#ifdef _3DS
#include <3ds_utils.h>
#endif
-#ifdef VITA
-#include <psp2/kernel/sysmem.h>
-static int sceBlock;
-#endif
#include "new_dynarec_config.h"
#include "../psxhle.h"
#include "../psxinterpreter.h"
#include "../gte.h"
#include "emu_if.h" // emulator interface
+#include "arm_features.h"
#define noinline __attribute__((noinline,noclone))
#ifndef ARRAY_SIZE
//#define DISASM
//#define ASSEM_PRINT
+//#define REG_ALLOC_PRINT
#ifdef ASSEM_PRINT
#define assem_debug printf
#define MAXBLOCK 4096
#define MAX_OUTPUT_BLOCK_SIZE 262144
+#ifdef VITA
+// apparently Vita has a 16MB limit, so either we cut tc in half,
+// or use this hack (it's a hack because tc size was designed to be power-of-2)
+#define TC_REDUCE_BYTES 4096
+#else
+#define TC_REDUCE_BYTES 0
+#endif
+
struct ndrc_mem
{
- u_char translation_cache[1 << TARGET_SIZE_2];
+ u_char translation_cache[(1 << TARGET_SIZE_2) - TC_REDUCE_BYTES];
struct
{
struct tramp_insns ops[2048 / sizeof(struct tramp_insns)];
INVCODE_STUB = 14,
};
+// regmap_pre[i] - regs before [i] insn starts; dirty things here that
+// don't match .regmap will be written back
+// [i].regmap_entry - regs that must be set up if someone jumps here
+// [i].regmap - regs [i] insn will read/(over)write
+// branch_regs[i].* - same as above but for branches, takes delay slot into account
struct regstat
{
- signed char regmap_entry[HOST_REGS]; // pre-insn + loop preloaded regs?
+ signed char regmap_entry[HOST_REGS];
signed char regmap[HOST_REGS];
uint64_t wasdirty;
uint64_t dirty;
uint64_t u;
- u_int wasconst;
- u_int isconst;
+ u_int wasconst; // before; for example 'lw r2, (r2)' wasconst is true
+ u_int isconst; // ... but isconst is false when r2 is known
u_int loadedconst; // host regs that have constants loaded
u_int waswritten; // MIPS regs that were used as store base before
};
u_char rs2;
u_char rt1;
u_char rt2;
- u_char lt1;
+ u_char use_lt1:1;
u_char bt:1;
u_char ooo:1;
u_char is_ds:1;
static struct ll_entry *jump_out[4096];
static u_int start;
static u_int *source;
- static char insn[MAXBLOCK][10];
static uint64_t gte_rs[MAXBLOCK]; // gte: 32 data and 32 ctl regs
static uint64_t gte_rt[MAXBLOCK];
static uint64_t gte_unneeded[MAXBLOCK];
static u_int ba[MAXBLOCK];
static uint64_t unneeded_reg[MAXBLOCK];
static uint64_t branch_unneeded_reg[MAXBLOCK];
- // pre-instruction [i], excluding loop-preload regs?
+ // see 'struct regstat' for a description
static signed char regmap_pre[MAXBLOCK][HOST_REGS];
// contains 'real' consts at [i] insn, but may differ from what's actually
// loaded in host reg as 'final' value is always loaded, see get_final_value()
static void *copy;
static int expirep;
static u_int stop_after_jal;
- static u_int f1_hack; // 0 - off, ~0 - capture address, else addr
+ static u_int f1_hack;
int new_dynarec_hacks;
int new_dynarec_hacks_pergame;
//#define FLOAT 19 // Floating point unit
//#define FCONV 20 // Convert integer to float
//#define FCOMP 21 // Floating point compare (sets FSREG)
-#define SYSCALL 22// SYSCALL
+#define SYSCALL 22// SYSCALL,BREAK
#define OTHER 23 // Other
#define SPAN 24 // Branch/delay slot spans 2 pages
#define NI 25 // Not implemented
void cc_interrupt();
void fp_exception();
void fp_exception_ds();
+void jump_syscall (u_int u0, u_int u1, u_int pc);
+void jump_syscall_ds(u_int u0, u_int u1, u_int pc);
+void jump_break (u_int u0, u_int u1, u_int pc);
+void jump_break_ds(u_int u0, u_int u1, u_int pc);
void jump_to_new_pc();
void call_gteStall();
void new_dyna_leave();
static void emit_far_jump(const void *f);
static void emit_far_call(const void *f);
+#ifdef VITA
+#include <psp2/kernel/sysmem.h>
+static int sceBlock;
+// note: this interacts with RetroArch's Vita bootstrap code: bootstrap/vita/sbrk.c
+extern int getVMBlock();
+int _newlib_vm_size_user = sizeof(*ndrc);
+#endif
+
static void mprotect_w_x(void *start, void *end, int is_x)
{
#ifdef NO_WRITE_EXEC
int cycle_multiplier = CYCLE_MULT_DEFAULT; // 100 for 1.0
int cycle_multiplier_override;
int cycle_multiplier_old;
+static int cycle_multiplier_active;
static int CLOCK_ADJUST(int x)
{
- int m = cycle_multiplier_override && cycle_multiplier == CYCLE_MULT_DEFAULT
- ? cycle_multiplier_override : cycle_multiplier;
- int s=(x>>31)|1;
+ int m = cycle_multiplier_active;
+ int s = (x >> 31) | 1;
return (x * m + s * 50) / 100;
}
//printf("TRACE: count=%d next=%d (get_addr no-match %x)\n",Count,next_interupt,vaddr);
int r=new_recompile_block(vaddr);
if(r==0) return get_addr(vaddr);
- // Execute in unmapped page, generate pagefault execption
+ // generate an address error
Status|=2;
- Cause=(vaddr<<31)|0x8;
+ Cause=(vaddr<<31)|(4<<2);
EPC=(vaddr&1)?vaddr-5:vaddr;
BadVAddr=(vaddr&~1);
- Context=(Context&0xFF80000F)|((BadVAddr>>9)&0x007FFFF0);
- EntryHi=BadVAddr&0xFFFFE000;
- return get_addr_ht(0x80000000);
+ return get_addr_ht(0x80000080);
}
// Look up address in hash table first
void *get_addr_ht(u_int vaddr)
return get_addr(vaddr);
}
-void clear_all_regs(signed char regmap[])
+static void clear_all_regs(signed char regmap[])
+{
+ memset(regmap, -1, sizeof(regmap[0]) * HOST_REGS);
+}
+
+#if defined(__arm__) && defined(HAVE_ARMV6) && HOST_REGS == 13 && EXCLUDE_REG == 11
+
+extern signed char get_reg(const signed char regmap[], signed char r);
+
+#else
+
+static signed char get_reg(const signed char regmap[], signed char r)
{
int hr;
- for (hr=0;hr<HOST_REGS;hr++) regmap[hr]=-1;
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ if (hr == EXCLUDE_REG)
+ continue;
+ if (regmap[hr] == r)
+ return hr;
+ }
+ return -1;
}
-static signed char get_reg(const signed char regmap[],int r)
+#endif
+
+static signed char get_reg_temp(const signed char regmap[])
{
int hr;
- for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&®map[hr]==r) return hr;
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ if (hr == EXCLUDE_REG)
+ continue;
+ if (regmap[hr] == (signed char)-1)
+ return hr;
+ }
return -1;
}
return -1;
}
-int count_free_regs(signed char regmap[])
+// reverse reg map: mips -> host
+#define RRMAP_SIZE 64
+static void make_rregs(const signed char regmap[], signed char rrmap[RRMAP_SIZE],
+ u_int *regs_can_change)
+{
+ u_int r, hr, hr_can_change = 0;
+ memset(rrmap, -1, RRMAP_SIZE);
+ for (hr = 0; hr < HOST_REGS; )
+ {
+ r = regmap[hr];
+ rrmap[r & (RRMAP_SIZE - 1)] = hr;
+ // only add mips $1-$31+$lo, others shifted out
+ hr_can_change |= (uint64_t)1 << (hr + ((r - 1) & 32));
+ hr++;
+ if (hr == EXCLUDE_REG)
+ hr++;
+ }
+ hr_can_change |= 1u << (rrmap[33] & 31);
+ hr_can_change |= 1u << (rrmap[CCREG] & 31);
+ hr_can_change &= ~(1u << 31);
+ *regs_can_change = hr_can_change;
+}
+
+// same as get_reg, but takes rrmap
+static signed char get_rreg(signed char rrmap[RRMAP_SIZE], signed char r)
+{
+ assert(0 <= r && r < RRMAP_SIZE);
+ return rrmap[r];
+}
+
+static int count_free_regs(const signed char regmap[])
{
int count=0;
int hr;
return count;
}
-void dirty_reg(struct regstat *cur,signed char reg)
+static void dirty_reg(struct regstat *cur, signed char reg)
{
int hr;
- if(!reg) return;
- for (hr=0;hr<HOST_REGS;hr++) {
- if((cur->regmap[hr]&63)==reg) {
- cur->dirty|=1<<hr;
- }
- }
+ if (!reg) return;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0)
+ cur->dirty |= 1<<hr;
}
static void set_const(struct regstat *cur, signed char reg, uint32_t value)
{
int hr;
- if(!reg) return;
- for (hr=0;hr<HOST_REGS;hr++) {
- if(cur->regmap[hr]==reg) {
- cur->isconst|=1<<hr;
- current_constmap[hr]=value;
- }
+ if (!reg) return;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0) {
+ cur->isconst |= 1<<hr;
+ current_constmap[hr] = value;
}
}
static void clear_const(struct regstat *cur, signed char reg)
{
int hr;
- if(!reg) return;
- for (hr=0;hr<HOST_REGS;hr++) {
- if((cur->regmap[hr]&63)==reg) {
- cur->isconst&=~(1<<hr);
- }
- }
+ if (!reg) return;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0)
+ cur->isconst &= ~(1<<hr);
}
-static int is_const(struct regstat *cur, signed char reg)
+static int is_const(const struct regstat *cur, signed char reg)
{
int hr;
- if(reg<0) return 0;
- if(!reg) return 1;
- for (hr=0;hr<HOST_REGS;hr++) {
- if((cur->regmap[hr]&63)==reg) {
- return (cur->isconst>>hr)&1;
- }
- }
+ if (reg < 0) return 0;
+ if (!reg) return 1;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0)
+ return (cur->isconst>>hr)&1;
return 0;
}
-static uint32_t get_const(struct regstat *cur, signed char reg)
+static uint32_t get_const(const struct regstat *cur, signed char reg)
{
int hr;
- if(!reg) return 0;
- for (hr=0;hr<HOST_REGS;hr++) {
- if(cur->regmap[hr]==reg) {
- return current_constmap[hr];
- }
- }
- SysPrintf("Unknown constant in r%d\n",reg);
+ if (!reg) return 0;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0)
+ return current_constmap[hr];
+
+ SysPrintf("Unknown constant in r%d\n", reg);
abort();
}
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG) {
