X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fnew_dynarec.c;h=77fc35794257a6bfb80f5611bcbc26a9c1826e88;hp=f45322a8f263e1ff59ef7f015d65d327d2764fba;hb=4b1c7cd1dd8da0be729bae21df9243a9fa33157b;hpb=32631e6a5d44d1e6aa5d53d5777a039b2d3d4300 diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c index f45322a8..77fc3579 100644 --- a/libpcsxcore/new_dynarec/new_dynarec.c +++ b/libpcsxcore/new_dynarec/new_dynarec.c @@ -116,7 +116,7 @@ enum stub_type { struct regstat { - signed char regmap_entry[HOST_REGS]; + signed char regmap_entry[HOST_REGS]; // pre-insn + loop preloaded regs? signed char regmap[HOST_REGS]; uint64_t wasdirty; uint64_t dirty; @@ -161,6 +161,25 @@ struct link_entry u_int ext; }; +static struct decoded_insn +{ + u_char itype; + u_char opcode; + u_char opcode2; + u_char rs1; + u_char rs2; + u_char rt1; + u_char rt2; + u_char lt1; + u_char bt:1; + u_char ooo:1; + u_char is_ds:1; + u_char is_jump:1; + u_char is_ujump:1; + u_char is_load:1; + u_char is_store:1; +} dops[MAXBLOCK]; + // used by asm: u_char *out; struct ht_entry hash_table[65536] __attribute__((aligned(16))); @@ -171,17 +190,6 @@ struct link_entry static u_int start; static u_int *source; static char insn[MAXBLOCK][10]; - static u_char itype[MAXBLOCK]; - static u_char opcode[MAXBLOCK]; - static u_char opcode2[MAXBLOCK]; - static u_char bt[MAXBLOCK]; - static u_char rs1[MAXBLOCK]; - static u_char rs2[MAXBLOCK]; - static u_char rt1[MAXBLOCK]; - static u_char rt2[MAXBLOCK]; - static u_char dep1[MAXBLOCK]; - static u_char dep2[MAXBLOCK]; - static u_char lt1[MAXBLOCK]; 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]; @@ -192,12 +200,10 @@ struct link_entry static u_int smrv_weak_next; static int imm[MAXBLOCK]; static u_int ba[MAXBLOCK]; - static char likely[MAXBLOCK]; - static char is_ds[MAXBLOCK]; - static char ooo[MAXBLOCK]; static uint64_t unneeded_reg[MAXBLOCK]; static uint64_t branch_unneeded_reg[MAXBLOCK]; - static signed char regmap_pre[MAXBLOCK][HOST_REGS]; // pre-instruction i? + // pre-instruction [i], excluding loop-preload regs? + 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 uint32_t current_constmap[HOST_REGS]; @@ -222,11 +228,7 @@ struct link_entry static void *copy; static int expirep; static u_int stop_after_jal; -#ifndef RAM_FIXED - static uintptr_t ram_offset; -#else - static const uintptr_t ram_offset=0; -#endif + static u_int f1_hack; // 0 - off, ~0 - capture address, else addr int new_dynarec_hacks; int new_dynarec_hacks_pergame; @@ -240,6 +242,7 @@ struct link_entry extern int pcaddr; extern int pending_exception; extern int branch_target; + extern uintptr_t ram_offset; extern uintptr_t mini_ht[32][2]; extern u_char restore_candidate[512]; @@ -252,7 +255,7 @@ struct link_entry #define CCREG 36 // Cycle count #define INVCP 37 // Pointer to invalid_code //#define MMREG 38 // Pointer to memory_map -//#define ROREG 39 // ram offset (if rdram!=0x80000000) +#define ROREG 39 // ram offset (if rdram!=0x80000000) #define TEMPREG 40 #define FTEMP 40 // FPU temporary register #define PTEMP 41 // Prefetch temporary register @@ -326,13 +329,13 @@ void call_gteStall(); void new_dyna_leave(); // Needed by assembler -static void wb_register(signed char r,signed char regmap[],uint64_t dirty); -static void wb_dirtys(signed char i_regmap[],uint64_t i_dirty); -static void wb_needed_dirtys(signed char i_regmap[],uint64_t i_dirty,int addr); -static void load_all_regs(signed char i_regmap[]); -static void load_needed_regs(signed char i_regmap[],signed char next_regmap[]); +static void wb_register(signed char r, const signed char regmap[], uint64_t dirty); +static void wb_dirtys(const signed char i_regmap[], uint64_t i_dirty); +static void wb_needed_dirtys(const signed char i_regmap[], uint64_t i_dirty, int addr); +static void load_all_regs(const signed char i_regmap[]); +static void load_needed_regs(const signed char i_regmap[], const signed char next_regmap[]); static void load_regs_entry(int t); -static void load_all_consts(signed char regmap[],u_int dirty,int i); +static void load_all_consts(const signed char regmap[], u_int dirty, int i); static u_int get_host_reglist(const signed char *regmap); static int verify_dirty(const u_int *ptr); @@ -342,7 +345,8 @@ static void add_stub(enum stub_type type, void *addr, void *retaddr, static void add_stub_r(enum stub_type type, void *addr, void *retaddr, int i, int addr_reg, const struct regstat *i_regs, int ccadj, u_int reglist); static void add_to_linker(void *addr, u_int target, int ext); -static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override); +static void *emit_fastpath_cmp_jump(int i, const struct regstat *i_regs, + int addr, int *offset_reg, int *addr_reg_override); static void *get_direct_memhandler(void *table, u_int addr, enum stub_type type, uintptr_t *addr_host); static void cop2_do_stall_check(u_int op, int i, const struct regstat *i_regs, u_int reglist); @@ -461,28 +465,21 @@ static void do_clear_cache(void) #define NO_CYCLE_PENALTY_THR 12 -int cycle_multiplier; // 100 for 1.0 +int cycle_multiplier = CYCLE_MULT_DEFAULT; // 100 for 1.0 int cycle_multiplier_override; int cycle_multiplier_old; static int CLOCK_ADJUST(int x) { - int m = cycle_multiplier_override + int m = cycle_multiplier_override && cycle_multiplier == CYCLE_MULT_DEFAULT ? cycle_multiplier_override : cycle_multiplier; int s=(x>>31)|1; return (x * m + s * 50) / 100; } -// is the op an unconditional jump? -static int is_ujump(int i) -{ - return itype[i] == UJUMP || itype[i] == RJUMP - || (source[i] >> 16) == 0x1000; // beq r0, r0, offset // b offset -} - -static int is_jump(int i) +static int ds_writes_rjump_rs(int i) { - return itype[i] == RJUMP || itype[i] == UJUMP || itype[i] == CJUMP || itype[i] == SJUMP; + return dops[i].rs1 != 0 && (dops[i].rs1 == dops[i+1].rt1 || dops[i].rs1 == dops[i+1].rt2); } static u_int get_page(u_int vaddr) @@ -692,7 +689,7 @@ void lsn(u_char hsn[], int i, int *preferred_reg) j=slen-i-1; break; } - if (is_ujump(i+j)) + if (dops[i+j].is_ujump) { // Don't go past an unconditonal jump j++; @@ -701,22 +698,23 @@ void lsn(u_char hsn[], int i, int *preferred_reg) } for(;j>=0;j--) { - if(rs1[i+j]) hsn[rs1[i+j]]=j; - if(rs2[i+j]) hsn[rs2[i+j]]=j; - if(rt1[i+j]) hsn[rt1[i+j]]=j; - if(rt2[i+j]) hsn[rt2[i+j]]=j; - if(itype[i+j]==STORE || itype[i+j]==STORELR) { + if(dops[i+j].rs1) hsn[dops[i+j].rs1]=j; + if(dops[i+j].rs2) hsn[dops[i+j].rs2]=j; + if(dops[i+j].rt1) hsn[dops[i+j].rt1]=j; + if(dops[i+j].rt2) hsn[dops[i+j].rt2]=j; + if(dops[i+j].itype==STORE || dops[i+j].itype==STORELR) { // Stores can allocate zero - hsn[rs1[i+j]]=j; - hsn[rs2[i+j]]=j; + hsn[dops[i+j].rs1]=j; + hsn[dops[i+j].rs2]=j; } + if (ram_offset && (dops[i+j].is_load || dops[i+j].is_store)) + hsn[ROREG] = j; // On some architectures stores need invc_ptr #if defined(HOST_IMM8) - if(itype[i+j]==STORE || itype[i+j]==STORELR || (opcode[i+j]&0x3b)==0x39 || (opcode[i+j]&0x3b)==0x3a) { - hsn[INVCP]=j; - } + if (dops[i+j].is_store) + hsn[INVCP] = j; #endif - if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP)) + if(i+j>=0&&(dops[i+j].itype==UJUMP||dops[i+j].itype==CJUMP||dops[i+j].itype==SJUMP)) { hsn[CCREG]=j; b=j; @@ -731,37 +729,37 @@ void lsn(u_char hsn[], int i, int *preferred_reg) j=7-b;if(t+j>=slen) j=slen-t-1; for(;j>=0;j--) { - if(rs1[t+j]) if(hsn[rs1[t+j]]>j+b+2) hsn[rs1[t+j]]=j+b+2; - if(rs2[t+j]) if(hsn[rs2[t+j]]>j+b+2) hsn[rs2[t+j]]=j+b+2; - //if(rt1[t+j]) if(hsn[rt1[t+j]]>j+b+2) hsn[rt1[t+j]]=j+b+2; - //if(rt2[t+j]) if(hsn[rt2[t+j]]>j+b+2) hsn[rt2[t+j]]=j+b+2; + if(dops[t+j].rs1) if(hsn[dops[t+j].rs1]>j+b+2) hsn[dops[t+j].rs1]=j+b+2; + if(dops[t+j].rs2) if(hsn[dops[t+j].rs2]>j+b+2) hsn[dops[t+j].rs2]=j+b+2; + //if(dops[t+j].rt1) if(hsn[dops[t+j].rt1]>j+b+2) hsn[dops[t+j].rt1]=j+b+2; + //if(dops[t+j].rt2) if(hsn[dops[t+j].rt2]>j+b+2) hsn[dops[t+j].rt2]=j+b+2; } } // TODO: preferred register based on backward branch } // Delay slot should preferably not overwrite branch conditions or cycle count - if (i > 0 && is_jump(i-1)) { - if(rs1[i-1]) if(hsn[rs1[i-1]]>1) hsn[rs1[i-1]]=1; - if(rs2[i-1]) if(hsn[rs2[i-1]]>1) hsn[rs2[i-1]]=1; + if (i > 0 && dops[i-1].is_jump) { + if(dops[i-1].rs1) if(hsn[dops[i-1].rs1]>1) hsn[dops[i-1].rs1]=1; + if(dops[i-1].rs2) if(hsn[dops[i-1].rs2]>1) hsn[dops[i-1].rs2]=1; hsn[CCREG]=1; // ...or hash tables hsn[RHASH]=1; hsn[RHTBL]=1; } // Coprocessor load/store needs FTEMP, even if not declared - if(itype[i]==C1LS||itype[i]==C2LS) { + if(dops[i].itype==C2LS) { hsn[FTEMP]=0; } // Load L/R also uses FTEMP as a temporary register - if(itype[i]==LOADLR) { + if(dops[i].itype==LOADLR) { hsn[FTEMP]=0; } // Also SWL/SWR/SDL/SDR - if(opcode[i]==0x2a||opcode[i]==0x2e||opcode[i]==0x2c||opcode[i]==0x2d) { + if(dops[i].opcode==0x2a||dops[i].opcode==0x2e||dops[i].opcode==0x2c||dops[i].opcode==0x2d) { hsn[FTEMP]=0; } // Don't remove the miniht registers - if(itype[i]==UJUMP||itype[i]==RJUMP) + if(dops[i].itype==UJUMP||dops[i].itype==RJUMP) { hsn[RHASH]=0; hsn[RHTBL]=0; @@ -775,7 +773,7 @@ int needed_again(int r, int i) int b=-1; int rn=10; - if (i > 0 && is_ujump(i-1)) + if (i > 0 && dops[i-1].is_ujump) { if(ba[i-1]start+slen*4-4) return 0; // Don't need any registers if exiting the block @@ -786,46 +784,27 @@ int needed_again(int r, int i) j=slen-i-1; break; } - if (is_ujump(i+j)) + if (dops[i+j].is_ujump) { // Don't go past an unconditonal jump j++; break; } - if(itype[i+j]==SYSCALL||itype[i+j]==HLECALL||itype[i+j]==INTCALL||((source[i+j]&0xfc00003f)==0x0d)) + if(dops[i+j].itype==SYSCALL||dops[i+j].itype==HLECALL||dops[i+j].itype==INTCALL||((source[i+j]&0xfc00003f)==0x0d)) { break; } } for(;j>=1;j--) { - if(rs1[i+j]==r) rn=j; - if(rs2[i+j]==r) rn=j; + if(dops[i+j].rs1==r) rn=j; + if(dops[i+j].rs2==r) rn=j; if((unneeded_reg[i+j]>>r)&1) rn=10; - if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP)) + if(i+j>=0&&(dops[i+j].itype==UJUMP||dops[i+j].itype==CJUMP||dops[i+j].itype==SJUMP)) { b=j; } } - /* - if(b>=0) - { - if(ba[i+b]>=start && ba[i+b]<(start+slen*4)) - { - // Follow first branch - int o=rn; - int t=(ba[i+b]-start)>>2; - j=7-b;if(t+j>=slen) j=slen-t-1; - for(;j>=0;j--) - { - if(!((unneeded_reg[t+j]>>r)&1)) { - if(rs1[t+j]==r) if(rn>j+b+2) rn=j+b+2; - if(rs2[t+j]==r) if(rn>j+b+2) rn=j+b+2; - } - else rn=o; - } - } - }*/ if(rn<10) return 1; (void)b; return 0; @@ -842,7 +821,7 @@ int loop_reg(int i, int r, int hr) j=slen-i-1; break; } - if (is_ujump(i+j)) + if (dops[i+j].is_ujump) { // Don't go past an unconditonal jump j++; @@ -851,14 +830,14 @@ int loop_reg(int i, int r, int hr) } k=0; if(i>0){ - if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP) + if(dops[i-1].itype==UJUMP||dops[i-1].itype==CJUMP||dops[i-1].itype==SJUMP) k--; } for(;k>r)&1) return hr; - if(i+k>=0&&(itype[i+k]==UJUMP||itype[i+k]==CJUMP||itype[i+k]==SJUMP)) + if(i+k>=0&&(dops[i+k].itype==UJUMP||dops[i+k].itype==CJUMP||dops[i+k].itype==SJUMP)) { if(ba[i+k]>=start && ba[i+k]<(start+i*4)) { @@ -881,8 +860,8 @@ void alloc_all(struct regstat *cur,int i) for(hr=0;hrregmap[hr]&63)!=rs1[i])&&((cur->regmap[hr]&63)!=rs2[i])&& - ((cur->regmap[hr]&63)!=rt1[i])&&((cur->regmap[hr]&63)!=rt2[i])) + 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)) { cur->regmap[hr]=-1; cur->dirty&=~(1<addr)>>shift)==(addr>>shift) || - ((uintptr_t)((*head)->addr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift)) + uintptr_t o1 = (u_char *)(*head)->addr - ndrc->translation_cache; + uintptr_t o2 = o1 - MAX_OUTPUT_BLOCK_SIZE; + if ((o1 >> shift) == base_offs_s || (o2 >> shift) == base_offs_s) { inv_debug("EXP: Remove pointer to %p (%x)\n",(*head)->addr,(*head)->vaddr); remove_hash((*head)->vaddr); @@ -1131,13 +1112,15 @@ void ll_clear(struct ll_entry **head) } // Dereference the pointers and remove if it matches -static void ll_kill_pointers(struct ll_entry *head,uintptr_t addr,int shift) +static void ll_kill_pointers(struct ll_entry *head, + uintptr_t base_offs_s, int shift) { while(head) { - uintptr_t ptr = (uintptr_t)get_pointer(head->addr); - inv_debug("EXP: Lookup pointer to %lx at %p (%x)\n",(long)ptr,head->addr,head->vaddr); - if(((ptr>>shift)==(addr>>shift)) || - (((ptr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift))) + u_char *ptr = get_pointer(head->addr); + uintptr_t o1 = ptr - ndrc->translation_cache; + uintptr_t o2 = o1 - MAX_OUTPUT_BLOCK_SIZE; + inv_debug("EXP: Lookup pointer to %p at %p (%x)\n",ptr,head->addr,head->vaddr); + if ((o1 >> shift) == base_offs_s || (o2 >> shift) == base_offs_s) { inv_debug("EXP: Kill pointer at %p (%x)\n",head->addr,head->vaddr); void *host_addr=find_extjump_insn(head->addr); @@ -1168,7 +1151,7 @@ static void invalidate_page(u_int page) inv_debug("INVALIDATE: kill pointer to %x (%p)\n",head->vaddr,head->addr); void *host_addr=find_extjump_insn(head->addr); mark_clear_cache(host_addr); - set_jump_target(host_addr, head->addr); + set_jump_target(host_addr, head->addr); // point back to dyna_linker next=head->next; free(head); head=next; @@ -1317,14 +1300,13 @@ static void do_invstub(int n) // Add an entry to jump_out after making a link // src should point to code by emit_extjump2() -void add_link(u_int vaddr,void *src) +void add_jump_out(u_int vaddr,void *src) { u_int page=get_page(vaddr); - inv_debug("add_link: %p -> %x (%d)\n",src,vaddr,page); + inv_debug("add_jump_out: %p -> %x (%d)\n",src,vaddr,page); check_extjump2(src); ll_add(jump_out+page,vaddr,src); - //void *ptr=get_pointer(src); - //inv_debug("add_link: Pointer is to %p\n",ptr); + //inv_debug("add_jump_out: to %p\n",get_pointer(src)); } // If a code block was found to be unmodified (bit was set in @@ -1383,9 +1365,11 @@ void clean_blocks(u_int page) static void alloc_reg(struct regstat *cur,int i,signed char reg) { int r,hr; - int preferred_reg = (reg&7); - if(reg==CCREG) preferred_reg=HOST_CCREG; - if(reg==PTEMP||reg==FTEMP) preferred_reg=12; + int preferred_reg = PREFERRED_REG_FIRST + + reg % (PREFERRED_REG_LAST - PREFERRED_REG_FIRST + 1); + if (reg == CCREG) preferred_reg = HOST_CCREG; + if (reg == PTEMP || reg == FTEMP) preferred_reg = 12; + assert(PREFERRED_REG_FIRST != EXCLUDE_REG && EXCLUDE_REG != HOST_REGS); // Don't allocate unused registers if((cur->u>>reg)&1) return; @@ -1429,28 +1413,47 @@ static void alloc_reg(struct regstat *cur,int i,signed char reg) if((cur->u>>r)&1) {cur->regmap[hr]=-1;break;} } } + // Try to allocate any available register, but prefer // registers that have not been used recently. - if(i>0) { - for(hr=0;hrregmap[hr]==-1) { - if(regs[i-1].regmap[hr]!=rs1[i-1]&®s[i-1].regmap[hr]!=rs2[i-1]&®s[i-1].regmap[hr]!=rt1[i-1]&®s[i-1].regmap[hr]!=rt2[i-1]) { + if (i > 0) { + for (hr = PREFERRED_REG_FIRST; ; ) { + if (cur->regmap[hr] < 0) { + int oldreg = regs[i-1].regmap[hr]; + if (oldreg < 0 || (oldreg != dops[i-1].rs1 && oldreg != dops[i-1].rs2 + && oldreg != dops[i-1].rt1 && oldreg != dops[i-1].rt2)) + { cur->regmap[hr]=reg; cur->dirty&=~(1<isconst&=~(1<regmap[hr]==-1) { + for (hr = PREFERRED_REG_FIRST; ; ) { + if (cur->regmap[hr] < 0) { cur->regmap[hr]=reg; cur->dirty&=~(1<isconst&=~(1<0) { // Don't evict the cycle count at entry points, otherwise the entry // stub will have to write it. - if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2; - if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP)) hsn[CCREG]=2; + if(dops[i].bt&&hsn[CCREG]>2) hsn[CCREG]=2; + if (i>1 && hsn[CCREG] > 2 && dops[i-2].is_jump) hsn[CCREG]=2; for(j=10;j>=3;j--) { // Alloc preferred register if available @@ -1483,7 +1486,7 @@ static void alloc_reg(struct regstat *cur,int i,signed char reg) } for(r=1;r<=MAXREG;r++) { - if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) { + if(hsn[r]==j&&r!=dops[i-1].rs1&&r!=dops[i-1].rs2&&r!=dops[i-1].rt1&&r!=dops[i-1].rt2) { for(hr=0;hrregmap[hr]==r) { @@ -1570,13 +1573,13 @@ static void alloc_reg_temp(struct regstat *cur,int i,signed char reg) if(i>0) { // Don't evict the cycle count at entry points, otherwise the entry // stub will have to write it. - if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2; - if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP)) hsn[CCREG]=2; + if(dops[i].bt&&hsn[CCREG]>2) hsn[CCREG]=2; + if (i>1 && hsn[CCREG] > 2 && dops[i-2].is_jump) hsn[CCREG]=2; for(j=10;j>=3;j--) { for(r=1;r<=MAXREG;r++) { - if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) { + if(hsn[r]==j&&r!=dops[i-1].rs1&&r!=dops[i-1].rs2&&r!=dops[i-1].rt1&&r!