#define COP2 27 // Coprocessor 2 move
#define C2LS 28 // Coprocessor 2 load/store
#define C2OP 29 // Coprocessor 2 operation
+#define INTCALL 30// Call interpreter to handle rare corner cases
/* stubs */
#define CC_STUB 1
void jump_syscall_hle();
void jump_eret();
void jump_hlecall();
+void jump_intcall();
void new_dyna_leave();
// TLB
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
{
#ifdef FORCE32
return get_addr(vaddr);
-#endif
+#else
//printf("TRACE: count=%d next=%d (get_addr_32 %x,flags %x)\n",Count,next_interupt,vaddr,flags);
int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
if(ht_bin[0]==vaddr) return (void *)ht_bin[1];
Context=(Context&0xFF80000F)|((BadVAddr>>9)&0x007FFFF0);
EntryHi=BadVAddr&0xFFFFE000;
return get_addr_ht(0x80000000);
+#endif
}
void clear_all_regs(signed char regmap[])
if(itype[i]==LOADLR) {
hsn[FTEMP]=0;
}
- // Also 64-bit SDL/SDR
- if(opcode[i]==0x2c||opcode[i]==0x2d) {
+ // Also SWL/SWR/SDL/SDR
+ if(opcode[i]==0x2a||opcode[i]==0x2e||opcode[i]==0x2c||opcode[i]==0x2d) {
hsn[FTEMP]=0;
}
// Don't remove the TLB registers either
j++;
break;
}
- if(itype[i+j]==SYSCALL||itype[i+j]==HLECALL||((source[i+j]&0xfc00003f)==0x0d))
+ if(itype[i+j]==SYSCALL||itype[i+j]==HLECALL||itype[i+j]==INTCALL||((source[i+j]&0xfc00003f)==0x0d))
{
break;
}
// Dereference the pointers and remove if it matches
void ll_kill_pointers(struct ll_entry *head,int addr,int shift)
{
+ u_int old_host_addr=0;
while(head) {
int ptr=get_pointer(head->addr);
inv_debug("EXP: Lookup pointer to %x at %x (%x)\n",(int)ptr,(int)head->addr,head->vaddr);
(((ptr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift)))
{
inv_debug("EXP: Kill pointer at %x (%x)\n",(int)head->addr,head->vaddr);
- kill_pointer(head->addr);
+ u_int host_addr=(u_int)kill_pointer(head->addr);
+
+ if((host_addr>>12)!=(old_host_addr>>12)) {
+ #ifdef __arm__
+ __clear_cache((void *)(old_host_addr&~0xfff),(void *)(old_host_addr|0xfff));
+ #endif
+ old_host_addr=host_addr;
+ }
}
head=head->next;
}
+ #ifdef __arm__
+ if (old_host_addr)
+ __clear_cache((void *)(old_host_addr&~0xfff),(void *)(old_host_addr|0xfff));
+ #endif
}
// This is called when we write to a compiled block (see do_invstub)
-int invalidate_page(u_int page)
+void invalidate_page(u_int page)
{
- int modified=0;
struct ll_entry *head;
struct ll_entry *next;
+ u_int old_host_addr=0;
head=jump_in[page];
jump_in[page]=0;
while(head!=NULL) {
jump_out[page]=0;
while(head!=NULL) {
inv_debug("INVALIDATE: kill pointer to %x (%x)\n",head->vaddr,(int)head->addr);
- kill_pointer(head->addr);
- modified=1;
+ u_int host_addr=(u_int)kill_pointer(head->addr);
+
+ if((host_addr>>12)!=(old_host_addr>>12)) {
+ #ifdef __arm__
+ __clear_cache((void *)(old_host_addr&~0xfff),(void *)(old_host_addr|0xfff));
+ #endif
+ old_host_addr=host_addr;
+ }
next=head->next;
free(head);
head=next;
}
- return modified;
