u_int ba[MAXBLOCK];
char likely[MAXBLOCK];
char is_ds[MAXBLOCK];
+ char ooo[MAXBLOCK];
uint64_t unneeded_reg[MAXBLOCK];
uint64_t unneeded_reg_upper[MAXBLOCK];
uint64_t branch_unneeded_reg[MAXBLOCK];
signed char regmap[MAXBLOCK][HOST_REGS];
signed char regmap_entry[MAXBLOCK][HOST_REGS];
uint64_t constmap[MAXBLOCK][HOST_REGS];
- uint64_t known_value[HOST_REGS];
- u_int known_reg;
struct regstat regs[MAXBLOCK];
struct regstat branch_regs[MAXBLOCK];
+ signed char minimum_free_regs[MAXBLOCK];
u_int needed_reg[MAXBLOCK];
uint64_t requires_32bit[MAXBLOCK];
u_int wont_dirty[MAXBLOCK];
#define CSREG 35 // Coprocessor status
#define CCREG 36 // Cycle count
#define INVCP 37 // Pointer to invalid_code
-#define TEMPREG 38
-#define FTEMP 38 // FPU/LDL/LDR temporary register
-#define PTEMP 39 // Prefetch temporary register
-#define TLREG 40 // TLB mapping offset
-#define RHASH 41 // Return address hash
-#define RHTBL 42 // Return address hash table address
-#define RTEMP 43 // JR/JALR address register
-#define MAXREG 43
-#define AGEN1 44 // Address generation temporary register
-#define AGEN2 45 // Address generation temporary register
-#define MGEN1 46 // Maptable address generation temporary register
-#define MGEN2 47 // Maptable address generation temporary register
-#define BTREG 48 // Branch target temporary register
+#define MMREG 38 // Pointer to memory_map
+#define ROREG 39 // ram offset (if rdram!=0x80000000)
+#define TEMPREG 40
+#define FTEMP 40 // FPU temporary register
+#define PTEMP 41 // Prefetch temporary register
+#define TLREG 42 // TLB mapping offset
+#define RHASH 43 // Return address hash
+#define RHTBL 44 // Return address hash table address
+#define RTEMP 45 // JR/JALR address register
+#define MAXREG 45
+#define AGEN1 46 // Address generation temporary register
+#define AGEN2 47 // Address generation temporary register
+#define MGEN1 48 // Maptable address generation temporary register
+#define MGEN2 49 // Maptable address generation temporary register
+#define BTREG 50 // Branch target temporary register
/* instruction types */
#define NOP 0 // No operation
// 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);
{
inv_debug("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;
- }
+ #ifdef __arm__
+ needs_clear_cache[(host_addr-(u_int)BASE_ADDR)>>17]|=1<<(((host_addr-(u_int)BASE_ADDR)>>12)&31);
+ #endif
}
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)
{
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) {
while(head!=NULL) {
inv_debug("INVALIDATE: kill pointer to %x (%x)\n",head->vaddr,(int)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;
- }
+ #ifdef __arm__
+ needs_clear_cache[(host_addr-(u_int)BASE_ADDR)>>17]|=1<<(((host_addr-(u_int)BASE_ADDR)>>12)&31);
+ #endif
next=head->next;
free(head);
head=next;
}
- #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)
{
for(first=page+1;first<last;first++) {
invalidate_page(first);
}
+ #ifdef __arm__
+ do_clear_cache();
+ #endif
// Don't trap writes
invalid_code[block]=1;
{
invalidate_block(addr>>12);
}
+// This is called when loading a save state.
+// Anything could have changed, so invalidate everything.
void invalidate_all_pages()
{
u_int page,n;
if(rs1[i]) alloc_reg(current,i,rs1[i]);
if(rs2[i]) alloc_reg(current,i,rs2[i]);
alloc_reg(current,i,rt1[i]);
- if(rt1[i]==rs2[i]) alloc_reg_temp(current,i,-1);
+ if(rt1[i]==rs2[i]) {
+ alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
+ }
current->is32|=1LL<<rt1[i];
} else { // DSLLV/DSRLV/DSRAV
if(rs1[i]) alloc_reg64(current,i,rs1[i]);
alloc_reg64(current,i,rt1[i]);
current->is32&=~(1LL<<rt1[i]);
if(opcode2[i]==0x16||opcode2[i]==0x17) // DSRLV and DSRAV need a temporary register
+ {
alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
+ }
}
clear_const(current,rs1[i]);
clear_const(current,rs2[i]);
if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
if(rt1[i]) {
alloc_reg(current,i,rt1[i]);
+ if(get_reg(current->regmap,rt1[i])<0) {
+ // dummy load, but we still need a register to calculate the address
+ alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
+ }
if(opcode[i]==0x27||opcode[i]==0x37) // LWU/LD
{
current->is32&=~(1LL<<rt1[i]);
alloc_reg64(current,i,rt1[i]);
alloc_all(current,i);
alloc_reg64(current,i,FTEMP);
+ minimum_free_regs[i]=HOST_REGS;
}
else current->is32|=1LL<<rt1[i];
dirty_reg(current,rt1[i]);
{
alloc_reg(current,i,FTEMP);
alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
}
}
else
{
// Load to r0 (dummy load)
// but we still need a register to calculate the address
+ if(opcode[i]==0x22||opcode[i]==0x26)
+ {
+ alloc_reg(current,i,FTEMP); // LWL/LWR need another temporary
+ }
alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
+ if(opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR
+ {
+ alloc_all(current,i);
+ alloc_reg64(current,i,FTEMP);
+ minimum_free_regs[i]=HOST_REGS;
+ }
}
}
}
// We need a temporary register for address generation
alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
}
void c1ls_alloc(struct regstat *current,int i)