=dops[i-1].rt2) { for(hr=0;hr2) { if(cur->regmap[hr]==r) { @@ -1612,58 +1615,58 @@ static void alloc_reg_temp(struct regstat *cur,int i,signed char reg) static void mov_alloc(struct regstat *current,int i) { - if (rs1[i] == HIREG || rs1[i] == LOREG) { + if (dops[i].rs1 == HIREG || dops[i].rs1 == LOREG) { // logically this is needed but just won't work, no idea why //alloc_cc(current,i); // for stalls //dirty_reg(current,CCREG); } // Note: Don't need to actually alloc the source registers - //alloc_reg(current,i,rs1[i]); - alloc_reg(current,i,rt1[i]); + //alloc_reg(current,i,dops[i].rs1); + alloc_reg(current,i,dops[i].rt1); - clear_const(current,rs1[i]); - clear_const(current,rt1[i]); - dirty_reg(current,rt1[i]); + clear_const(current,dops[i].rs1); + clear_const(current,dops[i].rt1); + dirty_reg(current,dops[i].rt1); } static void shiftimm_alloc(struct regstat *current,int i) { - if(opcode2[i]<=0x3) // SLL/SRL/SRA + if(dops[i].opcode2<=0x3) // SLL/SRL/SRA { - if(rt1[i]) { - if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); - else lt1[i]=rs1[i]; - alloc_reg(current,i,rt1[i]); - dirty_reg(current,rt1[i]); - if(is_const(current,rs1[i])) { - int v=get_const(current,rs1[i]); - if(opcode2[i]==0x00) set_const(current,rt1[i],v<>imm[i]); - if(opcode2[i]==0x03) set_const(current,rt1[i],v>>imm[i]); + 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; + alloc_reg(current,i,dops[i].rt1); + dirty_reg(current,dops[i].rt1); + if(is_const(current,dops[i].rs1)) { + int v=get_const(current,dops[i].rs1); + if(dops[i].opcode2==0x00) set_const(current,dops[i].rt1,v<>imm[i]); + if(dops[i].opcode2==0x03) set_const(current,dops[i].rt1,v>>imm[i]); } - else clear_const(current,rt1[i]); + else clear_const(current,dops[i].rt1); } } else { - clear_const(current,rs1[i]); - clear_const(current,rt1[i]); + clear_const(current,dops[i].rs1); + clear_const(current,dops[i].rt1); } - if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA + if(dops[i].opcode2>=0x38&&dops[i].opcode2<=0x3b) // DSLL/DSRL/DSRA { assert(0); } - if(opcode2[i]==0x3c) // DSLL32 + if(dops[i].opcode2==0x3c) // DSLL32 { assert(0); } - if(opcode2[i]==0x3e) // DSRL32 + if(dops[i].opcode2==0x3e) // DSRL32 { assert(0); } - if(opcode2[i]==0x3f) // DSRA32 + if(dops[i].opcode2==0x3f) // DSRA32 { assert(0); } @@ -1671,125 +1674,128 @@ static void shiftimm_alloc(struct regstat *current,int i) static void shift_alloc(struct regstat *current,int i) { - if(rt1[i]) { - if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV + if(dops[i].rt1) { + if(dops[i].opcode2<=0x07) // SLLV/SRLV/SRAV { - 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]) { + if(dops[i].rs1) alloc_reg(current,i,dops[i].rs1); + if(dops[i].rs2) alloc_reg(current,i,dops[i].rs2); + alloc_reg(current,i,dops[i].rt1); + if(dops[i].rt1==dops[i].rs2) { alloc_reg_temp(current,i,-1); minimum_free_regs[i]=1; } } else { // DSLLV/DSRLV/DSRAV assert(0); } - clear_const(current,rs1[i]); - clear_const(current,rs2[i]); - clear_const(current,rt1[i]); - dirty_reg(current,rt1[i]); + clear_const(current,dops[i].rs1); + clear_const(current,dops[i].rs2); + clear_const(current,dops[i].rt1); + dirty_reg(current,dops[i].rt1); } } static void alu_alloc(struct regstat *current,int i) { - if(opcode2[i]>=0x20&&opcode2[i]<=0x23) { // ADD/ADDU/SUB/SUBU - if(rt1[i]) { - if(rs1[i]&&rs2[i]) { - alloc_reg(current,i,rs1[i]); - alloc_reg(current,i,rs2[i]); + if(dops[i].opcode2>=0x20&&dops[i].opcode2<=0x23) { // ADD/ADDU/SUB/SUBU + if(dops[i].rt1) { + if(dops[i].rs1&&dops[i].rs2) { + alloc_reg(current,i,dops[i].rs1); + alloc_reg(current,i,dops[i].rs2); } else { - if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); - if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg(current,i,rs2[i]); + if(dops[i].rs1&&needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1); + if(dops[i].rs2&&needed_again(dops[i].rs2,i)) alloc_reg(current,i,dops[i].rs2); } - alloc_reg(current,i,rt1[i]); + alloc_reg(current,i,dops[i].rt1); } } - if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU - if(rt1[i]) { - alloc_reg(current,i,rs1[i]); - alloc_reg(current,i,rs2[i]); - alloc_reg(current,i,rt1[i]); + if(dops[i].opcode2==0x2a||dops[i].opcode2==0x2b) { // SLT/SLTU + if(dops[i].rt1) { + alloc_reg(current,i,dops[i].rs1); + alloc_reg(current,i,dops[i].rs2); + alloc_reg(current,i,dops[i].rt1); } } - if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR - if(rt1[i]) { - if(rs1[i]&&rs2[i]) { - alloc_reg(current,i,rs1[i]); - alloc_reg(current,i,rs2[i]); + if(dops[i].opcode2>=0x24&&dops[i].opcode2<=0x27) { // AND/OR/XOR/NOR + if(dops[i].rt1) { + if(dops[i].rs1&&dops[i].rs2) { + alloc_reg(current,i,dops[i].rs1); + alloc_reg(current,i,dops[i].rs2); } else { - if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); - if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg(current,i,rs2[i]); + if(dops[i].rs1&&needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1); + if(dops[i].rs2&&needed_again(dops[i].rs2,i)) alloc_reg(current,i,dops[i].rs2); } - alloc_reg(current,i,rt1[i]); + alloc_reg(current,i,dops[i].rt1); } } - if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU + if(dops[i].opcode2>=0x2c&&dops[i].opcode2<=0x2f) { // DADD/DADDU/DSUB/DSUBU assert(0); } - clear_const(current,rs1[i]); - clear_const(current,rs2[i]); - clear_const(current,rt1[i]); - dirty_reg(current,rt1[i]); + clear_const(current,dops[i].rs1); + clear_const(current,dops[i].rs2); + clear_const(current,dops[i].rt1); + dirty_reg(current,dops[i].rt1); } static void imm16_alloc(struct regstat *current,int i) { - if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); - else lt1[i]=rs1[i]; - if(rt1[i]) alloc_reg(current,i,rt1[i]); - if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU + if(dops[i].rs1&&needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1); + else dops[i].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); } - else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU - clear_const(current,rs1[i]); - clear_const(current,rt1[i]); + else if(dops[i].opcode==0x0a||dops[i].opcode==0x0b) { // SLTI/SLTIU + clear_const(current,dops[i].rs1); + clear_const(current,dops[i].rt1); } - else if(opcode[i]>=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI - if(is_const(current,rs1[i])) { - int v=get_const(current,rs1[i]); - if(opcode[i]==0x0c) set_const(current,rt1[i],v&imm[i]); - if(opcode[i]==0x0d) set_const(current,rt1[i],v|imm[i]); - if(opcode[i]==0x0e) set_const(current,rt1[i],v^imm[i]); + else if(dops[i].opcode>=0x0c&&dops[i].opcode<=0x0e) { // ANDI/ORI/XORI + if(is_const(current,dops[i].rs1)) { + int v=get_const(current,dops[i].rs1); + if(dops[i].opcode==0x0c) set_const(current,dops[i].rt1,v&imm[i]); + if(dops[i].opcode==0x0d) set_const(current,dops[i].rt1,v|imm[i]); + if(dops[i].opcode==0x0e) set_const(current,dops[i].rt1,v^imm[i]); } - else clear_const(current,rt1[i]); + else clear_const(current,dops[i].rt1); } - else if(opcode[i]==0x08||opcode[i]==0x09) { // ADDI/ADDIU - if(is_const(current,rs1[i])) { - int v=get_const(current,rs1[i]); - set_const(current,rt1[i],v+imm[i]); + else if(dops[i].opcode==0x08||dops[i].opcode==0x09) { // ADDI/ADDIU + if(is_const(current,dops[i].rs1)) { + int v=get_const(current,dops[i].rs1); + set_const(current,dops[i].rt1,v+imm[i]); } - else clear_const(current,rt1[i]); + else clear_const(current,dops[i].rt1); } else { - set_const(current,rt1[i],imm[i]<<16); // LUI + set_const(current,dops[i].rt1,imm[i]<<16); // LUI } - dirty_reg(current,rt1[i]); + dirty_reg(current,dops[i].rt1); } static void load_alloc(struct regstat *current,int i) { - clear_const(current,rt1[i]); - //if(rs1[i]!=rt1[i]&&needed_again(rs1[i],i)) clear_const(current,rs1[i]); // Does this help or hurt? - if(!rs1[i]) current->u&=~1LL; // Allow allocating r0 if it's the source register - if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); - if(rt1[i]&&!((current->u>>rt1[i])&1)) { - alloc_reg(current,i,rt1[i]); - assert(get_reg(current->regmap,rt1[i])>=0); - if(opcode[i]==0x27||opcode[i]==0x37) // LWU/LD + clear_const(current,dops[i].rt1); + //if(dops[i].rs1!=dops[i].rt1&&needed_again(dops[i].rs1,i)) clear_const(current,dops[i].rs1); // Does this help or hurt? + if(!dops[i].rs1) current->u&=~1LL; // Allow allocating r0 if it's the source register + if (needed_again(dops[i].rs1, i)) + alloc_reg(current, i, dops[i].rs1); + if (ram_offset) + alloc_reg(current, i, ROREG); + if(dops[i].rt1&&!((current->u>>dops[i].rt1)&1)) { + alloc_reg(current,i,dops[i].rt1); + assert(get_reg(current->regmap,dops[i].rt1)>=0); + if(dops[i].opcode==0x27||dops[i].opcode==0x37) // LWU/LD { assert(0); } - else if(opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR + else if(dops[i].opcode==0x1A||dops[i].opcode==0x1B) // LDL/LDR { assert(0); } - dirty_reg(current,rt1[i]); + dirty_reg(current,dops[i].rt1); // LWL/LWR need a temporary register for the old value - if(opcode[i]==0x22||opcode[i]==0x26) + if(dops[i].opcode==0x22||dops[i].opcode==0x26) { alloc_reg(current,i,FTEMP); alloc_reg_temp(current,i,-1); @@ -1800,13 +1806,13 @@ static void load_alloc(struct regstat *current,int i) { // Load to r0 or unneeded register (dummy load) // but we still need a register to calculate the address - if(opcode[i]==0x22||opcode[i]==0x26) + if(dops[i].opcode==0x22||dops[i].opcode==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 + if(dops[i].opcode==0x1A||dops[i].opcode==0x1B) // LDL/LDR { assert(0); } @@ -1815,18 +1821,20 @@ static void load_alloc(struct regstat *current,int i) void store_alloc(struct regstat *current,int i) { - clear_const(current,rs2[i]); - if(!(rs2[i])) current->u&=~1LL; // Allow allocating r0 if necessary - if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); - alloc_reg(current,i,rs2[i]); - if(opcode[i]==0x2c||opcode[i]==0x2d||opcode[i]==0x3f) { // 64-bit SDL/SDR/SD + clear_const(current,dops[i].rs2); + if(!(dops[i].rs2)) current->u&=~1LL; // Allow allocating r0 if necessary + if(needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1); + alloc_reg(current,i,dops[i].rs2); + if(dops[i].opcode==0x2c||dops[i].opcode==0x2d||dops[i].opcode==0x3f) { // 64-bit SDL/SDR/SD assert(0); } + if (ram_offset) + alloc_reg(current, i, ROREG); #if defined(HOST_IMM8) // On CPUs without 32-bit immediates we need a pointer to invalid_code - else alloc_reg(current,i,INVCP); + alloc_reg(current, i, INVCP); #endif - if(opcode[i]==0x2a||opcode[i]==0x2e||opcode[i]==0x2c||opcode[i]==0x2d) { // SWL/SWL/SDL/SDR + if(dops[i].opcode==0x2a||dops[i].opcode==0x2e||dops[i].opcode==0x2c||dops[i].opcode==0x2d) { // SWL/SWL/SDL/SDR alloc_reg(current,i,FTEMP); } // We need a temporary register for address generation @@ -1836,31 +1844,20 @@ void store_alloc(struct regstat *current,int i) void c1ls_alloc(struct regstat *current,int i) { - //clear_const(current,rs1[i]); // FIXME - clear_const(current,rt1[i]); - if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); + clear_const(current,dops[i].rt1); alloc_reg(current,i,CSREG); // Status - alloc_reg(current,i,FTEMP); - if(opcode[i]==0x35||opcode[i]==0x3d) { // 64-bit LDC1/SDC1 - assert(0); - } - #if defined(HOST_IMM8) - // On CPUs without 32-bit immediates we need a pointer to invalid_code - else if((opcode[i]&0x3b)==0x39) // SWC1/SDC1 - alloc_reg(current,i,INVCP); - #endif - // We need a temporary register for address generation - alloc_reg_temp(current,i,-1); } void c2ls_alloc(struct regstat *current,int i) { - clear_const(current,rt1[i]); - if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); + clear_const(current,dops[i].rt1); + if(needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1); alloc_reg(current,i,FTEMP); + if (ram_offset) + alloc_reg(current, i, ROREG); #if defined(HOST_IMM8) // On CPUs without 32-bit immediates we need a pointer to invalid_code - if((opcode[i]&0x3b)==0x3a) // SWC2/SDC2 + if (dops[i].opcode == 0x3a) // SWC2 alloc_reg(current,i,INVCP); #endif // We need a temporary register for address generation @@ -1879,19 +1876,19 @@ void multdiv_alloc(struct regstat *current,int i) // case 0x1D: DMULTU // case 0x1E: DDIV // case 0x1F: DDIVU - clear_const(current,rs1[i]); - clear_const(current,rs2[i]); + clear_const(current,dops[i].rs1); + clear_const(current,dops[i].rs2); alloc_cc(current,i); // for stalls - if(rs1[i]&&rs2[i]) + if(dops[i].rs1&&dops[i].rs2) { - if((opcode2[i]&4)==0) // 32-bit + if((dops[i].opcode2&4)==0) // 32-bit { current->u&=~(1LL<u&=~(1LL< 3) // MTC2/CTC2 + else if (dops[i].opcode2 > 3) // MTC2/CTC2 { - if(rs1[i]){ - clear_const(current,rs1[i]); - alloc_reg(current,i,rs1[i]); + if(dops[i].rs1){ + clear_const(current,dops[i].rs1); + alloc_reg(current,i,dops[i].rs1); } else { current->u&=~1LL; @@ -1991,7 +1988,7 @@ void syscall_alloc(struct regstat *current,int i) void delayslot_alloc(struct regstat *current,int i) { - switch(itype[i]) { + switch(dops[i].itype) { case UJUMP: case CJUMP: case SJUMP: @@ -2059,28 +2056,28 @@ static void pagespan_alloc(struct regstat *current,int i) alloc_all(current,i); alloc_cc(current,i); dirty_reg(current,CCREG); - if(opcode[i]==3) // JAL + if(dops[i].opcode==3) // JAL { alloc_reg(current,i,31); dirty_reg(current,31); } - if(opcode[i]==0&&(opcode2[i]&0x3E)==8) // JR/JALR + if(dops[i].opcode==0&&(dops[i].opcode2&0x3E)==8) // JR/JALR { - alloc_reg(current,i,rs1[i]); - if (rt1[i]!=0) { - alloc_reg(current,i,rt1[i]); - dirty_reg(current,rt1[i]); + alloc_reg(current,i,dops[i].rs1); + if (dops[i].rt1!=0) { + alloc_reg(current,i,dops[i].rt1); + dirty_reg(current,dops[i].rt1); } } - if((opcode[i]&0x2E)==4) // BEQ/BNE/BEQL/BNEL + if((dops[i].opcode&0x2E)==4) // BEQ/BNE/BEQL/BNEL { - if(rs1[i]) alloc_reg(current,i,rs1[i]); - if(rs2[i]) alloc_reg(current,i,rs2[i]); + if(dops[i].rs1) alloc_reg(current,i,dops[i].rs1); + if(dops[i].rs2) alloc_reg(current,i,dops[i].rs2); } else - if((opcode[i]&0x2E)==6) // BLEZ/BGTZ/BLEZL/BGTZL + if((dops[i].opcode&0x2E)==6) // BLEZ/BGTZ/BLEZL/BGTZL { - if(rs1[i]) alloc_reg(current,i,rs1[i]); + if(dops[i].rs1) alloc_reg(current,i,dops[i].rs1); } //else ... } @@ -2107,7 +2104,7 @@ static void add_stub_r(enum stub_type type, void *addr, void *retaddr, } // Write out a single register -static void wb_register(signed char r,signed char regmap[],uint64_t dirty) +static void wb_register(signed char r, const signed char regmap[], uint64_t dirty) { int hr; for(hr=0;hr=0x20&&opcode2[i]<=0x23) { // ADD/ADDU/SUB/SUBU - if(rt1[i]) { + if(dops[i].opcode2>=0x20&&dops[i].opcode2<=0x23) { // ADD/ADDU/SUB/SUBU + if(dops[i].rt1) { signed char s1,s2,t; - t=get_reg(i_regs->regmap,rt1[i]); + t=get_reg(i_regs->regmap,dops[i].rt1); if(t>=0) { - s1=get_reg(i_regs->regmap,rs1[i]); - s2=get_reg(i_regs->regmap,rs2[i]); - if(rs1[i]&&rs2[i]) { + s1=get_reg(i_regs->regmap,dops[i].rs1); + s2=get_reg(i_regs->regmap,dops[i].rs2); + if(dops[i].rs1&&dops[i].rs2) { assert(s1>=0); assert(s2>=0); - if(opcode2[i]&2) emit_sub(s1,s2,t); + if(dops[i].opcode2&2) emit_sub(s1,s2,t); else emit_add(s1,s2,t); } - else if(rs1[i]) { + else if(dops[i].rs1) { if(s1>=0) emit_mov(s1,t); - else emit_loadreg(rs1[i],t); + else emit_loadreg(dops[i].rs1,t); } - else if(rs2[i]) { + else if(dops[i].rs2) { if(s2>=0) { - if(opcode2[i]&2) emit_neg(s2,t); + if(dops[i].opcode2&2) emit_neg(s2,t); else emit_mov(s2,t); } else { - emit_loadreg(rs2[i],t); - if(opcode2[i]&2) emit_neg(t,t); + emit_loadreg(dops[i].rs2,t); + if(dops[i].opcode2&2) emit_neg(t,t); } } else emit_zeroreg(t); } } } - if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU + if(dops[i].opcode2>=0x2c&&dops[i].opcode2<=0x2f) { // DADD/DADDU/DSUB/DSUBU assert(0); } - if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU - if(rt1[i]) { + if(dops[i].opcode2==0x2a||dops[i].opcode2==0x2b) { // SLT/SLTU + if(dops[i].rt1) { signed char s1l,s2l,t; { - t=get_reg(i_regs->regmap,rt1[i]); + t=get_reg(i_regs->regmap,dops[i].rt1); //assert(t>=0); if(t>=0) { - s1l=get_reg(i_regs->regmap,rs1[i]); - s2l=get_reg(i_regs->regmap,rs2[i]); - if(rs2[i]==0) // rxregmap,dops[i].rs1); + s2l=get_reg(i_regs->regmap,dops[i].rs2); + if(dops[i].rs2==0) // rx=0); emit_shrimm(s1l,31,t); } else // SLTU (unsigned can not be less than zero, 0<0) emit_zeroreg(t); } - else if(rs1[i]==0) // r0=0); - if(opcode2[i]==0x2a) // SLT + if(dops[i].opcode2==0x2a) // SLT emit_set_gz32(s2l,t); else // SLTU (set if not zero) emit_set_nz32(s2l,t); } else{ assert(s1l>=0);assert(s2l>=0); - if(opcode2[i]==0x2a) // SLT + if(dops[i].opcode2==0x2a) // SLT emit_set_if_less32(s1l,s2l,t); else // SLTU emit_set_if_carry32(s1l,s2l,t); @@ -2235,61 +2232,61 @@ static void alu_assemble(int i,struct regstat *i_regs) } } } - if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR - if(rt1[i]) { + if(dops[i].opcode2>=0x24&&dops[i].opcode2<=0x27) { // AND/OR/XOR/NOR + if(dops[i].rt1) { signed char s1l,s2l,tl; - tl=get_reg(i_regs->regmap,rt1[i]); + tl=get_reg(i_regs->regmap,dops[i].rt1); { if(tl>=0) { - s1l=get_reg(i_regs->regmap,rs1[i]); - s2l=get_reg(i_regs->regmap,rs2[i]); - if(rs1[i]&&rs2[i]) { + s1l=get_reg(i_regs->regmap,dops[i].rs1); + s2l=get_reg(i_regs->regmap,dops[i].rs2); + if(dops[i].rs1&&dops[i].rs2) { assert(s1l>=0); assert(s2l>=0); - if(opcode2[i]==0x24) { // AND + if(dops[i].opcode2==0x24) { // AND emit_and(s1l,s2l,tl); } else - if(opcode2[i]==0x25) { // OR + if(dops[i].opcode2==0x25) { // OR emit_or(s1l,s2l,tl); } else - if(opcode2[i]==0x26) { // XOR + if(dops[i].opcode2==0x26) { // XOR emit_xor(s1l,s2l,tl); } else - if(opcode2[i]==0x27) { // NOR + if(dops[i].opcode2==0x27) { // NOR emit_or(s1l,s2l,tl); emit_not(tl,tl); } } else { - if(opcode2[i]==0x24) { // AND + if(dops[i].opcode2==0x24) { // AND emit_zeroreg(tl); } else - if(opcode2[i]==0x25||opcode2[i]==0x26) { // OR/XOR - if(rs1[i]){ + if(dops[i].opcode2==0x25||dops[i].opcode2==0x26) { // OR/XOR + if(dops[i].rs1){ if(s1l>=0) emit_mov(s1l,tl); - else emit_loadreg(rs1[i],tl); // CHECK: regmap_entry? + else emit_loadreg(dops[i].rs1,tl); // CHECK: regmap_entry? } else - if(rs2[i]){ + if(dops[i].rs2){ if(s2l>=0) emit_mov(s2l,tl); - else emit_loadreg(rs2[i],tl); // CHECK: regmap_entry? + else emit_loadreg(dops[i].rs2,tl); // CHECK: regmap_entry? } else emit_zeroreg(tl); } else - if(opcode2[i]==0x27) { // NOR - if(rs1[i]){ + if(dops[i].opcode2==0x27) { // NOR + if(dops[i].rs1){ if(s1l>=0) emit_not(s1l,tl); else { - emit_loadreg(rs1[i],tl); + emit_loadreg(dops[i].rs1,tl); emit_not(tl,tl); } } else - if(rs2[i]){ + if(dops[i].rs2){ if(s2l>=0) emit_not(s2l,tl); else { - emit_loadreg(rs2[i],tl); + emit_loadreg(dops[i].rs2,tl); emit_not(tl,tl); } } @@ -2302,12 +2299,12 @@ static void alu_assemble(int i,struct regstat *i_regs) } } -void imm16_assemble(int i,struct regstat *i_regs) +static void imm16_assemble(int i, const struct regstat *i_regs) { - if (opcode[i]==0x0f) { // LUI - if(rt1[i]) { + if (dops[i].opcode==0x0f) { // LUI + if(dops[i].rt1) { signed char t; - t=get_reg(i_regs->regmap,rt1[i]); + t=get_reg(i_regs->regmap,dops[i].rt1); //assert(t>=0); if(t>=0) { if(!