+ #ifdef __arm__
+ if (old_host_addr)
+ __clear_cache((void *)(old_host_addr&~0xfff),(void *)(old_host_addr|0xfff));
+ #endif
}
void invalidate_block(u_int block)
{
- int modified;
u_int page=get_page(block<<12);
u_int vpage=get_vpage(block<<12);
inv_debug("INVALIDATE: %x (%d)\n",block<<12,page);
if(vpage>2047||(head->vaddr>>12)==block) { // Ignore vaddr hash collision
get_bounds((int)head->addr,&start,&end);
//printf("start: %x end: %x\n",start,end);
- if(page<2048&&start>=0x80000000&&end<0x80800000) {
+ if(page<2048&&start>=0x80000000&&end<0x80000000+RAM_SIZE) {
if(((start-(u_int)rdram)>>12)<=page&&((end-1-(u_int)rdram)>>12)>=page) {
if((((start-(u_int)rdram)>>12)&2047)<first) first=((start-(u_int)rdram)>>12)&2047;
if((((end-1-(u_int)rdram)>>12)&2047)>last) last=((end-1-(u_int)rdram)>>12)&2047;
head=head->next;
}
//printf("first=%d last=%d\n",first,last);
- modified=invalidate_page(page);
+ invalidate_page(page);
assert(first+5>page); // NB: this assumes MAXBLOCK<=4096 (4 pages)
assert(last<page+5);
// Invalidate the adjacent pages if a block crosses a 4K boundary
}
else if(block>=0x80000&&block<0x80800) memory_map[block]=((u_int)rdram-0x80000000)>>2;
#endif
- #ifdef __arm__
- if(modified)
- __clear_cache((void *)BASE_ADDR,(void *)BASE_ADDR+(1<<TARGET_SIZE_2));
- #endif
+
#ifdef USE_MINI_HT
memset(mini_ht,-1,sizeof(mini_ht));
#endif
u_int i;
u_int inv=0;
get_bounds((int)head->addr,&start,&end);
- if(start-(u_int)rdram<0x800000) {
+ if(start-(u_int)rdram<RAM_SIZE) {
for(i=(start-(u_int)rdram+0x80000000)>>12;i<=(end-1-(u_int)rdram+0x80000000)>>12;i++) {
inv|=invalid_code[i];
}
//printf("addr=%x start=%x end=%x\n",addr,start,end);
if(addr<start||addr>=end) inv=1;
}
- else if((signed int)head->vaddr>=(signed int)0x80800000) {
+ else if((signed int)head->vaddr>=(signed int)0x80000000+RAM_SIZE) {
inv=1;
}
if(!inv) {
// On CPUs without 32-bit immediates we need a pointer to invalid_code
else alloc_reg(current,i,INVCP);
#endif
- if(opcode[i]==0x2c||opcode[i]==0x2d) { // 64-bit SDL/SDR
+ if(opcode[i]==0x2a||opcode[i]==0x2e||opcode[i]==0x2c||opcode[i]==0x2d) { // SWL/SWL/SDL/SDR
alloc_reg(current,i,FTEMP);
}
// We need a temporary register for address generation
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
int s,th,tl,addr,map=-1;
int offset;
int jaddr=0;
- int memtarget,c=0;
+ int memtarget=0,c=0;
u_int hr,reglist=0;
th=get_reg(i_regs->regmap,rt1[i]|64);
tl=get_reg(i_regs->regmap,rt1[i]);
if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
if(s>=0) {
c=(i_regs->wasconst>>s)&1;
- memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+ memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
}
- if(offset||s<0||c) addr=tl;
- else addr=s;
//printf("load_assemble: c=%d\n",c);
//if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset);
// FIXME: Even if the load is a NOP, we should check for pagefaults...
