ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

COP0: Correctly update performance counter cycles

This commit is contained in:
refractionpcsx2 2023-11-08 22:34:15 +00:00
parent 8e9fd91cee
commit a3eb1655c6
2 changed files with 369 additions and 291 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

154
pcsx2/COP0.cpp
View File

@ -21,7 +21,8 @@
// Updates the CPU's mode of operation (either, Kernel, Supervisor, or User modes).
// Currently the different modes are not implemented.
// Given this function is called so much, it's commented out for now. (rama)
__ri void cpuUpdateOperationMode() {
__ri void cpuUpdateOperationMode()
{
//u32 value = cpuRegs.CP0.n.Status.val;
@ -33,15 +34,15 @@ __ri void cpuUpdateOperationMode() {
//}
}
void WriteCP0Status(u32 value) {
//DMA_LOG("COP0 Status write = 0x%08x", value);
void WriteCP0Status(u32 value)
{
COP0_UpdatePCCR();
cpuRegs.CP0.n.Status.val = value;
cpuSetNextEventDelta(4);
}
void WriteCP0Config(u32 value) {
void WriteCP0Config(u32 value)
{
// Protect the read-only ICacheSize (IC) and DataCacheSize (DC) bits
cpuRegs.CP0.n.Config = value & ~0xFC0;
cpuRegs.CP0.n.Config |= 0x440;
@ -140,13 +141,13 @@ __fi void COP0_UpdatePCCR()
if (PERF_ShouldCountEvent(cpuRegs.PERF.n.pccr.b.Event0))
{
u32 incr = cpuRegs.cycle - cpuRegs.lastPERFCycle[0];
if( incr == 0 ) incr++;
if (incr == 0)
incr++;
// use prev/XOR method for one-time exceptions (but likely less correct)
//u32 prev = cpuRegs.PERF.n.pcr0;
cpuRegs.PERF.n.pcr0 += incr;
cpuRegs.lastPERFCycle[0] = cpuRegs.cycle;
//DevCon.Warning("PCR VAL %x", cpuRegs.PERF.n.pccr.val);
//prev ^= (1UL<<31); // XOR is fun!
//if( (prev & cpuRegs.PERF.n.pcr0) & (1UL<<31) )
if ((cpuRegs.PERF.n.pcr0 & 0x80000000))
@ -191,10 +192,10 @@ __fi void COP0_UpdatePCCR()
if (PERF_ShouldCountEvent(cpuRegs.PERF.n.pccr.b.Event1))
{
u32 incr = cpuRegs.cycle - cpuRegs.lastPERFCycle[1];
if( incr == 0 ) incr++;
if (incr == 0)
incr++;
cpuRegs.PERF.n.pcr1 += incr;
cpuRegs.lastPERFCycle[1] = cpuRegs.cycle;
if ((cpuRegs.PERF.n.pcr1 & 0x80000000))
{
@ -229,6 +230,8 @@ __fi void COP0_UpdatePCCR()
}
}
}
cpuRegs.lastPERFCycle[0] = cpuRegs.cycle;
cpuRegs.lastPERFCycle[1] = cpuRegs.cycle;
}
//////////////////////////////////////////////////////////////////////////////////////////
@ -249,27 +252,34 @@ void MapTLB(const tlbs& t, int i)
vtlb_VMapBuffer(t.VPN2, eeMem->Scratch, Ps2MemSize::Scratch);
}
if (t.VPN2 == 0x70000000) return; //uh uhh right ...
if (t.EntryLo0 & 0x2) {
if (t.VPN2 == 0x70000000)
return; //uh uhh right ...