- if(((cur->regmap[hr]&63)!=dops[i].rs1)&&((cur->regmap[hr]&63)!=dops[i].rs2)&&
- ((cur->regmap[hr]&63)!=dops[i].rt1)&&((cur->regmap[hr]&63)!=dops[i].rt2))
+ if((cur->regmap[hr]!=dops[i].rs1)&&(cur->regmap[hr]!=dops[i].rs2)&&
+ (cur->regmap[hr]!=dops[i].rt1)&&(cur->regmap[hr]!=dops[i].rt2))
{
cur->regmap[hr]=-1;
cur->dirty&=~(1<<hr);
}
// Don't need zeros
- if((cur->regmap[hr]&63)==0)
+ if(cur->regmap[hr]==0)
{
cur->regmap[hr]=-1;
cur->dirty&=~(1<<hr);
FUNCNAME(jump_handler_write32),
FUNCNAME(invalidate_addr),
FUNCNAME(jump_to_new_pc),
+ FUNCNAME(jump_break),
+ FUNCNAME(jump_break_ds),
+ FUNCNAME(jump_syscall),
+ FUNCNAME(jump_syscall_ds),
FUNCNAME(call_gteStall),
FUNCNAME(new_dyna_leave),
FUNCNAME(pcsx_mtc0),
if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
for(hr=0;hr<HOST_REGS;hr++) {
// Evict both parts of a 64-bit register
- if((cur->regmap[hr]&63)==r) {
+ if(cur->regmap[hr]==r) {
cur->regmap[hr]=-1;
cur->dirty&=~(1<<hr);
cur->isconst&=~(1<<hr);
{
if(dops[i].rt1) {
if(dops[i].rs1&&needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1);
- else dops[i].lt1=dops[i].rs1;
+ else dops[i].use_lt1=!!dops[i].rs1;
alloc_reg(current,i,dops[i].rt1);
dirty_reg(current,dops[i].rt1);
if(is_const(current,dops[i].rs1)) {
static void imm16_alloc(struct regstat *current,int i)
{
if(dops[i].rs1&&needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1);
- else dops[i].lt1=dops[i].rs1;
+ else dops[i].use_lt1=!!dops[i].rs1;
if(dops[i].rt1) alloc_reg(current,i,dops[i].rt1);
if(dops[i].opcode==0x18||dops[i].opcode==0x19) { // DADDI/DADDIU
assert(0);
int hr;
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG) {
- if((regmap[hr]&63)==r) {
+ if(regmap[hr]==r) {
if((dirty>>hr)&1) {
assert(regmap[hr]<64);
emit_storereg(r,hr);
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG) {
reg=pre[hr];
- if(((~u)>>(reg&63))&1) {
+ if(((~u)>>reg)&1) {
if(reg>0) {
if(((dirty_pre&~dirty)>>hr)&1) {
if(reg>0&®<34) {
// could be FIFO, must perform the read
// ||dummy read
assem_debug("(forced read)\n");
- tl=get_reg(i_regs->regmap,-1);
+ tl=get_reg_temp(i_regs->regmap);
assert(tl>=0);
}
if(offset||s<0||c) addr=tl;
else addr=s;
- //if(tl<0) tl=get_reg(i_regs->regmap,-1);
+ //if(tl<0) tl=get_reg_temp(i_regs->regmap);
if(tl>=0) {
//printf("load_assemble: c=%d\n",c);
//if(c) printf("load_assemble: const=%lx\n",(long)constmap[i][s]+offset);
u_int reglist=get_host_reglist(i_regs->regmap);
tl=get_reg(i_regs->regmap,dops[i].rt1);
s=get_reg(i_regs->regmap,dops[i].rs1);
- temp=get_reg(i_regs->regmap,-1);
+ temp=get_reg_temp(i_regs->regmap);
temp2=get_reg(i_regs->regmap,FTEMP);
addr=get_reg(i_regs->regmap,AGEN1+(i&1));
assert(addr<0);
tl=get_reg(i_regs->regmap,dops[i].rs2);
s=get_reg(i_regs->regmap,dops[i].rs1);
temp=get_reg(i_regs->regmap,agr);
- if(temp<0) temp=get_reg(i_regs->regmap,-1);
+ if(temp<0) temp=get_reg_temp(i_regs->regmap);
offset=imm[i];
if(s>=0) {
c=(i_regs->wasconst>>s)&1;
tl=get_reg(i_regs->regmap,dops[i].rs2);
s=get_reg(i_regs->regmap,dops[i].rs1);
temp=get_reg(i_regs->regmap,agr);
- if(temp<0) temp=get_reg(i_regs->regmap,-1);
+ if(temp<0) temp=get_reg_temp(i_regs->regmap);
offset=imm[i];
if(s>=0) {
c=(i_regs->isconst>>s)&1;
{
int j, known_cycles = 0;
u_int reglist = get_host_reglist(i_regs->regmap);
- int rtmp = get_reg(i_regs->regmap, -1);
+ int rtmp = get_reg_temp(i_regs->regmap);
if (rtmp < 0)
rtmp = reglist_find_free(reglist);
if (HACK_ENABLED(NDHACK_NO_STALLS))
// get the address
if (dops[i].opcode==0x3a) { // SWC2
ar=get_reg(i_regs->regmap,agr);
- if(ar<0) ar=get_reg(i_regs->regmap,-1);
+ if(ar<0) ar=get_reg_temp(i_regs->regmap);
reglist|=1<<ar;
} else { // LWC2
ar=tl;
static void cop2_assemble(int i, const struct regstat *i_regs)
{
u_int copr = (source[i]>>11) & 0x1f;
- signed char temp = get_reg(i_regs->regmap, -1);
+ signed char temp = get_reg_temp(i_regs->regmap);
if (!HACK_ENABLED(NDHACK_NO_STALLS)) {
u_int reglist = reglist_exclude(get_host_reglist(i_regs->regmap), temp, -1);
static void syscall_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
- emit_movimm(0x20,0); // cause code
- emit_movimm(0,1); // not in delay slot
- call_c_cpu_handler(i, i_regs, ccadj_, start+i*4, psxException);
+ // 'break' tends to be littered around to catch things like
+ // division by 0 and is almost never executed, so don't emit much code here
+ void *func = (dops[i].opcode2 == 0x0C)
+ ? (is_delayslot ? jump_syscall_ds : jump_syscall)
+ : (is_delayslot ? jump_break_ds : jump_break);
+ assert(get_reg(i_regs->regmap, CCREG) == HOST_CCREG);
+ emit_movimm(start + i*4, 2); // pc
+ emit_addimm(HOST_CCREG, ccadj_ + CLOCK_ADJUST(1), HOST_CCREG);
+ emit_far_jump(func);
}
static void hlecall_assemble(int i, const struct regstat *i_regs, int ccadj_)
case SPAN:
pagespan_assemble(i, i_regs);
break;
+ case NOP:
case OTHER:
case NI:
// not handled, just skip
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG) {
if(pre[hr]!=entry[hr]) {
- if(pre[hr]>=0&&(pre[hr]&63)<TEMPREG) {
+ if(pre[hr]>=0&&pre[hr]<TEMPREG) {
int nr;
if((nr=get_reg(entry,pre[hr]))>=0) {
emit_mov(hr,nr);
int agr=AGEN1+(i&1);
if(dops[i].itype==LOAD) {
ra=get_reg(i_regs->regmap,dops[i].rt1);
- if(ra<0) ra=get_reg(i_regs->regmap,-1);
+ if(ra<0) ra=get_reg_temp(i_regs->regmap);
assert(ra>=0);
}
if(dops[i].itype==LOADLR) {
}
if(dops[i].itype==STORE||dops[i].itype==STORELR) {
ra=get_reg(i_regs->regmap,agr);
- if(ra<0) ra=get_reg(i_regs->regmap,-1);
+ if(ra<0) ra=get_reg_temp(i_regs->regmap);
}
if(dops[i].itype==C2LS) {
if ((dops[i].opcode&0x3b)==0x31||(dops[i].opcode&0x3b)==0x32) // LWC1/LDC1/LWC2/LDC2
ra=get_reg(i_regs->regmap,FTEMP);
else { // SWC1/SDC1/SWC2/SDC2
ra=get_reg(i_regs->regmap,agr);
- if(ra<0) ra=get_reg(i_regs->regmap,-1);
+ if(ra<0) ra=get_reg_temp(i_regs->regmap);
}
}
int rs=get_reg(i_regs->regmap,dops[i].rs1);
emit_zeroreg(hr);
}
else
- if(i_regmap[hr]>0 && (i_regmap[hr]&63)<TEMPREG && i_regmap[hr]!=CCREG)
+ if(i_regmap[hr]>0 && i_regmap[hr]<TEMPREG && i_regmap[hr]!=CCREG)
{
emit_loadreg(i_regmap[hr],hr);
}
emit_zeroreg(hr);
}
else
- if(i_regmap[hr]>0 && (i_regmap[hr]&63)<TEMPREG && i_regmap[hr]!=CCREG)
+ if(i_regmap[hr]>0 && i_regmap[hr]<TEMPREG && i_regmap[hr]!=CCREG)
{
emit_loadreg(i_regmap[hr],hr);
}
while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
- (branch_regs[i].regmap[hr]&63)!=dops[i].rs1 &&
- (branch_regs[i].regmap[hr]&63)!=dops[i].rs2 )
+ branch_regs[i].regmap[hr]!=dops[i].rs1 &&
+ branch_regs[i].regmap[hr]!=dops[i].rs2 )
{
addr=hr++;break;
}
while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
- (branch_regs[i].regmap[hr]&63)!=dops[i].rs1 &&
- (branch_regs[i].regmap[hr]&63)!=dops[i].rs2 )
+ branch_regs[i].regmap[hr]!=dops[i].rs1 &&
+ branch_regs[i].regmap[hr]!=dops[i].rs2 )
{
alt=hr++;break;
}
while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
- (branch_regs[i].regmap[hr]&63)!=dops[i].rs1 &&
- (branch_regs[i].regmap[hr]&63)!=dops[i].rs2 )
+ branch_regs[i].regmap[hr]!=dops[i].rs1 &&
+ branch_regs[i].regmap[hr]!=dops[i].rs2 )
{
ntaddr=hr;break;
}
int cc;
int match;
match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
- assem_debug("smatch=%d\n",match);
+ assem_debug("smatch=%d ooo=%d\n", match, dops[i].ooo);
int s1l;
int unconditional=0,nevertaken=0;
int invert=0;
wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,ds_unneeded);
load_regs(regs[i].regmap,branch_regs[i].regmap,dops[i+1].rs1,dops[i+1].rs2);
address_generation(i+1,&branch_regs[i],0);
- load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG);
+ if (ram_offset)
+ load_regs(regs[i].regmap,branch_regs[i].regmap,ROREG,ROREG);
+ load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,INVCP);
ds_assemble(i+1,&branch_regs[i]);
cc=get_reg(branch_regs[i].regmap,CCREG);
if (cc == -1) {
while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
- (i_regs->regmap[hr]&63)!=dops[i].rs1 &&
- (i_regs->regmap[hr]&63)!=dops[i].rs2 )
+ i_regs->regmap[hr]!=dops[i].rs1 &&
+ i_regs->regmap[hr]!=dops[i].rs2 )
{
addr=hr++;break;
}
while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG && hr!=HOST_BTREG &&
- (i_regs->regmap[hr]&63)!=dops[i].rs1 &&
- (i_regs->regmap[hr]&63)!=dops[i].rs2 )
+ i_regs->regmap[hr]!=dops[i].rs1 &&
+ i_regs->regmap[hr]!=dops[i].rs2 )
{
alt=hr++;break;
}
while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG && hr!=HOST_BTREG &&
- (i_regs->regmap[hr]&63)!=dops[i].rs1 &&
- (i_regs->regmap[hr]&63)!=dops[i].rs2 )
+ i_regs->regmap[hr]!=dops[i].rs1 &&
+ i_regs->regmap[hr]!=dops[i].rs2 )
{
ntaddr=hr;break;
}
}
int btaddr=get_reg(regs[0].regmap,BTREG);
if(btaddr<0) {
- btaddr=get_reg(regs[0].regmap,-1);
+ btaddr=get_reg_temp(regs[0].regmap);
emit_readword(&branch_target,btaddr);
}
assert(btaddr!=HOST_CCREG);
load_regs_bt(regs[0].regmap,regs[0].dirty,start+4);
}
+static void check_regmap(signed char *regmap)
+{
+#ifndef NDEBUG
+ int i,j;
+ for (i = 0; i < HOST_REGS; i++) {
+ if (regmap[i] < 0)
+ continue;
+ for (j = i + 1; j < HOST_REGS; j++)