+#ifdef PCSX
+ if(tl<0&&(!c||(((u_int)constmap[i][s]+offset)>>16)==0x1f80)
+ ||rt1[i]==0) {
+ // could be FIFO, must perform the read
+ // ||dummy read
+ assem_debug("(forced read)\n");
+ tl=get_reg(i_regs->regmap,-1);
+ assert(tl>=0);
+ }
+#endif
+ if(offset||s<0||c) addr=tl;
+ else addr=s;
if(tl>=0) {
//assert(tl>=0);
//assert(rt1[i]);
//#define R29_HACK 1
#ifdef R29_HACK
// Strmnnrmn's speed hack
- if(rs1[i]!=29||start<0x80001000||start>=0x80800000)
+ if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
#endif
{
- emit_cmpimm(addr,0x800000);
+ emit_cmpimm(addr,RAM_SIZE);
jaddr=(int)out;
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
// Hint to branch predictor that the branch is unlikely to be taken
offset=imm[i];
if(s>=0) {
c=(i_regs->wasconst>>s)&1;
- memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+ memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
}
assert(tl>=0);
#ifdef R29_HACK
// Strmnnrmn's speed hack
memtarget=1;
- if(rs1[i]!=29||start<0x80001000||start>=0x80800000)
+ if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
#endif
- emit_cmpimm(addr,0x800000);
+ emit_cmpimm(addr,RAM_SIZE);
#ifdef DESTRUCTIVE_SHIFT
if(s==addr) emit_mov(s,temp);
#endif
#ifdef R29_HACK
- if(rs1[i]!=29||start<0x80001000||start>=0x80800000)
+ if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
#endif
{
jaddr=(int)out;
int case1,case2,case3;
int done0,done1,done2;
int memtarget,c=0;
+ int agr=AGEN1+(i&1);
u_int hr,reglist=0;
th=get_reg(i_regs->regmap,rs2[i]|64);
tl=get_reg(i_regs->regmap,rs2[i]);
s=get_reg(i_regs->regmap,rs1[i]);
- temp=get_reg(i_regs->regmap,-1);
+ temp=get_reg(i_regs->regmap,agr);
+ if(temp<0) temp=get_reg(i_regs->regmap,-1);
offset=imm[i];
if(s>=0) {
c=(i_regs->isconst>>s)&1;
- memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+ memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
}
assert(tl>=0);
assert(temp>=0);
if(!using_tlb) {
if(!c) {
- emit_cmpimm(s<0||offset?temp:s,0x800000);
+ emit_cmpimm(s<0||offset?temp:s,RAM_SIZE);
if(!offset&&s!=temp) emit_mov(s,temp);
jaddr=(int)out;
emit_jno(0);
set_jump_target(done0,(int)out);
}
if(!c||!memtarget)
- add_stub(STORELR_STUB,jaddr,(int)out,0,(int)i_regs,rs2[i],ccadj[i],reglist);
+ add_stub(STORELR_STUB,jaddr,(int)out,i,(int)i_regs,temp,ccadj[i],reglist);
}
if(!using_tlb) {
emit_addimm_no_flags((u_int)0x80000000-(u_int)rdram,temp);
// Generate address + offset
if(!using_tlb) {
if(!c)
- emit_cmpimm(offset||c||s<0?ar:s,0x800000);
+ emit_cmpimm(offset||c||s<0?ar:s,RAM_SIZE);
}
else
{
jaddr2=(int)out;
emit_jno(0);
}
- else if(((signed int)(constmap[i][s]+offset))>=(signed int)0x80800000) {
+ else if(((signed int)(constmap[i][s]+offset))>=(signed int)0x80000000+RAM_SIZE) {
jaddr2=(int)out;
emit_jmp(0); // inline_readstub/inline_writestub? Very rare case
}
int s,tl;
int ar;
int offset;
- int c=0;
+ int memtarget=0,c=0;
int jaddr,jaddr2=0,jaddr3,type;
int agr=AGEN1+(i&1);
u_int hr,reglist=0;
} else { // LWC2
ar=tl;
}
+ if(s>=0) c=(i_regs->wasconst>>s)&1;
+ memtarget=c&&(((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE);
if (!offset&&!c&&s>=0) ar=s;
assert(ar>=0);
if (opcode[i]==0x3a) { // SWC2
cop2_get_dreg(copr,tl,HOST_TEMPREG);
+ type=STOREW_STUB;
}
- if(s>=0) c=(i_regs->wasconst>>s)&1;
- if(!c) {
- emit_cmpimm(offset||c||s<0?ar:s,0x800000);
- jaddr2=(int)out;
- emit_jno(0);
- }
- else if(((signed int)(constmap[i][s]+offset))>=(signed int)0x80800000) {
- jaddr2=(int)out;
- emit_jmp(0); // inline_readstub/inline_writestub? Very rare case
- }
- if (opcode[i]==0x32) { // LWC2
- #ifdef HOST_IMM_ADDR32
- if(c) emit_readword_tlb(constmap[i][s]+offset,-1,tl);
- else
- #endif
- emit_readword_indexed(0,ar,tl);
+ else
type=LOADW_STUB;
+
+ if(c&&!memtarget) {
+ jaddr2=(int)out;
+ emit_jmp(0); // inline_readstub/inline_writestub?
