8 val)
{
if (!IsDSPIOEnabled()) return;
//if (!IsDSPIOEnabled()) return;
DSPCatchUp();
addr &= 0x3F;
@ -493,8 +488,8 @@ void Write8(u32 addr, u8 val)
}
void Write16(u32 addr, u16 val)
{
//printf("DSP WRITE16 %d %08X %08X %08X\n", IsDSPCoreEnabled(), addr, val, NDS::GetPC(0));
if (!IsDSPIOEnabled()) return;
printf("DSP WRITE16 %d %08X %08X %08X\n", IsDSPCoreEnabled(), addr, val, NDS::GetPC(0));
//if (!IsDSPIOEnabled()) return;
DSPCatchUp();
addr &= 0x3E;

View file

@ -53,9 +53,6 @@ void DSPCatchUpU32(u32 _);
bool IsRstReleased();
void SetRstLine(bool release);
// apply NWRAM settings
void OnMBKCfg(char bank, u32 slot, u8 oldcfg, u8 newcfg, u8* nwrambacking);
// DSP_* regs (0x040043xx) (NOTE: checks SCFG_EXT)
u8 Read8(u32 addr);
void Write8(u32 addr, u8 val);

View file

@ -7,6 +7,11 @@
namespace Teakra {
struct SharedMemoryCallback {
std::function<std::uint16_t(std::uint32_t address)> read16;
std::function<void(std::uint32_t address, std::uint16_t value)> write16;
};
struct AHBMCallback {
std::function<std::uint8_t(std::uint32_t address)> read8;
std::function<void(std::uint32_t address, std::uint8_t value)> write8;
@ -25,9 +30,6 @@ public:
void Reset();
std::array<std::uint8_t, 0x80000>& GetDspMemory();
const std::array<std::uint8_t, 0x80000>& GetDspMemory() const;
// APBP Data
bool SendDataIsEmpty(std::uint8_t index) const;
void SendData(std::uint8_t index, std::uint16_t value);
@ -70,6 +72,7 @@ public:
// core
void Run(unsigned cycle);
void SetSharedMemoryCallback(const SharedMemoryCallback& callback);
void SetAHBMCallback(const AHBMCallback& callback);
void SetAudioCallback(std::function<void(std::array<std::int16_t, 2>)> callback);

View file

@ -5,19 +5,21 @@
namespace Teakra {
struct SharedMemory {
std::array<u8, 0x80000> raw{};
u16 ReadWord(u32 word_address) const {
u32 byte_address = word_address * 2;
u8 low = raw[byte_address];
u8 high = raw[byte_address + 1];
return low | ((u16)high << 8);
return read_external16(word_address << 1);
}
void WriteWord(u32 word_address, u16 value) {
u8 low = value & 0xFF;
u8 high = value >> 8;
u32 byte_address = word_address * 2;
raw[byte_address] = low;
raw[byte_address + 1] = high;
write_external16(word_address << 1, value);
}
void SetExternalMemoryCallback(
std::function<u16(u32)> read16, std::function<void(u32, u16)> write16) {
read_external16 = std::move(read16);
write_external16 = std::move(write16);
}
std::function<u16(u32)> read_external16;
std::function<void(u32, u16)> write_external16;
};
} // namespace Teakra

View file

@ -50,7 +50,6 @@ struct Teakra::Impl {
}
void Reset() {
//shared_memory.raw.fill(0); // BAD!!!!
miu.Reset();
apbp_from_cpu.Reset();
apbp_from_dsp.Reset();
@ -71,14 +70,6 @@ void Teakra::Reset() {
impl->Reset();
}
std::array<std::uint8_t, 0x80000>& Teakra::GetDspMemory() {
return impl->shared_memory.raw;
}
const std::array<std::uint8_t, 0x80000>& Teakra::GetDspMemory() const {
return impl->shared_memory.raw;
}
void Teakra::Run(unsigned cycle) {
impl->processor.Run(cycle);
}
@ -117,6 +108,9 @@ void Teakra::ClearSemaphore(std::uint16_t value) {
void Teakra::MaskSemaphore(std::uint16_t value) {
impl->apbp_from_dsp.MaskSemaphore(value);
}
void Teakra::SetSharedMemoryCallback(const SharedMemoryCallback& callback) {
impl->shared_memory.SetExternalMemoryCallback(callback.read16, callback.write16);
}
void Teakra::SetAHBMCallback(const AHBMCallback& callback) {
impl->ahbm.SetExternalMemoryCallback(callback.read8, callback.write8,
callback.read16, callback.write16,