+ assert(regmap[i] != regmap[j]);
+ }
+#endif
+}
+
// Basic liveness analysis for MIPS registers
-void unneeded_registers(int istart,int iend,int r)
+static void unneeded_registers(int istart,int iend,int r)
{
int i;
uint64_t u,gte_u,b,gte_b;
// Write back dirty registers as soon as we will no longer modify them,
// so that we don't end up with lots of writes at the branches.
-void clean_registers(int istart,int iend,int wr)
+static void clean_registers(int istart, int iend, int wr)
{
int i;
int r;
}
for (i=iend;i>=istart;i--)
{
+ signed char rregmap_i[RRMAP_SIZE];
+ u_int hr_candirty = 0;
+ assert(HOST_REGS < 32);
+ make_rregs(regs[i].regmap, rregmap_i, &hr_candirty);
+ __builtin_prefetch(regs[i-1].regmap);
if(dops[i].is_jump)
{
+ signed char branch_rregmap_i[RRMAP_SIZE];
+ u_int branch_hr_candirty = 0;
+ make_rregs(branch_regs[i].regmap, branch_rregmap_i, &branch_hr_candirty);
if(ba[i]<start || ba[i]>=(start+slen*4))
{
// Branch out of this block, flush all regs
+ will_dirty_i = 0;
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ will_dirty_i &= branch_hr_candirty;
if (dops[i].is_ujump)
{
// Unconditional branch
- will_dirty_i=0;
- wont_dirty_i=0;
+ wont_dirty_i = 0;
// Merge in delay slot (will dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- }
- }
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
}
else
{
// Conditional branch
- will_dirty_i=0;
- wont_dirty_i=wont_dirty_next;
+ wont_dirty_i = wont_dirty_next;
// Merge in delay slot (will dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if (1) { // !dops[i].likely) {
- // Might not dirty if likely branch is not taken
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==0) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- //if((regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- //if((regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- }
- }
- }
+ // (the original code had no explanation why these 2 are commented out)
+ //will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ //will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
}
// Merge in delay slot (wont dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((regs[i].regmap[r]&63)==dops[i].rt1) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt2) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) wont_dirty_i|=1<<r;
- if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) wont_dirty_i|=1<<r;
- if(branch_regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- }
- }
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ wont_dirty_i &= ~(1u << 31);
if(wr) {
#ifndef DESTRUCTIVE_WRITEBACK
branch_regs[i].dirty&=wont_dirty_i;
temp_will_dirty=0;
temp_wont_dirty=0;
// Merge in delay slot (will dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
- if(branch_regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt1) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt2) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
- if(regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
- }
- }
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ temp_will_dirty &= branch_hr_candirty;
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ temp_will_dirty &= hr_candirty;
} else {
// Conditional branch (not taken case)
temp_will_dirty=will_dirty_next;
temp_wont_dirty=wont_dirty_next;
// Merge in delay slot (will dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if (1) { // !dops[i].likely) {
- // Will not dirty if likely branch is not taken
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
- if(branch_regs[i].regmap[r]==0) temp_will_dirty&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
- //if((regs[i].regmap[r]&63)==dops[i].rt1) temp_will_dirty|=1<<r;
- //if((regs[i].regmap[r]&63)==dops[i].rt2) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
- if(regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
- }
- }
- }
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ temp_will_dirty &= branch_hr_candirty;
+ //temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ //temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ temp_will_dirty &= hr_candirty;
}
// Merge in delay slot (wont dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((regs[i].regmap[r]&63)==dops[i].rt1) temp_wont_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt2) temp_wont_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) temp_wont_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) temp_wont_dirty|=1<<r;
- if(regs[i].regmap[r]==CCREG) temp_wont_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) temp_wont_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) temp_wont_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) temp_wont_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) temp_wont_dirty|=1<<r;
- if(branch_regs[i].regmap[r]==CCREG) temp_wont_dirty|=1<<r;
- }
- }
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ temp_wont_dirty &= ~(1u << 31);
// Deal with changed mappings
if(i<iend) {
for(r=0;r<HOST_REGS;r++) {
if(regs[i].regmap[r]!=regmap_pre[i][r]) {
temp_will_dirty&=~(1<<r);
temp_wont_dirty&=~(1<<r);
- if((regmap_pre[i][r]&63)>0 && (regmap_pre[i][r]&63)<34) {
- temp_will_dirty|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
- temp_wont_dirty|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
+ if(regmap_pre[i][r]>0 && regmap_pre[i][r]<34) {
+ temp_will_dirty|=((unneeded_reg[i]>>regmap_pre[i][r])&1)<<r;
+ temp_wont_dirty|=((unneeded_reg[i]>>regmap_pre[i][r])&1)<<r;
} else {
temp_will_dirty|=1<<r;
temp_wont_dirty|=1<<r;
wont_dirty_i|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
}
if(branch_regs[i].regmap[r]>=0) {
- will_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>(branch_regs[i].regmap[r]&63))&1)<<r;
- wont_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>(branch_regs[i].regmap[r]&63))&1)<<r;
+ will_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>branch_regs[i].regmap[r])&1)<<r;
+ wont_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>branch_regs[i].regmap[r])&1)<<r;
}
}
}
//}
// Merge in delay slot
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- }
- }
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ will_dirty_i &= branch_hr_candirty;
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
} else {
// Conditional branch
will_dirty_i=will_dirty_next;
wont_dirty_i=wont_dirty_next;
- //if(ba[i]>start+i*4) { // Disable recursion (for debugging)
+ //if(ba[i]>start+i*4) // Disable recursion (for debugging)
for(r=0;r<HOST_REGS;r++) {
if(r!=EXCLUDE_REG) {
signed char target_reg=branch_regs[i].regmap[r];
wont_dirty_i|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
}
else if(target_reg>=0) {
- will_dirty_i&=((unneeded_reg[(ba[i]-start)>>2]>>(target_reg&63))&1)<<r;
- wont_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>(target_reg&63))&1)<<r;
+ will_dirty_i&=((unneeded_reg[(ba[i]-start)>>2]>>target_reg)&1)<<r;
+ wont_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>target_reg)&1)<<r;
}
}
}
- //}
// Merge in delay slot
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if (1) { // !dops[i].likely) {
- // Might not dirty if likely branch is not taken
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- //if((regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- //if((regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- }
- }
- }
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ will_dirty_i &= branch_hr_candirty;
+ //will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ //will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
}
// Merge in delay slot (won't dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((regs[i].regmap[r]&63)==dops[i].rt1) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt2) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt1) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i+1].rt2) wont_dirty_i|=1<<r;
- if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt1) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i].rt2) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt1) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==dops[i+1].rt2) wont_dirty_i|=1<<r;
- if(branch_regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- }
- }
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ wont_dirty_i &= ~(1u << 31);
if(wr) {
#ifndef DESTRUCTIVE_WRITEBACK
branch_regs[i].dirty&=wont_dirty_i;
}
will_dirty_next=will_dirty_i;
wont_dirty_next=wont_dirty_i;
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((regs[i].regmap[r]&63)==dops[i].rt1) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt2) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt1) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i].rt2) wont_dirty_i|=1<<r;
- if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- if(i>istart) {
- if (!dops[i].is_jump)
- {
- // Don't store a register immediately after writing it,
- // may prevent dual-issue.