((i_regs->isconst>>t)&1)) @@ -2315,18 +2312,18 @@ void imm16_assemble(int i,struct regstat *i_regs) } } } - if(opcode[i]==0x08||opcode[i]==0x09) { // ADDI/ADDIU - if(rt1[i]) { + if(dops[i].opcode==0x08||dops[i].opcode==0x09) { // ADDI/ADDIU + if(dops[i].rt1) { signed char s,t; - t=get_reg(i_regs->regmap,rt1[i]); - s=get_reg(i_regs->regmap,rs1[i]); - if(rs1[i]) { + t=get_reg(i_regs->regmap,dops[i].rt1); + s=get_reg(i_regs->regmap,dops[i].rs1); + if(dops[i].rs1) { //assert(t>=0); //assert(s>=0); if(t>=0) { if(!((i_regs->isconst>>t)&1)) { if(s<0) { - if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t); + if(i_regs->regmap_entry[t]!=dops[i].rs1) emit_loadreg(dops[i].rs1,t); emit_addimm(t,imm[i],t); }else{ if(!((i_regs->wasconst>>s)&1)) @@ -2344,13 +2341,13 @@ void imm16_assemble(int i,struct regstat *i_regs) } } } - if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU - if(rt1[i]) { + if(dops[i].opcode==0x18||dops[i].opcode==0x19) { // DADDI/DADDIU + if(dops[i].rt1) { signed char sl,tl; - tl=get_reg(i_regs->regmap,rt1[i]); - sl=get_reg(i_regs->regmap,rs1[i]); + tl=get_reg(i_regs->regmap,dops[i].rt1); + sl=get_reg(i_regs->regmap,dops[i].rs1); if(tl>=0) { - if(rs1[i]) { + if(dops[i].rs1) { assert(sl>=0); emit_addimm(sl,imm[i],tl); } else { @@ -2359,18 +2356,18 @@ void imm16_assemble(int i,struct regstat *i_regs) } } } - else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU - if(rt1[i]) { - //assert(rs1[i]!=0); // r0 might be valid, but it's probably a bug + else if(dops[i].opcode==0x0a||dops[i].opcode==0x0b) { // SLTI/SLTIU + if(dops[i].rt1) { + //assert(dops[i].rs1!=0); // r0 might be valid, but it's probably a bug signed char sl,t; - t=get_reg(i_regs->regmap,rt1[i]); - sl=get_reg(i_regs->regmap,rs1[i]); + t=get_reg(i_regs->regmap,dops[i].rt1); + sl=get_reg(i_regs->regmap,dops[i].rs1); //assert(t>=0); if(t>=0) { - if(rs1[i]>0) { - if(opcode[i]==0x0a) { // SLTI + if(dops[i].rs1>0) { + if(dops[i].opcode==0x0a) { // SLTI if(sl<0) { - if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t); + if(i_regs->regmap_entry[t]!=dops[i].rs1) emit_loadreg(dops[i].rs1,t); emit_slti32(t,imm[i],t); }else{ emit_slti32(sl,imm[i],t); @@ -2378,7 +2375,7 @@ void imm16_assemble(int i,struct regstat *i_regs) } else { // SLTIU if(sl<0) { - if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t); + if(i_regs->regmap_entry[t]!=dops[i].rs1) emit_loadreg(dops[i].rs1,t); emit_sltiu32(t,imm[i],t); }else{ emit_sltiu32(sl,imm[i],t); @@ -2387,7 +2384,7 @@ void imm16_assemble(int i,struct regstat *i_regs) }else{ // SLTI(U) with r0 is just stupid, // nonetheless examples can be found - if(opcode[i]==0x0a) // SLTI + if(dops[i].opcode==0x0a) // SLTI if(0=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI - if(rt1[i]) { + else if(dops[i].opcode>=0x0c&&dops[i].opcode<=0x0e) { // ANDI/ORI/XORI + if(dops[i].rt1) { signed char sl,tl; - tl=get_reg(i_regs->regmap,rt1[i]); - sl=get_reg(i_regs->regmap,rs1[i]); + tl=get_reg(i_regs->regmap,dops[i].rt1); + sl=get_reg(i_regs->regmap,dops[i].rs1); if(tl>=0 && !((i_regs->isconst>>tl)&1)) { - if(opcode[i]==0x0c) //ANDI + if(dops[i].opcode==0x0c) //ANDI { - if(rs1[i]) { + if(dops[i].rs1) { if(sl<0) { - if(i_regs->regmap_entry[tl]!=rs1[i]) emit_loadreg(rs1[i],tl); + if(i_regs->regmap_entry[tl]!=dops[i].rs1) emit_loadreg(dops[i].rs1,tl); emit_andimm(tl,imm[i],tl); }else{ if(!((i_regs->wasconst>>sl)&1)) @@ -2423,11 +2420,11 @@ void imm16_assemble(int i,struct regstat *i_regs) } else { - if(rs1[i]) { + if(dops[i].rs1) { if(sl<0) { - if(i_regs->regmap_entry[tl]!=rs1[i]) emit_loadreg(rs1[i],tl); + if(i_regs->regmap_entry[tl]!=dops[i].rs1) emit_loadreg(dops[i].rs1,tl); } - if(opcode[i]==0x0d) { // ORI + if(dops[i].opcode==0x0d) { // ORI if(sl<0) { emit_orimm(tl,imm[i],tl); }else{ @@ -2437,7 +2434,7 @@ void imm16_assemble(int i,struct regstat *i_regs) emit_movimm(constmap[i][sl]|imm[i],tl); } } - if(opcode[i]==0x0e) { // XORI + if(dops[i].opcode==0x0e) { // XORI if(sl<0) { emit_xorimm(tl,imm[i],tl); }else{ @@ -2457,33 +2454,33 @@ void imm16_assemble(int i,struct regstat *i_regs) } } -void shiftimm_assemble(int i,struct regstat *i_regs) +static void shiftimm_assemble(int i, const struct regstat *i_regs) { - if(opcode2[i]<=0x3) // SLL/SRL/SRA + if(dops[i].opcode2<=0x3) // SLL/SRL/SRA { - if(rt1[i]) { + if(dops[i].rt1) { signed char s,t; - t=get_reg(i_regs->regmap,rt1[i]); - s=get_reg(i_regs->regmap,rs1[i]); + t=get_reg(i_regs->regmap,dops[i].rt1); + s=get_reg(i_regs->regmap,dops[i].rs1); //assert(t>=0); if(t>=0&&!((i_regs->isconst>>t)&1)){ - if(rs1[i]==0) + if(dops[i].rs1==0) { emit_zeroreg(t); } else { - if(s<0&&i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t); + if(s<0&&i_regs->regmap_entry[t]!=dops[i].rs1) emit_loadreg(dops[i].rs1,t); if(imm[i]) { - if(opcode2[i]==0) // SLL + if(dops[i].opcode2==0) // SLL { emit_shlimm(s<0?t:s,imm[i],t); } - if(opcode2[i]==2) // SRL + if(dops[i].opcode2==2) // SRL { emit_shrimm(s<0?t:s,imm[i],t); } - if(opcode2[i]==3) // SRA + if(dops[i].opcode2==3) // SRA { emit_sarimm(s<0?t:s,imm[i],t); } @@ -2493,50 +2490,50 @@ void shiftimm_assemble(int i,struct regstat *i_regs) } } } - //emit_storereg(rt1[i],t); //DEBUG + //emit_storereg(dops[i].rt1,t); //DEBUG } } - if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA + if(dops[i].opcode2>=0x38&&dops[i].opcode2<=0x3b) // DSLL/DSRL/DSRA { assert(0); } - if(opcode2[i]==0x3c) // DSLL32 + if(dops[i].opcode2==0x3c) // DSLL32 { assert(0); } - if(opcode2[i]==0x3e) // DSRL32 + if(dops[i].opcode2==0x3e) // DSRL32 { assert(0); } - if(opcode2[i]==0x3f) // DSRA32 + if(dops[i].opcode2==0x3f) // DSRA32 { assert(0); } } #ifndef shift_assemble -static void shift_assemble(int i,struct regstat *i_regs) +static void shift_assemble(int i, const struct regstat *i_regs) { signed char s,t,shift; - if (rt1[i] == 0) + if (dops[i].rt1 == 0) return; - assert(opcode2[i]<=0x07); // SLLV/SRLV/SRAV - t = get_reg(i_regs->regmap, rt1[i]); - s = get_reg(i_regs->regmap, rs1[i]); - shift = get_reg(i_regs->regmap, rs2[i]); + assert(dops[i].opcode2<=0x07); // SLLV/SRLV/SRAV + t = get_reg(i_regs->regmap, dops[i].rt1); + s = get_reg(i_regs->regmap, dops[i].rs1); + shift = get_reg(i_regs->regmap, dops[i].rs2); if (t < 0) return; - if(rs1[i]==0) + if(dops[i].rs1==0) emit_zeroreg(t); - else if(rs2[i]==0) { + else if(dops[i].rs2==0) { assert(s>=0); if(s!=t) emit_mov(s,t); } else { host_tempreg_acquire(); emit_andimm(shift,31,HOST_TEMPREG); - switch(opcode2[i]) { + switch(dops[i].opcode2) { case 4: // SLLV emit_shl(s,HOST_TEMPREG,t); break; @@ -2581,11 +2578,25 @@ static int get_ptr_mem_type(u_int a) return MTYPE_8000; } -static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override) +static int get_ro_reg(const struct regstat *i_regs, int host_tempreg_free) +{ + int r = get_reg(i_regs->regmap, ROREG); + if (r < 0 && host_tempreg_free) { + host_tempreg_acquire(); + emit_loadreg(ROREG, r = HOST_TEMPREG); + } + if (r < 0) + abort(); + return r; +} + +static void *emit_fastpath_cmp_jump(int i, const struct regstat *i_regs, + int addr, int *offset_reg, int *addr_reg_override) { void *jaddr = NULL; - int type=0; - int mr=rs1[i]; + int type = 0; + int mr = dops[i].rs1; + *offset_reg = -1; if(((smrv_strong|smrv_weak)>>mr)&1) { type=get_ptr_mem_type(smrv[mr]); //printf("set %08x @%08x r%d %d\n", smrv[mr], start+i*4, mr, type); @@ -2629,22 +2640,19 @@ static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override) } } - if(type==0) + if (type == 0) // need ram check { emit_cmpimm(addr,RAM_SIZE); - jaddr=out; + jaddr = out; #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK // Hint to branch predictor that the branch is unlikely to be taken - if(rs1[i]>=28) + if (dops[i].rs1 >= 28) emit_jno_unlikely(0); else #endif emit_jno(0); - if(ram_offset!=0) { - host_tempreg_acquire(); - emit_addimm(addr,ram_offset,HOST_TEMPREG); - addr=*addr_reg_override=HOST_TEMPREG; - } + if (ram_offset != 0) + *offset_reg = get_ro_reg(i_regs, 0); } return jaddr; @@ -2654,9 +2662,10 @@ static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override) static void *get_direct_memhandler(void *table, u_int addr, enum stub_type type, uintptr_t *addr_host) { + uintptr_t msb = 1ull << (sizeof(uintptr_t)*8 - 1); uintptr_t l1, l2 = 0; l1 = ((uintptr_t *)table)[addr>>12]; - if ((l1 & (1ul << (sizeof(l1)*8-1))) == 0) { + if (!(l1 & msb)) { uintptr_t v = l1 << 1; *addr_host = v + addr; return NULL; @@ -2666,10 +2675,10 @@ static void *get_direct_memhandler(void *table, u_int addr, if (type == LOADB_STUB || type == LOADBU_STUB || type == STOREB_STUB) l2 = ((uintptr_t *)l1)[0x1000/4 + 0x1000/2 + (addr&0xfff)]; else if (type == LOADH_STUB || type == LOADHU_STUB || type == STOREH_STUB) - l2=((uintptr_t *)l1)[0x1000/4 + (addr&0xfff)/2]; + l2 = ((uintptr_t *)l1)[0x1000/4 + (addr&0xfff)/2]; else - l2=((uintptr_t *)l1)[(addr&0xfff)/4]; - if ((l2 & (1<<31)) == 0) { + l2 = ((uintptr_t *)l1)[(addr&0xfff)/4]; + if (!(l2 & msb)) { uintptr_t v = l2 << 1; *addr_host = v + (addr&0xfff); return NULL; @@ -2706,16 +2715,59 @@ static int reglist_find_free(u_int reglist) return __builtin_ctz(free_regs); } -static void load_assemble(int i, const struct regstat *i_regs) +static void do_load_word(int a, int rt, int offset_reg) +{ + if (offset_reg >= 0) + emit_ldr_dualindexed(offset_reg, a, rt); + else + emit_readword_indexed(0, a, rt); +} + +static void do_store_word(int a, int ofs, int rt, int offset_reg, int preseve_a) +{ + if (offset_reg < 0) { + emit_writeword_indexed(rt, ofs, a); + return; + } + if (ofs != 0) + emit_addimm(a, ofs, a); + emit_str_dualindexed(offset_reg, a, rt); + if (ofs != 0 && preseve_a) + emit_addimm(a, -ofs, a); +} + +static void do_store_hword(int a, int ofs, int rt, int offset_reg, int preseve_a) +{ + if (offset_reg < 0) { + emit_writehword_indexed(rt, ofs, a); + return; + } + if (ofs != 0) + emit_addimm(a, ofs, a); + emit_strh_dualindexed(offset_reg, a, rt); + if (ofs != 0 && preseve_a) + emit_addimm(a, -ofs, a); +} + +static void do_store_byte(int a, int rt, int offset_reg) +{ + if (offset_reg >= 0) + emit_strb_dualindexed(offset_reg, a, rt); + else + emit_writebyte_indexed(rt, 0, a); +} + +static void load_assemble(int i, const struct regstat *i_regs, int ccadj_) { int s,tl,addr; int offset; void *jaddr=0; int memtarget=0,c=0; - int fastio_reg_override=-1; + int offset_reg = -1; + int fastio_reg_override = -1; u_int reglist=get_host_reglist(i_regs->regmap); - tl=get_reg(i_regs->regmap,rt1[i]); - s=get_reg(i_regs->regmap,rs1[i]); + tl=get_reg(i_regs->regmap,dops[i].rt1); + s=get_reg(i_regs->regmap,dops[i].rs1); offset=imm[i]; if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<=0) { @@ -2728,7 +2780,7 @@ static void load_assemble(int i, const struct regstat *i_regs) //if(c) printf("load_assemble: const=%lx\n",(long)constmap[i][s]+offset); // FIXME: Even if the load is a NOP, we should check for pagefaults... if((tl<0&&(!c||(((u_int)constmap[i][s]+offset)>>16)==0x1f80)) - ||rt1[i]==0) { + ||dops[i].rt1==0) { // could be FIFO, must perform the read // ||dummy read assem_debug("(forced read)\n"); @@ -2746,113 +2798,128 @@ static void load_assemble(int i, const struct regstat *i_regs) if(!c) { #ifdef R29_HACK // Strmnnrmn's speed hack - if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE) + if(dops[i].rs1!=29||start<0x80001000||start>=0x80000000+RAM_SIZE) #endif { - jaddr=emit_fastpath_cmp_jump(i,addr,&fastio_reg_override); + jaddr = emit_fastpath_cmp_jump(i, i_regs, addr, + &offset_reg, &fastio_reg_override); } } - else if(ram_offset&&memtarget) { - host_tempreg_acquire(); - emit_addimm(addr,ram_offset,HOST_TEMPREG); - fastio_reg_override=HOST_TEMPREG; + else if (ram_offset && memtarget) { + offset_reg = get_ro_reg(i_regs, 0); } - 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 + int dummy=(dops[i].rt1==0)||(tl!=get_reg(i_regs->regmap,dops[i].rt1)); // ignore loads to r0 and unneeded reg + switch (dops[i].opcode) { + case 0x20: // LB if(!c||memtarget) { if(!dummy) { - { - int x=0,a=tl; - if(!c) a=addr; - if(fastio_reg_override>=0) a=fastio_reg_override; + int a = tl; + if (!c) a = addr; + if (fastio_reg_override >= 0) + a = fastio_reg_override; - emit_movsbl_indexed(x,a,tl); - } + if (offset_reg >= 0) + emit_ldrsb_dualindexed(offset_reg, a, tl); + else + emit_movsbl_indexed(0, a, tl); } if(jaddr) - add_stub_r(LOADB_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist); + add_stub_r(LOADB_STUB,jaddr,out,i,addr,i_regs,ccadj_,reglist); } else - inline_readstub(LOADB_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); - } - if (opcode[i]==0x21) { // LH + inline_readstub(LOADB_STUB,i,constmap[i][s]+offset,i_regs->regmap,dops[i].rt1,ccadj_,reglist); + break; + case 0x21: // LH if(!c||memtarget) { if(!dummy) { - int x=0,a=tl; - if(!c) a=addr; - if(fastio_reg_override>=0) a=fastio_reg_override; - emit_movswl_indexed(x,a,tl); + int a = tl; + if (!c) a = addr; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + if (offset_reg >= 0) + emit_ldrsh_dualindexed(offset_reg, a, tl); + else + emit_movswl_indexed(0, a, tl); } if(jaddr) - add_stub_r(LOADH_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist); + add_stub_r(LOADH_STUB,jaddr,out,i,addr,i_regs,ccadj_,reglist); } else - inline_readstub(LOADH_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); - } - if (opcode[i]==0x23) { // LW + inline_readstub(LOADH_STUB,i,constmap[i][s]+offset,i_regs->regmap,dops[i].rt1,ccadj_,reglist); + break; + case 0x23: // LW if(!c||memtarget) { if(!dummy) { - int a=addr; - if(fastio_reg_override>=0) a=fastio_reg_override; - emit_readword_indexed(0,a,tl); + int a = addr; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + do_load_word(a, tl, offset_reg); } if(jaddr) - add_stub_r(LOADW_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist); + add_stub_r(LOADW_STUB,jaddr,out,i,addr,i_regs,ccadj_,reglist); } else - inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); - } - if (opcode[i]==0x24) { // LBU + inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,dops[i].rt1,ccadj_,reglist); + break; + case 0x24: // LBU if(!c||memtarget) { if(!dummy) { - int x=0,a=tl; - if(!c) a=addr; - if(fastio_reg_override>=0) a=fastio_reg_override; + int a = tl; + if (!c) a = addr; + if (fastio_reg_override >= 0) + a = fastio_reg_override; - emit_movzbl_indexed(x,a,tl); + if (offset_reg >= 0) + emit_ldrb_dualindexed(offset_reg, a, tl); + else + emit_movzbl_indexed(0, a, tl); } if(jaddr) - add_stub_r(LOADBU_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist); + add_stub_r(LOADBU_STUB,jaddr,out,i,addr,i_regs,ccadj_,reglist); } else - inline_readstub(LOADBU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); - } - if (opcode[i]==0x25) { // LHU + inline_readstub(LOADBU_STUB,i,constmap[i][s]+offset,i_regs->regmap,dops[i].rt1,ccadj_,reglist); + break; + case 0x25: // LHU if(!c||memtarget) { if(!dummy) { - int x=0,a=tl; - if(!c) a=addr; - if(fastio_reg_override>=0) a=fastio_reg_override; - emit_movzwl_indexed(x,a,tl); + int a = tl; + if(!c) a = addr; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + if (offset_reg >= 0) + emit_ldrh_dualindexed(offset_reg, a, tl); + else + emit_movzwl_indexed(0, a, tl); } if(jaddr) - add_stub_r(LOADHU_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist); + add_stub_r(LOADHU_STUB,jaddr,out,i,addr,i_regs,ccadj_,reglist); } else - inline_readstub(LOADHU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); - } - if (opcode[i]==0x27) { // LWU - assert(0); - } - if (opcode[i]==0x37) { // LD + inline_readstub(LOADHU_STUB,i,constmap[i][s]+offset,i_regs->regmap,dops[i].rt1,ccadj_,reglist); + break; + case 0x27: // LWU + case 0x37: // LD + default: assert(0); } } - if (fastio_reg_override == HOST_TEMPREG) + if (fastio_reg_override == HOST_TEMPREG || offset_reg == HOST_TEMPREG) host_tempreg_release(); } #ifndef loadlr_assemble -static void loadlr_assemble(int i, const struct regstat *i_regs) +static void loadlr_assemble(int i, const struct regstat *i_regs, int ccadj_) { int s,tl,temp,temp2,addr; int offset; void *jaddr=0; int memtarget=0,c=0; - int fastio_reg_override=-1; + int offset_reg = -1; + int fastio_reg_override = -1; u_int reglist=get_host_reglist(i_regs->regmap); - tl=get_reg(i_regs->regmap,rt1[i]); - s=get_reg(i_regs->regmap,rs1[i]); + 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); temp2=get_reg(i_regs->regmap,FTEMP); addr=get_reg(i_regs->regmap,AGEN1+(i&1)); @@ -2869,43 +2936,44 @@ static void loadlr_assemble(int i, const struct regstat *i_regs) } if(!c) { emit_shlimm(addr,3,temp); - if (opcode[i]==0x22||opcode[i]==0x26) { + if (dops[i].opcode==0x22||dops[i].opcode==0x26) { emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR }else{ emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR } - jaddr=emit_fastpath_cmp_jump(i,temp2,&fastio_reg_override); + jaddr = emit_fastpath_cmp_jump(i, i_regs, temp2, + &offset_reg, &fastio_reg_override); } else { - if(ram_offset&&memtarget) { - host_tempreg_acquire(); - emit_addimm(temp2,ram_offset,HOST_TEMPREG); - fastio_reg_override=HOST_TEMPREG; + if (ram_offset && memtarget) { + offset_reg = get_ro_reg(i_regs, 0); } - if (opcode[i]==0x22||opcode[i]==0x26) { + if (dops[i].opcode==0x22||dops[i].opcode==0x26) { emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR }else{ emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR } } - if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR + if (dops[i].opcode==0x22||dops[i].opcode==0x26) { // LWL/LWR if(!c||memtarget) { - int a=temp2; - if(fastio_reg_override>=0) a=fastio_reg_override; - emit_readword_indexed(0,a,temp2); - if(fastio_reg_override==HOST_TEMPREG) host_tempreg_release(); - if(jaddr) add_stub_r(LOADW_STUB,jaddr,out,i,temp2,i_regs,ccadj[i],reglist); + int a = temp2; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + do_load_word(a, temp2, offset_reg); + if (fastio_reg_override == HOST_TEMPREG || offset_reg == HOST_TEMPREG) + host_tempreg_release(); + if(jaddr) add_stub_r(LOADW_STUB,jaddr,out,i,temp2,i_regs,ccadj_,reglist); } else - inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist); - if(rt1[i]) { + inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj_,reglist); + if(dops[i].rt1) { assert(tl>=0); emit_andimm(temp,24,temp); - if (opcode[i]==0x22) // LWL + if (dops[i].opcode==0x22) // LWL emit_xorimm(temp,24,temp); host_tempreg_acquire(); emit_movimm(-1,HOST_TEMPREG); - if (opcode[i]==0x26) { + if (dops[i].opcode==0x26) { emit_shr(temp2,temp,temp2); emit_bic_lsr(tl,HOST_TEMPREG,temp,tl); }else{ @@ -2915,27 +2983,28 @@ static void loadlr_assemble(int i, const struct regstat *i_regs) host_tempreg_release(); emit_or(temp2,tl,tl); } - //emit_storereg(rt1[i],tl); // DEBUG + //emit_storereg(dops[i].rt1,tl); // DEBUG } - if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR + if (dops[i].opcode==0x1A||dops[i].opcode==0x1B) { // LDL/LDR assert(0); } } #endif -void store_assemble(int i, const struct regstat *i_regs) +static void store_assemble(int i, const struct regstat *i_regs, int ccadj_) { int s,tl; int addr,temp; int offset; void *jaddr=0; - enum stub_type type; + enum stub_type type=0; int memtarget=0,c=0; int agr=AGEN1+(i&1); - int fastio_reg_override=-1; + int offset_reg = -1; + int fastio_reg_override = -1; u_int reglist=get_host_reglist(i_regs->regmap); - tl=get_reg(i_regs->regmap,rs2[i]); - s=get_reg(i_regs->regmap,rs1[i]); + 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); offset=imm[i]; @@ -2950,54 +3019,57 @@ void store_assemble(int i, const struct regstat *i_regs) if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<=0) a=fastio_reg_override; - emit_writebyte_indexed(tl,x,a); - } - type=STOREB_STUB; - } - if (opcode[i]==0x29) { // SH + int a = temp; + if (!c) a = addr; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + do_store_byte(a, tl, offset_reg); + } + type = STOREB_STUB; + break; + case 0x29: // SH if(!c||memtarget) { - int x=0,a=temp; - if(!c) a=addr; - if(fastio_reg_override>=0) a=fastio_reg_override; - emit_writehword_indexed(tl,x,a); - } - type=STOREH_STUB; - } - if (opcode[i]==0x2B) { // SW + int a = temp; + if (!c) a = addr; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + do_store_hword(a, 0, tl, offset_reg, 1); + } + type = STOREH_STUB; + break; + case 0x2B: // SW if(!c||memtarget) { - int a=addr; - if(fastio_reg_override>=0) a=fastio_reg_override; - emit_writeword_indexed(tl,0,a); - } - type=STOREW_STUB; - } - if (opcode[i]==0x3F) { // SD + int a = addr; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + do_store_word(a, 0, tl, offset_reg, 1); + } + type = STOREW_STUB; + break; + case 0x3F: // SD + default: assert(0); - type=STORED_STUB; } - if(fastio_reg_override==HOST_TEMPREG) + if (fastio_reg_override == HOST_TEMPREG || offset_reg == HOST_TEMPREG) host_tempreg_release(); if(jaddr) { // PCSX store handlers don't check invcode again reglist|=1<waswritten&(1<waswritten&(1<regmap,rs2[i],ccadj[i],reglist); + inline_writestub(type,i,addr_val,i_regs->regmap,dops[i].rs2,ccadj_,reglist); } // basic current block modification detection.. // not looking back as that should be in mips cache already @@ -3044,19 +3116,20 @@ void store_assemble(int i, const struct regstat *i_regs) } } -static void storelr_assemble(int i, const struct regstat *i_regs) +static void storelr_assemble(int i, const struct regstat *i_regs, int ccadj_) { int s,tl; int temp; int offset; void *jaddr=0; - void *case1, *case2, *case3; + void *case1, *case23, *case3; void *done0, *done1, *done2; int memtarget=0,c=0; int agr=AGEN1+(i&1); + int offset_reg = -1; u_int reglist=get_host_reglist(i_regs->regmap); - tl=get_reg(i_regs->regmap,rs2[i]); - s=get_reg(i_regs->regmap,rs1[i]); + 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); offset=imm[i]; @@ -3076,91 +3149,90 @@ static void storelr_assemble(int i, const struct regstat *i_regs) } else { - if(!memtarget||!rs1[i]) { + if(!memtarget||!dops[i].rs1) { jaddr=out; emit_jmp(0); } } - if(ram_offset) - emit_addimm_no_flags(ram_offset,temp); + if (ram_offset) + offset_reg = get_ro_reg(i_regs, 0); - if (opcode[i]==0x2C||opcode[i]==0x2D) { // SDL/SDR + if (dops[i].opcode==0x2C||dops[i].opcode==0x2D) { // SDL/SDR assert(0); } - emit_xorimm(temp,3,temp); emit_testimm(temp,2); - case2=out; + case23=out; emit_jne(0); emit_testimm(temp,1); case1=out; emit_jne(0); // 0 - if (opcode[i]==0x2A) { // SWL - emit_writeword_indexed(tl,0,temp); + if (dops[i].opcode == 0x2A) { // SWL + // Write msb into least significant byte + if (dops[i].rs2) emit_rorimm(tl, 24, tl); + do_store_byte(temp, tl, offset_reg); + if (dops[i].rs2) emit_rorimm(tl, 8, tl); } - else if (opcode[i]==0x2E) { // SWR - emit_writebyte_indexed(tl,3,temp); + else if (dops[i].opcode == 0x2E) { // SWR + // Write entire word + do_store_word(temp, 0, tl, offset_reg, 1); } - else - assert(0); - done0=out; + done0 = out; emit_jmp(0); // 1 set_jump_target(case1, 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 (dops[i].opcode == 0x2A) { // SWL + // Write two msb into two least significant bytes + if (dops[i].rs2) emit_rorimm(tl, 16, tl); + do_store_hword(temp, -1, tl, offset_reg, 0); + if (dops[i].rs2) emit_rorimm(tl, 16, tl); } - else if (opcode[i]==0x2E) { // SWR - // Write two lsb into two most significant bytes - emit_writehword_indexed(tl,1,temp); + else if (dops[i].opcode == 0x2E) { // SWR + // Write 3 lsb into three most significant bytes + do_store_byte(temp, tl, offset_reg); + if (dops[i].rs2) emit_rorimm(tl, 8, tl); + do_store_hword(temp, 1, tl, offset_reg, 0); + if (dops[i].rs2) emit_rorimm(tl, 24, tl); } done1=out; emit_jmp(0); - // 2 - set_jump_target(case2, out); + // 2,3 + set_jump_target(case23, out); emit_testimm(temp,1); - case3=out; + case3 = 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); + // 2 + if (dops[i].opcode==0x2A) { // SWL + // Write 3 msb into three least significant bytes + if (dops[i].rs2) emit_rorimm(tl, 8, tl); + do_store_hword(temp, -2, tl, offset_reg, 1); + if (dops[i].rs2) emit_rorimm(tl, 16, tl); + do_store_byte(temp, tl, offset_reg); + if (dops[i].rs2) emit_rorimm(tl, 8, tl); } - else 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); + else if (dops[i].opcode == 0x2E) { // SWR + // Write two lsb into two most significant bytes + do_store_hword(temp, 0, tl, offset_reg, 1); } - done2=out; + done2 = out; emit_jmp(0); // 3 set_jump_target(case3, 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 (dops[i].opcode == 0x2A) { // SWL + do_store_word(temp, -3, tl, offset_reg, 0); } - else if (opcode[i]==0x2E) { // SWR - // Write entire word - emit_writeword_indexed(tl,-3,temp); + else if (dops[i].opcode == 0x2E) { // SWR + do_store_byte(temp, tl, offset_reg); } set_jump_target(done0, out); set_jump_target(done1, out); set_jump_target(done2, out); + if (offset_reg == HOST_TEMPREG) + host_tempreg_release(); if(!c||!memtarget) - add_stub_r(STORELR_STUB,jaddr,out,i,temp,i_regs,ccadj[i],reglist); - if(!(i_regs->waswritten&(1<waswritten&(1<regmap,INVCP); assert(ir>=0); @@ -3178,28 +3250,28 @@ static void storelr_assemble(int i, const struct regstat *i_regs) } } -static void cop0_assemble(int i,struct regstat *i_regs) +static void cop0_assemble(int i, const struct regstat *i_regs, int ccadj_) { - if(opcode2[i]==0) // MFC0 + if(dops[i].opcode2==0) // MFC0 { - signed char t=get_reg(i_regs->regmap,rt1[i]); + signed char t=get_reg(i_regs->regmap,dops[i].rt1); u_int copr=(source[i]>>11)&0x1f; //assert(t>=0); // Why does this happen? OOT is weird - if(t>=0&&rt1[i]!=0) { + if(t>=0&&dops[i].rt1!=0) { emit_readword(®_cop0[copr],t); } } - else if(opcode2[i]==4) // MTC0 + else if(dops[i].opcode2==4) // MTC0 { - signed char s=get_reg(i_regs->regmap,rs1[i]); + signed char s=get_reg(i_regs->regmap,dops[i].rs1); char copr=(source[i]>>11)&0x1f; assert(s>=0); - wb_register(rs1[i],i_regs->regmap,i_regs->dirty); + wb_register(dops[i].rs1,i_regs->regmap,i_regs->dirty); if(copr==9||copr==11||copr==12||copr==13) { emit_readword(&last_count,HOST_TEMPREG); emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc emit_add(HOST_CCREG,HOST_TEMPREG,HOST_CCREG); - emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); + emit_addimm(HOST_CCREG,ccadj_,HOST_CCREG); emit_writeword(HOST_CCREG,&Count); } // What a mess. The status register (12) can enable interrupts, @@ -3214,10 +3286,10 @@ static void cop0_assemble(int i,struct regstat *i_regs) emit_writeword(HOST_CCREG,&last_count); emit_movimm(0,HOST_CCREG); emit_storereg(CCREG,HOST_CCREG); - emit_loadreg(rs1[i],1); + emit_loadreg(dops[i].rs1,1); emit_movimm(copr,0); emit_far_call(pcsx_mtc0_ds); - emit_loadreg(rs1[i],s); + emit_loadreg(dops[i].rs1,s); return; } emit_movimm(start+i*4+4,HOST_TEMPREG); @@ -3226,7 +3298,7 @@ static void cop0_assemble(int i,struct regstat *i_regs) emit_writeword(HOST_TEMPREG,&pending_exception); } if(s==HOST_CCREG) - emit_loadreg(rs1[i],1); + emit_loadreg(dops[i].rs1,1); else if(s!=1) emit_mov(s,1); emit_movimm(copr,0); @@ -3234,7 +3306,7 @@ static void cop0_assemble(int i,struct regstat *i_regs) if(copr==9||copr==11||copr==12||copr==13) { emit_readword(&Count,HOST_CCREG); emit_readword(&next_interupt,HOST_TEMPREG); - emit_addimm(HOST_CCREG,-CLOCK_ADJUST(ccadj[i]),HOST_CCREG); + emit_addimm(HOST_CCREG,-ccadj_,HOST_CCREG); emit_sub(HOST_CCREG,HOST_TEMPREG,HOST_CCREG); emit_writeword(HOST_TEMPREG,&last_count); emit_storereg(CCREG,HOST_CCREG); @@ -3251,11 +3323,11 @@ static void cop0_assemble(int i,struct regstat *i_regs) emit_jmpreg(0); set_jump_target(jaddr, out); } - emit_loadreg(rs1[i],s); + emit_loadreg(dops[i].rs1,s); } else { - assert(opcode2[i]==0x10); + assert(dops[i].opcode2==0x10); //if((source[i]&0x3f)==0x10) // RFE { emit_readword(&Status,0); @@ -3267,7 +3339,7 @@ static void cop0_assemble(int i,struct regstat *i_regs) } } -static void cop1_unusable(int i,struct regstat *i_regs) +static void cop1_unusable(int i, const struct regstat *i_regs) { // XXX: should just just do the exception instead //if(!cop1_usable) @@ -3278,12 +3350,12 @@ static void cop1_unusable(int i,struct regstat *i_regs) } } -static void cop1_assemble(int i,struct regstat *i_regs) +static void cop1_assemble(int i, const struct regstat *i_regs) { cop1_unusable(i, i_regs); } -static void c1ls_assemble(int i,struct regstat *i_regs) +static void c1ls_assemble(int i, const struct regstat *i_regs) { cop1_unusable(i, i_regs); } @@ -3306,21 +3378,21 @@ static void do_cop1stub(int n) wb_dirtys(i_regs->regmap_entry,i_regs->wasdirty); if(regs[i].regmap_entry[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG); emit_movimm(start+(i-ds)*4,EAX); // Get PC - emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right? There should probably be an extra cycle... + emit_addimm(HOST_CCREG,ccadj[i],HOST_CCREG); // CHECK: is this right? There should probably be an extra cycle... emit_far_jump(ds?fp_exception_ds:fp_exception); } static int cop2_is_stalling_op(int i, int *cycles) { - if (opcode[i] == 0x3a) { // SWC2 + if (dops[i].opcode == 0x3a) { // SWC2 *cycles = 0; return 1; } - if (itype[i] == COP2 && (opcode2[i] == 0 || opcode2[i] == 2)) { // MFC2/CFC2 + if (dops[i].itype == COP2 && (dops[i].opcode2 == 0 || dops[i].opcode2 == 2)) { // MFC2/CFC2 *cycles = 0; return 1; } - if (itype[i] == C2OP) { + if (dops[i].itype == C2OP) { *cycles = gte_cycletab[source[i] & 0x3f]; return 1; } @@ -3342,7 +3414,7 @@ static void emit_log_gte_stall(int i, int stall, u_int reglist) emit_movimm(stall, 0); else emit_mov(HOST_TEMPREG, 0); - emit_addimm(HOST_CCREG, CLOCK_ADJUST(ccadj[i]), 1); + emit_addimm(HOST_CCREG, ccadj[i], 1); emit_far_call(log_gte_stall); restore_regs(reglist); } @@ -3360,15 +3432,17 @@ static void cop2_do_stall_check(u_int op, int i, const struct regstat *i_regs, u //printf("no cc %08x\n", start + i*4); return; } - if (!bt[i]) { + if (!dops[i].bt) { for (j = i - 1; j >= 0; j--) { - //if (is_ds[j]) break; - if (cop2_is_stalling_op(j, &other_gte_op_cycles) || bt[j]) + //if (dops[j].is_ds) break; + if (cop2_is_stalling_op(j, &other_gte_op_cycles) || dops[j].bt) + break; + if (j > 0 && ccadj[j - 1] > ccadj[j]) break; } j = max(j, 0); } - cycles_passed = CLOCK_ADJUST(ccadj[i] - ccadj[j]); + cycles_passed = ccadj[i] - ccadj[j]; if (other_gte_op_cycles >= 0) stall = other_gte_op_cycles - cycles_passed; else if (cycles_passed >= 44) @@ -3379,13 +3453,13 @@ static void cop2_do_stall_check(u_int op, int i, const struct regstat *i_regs, u #if 0 // too slow save_regs(reglist); emit_movimm(gte_cycletab[op], 0); - emit_addimm(HOST_CCREG, CLOCK_ADJUST(ccadj[i]), 1); + emit_addimm(HOST_CCREG, ccadj[i], 1); emit_far_call(call_gteStall); restore_regs(reglist); #else host_tempreg_acquire(); emit_readword(&psxRegs.gteBusyCycle, rtmp); - emit_addimm(rtmp, -CLOCK_ADJUST(ccadj[i]), rtmp); + emit_addimm(rtmp, -ccadj[i], rtmp); emit_sub(rtmp, HOST_CCREG, HOST_TEMPREG); emit_cmpimm(HOST_TEMPREG, 44); emit_cmovb_reg(rtmp, HOST_CCREG); @@ -3405,7 +3479,7 @@ static void cop2_do_stall_check(u_int op, int i, const struct regstat *i_regs, u for (j = i + 1; j < slen; j++) { if (cop2_is_stalling_op(j, &other_gte_op_cycles)) break; - if (is_jump(j)) { + if (dops[j].is_jump) { // check ds if (j + 1 < slen && cop2_is_stalling_op(j + 1, &other_gte_op_cycles)) j++; @@ -3415,7 +3489,7 @@ static void cop2_do_stall_check(u_int op, int i, const struct regstat *i_regs, u if (other_gte_op_cycles >= 0) // will handle stall when assembling that op return; - cycles_passed = CLOCK_ADJUST(ccadj[min(j, slen -1)] - ccadj[i]); + cycles_passed = ccadj[min(j, slen -1)] - ccadj[i]; if (cycles_passed >= 44) return; assem_debug("; save gteBusyCycle\n"); @@ -3423,11 +3497,11 @@ static void cop2_do_stall_check(u_int op, int i, const struct regstat *i_regs, u #if 0 emit_readword(&last_count, HOST_TEMPREG); emit_add(HOST_TEMPREG, HOST_CCREG, HOST_TEMPREG); - emit_addimm(HOST_TEMPREG, CLOCK_ADJUST(ccadj[i]), HOST_TEMPREG); + emit_addimm(HOST_TEMPREG, ccadj[i], HOST_TEMPREG); emit_addimm(HOST_TEMPREG, gte_cycletab[op]), HOST_TEMPREG); emit_writeword(HOST_TEMPREG, &psxRegs.gteBusyCycle); #else - emit_addimm(HOST_CCREG, CLOCK_ADJUST(ccadj[i]) + gte_cycletab[op], HOST_TEMPREG); + emit_addimm(HOST_CCREG, ccadj[i] + gte_cycletab[op], HOST_TEMPREG); emit_writeword(HOST_TEMPREG, &psxRegs.gteBusyCycle); #endif host_tempreg_release(); @@ -3435,21 +3509,21 @@ static void cop2_do_stall_check(u_int op, int i, const struct regstat *i_regs, u static int is_mflohi(int i) { - return (itype[i] == MOV && (rs1[i] == HIREG || rs1[i] == LOREG)); + return (dops[i].itype == MOV && (dops[i].rs1 == HIREG || dops[i].rs1 == LOREG)); } static int check_multdiv(int i, int *cycles) { - if (itype[i] != MULTDIV) + if (dops[i].itype != MULTDIV) return 0; - if (opcode2[i] == 0x18 || opcode2[i] == 0x19) // MULT(U) + if (dops[i].opcode2 == 0x18 || dops[i].opcode2 == 0x19) // MULT(U) *cycles = 11; // approx from 7 11 14 else *cycles = 37; return 1; } -static void multdiv_prepare_stall(int i, const struct regstat *i_regs) +static void multdiv_prepare_stall(int i, const struct regstat *i_regs, int ccadj_) { int j, found = 0, c = 0; if (HACK_ENABLED(NDHACK_NO_STALLS)) @@ -3459,11 +3533,11 @@ static void multdiv_prepare_stall(int i, const struct regstat *i_regs) return; } for (j = i + 1; j < slen; j++) { - if (bt[j]) + if (dops[j].bt) break; if ((found = is_mflohi(j))) break; - if (is_jump(j)) { + if (dops[j].is_jump) { // check ds if (j + 1 < slen && (found = is_mflohi(j + 1))) j++; @@ -3477,7 +3551,7 @@ static void multdiv_prepare_stall(int i, const struct regstat *i_regs) assert(c > 0); assem_debug("; muldiv prepare stall %d\n", c); host_tempreg_acquire(); - emit_addimm(HOST_CCREG, CLOCK_ADJUST(ccadj[i]) + c, HOST_TEMPREG); + emit_addimm(HOST_CCREG, ccadj_ + c, HOST_TEMPREG); emit_writeword(HOST_TEMPREG, &psxRegs.muldivBusyCycle); host_tempreg_release(); } @@ -3496,19 +3570,21 @@ static void multdiv_do_stall(int i, const struct regstat *i_regs) //printf("no cc/rtmp %08x\n", start + i*4); return; } - if (!bt[i]) { + if (!dops[i].bt) { for (j = i - 1; j >= 0; j--) { - if (is_ds[j]) break; - if (check_multdiv(j, &known_cycles) || bt[j]) + if (dops[j].is_ds) break; + if (check_multdiv(j, &known_cycles)) break; if (is_mflohi(j)) // already handled by this op return; + if (dops[j].bt || (j > 0 && ccadj[j - 1] > ccadj[j])) + break; } j = max(j, 0); } if (known_cycles > 0) { - known_cycles -= CLOCK_ADJUST(ccadj[i] - ccadj[j]); + known_cycles -= ccadj[i] - ccadj[j]; assem_debug("; muldiv stall resolved %d\n", known_cycles); if (known_cycles > 0) emit_addimm(HOST_CCREG, known_cycles, HOST_CCREG); @@ -3517,7 +3593,7 @@ static void multdiv_do_stall(int i, const struct regstat *i_regs) assem_debug("; muldiv stall unresolved\n"); host_tempreg_acquire(); emit_readword(&psxRegs.muldivBusyCycle, rtmp); - emit_addimm(rtmp, -CLOCK_ADJUST(ccadj[i]), rtmp); + emit_addimm(rtmp, -ccadj[i], rtmp); emit_sub(rtmp, HOST_CCREG, HOST_TEMPREG); emit_cmpimm(HOST_TEMPREG, 37); emit_cmovb_reg(rtmp, HOST_CCREG); @@ -3608,7 +3684,7 @@ static void cop2_put_dreg(u_int copr,signed char sl,signed char temp) } } -static void c2ls_assemble(int i, const struct regstat *i_regs) +static void c2ls_assemble(int i, const struct regstat *i_regs, int ccadj_) { int s,tl; int ar; @@ -3617,20 +3693,21 @@ static void c2ls_assemble(int i, const struct regstat *i_regs) void *jaddr2=NULL; enum stub_type type; int agr=AGEN1+(i&1); - int fastio_reg_override=-1; + int offset_reg = -1; + int fastio_reg_override = -1; u_int reglist=get_host_reglist(i_regs->regmap); u_int copr=(source[i]>>16)&0x1f; - s=get_reg(i_regs->regmap,rs1[i]); + s=get_reg(i_regs->regmap,dops[i].rs1); tl=get_reg(i_regs->regmap,FTEMP); offset=imm[i]; - assert(rs1[i]>0); + assert(dops[i].rs1>0); assert(tl>=0); if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<regmap,agr); if(ar<0) ar=get_reg(i_regs->regmap,-1); reglist|=1<=0) a=fastio_reg_override; - emit_readword_indexed(0,a,tl); + jaddr2 = emit_fastpath_cmp_jump(i, i_regs, ar, + &offset_reg, &fastio_reg_override); + } + else if (ram_offset && memtarget) { + offset_reg = get_ro_reg(i_regs, 0); + } + switch (dops[i].opcode) { + case 0x32: { // LWC2 + int a = ar; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + do_load_word(a, tl, offset_reg); + break; } - if (opcode[i]==0x3a) { // SWC2 + case 0x3a: { // SWC2 #ifdef DESTRUCTIVE_SHIFT if(!offset&&!c&&s>=0) emit_mov(s,ar); #endif - int a=ar; - if(fastio_reg_override>=0) a=fastio_reg_override; - emit_writeword_indexed(tl,0,a); + int a = ar; + if (fastio_reg_override >= 0) + a = fastio_reg_override; + do_store_word(a, 0, tl, offset_reg, 1); + break; + } + default: + assert(0); } } - if(fastio_reg_override==HOST_TEMPREG) + if (fastio_reg_override == HOST_TEMPREG || offset_reg == HOST_TEMPREG) host_tempreg_release(); if(jaddr2) - add_stub_r(type,jaddr2,out,i,ar,i_regs,ccadj[i],reglist); - if(opcode[i]==0x3a) // SWC2 - if(!(i_regs->waswritten&(1<waswritten&(1<regmap,INVCP); assert(ir>=0); @@ -3699,7 +3783,7 @@ static void c2ls_assemble(int i, const struct regstat *i_regs) add_stub(INVCODE_STUB,jaddr3,out,reglist|(1<regmap), temp, -1); - if (opcode2[i] == 0 || opcode2[i] == 2) { // MFC2/CFC2 - signed char tl = get_reg(i_regs->regmap, rt1[i]); + if (dops[i].opcode2 == 0 || dops[i].opcode2 == 2) { // MFC2/CFC2 + signed char tl = get_reg(i_regs->regmap, dops[i].rt1); reglist = reglist_exclude(reglist, tl, -1); } cop2_do_stall_check(0, i, i_regs, reglist); } - if (opcode2[i]==0) { // MFC2 - signed char tl=get_reg(i_regs->regmap,rt1[i]); - if(tl>=0&&rt1[i]!=0) + if (dops[i].opcode2==0) { // MFC2 + signed char tl=get_reg(i_regs->regmap,dops[i].rt1); + if(tl>=0&&dops[i].rt1!=0) cop2_get_dreg(copr,tl,temp); } - else if (opcode2[i]==4) { // MTC2 - signed char sl=get_reg(i_regs->regmap,rs1[i]); + else if (dops[i].opcode2==4) { // MTC2 + signed char sl=get_reg(i_regs->regmap,dops[i].rs1); cop2_put_dreg(copr,sl,temp); } - else if (opcode2[i]==2) // CFC2 + else if (dops[i].opcode2==2) // CFC2 { - signed char tl=get_reg(i_regs->regmap,rt1[i]); - if(tl>=0&&rt1[i]!=0) + signed char tl=get_reg(i_regs->regmap,dops[i].rt1); + if(tl>=0&&dops[i].rt1!=0) emit_readword(®_cop2c[copr],tl); } - else if (opcode2[i]==6) // CTC2 + else if (dops[i].opcode2==6) // CTC2 { - signed char sl=get_reg(i_regs->regmap,rs1[i]); + signed char sl=get_reg(i_regs->regmap,dops[i].rs1); switch(copr) { case 4: case 12: @@ -3772,75 +3856,26 @@ static void do_unalignedwritestub(int n) signed char *i_regmap=i_regs->regmap; int temp2=get_reg(i_regmap,FTEMP); int rt; - rt=get_reg(i_regmap,rs2[i]); + rt=get_reg(i_regmap,dops[i].rs2); assert(rt>=0); assert(addr>=0); - assert(opcode[i]==0x2a||opcode[i]==0x2e); // SWL/SWR only implemented + assert(dops[i].opcode==0x2a||dops[i].opcode==0x2e); // SWL/SWR only implemented reglist|=(1<regmap,rt1[i]); + tl=get_reg(i_regs->regmap,dops[i].rt1); //assert(tl>=0); if(tl>=0) { - sl=get_reg(i_regs->regmap,rs1[i]); + sl=get_reg(i_regs->regmap,dops[i].rs1); if(sl>=0) emit_mov(sl,tl); - else emit_loadreg(rs1[i],tl); + else emit_loadreg(dops[i].rs1,tl); } } - if (rs1[i] == HIREG || rs1[i] == LOREG) // MFHI/MFLO + if (dops[i].rs1 == HIREG || dops[i].rs1 == LOREG) // MFHI/MFLO multdiv_do_stall(i, i_regs); } // call interpreter, exception handler, things that change pc/regs/cycles ... -static void call_c_cpu_handler(int i, const struct regstat *i_regs, u_int pc, void *func) +static void call_c_cpu_handler(int i, const struct regstat *i_regs, int ccadj_, u_int pc, void *func) { signed char ccreg=get_reg(i_regs->regmap,CCREG); assert(ccreg==HOST_CCREG); @@ -3880,33 +3915,33 @@ static void call_c_cpu_handler(int i, const struct regstat *i_regs, u_int pc, vo emit_movimm(pc,3); // Get PC emit_readword(&last_count,2); emit_writeword(3,&psxRegs.pc); - emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // XXX + emit_addimm(HOST_CCREG,ccadj_,HOST_CCREG); emit_add(2,HOST_CCREG,2); emit_writeword(2,&psxRegs.cycle); emit_far_call(func); emit_far_jump(jump_to_new_pc); } -static void syscall_assemble(int i,struct regstat *i_regs) +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,start+i*4,psxException); + call_c_cpu_handler(i, i_regs, ccadj_, start+i*4, psxException); } -static void hlecall_assemble(int i,struct regstat *i_regs) +static void hlecall_assemble(int i, const struct regstat *i_regs, int ccadj_) { void *hlefunc = psxNULL; uint32_t hleCode = source[i] & 0x03ffffff; if (hleCode < ARRAY_SIZE(psxHLEt)) hlefunc = psxHLEt[hleCode]; - call_c_cpu_handler(i,i_regs,start+i*4+4,hlefunc); + call_c_cpu_handler(i, i_regs, ccadj_, start + i*4+4, hlefunc); } -static void intcall_assemble(int i,struct regstat *i_regs) +static void intcall_assemble(int i, const struct regstat *i_regs, int ccadj_) { - call_c_cpu_handler(i,i_regs,start+i*4,execI); + call_c_cpu_handler(i, i_regs, ccadj_, start + i*4, execI); } static void speculate_mov(int rs,int rt) @@ -3936,61 +3971,61 @@ static void speculate_register_values(int i) } smrv_strong=smrv_strong_next; smrv_weak=smrv_weak_next; - switch(itype[i]) { + switch(dops[i].itype) { case ALU: - if ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]); - else if((smrv_strong>>rs2[i])&1) speculate_mov(rs2[i],rt1[i]); - else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]); - else if((smrv_weak>>rs2[i])&1) speculate_mov_weak(rs2[i],rt1[i]); + if ((smrv_strong>>dops[i].rs1)&1) speculate_mov(dops[i].rs1,dops[i].rt1); + else if((smrv_strong>>dops[i].rs2)&1) speculate_mov(dops[i].rs2,dops[i].rt1); + else if((smrv_weak>>dops[i].rs1)&1) speculate_mov_weak(dops[i].rs1,dops[i].rt1); + else if((smrv_weak>>dops[i].rs2)&1) speculate_mov_weak(dops[i].rs2,dops[i].rt1); else { - smrv_strong_next&=~(1<=0) { if(get_final_value(hr,i,&value)) - smrv[rt1[i]]=value; - else smrv[rt1[i]]=constmap[i][hr]; - smrv_strong_next|=1<>rs1[i])&1) speculate_mov(rs1[i],rt1[i]); - else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]); + if ((smrv_strong>>dops[i].rs1)&1) speculate_mov(dops[i].rs1,dops[i].rt1); + else if((smrv_weak>>dops[i].rs1)&1) speculate_mov_weak(dops[i].rs1,dops[i].rt1); } break; case LOAD: - if(start<0x2000&&(rt1[i]==26||(smrv[rt1[i]]>>24)==0xa0)) { + if(start<0x2000&&(dops[i].rt1==26||(smrv[dops[i].rt1]>>24)==0xa0)) { // special case for BIOS - smrv[rt1[i]]=0xa0000000; - smrv_strong_next|=1<regmap,rt1[i]); + if(dops[i].itype==LOAD) { + ra=get_reg(i_regs->regmap,dops[i].rt1); if(ra<0) ra=get_reg(i_regs->regmap,-1); assert(ra>=0); } - if(itype[i]==LOADLR) { + if(dops[i].itype==LOADLR) { ra=get_reg(i_regs->regmap,FTEMP); } - if(itype[i]==STORE||itype[i]==STORELR) { + 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(itype[i]==C1LS||itype[i]==C2LS) { - if ((opcode[i]&0x3b)==0x31||(opcode[i]&0x3b)==0x32) // LWC1/LDC1/LWC2/LDC2 + 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); } } - int rs=get_reg(i_regs->regmap,rs1[i]); + int rs=get_reg(i_regs->regmap,dops[i].rs1); if(ra>=0) { int offset=imm[i]; int c=(i_regs->wasconst>>rs)&1; - if(rs1[i]==0) { + if(dops[i].rs1==0) { // Using r0 as a base address if(!entry||entry[ra]!=agr) { - if (opcode[i]==0x22||opcode[i]==0x26) { + if (dops[i].opcode==0x22||dops[i].opcode==0x26) { emit_movimm(offset&0xFFFFFFFC,ra); // LWL/LWR - }else if (opcode[i]==0x1a||opcode[i]==0x1b) { + }else if (dops[i].opcode==0x1a||dops[i].opcode==0x1b) { emit_movimm(offset&0xFFFFFFF8,ra); // LDL/LDR }else{ emit_movimm(offset,ra); @@ -4196,17 +4297,17 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[]) } // else did it in the previous cycle } else if(rs<0) { - if(!entry||entry[ra]!=rs1[i]) - emit_loadreg(rs1[i],ra); - //if(!entry||entry[ra]!=rs1[i]) + if(!entry||entry[ra]!=dops[i].rs1) + emit_loadreg(dops[i].rs1,ra); + //if(!entry||entry[ra]!=dops[i].rs1) // printf("poor load scheduling!\n"); } else if(c) { - if(rs1[i]!=rt1[i]||itype[i]!=LOAD) { + if(dops[i].rs1!=dops[i].rt1||dops[i].itype!=LOAD) { if(!entry||entry[ra]!=agr) { - if (opcode[i]==0x22||opcode[i]==0x26) { + if (dops[i].opcode==0x22||dops[i].opcode==0x26) { emit_movimm((constmap[i][rs]+offset)&0xFFFFFFFC,ra); // LWL/LWR - }else if (opcode[i]==0x1a||opcode[i]==0x1b) { + }else if (dops[i].opcode==0x1a||dops[i].opcode==0x1b) { emit_movimm((constmap[i][rs]+offset)&0xFFFFFFF8,ra); // LDL/LDR }else{ emit_movimm(constmap[i][rs]+offset,ra); @@ -4215,7 +4316,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[]) } // else did it in the previous cycle } // else load_consts already did it } - if(offset&&!c&&rs1[i]) { + if(offset&&!c&&dops[i].rs1) { if(rs>=0) { emit_addimm(rs,offset,ra); }else{ @@ -4225,30 +4326,30 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[]) } } // Preload constants for next instruction - if(itype[i+1]==LOAD||itype[i+1]==LOADLR||itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS||itype[i+1]==C2LS) { + if (dops[i+1].is_load || dops[i+1].is_store) { int agr,ra; // Actual address agr=AGEN1+((i+1)&1); ra=get_reg(i_regs->regmap,agr); if(ra>=0) { - int rs=get_reg(regs[i+1].regmap,rs1[i+1]); + int rs=get_reg(regs[i+1].regmap,dops[i+1].rs1); int offset=imm[i+1]; int c=(regs[i+1].wasconst>>rs)&1; - if(c&&(rs1[i+1]!=rt1[i+1]||itype[i+1]!=LOAD)) { - if (opcode[i+1]==0x22||opcode[i+1]==0x26) { + if(c&&(dops[i+1].rs1!=dops[i+1].rt1||dops[i+1].itype!=LOAD)) { + if (dops[i+1].opcode==0x22||dops[i+1].opcode==0x26) { emit_movimm((constmap[i+1][rs]+offset)&0xFFFFFFFC,ra); // LWL/LWR - }else if (opcode[i+1]==0x1a||opcode[i+1]==0x1b) { + }else if (dops[i+1].opcode==0x1a||dops[i+1].opcode==0x1b) { emit_movimm((constmap[i+1][rs]+offset)&0xFFFFFFF8,ra); // LDL/LDR }else{ emit_movimm(constmap[i+1][rs]+offset,ra); regs[i+1].loadedconst|=1<>hr)&1)) break; - if(bt[i+1]) break; + if(dops[i+1].bt) break; i++; } if(i>hr)&1)) + if(dops[i+2].itype==LOAD&&dops[i+2].rs1==reg&&dops[i+2].rt1==reg&&((regs[i+1].wasconst>>hr)&1)) { // Precompute load address *value=constmap[i][hr]+imm[i+2]; return 1; } } - if(itype[i+1]==LOAD&&rs1[i+1]==reg&&rt1[i+1]==reg) + if(dops[i+1].itype==LOAD&&dops[i+1].rs1==reg&&dops[i+1].rt1==reg) { // Precompute load address *value=constmap[i][hr]+imm[i+1]; @@ -4303,7 +4404,7 @@ static void load_consts(signed char pre[],signed char regmap[],int i) { int hr,hr2; // propagate loaded constant flags - if(i==0||bt[i]) + if(i==0||dops[i].bt) regs[i].loadedconst=0; else { for(hr=0;hr>2; @@ -4414,7 +4515,7 @@ void wb_needed_dirtys(signed char i_regmap[],uint64_t i_dirty,int addr) } // Load all registers (except cycle count) -void load_all_regs(signed char i_regmap[]) +static void load_all_regs(const signed char i_regmap[]) { int hr; for(hr=0;hr0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP)) return 0; + //if(t>0&&(dops[t-1].is_jump) return 0; // Delay slots require additional processing, so do not match - if(is_ds[t]) return 0; + if(dops[t].is_ds) return 0; } else { @@ -4611,7 +4712,7 @@ static int match_bt(signed char i_regmap[],uint64_t i_dirty,int addr) } #ifdef DRC_DBG -static void drc_dbg_emit_do_cmp(int i) +static void drc_dbg_emit_do_cmp(int i, int ccadj_) { extern void do_insn_cmp(); //extern int cycle; @@ -4620,9 +4721,9 @@ static void drc_dbg_emit_do_cmp(int i) assem_debug("//do_insn_cmp %08x\n", start+i*4); save_regs(reglist); // write out changed consts to match the interpreter - if (i > 0 && !bt[i]) { + if (i > 0 && !dops[i].bt) { for (hr = 0; hr < HOST_REGS; hr++) { - int reg = regs[i-1].regmap[hr]; + int reg = regs[i].regmap_entry[hr]; // regs[i-1].regmap[hr]; if (hr == EXCLUDE_REG || reg < 0) continue; if (!((regs[i-1].isconst >> hr) & 1)) @@ -4635,6 +4736,11 @@ static void drc_dbg_emit_do_cmp(int i) } emit_movimm(start+i*4,0); emit_writeword(0,&pcaddr); + int cc = get_reg(regs[i].regmap_entry, CCREG); + if (cc < 0) + emit_loadreg(CCREG, cc = 0); + emit_addimm(cc, ccadj_, 0); + emit_writeword(0, &psxRegs.cycle); emit_far_call(do_insn_cmp); //emit_readword(&cycle,0); //emit_addimm(0,2,0); @@ -4644,60 +4750,29 @@ static void drc_dbg_emit_do_cmp(int i) assem_debug("\\\\do_insn_cmp\n"); } #else -#define drc_dbg_emit_do_cmp(x) +#define drc_dbg_emit_do_cmp(x,y) #endif // Used when a branch jumps into the delay slot of another branch static void ds_assemble_entry(int i) { - int t=(ba[i]-start)>>2; + int t = (ba[i] - start) >> 2; + int ccadj_ = -CLOCK_ADJUST(1); if (!instr_addr[t]) instr_addr[t] = out; assem_debug("Assemble delay slot at %x\n",ba[i]); assem_debug("<->\n"); - drc_dbg_emit_do_cmp(t); + drc_dbg_emit_do_cmp(t, ccadj_); if(regs[t].regmap_entry[HOST_CCREG]==CCREG&®s[t].regmap[HOST_CCREG]!=CCREG) wb_register(CCREG,regs[t].regmap_entry,regs[t].wasdirty); - load_regs(regs[t].regmap_entry,regs[t].regmap,rs1[t],rs2[t]); + load_regs(regs[t].regmap_entry,regs[t].regmap,dops[t].rs1,dops[t].rs2); address_generation(t,®s[t],regs[t].regmap_entry); - if(itype[t]==STORE||itype[t]==STORELR||(opcode[t]&0x3b)==0x39||(opcode[t]&0x3b)==0x3a) + if (ram_offset && (dops[t].is_load || dops[t].is_store)) + load_regs(regs[t].regmap_entry,regs[t].regmap,ROREG,ROREG); + if (dops[t].is_store) load_regs(regs[t].regmap_entry,regs[t].regmap,INVCP,INVCP); is_delayslot=0; - switch(itype[t]) { - case ALU: - alu_assemble(t,®s[t]);break; - case IMM16: - imm16_assemble(t,®s[t]);break; - case SHIFT: - shift_assemble(t,®s[t]);break; - case SHIFTIMM: - shiftimm_assemble(t,®s[t]);break; - case LOAD: - load_assemble(t,®s[t]);break; - case LOADLR: - loadlr_assemble(t,®s[t]);break; - case STORE: - store_assemble(t,®s[t]);break; - case STORELR: - storelr_assemble(t,®s[t]);break; - case COP0: - cop0_assemble(t,®s[t]);break; - case COP1: - cop1_assemble(t,®s[t]);break; - case C1LS: - c1ls_assemble(t,®s[t]);break; - case COP2: - cop2_assemble(t,®s[t]);break; - case C2LS: - c2ls_assemble(t,®s[t]);break; - case C2OP: - c2op_assemble(t,®s[t]);break; - case MULTDIV: - multdiv_assemble(t,®s[t]); - multdiv_prepare_stall(i,®s[t]); - break; - case MOV: - mov_assemble(t,®s[t]);break; + switch (dops[t].itype) { case SYSCALL: case HLECALL: case INTCALL: @@ -4707,6 +4782,9 @@ static void ds_assemble_entry(int i) case CJUMP: case SJUMP: SysPrintf("Jump in the delay slot. This is probably a bug.