static u_int get_page(u_int vaddr)
{
+#ifndef PCSX
u_int page=(vaddr^0x80000000)>>12;
+#else
+ u_int page=vaddr&~0xe0000000;
+ if (page < 0x1000000)
+ page &= ~0x0e00000; // RAM mirrors
+ page>>=12;
+#endif
#ifndef DISABLE_TLB
if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[vaddr>>12]^0x80000000)>>12;
#endif
alloc_reg(current,i,CSREG); // Load status
if(opcode2[i]<3) // MFC1/DMFC1/CFC1
{
- assert(rt1[i]);
- clear_const(current,rt1[i]);
- if(opcode2[i]==1) {
- alloc_reg64(current,i,rt1[i]); // DMFC1
- current->is32&=~(1LL<<rt1[i]);
- }else{
- alloc_reg(current,i,rt1[i]); // MFC1/CFC1
- current->is32|=1LL<<rt1[i];
+ if(rt1[i]){
+ clear_const(current,rt1[i]);
+ if(opcode2[i]==1) {
+ alloc_reg64(current,i,rt1[i]); // DMFC1
+ current->is32&=~(1LL<<rt1[i]);
+ }else{
+ alloc_reg(current,i,rt1[i]); // MFC1/CFC1
+ current->is32|=1LL<<rt1[i];
+ }
+ dirty_reg(current,rt1[i]);
}
- dirty_reg(current,rt1[i]);
alloc_reg_temp(current,i,-1);
}
else if(opcode2[i]>3) // MTC1/DMTC1/CTC1
else printf("optimizable: yes\n");
}*/
//if(is32[t]&~unneeded_reg_upper[t]&~i_is32) return 0;
+#ifndef FORCE32
if(requires_32bit[t]&~i_is32) return 0;
- else return 1;
+ else
+#endif
+ return 1;
}
return 0;
}
}
}
//if(is32[t]&~unneeded_reg_upper[t]&~i_is32) return 0;
+#ifndef FORCE32
if(requires_32bit[t]&~i_is32) return 0;
+#endif
// Delay slots are not valid branch targets
//if(t>0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP||itype[t-1]==FJUMP)) return 0;
// Delay slots require additional processing, so do not match
//if(opcode2[i]>=0x10) return; // FIXME (BxxZAL)
//assert(opcode2[i]<0x10||rs1[i]==0); // FIXME (BxxZAL)
- if(ooo)
+ if(ooo) {
if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))
- {
- // Write-after-read dependency prevents out of order execution
- // First test branch condition, then execute delay slot, then branch
- ooo=0;
+ {
+ // Write-after-read dependency prevents out of order execution
+ // First test branch condition, then execute delay slot, then branch
+ ooo=0;
+ }
+ if(rt1[i]==31&&(rs1[i+1]==31||rs2[i+1]==31||rt1[i+1]==31||rt2[i+1]==31))
+ // BxxZAL $ra is available to delay insn, so do it in order
+ ooo=0;
}
- assert(opcode2[i]<0x10||ooo); // FIXME (BxxZALL)
if(ooo) {
s1l=get_reg(branch_regs[i].regmap,rs1[i]);
load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
if(rt1[i]==31) {
int rt,return_address;
- assert(rt1[i+1]!=31);
- assert(rt2[i+1]!=31);
rt=get_reg(branch_regs[i].regmap,31);
assem_debug("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
if(rt>=0) {
// In-order execution (branch first)
//printf("IOE\n");
int nottaken=0;
+ if(rt1[i]==31) {
+ int rt,return_address;
+ rt=get_reg(branch_regs[i].regmap,31);
+ if(rt>=0) {
+ // Save the PC even if the branch is not taken
+ return_address=start+i*4+8;
+ emit_movimm(return_address,rt); // PC into link register
+ #ifdef IMM_PREFETCH
+ emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+ #endif
+ }
+ }
if(!unconditional) {
//printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
if(!only32)
{
assert(s1h>=0);
- if((opcode2[i]&0x1d)==0) // BLTZ/BLTZL
