#include "r3000a.h"
#include "gte.h"
#include "psxhle.h"
+#include "debug.h"
+#include "psxinterpreter.h"
+#include <assert.h>
static int branch = 0;
static int branch2 = 0;
#define debugI()
#endif
-inline void execI();
-
// Subsets
void (*psxBSC[64])();
void (*psxSPC[64])();
void (*psxREG[32])();
void (*psxCP0[32])();
-void (*psxCP2[64])();
+void (*psxCP2[64])(struct psxCP2Regs *regs);
void (*psxCP2BSC[32])();
+static u32 fetchNoCache(u32 pc)
+{
+ u32 *code = (u32 *)PSXM(pc);
+ return ((code == NULL) ? 0 : SWAP32(*code));
+}
+
+/*
+Formula One 2001 :
+Use old CPU cache code when the RAM location is updated with new code (affects in-game racing)
+*/
+static struct cache_entry {
+ u32 tag;
+ u32 data[4];
+} ICache[256];
+
+static u32 fetchICache(u32 pc)
+{
+ // cached?
+ if (pc < 0xa0000000)
+ {
+ // this is not how the hardware works but whatever
+ struct cache_entry *entry = &ICache[(pc & 0xff0) >> 4];
+
+ if (((entry->tag ^ pc) & 0xfffffff0) != 0 || pc < entry->tag)
+ {
+ u32 *code = (u32 *)PSXM(pc & ~0x0f);
+ if (!code)
+ return 0;
+
+ entry->tag = pc;
+ // treat as 4 words, although other configurations are said to be possible
+ switch (pc & 0x0c)
+ {
+ case 0x00: entry->data[0] = SWAP32(code[0]);
+ case 0x04: entry->data[1] = SWAP32(code[1]);
+ case 0x08: entry->data[2] = SWAP32(code[2]);
+ case 0x0c: entry->data[3] = SWAP32(code[3]);
+ }
+ }
+ return entry->data[(pc & 0x0f) >> 2];
+ }
+
+ return fetchNoCache(pc);
+}
+
+u32 (*fetch)(u32 pc) = fetchNoCache;
+
static void delayRead(int reg, u32 bpc) {
u32 rold, rnew;
psxRegs.pc = bpc;
- psxBranchTest();
+ branch = 0;
psxRegs.GPR.r[reg] = rold;
execI(); // first branch opcode
psxRegs.GPR.r[reg] = rnew;
- branch = 0;
+ psxBranchTest();
}
static void delayWrite(int reg, u32 bpc) {
case 0x01: // REGIMM
switch (_tRt_) {
- case 0x00: case 0x02:
- case 0x10: case 0x12: // BLTZ/BGEZ...
+ case 0x00: case 0x01:
+ case 0x10: case 0x11: // BLTZ/BGEZ...
+ // Xenogears - lbu v0 / beq v0
+ // - no load delay (fixes battle loading)
+ break;
+
if (_tRs_ == reg) return 2;
break;
}
break;
case 0x04: case 0x05: // BEQ/BNE
+ // Xenogears - lbu v0 / beq v0
+ // - no load delay (fixes battle loading)
+ break;
+
if (_tRs_ == reg || _tRt_ == reg) return 2;
break;
case 0x06: case 0x07: // BLEZ/BGTZ
+ // Xenogears - lbu v0 / beq v0
+ // - no load delay (fixes battle loading)
+ break;
+
if (_tRs_ == reg) return 2;
break;
}
void psxDelayTest(int reg, u32 bpc) {
- u32 *code;
- u32 tmp;
-
- code = (u32 *)PSXM(bpc);
- tmp = ((code == NULL) ? 0 : SWAP32(*code));
+ u32 tmp = fetch(psxRegs.pc);
branch = 1;
switch (psxTestLoadDelay(reg, tmp)) {
psxBranchTest();
}
-__inline void doBranch(u32 tar) {
- u32 *code;
+static u32 psxBranchNoDelay(void) {
+ u32 temp;
+
+ psxRegs.code = fetch(psxRegs.pc);
+ switch (_Op_) {
+ case 0x00: // SPECIAL
+ switch (_Funct_) {
+ case 0x08: // JR
+ return _u32(_rRs_);