- if((regs[i].regmap[r]&63)==dops[i-1].rt1) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==dops[i-1].rt2) wont_dirty_i|=1<<r;
- }
- }
- }
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ wont_dirty_i &= ~(1u << 31);
+ if (i > istart && !dops[i].is_jump) {
+ // Don't store a register immediately after writing it,
+ // may prevent dual-issue.
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i-1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i-1].rt2) & 31);
}
// Save it
will_dirty[i]=will_dirty_i;
}
}
#endif
- //}
}
// Deal with changed mappings
temp_will_dirty=will_dirty_i;
regs[i].wasdirty|=will_dirty_i&(1<<r);
}
}
- else if(regmap_pre[i][r]>=0&&(nr=get_reg(regs[i].regmap,regmap_pre[i][r]))>=0) {
+ else if(regmap_pre[i][r]>=0&&(nr=get_rreg(rregmap_i,regmap_pre[i][r]))>=0) {
// Register moved to a different register
will_dirty_i&=~(1<<r);
wont_dirty_i&=~(1<<r);
else {
will_dirty_i&=~(1<<r);
wont_dirty_i&=~(1<<r);
- if((regmap_pre[i][r]&63)>0 && (regmap_pre[i][r]&63)<34) {
- will_dirty_i|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
- wont_dirty_i|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
+ if(regmap_pre[i][r]>0 && regmap_pre[i][r]<34) {
+ will_dirty_i|=((unneeded_reg[i]>>regmap_pre[i][r])&1)<<r;
+ wont_dirty_i|=((unneeded_reg[i]>>regmap_pre[i][r])&1)<<r;
} else {
wont_dirty_i|=1<<r;
/*printf("i: %x (%d) mismatch: %d\n",start+i*4,i,r);assert(!((will_dirty>>r)&1));*/
}
#ifdef DISASM
+#include <inttypes.h>
+static char insn[MAXBLOCK][10];
+
+#define set_mnemonic(i_, n_) \
+ strcpy(insn[i_], n_)
+
+void print_regmap(const char *name, const signed char *regmap)
+{
+ char buf[5];
+ int i, l;
+ fputs(name, stdout);
+ for (i = 0; i < HOST_REGS; i++) {
+ l = 0;
+ if (regmap[i] >= 0)
+ l = snprintf(buf, sizeof(buf), "$%d", regmap[i]);
+ for (; l < 3; l++)
+ buf[l] = ' ';
+ buf[l] = 0;
+ printf(" r%d=%s", i, buf);
+ }
+ fputs("\n", stdout);
+}
+
/* disassembly */
void disassemble_inst(int i)
{
//printf (" %s %8x\n",insn[i],source[i]);
printf (" %x: %s\n",start+i*4,insn[i]);
}
+ return;
+ printf("D: %"PRIu64" WD: %"PRIu64" U: %"PRIu64"\n",
+ regs[i].dirty, regs[i].wasdirty, unneeded_reg[i]);
+ print_regmap("pre: ", regmap_pre[i]);
+ print_regmap("entry: ", regs[i].regmap_entry);
+ print_regmap("map: ", regs[i].regmap);
+ if (dops[i].is_jump) {
+ print_regmap("bentry:", branch_regs[i].regmap_entry);
+ print_regmap("bmap: ", branch_regs[i].regmap);
+ }
}
#else
+#define set_mnemonic(i_, n_)
static void disassemble_inst(int i) {}
#endif // DISASM
SysPrintf("linkage_arm* miscompilation/breakage detected.\n");
}
- SysPrintf("testing if we can run recompiled code...\n");
+ SysPrintf("testing if we can run recompiled code @%p...\n", out);
((volatile u_int *)out)[0]++; // make cache dirty
for (i = 0; i < ARRAY_SIZE(ret); i++) {
literalcount=0;
stop_after_jal=0;
inv_code_start=inv_code_end=~0;
+ hack_addr=0;
f1_hack=0;
// TLB
for(n=0;n<4096;n++) ll_clear(jump_in+n);
void new_dynarec_init(void)
{
- SysPrintf("Init new dynarec\n");
+ SysPrintf("Init new dynarec, ndrc size %x\n", (int)sizeof(*ndrc));
+#ifdef _3DS
+ check_rosalina();
+#endif
#ifdef BASE_ADDR_DYNAMIC
#ifdef VITA
- sceBlock = sceKernelAllocMemBlockForVM("code", 1 << TARGET_SIZE_2);
- if (sceBlock < 0)
- SysPrintf("sceKernelAllocMemBlockForVM failed\n");
+ sceBlock = getVMBlock(); //sceKernelAllocMemBlockForVM("code", sizeof(*ndrc));
+ if (sceBlock <= 0)
+ SysPrintf("sceKernelAllocMemBlockForVM failed: %x\n", sceBlock);
int ret = sceKernelGetMemBlockBase(sceBlock, (void **)&ndrc);
if (ret < 0)
- SysPrintf("sceKernelGetMemBlockBase failed\n");
+ SysPrintf("sceKernelGetMemBlockBase failed: %x\n", ret);
+ sceKernelOpenVMDomain();
+ sceClibPrintf("translation_cache = 0x%08lx\n ", (long)ndrc->translation_cache);
+ #elif defined(_MSC_VER)
+ ndrc = VirtualAlloc(NULL, sizeof(*ndrc), MEM_COMMIT | MEM_RESERVE,
+ PAGE_EXECUTE_READWRITE);
#else
uintptr_t desired_addr = 0;
#ifdef __ELF__
#else
#ifndef NO_WRITE_EXEC
// not all systems allow execute in data segment by default
- if (mprotect(ndrc, sizeof(ndrc->translation_cache) + sizeof(ndrc->tramp.ops),
+ // size must be 4K aligned for 3DS?