\n"); + break; + default: + assemble(t, ®s[t], ccadj_); } store_regs_bt(regs[t].regmap,regs[t].dirty,ba[i]+4); load_regs_bt(regs[t].regmap,regs[t].dirty,ba[i]+4); @@ -4736,13 +4814,14 @@ static void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2) emit_movimm_from(imm1,rt1,imm2,rt2); } -void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert) +static void do_cc(int i, const signed char i_regmap[], int *adj, + int addr, int taken, int invert) { - int count; + int count, count_plus2; void *jaddr; void *idle=NULL; int t=0; - if(itype[i]==RJUMP) + if(dops[i].itype==RJUMP) { *adj=0; } @@ -4750,14 +4829,15 @@ void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert) if(internal_branch(ba[i])) { t=(ba[i]-start)>>2; - if(is_ds[t]) *adj=-1; // Branch into delay slot adds an extra cycle + if(dops[t].is_ds) *adj=-CLOCK_ADJUST(1); // Branch into delay slot adds an extra cycle else *adj=ccadj[t]; } else { *adj=0; } - count=ccadj[i]; + count = ccadj[i]; + count_plus2 = count + CLOCK_ADJUST(2); if(taken==TAKEN && i==(ba[i]-start)>>2 && source[i+1]==0) { // Idle loop if(count&1) emit_addimm_and_set_flags(2*(count+2),HOST_CCREG); @@ -4768,26 +4848,26 @@ void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert) emit_jmp(0); } else if(*adj==0||invert) { - int cycles=CLOCK_ADJUST(count+2); + int cycles = count_plus2; // faster loop HACK #if 0 if (t&&*adj) { int rel=t-i; if(-NO_CYCLE_PENALTY_THR=0); #ifdef DESTRUCTIVE_WRITEBACK - if(rs1[i]) { + if(dops[i].rs1) { if((branch_regs[i].dirty>>s1l)&&1) - emit_loadreg(rs1[i],s1l); + emit_loadreg(dops[i].rs1,s1l); } else { if((branch_regs[i].dirty>>s1l)&1) - emit_loadreg(rs2[i],s1l); + emit_loadreg(dops[i].rs2,s1l); } if(s2l>=0) if((branch_regs[i].dirty>>s2l)&1) - emit_loadreg(rs2[i],s2l); + emit_loadreg(dops[i].rs2,s2l); #endif int hr=0; int addr=-1,alt=-1,ntaddr=-1; while(hr=0) emit_cmp(s1l,s2l); @@ -4892,7 +4972,7 @@ static void do_ccstub(int n) emit_cmovne_reg(alt,addr); #endif } - if((opcode[i]&0x2f)==5) // BNE + if((dops[i].opcode&0x2f)==5) // BNE { #ifdef HAVE_CMOV_IMM if(s2l>=0) emit_cmp(s1l,s2l); @@ -4905,7 +4985,7 @@ static void do_ccstub(int n) emit_cmovne_reg(alt,addr); #endif } - if((opcode[i]&0x2f)==6) // BLEZ + if((dops[i].opcode&0x2f)==6) // BLEZ { //emit_movimm(ba[i],alt); //emit_movimm(start+i*4+8,addr); @@ -4913,7 +4993,7 @@ static void do_ccstub(int n) emit_cmpimm(s1l,1); emit_cmovl_reg(alt,addr); } - if((opcode[i]&0x2f)==7) // BGTZ + if((dops[i].opcode&0x2f)==7) // BGTZ { //emit_movimm(ba[i],addr); //emit_movimm(start+i*4+8,ntaddr); @@ -4921,7 +5001,7 @@ static void do_ccstub(int n) emit_cmpimm(s1l,1); emit_cmovl_reg(ntaddr,addr); } - if((opcode[i]==1)&&(opcode2[i]&0x2D)==0) // BLTZ + if((dops[i].opcode==1)&&(dops[i].opcode2&0x2D)==0) // BLTZ { //emit_movimm(ba[i],alt); //emit_movimm(start+i*4+8,addr); @@ -4929,7 +5009,7 @@ static void do_ccstub(int n) emit_test(s1l,s1l); emit_cmovs_reg(alt,addr); } - if((opcode[i]==1)&&(opcode2[i]&0x2D)==1) // BGEZ + if((dops[i].opcode==1)&&(dops[i].opcode2&0x2D)==1) // BGEZ { //emit_movimm(ba[i],addr); //emit_movimm(start+i*4+8,alt); @@ -4937,7 +5017,7 @@ static void do_ccstub(int n) emit_test(s1l,s1l); emit_cmovs_reg(alt,addr); } - if(opcode[i]==0x11 && opcode2[i]==0x08 ) { + if(dops[i].opcode==0x11 && dops[i].opcode2==0x08 ) { if(source[i]&0x10000) // BC1T { //emit_movimm(ba[i],alt); @@ -4958,10 +5038,10 @@ static void do_ccstub(int n) emit_writeword(addr,&pcaddr); } else - if(itype[i]==RJUMP) + if(dops[i].itype==RJUMP) { - int r=get_reg(branch_regs[i].regmap,rs1[i]); - if(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]) { + int r=get_reg(branch_regs[i].regmap,dops[i].rs1); + if (ds_writes_rjump_rs(i)) { r=get_reg(branch_regs[i].regmap,RTEMP); } emit_writeword(r,&pcaddr); @@ -4970,17 +5050,17 @@ static void do_ccstub(int n) } // Update cycle count assert(branch_regs[i].regmap[HOST_CCREG]==CCREG||branch_regs[i].regmap[HOST_CCREG]==-1); - if(stubs[n].a) emit_addimm(HOST_CCREG,CLOCK_ADJUST((signed int)stubs[n].a),HOST_CCREG); + if(stubs[n].a) emit_addimm(HOST_CCREG,(int)stubs[n].a,HOST_CCREG); emit_far_call(cc_interrupt); - if(stubs[n].a) emit_addimm(HOST_CCREG,-CLOCK_ADJUST((signed int)stubs[n].a),HOST_CCREG); + if(stubs[n].a) emit_addimm(HOST_CCREG,-(int)stubs[n].a,HOST_CCREG); if(stubs[n].d==TAKEN) { if(internal_branch(ba[i])) load_needed_regs(branch_regs[i].regmap,regs[(ba[i]-start)>>2].regmap_entry); - else if(itype[i]==RJUMP) { + else if(dops[i].itype==RJUMP) { if(get_reg(branch_regs[i].regmap,RTEMP)>=0) emit_readword(&pcaddr,get_reg(branch_regs[i].regmap,RTEMP)); else - emit_loadreg(rs1[i],get_reg(branch_regs[i].regmap,rs1[i])); + emit_loadreg(dops[i].rs1,get_reg(branch_regs[i].regmap,dops[i].rs1)); } }else if(stubs[n].d==NOTTAKEN) { if(i=0) { #ifdef USE_MINI_HT - if(internal_branch(return_address)&&rt1[i+1]!=31) { + if(internal_branch(return_address)&&dops[i+1].rt1!=31) { int temp=-1; // note: must be ds-safe #ifdef HOST_TEMPREG temp=HOST_TEMPREG; @@ -5042,14 +5122,14 @@ static void ujump_assemble_write_ra(int i) } } -static void ujump_assemble(int i,struct regstat *i_regs) +static void ujump_assemble(int i, const struct regstat *i_regs) { int ra_done=0; if(i==(ba[i]-start)>>2) assem_debug("idle loop\n"); address_generation(i+1,i_regs,regs[i].regmap_entry); #ifdef REG_PREFETCH int temp=get_reg(branch_regs[i].regmap,PTEMP); - if(rt1[i]==31&&temp>=0) + if(dops[i].rt1==31&&temp>=0) { signed char *i_regmap=i_regs->regmap; int return_address=start+i*4+8; @@ -5057,32 +5137,32 @@ static void ujump_assemble(int i,struct regstat *i_regs) if(i_regmap[temp]==PTEMP) emit_movimm((uintptr_t)hash_table_get(return_address),temp); } #endif - if(rt1[i]==31&&(rt1[i]==rs1[i+1]||rt1[i]==rs2[i+1])) { + if(dops[i].rt1==31&&(dops[i].rt1==dops[i+1].rs1||dops[i].rt1==dops[i+1].rs2)) { ujump_assemble_write_ra(i); // writeback ra for DS ra_done=1; } ds_assemble(i+1,i_regs); uint64_t bc_unneeded=branch_regs[i].u; - bc_unneeded|=1|(1LL<=0) emit_prefetchreg(temp); + if(dops[i].rt1==31&&temp>=0) emit_prefetchreg(temp); #endif do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0); - if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc); + if(adj) emit_addimm(cc, ccadj[i] + CLOCK_ADJUST(2) - adj, cc); load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); if(internal_branch(ba[i])) assem_debug("branch: internal\n"); else assem_debug("branch: external\n"); - if(internal_branch(ba[i])&&is_ds[(ba[i]-start)>>2]) { + if (internal_branch(ba[i]) && dops[(ba[i]-start)>>2].is_ds) { ds_assemble_entry(i); } else { @@ -5094,9 +5174,9 @@ static void ujump_assemble(int i,struct regstat *i_regs) static void rjump_assemble_write_ra(int i) { int rt,return_address; - assert(rt1[i+1]!=rt1[i]); - assert(rt2[i+1]!=rt1[i]); - rt=get_reg(branch_regs[i].regmap,rt1[i]); + assert(dops[i+1].rt1!=dops[i].rt1); + assert(dops[i+1].rt2!=dops[i].rt1); + rt=get_reg(branch_regs[i].regmap,dops[i].rt1); assem_debug("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]); assert(rt>=0); return_address=start+i*4+8; @@ -5112,14 +5192,14 @@ static void rjump_assemble_write_ra(int i) #endif } -static void rjump_assemble(int i,struct regstat *i_regs) +static void rjump_assemble(int i, const struct regstat *i_regs) { int temp; int rs,cc; int ra_done=0; - rs=get_reg(branch_regs[i].regmap,rs1[i]); + rs=get_reg(branch_regs[i].regmap,dops[i].rs1); assert(rs>=0); - if(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]) { + if (ds_writes_rjump_rs(i)) { // Delay slot abuse, make a copy of the branch address register temp=get_reg(branch_regs[i].regmap,RTEMP); assert(temp>=0); @@ -5129,7 +5209,7 @@ static void rjump_assemble(int i,struct regstat *i_regs) } address_generation(i+1,i_regs,regs[i].regmap_entry); #ifdef REG_PREFETCH - if(rt1[i]==31) + if(dops[i].rt1==31) { if((temp=get_reg(branch_regs[i].regmap,PTEMP))>=0) { signed char *i_regmap=i_regs->regmap; @@ -5139,22 +5219,22 @@ static void rjump_assemble(int i,struct regstat *i_regs) } #endif #ifdef USE_MINI_HT - if(rs1[i]==31) { + if(dops[i].rs1==31) { int rh=get_reg(regs[i].regmap,RHASH); if(rh>=0) do_preload_rhash(rh); } #endif - if(rt1[i]!=0&&(rt1[i]==rs1[i+1]||rt1[i]==rs2[i+1])) { + if(dops[i].rt1!=0&&(dops[i].rt1==dops[i+1].rs1||dops[i].rt1==dops[i+1].rs2)) { rjump_assemble_write_ra(i); ra_done=1; } ds_assemble(i+1,i_regs); uint64_t bc_unneeded=branch_regs[i].u; - bc_unneeded|=1|(1LL<>rs)&1) { - if(rs1[i]!=rt1[i+1]&&rs1[i]!=rt2[i+1]) { - emit_loadreg(rs1[i],rs); + if(dops[i].rs1!=dops[i+1].rt1&&dops[i].rs1!=dops[i+1].rt2) { + emit_loadreg(dops[i].rs1,rs); } } #endif #ifdef REG_PREFETCH - if(rt1[i]==31&&temp>=0) emit_prefetchreg(temp); + if(dops[i].rt1==31&&temp>=0) emit_prefetchreg(temp); #endif #ifdef USE_MINI_HT - if(rs1[i]==31) { + if(dops[i].rs1==31) { do_miniht_load(ht,rh); } #endif //do_cc(i,branch_regs[i].regmap,&adj,-1,TAKEN); //if(adj) emit_addimm(cc,2*(ccadj[i]+2-adj),cc); // ??? - Shouldn't happen //assert(adj==0); - emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG); + emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG); add_stub(CC_STUB,out,NULL,0,i,-1,TAKEN,rs); - if(itype[i+1]==COP0&&(source[i+1]&0x3f)==0x10) + if(dops[i+1].itype==COP0&&(source[i+1]&0x3f)==0x10) // special case for RFE emit_jmp(0); else emit_jns(0); //load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,-1); #ifdef USE_MINI_HT - if(rs1[i]==31) { + if(dops[i].rs1==31) { do_miniht_jump(rs,rh,ht); } else @@ -5205,13 +5285,13 @@ static void rjump_assemble(int i,struct regstat *i_regs) do_jump_vaddr(rs); } #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK - if(rt1[i]!=31&&iregmap; + const signed char *i_regmap = i_regs->regmap; int cc; int match; match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); @@ -5229,44 +5309,44 @@ static void cjump_assemble(int i,struct regstat *i_regs) invert=1; // because of near cond. branches #endif - if(ooo[i]) { - s1l=get_reg(branch_regs[i].regmap,rs1[i]); - s2l=get_reg(branch_regs[i].regmap,rs2[i]); + if(dops[i].ooo) { + s1l=get_reg(branch_regs[i].regmap,dops[i].rs1); + s2l=get_reg(branch_regs[i].regmap,dops[i].rs2); } else { - s1l=get_reg(i_regmap,rs1[i]); - s2l=get_reg(i_regmap,rs2[i]); + s1l=get_reg(i_regmap,dops[i].rs1); + s2l=get_reg(i_regmap,dops[i].rs2); } - if(rs1[i]==0&&rs2[i]==0) + if(dops[i].rs1==0&&dops[i].rs2==0) { - if(opcode[i]&1) nop=1; + if(dops[i].opcode&1) nop=1; else unconditional=1; - //assert(opcode[i]!=5); - //assert(opcode[i]!=7); - //assert(opcode[i]!=0x15); - //assert(opcode[i]!=0x17); + //assert(dops[i].opcode!=5); + //assert(dops[i].opcode!=7); + //assert(dops[i].opcode!=0x15); + //assert(dops[i].opcode!=0x17); } - else if(rs1[i]==0) + else if(dops[i].rs1==0) { s1l=s2l; s2l=-1; } - else if(rs2[i]==0) + else if(dops[i].rs2==0) { s2l=-1; } - if(ooo[i]) { + if(dops[i].ooo) { // Out of order execution (delay slot first) //printf("OOOE\n"); address_generation(i+1,i_regs,regs[i].regmap_entry); ds_assemble(i+1,i_regs); int adj; uint64_t bc_unneeded=branch_regs[i].u; - bc_unneeded&=~((1LL<>2 || source[i+1]!=0) { - if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc); + if(adj) emit_addimm(cc, ccadj[i] + CLOCK_ADJUST(2) - adj, cc); load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); if(internal) assem_debug("branch: internal\n"); else assem_debug("branch: external\n"); - if(internal&&is_ds[(ba[i]-start)>>2]) { + if (internal && dops[(ba[i]-start)>>2].is_ds) { ds_assemble_entry(i); } else { @@ -5296,7 +5376,7 @@ static void cjump_assemble(int i,struct regstat *i_regs) } } else if(nop) { - emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc); + emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc); void *jaddr=out; emit_jns(0); add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0); @@ -5304,11 +5384,11 @@ static void cjump_assemble(int i,struct regstat *i_regs) else { void *taken = NULL, *nottaken = NULL, *nottaken1 = NULL; do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert); - if(adj&&!invert) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc); + if(adj&&!invert) emit_addimm(cc, ccadj[i] + CLOCK_ADJUST(2) - adj, cc); //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]); assert(s1l>=0); - if(opcode[i]==4) // BEQ + if(dops[i].opcode==4) // BEQ { if(s2l>=0) emit_cmp(s1l,s2l); else emit_test(s1l,s1l); @@ -5320,7 +5400,7 @@ static void cjump_assemble(int i,struct regstat *i_regs) emit_jeq(0); } } - if(opcode[i]==5) // BNE + if(dops[i].opcode==5) // BNE { if(s2l>=0) emit_cmp(s1l,s2l); else emit_test(s1l,s1l); @@ -5332,7 +5412,7 @@ static void cjump_assemble(int i,struct regstat *i_regs) emit_jne(0); } } - if(opcode[i]==6) // BLEZ + if(dops[i].opcode==6) // BLEZ { emit_cmpimm(s1l,1); if(invert){ @@ -5343,7 +5423,7 @@ static void cjump_assemble(int i,struct regstat *i_regs) emit_jl(0); } } - if(opcode[i]==7) // BGTZ + if(dops[i].opcode==7) // BGTZ { emit_cmpimm(s1l,1); if(invert){ @@ -5357,9 +5437,9 @@ static void cjump_assemble(int i,struct regstat *i_regs) if(invert) { if(taken) set_jump_target(taken, out); #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK - if(match&&(!internal||!is_ds[(ba[i]-start)>>2])) { + if (match && (!internal || !dops[(ba[i]-start)>>2].is_ds)) { if(adj) { - emit_addimm(cc,-CLOCK_ADJUST(adj),cc); + emit_addimm(cc,-adj,cc); add_to_linker(out,ba[i],internal); }else{ emit_addnop(13); @@ -5369,14 +5449,14 @@ static void cjump_assemble(int i,struct regstat *i_regs) }else #endif { - if(adj) emit_addimm(cc,-CLOCK_ADJUST(adj),cc); + if(adj) emit_addimm(cc,-adj,cc); store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); if(internal) assem_debug("branch: internal\n"); else assem_debug("branch: external\n"); - if(internal&&is_ds[(ba[i]-start)>>2]) { + if (internal && dops[(ba[i] - start) >> 2].is_ds) { ds_assemble_entry(i); } else { @@ -5389,41 +5469,38 @@ static void cjump_assemble(int i,struct regstat *i_regs) if(nottaken1) set_jump_target(nottaken1, out); if(adj) { - if(!invert) emit_addimm(cc,CLOCK_ADJUST(adj),cc); + if(!invert) emit_addimm(cc,adj,cc); } } // (!unconditional) } // if(ooo) else { // In-order execution (branch first) - //if(likely[i]) printf("IOL\n"); - //else - //printf("IOE\n"); void *taken = NULL, *nottaken = NULL, *nottaken1 = NULL; if(!unconditional&&!nop) { //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]); assert(s1l>=0); - if((opcode[i]&0x2f)==4) // BEQ + if((dops[i].opcode&0x2f)==4) // BEQ { if(s2l>=0) emit_cmp(s1l,s2l); else emit_test(s1l,s1l); nottaken=out; emit_jne(DJT_2); } - if((opcode[i]&0x2f)==5) // BNE + if((dops[i].opcode&0x2f)==5) // BNE { if(s2l>=0) emit_cmp(s1l,s2l); else emit_test(s1l,s1l); nottaken=out; emit_jeq(DJT_2); } - if((opcode[i]&0x2f)==6) // BLEZ + if((dops[i].opcode&0x2f)==6) // BLEZ { emit_cmpimm(s1l,1); nottaken=out; emit_jge(DJT_2); } - if((opcode[i]&0x2f)==7) // BGTZ + if((dops[i].opcode&0x2f)==7) // BGTZ { emit_cmpimm(s1l,1); nottaken=out; @@ -5432,7 +5509,7 @@ static void cjump_assemble(int i,struct regstat *i_regs) } // if(!unconditional) int adj; uint64_t ds_unneeded=branch_regs[i].u; - ds_unneeded&=~((1LL<>2]) { + if (internal && dops[(ba[i] - start) >> 2].is_ds) { ds_assemble_entry(i); } else { @@ -5472,18 +5551,19 @@ static void cjump_assemble(int i,struct regstat *i_regs) if(nottaken1) set_jump_target(nottaken1, out); set_jump_target(nottaken, out); assem_debug("2:\n"); - if(!likely[i]) { - wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,ds_unneeded); - load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i+1],rs2[i+1]); - address_generation(i+1,&branch_regs[i],0); - load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG); - ds_assemble(i+1,&branch_regs[i]); - } + wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,ds_unneeded); + // load regs + 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); + 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&&!likely[i]) { + if (cc == -1) { // Cycle count isn't in a register, temporarily load it then write it out emit_loadreg(CCREG,HOST_CCREG); - emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG); + emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG); void *jaddr=out; emit_jns(0); add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0); @@ -5492,18 +5572,18 @@ static void cjump_assemble(int i,struct regstat *i_regs) else{ cc=get_reg(i_regmap,CCREG); assert(cc==HOST_CCREG); - emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc); + emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc); void *jaddr=out; emit_jns(0); - add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0); + add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0); } } } } -static void sjump_assemble(int i,struct regstat *i_regs) +static void sjump_assemble(int i, const struct regstat *i_regs) { - signed char *i_regmap=i_regs->regmap; + const signed char *i_regmap = i_regs->regmap; int cc; int match; match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); @@ -5521,39 +5601,39 @@ static void sjump_assemble(int i,struct regstat *i_regs) invert=1; // because of near cond. branches #endif - //if(opcode2[i]>=0x10) return; // FIXME (BxxZAL) - //assert(opcode2[i]<0x10||rs1[i]==0); // FIXME (BxxZAL) + //if(dops[i].opcode2>=0x10) return; // FIXME (BxxZAL) + //assert(dops[i].opcode2<0x10||dops[i].rs1==0); // FIXME (BxxZAL) - if(ooo[i]) { - s1l=get_reg(branch_regs[i].regmap,rs1[i]); + if(dops[i].ooo) { + s1l=get_reg(branch_regs[i].regmap,dops[i].rs1); } else { - s1l=get_reg(i_regmap,rs1[i]); + s1l=get_reg(i_regmap,dops[i].rs1); } - if(rs1[i]==0) + if(dops[i].rs1==0) { - if(opcode2[i]&1) unconditional=1; + if(dops[i].opcode2&1) unconditional=1; else nevertaken=1; // These are never taken (r0 is never less than zero) - //assert(opcode2[i]!=0); - //assert(opcode2[i]!=2); - //assert(opcode2[i]!=0x10); - //assert(opcode2[i]!=0x12); + //assert(dops[i].opcode2!=0); + //assert(dops[i].opcode2!=2); + //assert(dops[i].opcode2!=0x10); + //assert(dops[i].opcode2!=0x12); } - if(ooo[i]) { + if(dops[i].ooo) { // Out of order execution (delay slot first) //printf("OOOE\n"); address_generation(i+1,i_regs,regs[i].regmap_entry); ds_assemble(i+1,i_regs); int adj; uint64_t bc_unneeded=branch_regs[i].u; - bc_unneeded&=~((1LL<>2 || source[i+1]!