}
- if (opcode[i]==0x3a) { // SWC2
-#ifdef DESTRUCTIVE_SHIFT
- if(!offset&&!c&&s>=0) emit_mov(s,ar);
-#endif
- emit_writeword_indexed(tl,0,ar);
- type=STOREW_STUB;
+ else {
+ if(!c) {
+ emit_cmpimm(offset||c||s<0?ar:s,RAM_SIZE);
+ jaddr2=(int)out;
+ emit_jno(0);
+ }
+ if (opcode[i]==0x32) { // LWC2
+ #ifdef HOST_IMM_ADDR32
+ if(c) emit_readword_tlb(constmap[i][s]+offset,-1,tl);
+ else
+ #endif
+ emit_readword_indexed(0,ar,tl);
+ }
+ if (opcode[i]==0x3a) { // SWC2
+ #ifdef DESTRUCTIVE_SHIFT
+ if(!offset&&!c&&s>=0) emit_mov(s,ar);
+ #endif
+ emit_writeword_indexed(tl,0,ar);
+ }
}
if(jaddr2)
add_stub(type,jaddr2,(int)out,i,ar,(int)i_regs,ccadj[i],reglist);
{
//if(opcode2[i]==0x10||opcode2[i]==0x12) { // MFHI/MFLO
//if(opcode2[i]==0x11||opcode2[i]==0x13) { // MTHI/MTLO
- assert(rt1[i]>0);
+ //assert(rt1[i]>0);
if(rt1[i]) {
signed char sh,sl,th,tl;
th=get_reg(i_regs->regmap,rt1[i]|64);
emit_jmp((int)jump_hlecall);
}
+void intcall_assemble(int i,struct regstat *i_regs)
+{
+ signed char ccreg=get_reg(i_regs->regmap,CCREG);
+ assert(ccreg==HOST_CCREG);
+ assert(!is_delayslot);
+ emit_movimm(start+i*4,0); // Get PC
+ emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+ emit_jmp((int)jump_intcall);
+}
+
void ds_assemble(int i,struct regstat *i_regs)
{
is_delayslot=1;
mov_assemble(i,i_regs);break;
case SYSCALL:
case HLECALL:
+ case INTCALL:
case SPAN:
case UJUMP:
case RJUMP:
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(itype[i]==C1LS||itype[i]==C2LS) {
if ((opcode[i]&0x3b)==0x31||(opcode[i]&0x3b)==0x32) // LWC1/LDC1/LWC2/LDC2
ra=get_reg(i_regs->regmap,FTEMP);
- else { // SWC1/SDC1
+ else { // SWC1/SDC1/SWC2/SDC2
ra=get_reg(i_regs->regmap,agr);
if(ra<0) ra=get_reg(i_regs->regmap,-1);
}
// Stores to memory go thru the mapper to detect self-modifying
// code, loads don't.
if((unsigned int)(constmap[i][rs]+offset)>=0xC0000000 ||
- (unsigned int)(constmap[i][rs]+offset)<0x80800000 )
+ (unsigned int)(constmap[i][rs]+offset)<0x80000000+RAM_SIZE )
generate_map_const(constmap[i][rs]+offset,rm);
}else{
if((signed int)(constmap[i][rs]+offset)>=(signed int)0xC0000000)
// Stores to memory go thru the mapper to detect self-modifying
// code, loads don't.