+ if (mprotect(ndrc, sizeof(*ndrc),
PROT_READ | PROT_WRITE | PROT_EXEC) != 0)
SysPrintf("mprotect() failed: %s\n", strerror(errno));
#endif
int n;
#ifdef BASE_ADDR_DYNAMIC
#ifdef VITA
- sceKernelFreeMemBlock(sceBlock);
- sceBlock = -1;
+ // sceBlock is managed by retroarch's bootstrap code
+ //sceKernelFreeMemBlock(sceBlock);
+ //sceBlock = -1;
#else
if (munmap(ndrc, sizeof(*ndrc)) < 0)
SysPrintf("munmap() failed\n");
static u_int *get_source_start(u_int addr, u_int *limit)
{
- if (!HACK_ENABLED(NDHACK_OVERRIDE_CYCLE_M))
- cycle_multiplier_override = 0;
-
if (addr < 0x00200000 ||
(0xa0000000 <= addr && addr < 0xa0200000))
{
// BIOS. The multiplier should be much higher as it's uncached 8bit mem,
// but timings in PCSX are too tied to the interpreter's BIAS
if (!HACK_ENABLED(NDHACK_OVERRIDE_CYCLE_M))
- cycle_multiplier_override = 200;
+ cycle_multiplier_active = 200;
*limit = (addr & 0xfff00000) | 0x80000;
return (u_int *)((u_char *)psxR + (addr&0x7ffff));
memcpy(&psxRegs.GPR, regs_save, sizeof(regs_save));
}
+static int apply_hacks(void)
+{
+ int i;
+ if (HACK_ENABLED(NDHACK_NO_COMPAT_HACKS))
+ return 0;
+ /* special hack(s) */
+ for (i = 0; i < slen - 4; i++)
+ {
+ // lui a4, 0xf200; jal <rcnt_read>; addu a0, 2; slti v0, 28224
+ if (source[i] == 0x3c04f200 && dops[i+1].itype == UJUMP
+ && source[i+2] == 0x34840002 && dops[i+3].opcode == 0x0a
+ && imm[i+3] == 0x6e40 && dops[i+3].rs1 == 2)
+ {
+ SysPrintf("PE2 hack @%08x\n", start + (i+3)*4);
+ dops[i + 3].itype = NOP;
+ }
+ }
+ i = slen;
+ if (i > 10 && source[i-1] == 0 && source[i-2] == 0x03e00008
+ && source[i-4] == 0x8fbf0018 && source[i-6] == 0x00c0f809
+ && dops[i-7].itype == STORE)
+ {
+ i = i-8;
+ if (dops[i].itype == IMM16)
+ i--;
+ // swl r2, 15(r6); swr r2, 12(r6); sw r6, *; jalr r6
+ if (dops[i].itype == STORELR && dops[i].rs1 == 6
+ && dops[i-1].itype == STORELR && dops[i-1].rs1 == 6)
+ {
+ SysPrintf("F1 hack from %08x, old dst %08x\n", start, hack_addr);
+ f1_hack = 1;
+ return 1;
+ }
+ }
+ return 0;
+}
+
int new_recompile_block(u_int addr)
{
u_int pagelimit = 0;
ll_add_flags(jump_in+page,start,state_rflags,(void *)beginning);
return 0;
}
- else if (f1_hack == ~0u || (f1_hack != 0 && start == f1_hack)) {
+ else if (f1_hack && hack_addr == 0) {
void *beginning = start_block();
u_int page = get_page(start);
+ emit_movimm(start, 0);
+ emit_writeword(0, &hack_addr);
emit_readword(&psxRegs.GPR.n.sp, 0);
emit_readptr(&mem_rtab, 1);
emit_shrimm(0, 12, 2);
ll_add_flags(jump_in + page, start, state_rflags, beginning);
SysPrintf("F1 hack to %08x\n", start);
- f1_hack = start;
return 0;
}
+ cycle_multiplier_active = cycle_multiplier_override && cycle_multiplier == CYCLE_MULT_DEFAULT
+ ? cycle_multiplier_override : cycle_multiplier;
+
source = get_source_start(start, &pagelimit);
if (source == NULL) {
- SysPrintf("Compile at bogus memory address: %08x\n", addr);
- abort();
+ if (addr != hack_addr) {
+ SysPrintf("Compile at bogus memory address: %08x\n", addr);
+ hack_addr = addr;
+ }
+ //abort();
+ return -1;
}
/* Pass 1: disassemble */
/* Pass 10: garbage collection / free memory */
int j;
- int done=0;
+ int done = 0, ni_count = 0;
unsigned int type,op,op2;
//printf("addr = %x source = %x %x\n", addr,source,source[0]);
/* Pass 1 disassembly */
- for(i=0;!done;i++) {
- dops[i].bt=0;
- dops[i].ooo=0;
+ for (i = 0; !done; i++)
+ {
+ memset(&dops[i], 0, sizeof(dops[i]));
op2=0;
minimum_free_regs[i]=0;
dops[i].opcode=op=source[i]>>26;
switch(op)
{
- case 0x00: strcpy(insn[i],"special"); type=NI;
+ case 0x00: set_mnemonic(i, "special"); type=NI;
op2=source[i]&0x3f;
switch(op2)
{
- case 0x00: strcpy(insn[i],"SLL"); type=SHIFTIMM; break;
- case 0x02: strcpy(insn[i],"SRL"); type=SHIFTIMM; break;
- case 0x03: strcpy(insn[i],"SRA"); type=SHIFTIMM; break;
- case 0x04: strcpy(insn[i],"SLLV"); type=SHIFT; break;
- case 0x06: strcpy(insn[i],"SRLV"); type=SHIFT; break;
- case 0x07: strcpy(insn[i],"SRAV"); type=SHIFT; break;
- case 0x08: strcpy(insn[i],"JR"); type=RJUMP; break;
- case 0x09: strcpy(insn[i],"JALR"); type=RJUMP; break;
- case 0x0C: strcpy(insn[i],"SYSCALL"); type=SYSCALL; break;
- case 0x0D: strcpy(insn[i],"BREAK"); type=OTHER; break;
- case 0x0F: strcpy(insn[i],"SYNC"); type=OTHER; break;
- case 0x10: strcpy(insn[i],"MFHI"); type=MOV; break;
- case 0x11: strcpy(insn[i],"MTHI"); type=MOV; break;
- case 0x12: strcpy(insn[i],"MFLO"); type=MOV; break;
- case 0x13: strcpy(insn[i],"MTLO"); type=MOV; break;
- case 0x18: strcpy(insn[i],"MULT"); type=MULTDIV; break;
- case 0x19: strcpy(insn[i],"MULTU"); type=MULTDIV; break;
- case 0x1A: strcpy(insn[i],"DIV"); type=MULTDIV; break;
- case 0x1B: strcpy(insn[i],"DIVU"); type=MULTDIV; break;
- case 0x20: strcpy(insn[i],"ADD"); type=ALU; break;
- case 0x21: strcpy(insn[i],"ADDU"); type=ALU; break;
- case 0x22: strcpy(insn[i],"SUB"); type=ALU; break;
- case 0x23: strcpy(insn[i],"SUBU"); type=ALU; break;
- case 0x24: strcpy(insn[i],"AND"); type=ALU; break;
- case 0x25: strcpy(insn[i],"OR"); type=ALU; break;
- case 0x26: strcpy(insn[i],"XOR"); type=ALU; break;
- case 0x27: strcpy(insn[i],"NOR"); type=ALU; break;
- case 0x2A: strcpy(insn[i],"SLT"); type=ALU; break;
- case 0x2B: strcpy(insn[i],"SLTU"); type=ALU; break;
- case 0x30: strcpy(insn[i],"TGE"); type=NI; break;
- case 0x31: strcpy(insn[i],"TGEU"); type=NI; break;
- case 0x32: strcpy(insn[i],"TLT"); type=NI; break;
- case 0x33: strcpy(insn[i],"TLTU"); type=NI; break;
- case 0x34: strcpy(insn[i],"TEQ"); type=NI; break;
- case 0x36: strcpy(insn[i],"TNE"); type=NI; break;
+ case 0x00: set_mnemonic(i, "SLL"); type=SHIFTIMM; break;
+ case 0x02: set_mnemonic(i, "SRL"); type=SHIFTIMM; break;
+ case 0x03: set_mnemonic(i, "SRA"); type=SHIFTIMM; break;
+ case 0x04: set_mnemonic(i, "SLLV"); type=SHIFT; break;
+ case 0x06: set_mnemonic(i, "SRLV"); type=SHIFT; break;
+ case 0x07: set_mnemonic(i, "SRAV"); type=SHIFT; break;
+ case 0x08: set_mnemonic(i, "JR"); type=RJUMP; break;
+ case 0x09: set_mnemonic(i, "JALR"); type=RJUMP; break;
+ case 0x0C: set_mnemonic(i, "SYSCALL"); type=SYSCALL; break;
+ case 0x0D: set_mnemonic(i, "BREAK"); type=SYSCALL; break;
+ case 0x0F: set_mnemonic(i, "SYNC"); type=OTHER; break;
+ case 0x10: set_mnemonic(i, "MFHI"); type=MOV; break;
+ case 0x11: set_mnemonic(i, "MTHI"); type=MOV; break;
+ case 0x12: set_mnemonic(i, "MFLO"); type=MOV; break;
+ case 0x13: set_mnemonic(i, "MTLO"); type=MOV; break;
+ case 0x18: set_mnemonic(i, "MULT"); type=MULTDIV; break;
+ case 0x19: set_mnemonic(i, "MULTU"); type=MULTDIV; break;
+ case 0x1A: set_mnemonic(i, "DIV"); type=MULTDIV; break;
+ case 0x1B: set_mnemonic(i, "DIVU"); type=MULTDIV; break;
+ case 0x20: set_mnemonic(i, "ADD"); type=ALU; break;
+ case 0x21: set_mnemonic(i, "ADDU"); type=ALU; break;
+ case 0x22: set_mnemonic(i, "SUB"); type=ALU; break;
+ case 0x23: set_mnemonic(i, "SUBU"); type=ALU; break;
+ case 0x24: set_mnemonic(i, "AND"); type=ALU; break;
+ case 0x25: set_mnemonic(i, "OR"); type=ALU; break;
+ case 0x26: set_mnemonic(i, "XOR"); type=ALU; break;
+ case 0x27: set_mnemonic(i, "NOR"); type=ALU; break;
+ case 0x2A: set_mnemonic(i, "SLT"); type=ALU; break;
+ case 0x2B: set_mnemonic(i, "SLTU"); type=ALU; break;
+ case 0x30: set_mnemonic(i, "TGE"); type=NI; break;
+ case 0x31: set_mnemonic(i, "TGEU"); type=NI; break;
+ case 0x32: set_mnemonic(i, "TLT"); type=NI; break;
+ case 0x33: set_mnemonic(i, "TLTU"); type=NI; break;
+ case 0x34: set_mnemonic(i, "TEQ"); type=NI; break;
+ case 0x36: set_mnemonic(i, "TNE"); type=NI; break;