+ if((opcode2[i]&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL
{
emit_test(s1h,s1h);
nottaken=(int)out;
emit_jns(1);
}
- if((opcode2[i]&0x1d)==1) // BGEZ/BGEZL
+ if((opcode2[i]&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL
{
emit_test(s1h,s1h);
nottaken=(int)out;
else
{
assert(s1l>=0);
- if((opcode2[i]&0x1d)==0) // BLTZ/BLTZL
+ if((opcode2[i]&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL
{
emit_test(s1l,s1l);
nottaken=(int)out;
emit_jns(1);
}
- if((opcode2[i]&0x1d)==1) // BGEZ/BGEZL
+ if((opcode2[i]&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL
{
emit_test(s1l,s1l);
nottaken=(int)out;
else ba[i]=-1;
/* Is this the end of the block? */
if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)) {
+#ifdef PCSX
+ // check for link register access in delay slot
+ int rt1_=rt1[i-1];
+ if(rt1_!=0&&(rs1[i]==rt1_||rs2[i]==rt1_||rt1[i]==rt1_||rt2[i]==rt1_)) {
+ printf("link access in delay slot @%08x (%08x)\n", addr + i*4, addr);
+ ba[i-1]=-1;
+ itype[i-1]=INTCALL;
+ done=2;
+ }
+ else
+#endif
if(rt1[i-1]==0) { // Continue past subroutine call (JAL)
done=2;
}
int ds=0;
int cc=0;
int hr;
-
+
+#ifndef FORCE32
provisional_32bit();
-
+#endif
if((u_int)addr&1) {
// First instruction is delay slot
cc=-1;
}
}
}
+#ifndef FORCE32
// If something jumps here with 64-bit values
// then promote those registers to 64 bits
if(bt[i])
current.is32=temp_is32;
}
}
-#ifdef FORCE32
- memset(p32, 0xff, sizeof(p32));
+#else
current.is32=-1LL;
#endif
regs[i].wasconst=current.isconst;
regs[i].was32=current.is32;
regs[i].wasdirty=current.dirty;
- #ifdef DESTRUCTIVE_WRITEBACK
+ #if defined(DESTRUCTIVE_WRITEBACK) && !defined(FORCE32)
// To change a dirty register from 32 to 64 bits, we must write
// it out during the previous cycle (for branches, 2 cycles)
if(i<slen-1&&bt[i+1]&&itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP)
if (rt1[i]==31) {
alloc_reg(¤t,i,31);
dirty_reg(¤t,31);
- //assert(rs1[i+1]!=31&&rs2[i+1]!=31);
+ assert(rs1[i+1]!=31&&rs2[i+1]!=31);
assert(rt1[i+1]!=rt1[i]);
#ifdef REG_PREFETCH
alloc_reg(¤t,i,PTEMP);
if (rt1[i]!=0) {
alloc_reg(¤t,i,rt1[i]);
dirty_reg(¤t,rt1[i]);
- //assert(rs1[i+1]!=31&&rs2[i+1]!=31);
+ assert(rs1[i+1]!=rt1[i]&&rs2[i+1]!=rt1[i]);
assert(rt1[i+1]!=rt1[i]);
#ifdef REG_PREFETCH
alloc_reg(¤t,i,PTEMP);
if (rt1[i]==31) { // BLTZAL/BGEZAL
alloc_reg(¤t,i,31);
dirty_reg(¤t,31);
- assert(rs1[i+1]!=31&&rs2[i+1]!=31);
//#ifdef REG_PREFETCH
//alloc_reg(¤t,i,PTEMP);
//#endif
clean_registers(0,slen-1,1);
/* Pass 7 - Identify 32-bit registers */
-
+#ifndef FORCE32
provisional_r32();
u_int r32=0;
}
//requires_32bit[i]=is32[i]&~unneeded_reg_upper[i]; // DEBUG
}
+#endif
if(itype[slen-1]==SPAN) {
bt[slen-1]=1; // Mark as a branch target so instruction can restart after exception
u_int vpage=get_vpage(vaddr);
literal_pool(256);
//if(!(is32[i]&(~unneeded_reg_upper[i])&~(1LL<<CCREG)))
+#ifndef FORCE32
if(!requires_32bit[i])
+#else
+ if(1)
+#endif
{
assem_debug("%8x (%d) <- %8x\n",instr_addr[i],i,start+i*4);
assem_debug("jump_in: %x\n",start+i*4);
notaz.gp2x.de / pcsx_rearmed.git / blobdiff