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
// 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);
if(((ptr>>shift)==(addr>>shift)) ||
(((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);
+ printf("EXP: Kill pointer at %x (%x)\n",(int)head->addr,head->vaddr);
+ 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
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;
}
//printf("load_assemble: c=%d\n",c);
//#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(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)
#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(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))
// First test branch condition, then execute delay slot, then branch
ooo=0;
}
- // TODO: Conditional branches w/link must execute in-order so that
- // condition test and write to r31 occur before cycle count test
+ assert(opcode2[i]<0x10||ooo); // FIXME (BxxZALL)
if(ooo) {
s1l=get_reg(branch_regs[i].regmap,rs1[i]);
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){
{
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
#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) {
case 0x15: strcpy(insn[i],"BNEL"); type=CJUMP; break;
case 0x16: strcpy(insn[i],"BLEZL"); type=CJUMP; break;
case 0x17: strcpy(insn[i],"BGTZL"); type=CJUMP; break;
+#ifndef FORCE32
case 0x18: strcpy(insn[i],"DADDI"); type=IMM16; 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;
op2=(source[i]>>21)&0x1f;
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;
}
itype[i]=type;
// 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) done=1;
+ //assert(i<MAXBLOCK-1);
if(start+i*4==pagelimit-4) done=1;
assert(start+i*4<pagelimit);
if (i==MAXBLOCK-1) done=1;
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;