#endif
// We need a temporary register for address generation
alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
}
#ifndef multdiv_alloc
current->is32&=~(1LL<<LOREG);
dirty_reg(current,HIREG);
dirty_reg(current,LOREG);
+ minimum_free_regs[i]=HOST_REGS;
}
}
else
assert(opcode2[i]==0x10);
alloc_all(current,i);
}
+ minimum_free_regs[i]=HOST_REGS;
}
void cop1_alloc(struct regstat *current,int i)
alloc_reg_temp(current,i,-1);
}
}
+ minimum_free_regs[i]=1;
}
void fconv_alloc(struct regstat *current,int i)
{
alloc_reg(current,i,CSREG); // Load status
alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
}
void float_alloc(struct regstat *current,int i)
{
alloc_reg(current,i,CSREG); // Load status
alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
}
void c2op_alloc(struct regstat *current,int i)
{
alloc_reg(current,i,FSREG); // Load flags
dirty_reg(current,FSREG); // Flag will be modified
alloc_reg_temp(current,i,-1);
+ minimum_free_regs[i]=1;
}
void syscall_alloc(struct regstat *current,int i)
alloc_cc(current,i);
dirty_reg(current,CCREG);
alloc_all(current,i);
+ minimum_free_regs[i]=HOST_REGS;
current->isconst=0;
}
current->isconst=0;
current->wasconst=0;
regs[i].wasconst=0;
+ minimum_free_regs[i]=HOST_REGS;
alloc_all(current,i);
alloc_cc(current,i);
dirty_reg(current,CCREG);
#endif
if(offset||s<0||c) addr=tl;
else addr=s;
- if(tl>=0) {
- //assert(tl>=0);
- //assert(rt1[i]);
- reglist&=~(1<<tl);
- if(th>=0) reglist&=~(1<<th);
- if(!using_tlb) {
- if(!c) {
+ //if(tl<0) tl=get_reg(i_regs->regmap,-1);
+ if(tl>=0) {
+ //printf("load_assemble: c=%d\n",c);
+ //if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset);
+ assert(tl>=0); // Even if the load is a NOP, we must check for pagefaults and I/O
+ reglist&=~(1<<tl);
+ if(th>=0) reglist&=~(1<<th);
+ if(!using_tlb) {
+ if(!c) {
+ #ifdef RAM_OFFSET
+ map=get_reg(i_regs->regmap,ROREG);
+ if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG);
+ #endif
//#define R29_HACK 1
- #ifdef R29_HACK
- // Strmnnrmn's speed hack
- if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
+ #ifdef R29_HACK
+ // Strmnnrmn's speed hack
+ if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
+ #endif
+ {
+ 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
+ if(rs1[i]>=28)
+ emit_jno_unlikely(0);
+ else
#endif
- {
- 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
- if(rs1[i]>=28)
- emit_jno_unlikely(0);
- else
- #endif
- emit_jno(0);
- }
+ emit_jno(0);
}
- }else{ // using tlb
- int x=0;
- if (opcode[i]==0x20||opcode[i]==0x24) x=3; // LB/LBU
- if (opcode[i]==0x21||opcode[i]==0x25) x=2; // LH/LHU
- map=get_reg(i_regs->regmap,TLREG);
- assert(map>=0);
- map=do_tlb_r(addr,tl,map,x,-1,-1,c,constmap[i][s]+offset);
- do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr);
- }
- if (opcode[i]==0x20) { // LB
- if(!c||memtarget) {
+ }
+ }else{ // using tlb
+ int x=0;
+ if (opcode[i]==0x20||opcode[i]==0x24) x=3; // LB/LBU
+ if (opcode[i]==0x21||opcode[i]==0x25) x=2; // LH/LHU
+ map=get_reg(i_regs->regmap,TLREG);
+ assert(map>=0);
+ map=do_tlb_r(addr,tl,map,x,-1,-1,c,constmap[i][s]+offset);
+ do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr);
+ }
+ int dummy=(rt1[i]==0)||(tl!=get_reg(i_regs->regmap,rt1[i])); // ignore loads to r0 and unneeded reg
+ if (opcode[i]==0x20) { // LB
+ if(!c||memtarget) {
+ if(!dummy) {
#ifdef HOST_IMM_ADDR32
if(c)
emit_movsbl_tlb((constmap[i][s]+offset)^3,map,tl);
//emit_xorimm(addr,3,tl);
//gen_tlb_addr_r(tl,map);
//emit_movsbl_indexed((int)rdram-0x80000000,tl,tl);
- int x=0;
+ int x=0,a=tl;
#ifdef BIG_ENDIAN_MIPS
if(!c) emit_xorimm(addr,3,tl);
else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
#else
- if(c) x=(constmap[i][s]+offset)-(constmap[i][s]+offset);
- else if (tl!=addr) emit_mov(addr,tl);
+ if(!c) a=addr;
#endif
- emit_movsbl_indexed_tlb(x,tl,map,tl);
+ emit_movsbl_indexed_tlb(x,a,map,tl);
}
- if(jaddr)
- add_stub(LOADB_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- else
- inline_readstub(LOADB_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ if(jaddr)
+ add_stub(LOADB_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- if (opcode[i]==0x21) { // LH
- if(!c||memtarget) {
+ else
+ inline_readstub(LOADB_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ }
+ if (opcode[i]==0x21) { // LH
+ if(!c||memtarget) {
+ if(!dummy) {
#ifdef HOST_IMM_ADDR32
if(c)
emit_movswl_tlb((constmap[i][s]+offset)^2,map,tl);
else
#endif
{
- int x=0;
+ int x=0,a=tl;
#ifdef BIG_ENDIAN_MIPS
if(!c) emit_xorimm(addr,2,tl);
else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
#else
- if(c) x=(constmap[i][s]+offset)-(constmap[i][s]+offset);
- else if (tl!=addr) emit_mov(addr,tl);
+ if(!c) a=addr;
#endif
//#ifdef
//emit_movswl_indexed_tlb(x,tl,map,tl);
//else
if(map>=0) {
- gen_tlb_addr_r(tl,map);
- emit_movswl_indexed(x,tl,tl);
- }else
- emit_movswl_indexed((int)rdram-0x80000000+x,tl,tl);
+ gen_tlb_addr_r(a,map);
+ emit_movswl_indexed(x,a,tl);
+ }else{
+ #ifdef RAM_OFFSET
+ emit_movswl_indexed(x,a,tl);
+ #else