+ case 0x09: // JALR
+ temp = _u32(_rRs_);
+ if (_Rd_) { _SetLink(_Rd_); }
+ return temp;
+ }
+ break;
+ case 0x01: // REGIMM
+ switch (_Rt_) {
+ case 0x00: // BLTZ
+ if (_i32(_rRs_) < 0)
+ return _BranchTarget_;
+ break;
+ case 0x01: // BGEZ
+ if (_i32(_rRs_) >= 0)
+ return _BranchTarget_;
+ break;
+ case 0x08: // BLTZAL
+ if (_i32(_rRs_) < 0) {
+ _SetLink(31);
+ return _BranchTarget_;
+ }
+ break;
+ case 0x09: // BGEZAL
+ if (_i32(_rRs_) >= 0) {
+ _SetLink(31);
+ return _BranchTarget_;
+ }
+ break;
+ }
+ break;
+ case 0x02: // J
+ return _JumpTarget_;
+ case 0x03: // JAL
+ _SetLink(31);
+ return _JumpTarget_;
+ case 0x04: // BEQ
+ if (_i32(_rRs_) == _i32(_rRt_))
+ return _BranchTarget_;
+ break;
+ case 0x05: // BNE
+ if (_i32(_rRs_) != _i32(_rRt_))
+ return _BranchTarget_;
+ break;
+ case 0x06: // BLEZ
+ if (_i32(_rRs_) <= 0)
+ return _BranchTarget_;
+ break;
+ case 0x07: // BGTZ
+ if (_i32(_rRs_) > 0)
+ return _BranchTarget_;
+ break;
+ }
+
+ return (u32)-1;
+}
+
+static int psxDelayBranchExec(u32 tar) {
+ execI();
+
+ branch = 0;
+ psxRegs.pc = tar;
+ psxRegs.cycle += BIAS;
+ psxBranchTest();
+ return 1;
+}
+
+static int psxDelayBranchTest(u32 tar1) {
+ u32 tar2, tmp1, tmp2;
+
+ tar2 = psxBranchNoDelay();
+ if (tar2 == (u32)-1)
+ return 0;
+
+ debugI();
+
+ /*
+ * Branch in delay slot:
+ * - execute 1 instruction at tar1
+ * - jump to tar2 (target of branch in delay slot; this branch
+ * has no normal delay slot, instruction at tar1 was fetched instead)
+ */
+ psxRegs.pc = tar1;
+ tmp1 = psxBranchNoDelay();
+ if (tmp1 == (u32)-1) {
+ return psxDelayBranchExec(tar2);
+ }
+ debugI();
+ psxRegs.cycle += BIAS;
+
+ /*
+ * Got a branch at tar1:
+ * - execute 1 instruction at tar2
+ * - jump to target of that branch (tmp1)
+ */
+ psxRegs.pc = tar2;
+ tmp2 = psxBranchNoDelay();
+ if (tmp2 == (u32)-1) {
+ return psxDelayBranchExec(tmp1);
+ }
+ debugI();
+ psxRegs.cycle += BIAS;
+
+ /*
+ * Got a branch at tar2:
+ * - execute 1 instruction at tmp1
+ * - jump to target of that branch (tmp2)
+ */
+ psxRegs.pc = tmp1;
+ return psxDelayBranchExec(tmp2);
+}
+
+static void doBranch(u32 tar) {
u32 tmp;
branch2 = branch = 1;
branchPC = tar;
- code = (u32 *)PSXM(psxRegs.pc);
- psxRegs.code = ((code == NULL) ? 0 : SWAP32(*code));
+ // check for branch in delay slot
+ if (psxDelayBranchTest(tar))
+ return;
+
+ psxRegs.code = fetch(psxRegs.pc);
debugI();
* Format: OP rs, rt *
*********************************************************/
void psxDIV() {
- if (_i32(_rRt_) != 0) {
- _i32(_rLo_) = _i32(_rRs_) / _i32(_rRt_);
- _i32(_rHi_) = _i32(_rRs_) % _i32(_rRt_);
- }
+ if (!_i32(_rRt_)) {
+ _i32(_rHi_) = _i32(_rRs_);
+ if (_i32(_rRs_) & 0x80000000) {
+ _i32(_rLo_) = 1;
+ } else {
+ _i32(_rLo_) = 0xFFFFFFFF;
+ }
+/*
+ * Notaz said that this was "not needed" for ARM platforms and could slow it down so let's disable for ARM.