#if 0
- case 0x14: strcpy(insn[i],"DSLLV"); type=SHIFT; break;
- case 0x16: strcpy(insn[i],"DSRLV"); type=SHIFT; break;
- case 0x17: strcpy(insn[i],"DSRAV"); type=SHIFT; break;
- case 0x1C: strcpy(insn[i],"DMULT"); type=MULTDIV; break;
- case 0x1D: strcpy(insn[i],"DMULTU"); type=MULTDIV; break;
- case 0x1E: strcpy(insn[i],"DDIV"); type=MULTDIV; break;
- case 0x1F: strcpy(insn[i],"DDIVU"); type=MULTDIV; break;
- case 0x2C: strcpy(insn[i],"DADD"); type=ALU; break;
- case 0x2D: strcpy(insn[i],"DADDU"); type=ALU; break;
- case 0x2E: strcpy(insn[i],"DSUB"); type=ALU; break;
- case 0x2F: strcpy(insn[i],"DSUBU"); type=ALU; break;
- case 0x38: strcpy(insn[i],"DSLL"); type=SHIFTIMM; break;
- case 0x3A: strcpy(insn[i],"DSRL"); type=SHIFTIMM; break;
- case 0x3B: strcpy(insn[i],"DSRA"); type=SHIFTIMM; break;
- case 0x3C: strcpy(insn[i],"DSLL32"); type=SHIFTIMM; break;
- case 0x3E: strcpy(insn[i],"DSRL32"); type=SHIFTIMM; break;
- case 0x3F: strcpy(insn[i],"DSRA32"); type=SHIFTIMM; break;
+ case 0x14: set_mnemonic(i, "DSLLV"); type=SHIFT; break;
+ case 0x16: set_mnemonic(i, "DSRLV"); type=SHIFT; break;
+ case 0x17: set_mnemonic(i, "DSRAV"); type=SHIFT; break;
+ case 0x1C: set_mnemonic(i, "DMULT"); type=MULTDIV; break;
+ case 0x1D: set_mnemonic(i, "DMULTU"); type=MULTDIV; break;
+ case 0x1E: set_mnemonic(i, "DDIV"); type=MULTDIV; break;
+ case 0x1F: set_mnemonic(i, "DDIVU"); type=MULTDIV; break;
+ case 0x2C: set_mnemonic(i, "DADD"); type=ALU; break;
+ case 0x2D: set_mnemonic(i, "DADDU"); type=ALU; break;
+ case 0x2E: set_mnemonic(i, "DSUB"); type=ALU; break;
+ case 0x2F: set_mnemonic(i, "DSUBU"); type=ALU; break;
+ case 0x38: set_mnemonic(i, "DSLL"); type=SHIFTIMM; break;
+ case 0x3A: set_mnemonic(i, "DSRL"); type=SHIFTIMM; break;
+ case 0x3B: set_mnemonic(i, "DSRA"); type=SHIFTIMM; break;
+ case 0x3C: set_mnemonic(i, "DSLL32"); type=SHIFTIMM; break;
+ case 0x3E: set_mnemonic(i, "DSRL32"); type=SHIFTIMM; break;
+ case 0x3F: set_mnemonic(i, "DSRA32"); type=SHIFTIMM; break;
#endif
}
break;
- case 0x01: strcpy(insn[i],"regimm"); type=NI;
+ case 0x01: set_mnemonic(i, "regimm"); type=NI;
op2=(source[i]>>16)&0x1f;
switch(op2)
{
- case 0x00: strcpy(insn[i],"BLTZ"); type=SJUMP; break;
- case 0x01: strcpy(insn[i],"BGEZ"); type=SJUMP; break;
- //case 0x02: strcpy(insn[i],"BLTZL"); type=SJUMP; break;
- //case 0x03: strcpy(insn[i],"BGEZL"); type=SJUMP; break;
- //case 0x08: strcpy(insn[i],"TGEI"); type=NI; break;
- //case 0x09: strcpy(insn[i],"TGEIU"); type=NI; break;
- //case 0x0A: strcpy(insn[i],"TLTI"); type=NI; break;
- //case 0x0B: strcpy(insn[i],"TLTIU"); type=NI; break;
- //case 0x0C: strcpy(insn[i],"TEQI"); type=NI; break;
- //case 0x0E: strcpy(insn[i],"TNEI"); type=NI; break;
- case 0x10: strcpy(insn[i],"BLTZAL"); type=SJUMP; break;
- case 0x11: strcpy(insn[i],"BGEZAL"); type=SJUMP; break;
- //case 0x12: strcpy(insn[i],"BLTZALL"); type=SJUMP; break;
- //case 0x13: strcpy(insn[i],"BGEZALL"); type=SJUMP; break;
+ case 0x00: set_mnemonic(i, "BLTZ"); type=SJUMP; break;
+ case 0x01: set_mnemonic(i, "BGEZ"); type=SJUMP; break;
+ //case 0x02: set_mnemonic(i, "BLTZL"); type=SJUMP; break;
+ //case 0x03: set_mnemonic(i, "BGEZL"); type=SJUMP; break;
+ //case 0x08: set_mnemonic(i, "TGEI"); type=NI; break;
+ //case 0x09: set_mnemonic(i, "TGEIU"); type=NI; break;
+ //case 0x0A: set_mnemonic(i, "TLTI"); type=NI; break;
+ //case 0x0B: set_mnemonic(i, "TLTIU"); type=NI; break;
+ //case 0x0C: set_mnemonic(i, "TEQI"); type=NI; break;
+ //case 0x0E: set_mnemonic(i, "TNEI"); type=NI; break;
+ case 0x10: set_mnemonic(i, "BLTZAL"); type=SJUMP; break;
+ case 0x11: set_mnemonic(i, "BGEZAL"); type=SJUMP; break;
+ //case 0x12: set_mnemonic(i, "BLTZALL"); type=SJUMP; break;
+ //case 0x13: set_mnemonic(i, "BGEZALL"); type=SJUMP; break;
}
break;
- case 0x02: strcpy(insn[i],"J"); type=UJUMP; break;
- case 0x03: strcpy(insn[i],"JAL"); type=UJUMP; break;
- case 0x04: strcpy(insn[i],"BEQ"); type=CJUMP; break;
- case 0x05: strcpy(insn[i],"BNE"); type=CJUMP; break;
- case 0x06: strcpy(insn[i],"BLEZ"); type=CJUMP; break;
- case 0x07: strcpy(insn[i],"BGTZ"); type=CJUMP; break;
- case 0x08: strcpy(insn[i],"ADDI"); type=IMM16; break;
- case 0x09: strcpy(insn[i],"ADDIU"); type=IMM16; break;
- case 0x0A: strcpy(insn[i],"SLTI"); type=IMM16; break;
- case 0x0B: strcpy(insn[i],"SLTIU"); type=IMM16; break;
- case 0x0C: strcpy(insn[i],"ANDI"); type=IMM16; break;
- case 0x0D: strcpy(insn[i],"ORI"); type=IMM16; break;
- case 0x0E: strcpy(insn[i],"XORI"); type=IMM16; break;
- case 0x0F: strcpy(insn[i],"LUI"); type=IMM16; break;
- case 0x10: strcpy(insn[i],"cop0"); type=NI;
+ case 0x02: set_mnemonic(i, "J"); type=UJUMP; break;
+ case 0x03: set_mnemonic(i, "JAL"); type=UJUMP; break;
+ case 0x04: set_mnemonic(i, "BEQ"); type=CJUMP; break;
+ case 0x05: set_mnemonic(i, "BNE"); type=CJUMP; break;
+ case 0x06: set_mnemonic(i, "BLEZ"); type=CJUMP; break;
+ case 0x07: set_mnemonic(i, "BGTZ"); type=CJUMP; break;
+ case 0x08: set_mnemonic(i, "ADDI"); type=IMM16; break;
+ case 0x09: set_mnemonic(i, "ADDIU"); type=IMM16; break;
+ case 0x0A: set_mnemonic(i, "SLTI"); type=IMM16; break;
+ case 0x0B: set_mnemonic(i, "SLTIU"); type=IMM16; break;
+ case 0x0C: set_mnemonic(i, "ANDI"); type=IMM16; break;
+ case 0x0D: set_mnemonic(i, "ORI"); type=IMM16; break;
+ case 0x0E: set_mnemonic(i, "XORI"); type=IMM16; break;
+ case 0x0F: set_mnemonic(i, "LUI"); type=IMM16; break;
+ case 0x10: set_mnemonic(i, "cop0"); type=NI;
op2=(source[i]>>21)&0x1f;
switch(op2)
{
- case 0x00: strcpy(insn[i],"MFC0"); type=COP0; break;
- case 0x02: strcpy(insn[i],"CFC0"); type=COP0; break;
- case 0x04: strcpy(insn[i],"MTC0"); type=COP0; break;
- case 0x06: strcpy(insn[i],"CTC0"); type=COP0; break;
- case 0x10: strcpy(insn[i],"RFE"); type=COP0; break;
+ case 0x00: set_mnemonic(i, "MFC0"); type=COP0; break;
+ case 0x02: set_mnemonic(i, "CFC0"); type=COP0; break;
+ case 0x04: set_mnemonic(i, "MTC0"); type=COP0; break;
+ case 0x06: set_mnemonic(i, "CTC0"); type=COP0; break;
+ case 0x10: set_mnemonic(i, "RFE"); type=COP0; break;
}
break;
- case 0x11: strcpy(insn[i],"cop1"); type=COP1;
+ case 0x11: set_mnemonic(i, "cop1"); type=COP1;
op2=(source[i]>>21)&0x1f;
break;
#if 0
- 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 0x17: strcpy(insn[i],"BGTZL"); type=CJUMP; break;
- 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;
+ case 0x14: set_mnemonic(i, "BEQL"); type=CJUMP; break;
+ case 0x15: set_mnemonic(i, "BNEL"); type=CJUMP; break;
+ case 0x16: set_mnemonic(i, "BLEZL"); type=CJUMP; break;
+ case 0x17: set_mnemonic(i, "BGTZL"); type=CJUMP; break;
+ case 0x18: set_mnemonic(i, "DADDI"); type=IMM16; break;
+ case 0x19: set_mnemonic(i, "DADDIU"); type=IMM16; break;
+ case 0x1A: set_mnemonic(i, "LDL"); type=LOADLR; break;
+ case 0x1B: set_mnemonic(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 0x23: strcpy(insn[i],"LW"); type=LOAD; break;
- case 0x24: strcpy(insn[i],"LBU"); type=LOAD; break;
- case 0x25: strcpy(insn[i],"LHU"); type=LOAD; break;
- case 0x26: strcpy(insn[i],"LWR"); type=LOADLR; break;
+ case 0x20: set_mnemonic(i, "LB"); type=LOAD; break;
+ case 0x21: set_mnemonic(i, "LH"); type=LOAD; break;
+ case 0x22: set_mnemonic(i, "LWL"); type=LOADLR; break;
+ case 0x23: set_mnemonic(i, "LW"); type=LOAD; break;
+ case 0x24: set_mnemonic(i, "LBU"); type=LOAD; break;
+ case 0x25: set_mnemonic(i, "LHU"); type=LOAD; break;
+ case 0x26: set_mnemonic(i, "LWR"); type=LOADLR; break;
#if 0