if((unsigned int)(constmap[i+1][rs]+offset)>=0xC0000000 ||
- (unsigned int)(constmap[i+1][rs]+offset)<0x80800000 )
+ (unsigned int)(constmap[i+1][rs]+offset)<0x80000000+RAM_SIZE )
generate_map_const(constmap[i+1][rs]+offset,ra);
}else{
if((signed int)(constmap[i+1][rs]+offset)>=(signed int)0xC0000000)
}
}
//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
mov_assemble(t,®s[t]);break;
case SYSCALL:
case HLECALL:
+ case INTCALL:
case SPAN:
case UJUMP:
case RJUMP:
#endif
//if(opcode2[i]>=0x10) return; // FIXME (BxxZAL)
- assert(opcode2[i]<0x10||rs1[i]==0); // 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;
}
- // TODO: Conditional branches w/link must execute in-order so that
- // condition test and write to r31 occur before cycle count test
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) {
if(!only32)
{
assert(s1h>=0);
- if(opcode2[i]==0) // BLTZ
+ if((opcode2[i]&0xf)==0) // BLTZ/BLTZAL
{
emit_test(s1h,s1h);
if(invert){
emit_js(0);
}
}
- if(opcode2[i]==1) // BGEZ
+ if((opcode2[i]&0xf)==1) // BGEZ/BLTZAL
{
emit_test(s1h,s1h);
if(invert){
else
{
assert(s1l>=0);
- if(opcode2[i]==0) // BLTZ
+ if((opcode2[i]&0xf)==0) // BLTZ/BLTZAL
{
emit_test(s1l,s1l);
if(invert){
emit_js(0);
}
}
- if(opcode2[i]==1) // BGEZ
+ if((opcode2[i]&0xf)==1) // BGEZ/BLTZAL
{
emit_test(s1l,s1l);
if(invert){
// 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;
mov_assemble(0,®s[0]);break;
case SYSCALL:
case HLECALL:
+ case INTCALL:
case SPAN:
case UJUMP:
case RJUMP:
{
uu=u=0x300C0F3; // Discard at, a0-a3, t6-t9
}
- if(start>0x80000400&&start<0x80800000) {
+ if(start>0x80000400&&start<0x80000000+RAM_SIZE) {
if(itype[i]==UJUMP&&rt1[i]==31)
{
//uu=u=0x30300FF0FLL; // Discard at, v0-v1, t0-t9, lo, hi
}
}
}
- else if(itype[i]==SYSCALL||itype[i]==HLECALL)
+ else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
{
// SYSCALL instruction (software interrupt)
u=1;
// Save it
unneeded_reg[i]=u;
unneeded_reg_upper[i]=uu;
-#ifdef FORCE32
- unneeded_reg_upper[i]=-1LL;
-#endif
/*
printf("ur (%d,%d) %x: ",istart,iend,start+i*4);
printf("U:");
}
printf("\n");*/
}
+#ifdef FORCE32
+ for (i=iend;i>=istart;i--)
+ {
+ unneeded_reg_upper[i]=branch_unneeded_reg_upper[i]=-1LL;
+ }
+#endif
}
// Identify registers which are likely to contain 32-bit values
if((regs[i].was32>>dep2[i+1])&1) r32|=1LL<<dep2[i+1];
}
}
- else if(itype[i]==SYSCALL||itype[i]==HLECALL)
+ else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
{
// SYSCALL instruction (software interrupt)
r32=0;
}
}
}
- else if(itype[i]==SYSCALL||itype[i]==HLECALL)
+ else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
{
// SYSCALL instruction (software interrupt)
will_dirty_i=0;
case FJUMP:
printf (" %x: %s %8x\n",start+i*4,insn[i],ba[i]);break;
case RJUMP:
- if (rt1[i]!=31)
+ if (opcode[i]==0x9&&rt1[i]!=31)
printf (" %x: %s r%d,r%d\n",start+i*4,insn[i],rt1[i],rs1[i]);
else
printf (" %x: %s r%d\n",start+i*4,insn[i],rs1[i]);
case C2LS:
printf (" %x: %s cpr2[%d],r%d+%x\n",start+i*4,insn[i],(source[i]>>16)&0x1f,rs1[i],imm[i]);
break;
+ case INTCALL:
+ printf (" %x: %s (INTCALL)\n",start+i*4,insn[i]);
+ break;
default:
//printf (" %s %8x\n",insn[i],source[i]);
printf (" %x: %s\n",start+i*4,insn[i]);
start = (u_int)addr&~3;
//assert(((u_int)addr&1)==0);
#ifdef PCSX
- if (Config.HLE && start == 0x80001000) {
+ if (Config.HLE && start == 0x80001000) // hlecall
+ {
// XXX: is this enough? Maybe check hleSoftCall?
+ u_int beginning=(u_int)out;
u_int page=get_page(start);
- ll_add(jump_in+page,start,out);
invalid_code[start>>12]=0;
emit_movimm(start,0);
emit_writeword(0,(int)&pcaddr);
- emit_jmp((int)new_dyna_leave); // enough??
+ emit_jmp((int)new_dyna_leave);
+#ifdef __arm__
+ __clear_cache((void *)beginning,out);
+#endif
+ ll_add(jump_in+page,start,(void *)beginning);
return 0;
}
- else if ((u_int)addr < 0x00200000) {
+ else if ((u_int)addr < 0x00200000 ||
+ (0xa0000000 <= addr && addr < 0xa0200000)) {
// used for BIOS calls mostly?