+ emit_movswl_indexed((int)rdram-0x80000000+x,a,tl);
+ #endif
+ }
}
- if(jaddr)
- add_stub(LOADH_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- else
- inline_readstub(LOADH_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ if(jaddr)
+ add_stub(LOADH_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- if (opcode[i]==0x23) { // LW
- if(!c||memtarget) {
+ else
+ inline_readstub(LOADH_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ }
+ if (opcode[i]==0x23) { // LW
+ if(!c||memtarget) {
+ if(!dummy) {
//emit_readword_indexed((int)rdram-0x80000000,addr,tl);
#ifdef HOST_IMM_ADDR32
if(c)
else
#endif
emit_readword_indexed_tlb(0,addr,map,tl);
- if(jaddr)
- add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- else
- inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ if(jaddr)
+ add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- if (opcode[i]==0x24) { // LBU
- if(!c||memtarget) {
+ else
+ inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ }
+ if (opcode[i]==0x24) { // LBU
+ if(!c||memtarget) {
+ if(!dummy) {
#ifdef HOST_IMM_ADDR32
if(c)
emit_movzbl_tlb((constmap[i][s]+offset)^3,map,tl);
//emit_xorimm(addr,3,tl);
//gen_tlb_addr_r(tl,map);
//emit_movzbl_indexed((int)rdram-0x80000000,tl,tl);
- int x=0;
+ int x=0,a=tl;
#ifdef BIG_ENDIAN_MIPS
if(!c) emit_xorimm(addr,3,tl);
else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
#else
- if(c) x=(constmap[i][s]+offset)-(constmap[i][s]+offset);
- else if (tl!=addr) emit_mov(addr,tl);
+ if(!c) a=addr;
#endif
- emit_movzbl_indexed_tlb(x,tl,map,tl);
+ emit_movzbl_indexed_tlb(x,a,map,tl);
}
- if(jaddr)
- add_stub(LOADBU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- else
- inline_readstub(LOADBU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ if(jaddr)
+ add_stub(LOADBU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- if (opcode[i]==0x25) { // LHU
- if(!c||memtarget) {
+ else
+ inline_readstub(LOADBU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ }
+ if (opcode[i]==0x25) { // LHU
+ if(!c||memtarget) {
+ if(!dummy) {
#ifdef HOST_IMM_ADDR32
if(c)
emit_movzwl_tlb((constmap[i][s]+offset)^2,map,tl);
else
#endif
{
- int x=0;
+ int x=0,a=tl;
#ifdef BIG_ENDIAN_MIPS
if(!c) emit_xorimm(addr,2,tl);
else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
#else
- if(c) x=(constmap[i][s]+offset)-(constmap[i][s]+offset);
- else if (tl!=addr) emit_mov(addr,tl);
+ if(!c) a=addr;
#endif
//#ifdef
//emit_movzwl_indexed_tlb(x,tl,map,tl);
//#else
if(map>=0) {
- gen_tlb_addr_r(tl,map);
- emit_movzwl_indexed(x,tl,tl);
- }else
- emit_movzwl_indexed((int)rdram-0x80000000+x,tl,tl);
- if(jaddr)
- add_stub(LOADHU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+ gen_tlb_addr_r(a,map);
+ emit_movzwl_indexed(x,a,tl);
+ }else{
+ #ifdef RAM_OFFSET
+ emit_movzwl_indexed(x,a,tl);
+ #else
+ emit_movzwl_indexed((int)rdram-0x80000000+x,a,tl);
+ #endif
+ }
}
}
- else
- inline_readstub(LOADHU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ if(jaddr)
+ add_stub(LOADHU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- if (opcode[i]==0x27) { // LWU
- assert(th>=0);
- if(!c||memtarget) {
+ else
+ inline_readstub(LOADHU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ }
+ if (opcode[i]==0x27) { // LWU
+ assert(th>=0);
+ if(!c||memtarget) {
+ if(!dummy) {
//emit_readword_indexed((int)rdram-0x80000000,addr,tl);
#ifdef HOST_IMM_ADDR32
if(c)
else
#endif
emit_readword_indexed_tlb(0,addr,map,tl);
- if(jaddr)
- add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- else {
- inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
- }
- emit_zeroreg(th);
+ if(jaddr)
+ add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+ }
+ else {
+ inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
}
- if (opcode[i]==0x37) { // LD
- if(!c||memtarget) {
+ emit_zeroreg(th);
+ }
+ if (opcode[i]==0x37) { // LD
+ if(!c||memtarget) {
+ if(!dummy) {
//gen_tlb_addr_r(tl,map);
//if(th>=0) emit_readword_indexed((int)rdram-0x80000000,addr,th);
//emit_readword_indexed((int)rdram-0x7FFFFFFC,addr,tl);
else
#endif
emit_readdword_indexed_tlb(0,addr,map,th,tl);
- if(jaddr)
- add_stub(LOADD_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- else
- inline_readstub(LOADD_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ if(jaddr)
+ add_stub(LOADD_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
}
- //emit_storereg(rt1[i],tl); // DEBUG
+ else
+ inline_readstub(LOADD_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
}
+ }
+ //emit_storereg(rt1[i],tl); // DEBUG
//if(opcode[i]==0x23)
//if(opcode[i]==0x24)
//if(opcode[i]==0x23||opcode[i]==0x24)
}
type=STORED_STUB;
}
- if(!using_tlb&&(!c||memtarget))
- // addr could be a temp, make sure it survives STORE*_STUB
- reglist|=1<<addr;
- if(jaddr) {
- add_stub(type,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
- } else if(!memtarget) {
- inline_writestub(type,i,constmap[i][s]+offset,i_regs->regmap,rs2[i],ccadj[i],reglist);
- }
if(!using_tlb) {
if(!c||memtarget) {
#ifdef DESTRUCTIVE_SHIFT
#else