if (t.EntryLo0 & 0x2)
{
mask = ((~t.Mask) << 1) & 0xfffff;
saddr = t.VPN2 >> 12;
eaddr = saddr + t.Mask + 1;
for (addr=saddr; addr<eaddr; addr++) {
if ((addr & mask) == ((t.VPN2 >> 12) & mask)) { //match
for (addr = saddr; addr < eaddr; addr++)
{
if ((addr & mask) == ((t.VPN2 >> 12) & mask))
{ //match
memSetPageAddr(addr << 12, t.PFN0 + ((addr - saddr) << 12));
Cpu->Clear(addr << 12, 0x400);
}
}
}
if (t.EntryLo1 & 0x2) {
if (t.EntryLo1 & 0x2)
{
mask = ((~t.Mask) << 1) & 0xfffff;
saddr = (t.VPN2 >> 12) + t.Mask + 1;
eaddr = saddr + t.Mask + 1;
for (addr=saddr; addr<eaddr; addr++) {
if ((addr & mask) == ((t.VPN2 >> 12) & mask)) { //match
for (addr = saddr; addr < eaddr; addr++)
{
if ((addr & mask) == ((t.VPN2 >> 12) & mask))
{ //match
memSetPageAddr(addr << 12, t.PFN1 + ((addr - saddr) << 12));
Cpu->Clear(addr << 12, 0x400);
}
@ -295,21 +305,26 @@ void UnmapTLB(const tlbs& t, int i)
saddr = t.VPN2 >> 12;
eaddr = saddr + t.Mask + 1;
// Console.WriteLn("Clear TLB: %08x ~ %08x",saddr,eaddr-1);
for (addr=saddr; addr<eaddr; addr++) {
if ((addr & mask) == ((t.VPN2 >> 12) & mask)) { //match
for (addr = saddr; addr < eaddr; addr++)
{
if ((addr & mask) == ((t.VPN2 >> 12) & mask))
{ //match
memClearPageAddr(addr << 12);
Cpu->Clear(addr << 12, 0x400);
}
}
}
if (t.EntryLo1 & 0x2) {
if (t.EntryLo1 & 0x2)
{
mask = ((~t.Mask) << 1) & 0xfffff;
saddr = (t.VPN2 >> 12) + t.Mask + 1;
eaddr = saddr + t.Mask + 1;
// Console.WriteLn("Clear TLB: %08x ~ %08x",saddr,eaddr-1);
for (addr=saddr; addr<eaddr; addr++) {
if ((addr & mask) == ((t.VPN2 >> 12) & mask)) { //match
for (addr = saddr; addr < eaddr; addr++)
{
if ((addr & mask) == ((t.VPN2 >> 12) & mask))
{ //match
memClearPageAddr(addr << 12);
Cpu->Clear(addr << 12, 0x400);
}
@ -341,7 +356,8 @@ namespace Interpreter {
namespace OpcodeImpl {
namespace COP0 {
void TLBR() {
void TLBR()
{
COP0_LOG("COP0_TLBR %d:%x,%x,%x,%x",
cpuRegs.CP0.n.Index, cpuRegs.CP0.n.PageMask, cpuRegs.CP0.n.EntryHi,
cpuRegs.CP0.n.EntryLo0, cpuRegs.CP0.n.EntryLo1);
@ -354,7 +370,8 @@ void TLBR() {
cpuRegs.CP0.n.EntryLo1 = (tlb[i].EntryLo1 & ~1) | ((tlb[i].EntryHi >> 12) & 1);
}
void TLBWI() {
void TLBWI()
{
int j = cpuRegs.CP0.n.Index & 0x3f;
//if (j > 48) return;
@ -371,7 +388,8 @@ void TLBWI() {
WriteTLB(j);
}
void TLBWR() {
void TLBWR()
{
int j = cpuRegs.CP0.n.Random & 0x3f;