=0) { - if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc); + if(adj) emit_addimm(cc, ccadj[i] + CLOCK_ADJUST(2) - adj, cc); load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); if(internal) assem_debug("branch: internal\n"); else assem_debug("branch: external\n"); - if(internal&&is_ds[(ba[i]-start)>>2]) { + if (internal && dops[(ba[i] - start) >> 2].is_ds) { ds_assemble_entry(i); } else { @@ -5594,7 +5674,7 @@ static void sjump_assemble(int i,struct regstat *i_regs) } } else if(nevertaken) { - emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc); + emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc); void *jaddr=out; emit_jns(0); add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0); @@ -5602,10 +5682,10 @@ static void sjump_assemble(int i,struct regstat *i_regs) else { void *nottaken = NULL; do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert); - if(adj&&!invert) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc); + if(adj&&!invert) emit_addimm(cc, ccadj[i] + CLOCK_ADJUST(2) - adj, cc); { assert(s1l>=0); - if((opcode2[i]&0xf)==0) // BLTZ/BLTZAL + if((dops[i].opcode2&0xf)==0) // BLTZ/BLTZAL { emit_test(s1l,s1l); if(invert){ @@ -5616,7 +5696,7 @@ static void sjump_assemble(int i,struct regstat *i_regs) emit_js(0); } } - if((opcode2[i]&0xf)==1) // BGEZ/BLTZAL + if((dops[i].opcode2&0xf)==1) // BGEZ/BLTZAL { emit_test(s1l,s1l); if(invert){ @@ -5631,9 +5711,9 @@ static void sjump_assemble(int i,struct regstat *i_regs) if(invert) { #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK - if(match&&(!internal||!is_ds[(ba[i]-start)>>2])) { + if (match && (!internal || !dops[(ba[i] - start) >> 2].is_ds)) { if(adj) { - emit_addimm(cc,-CLOCK_ADJUST(adj),cc); + emit_addimm(cc,-adj,cc); add_to_linker(out,ba[i],internal); }else{ emit_addnop(13); @@ -5643,14 +5723,14 @@ static void sjump_assemble(int i,struct regstat *i_regs) }else #endif { - if(adj) emit_addimm(cc,-CLOCK_ADJUST(adj),cc); + if(adj) emit_addimm(cc,-adj,cc); store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]); if(internal) assem_debug("branch: internal\n"); else assem_debug("branch: external\n"); - if(internal&&is_ds[(ba[i]-start)>>2]) { + if (internal && dops[(ba[i] - start) >> 2].is_ds) { ds_assemble_entry(i); } else { @@ -5662,7 +5742,7 @@ static void sjump_assemble(int i,struct regstat *i_regs) } if(adj) { - if(!invert) emit_addimm(cc,CLOCK_ADJUST(adj),cc); + if(!invert) emit_addimm(cc,adj,cc); } } // (!unconditional) } // if(ooo) @@ -5671,7 +5751,7 @@ static void sjump_assemble(int i,struct regstat *i_regs) // In-order execution (branch first) //printf("IOE\n"); void *nottaken = NULL; - if(rt1[i]==31) { + if(dops[i].rt1==31) { int rt,return_address; rt=get_reg(branch_regs[i].regmap,31); if(rt>=0) { @@ -5686,13 +5766,13 @@ static void sjump_assemble(int i,struct regstat *i_regs) if(!unconditional) { //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]); assert(s1l>=0); - if((opcode2[i]&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL + if((dops[i].opcode2&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL { emit_test(s1l,s1l); nottaken=out; emit_jns(DJT_1); } - if((opcode2[i]&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL + if((dops[i].opcode2&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL { emit_test(s1l,s1l); nottaken=out; @@ -5701,15 +5781,17 @@ static void sjump_assemble(int i,struct regstat *i_regs) } // if(!unconditional) int adj; uint64_t ds_unneeded=branch_regs[i].u; - ds_unneeded&=~((1LL<>2]) { + if (internal && dops[(ba[i] - start) >> 2].is_ds) { ds_assemble_entry(i); } else { @@ -5739,18 +5821,16 @@ static void sjump_assemble(int i,struct regstat *i_regs) if(!unconditional) { set_jump_target(nottaken, out); assem_debug("1:\n"); - if(!likely[i]) { - wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,ds_unneeded); - load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i+1],rs2[i+1]); - address_generation(i+1,&branch_regs[i],0); - load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG); - ds_assemble(i+1,&branch_regs[i]); - } + 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); + ds_assemble(i+1,&branch_regs[i]); cc=get_reg(branch_regs[i].regmap,CCREG); - if(cc==-1&&!likely[i]) { + if (cc == -1) { // Cycle count isn't in a register, temporarily load it then write it out emit_loadreg(CCREG,HOST_CCREG); - emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG); + emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG); void *jaddr=out; emit_jns(0); add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0); @@ -5759,28 +5839,28 @@ static void sjump_assemble(int i,struct regstat *i_regs) else{ cc=get_reg(i_regmap,CCREG); assert(cc==HOST_CCREG); - emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc); + emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc); void *jaddr=out; emit_jns(0); - add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0); + add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0); } } } } -static void pagespan_assemble(int i,struct regstat *i_regs) +static void pagespan_assemble(int i, const struct regstat *i_regs) { - int s1l=get_reg(i_regs->regmap,rs1[i]); - int s2l=get_reg(i_regs->regmap,rs2[i]); + int s1l=get_reg(i_regs->regmap,dops[i].rs1); + int s2l=get_reg(i_regs->regmap,dops[i].rs2); void *taken = NULL; void *nottaken = NULL; int unconditional=0; - if(rs1[i]==0) + if(dops[i].rs1==0) { s1l=s2l; s2l=-1; } - else if(rs2[i]==0) + else if(dops[i].rs2==0) { s2l=-1; } @@ -5791,8 +5871,8 @@ static void pagespan_assemble(int i,struct regstat *i_regs) while(hrregmap[hr]&63)!=rs1[i] && - (i_regs->regmap[hr]&63)!=rs2[i] ) + (i_regs->regmap[hr]&63)!=dops[i].rs1 && + (i_regs->regmap[hr]&63)!=dops[i].rs2 ) { addr=hr++;break; } @@ -5802,20 +5882,20 @@ static void pagespan_assemble(int i,struct regstat *i_regs) while(hrregmap[hr]&63)!=rs1[i] && - (i_regs->regmap[hr]&63)!=rs2[i] ) + (i_regs->regmap[hr]&63)!=dops[i].rs1 && + (i_regs->regmap[hr]&63)!=dops[i].rs2 ) { alt=hr++;break; } hr++; } - if((opcode[i]&0x2E)==6) // BLEZ/BGTZ needs another register + if((dops[i].opcode&0x2E)==6) // BLEZ/BGTZ needs another register { while(hrregmap[hr]&63)!=rs1[i] && - (i_regs->regmap[hr]&63)!=rs2[i] ) + (i_regs->regmap[hr]&63)!=dops[i].rs1 && + (i_regs->regmap[hr]&63)!=dops[i].rs2 ) { ntaddr=hr;break; } @@ -5823,33 +5903,33 @@ static void pagespan_assemble(int i,struct regstat *i_regs) } } assert(hrregmap,31); emit_movimm(start+i*4+8,rt); unconditional=1; } - if(opcode[i]==0&&(opcode2[i]&0x3E)==8) // JR/JALR + if(dops[i].opcode==0&&(dops[i].opcode2&0x3E)==8) // JR/JALR { emit_mov(s1l,addr); - if(opcode2[i]==9) // JALR + if(dops[i].opcode2==9) // JALR { - int rt=get_reg(i_regs->regmap,rt1[i]); + int rt=get_reg(i_regs->regmap,dops[i].rt1); emit_movimm(start+i*4+8,rt); } } - if((opcode[i]&0x3f)==4) // BEQ + if((dops[i].opcode&0x3f)==4) // BEQ { - if(rs1[i]==rs2[i]) + if(dops[i].rs1==dops[i].rs2) { unconditional=1; } @@ -5870,7 +5950,7 @@ static void pagespan_assemble(int i,struct regstat *i_regs) emit_cmovne_reg(alt,addr); } } - if((opcode[i]&0x3f)==5) // BNE + if((dops[i].opcode&0x3f)==5) // BNE { #ifdef HAVE_CMOV_IMM if(s2l>=0) emit_cmp(s1l,s2l); @@ -5884,7 +5964,7 @@ static void pagespan_assemble(int i,struct regstat *i_regs) emit_cmovne_reg(alt,addr); #endif } - if((opcode[i]&0x3f)==0x14) // BEQL + if((dops[i].opcode&0x3f)==0x14) // BEQL { if(s2l>=0) emit_cmp(s1l,s2l); else emit_test(s1l,s1l); @@ -5892,7 +5972,7 @@ static void pagespan_assemble(int i,struct regstat *i_regs) nottaken=out; emit_jne(0); } - if((opcode[i]&0x3f)==0x15) // BNEL + if((dops[i].opcode&0x3f)==0x15) // BNEL { if(s2l>=0) emit_cmp(s1l,s2l); else emit_test(s1l,s1l); @@ -5900,30 +5980,30 @@ static void pagespan_assemble(int i,struct regstat *i_regs) emit_jeq(0); if(taken) set_jump_target(taken, out); } - if((opcode[i]&0x3f)==6) // BLEZ + if((dops[i].opcode&0x3f)==6) // BLEZ { emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr); emit_cmpimm(s1l,1); emit_cmovl_reg(alt,addr); } - if((opcode[i]&0x3f)==7) // BGTZ + if((dops[i].opcode&0x3f)==7) // BGTZ { emit_mov2imm_compact(ba[i],addr,start+i*4+8,ntaddr); emit_cmpimm(s1l,1); emit_cmovl_reg(ntaddr,addr); } - if((opcode[i]&0x3f)==0x16) // BLEZL + if((dops[i].opcode&0x3f)==0x16) // BLEZL { - assert((opcode[i]&0x3f)!=0x16); + assert((dops[i].opcode&0x3f)!=0x16); } - if((opcode[i]&0x3f)==0x17) // BGTZL + if((dops[i].opcode&0x3f)==0x17) // BGTZL { - assert((opcode[i]&0x3f)!=0x17); + assert((dops[i].opcode&0x3f)!=0x17); } - assert(opcode[i]!=1); // BLTZ/BGEZ + assert(dops[i].opcode!=1); // BLTZ/BGEZ //FIXME: Check CSREG - if(opcode[i]==0x11 && opcode2[i]==0x08 ) { + if(dops[i].opcode==0x11 && dops[i].opcode2==0x08 ) { if((source[i]&0x30000)==0) // BC1F { emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt); @@ -5952,7 +6032,7 @@ static void pagespan_assemble(int i,struct regstat *i_regs) assert(i_regs->regmap[HOST_CCREG]==CCREG); wb_dirtys(regs[i].regmap,regs[i].dirty); - if(likely[i]||unconditional) + if(unconditional) { emit_movimm(ba[i],HOST_BTREG); } @@ -5968,25 +6048,9 @@ static void pagespan_assemble(int i,struct regstat *i_regs) emit_extjump_ds(branch_addr, target_addr); if(compiled_target_addr) { set_jump_target(branch_addr, compiled_target_addr); - add_link(target_addr,stub); + add_jump_out(target_addr,stub); } else set_jump_target(branch_addr, stub); - if(likely[i]) { - // Not-taken path - set_jump_target(nottaken, out); - wb_dirtys(regs[i].regmap,regs[i].dirty); - void *branch_addr=out; - emit_jmp(0); - int target_addr=start+i*4+8; - void *stub=out; - void *compiled_target_addr=check_addr(target_addr); - emit_extjump_ds(branch_addr, target_addr); - if(compiled_target_addr) { - set_jump_target(branch_addr, compiled_target_addr); - add_link(target_addr,stub); - } - else set_jump_target(branch_addr, stub); - } } // Assemble the delay slot for the above @@ -5997,53 +6061,21 @@ static void pagespan_ds() u_int page=get_page(vaddr); u_int vpage=get_vpage(vaddr); ll_add(jump_dirty+vpage,vaddr,(void *)out); - do_dirty_stub_ds(); + do_dirty_stub_ds(slen*4); ll_add(jump_in+page,vaddr,(void *)out); assert(regs[0].regmap_entry[HOST_CCREG]==CCREG); if(regs[0].regmap[HOST_CCREG]!=CCREG) wb_register(CCREG,regs[0].regmap_entry,regs[0].wasdirty); if(regs[0].regmap[HOST_BTREG]!=BTREG) emit_writeword(HOST_BTREG,&branch_target); - load_regs(regs[0].regmap_entry,regs[0].regmap,rs1[0],rs2[0]); + load_regs(regs[0].regmap_entry,regs[0].regmap,dops[0].rs1,dops[0].rs2); address_generation(0,®s[0],regs[0].regmap_entry); - if(itype[0]==STORE||itype[0]==STORELR||(opcode[0]&0x3b)==0x39||(opcode[0]&0x3b)==0x3a) + if (ram_offset && (dops[0].is_load || dops[0].is_store)) + load_regs(regs[0].regmap_entry,regs[0].regmap,ROREG,ROREG); + if (dops[0].is_store) load_regs(regs[0].regmap_entry,regs[0].regmap,INVCP,INVCP); is_delayslot=0; - switch(itype[0]) { - case ALU: - alu_assemble(0,®s[0]);break; - case IMM16: - imm16_assemble(0,®s[0]);break; - case SHIFT: - shift_assemble(0,®s[0]);break; - case SHIFTIMM: - shiftimm_assemble(0,®s[0]);break; - case LOAD: - load_assemble(0,®s[0]);break; - case LOADLR: - loadlr_assemble(0,®s[0]);break; - case STORE: - store_assemble(0,®s[0]);break; - case STORELR: - storelr_assemble(0,®s[0]);break; - case COP0: - cop0_assemble(0,®s[0]);break; - case COP1: - cop1_assemble(0,®s[0]);break; - case C1LS: - c1ls_assemble(0,®s[0]);break; - case COP2: - cop2_assemble(0,®s[0]);break; - case C2LS: - c2ls_assemble(0,®s[0]);break; - case C2OP: - c2op_assemble(0,®s[0]);break; - case MULTDIV: - multdiv_assemble(0,®s[0]); - multdiv_prepare_stall(0,®s[0]); - break; - case MOV: - mov_assemble(0,®s[0]);break; + switch (dops[0].itype) { case SYSCALL: case HLECALL: case INTCALL: @@ -6053,6 +6085,9 @@ static void pagespan_ds() case CJUMP: case SJUMP: SysPrintf("Jump in the delay slot. This is probably a bug.\n"); + break; + default: + assemble(0, ®s[0], 0); } int btaddr=get_reg(regs[0].regmap,BTREG); if(btaddr<0) { @@ -6099,10 +6134,10 @@ void unneeded_registers(int istart,int iend,int r) for (i=iend;i>=istart;i--) { //printf("unneeded registers i=%d (%d,%d) r=%d\n",i,istart,iend,r); - if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP) + if(dops[i].is_jump) { // If subroutine call, flag return address as a possible branch target - if(rt1[i]==31 && i=(start+slen*4)) { @@ -6111,32 +6146,19 @@ void unneeded_registers(int istart,int iend,int r) gte_u=gte_u_unknown; branch_unneeded_reg[i]=u; // Merge in delay slot - u|=(1LL<>2]=1; + dops[(ba[i]-start)>>2].bt=1; if(ba[i]<=start+i*4) { // Backward branch - if(is_ujump(i)) + if(dops[i].is_ujump) { // Unconditional branch temp_u=1; @@ -6147,26 +6169,13 @@ void unneeded_registers(int istart,int iend,int r) temp_gte_u&=gte_unneeded[i+2]; } // Merge in delay slot - temp_u|=(1LL<>2]=gte_u_unknown; } } /*else*/ if(1) { - if (is_ujump(i)) + if (dops[i].is_ujump) { // Unconditional branch u=unneeded_reg[(ba[i]-start)>>2]; gte_u=gte_unneeded[(ba[i]-start)>>2]; branch_unneeded_reg[i]=u; // Merge in delay slot - u|=(1LL<>2]; branch_unneeded_reg[i]=b; // Branch delay slot - b|=(1LL<=istart;i--) { - if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP) + if(dops[i].is_jump) { if(ba[i]=(start+slen*4)) { // Branch out of this block, flush all regs - if (is_ujump(i)) + if (dops[i].is_ujump) { // Unconditional branch will_dirty_i=0; @@ -6299,17 +6297,17 @@ void clean_registers(int istart,int iend,int wr) // Merge in delay slot (will dirty) for(r=0;r33) will_dirty_i&=~(1<33) will_dirty_i&=~(1<33) will_dirty_i&=~(1<33) will_dirty_i&=~(1<33) temp_will_dirty&=~(1<33) temp_will_dirty&=~(1<33) temp_will_dirty&=~(1<33) temp_will_dirty&=~(1<33) will_dirty_i&=~(1<33) will_dirty_i&=~(1<>2]>>(target_reg&63))&1)<>2]>>(target_reg&63))&1)<>2].regmap_entry[r]) { - will_dirty[i+1]&=will_dirty[(ba[i]-start)>>2]&(1<>2]&(1<33) will_dirty_i&=~(1<33) will_dirty_i&=~(1<33) will_dirty_i&=~(1<istart) { - if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP) + if (!dops[i].is_jump) { // Don't store a register immediately after writing it, // may prevent dual-issue. - if((regs[i].regmap[r]&63)==rt1[i-1]) wont_dirty_i|=1<>r)&1) { - printf(" r%d",r); - } - } - printf("\n");*/ - - //if(i==istart||(itype[i-1]!=RJUMP&&itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP)) { regs[i].dirty|=will_dirty_i; #ifndef DESTRUCTIVE_WRITEBACK regs[i].dirty&=wont_dirty_i; - if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP) + if(dops[i].is_jump) { - if (i < iend-1 && !is_ujump(i)) { + if (i < iend-1 && !dops[i].is_ujump) { for(r=0;r>14):*ba);break; + printf (" %x: %s r%d,r%d,%8x\n",start+i*4,insn[i],dops[i].rs1,dops[i].rs2,i?start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14):*ba);break; case SJUMP: - printf (" %x: %s r%d,%8x\n",start+i*4,insn[i],rs1[i],start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14));break; + printf (" %x: %s r%d,%8x\n",start+i*4,insn[i],dops[i].rs1,start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14));break; case RJUMP: - if (opcode[i]==0x9&&rt1[i]!=31) - printf (" %x: %s r%d,r%d\n",start+i*4,insn[i],rt1[i],rs1[i]); + if (dops[i].opcode==0x9&&dops[i].rt1!=31) + printf (" %x: %s r%d,r%d\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1); else - printf (" %x: %s r%d\n",start+i*4,insn[i],rs1[i]); + printf (" %x: %s r%d\n",start+i*4,insn[i],dops[i].rs1); break; case SPAN: - printf (" %x: %s (pagespan) r%d,r%d,%8x\n",start+i*4,insn[i],rs1[i],rs2[i],ba[i]);break; + printf (" %x: %s (pagespan) r%d,r%d,%8x\n",start+i*4,insn[i],dops[i].rs1,dops[i].rs2,ba[i]);break; case IMM16: - if(opcode[i]==0xf) //LUI - printf (" %x: %s r%d,%4x0000\n",start+i*4,insn[i],rt1[i],imm[i]&0xffff); + if(dops[i].opcode==0xf) //LUI + printf (" %x: %s r%d,%4x0000\n",start+i*4,insn[i],dops[i].rt1,imm[i]&0xffff); else - printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],rt1[i],rs1[i],imm[i]); + printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,imm[i]); break; case LOAD: case LOADLR: - printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],rt1[i],rs1[i],imm[i]); + printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,imm[i]); break; case STORE: case STORELR: - printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],rs2[i],rs1[i],imm[i]); + printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],dops[i].rs2,dops[i].rs1,imm[i]); break; case ALU: case SHIFT: - printf (" %x: %s r%d,r%d,r%d\n",start+i*4,insn[i],rt1[i],rs1[i],rs2[i]); + printf (" %x: %s r%d,r%d,r%d\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,dops[i].rs2); break; case MULTDIV: - printf (" %x: %s r%d,r%d\n",start+i*4,insn[i],rs1[i],rs2[i]); + printf (" %x: %s r%d,r%d\n",start+i*4,insn[i],dops[i].rs1,dops[i].rs2); break; case SHIFTIMM: - printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],rt1[i],rs1[i],imm[i]); + printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,imm[i]); break; case MOV: - if((opcode2[i]&0x1d)==0x10) - printf (" %x: %s r%d\n",start+i*4,insn[i],rt1[i]); - else if((opcode2[i]&0x1d)==0x11) - printf (" %x: %s r%d\n",start+i*4,insn[i],rs1[i]); + if((dops[i].opcode2&0x1d)==0x10) + printf (" %x: %s r%d\n",start+i*4,insn[i],dops[i].rt1); + else if((dops[i].opcode2&0x1d)==0x11) + printf (" %x: %s r%d\n",start+i*4,insn[i],dops[i].rs1); else printf (" %x: %s\n",start+i*4,insn[i]); break; case COP0: - if(opcode2[i]==0) - printf (" %x: %s r%d,cpr0[%d]\n",start+i*4,insn[i],rt1[i],(source[i]>>11)&0x1f); // MFC0 - else if(opcode2[i]==4) - printf (" %x: %s r%d,cpr0[%d]\n",start+i*4,insn[i],rs1[i],(source[i]>>11)&0x1f); // MTC0 + if(dops[i].opcode2==0) + printf (" %x: %s r%d,cpr0[%d]\n",start+i*4,insn[i],dops[i].rt1,(source[i]>>11)&0x1f); // MFC0 + else if(dops[i].opcode2==4) + printf (" %x: %s r%d,cpr0[%d]\n",start+i*4,insn[i],dops[i].rs1,(source[i]>>11)&0x1f); // MTC0 else printf (" %x: %s\n",start+i*4,insn[i]); break; case COP1: - if(opcode2[i]<3) - printf (" %x: %s r%d,cpr1[%d]\n",start+i*4,insn[i],rt1[i],(source[i]>>11)&0x1f); // MFC1 - else if(opcode2[i]>3) - printf (" %x: %s r%d,cpr1[%d]\n",start+i*4,insn[i],rs1[i],(source[i]>>11)&0x1f); // MTC1 + if(dops[i].opcode2<3) + printf (" %x: %s r%d,cpr1[%d]\n",start+i*4,insn[i],dops[i].rt1,(source[i]>>11)&0x1f); // MFC1 + else if(dops[i].opcode2>3) + printf (" %x: %s r%d,cpr1[%d]\n",start+i*4,insn[i],dops[i].rs1,(source[i]>>11)&0x1f); // MTC1 else printf (" %x: %s\n",start+i*4,insn[i]); break; case COP2: - if(opcode2[i]<3) - printf (" %x: %s r%d,cpr2[%d]\n",start+i*4,insn[i],rt1[i],(source[i]>>11)&0x1f); // MFC2 - else if(opcode2[i]>3) - printf (" %x: %s r%d,cpr2[%d]\n",start+i*4,insn[i],rs1[i],(source[i]>>11)&0x1f); // MTC2 + if(dops[i].opcode2<3) + printf (" %x: %s r%d,cpr2[%d]\n",start+i*4,insn[i],dops[i].rt1,(source[i]>>11)&0x1f); // MFC2 + else if(dops[i].opcode2>3) + printf (" %x: %s r%d,cpr2[%d]\n",start+i*4,insn[i],dops[i].rs1,(source[i]>>11)&0x1f); // MTC2 else printf (" %x: %s\n",start+i*4,insn[i]); break; case C1LS: - printf (" %x: %s cpr1[%d],r%d+%x\n",start+i*4,insn[i],(source[i]>>16)&0x1f,rs1[i],imm[i]); + printf (" %x: %s cpr1[%d],r%d+%x\n",start+i*4,insn[i],(source[i]>>16)&0x1f,dops[i].rs1,imm[i]); break; case C2LS: - printf (" %x: %s cpr2[%d],r%d+%x\n",start+i*4,insn[i],(source[i]>>16)&0x1f,rs1[i],imm[i]); + printf (" %x: %s cpr2[%d],r%d+%x\n",start+i*4,insn[i],(source[i]>>16)&0x1f,dops[i].rs1,imm[i]); break; case INTCALL: printf (" %x: %s (INTCALL)\n",start+i*4,insn[i]); @@ -6847,6 +6827,7 @@ void new_dynarec_clear_full(void) literalcount=0; stop_after_jal=0; inv_code_start=inv_code_end=~0; + f1_hack=0; // TLB for(n=0;n<4096;n++) ll_clear(jump_in+n); for(n=0;n<4096;n++) ll_clear(jump_out+n); @@ -6899,9 +6880,7 @@ void new_dynarec_init(void) #endif arch_init(); new_dynarec_test(); -#ifndef RAM_FIXED ram_offset=(uintptr_t)rdram-0x80000000; -#endif if (ram_offset!