- source = (u_int *)((u_int)rdram+start-0);
- pagelimit = 0x00200000;
+ source = (u_int *)((u_int)rdram+(start&0x1fffff));
+ pagelimit = (addr&0xa0000000)|0x00200000;
+ }
+ else if (!Config.HLE && (
+/* (0x9fc00000 <= addr && addr < 0x9fc80000) ||*/
+ (0xbfc00000 <= addr && addr < 0xbfc80000))) {
+ // BIOS
+ source = (u_int *)((u_int)psxR+(start&0x7ffff));
+ pagelimit = (addr&0xfff00000)|0x80000;
}
else
#endif
}
else
#endif
- if ((int)addr >= 0x80000000 && (int)addr < 0x80800000) {
+ if ((int)addr >= 0x80000000 && (int)addr < 0x80000000+RAM_SIZE) {
source = (u_int *)((u_int)rdram+start-0x80000000);
- pagelimit = 0x80800000;
+ pagelimit = 0x80000000+RAM_SIZE;
}
#ifndef DISABLE_TLB
else if ((signed int)addr >= (signed int)0xC0000000) {
else {
assem_debug("Compile at unmapped memory address: %x \n", (int)addr);
//assem_debug("start: %x next: %x\n",memory_map[start>>12],memory_map[(start+4096)>>12]);
- return 1; // Caller will invoke exception handler
+ return -1; // Caller will invoke exception handler
}
//printf("source= %x\n",(int)source);
}
case 0x02: strcpy(insn[i],"TLBWI"); type=COP0; break;
case 0x06: strcpy(insn[i],"TLBWR"); type=COP0; break;
case 0x08: strcpy(insn[i],"TLBP"); type=COP0; break;
+#ifdef PCSX
+ case 0x10: strcpy(insn[i],"RFE"); type=COP0; break;
+#else
case 0x18: strcpy(insn[i],"ERET"); type=COP0; break;
+#endif
}
}
break;
break;
}
break;
+#ifndef FORCE32
case 0x14: strcpy(insn[i],"BEQL"); type=CJUMP; break;
case 0x15: strcpy(insn[i],"BNEL"); type=CJUMP; break;
case 0x16: strcpy(insn[i],"BLEZL"); type=CJUMP; break;
case 0x19: strcpy(insn[i],"DADDIU"); type=IMM16; break;
case 0x1A: strcpy(insn[i],"LDL"); type=LOADLR; break;
case 0x1B: strcpy(insn[i],"LDR"); type=LOADLR; break;
+#endif
case 0x20: strcpy(insn[i],"LB"); type=LOAD; break;
case 0x21: strcpy(insn[i],"LH"); type=LOAD; break;
case 0x22: strcpy(insn[i],"LWL"); type=LOADLR; break;
case 0x29: strcpy(insn[i],"SH"); type=STORE; break;
case 0x2A: strcpy(insn[i],"SWL"); type=STORELR; break;
case 0x2B: strcpy(insn[i],"SW"); type=STORE; break;
+#ifndef FORCE32
case 0x2C: strcpy(insn[i],"SDL"); type=STORELR; break;
case 0x2D: strcpy(insn[i],"SDR"); type=STORELR; break;
+#endif
case 0x2E: strcpy(insn[i],"SWR"); type=STORELR; break;
case 0x2F: strcpy(insn[i],"CACHE"); type=NOP; break;
case 0x30: strcpy(insn[i],"LL"); type=NI; break;
case 0x31: strcpy(insn[i],"LWC1"); type=C1LS; break;
+#ifndef FORCE32
case 0x34: strcpy(insn[i],"LLD"); type=NI; break;
case 0x35: strcpy(insn[i],"LDC1"); type=C1LS; break;