emit_cmpmem_indexedsr12_imm((int)invalid_code,addr,1);
#endif
+ #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
+ emit_callne(invalidate_addr_reg[addr]);
+ #else
jaddr2=(int)out;
emit_jne(0);
add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),addr,0,0,0);
+ #endif
}
}
+ if(jaddr) {
+ add_stub(type,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+ } else if(c&&!memtarget) {
+ inline_writestub(type,i,constmap[i][s]+offset,i_regs->regmap,rs2[i],ccadj[i],reglist);
+ }
//if(opcode[i]==0x2B || opcode[i]==0x3F)
//if(opcode[i]==0x2B || opcode[i]==0x28)
//if(opcode[i]==0x2B || opcode[i]==0x29)
for(hr=0;hr<HOST_REGS;hr++) {
if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
}
- if(tl>=0) {
- assert(temp>=0);
- if(!using_tlb) {
- if(!c) {
- emit_cmpimm(s<0||offset?temp:s,RAM_SIZE);
- if(!offset&&s!=temp) emit_mov(s,temp);
- jaddr=(int)out;
- emit_jno(0);
- }
- else
- {
- if(!memtarget||!rs1[i]) {
- jaddr=(int)out;
- emit_jmp(0);
- }
- }
- if((u_int)rdram!=0x80000000)
- emit_addimm_no_flags((u_int)rdram-(u_int)0x80000000,temp);
- }else{ // using tlb
- int map=get_reg(i_regs->regmap,TLREG);
- assert(map>=0);
- map=do_tlb_w(c||s<0||offset?temp:s,temp,map,0,c,constmap[i][s]+offset);
- if(!c&&!offset&&s>=0) emit_mov(s,temp);
- do_tlb_w_branch(map,c,constmap[i][s]+offset,&jaddr);
- if(!jaddr&&!memtarget) {
+ assert(temp>=0);
+ if(!using_tlb) {
+ if(!c) {
+ emit_cmpimm(s<0||offset?temp:s,RAM_SIZE);
+ if(!offset&&s!=temp) emit_mov(s,temp);
+ jaddr=(int)out;
+ emit_jno(0);
+ }
+ else
+ {
+ if(!memtarget||!rs1[i]) {
jaddr=(int)out;
emit_jmp(0);
}
- gen_tlb_addr_w(temp,map);
}
-
- if (opcode[i]==0x2C||opcode[i]==0x2D) { // SDL/SDR
- temp2=get_reg(i_regs->regmap,FTEMP);
- if(!rs2[i]) temp2=th=tl;
+ #ifdef RAM_OFFSET
+ int map=get_reg(i_regs->regmap,ROREG);
+ if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG);
+ gen_tlb_addr_w(temp,map);
+ #else
+ if((u_int)rdram!=0x80000000)
+ emit_addimm_no_flags((u_int)rdram-(u_int)0x80000000,temp);
+ #endif
+ }else{ // using tlb
+ int map=get_reg(i_regs->regmap,TLREG);
+ assert(map>=0);
+ map=do_tlb_w(c||s<0||offset?temp:s,temp,map,0,c,constmap[i][s]+offset);
+ if(!c&&!offset&&s>=0) emit_mov(s,temp);
+ do_tlb_w_branch(map,c,constmap[i][s]+offset,&jaddr);
+ if(!jaddr&&!memtarget) {
+ jaddr=(int)out;
+ emit_jmp(0);
}
+ gen_tlb_addr_w(temp,map);
+ }
+
+ if (opcode[i]==0x2C||opcode[i]==0x2D) { // SDL/SDR
+ temp2=get_reg(i_regs->regmap,FTEMP);
+ if(!rs2[i]) temp2=th=tl;
+ }
#ifndef BIG_ENDIAN_MIPS
emit_xorimm(temp,3,temp);
#endif
- emit_testimm(temp,2);
- case2=(int)out;
- emit_jne(0);
- emit_testimm(temp,1);
- case1=(int)out;
- emit_jne(0);
- // 0
- if (opcode[i]==0x2A) { // SWL
- emit_writeword_indexed(tl,0,temp);
- }
- if (opcode[i]==0x2E) { // SWR
- emit_writebyte_indexed(tl,3,temp);
- }
- if (opcode[i]==0x2C) { // SDL
- emit_writeword_indexed(th,0,temp);
- if(rs2[i]) emit_mov(tl,temp2);
- }
- if (opcode[i]==0x2D) { // SDR
- emit_writebyte_indexed(tl,3,temp);
- if(rs2[i]) emit_shldimm(th,tl,24,temp2);
- }
+ emit_testimm(temp,2);
+ case2=(int)out;
+ emit_jne(0);
+ emit_testimm(temp,1);
+ case1=(int)out;
+ emit_jne(0);
+ // 0
+ if (opcode[i]==0x2A) { // SWL
+ emit_writeword_indexed(tl,0,temp);
+ }
+ if (opcode[i]==0x2E) { // SWR
+ emit_writebyte_indexed(tl,3,temp);
+ }
+ if (opcode[i]==0x2C) { // SDL
+ emit_writeword_indexed(th,0,temp);
+ if(rs2[i]) emit_mov(tl,temp2);
+ }
+ if (opcode[i]==0x2D) { // SDR
+ emit_writebyte_indexed(tl,3,temp);
+ if(rs2[i]) emit_shldimm(th,tl,24,temp2);
+ }
+ done0=(int)out;
+ emit_jmp(0);
+ // 1
+ set_jump_target(case1,(int)out);
+ if (opcode[i]==0x2A) { // SWL
+ // Write 3 msb into three least significant bytes
+ if(rs2[i]) emit_rorimm(tl,8,tl);
+ emit_writehword_indexed(tl,-1,temp);
+ if(rs2[i]) emit_rorimm(tl,16,tl);
+ emit_writebyte_indexed(tl,1,temp);
+ if(rs2[i]) emit_rorimm(tl,8,tl);
+ }
+ if (opcode[i]==0x2E) { // SWR
+ // Write two lsb into two most significant bytes
+ emit_writehword_indexed(tl,1,temp);
+ }
+ if (opcode[i]==0x2C) { // SDL
+ if(rs2[i]) emit_shrdimm(tl,th,8,temp2);
+ // Write 3 msb into three least significant bytes
+ if(rs2[i]) emit_rorimm(th,8,th);
+ emit_writehword_indexed(th,-1,temp);
+ if(rs2[i]) emit_rorimm(th,16,th);
+ emit_writebyte_indexed(th,1,temp);
+ if(rs2[i]) emit_rorimm(th,8,th);
+ }
+ if (opcode[i]==0x2D) { // SDR
+ if(rs2[i]) emit_shldimm(th,tl,16,temp2);
+ // Write two lsb into two most significant bytes
+ emit_writehword_indexed(tl,1,temp);
+ }
+ done1=(int)out;
+ emit_jmp(0);
+ // 2
+ set_jump_target(case2,(int)out);
+ emit_testimm(temp,1);
+ case3=(int)out;
+ emit_jne(0);
+ if (opcode[i]==0x2A) { // SWL
+ // Write two msb into two least significant bytes
+ if(rs2[i]) emit_rorimm(tl,16,tl);
+ emit_writehword_indexed(tl,-2,temp);
+ if(rs2[i]) emit_rorimm(tl,16,tl);
+ }
+ if (opcode[i]==0x2E) { // SWR
+ // Write 3 lsb into three most significant bytes
+ emit_writebyte_indexed(tl,-1,temp);
+ if(rs2[i]) emit_rorimm(tl,8,tl);
+ emit_writehword_indexed(tl,0,temp);
+ if(rs2[i]) emit_rorimm(tl,24,tl);
+ }