+ * This fixes a crash issue that can happen when running Amidog's CPU test.
+ * (It still stays stuck to a black screen but at least it doesn't crash anymore)
+ */
+#if !defined(__arm__) && !defined(__aarch64__)
+ } else if (_i32(_rRs_) == 0x80000000 && _i32(_rRt_) == 0xFFFFFFFF) {
+ _i32(_rLo_) = 0x80000000;
+ _i32(_rHi_) = 0;
+#endif
+ } else {
+ _i32(_rLo_) = _i32(_rRs_) / _i32(_rRt_);
+ _i32(_rHi_) = _i32(_rRs_) % _i32(_rRt_);
+ }
+}
+
+void psxDIV_stall() {
+ psxRegs.muldivBusyCycle = psxRegs.cycle + 37;
+ psxDIV();
}
void psxDIVU() {
_rLo_ = _rRs_ / _rRt_;
_rHi_ = _rRs_ % _rRt_;
}
+ else {
+ _i32(_rLo_) = 0xffffffff;
+ _i32(_rHi_) = _i32(_rRs_);
+ }
+}
+
+void psxDIVU_stall() {
+ psxRegs.muldivBusyCycle = psxRegs.cycle + 37;
+ psxDIVU();
}
void psxMULT() {
psxRegs.GPR.n.hi = (u32)((res >> 32) & 0xffffffff);
}
+void psxMULT_stall() {
+ // approximate, but maybe good enough
+ u32 rs = _rRs_;
+ u32 lz = __builtin_clz(((rs ^ ((s32)rs >> 21)) | 1));
+ u32 c = 7 + (2 - (lz / 11)) * 4;
+ psxRegs.muldivBusyCycle = psxRegs.cycle + c;
+ psxMULT();
+}
+
void psxMULTU() {
u64 res = (u64)((u64)_u32(_rRs_) * (u64)_u32(_rRt_));
psxRegs.GPR.n.hi = (u32)((res >> 32) & 0xffffffff);
}
+void psxMULTU_stall() {
+ // approximate, but maybe good enough
+ u32 lz = __builtin_clz(_rRs_ | 1);
+ u32 c = 7 + (2 - (lz / 11)) * 4;
+ psxRegs.muldivBusyCycle = psxRegs.cycle + c;
+ psxMULTU();
+}
+
/*********************************************************
* Register branch logic *
* Format: OP rs, offset *
*********************************************************/
#define RepZBranchi32(op) if(_i32(_rRs_) op 0) doBranch(_BranchTarget_);
-#define RepZBranchLinki32(op) if(_i32(_rRs_) op 0) { _SetLink(31); doBranch(_BranchTarget_); }
+#define RepZBranchLinki32(op) { _SetLink(31); if(_i32(_rRs_) op 0) { doBranch(_BranchTarget_); } }
void psxBGEZ() { RepZBranchi32(>=) } // Branch if Rs >= 0
void psxBGEZAL() { RepZBranchLinki32(>=) } // Branch if Rs >= 0 and link
* Shift arithmetic with variant register shift *
* Format: OP rd, rt, rs *
*********************************************************/
-void psxSLLV() { if (!_Rd_) return; _u32(_rRd_) = _u32(_rRt_) << _u32(_rRs_); } // Rd = Rt << rs
-void psxSRAV() { if (!_Rd_) return; _i32(_rRd_) = _i32(_rRt_) >> _u32(_rRs_); } // Rd = Rt >> rs (arithmetic)
-void psxSRLV() { if (!_Rd_) return; _u32(_rRd_) = _u32(_rRt_) >> _u32(_rRs_); } // Rd = Rt >> rs (logical)
+void psxSLLV() { if (!_Rd_) return; _u32(_rRd_) = _u32(_rRt_) << (_u32(_rRs_) & 0x1F); } // Rd = Rt << rs
+void psxSRAV() { if (!_Rd_) return; _i32(_rRd_) = _i32(_rRt_) >> (_u32(_rRs_) & 0x1F); } // Rd = Rt >> rs (arithmetic)
+void psxSRLV() { if (!_Rd_) return; _u32(_rRd_) = _u32(_rRt_) >> (_u32(_rRs_) & 0x1F); } // Rd = Rt >> rs (logical)
/*********************************************************
* Load higher 16 bits of the first word in GPR with imm *