case 0x37: strcpy(insn[i],"LD"); type=LOAD; break;
+#endif
case 0x38: strcpy(insn[i],"SC"); type=NI; break;
case 0x39: strcpy(insn[i],"SWC1"); type=C1LS; break;
+#ifndef FORCE32
case 0x3C: strcpy(insn[i],"SCD"); type=NI; break;
case 0x3D: strcpy(insn[i],"SDC1"); type=C1LS; break;
case 0x3F: strcpy(insn[i],"SD"); type=STORE; break;
+#endif
#ifdef PCSX
case 0x12: strcpy(insn[i],"COP2"); type=NI;
+ // note: COP MIPS-1 encoding differs from MIPS32
op2=(source[i]>>21)&0x1f;
- switch(op2)
+ if (source[i]&0x3f) {
+ if (gte_handlers[source[i]&0x3f]!=NULL) {
+ snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
+ type=C2OP;
+ }
+ }
+ else switch(op2)
{
case 0x00: strcpy(insn[i],"MFC2"); type=COP2; break;
case 0x02: strcpy(insn[i],"CFC2"); type=COP2; break;
case 0x04: strcpy(insn[i],"MTC2"); type=COP2; break;
case 0x06: strcpy(insn[i],"CTC2"); type=COP2; break;
- default:
- if (gte_handlers[source[i]&0x3f]!=NULL) {
- snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
- type=C2OP;
- }
- break;
}
break;
case 0x32: strcpy(insn[i],"LWC2"); type=C2LS; break;
case 0x3B: strcpy(insn[i],"HLECALL"); type=HLECALL; break;
#endif
default: strcpy(insn[i],"???"); type=NI;
- printf("NI %08x @%08x\n", source[i], addr + i*4);
+ printf("NI %08x @%08x (%08x)\n", source[i], addr + i*4, addr);
break;
}
+#ifdef PCSX
+ /* detect branch in delay slot early */
+ if(type==RJUMP||type==UJUMP||type==CJUMP||type==SJUMP||type==FJUMP) {
+ opcode[i+1]=source[i+1]>>26;
+ opcode2[i+1]=source[i+1]&0x3f;
+ if((0<opcode[i+1]&&opcode[i+1]<8)||(opcode[i+1]==0&&(opcode2[i+1]==8||opcode2[i+1]==9))) {
+ printf("branch in delay slot @%08x (%08x)\n", addr + i*4+4, addr);
+ // don't handle first branch and call interpreter if it's hit
+ type=INTCALL;
+ }
+ }
+#endif
itype[i]=type;
opcode2[i]=op2;
/* Get registers/immediates */
break;
case SYSCALL:
case HLECALL:
+ case INTCALL:
rs1[i]=CCREG;
rs2[i]=0;
rt1[i]=0;
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=1;
- // Does the block continue due to a branch?
- for(j=i-1;j>=0;j--)
- {
- if(ba[j]==start+i*4+4) done=j=0;
- if(ba[j]==start+i*4+8) done=j=0;
- }
+ done=2;
}
else {
if(stop_after_jal) done=1;
// Don't get too close to the limit
if(i>MAXBLOCK/2) done=1;
}
- if(i>0&&itype[i-1]==SYSCALL&&stop_after_jal) done=1;
- if(itype[i-1]==HLECALL) done=1;
- assert(i<MAXBLOCK-1);
+ if(itype[i]==SYSCALL&&stop_after_jal) done=1;
+ if(itype[i]==HLECALL||itype[i]==INTCALL) done=2;
+ if(done==2) {
+ // Does the block continue due to a branch?