+ if (opcode[i]==0x2C) { // SDL
+ if(rs2[i]) emit_shrdimm(tl,th,16,temp2);
+ // Write two msb into two least significant bytes
+ if(rs2[i]) emit_rorimm(th,16,th);
+ emit_writehword_indexed(th,-2,temp);
+ if(rs2[i]) emit_rorimm(th,16,th);
+ }
+ if (opcode[i]==0x2D) { // SDR
+ if(rs2[i]) emit_shldimm(th,tl,8,temp2);
+ // Write 3 lsb into three most significant bytes
+ emit_writebyte_indexed(tl,-1,temp);
+ if(rs2[i]) emit_rorimm(tl,8,tl);
+ emit_writehword_indexed(tl,0,temp);
+ if(rs2[i]) emit_rorimm(tl,24,tl);
+ }
+ done2=(int)out;
+ emit_jmp(0);
+ // 3
+ set_jump_target(case3,(int)out);
+ if (opcode[i]==0x2A) { // SWL
+ // Write msb into least significant byte
+ if(rs2[i]) emit_rorimm(tl,24,tl);
+ emit_writebyte_indexed(tl,-3,temp);
+ if(rs2[i]) emit_rorimm(tl,8,tl);
+ }
+ if (opcode[i]==0x2E) { // SWR
+ // Write entire word
+ emit_writeword_indexed(tl,-3,temp);
+ }
+ if (opcode[i]==0x2C) { // SDL
+ if(rs2[i]) emit_shrdimm(tl,th,24,temp2);
+ // Write msb into least significant byte
+ if(rs2[i]) emit_rorimm(th,24,th);
+ emit_writebyte_indexed(th,-3,temp);
+ if(rs2[i]) emit_rorimm(th,8,th);
+ }
+ if (opcode[i]==0x2D) { // SDR
+ if(rs2[i]) emit_mov(th,temp2);
+ // Write entire word
+ emit_writeword_indexed(tl,-3,temp);
+ }
+ set_jump_target(done0,(int)out);
+ set_jump_target(done1,(int)out);
+ set_jump_target(done2,(int)out);
+ if (opcode[i]==0x2C) { // SDL
+ emit_testimm(temp,4);
done0=(int)out;
- emit_jmp(0);
- // 1
- set_jump_target(case1,(int)out);
- if (opcode[i]==0x2A) { // SWL
- // Write 3 msb into three least significant bytes
- if(rs2[i]) emit_rorimm(tl,8,tl);
- emit_writehword_indexed(tl,-1,temp);
- if(rs2[i]) emit_rorimm(tl,16,tl);
- emit_writebyte_indexed(tl,1,temp);
- if(rs2[i]) emit_rorimm(tl,8,tl);
- }
- if (opcode[i]==0x2E) { // SWR
- // Write two lsb into two most significant bytes
- emit_writehword_indexed(tl,1,temp);
- }
- if (opcode[i]==0x2C) { // SDL
- if(rs2[i]) emit_shrdimm(tl,th,8,temp2);
- // Write 3 msb into three least significant bytes
- if(rs2[i]) emit_rorimm(th,8,th);
- emit_writehword_indexed(th,-1,temp);
- if(rs2[i]) emit_rorimm(th,16,th);
- emit_writebyte_indexed(th,1,temp);
- if(rs2[i]) emit_rorimm(th,8,th);
- }
- if (opcode[i]==0x2D) { // SDR
- if(rs2[i]) emit_shldimm(th,tl,16,temp2);
- // Write two lsb into two most significant bytes
- emit_writehword_indexed(tl,1,temp);
- }
- done1=(int)out;
- emit_jmp(0);
- // 2
- set_jump_target(case2,(int)out);
- emit_testimm(temp,1);
- case3=(int)out;
emit_jne(0);
- if (opcode[i]==0x2A) { // SWL
- // Write two msb into two least significant bytes
- if(rs2[i]) emit_rorimm(tl,16,tl);
- emit_writehword_indexed(tl,-2,temp);
- if(rs2[i]) emit_rorimm(tl,16,tl);
- }
- if (opcode[i]==0x2E) { // SWR
- // Write 3 lsb into three most significant bytes
- emit_writebyte_indexed(tl,-1,temp);
- if(rs2[i]) emit_rorimm(tl,8,tl);
- emit_writehword_indexed(tl,0,temp);
- if(rs2[i]) emit_rorimm(tl,24,tl);
- }
- if (opcode[i]==0x2C) { // SDL
- if(rs2[i]) emit_shrdimm(tl,th,16,temp2);
- // Write two msb into two least significant bytes
- if(rs2[i]) emit_rorimm(th,16,th);
- emit_writehword_indexed(th,-2,temp);
- if(rs2[i]) emit_rorimm(th,16,th);
- }
- if (opcode[i]==0x2D) { // SDR
- if(rs2[i]) emit_shldimm(th,tl,8,temp2);
- // Write 3 lsb into three most significant bytes
- emit_writebyte_indexed(tl,-1,temp);
- if(rs2[i]) emit_rorimm(tl,8,tl);
- emit_writehword_indexed(tl,0,temp);
- if(rs2[i]) emit_rorimm(tl,24,tl);
- }
- done2=(int)out;
- emit_jmp(0);
- // 3
- set_jump_target(case3,(int)out);
- if (opcode[i]==0x2A) { // SWL
- // Write msb into least significant byte
- if(rs2[i]) emit_rorimm(tl,24,tl);
- emit_writebyte_indexed(tl,-3,temp);
- if(rs2[i]) emit_rorimm(tl,8,tl);
- }
- if (opcode[i]==0x2E) { // SWR
- // Write entire word
- emit_writeword_indexed(tl,-3,temp);
- }
- if (opcode[i]==0x2C) { // SDL
- if(rs2[i]) emit_shrdimm(tl,th,24,temp2);
- // Write msb into least significant byte
- if(rs2[i]) emit_rorimm(th,24,th);
- emit_writebyte_indexed(th,-3,temp);
- if(rs2[i]) emit_rorimm(th,8,th);
- }
- if (opcode[i]==0x2D) { // SDR
- if(rs2[i]) emit_mov(th,temp2);
- // Write entire word
- emit_writeword_indexed(tl,-3,temp);
- }
+ emit_andimm(temp,~3,temp);
+ emit_writeword_indexed(temp2,4,temp);
set_jump_target(done0,(int)out);
- set_jump_target(done1,(int)out);
- set_jump_target(done2,(int)out);
- if (opcode[i]==0x2C) { // SDL
- emit_testimm(temp,4);
- done0=(int)out;
- emit_jne(0);
- emit_andimm(temp,~3,temp);
- emit_writeword_indexed(temp2,4,temp);
- set_jump_target(done0,(int)out);
- }
- if (opcode[i]==0x2D) { // SDR
- emit_testimm(temp,4);
- done0=(int)out;
- emit_jeq(0);
- emit_andimm(temp,~3,temp);
- emit_writeword_indexed(temp2,-4,temp);
- set_jump_target(done0,(int)out);
- }
- if(!c||!memtarget)
- add_stub(STORELR_STUB,jaddr,(int)out,i,(int)i_regs,temp,ccadj[i],reglist);
}
+ if (opcode[i]==0x2D) { // SDR
+ emit_testimm(temp,4);
+ done0=(int)out;
+ emit_jeq(0);
+ emit_andimm(temp,~3,temp);
+ emit_writeword_indexed(temp2,-4,temp);
+ set_jump_target(done0,(int)out);