//if (j > 48) return;
@ -390,11 +408,14 @@ DevCon.Warning("COP0_TLBWR %d:%x,%x,%x,%x\n",
WriteTLB(j);
}
void TLBP() {
void TLBP()
{
int i;
union {
struct {
union
{
struct
{
u32 VPN2 : 19;
u32 VPN2X : 2;
u32 G : 3;
@ -406,20 +427,23 @@ void TLBP() {
EntryHi32.u = cpuRegs.CP0.n.EntryHi;
cpuRegs.CP0.n.Index = 0xFFFFFFFF;
for(i=0;i<48;i++){
if (tlb[i].VPN2 == ((~tlb[i].Mask) & (EntryHi32.s.VPN2))
&& ((tlb[i].G&1) || ((tlb[i].ASID & 0xff) == EntryHi32.s.ASID))) {
for (i = 0; i < 48; i++)
{
if (tlb[i].VPN2 == ((~tlb[i].Mask) & (EntryHi32.s.VPN2)) && ((tlb[i].G & 1) || ((tlb[i].ASID & 0xff) == EntryHi32.s.ASID)))
{
cpuRegs.CP0.n.Index = i;
break;
}
}
if(cpuRegs.CP0.n.Index == 0xFFFFFFFF) cpuRegs.CP0.n.Index = 0x80000000;
if (cpuRegs.CP0.n.Index == 0xFFFFFFFF)
cpuRegs.CP0.n.Index = 0x80000000;
}
void MFC0()
{
// Note on _Rd_ Condition 9: CP0.Count should be updated even if _Rt_ is 0.
if ((_Rd_ != 9) && !_Rt_ ) return;
if ((_Rd_ != 9) && !_Rt_)
return;
//if(bExecBIOS == FALSE && _Rd_ == 25) Console.WriteLn("MFC0 _Rd_ %x = %x", _Rd_, cpuRegs.CP0.r[_Rd_]);
switch (_Rd_)
@ -454,10 +478,12 @@ void MFC0()
case 9:
{
u32 incr = cpuRegs.cycle - cpuRegs.lastCOP0Cycle;
if( incr == 0 ) incr++;
if (incr == 0)
incr++;
cpuRegs.CP0.n.Count += incr;
cpuRegs.lastCOP0Cycle = cpuRegs.cycle;
if( !_Rt_ ) break;
if (!_Rt_)
break;
}
[[fallthrough]];
@ -518,36 +544,43 @@ void MTC0()
}
}
int CPCOND0() {
int CPCOND0()
{
return (((dmacRegs.stat.CIS | ~dmacRegs.pcr.CPC) & 0x3FF) == 0x3ff);
}
//#define CPCOND0 1
void BC0F() {
if (CPCOND0() == 0) intDoBranch(_BranchTarget_);
void BC0F()
{
if (CPCOND0() == 0)
intDoBranch(_BranchTarget_);
}
void BC0T() {
if (CPCOND0() == 1) intDoBranch(_BranchTarget_);
void BC0T()
{
if (CPCOND0() == 1)
intDoBranch(_BranchTarget_);
}
void BC0FL() {
void BC0FL()
{
if (CPCOND0() == 0)
intDoBranch(_BranchTarget_);
else
cpuRegs.pc += 4;
}
void BC0TL() {
void BC0TL()
{
if (CPCOND0() == 1)
intDoBranch(_BranchTarget_);
else
cpuRegs.pc += 4;
}
void ERET() {
void ERET()
{
#ifdef ENABLE_VTUNE
// Allow to stop vtune in a predictable way to compare runs
// Of course, the limit will depend on the game.
@ -557,19 +590,25 @@ void ERET() {
// quick_exit vs exit: quick_exit won't call static storage destructor (OS will manage). It helps
// avoiding the race condition between threads destruction.