- case 0x27: strcpy(insn[i],"LWU"); type=LOAD; break;
+ case 0x27: set_mnemonic(i, "LWU"); type=LOAD; break;
#endif
- case 0x28: strcpy(insn[i],"SB"); type=STORE; 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;
+ case 0x28: set_mnemonic(i, "SB"); type=STORE; break;
+ case 0x29: set_mnemonic(i, "SH"); type=STORE; break;
+ case 0x2A: set_mnemonic(i, "SWL"); type=STORELR; break;
+ case 0x2B: set_mnemonic(i, "SW"); type=STORE; break;
#if 0
- case 0x2C: strcpy(insn[i],"SDL"); type=STORELR; break;
- case 0x2D: strcpy(insn[i],"SDR"); type=STORELR; break;
+ case 0x2C: set_mnemonic(i, "SDL"); type=STORELR; break;
+ case 0x2D: set_mnemonic(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;
+ case 0x2E: set_mnemonic(i, "SWR"); type=STORELR; break;
+ case 0x2F: set_mnemonic(i, "CACHE"); type=NOP; break;
+ case 0x30: set_mnemonic(i, "LL"); type=NI; break;
+ case 0x31: set_mnemonic(i, "LWC1"); type=C1LS; break;
#if 0
- 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;
+ case 0x34: set_mnemonic(i, "LLD"); type=NI; break;
+ case 0x35: set_mnemonic(i, "LDC1"); type=C1LS; break;
+ case 0x37: set_mnemonic(i, "LD"); type=LOAD; break;
#endif
- case 0x38: strcpy(insn[i],"SC"); type=NI; break;
- case 0x39: strcpy(insn[i],"SWC1"); type=C1LS; break;
+ case 0x38: set_mnemonic(i, "SC"); type=NI; break;
+ case 0x39: set_mnemonic(i, "SWC1"); type=C1LS; break;
#if 0
- 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;
+ case 0x3C: set_mnemonic(i, "SCD"); type=NI; break;
+ case 0x3D: set_mnemonic(i, "SDC1"); type=C1LS; break;
+ case 0x3F: set_mnemonic(i, "SD"); type=STORE; break;
#endif
- case 0x12: strcpy(insn[i],"COP2"); type=NI;
+ case 0x12: set_mnemonic(i, "COP2"); type=NI;
op2=(source[i]>>21)&0x1f;
//if (op2 & 0x10)
if (source[i]&0x3f) { // use this hack to support old savestates with patched gte insns
if (gte_handlers[source[i]&0x3f]!=NULL) {
+#ifdef DISASM
if (gte_regnames[source[i]&0x3f]!=NULL)
strcpy(insn[i],gte_regnames[source[i]&0x3f]);
else
snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
+#endif
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;
+ case 0x00: set_mnemonic(i, "MFC2"); type=COP2; break;
+ case 0x02: set_mnemonic(i, "CFC2"); type=COP2; break;
+ case 0x04: set_mnemonic(i, "MTC2"); type=COP2; break;
+ case 0x06: set_mnemonic(i, "CTC2"); type=COP2; break;
}
break;
- case 0x32: strcpy(insn[i],"LWC2"); type=C2LS; break;
- case 0x3A: strcpy(insn[i],"SWC2"); type=C2LS; break;
- case 0x3B: strcpy(insn[i],"HLECALL"); type=HLECALL; break;
- default: strcpy(insn[i],"???"); type=NI;
+ case 0x32: set_mnemonic(i, "LWC2"); type=C2LS; break;
+ case 0x3A: set_mnemonic(i, "SWC2"); type=C2LS; break;
+ case 0x3B: set_mnemonic(i, "HLECALL"); type=HLECALL; break;
+ default: set_mnemonic(i, "???"); type=NI;
SysPrintf("NI %08x @%08x (%08x)\n", source[i], addr + i*4, addr);
break;
}
dops[i].itype=type;
dops[i].opcode2=op2;
/* Get registers/immediates */
- dops[i].lt1=0;
+ dops[i].use_lt1=0;
gte_rs[i]=gte_rt[i]=0;
switch(type) {
case LOAD:
do_in_intrp=1;
}
}
- if(do_in_intrp) {
- dops[i-1].rs1=CCREG;
- dops[i-1].rs2=dops[i-1].rt1=dops[i-1].rt2=0;
- ba[i-1]=-1;
- dops[i-1].itype=INTCALL;
- done=2;
+ if (do_in_intrp) {
+ memset(&dops[i-1], 0, sizeof(dops[i-1]));
+ dops[i-1].itype = INTCALL;
+ dops[i-1].rs1 = CCREG;
+ ba[i-1] = -1;
+ done = 2;
i--; // don't compile the DS
}
}
// Don't get too close to the limit
if(i>MAXBLOCK/2) done=1;
}
- if(dops[i].itype==SYSCALL&&stop_after_jal) done=1;
- if(dops[i].itype==HLECALL||dops[i].itype==INTCALL) done=2;
- if(done==2) {
+ if (dops[i].itype == SYSCALL || dops[i].itype == HLECALL || dops[i].itype == INTCALL)
+ done = stop_after_jal ? 1 : 2;
+ if (done == 2) {
// Does the block continue due to a branch?
for(j=i-1;j>=0;j--)
{
assert(start+i*4<pagelimit);
if (i==MAXBLOCK-1) done=1;
// Stop if we're compiling junk
- if(dops[i].itype==NI&&dops[i].opcode==0x11) {
+ if(dops[i].itype == NI && (++ni_count > 8 || dops[i].opcode == 0x11)) {
done=stop_after_jal=1;
SysPrintf("Disabled speculative precompilation\n");
}
}
assert(slen>0);
- /* spacial hack(s) */
- if (i > 10 && source[i-1] == 0 && source[i-2] == 0x03e00008
- && source[i-4] == 0x8fbf0018 && source[i-6] == 0x00c0f809
- && dops[i-7].itype == STORE)
- {
- i = i-8;
- if (dops[i].itype == IMM16)
- i--;
- // swl r2, 15(r6); swr r2, 12(r6); sw r6, *; jalr r6
- if (dops[i].itype == STORELR && dops[i].rs1 == 6
- && dops[i-1].itype == STORELR && dops[i-1].rs1 == 6)
- {
- SysPrintf("F1 hack from %08x\n", start);
- if (f1_hack == 0)
- f1_hack = ~0u;
- }
- }
+ int clear_hack_addr = apply_hacks();
/* Pass 2 - Register dependencies and branch targets */
/* Pass 3 - Register allocation */
struct regstat current; // Current register allocations/status
- current.dirty=0;
- current.u=unneeded_reg[0];
+ clear_all_regs(current.regmap_entry);
clear_all_regs(current.regmap);
- alloc_reg(¤t,0,CCREG);
- dirty_reg(¤t,CCREG);
- current.isconst=0;
- current.wasconst=0;
- current.waswritten=0;
+ current.wasdirty = current.dirty = 0;
+ current.u = unneeded_reg[0];
+ alloc_reg(¤t, 0, CCREG);
+ dirty_reg(¤t, CCREG);
+ current.wasconst = 0;
+ current.isconst = 0;
+ current.loadedconst = 0;
+ current.waswritten = 0;
int ds=0;
int cc=0;
int hr=-1;
memcpy(regmap_pre[i],current.regmap,sizeof(current.regmap));
regs[i].wasconst=current.isconst;
regs[i].wasdirty=current.dirty;
+ regs[i].dirty=0;
+ regs[i].u=0;
+ regs[i].isconst=0;
regs[i].loadedconst=0;
if (!dops[i].is_jump) {
if(i+1<slen) {
current.u=branch_unneeded_reg[i]&~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
current.u&=~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
current.u|=1;
- } else { SysPrintf("oops, branch at end of block with no delay slot\n");abort(); }
+ } else {
+ SysPrintf("oops, branch at end of block with no delay slot @%08x\n", start + i*4);
+ abort();
+ }
}
dops[i].is_ds=ds;
if(ds) {
if(r!=regmap_pre[i][hr]) {
// TODO: delay slot (?)
or=get_reg(regmap_pre[i],r); // Get old mapping for this register
- if(or<0||(r&63)>=TEMPREG){
+ if(or<0||r>=TEMPREG){
regs[i].regmap_entry[hr]=-1;
}
else
// Just move it to a different register
regs[i].regmap_entry[hr]=r;
// If it was dirty before, it's still dirty
- if((regs[i].wasdirty>>or)&1) dirty_reg(¤t,r&63);
+ if((regs[i].wasdirty>>or)&1) dirty_reg(¤t,r);
}
}
else
// Merge in delay slot
for(hr=0;hr<HOST_REGS;hr++)
{
- if(dops[i+1].rt1&&dops[i+1].rt1==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
- if(dops[i+1].rt2&&dops[i+1].rt2==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
+ if(dops[i+1].rt1&&dops[i+1].rt1==regs[i].regmap[hr]) nr&=~(1<<hr);
+ if(dops[i+1].rt2&&dops[i+1].rt2==regs[i].regmap[hr]) nr&=~(1<<hr);
if(dops[i+1].rs1==regmap_pre[i][hr]) nr|=1<<hr;
if(dops[i+1].rs2==regmap_pre[i][hr]) nr|=1<<hr;
if(dops[i+1].rs1==regs[i].regmap_entry[hr]) nr|=1<<hr;
for(hr=0;hr<HOST_REGS;hr++)
{
// Overwritten registers are not needed
- if(dops[i].rt1&&dops[i].rt1==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
- if(dops[i].rt2&&dops[i].rt2==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
- if(FTEMP==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
+ if(dops[i].rt1&&dops[i].rt1==regs[i].regmap[hr]) nr&=~(1<<hr);
+ if(dops[i].rt2&&dops[i].rt2==regs[i].regmap[hr]) nr&=~(1<<hr);
+ if(FTEMP==regs[i].regmap[hr]) nr&=~(1<<hr);
// Source registers are needed
if(dops[i].rs1==regmap_pre[i][hr]) nr|=1<<hr;
if(dops[i].rs2==regmap_pre[i][hr]) nr|=1<<hr;
// might have to load the register before the branch.