+ }
+ if(!c||!memtarget)
+ add_stub(STORELR_STUB,jaddr,(int)out,i,(int)i_regs,temp,ccadj[i],reglist);
if(!using_tlb) {
+ #ifdef RAM_OFFSET
+ int map=get_reg(i_regs->regmap,ROREG);
+ if(map<0) map=HOST_TEMPREG;
+ gen_orig_addr_w(temp,map);
+ #else
emit_addimm_no_flags((u_int)0x80000000-(u_int)rdram,temp);
+ #endif
#if defined(HOST_IMM8)
int ir=get_reg(i_regs->regmap,INVCP);
assert(ir>=0);
#else
emit_cmpmem_indexedsr12_imm((int)invalid_code,temp,1);
#endif
+ #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
+ emit_callne(invalidate_addr_reg[temp]);
+ #else
jaddr2=(int)out;
emit_jne(0);
add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),temp,0,0,0);
+ #endif
}
/*
emit_pusha();
#else
emit_cmpmem_indexedsr12_imm((int)invalid_code,temp,1);
#endif
+ #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
+ emit_callne(invalidate_addr_reg[temp]);
+ #else
jaddr3=(int)out;
emit_jne(0);
add_stub(INVCODE_STUB,jaddr3,(int)out,reglist|(1<<HOST_CCREG),temp,0,0,0);
+ #endif
}
}
if(jaddr2) add_stub(type,jaddr2,(int)out,i,offset||c||s<0?ar:s,(int)i_regs,ccadj[i],reglist);
#else
emit_cmpmem_indexedsr12_imm((int)invalid_code,ar,1);
#endif
+ #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
+ emit_callne(invalidate_addr_reg[ar]);
+ #else
jaddr3=(int)out;
emit_jne(0);
add_stub(INVCODE_STUB,jaddr3,(int)out,reglist|(1<<HOST_CCREG),ar,0,0,0);
+ #endif
}
if (opcode[i]==0x32) { // LWC2
cop2_put_dreg(copr,tl,HOST_TEMPREG);
{
//if(opcode2[i]==0x10||opcode2[i]==0x12) { // MFHI/MFLO
//if(opcode2[i]==0x11||opcode2[i]==0x13) { // MTHI/MTLO
- //assert(rt1[i]>0);
if(rt1[i]) {
signed char sh,sl,th,tl;
th=get_reg(i_regs->regmap,rt1[i]|64);
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;
}
int mgr=MGEN1+(i&1);
if(itype[i]==LOAD) {
ra=get_reg(i_regs->regmap,rt1[i]);
- //if(rt1[i]) assert(ra>=0);
+ if(ra<0) ra=get_reg(i_regs->regmap,-1);
+ assert(ra>=0);
}
if(itype[i]==LOADLR) {
ra=get_reg(i_regs->regmap,FTEMP);
}
}
//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
int prev_cop1_usable=cop1_usable;
int unconditional=0,nop=0;
int only32=0;
- int ooo=1;
int invert=0;
int internal=internal_branch(branch_regs[i].is32,ba[i]);
if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
- if(likely[i]) ooo=0;
if(!match) invert=1;
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
if(i>(ba[i]-start)>>2) invert=1;
#endif
-
- if(ooo)
- if((rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))||
- (rs2[i]&&(rs2[i]==rt1[i+1]||rs2[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;
- }
-
- if(ooo) {
+
+ if(ooo[i]) {
s1l=get_reg(branch_regs[i].regmap,rs1[i]);
s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
s2l=get_reg(branch_regs[i].regmap,rs2[i]);
only32=(regs[i].was32>>rs1[i])&(regs[i].was32>>rs2[i])&1;
}
- if(ooo) {
+ if(ooo[i]) {
// Out of order execution (delay slot first)
//printf("OOOE\n");
address_generation(i+1,i_regs,regs[i].regmap_entry);
int prev_cop1_usable=cop1_usable;
int unconditional=0,nevertaken=0;
int only32=0;
- int ooo=1;
int invert=0;
int internal=internal_branch(branch_regs[i].is32,ba[i]);
if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
- if(likely[i]) ooo=0;
if(!match) invert=1;
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
if(i>(ba[i]-start)>>2) invert=1;
//if(opcode2[i]>=0x10) return; // 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]))
- {
- // 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;
- }
-
- if(ooo) {
+ if(ooo[i]) {
s1l=get_reg(branch_regs[i].regmap,rs1[i]);
s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
}
only32=(regs[i].was32>>rs1[i])&1;
}
- if(ooo) {
+ if(ooo[i]) {
// Out of order execution (delay slot first)
//printf("OOOE\n");
address_generation(i+1,i_regs,regs[i].regmap_entry);
assem_debug("fmatch=%d\n",match);
int fs,cs;
int eaddr;
- int ooo=1;
int invert=0;
int internal=internal_branch(branch_regs[i].is32,ba[i]);
if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
- if(likely[i]) ooo=0;
if(!match) invert=1;
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
if(i>(ba[i]-start)>>2) invert=1;
#endif
- if(ooo)
- if(itype[i+1]==FCOMP)
- {
- // Write-after-read dependency prevents out of order execution
- // First test branch condition, then execute delay slot, then branch
- ooo=0;
- }
-
- if(ooo) {
+ if(ooo[i]) {
fs=get_reg(branch_regs[i].regmap,FSREG);
address_generation(i+1,i_regs,regs[i].regmap_entry); // Is this okay?
}
cop1_usable=1;
}
- if(ooo) {
+ if(ooo[i]) {
// Out of order execution (delay slot first)
//printf("OOOE\n");
ds_assemble(i+1,i_regs);
/* Pass 1 disassembly */
for(i=0;!done;i++) {
- bt[i]=0;likely[i]=0;op2=0;
+ bt[i]=0;likely[i]=0;ooo[i]=0;op2=0;
+ minimum_free_regs[i]=0;
opcode[i]=op=source[i]>>26;
switch(op)
{
#endif
//current.is32|=1LL<<rt1[i];
}
+ ooo[i]=1;
delayslot_alloc(¤t,i+1);
//current.isconst=0; // DEBUG
ds=1;
alloc_reg(¤t,i,RTEMP);
}
//current.isconst=0; // DEBUG
+ ooo[i]=1;
ds=1;
break;
case CJUMP:
(rs2[i]&&(rs2[i]==rt1[i+1]||rs2[i]==rt2[i+1]))) {
// The delay slot overwrites one of our conditions.