if (vtune > 30 * million) {
if (vtune > 30 * million)
{
Console.WriteLn("VTUNE: quick_exit");
std::quick_exit(EXIT_SUCCESS);
} else if (!(vtune % million)) {
}
else if (!(vtune % million))
{
Console.WriteLn("VTUNE: ERET was called %uM times", vtune / million);
}
#endif
if (cpuRegs.CP0.n.Status.b.ERL) {
if (cpuRegs.CP0.n.Status.b.ERL)
{
cpuRegs.pc = cpuRegs.CP0.n.ErrorEPC;
cpuRegs.CP0.n.Status.b.ERL = 0;
} else {
}
else
{
cpuRegs.pc = cpuRegs.CP0.n.EPC;
cpuRegs.CP0.n.Status.b.EXL = 0;
}
@ -578,22 +617,29 @@ void ERET() {
intSetBranch();
}
void DI() {
void DI()
{
if (cpuRegs.CP0.n.Status.b._EDI || cpuRegs.CP0.n.Status.b.EXL ||
cpuRegs.CP0.n.Status.b.ERL || (cpuRegs.CP0.n.Status.b.KSU == 0)) {
cpuRegs.CP0.n.Status.b.ERL || (cpuRegs.CP0.n.Status.b.KSU == 0))
{
cpuRegs.CP0.n.Status.b.EIE = 0;
// IRQs are disabled so no need to do a cpu exception/event test...
//cpuSetNextEventDelta();
}
}
void EI() {
void EI()
{
if (cpuRegs.CP0.n.Status.b._EDI || cpuRegs.CP0.n.Status.b.EXL ||
cpuRegs.CP0.n.Status.b.ERL || (cpuRegs.CP0.n.Status.b.KSU == 0)) {
cpuRegs.CP0.n.Status.b.ERL || (cpuRegs.CP0.n.Status.b.KSU == 0))
{
cpuRegs.CP0.n.Status.b.EIE = 1;
// schedule an event test, which will check for and raise pending IRQs.
cpuSetNextEventDelta(4);
}
}
} } } } // end namespace R5900::Interpreter::OpcodeImpl
} // namespace COP0
} // namespace OpcodeImpl
} // namespace Interpreter
} // namespace R5900

46
pcsx2/x86/iCOP0.cpp
View File

@ -150,7 +150,9 @@ void recMFC0()
if (_Rd_ == 9)
{
// This case needs to be handled even if the write-back is ignored (_Rt_ == 0 )
xMOV(ecx, ptr[&cpuRegs.cycle]);
xMOV(ecx, ptr32[&cpuRegs.cycle]);
xADD(ecx, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], ecx); // update cycles
xMOV(eax, ecx);
xSUB(eax, ptr[&cpuRegs.lastCOP0Cycle]);
u8* skipInc = JNZ8(0);
@ -180,6 +182,9 @@ void recMFC0()
else if (0 == (_Imm_ & 2)) // MFPC 0, only LSB of register matters
{
iFlushCall(FLUSH_INTERPRETER);
xMOV(eax, ptr32[&cpuRegs.cycle]);
xADD(eax, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], eax); // update cycles
xFastCall((void*)COP0_UpdatePCCR);
const int regt = _allocX86reg(X86TYPE_GPR, _Rt_, MODE_WRITE);
@ -188,6 +193,9 @@ void recMFC0()
else // MFPC 1
{
iFlushCall(FLUSH_INTERPRETER);
xMOV(eax, ptr32[&cpuRegs.cycle]);
xADD(eax, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], eax); // update cycles
xFastCall((void*)COP0_UpdatePCCR);
const int regt = _allocX86reg(X86TYPE_GPR, _Rt_, MODE_WRITE);
@ -214,6 +222,9 @@ void recMTC0()
{
case 12:
iFlushCall(FLUSH_INTERPRETER);
xMOV(eax, ptr32[&cpuRegs.cycle]);
xADD(eax, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], eax); // update cycles
xFastCall((void*)WriteCP0Status, g_cpuConstRegs[_Rt_].UL[0]);
break;
@ -223,7 +234,9 @@ void recMTC0()
break;
case 9:
xMOV(ecx, ptr[&cpuRegs.cycle]);
xMOV(ecx, ptr32[&cpuRegs.cycle]);
xADD(ecx, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], ecx); // update cycles
xMOV(ptr[&cpuRegs.lastCOP0Cycle], ecx);
xMOV(ptr32[&cpuRegs.CP0.r[9]], g_cpuConstRegs[_Rt_].UL[0]);
break;
@ -235,19 +248,26 @@ void recMTC0()
break;
// Updates PCRs and sets the PCCR.