+ for(j=i-1;j>=0;j--)
+ {
+ if(ba[j]==start+i*4+4) done=j=0;
+ if(ba[j]==start+i*4+8) done=j=0;
+ }
+ }
+ //assert(i<MAXBLOCK-1);
if(start+i*4==pagelimit-4) done=1;
assert(start+i*4<pagelimit);
if (i==MAXBLOCK-1) done=1;
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)
alloc_reg(¤t,i,31);
dirty_reg(¤t,31);
assert(rs1[i+1]!=31&&rs2[i+1]!=31);
+ assert(rt1[i+1]!=rt1[i]);
#ifdef REG_PREFETCH
alloc_reg(¤t,i,PTEMP);
#endif
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);
#endif
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
break;
case SYSCALL:
case HLECALL:
+ case INTCALL:
syscall_alloc(¤t,i);
break;
case SPAN:
}
}
}
- else if(itype[i]==SYSCALL||itype[i]==HLECALL)
+ else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
{
// SYSCALL instruction (software interrupt)
nr=0;
f_regmap[hr]=regs[i].regmap[hr];
else f_regmap[hr]=-1;
}
- else if(regs[i].regmap[hr]>=0) f_regmap[hr]=regs[i].regmap[hr];
+ else if(regs[i].regmap[hr]>=0) {
+ if(f_regmap[hr]!=regs[i].regmap[hr]) {
+ // dealloc old register
+ int n;
+ for(n=0;n<HOST_REGS;n++)
+ {
+ if(f_regmap[n]==regs[i].regmap[hr]) {f_regmap[n]=-1;}
+ }
+ // and alloc new one
+ f_regmap[hr]=regs[i].regmap[hr];
+ }
+ }
if(branch_regs[i].regmap[hr]>64) {
if(!((branch_regs[i].dirty>>hr)&1))
f_regmap[hr]=branch_regs[i].regmap[hr];
else f_regmap[hr]=-1;
}
- else if(branch_regs[i].regmap[hr]>=0) f_regmap[hr]=branch_regs[i].regmap[hr];
+ else if(branch_regs[i].regmap[hr]>=0) {
+ if(f_regmap[hr]!=branch_regs[i].regmap[hr]) {
+ // dealloc old register
+ int n;
+ for(n=0;n<HOST_REGS;n++)
+ {
+ if(f_regmap[n]==branch_regs[i].regmap[hr]) {f_regmap[n]=-1;}
+ }
+ // and alloc new one
+ f_regmap[hr]=branch_regs[i].regmap[hr];
+ }
+ }
if(itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS
||itype[i+1]==SHIFT||itype[i+1]==COP1||itype[i+1]==FLOAT
||itype[i+1]==FCOMP||itype[i+1]==FCONV
if(!((regs[i].dirty>>hr)&1))
f_regmap[hr]=regs[i].regmap[hr];
}
- else if(regs[i].regmap[hr]>=0) f_regmap[hr]=regs[i].regmap[hr];
+ else if(regs[i].regmap[hr]>=0) {
+ if(f_regmap[hr]!=regs[i].regmap[hr]) {
+ // dealloc old register
+ int n;
+ for(n=0;n<HOST_REGS;n++)
+ {
+ if(f_regmap[n]==regs[i].regmap[hr]) {f_regmap[n]=-1;}
+ }
+ // and alloc new one
+ f_regmap[hr]=regs[i].regmap[hr];
+ }
+ }
else if(regs[i].regmap[hr]<0) count++;
}
}
clean_registers(0,slen-1,1);
/* Pass 7 - Identify 32-bit registers */
-
+#ifndef FORCE32
provisional_r32();
u_int r32=0;
if((regs[i].was32>>dep2[i+1])&1) r32|=1LL<<dep2[i+1];
}
}
- else if(itype[i]==SYSCALL||itype[i]==HLECALL)
+ else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
{
// SYSCALL instruction (software interrupt)
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
ds=1;
pagespan_ds();
}
+ u_int instr_addr0_override=0;
+
+#ifdef PCSX
+ if (start == 0x80030000) {
+ // nasty hack for fastbios thing
+ instr_addr0_override=(u_int)out;
+ emit_movimm(start,0);
+ emit_readword((int)&pcaddr,1);
+ emit_writeword(0,(int)&pcaddr);
+ emit_cmp(0,1);
+ emit_jne((int)new_dyna_leave);
+ }
+#endif
for(i=0;i<slen;i++)
{
//if(ds) printf("ds: ");
syscall_assemble(i,®s[i]);break;
case HLECALL:
hlecall_assemble(i,®s[i]);break;
+ case INTCALL:
+ intcall_assemble(i,®s[i]);break;
case UJUMP:
ujump_assemble(i,®s[i]);ds=1;break;
case RJUMP:
}
}
+ if (instr_addr0_override)
+ instr_addr[0] = instr_addr0_override;
+
/* Pass 9 - Linker */
for(i=0;i<linkcount;i++)
{
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);
break;
case 3:
// Clear jump_out
- #ifdef __arm__
- if((expirep&2047)==0)
- __clear_cache((void *)BASE_ADDR,(void *)BASE_ADDR+(1<<TARGET_SIZE_2));
- #endif
ll_remove_matching_addrs(jump_out+(expirep&2047),base,shift);
ll_remove_matching_addrs(jump_out+2048+(expirep&2047),base,shift);
break;
notaz.gp2x.de / pcsx_rearmed.git / blobdiff