// Allocate the branch condition registers instead.
- // Note that such a sequence of instructions could
- // be considered a bug since the branch can not be
- // re-executed if an exception occurs.
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
if(rs2[i]) alloc_reg64(¤t,i,rs2[i]);
}
}
- else delayslot_alloc(¤t,i+1);
+ else
+ {
+ ooo[i]=1;
+ delayslot_alloc(¤t,i+1);
+ }
}
else
if((opcode[i]&0x3E)==6) // BLEZ/BGTZ
if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) {
// The delay slot overwrites one of our conditions.
// Allocate the branch condition registers instead.
- // Note that such a sequence of instructions could
- // be considered a bug since the branch can not be
- // re-executed if an exception occurs.
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
if(rs1[i]) alloc_reg64(¤t,i,rs1[i]);
}
}
- else delayslot_alloc(¤t,i+1);
+ else
+ {
+ ooo[i]=1;
+ delayslot_alloc(¤t,i+1);
+ }
}
else
// Don't alloc the delay slot yet because we might not execute it
//#endif
//current.is32|=1LL<<rt1[i];
}
- if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) {
- // The delay slot overwrites the branch condition.
+ if((rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) // The delay slot overwrites the branch condition.
+ ||(rt1[i]==31&&(rs1[i+1]==31||rs2[i+1]==31||rt1[i+1]==31||rt2[i+1]==31))) { // DS touches $ra
// Allocate the branch condition registers instead.
- // Note that such a sequence of instructions could
- // be considered a bug since the branch can not be
- // re-executed if an exception occurs.
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
if(rs1[i]) alloc_reg64(¤t,i,rs1[i]);
}
}
- else delayslot_alloc(¤t,i+1);
+ else
+ {
+ ooo[i]=1;
+ delayslot_alloc(¤t,i+1);
+ }
}
else
// Don't alloc the delay slot yet because we might not execute it
if(itype[i+1]==FCOMP) {
// The delay slot overwrites the branch condition.
// Allocate the branch condition registers instead.
- // Note that such a sequence of instructions could
- // be considered a bug since the branch can not be
- // re-executed if an exception occurs.
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
alloc_reg(¤t,i,CSREG);
alloc_reg(¤t,i,FSREG);
}
else {
+ ooo[i]=1;
delayslot_alloc(¤t,i+1);
alloc_reg(¤t,i+1,CSREG);
}
{
cc=0;
}
+#ifdef PCSX
+ else if(/*itype[i]==LOAD||*/itype[i]==STORE||itype[i]==C1LS) // load causes weird timing issues
+ {
+ cc+=2; // 2 cycle penalty (after CLOCK_DIVIDER)
+ }
+ else if(itype[i]==C2LS)
+ {
+ cc+=4;
+ }
+#endif
else
{
cc++;
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
- ||itype[i+1]==COP2||itype[i+1]==C2LS||itype[i+1]==C2OP)
- {
- // Test both in case the delay slot is ooo,
- // could be done better...
- if(count_free_regs(branch_regs[i].regmap)<2
- ||count_free_regs(regs[i].regmap)<2)
+ if(ooo[i]) {
+ if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i+1])
+ f_regmap[hr]=branch_regs[i].regmap[hr];
+ }else{
+ if(count_free_regs(branch_regs[i].regmap)<=minimum_free_regs[i+1])
f_regmap[hr]=branch_regs[i].regmap[hr];
}
// Avoid dirty->clean transition
- // #ifdef DESTRUCTIVE_WRITEBACK here?
+ #ifdef DESTRUCTIVE_WRITEBACK
if(t>0) if(get_reg(regmap_pre[t],f_regmap[hr])>=0) if((regs[t].wasdirty>>get_reg(regmap_pre[t],f_regmap[hr]))&1) f_regmap[hr]=-1;
+ #endif
+ // This check is only strictly required in the DESTRUCTIVE_WRITEBACK
+ // case above, however it's always a good idea. We can't hoist the
+ // load if the register was already allocated, so there's no point
+ // wasting time analyzing most of these cases. It only "succeeds"
+ // when the mapping was different and the load can be replaced with
+ // a mov, which is of negligible benefit. So such cases are
+ // skipped below.
if(f_regmap[hr]>0) {
- if(regs[t].regmap_entry[hr]<0) {
+ if(regs[t].regmap_entry[hr]<0&&get_reg(regmap_pre[t],f_regmap[hr])<0) {
int r=f_regmap[hr];
for(j=t;j<=i;j++)
{
// register is lower numbered than the lower-half
// register. Not sure if it's worth fixing...