iFlushCall(FLUSH_INTERPRETER);
xMOV(eax, ptr32[&cpuRegs.cycle]);
xADD(eax, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], eax); // update cycles
xFastCall((void*)COP0_UpdatePCCR);
xMOV(ptr32[&cpuRegs.PERF.n.pccr], g_cpuConstRegs[_Rt_].UL[0]);
xFastCall((void*)COP0_DiagnosticPCCR);
}
else if (0 == (_Imm_ & 2)) // MTPC 0, only LSB of register matters
{
xMOV(eax, ptr[&cpuRegs.cycle]);
xMOV(eax, ptr32[&cpuRegs.cycle]);
xADD(eax, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], eax); // update cycles
xMOV(ptr32[&cpuRegs.PERF.n.pcr0], g_cpuConstRegs[_Rt_].UL[0]);
xMOV(ptr[&cpuRegs.lastPERFCycle[0]], eax);
}
else // MTPC 1
{
xMOV(eax, ptr[&cpuRegs.cycle]);
xMOV(eax, ptr32[&cpuRegs.cycle]);
xADD(eax, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], eax); // update cycles
xMOV(ptr32[&cpuRegs.PERF.n.pcr1], g_cpuConstRegs[_Rt_].UL[0]);
xMOV(ptr[&cpuRegs.lastPERFCycle[1]], eax);
}
@ -269,6 +289,9 @@ void recMTC0()
case 12:
_eeMoveGPRtoR(arg1reg, _Rt_);
iFlushCall(FLUSH_INTERPRETER);
xMOV(eax, ptr32[&cpuRegs.cycle]);
xADD(eax, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], eax); // update cycles
xFastCall((void*)WriteCP0Status);
break;
@ -279,7 +302,9 @@ void recMTC0()
break;
case 9:
xMOV(ecx, ptr[&cpuRegs.cycle]);
xMOV(ecx, ptr32[&cpuRegs.cycle]);
xADD(ecx, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], ecx); // update cycles
_eeMoveGPRtoM((uptr)&cpuRegs.CP0.r[9], _Rt_);
xMOV(ptr[&cpuRegs.lastCOP0Cycle], ecx);
break;
@ -290,19 +315,26 @@ void recMTC0()
if (0 != (_Imm_ & 0x3E)) // only effective when the register is 0
break;
iFlushCall(FLUSH_INTERPRETER);
xMOV(eax, ptr32[&cpuRegs.cycle]);
xADD(eax, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], eax); // update cycles
xFastCall((void*)COP0_UpdatePCCR);
_eeMoveGPRtoM((uptr)&cpuRegs.PERF.n.pccr, _Rt_);
xFastCall((void*)COP0_DiagnosticPCCR);
}
else if (0 == (_Imm_ & 2)) // MTPC 0, only LSB of register matters
{
xMOV(ecx, ptr[&cpuRegs.cycle]);
xMOV(ecx, ptr32[&cpuRegs.cycle]);
xADD(ecx, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], ecx); // update cycles
_eeMoveGPRtoM((uptr)&cpuRegs.PERF.n.pcr0, _Rt_);
xMOV(ptr[&cpuRegs.lastPERFCycle[0]], ecx);
}
else // MTPC 1
{
xMOV(ecx, ptr[&cpuRegs.cycle]);
xMOV(ecx, ptr32[&cpuRegs.cycle]);
xADD(ecx, scaleblockcycles_clear());
xMOV(ptr32[&cpuRegs.cycle], ecx); // update cycles
_eeMoveGPRtoM((uptr)&cpuRegs.PERF.n.pcr1, _Rt_);
xMOV(ptr[&cpuRegs.lastPERFCycle[1]], ecx);
}