void psxMFHI() { if (!_Rd_) return; _rRd_ = _rHi_; } // Rd = Hi
void psxMFLO() { if (!_Rd_) return; _rRd_ = _rLo_; } // Rd = Lo
+static void mflohiCheckStall(void)
+{
+ u32 left = psxRegs.muldivBusyCycle - psxRegs.cycle;
+ if (left <= 37) {
+ //printf("muldiv stall %u\n", left);
+ psxRegs.cycle = psxRegs.muldivBusyCycle;
+ }
+}
+
+void psxMFHI_stall() { mflohiCheckStall(); psxMFHI(); }
+void psxMFLO_stall() { mflohiCheckStall(); psxMFLO(); }
+
/*********************************************************
* Move to GPR to HI/LO & Register jump *
* Format: OP rs *
* Format: OP *
*********************************************************/
void psxBREAK() {
- // Break exception - psx rom doens't handles this
+ psxRegs.pc -= 4;
+ psxException(0x24, branch);
}
void psxSYSCALL() {
// SysPrintf("psxRFE\n");
psxRegs.CP0.n.Status = (psxRegs.CP0.n.Status & 0xfffffff0) |
((psxRegs.CP0.n.Status & 0x3c) >> 2);
+ psxTestSWInts();
}
/*********************************************************
* Format: OP rs, rd *
*********************************************************/
void psxJR() {
- doBranch(_u32(_rRs_));
+ doBranch(_rRs_ & ~3);
psxJumpTest();
}
void psxJALR() {
u32 temp = _u32(_rRs_);
if (_Rd_) { _SetLink(_Rd_); }
- doBranch(temp);
+ doBranch(temp & ~3);
}
/*********************************************************
*/
}
-void psxSB() { psxMemWrite8 (_oB_, _u8 (_rRt_)); }
-void psxSH() { psxMemWrite16(_oB_, _u16(_rRt_)); }
-void psxSW() { psxMemWrite32(_oB_, _u32(_rRt_)); }
+void psxSB() { psxMemWrite8 (_oB_, _rRt_ & 0xff); }
+void psxSH() { psxMemWrite16(_oB_, _rRt_ & 0xffff); }
+void psxSW() { psxMemWrite32(_oB_, _rRt_); }
u32 SWL_MASK[4] = { 0xffffff00, 0xffff0000, 0xff000000, 0 };
u32 SWL_SHIFT[4] = { 24, 16, 8, 0 };
void psxCFC0() { if (!_Rt_) return; _i32(_rRt_) = (int)_rFs_; }
void psxTestSWInts() {
- // the next code is untested, if u know please
- // tell me if it works ok or not (linuzappz)
if (psxRegs.CP0.n.Cause & psxRegs.CP0.n.Status & 0x0300 &&
- psxRegs.CP0.n.Status & 0x1) {
+ psxRegs.CP0.n.Status & 0x1) {
+ psxRegs.CP0.n.Cause &= ~0x7c;
psxException(psxRegs.CP0.n.Cause, branch);
}
}
-__inline void MTC0(int reg, u32 val) {
+void MTC0(int reg, u32 val) {
// SysPrintf("MTC0 %d: %x\n", reg, val);
switch (reg) {
case 12: // Status
break;
case 13: // Cause
- psxRegs.CP0.n.Cause = val & ~(0xfc00);
+ psxRegs.CP0.n.Cause &= ~0x0300;
+ psxRegs.CP0.n.Cause |= val & 0x0300;
psxTestSWInts();
break;
}
void psxCOP2() {
- psxCP2[_Funct_]();
+ psxCP2[_Funct_]((struct psxCP2Regs *)&psxRegs.CP2D);
+}
+
+void psxCOP2_stall() {
+ u32 f = _Funct_;
+ gteCheckStall(f);
+ psxCP2[f]((struct psxCP2Regs *)&psxRegs.CP2D);
}
-void psxBASIC() {
+void psxBASIC(struct psxCP2Regs *regs) {
psxCP2BSC[_Rs_]();
}
void psxHLE() {
// psxHLEt[psxRegs.code & 0xffff]();
- psxHLEt[psxRegs.code & 0x07](); // HDHOSHY experimental patch
+// psxHLEt[psxRegs.code & 0x07](); // HDHOSHY experimental patch