if(get_reg(regs[j].regmap,r&63)<0) break;
+ if(get_reg(regs[j].regmap_entry,r&63)<0) break;
if(regs[j].is32&(1LL<<(r&63))) break;
}
if(regs[j].regmap[hr]==f_regmap[hr]&&(f_regmap[hr]&63)<TEMPREG) {
}
k=i;
while(k>1&®s[k-1].regmap[hr]==-1) {
- if(itype[k-1]==STORE||itype[k-1]==STORELR
- ||itype[k-1]==C1LS||itype[k-1]==SHIFT||itype[k-1]==COP1
- ||itype[k-1]==FLOAT||itype[k-1]==FCONV||itype[k-1]==FCOMP
- ||itype[k-1]==COP2||itype[k-1]==C2LS||itype[k-1]==C2OP) {
- if(count_free_regs(regs[k-1].regmap)<2) {
- //printf("no free regs for store %x\n",start+(k-1)*4);
- break;
- }
+ if(count_free_regs(regs[k-1].regmap)<=minimum_free_regs[k-1]) {
+ //printf("no free regs for store %x\n",start+(k-1)*4);
+ break;
}
- else
- if(itype[k-1]!=NOP&&itype[k-1]!=MOV&&itype[k-1]!=ALU&&itype[k-1]!=SHIFTIMM&&itype[k-1]!=IMM16&&itype[k-1]!=LOAD) break;
if(get_reg(regs[k-1].regmap,f_regmap[hr])>=0) {
//printf("no-match due to different register\n");
break;
}
}
for(k=t;k<j;k++) {
+ // Alloc register clean at beginning of loop,
+ // but may dirty it in pass 6
regs[k].regmap_entry[hr]=f_regmap[hr];
regs[k].regmap[hr]=f_regmap[hr];
- regmap_pre[k+1][hr]=f_regmap[hr];
- regs[k+1].wasdirty&=~(1<<hr);
regs[k].dirty&=~(1<<hr);
regs[k].wasconst&=~(1<<hr);
regs[k].isconst&=~(1<<hr);
+ if(itype[k]==UJUMP||itype[k]==RJUMP||itype[k]==CJUMP||itype[k]==SJUMP||itype[k]==FJUMP) {
+ branch_regs[k].regmap_entry[hr]=f_regmap[hr];
+ branch_regs[k].regmap[hr]=f_regmap[hr];
+ branch_regs[k].dirty&=~(1<<hr);
+ branch_regs[k].wasconst&=~(1<<hr);
+ branch_regs[k].isconst&=~(1<<hr);
+ if(itype[k]!=RJUMP&&itype[k]!=UJUMP&&(source[k]>>16)!=0x1000) {
+ regmap_pre[k+2][hr]=f_regmap[hr];
+ regs[k+2].wasdirty&=~(1<<hr);
+ assert((branch_regs[k].is32&(1LL<<f_regmap[hr]))==
+ (regs[k+2].was32&(1LL<<f_regmap[hr])));
+ }
+ }
+ else
+ {
+ regmap_pre[k+1][hr]=f_regmap[hr];
+ regs[k+1].wasdirty&=~(1<<hr);
+ }
}
if(regs[j].regmap[hr]==f_regmap[hr])
regs[j].regmap_entry[hr]=f_regmap[hr];
//printf("32/64 mismatch %x %d\n",start+j*4,hr);
break;
}
- if(itype[j]==STORE||itype[j]==STORELR||itype[j]==C1LS
- ||itype[j]==SHIFT||itype[j]==COP1||itype[j]==FLOAT
- ||itype[j]==FCOMP||itype[j]==FCONV
- ||itype[j]==COP2||itype[j]==C2LS||itype[j]==C2OP) {
- if(count_free_regs(regs[j].regmap)<2) {
- //printf("No free regs for store %x\n",start+j*4);
+ if(itype[j]==UJUMP||itype[j]==RJUMP||(source[j]>>16)==0x1000)
+ {
+ // Stop on unconditional branch
+ break;
+ }
+ if(itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP)
+ {
+ if(ooo[j]) {
+ if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j+1])
+ break;
+ }else{
+ if(count_free_regs(branch_regs[j].regmap)<=minimum_free_regs[j+1])
+ break;
+ }
+ if(get_reg(branch_regs[j].regmap,f_regmap[hr])>=0) {
+ //printf("no-match due to different register (branch)\n");
break;
}
}
- else if(itype[j]!=NOP&&itype[j]!=MOV&&itype[j]!=ALU&&itype[j]!=SHIFTIMM&&itype[j]!=IMM16&&itype[j]!=LOAD) break;
+ if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j]) {
+ //printf("No free regs for store %x\n",start+j*4);
+ break;
+ }
if(f_regmap[hr]>=64) {
if(regs[j].is32&(1LL<<(f_regmap[hr]&63))) {
break;
if(bt[i]) {
for(j=i;j<slen-1;j++) {
if(regs[j].regmap[HOST_CCREG]!=-1) break;
- if(itype[j]==STORE||itype[j]==STORELR||itype[j]==C1LS
- ||itype[j]==SHIFT||itype[j]==COP1||itype[j]==FLOAT
- ||itype[j]==FCOMP||itype[j]==FCONV
- ||itype[j]==COP2||itype[j]==C2LS||itype[j]==C2OP) {
- if(count_free_regs(regs[j].regmap)<2) {
- //printf("no free regs for store %x\n",start+j*4);
- break;
- }
+ if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j]) {
+ //printf("no free regs for store %x\n",start+j*4);
+ break;
}
- else
- if(itype[j]!=NOP&&itype[j]!=MOV&&itype[j]!=ALU&&itype[j]!=SHIFTIMM&&itype[j]!=IMM16&&itype[j]!=LOAD) break;
}
if(regs[j].regmap[HOST_CCREG]==CCREG) {
int k=i;
int k;
k=i;
while(regs[k-1].regmap[HOST_CCREG]==-1) {
- if(itype[k-1]==STORE||itype[k-1]==STORELR||itype[k-1]==C1LS
- ||itype[k-1]==SHIFT||itype[k-1]==COP1||itype[k-1]==FLOAT
- ||itype[k-1]==FCONV||itype[k-1]==FCOMP
- ||itype[k-1]==COP2||itype[k-1]==C2LS||itype[k-1]==C2OP) {
- if(count_free_regs(regs[k-1].regmap)<2) {
- //printf("no free regs for store %x\n",start+(k-1)*4);
- break;
- }
+ if(count_free_regs(regs[k-1].regmap)<=minimum_free_regs[k-1]) {
+ //printf("no free regs for store %x\n",start+(k-1)*4);
+ break;
}
- else
- if(itype[k-1]!=NOP&&itype[k-1]!=MOV&&itype[k-1]!=ALU&&itype[k-1]!=SHIFTIMM&&itype[k-1]!=IMM16&&itype[k-1]!=LOAD) break;
k--;
}
if(regs[k-1].regmap[HOST_CCREG]==CCREG) {
if(itype[i]!=STORE&&itype[i]!=STORELR&&itype[i]!=C1LS&&itype[i]!=SHIFT&&
itype[i]!=NOP&&itype[i]!=MOV&&itype[i]!=ALU&&itype[i]!=SHIFTIMM&&
itype[i]!=IMM16&&itype[i]!=LOAD&&itype[i]!=COP1&&itype[i]!=FLOAT&&
- itype[i]!=FCONV&&itype[i]!=FCOMP&&
- itype[i]!=COP2&&itype[i]!=C2LS&&itype[i]!=C2OP)
+ itype[i]!=FCONV&&itype[i]!=FCOMP)
{
memcpy(f_regmap,regs[i].regmap,sizeof(f_regmap));
}
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);
break;
case 3:
// Clear jump_out
+ #ifdef __arm__
+ if((expirep&2047)==0)
+ do_clear_cache();
+ #endif
ll_remove_matching_addrs(jump_out+(expirep&2047),base,shift);
ll_remove_matching_addrs(jump_out+2048+(expirep&2047),base,shift);
break;