if(i>0&&!dops[i].bt&&((regs[i].wasdirty>>hr)&1)) {
if((regmap_pre[i][hr]>0&&!((unneeded_reg[i]>>regmap_pre[i][hr])&1))) {
- if(dops[i-1].rt1==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(dops[i-1].rt2==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+ if(dops[i-1].rt1==regmap_pre[i][hr]) nr|=1<<hr;
+ if(dops[i-1].rt2==regmap_pre[i][hr]) nr|=1<<hr;
}
if((regs[i].regmap_entry[hr]>0&&!((unneeded_reg[i]>>regs[i].regmap_entry[hr])&1))) {
- if(dops[i-1].rt1==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- if(dops[i-1].rt2==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+ if(dops[i-1].rt1==regs[i].regmap_entry[hr]) nr|=1<<hr;
+ if(dops[i-1].rt2==regs[i].regmap_entry[hr]) nr|=1<<hr;
}
}
}
map2 = INVCP;
if(dops[i+1].itype==LOADLR || dops[i+1].itype==STORELR || dops[i+1].itype==C2LS)
temp = FTEMP;
- if((regs[i].regmap[hr]&63)!=dops[i].rs1 && (regs[i].regmap[hr]&63)!=dops[i].rs2 &&
- (regs[i].regmap[hr]&63)!=dops[i].rt1 && (regs[i].regmap[hr]&63)!=dops[i].rt2 &&
- (regs[i].regmap[hr]&63)!=dops[i+1].rt1 && (regs[i].regmap[hr]&63)!=dops[i+1].rt2 &&
+ if(regs[i].regmap[hr]!=dops[i].rs1 && regs[i].regmap[hr]!=dops[i].rs2 &&
+ regs[i].regmap[hr]!=dops[i].rt1 && regs[i].regmap[hr]!=dops[i].rt2 &&
+ regs[i].regmap[hr]!=dops[i+1].rt1 && regs[i].regmap[hr]!=dops[i+1].rt2 &&
regs[i].regmap[hr]!=dops[i+1].rs1 && regs[i].regmap[hr]!=dops[i+1].rs2 &&
- (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=PTEMP &&
+ regs[i].regmap[hr]!=temp && regs[i].regmap[hr]!=PTEMP &&
regs[i].regmap[hr]!=RHASH && regs[i].regmap[hr]!=RHTBL &&
regs[i].regmap[hr]!=RTEMP && regs[i].regmap[hr]!=CCREG &&
regs[i].regmap[hr]!=map1 && regs[i].regmap[hr]!=map2)
{
regs[i].regmap[hr]=-1;
regs[i].isconst&=~(1<<hr);
- if((branch_regs[i].regmap[hr]&63)!=dops[i].rs1 && (branch_regs[i].regmap[hr]&63)!=dops[i].rs2 &&
- (branch_regs[i].regmap[hr]&63)!=dops[i].rt1 && (branch_regs[i].regmap[hr]&63)!=dops[i].rt2 &&
- (branch_regs[i].regmap[hr]&63)!=dops[i+1].rt1 && (branch_regs[i].regmap[hr]&63)!=dops[i+1].rt2 &&
+ regs[i].dirty&=~(1<<hr);
+ regs[i+1].wasdirty&=~(1<<hr);
+ if(branch_regs[i].regmap[hr]!=dops[i].rs1 && branch_regs[i].regmap[hr]!=dops[i].rs2 &&
+ branch_regs[i].regmap[hr]!=dops[i].rt1 && branch_regs[i].regmap[hr]!=dops[i].rt2 &&
+ branch_regs[i].regmap[hr]!=dops[i+1].rt1 && branch_regs[i].regmap[hr]!=dops[i+1].rt2 &&
branch_regs[i].regmap[hr]!=dops[i+1].rs1 && branch_regs[i].regmap[hr]!=dops[i+1].rs2 &&
- (branch_regs[i].regmap[hr]&63)!=temp && branch_regs[i].regmap[hr]!=PTEMP &&
+ branch_regs[i].regmap[hr]!=temp && branch_regs[i].regmap[hr]!=PTEMP &&
branch_regs[i].regmap[hr]!=RHASH && branch_regs[i].regmap[hr]!=RHTBL &&
branch_regs[i].regmap[hr]!=RTEMP && branch_regs[i].regmap[hr]!=CCREG &&
branch_regs[i].regmap[hr]!=map1 && branch_regs[i].regmap[hr]!=map2)
map2 = INVCP;
if (dops[i].itype==LOADLR || dops[i].itype==STORELR || dops[i].itype==C2LS)
temp = FTEMP;
- if((regs[i].regmap[hr]&63)!=dops[i].rt1 && (regs[i].regmap[hr]&63)!=dops[i].rt2 &&
+ if(regs[i].regmap[hr]!=dops[i].rt1 && regs[i].regmap[hr]!=dops[i].rt2 &&
regs[i].regmap[hr]!=dops[i].rs1 && regs[i].regmap[hr]!=dops[i].rs2 &&
- (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=map1 && regs[i].regmap[hr]!=map2 &&
+ regs[i].regmap[hr]!=temp && regs[i].regmap[hr]!=map1 && regs[i].regmap[hr]!=map2 &&
//(dops[i].itype!=SPAN||regs[i].regmap[hr]!=CCREG)
regs[i].regmap[hr] != CCREG)
{
}
regs[i].regmap[hr]=-1;
regs[i].isconst&=~(1<<hr);
+ regs[i].dirty&=~(1<<hr);
+ regs[i+1].wasdirty&=~(1<<hr);
}
}
}
//printf("Test %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
if(r<34&&((unneeded_reg[j]>>r)&1)) break;
assert(r < 64);
- if(regs[j].regmap[hr]==f_regmap[hr]&&(f_regmap[hr]&63)<TEMPREG) {
+ if(regs[j].regmap[hr]==f_regmap[hr]&&f_regmap[hr]<TEMPREG) {
//printf("Hit %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
int k;
if(regs[i].regmap[hr]==-1&&branch_regs[i].regmap[hr]==-1) {
+ if(get_reg(regs[i].regmap,f_regmap[hr])>=0) break;
if(get_reg(regs[i+2].regmap,f_regmap[hr])>=0) break;
- if(r>63) {
- if(get_reg(regs[i].regmap,r&63)<0) break;
- if(get_reg(branch_regs[i].regmap,r&63)<0) break;
- }
k=i;
while(k>1&®s[k-1].regmap[hr]==-1) {
if(count_free_regs(regs[k-1].regmap)<=minimum_free_regs[k-1]) {
if(k>2&&(dops[k-3].itype==UJUMP||dops[k-3].itype==RJUMP)&&dops[k-3].rt1==31) {
break;
}
- assert(r < 64);
k--;
}
if(regs[k-1].regmap[hr]==f_regmap[hr]&®map_pre[k][hr]==f_regmap[hr]) {
}
}
// Load source into target register
- if(dops[i+1].lt1&&get_reg(regs[i+1].regmap,dops[i+1].rs1)<0) {
+ if(dops[i+1].use_lt1&&get_reg(regs[i+1].regmap,dops[i+1].rs1)<0) {
if((hr=get_reg(regs[i+1].regmap,dops[i+1].rt1))>=0)
{
if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0)
||(dops[i+1].opcode&0x3b)==0x39||(dops[i+1].opcode&0x3b)==0x3a) { // SB/SH/SW/SD/SWC1/SDC1/SWC2/SDC2
if(get_reg(regs[i+1].regmap,dops[i+1].rs1)<0) {
hr=get_reg2(regs[i].regmap,regs[i+1].regmap,-1);
- if(hr<0) hr=get_reg(regs[i+1].regmap,-1);
- else {regs[i+1].regmap[hr]=AGEN1+((i+1)&1);regs[i+1].isconst&=~(1<<hr);}
+ if(hr<0) hr=get_reg_temp(regs[i+1].regmap);
+ else {
+ regs[i+1].regmap[hr]=AGEN1+((i+1)&1);
+ regs[i+1].isconst&=~(1<<hr);
+ }
assert(hr>=0);
if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0)
{
hr=get_reg(regs[i+1].regmap,FTEMP);
if(dops[i+1].itype==STORE||dops[i+1].itype==STORELR||(dops[i+1].opcode&0x3b)==0x39||(dops[i+1].opcode&0x3b)==0x3a) { // SWC1/SDC1/SWC2/SDC2
hr=get_reg(regs[i+1].regmap,AGEN1+((i+1)&1));
- if(hr<0) hr=get_reg(regs[i+1].regmap,-1);
+ if(hr<0) hr=get_reg_temp(regs[i+1].regmap);
}
if(hr>=0&®s[i].regmap[hr]<0) {
int rs=get_reg(regs[i+1].regmap,dops[i+1].rs1);
dops[slen-1].bt=1; // Mark as a branch target so instruction can restart after exception
}
-#ifdef DISASM
+#ifdef REG_ALLOC_PRINT
/* Debug/disassembly */
for(i=0;i<slen;i++)
{
#endif
}
}
-#endif // DISASM
+#endif // REG_ALLOC_PRINT
/* Pass 8 - Assembly */
linkcount=0;stubcount=0;
}
for(i=0;i<slen;i++)
{
+ __builtin_prefetch(regs[i+1].regmap);
+ check_regmap(regmap_pre[i]);
+ check_regmap(regs[i].regmap_entry);
+ check_regmap(regs[i].regmap);
//if(ds) printf("ds: ");
disassemble_inst(i);
if(ds) {
instr_addr[i] = out;
assem_debug("<->\n");
drc_dbg_emit_do_cmp(i, ccadj[i]);
+ if (clear_hack_addr) {
+ emit_movimm(0, 0);
+ emit_writeword(0, &hack_addr);
+ clear_hack_addr = 0;
+ }
// load regs
if(regs[i].regmap_entry[HOST_CCREG]==CCREG&®s[i].regmap[HOST_CCREG]!=CCREG)