+ uint32_t hleCode = psxRegs.code & 0x03ffffff;
+ if (hleCode >= (sizeof(psxHLEt) / sizeof(psxHLEt[0]))) {
+ psxNULL();
+ } else {
+ psxHLEt[hleCode]();
+ }
}
void (*psxBSC[64])() = {
psxNULL, psxNULL, psxNULL, psxNULL, psxNULL, psxNULL, psxNULL, psxNULL
};
-void (*psxCP2[64])() = {
+void (*psxCP2[64])(struct psxCP2Regs *regs) = {
psxBASIC, gteRTPS , psxNULL , psxNULL, psxNULL, psxNULL , gteNCLIP, psxNULL, // 00
psxNULL , psxNULL , psxNULL , psxNULL, gteOP , psxNULL , psxNULL , psxNULL, // 08
gteDPCS , gteINTPL, gteMVMVA, gteNCDS, gteCDP , psxNULL , gteNCDT , psxNULL, // 10
}
static void intReset() {
+ memset(&ICache, 0xff, sizeof(ICache));
}
-static void intExecute() {
- for (;;)
+void intExecute() {
+ extern int stop;
+ for (;!stop;)
execI();
}
-static void intExecuteBlock() {
+void intExecuteBlock() {
branch2 = 0;
while (!branch2) execI();
}
static void intClear(u32 Addr, u32 Size) {
}
+void intNotify (int note, void *data) {
+ /* Gameblabla - Only clear the icache if it's isolated */
+ if (note == R3000ACPU_NOTIFY_CACHE_ISOLATED)
+ {
+ memset(&ICache, 0xff, sizeof(ICache));
+ }
+}
+
+void intApplyConfig() {
+ assert(psxBSC[18] == psxCOP2 || psxBSC[18] == psxCOP2_stall);
+ assert(psxBSC[50] == gteLWC2 || psxBSC[50] == gteLWC2_stall);
+ assert(psxBSC[58] == gteSWC2 || psxBSC[58] == gteSWC2_stall);
+ assert(psxSPC[16] == psxMFHI || psxSPC[16] == psxMFHI_stall);
+ assert(psxSPC[18] == psxMFLO || psxSPC[18] == psxMFLO_stall);
+ assert(psxSPC[24] == psxMULT || psxSPC[24] == psxMULT_stall);
+ assert(psxSPC[25] == psxMULTU || psxSPC[25] == psxMULTU_stall);
+ assert(psxSPC[26] == psxDIV || psxSPC[26] == psxDIV_stall);
+ assert(psxSPC[27] == psxDIVU || psxSPC[27] == psxDIVU_stall);
+
+ if (Config.DisableStalls) {
+ psxBSC[18] = psxCOP2;
+ psxBSC[50] = gteLWC2;
+ psxBSC[58] = gteSWC2;
+ psxSPC[16] = psxMFHI;
+ psxSPC[18] = psxMFLO;
+ psxSPC[24] = psxMULT;
+ psxSPC[25] = psxMULTU;
+ psxSPC[26] = psxDIV;
+ psxSPC[27] = psxDIVU;
+ } else {
+ psxBSC[18] = psxCOP2_stall;
+ psxBSC[50] = gteLWC2_stall;
+ psxBSC[58] = gteSWC2_stall;
+ psxSPC[16] = psxMFHI_stall;
+ psxSPC[18] = psxMFLO_stall;
+ psxSPC[24] = psxMULT_stall;
+ psxSPC[25] = psxMULTU_stall;
+ psxSPC[26] = psxDIV_stall;
+ psxSPC[27] = psxDIVU_stall;
+ }
+
+ // dynarec may occasionally call the interpreter, in such a case the
+ // cache won't work (cache only works right if all fetches go through it)
+ if (!Config.icache_emulation || psxCpu != &psxInt)
+ fetch = fetchNoCache;
+ else
+ fetch = fetchICache;
+}
+
static void intShutdown() {
}
// interpreter execution
-inline void execI() {
- u32 *code = (u32 *)PSXM(psxRegs.pc);
- psxRegs.code = ((code == NULL) ? 0 : SWAP32(*code));
+void execI() {
+ psxRegs.code = fetch(psxRegs.pc);
debugI();
intExecute,
intExecuteBlock,
intClear,
+ intNotify,
+ intApplyConfig,
intShutdown
};
notaz.gp2x.de / pcsx_rearmed.git / blobdiff