=0) SysPrintf("warning: RAM is not directly mapped, performance will suffer\n"); } @@ -7092,6 +7071,27 @@ int new_recompile_block(u_int addr) ll_add_flags(jump_in+page,start,state_rflags,(void *)beginning); return 0; } + else if (f1_hack == ~0u || (f1_hack != 0 && start == f1_hack)) { + void *beginning = start_block(); + u_int page = get_page(start); + emit_readword(&psxRegs.GPR.n.sp, 0); + emit_readptr(&mem_rtab, 1); + emit_shrimm(0, 12, 2); + emit_readptr_dualindexedx_ptrlen(1, 2, 1); + emit_addimm(0, 0x18, 0); + emit_adds_ptr(1, 1, 1); + emit_ldr_dualindexed(1, 0, 0); + emit_writeword(0, &psxRegs.GPR.r[26]); // lw k0, 0x18(sp) + emit_far_call(get_addr_ht); + emit_jmpreg(0); // jr k0 + literal_pool(0); + end_block(beginning); + + ll_add_flags(jump_in + page, start, state_rflags, beginning); + SysPrintf("F1 hack to %08x\n", start); + f1_hack = start; + return 0; + } source = get_source_start(start, &pagelimit); if (source == NULL) { @@ -7119,9 +7119,11 @@ int new_recompile_block(u_int addr) /* Pass 1 disassembly */ for(i=0;!done;i++) { - bt[i]=0;likely[i]=0;ooo[i]=0;op2=0; + dops[i].bt=0; + dops[i].ooo=0; + op2=0; minimum_free_regs[i]=0; - opcode[i]=op=source[i]>>26; + dops[i].opcode=op=source[i]>>26; switch(op) { case 0x00: strcpy(insn[i],"special"); type=NI; @@ -7190,18 +7192,18 @@ int new_recompile_block(u_int addr) { 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 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 0x12: strcpy(insn[i],"BLTZALL"); type=SJUMP; break; + //case 0x13: strcpy(insn[i],"BGEZALL"); type=SJUMP; break; } break; case 0x02: strcpy(insn[i],"J"); type=UJUMP; break; @@ -7303,168 +7305,155 @@ int new_recompile_block(u_int addr) SysPrintf("NI %08x @%08x (%08x)\n", source[i], addr + i*4, addr); break; } - itype[i]=type; - opcode2[i]=op2; + dops[i].itype=type; + dops[i].opcode2=op2; /* Get registers/immediates */ - lt1[i]=0; - dep1[i]=0; - dep2[i]=0; + dops[i].lt1=0; gte_rs[i]=gte_rt[i]=0; switch(type) { case LOAD: - rs1[i]=(source[i]>>21)&0x1f; - rs2[i]=0; - rt1[i]=(source[i]>>16)&0x1f; - rt2[i]=0; + dops[i].rs1=(source[i]>>21)&0x1f; + dops[i].rs2=0; + dops[i].rt1=(source[i]>>16)&0x1f; + dops[i].rt2=0; imm[i]=(short)source[i]; break; case STORE: case STORELR: - rs1[i]=(source[i]>>21)&0x1f; - rs2[i]=(source[i]>>16)&0x1f; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=(source[i]>>21)&0x1f; + dops[i].rs2=(source[i]>>16)&0x1f; + dops[i].rt1=0; + dops[i].rt2=0; imm[i]=(short)source[i]; break; case LOADLR: // LWL/LWR only load part of the register, // therefore the target register must be treated as a source too - rs1[i]=(source[i]>>21)&0x1f; - rs2[i]=(source[i]>>16)&0x1f; - rt1[i]=(source[i]>>16)&0x1f; - rt2[i]=0; + dops[i].rs1=(source[i]>>21)&0x1f; + dops[i].rs2=(source[i]>>16)&0x1f; + dops[i].rt1=(source[i]>>16)&0x1f; + dops[i].rt2=0; imm[i]=(short)source[i]; - if(op==0x26) dep1[i]=rt1[i]; // LWR break; case IMM16: - if (op==0x0f) rs1[i]=0; // LUI instruction has no source register - else rs1[i]=(source[i]>>21)&0x1f; - rs2[i]=0; - rt1[i]=(source[i]>>16)&0x1f; - rt2[i]=0; + if (op==0x0f) dops[i].rs1=0; // LUI instruction has no source register + else dops[i].rs1=(source[i]>>21)&0x1f; + dops[i].rs2=0; + dops[i].rt1=(source[i]>>16)&0x1f; + dops[i].rt2=0; if(op>=0x0c&&op<=0x0e) { // ANDI/ORI/XORI imm[i]=(unsigned short)source[i]; }else{ imm[i]=(short)source[i]; } - if(op==0x0d||op==0x0e) dep1[i]=rs1[i]; // ORI/XORI break; case UJUMP: - rs1[i]=0; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=0; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; // The JAL instruction writes to r31. if (op&1) { - rt1[i]=31; + dops[i].rt1=31; } - rs2[i]=CCREG; + dops[i].rs2=CCREG; break; case RJUMP: - rs1[i]=(source[i]>>21)&0x1f; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=(source[i]>>21)&0x1f; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; // The JALR instruction writes to rd. if (op2&1) { - rt1[i]=(source[i]>>11)&0x1f; + dops[i].rt1=(source[i]>>11)&0x1f; } - rs2[i]=CCREG; + dops[i].rs2=CCREG; break; case CJUMP: - rs1[i]=(source[i]>>21)&0x1f; - rs2[i]=(source[i]>>16)&0x1f; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=(source[i]>>21)&0x1f; + dops[i].rs2=(source[i]>>16)&0x1f; + dops[i].rt1=0; + dops[i].rt2=0; if(op&2) { // BGTZ/BLEZ - rs2[i]=0; + dops[i].rs2=0; } - likely[i]=op>>4; break; case SJUMP: - rs1[i]=(source[i]>>21)&0x1f; - rs2[i]=CCREG; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=(source[i]>>21)&0x1f; + dops[i].rs2=CCREG; + dops[i].rt1=0; + dops[i].rt2=0; if(op2&0x10) { // BxxAL - rt1[i]=31; + dops[i].rt1=31; // NOTE: If the branch is not taken, r31 is still overwritten } - likely[i]=(op2&2)>>1; break; case ALU: - rs1[i]=(source[i]>>21)&0x1f; // source - rs2[i]=(source[i]>>16)&0x1f; // subtract amount - rt1[i]=(source[i]>>11)&0x1f; // destination - rt2[i]=0; - if(op2>=0x24&&op2<=0x27) { // AND/OR/XOR/NOR - dep1[i]=rs1[i];dep2[i]=rs2[i]; - } - else if(op2>=0x2c&&op2<=0x2f) { // DADD/DSUB - dep1[i]=rs1[i];dep2[i]=rs2[i]; - } + dops[i].rs1=(source[i]>>21)&0x1f; // source + dops[i].rs2=(source[i]>>16)&0x1f; // subtract amount + dops[i].rt1=(source[i]>>11)&0x1f; // destination + dops[i].rt2=0; break; case MULTDIV: - rs1[i]=(source[i]>>21)&0x1f; // source - rs2[i]=(source[i]>>16)&0x1f; // divisor - rt1[i]=HIREG; - rt2[i]=LOREG; + dops[i].rs1=(source[i]>>21)&0x1f; // source + dops[i].rs2=(source[i]>>16)&0x1f; // divisor + dops[i].rt1=HIREG; + dops[i].rt2=LOREG; break; case MOV: - rs1[i]=0; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; - if(op2==0x10) rs1[i]=HIREG; // MFHI - if(op2==0x11) rt1[i]=HIREG; // MTHI - if(op2==0x12) rs1[i]=LOREG; // MFLO - if(op2==0x13) rt1[i]=LOREG; // MTLO - if((op2&0x1d)==0x10) rt1[i]=(source[i]>>11)&0x1f; // MFxx - if((op2&0x1d)==0x11) rs1[i]=(source[i]>>21)&0x1f; // MTxx - dep1[i]=rs1[i]; + dops[i].rs1=0; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; + if(op2==0x10) dops[i].rs1=HIREG; // MFHI + if(op2==0x11) dops[i].rt1=HIREG; // MTHI + if(op2==0x12) dops[i].rs1=LOREG; // MFLO + if(op2==0x13) dops[i].rt1=LOREG; // MTLO + if((op2&0x1d)==0x10) dops[i].rt1=(source[i]>>11)&0x1f; // MFxx + if((op2&0x1d)==0x11) dops[i].rs1=(source[i]>>21)&0x1f; // MTxx break; case SHIFT: - rs1[i]=(source[i]>>16)&0x1f; // target of shift - rs2[i]=(source[i]>>21)&0x1f; // shift amount - rt1[i]=(source[i]>>11)&0x1f; // destination - rt2[i]=0; + dops[i].rs1=(source[i]>>16)&0x1f; // target of shift + dops[i].rs2=(source[i]>>21)&0x1f; // shift amount + dops[i].rt1=(source[i]>>11)&0x1f; // destination + dops[i].rt2=0; break; case SHIFTIMM: - rs1[i]=(source[i]>>16)&0x1f; - rs2[i]=0; - rt1[i]=(source[i]>>11)&0x1f; - rt2[i]=0; + dops[i].rs1=(source[i]>>16)&0x1f; + dops[i].rs2=0; + dops[i].rt1=(source[i]>>11)&0x1f; + dops[i].rt2=0; imm[i]=(source[i]>>6)&0x1f; // DSxx32 instructions if(op2>=0x3c) imm[i]|=0x20; break; case COP0: - rs1[i]=0; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; - if(op2==0||op2==2) rt1[i]=(source[i]>>16)&0x1F; // MFC0/CFC0 - if(op2==4||op2==6) rs1[i]=(source[i]>>16)&0x1F; // MTC0/CTC0 - if(op2==4&&((source[i]>>11)&0x1f)==12) rt2[i]=CSREG; // Status - if(op2==16) if((source[i]&0x3f)==0x18) rs2[i]=CCREG; // ERET + dops[i].rs1=0; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; + if(op2==0||op2==2) dops[i].rt1=(source[i]>>16)&0x1F; // MFC0/CFC0 + if(op2==4||op2==6) dops[i].rs1=(source[i]>>16)&0x1F; // MTC0/CTC0 + if(op2==4&&((source[i]>>11)&0x1f)==12) dops[i].rt2=CSREG; // Status + if(op2==16) if((source[i]&0x3f)==0x18) dops[i].rs2=CCREG; // ERET break; case COP1: - rs1[i]=0; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; - if(op2<3) rt1[i]=(source[i]>>16)&0x1F; // MFC1/DMFC1/CFC1 - if(op2>3) rs1[i]=(source[i]>>16)&0x1F; // MTC1/DMTC1/CTC1 - rs2[i]=CSREG; + dops[i].rs1=0; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; + if(op2<3) dops[i].rt1=(source[i]>>16)&0x1F; // MFC1/DMFC1/CFC1 + if(op2>3) dops[i].rs1=(source[i]>>16)&0x1F; // MTC1/DMTC1/CTC1 + dops[i].rs2=CSREG; break; case COP2: - rs1[i]=0; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; - if(op2<3) rt1[i]=(source[i]>>16)&0x1F; // MFC2/CFC2 - if(op2>3) rs1[i]=(source[i]>>16)&0x1F; // MTC2/CTC2 - rs2[i]=CSREG; + dops[i].rs1=0; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; + if(op2<3) dops[i].rt1=(source[i]>>16)&0x1F; // MFC2/CFC2 + if(op2>3) dops[i].rs1=(source[i]>>16)&0x1F; // MTC2/CTC2 + dops[i].rs2=CSREG; int gr=(source[i]>>11)&0x1F; switch(op2) { @@ -7475,26 +7464,26 @@ int new_recompile_block(u_int addr) } break; case C1LS: - rs1[i]=(source[i]>>21)&0x1F; - rs2[i]=CSREG; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=(source[i]>>21)&0x1F; + dops[i].rs2=CSREG; + dops[i].rt1=0; + dops[i].rt2=0; imm[i]=(short)source[i]; break; case C2LS: - rs1[i]=(source[i]>>21)&0x1F; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=(source[i]>>21)&0x1F; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; imm[i]=(short)source[i]; if(op==0x32) gte_rt[i]=1ll<<((source[i]>>16)&0x1F); // LWC2 else gte_rs[i]=1ll<<((source[i]>>16)&0x1F); // SWC2 break; case C2OP: - rs1[i]=0; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=0; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; gte_rs[i]=gte_reg_reads[source[i]&0x3f]; gte_rt[i]=gte_reg_writes[source[i]&0x3f]; gte_rt[i]|=1ll<<63; // every op changes flags @@ -7508,63 +7497,82 @@ int new_recompile_block(u_int addr) case SYSCALL: case HLECALL: case INTCALL: - rs1[i]=CCREG; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=CCREG; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; break; default: - rs1[i]=0; - rs2[i]=0; - rt1[i]=0; - rt2[i]=0; + dops[i].rs1=0; + dops[i].rs2=0; + dops[i].rt1=0; + dops[i].rt2=0; } /* Calculate branch target addresses */ if(type==UJUMP) ba[i]=((start+i*4+4)&0xF0000000)|(((unsigned int)source[i]<<6)>>4); - else if(type==CJUMP&&rs1[i]==rs2[i]&&(op&1)) + else if(type==CJUMP&&dops[i].rs1==dops[i].rs2&&(op&1)) ba[i]=start+i*4+8; // Ignore never taken branch - else if(type==SJUMP&&rs1[i]==0&&!(op2&1)) + else if(type==SJUMP&&dops[i].rs1==0&&!(op2&1)) ba[i]=start+i*4+8; // Ignore never taken branch else if(type==CJUMP||type==SJUMP) ba[i]=start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14); else ba[i]=-1; - if (i > 0 && is_jump(i-1)) { + + /* simplify always (not)taken branches */ + if (type == CJUMP && dops[i].rs1 == dops[i].rs2) { + dops[i].rs1 = dops[i].rs2 = 0; + if (!(op & 1)) { + dops[i].itype = type = UJUMP; + dops[i].rs2 = CCREG; + } + } + else if (type == SJUMP && dops[i].rs1 == 0 && (op2 & 1)) + dops[i].itype = type = UJUMP; + + dops[i].is_jump = (dops[i].itype == RJUMP || dops[i].itype == UJUMP || dops[i].itype == CJUMP || dops[i].itype == SJUMP); + dops[i].is_ujump = (dops[i].itype == RJUMP || dops[i].itype == UJUMP); // || (source[i] >> 16) == 0x1000 // beq r0,r0 + dops[i].is_load = (dops[i].itype == LOAD || dops[i].itype == LOADLR || op == 0x32); // LWC2 + dops[i].is_store = (dops[i].itype == STORE || dops[i].itype == STORELR || op == 0x3a); // SWC2 + + /* messy cases to just pass over to the interpreter */ + if (i > 0 && dops[i-1].is_jump) { int do_in_intrp=0; // branch in delay slot? - if(type==RJUMP||type==UJUMP||type==CJUMP||type==SJUMP) { + if (dops[i].is_jump) { // don't handle first branch and call interpreter if it's hit SysPrintf("branch in delay slot @%08x (%08x)\n", addr + i*4, addr); do_in_intrp=1; } // basic load delay detection - else if((type==LOAD||type==LOADLR||type==COP0||type==COP2||type==C2LS)&&rt1[i]!=0) { + else if((type==LOAD||type==LOADLR||type==COP0||type==COP2||type==C2LS)&&dops[i].rt1!=0) { int t=(ba[i-1]-start)/4; - if(0 <= t && t < i &&(rt1[i]==rs1[t]||rt1[i]==rs2[t])&&itype[t]!=CJUMP&&itype[t]!=SJUMP) { + if(0 <= t && t < i &&(dops[i].rt1==dops[t].rs1||dops[i].rt1==dops[t].rs2)&&dops[t].itype!=CJUMP&&dops[t].itype!=SJUMP) { // jump target wants DS result - potential load delay effect SysPrintf("load delay @%08x (%08x)\n", addr + i*4, addr); do_in_intrp=1; - bt[t+1]=1; // expected return from interpreter + dops[t+1].bt=1; // expected return from interpreter } - else if(i>=2&&rt1[i-2]==2&&rt1[i]==2&&rs1[i]!=2&&rs2[i]!=2&&rs1[i-1]!=2&&rs2[i-1]!=2&& - !(i>=3&&is_jump(i-3))) { + else if(i>=2&&dops[i-2].rt1==2&&dops[i].rt1==2&&dops[i].rs1!=2&&dops[i].rs2!=2&&dops[i-1].rs1!=2&&dops[i-1].rs2!=2&& + !(i>=3&&dops[i-3].is_jump)) { // v0 overwrite like this is a sign of trouble, bail out SysPrintf("v0 overwrite @%08x (%08x)\n", addr + i*4, addr); do_in_intrp=1; } } if(do_in_intrp) { - rs1[i-1]=CCREG; - rs2[i-1]=rt1[i-1]=rt2[i-1]=0; + dops[i-1].rs1=CCREG; + dops[i-1].rs2=dops[i-1].rt1=dops[i-1].rt2=0; ba[i-1]=-1; - itype[i-1]=INTCALL; + dops[i-1].itype=INTCALL; done=2; i--; // don't compile the DS } } + /* Is this the end of the block? */ - if (i > 0 && is_ujump(i-1)) { - if(rt1[i-1]==0) { // Continue past subroutine call (JAL) + if (i > 0 && dops[i-1].is_ujump) { + if(dops[i-1].rt1==0) { // Continue past subroutine call (JAL) done=2; } else { @@ -7577,8 +7585,8 @@ int new_recompile_block(u_int addr) // Don't get too close to the limit if(i>MAXBLOCK/2) done=1; } - if(itype[i]==SYSCALL&&stop_after_jal) done=1; - if(itype[i]==HLECALL||itype[i]==INTCALL) done=2; + if(dops[i].itype==SYSCALL&&stop_after_jal) done=1; + if(dops[i].itype==HLECALL||dops[i].itype==INTCALL) done=2; if(done==2) { // Does the block continue due to a branch? for(j=i-1;j>=0;j--) @@ -7593,19 +7601,37 @@ int new_recompile_block(u_int addr) assert(start+i*40); + /* 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; + } + } + /* Pass 2 - Register dependencies and branch targets */ unneeded_registers(0,slen-1,0); @@ -7628,7 +7654,7 @@ int new_recompile_block(u_int addr) if((u_int)addr&1) { // First instruction is delay slot cc=-1; - bt[1]=1; + dops[1].bt=1; ds=1; unneeded_reg[0]=1; current.regmap[HOST_BTREG]=BTREG; @@ -7636,7 +7662,7 @@ int new_recompile_block(u_int addr) for(i=0;i0&&(itype[i-1]==STORE||itype[i-1]==STORELR||(itype[i-1]==C2LS&&opcode[i-1]==0x3a))&&(u_int)imm[i-1]<0x800) - current.waswritten|=1<=0x800) - current.waswritten&=~(1<0&&(dops[i-1].itype==STORE||dops[i-1].itype==STORELR||(dops[i-1].itype==C2LS&&dops[i-1].opcode==0x3a))&&(u_int)imm[i-1]<0x800) + current.waswritten|=1<=0x800) + current.waswritten&=~(1<0) { current.wasdirty=current.dirty; - switch(itype[i-1]) { + switch(dops[i-1].itype) { case UJUMP: memcpy(&branch_regs[i-1],¤t,sizeof(current)); branch_regs[i-1].isconst=0; branch_regs[i-1].wasconst=0; - branch_regs[i-1].u=branch_unneeded_reg[i-1]&~((1LL<0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i]==SYSCALL||itype[i]==HLECALL)) + ccadj[i] = CLOCK_ADJUST(cc); + if (i > 0 && (dops[i-1].is_jump || dops[i].itype == SYSCALL || dops[i].itype == HLECALL)) { cc=0; } #if !defined(DRC_DBG) - else if(itype[i]==C2OP&>e_cycletab[source[i]&0x3f]>2) + else if(dops[i].itype==C2OP&>e_cycletab[source[i]&0x3f]>2) { // this should really be removed since the real stalls have been implemented, // but doing so causes sizeable perf regression against the older version u_int gtec = gte_cycletab[source[i] & 0x3f]; cc += HACK_ENABLED(NDHACK_NO_STALLS) ? gtec/2 : 2; } - else if(i>1&&itype[i]==STORE&&itype[i-1]==STORE&&itype[i-2]==STORE&&!bt[i]) + else if(i>1&&dops[i].itype==STORE&&dops[i-1].itype==STORE&&dops[i-2].itype==STORE&&!dops[i].bt) { cc+=4; } - else if(itype[i]==C2LS) + else if(dops[i].itype==C2LS) { // same as with C2OP cc += HACK_ENABLED(NDHACK_NO_STALLS) ? 4 : 2; @@ -8245,7 +8271,7 @@ int new_recompile_block(u_int addr) cc++; } - if(!is_ds[i]) { + if(!dops[i].is_ds) { regs[i].dirty=current.dirty; regs[i].isconst=current.isconst; memcpy(constmap[i],current_constmap,sizeof(constmap[i])); @@ -8268,7 +8294,7 @@ int new_recompile_block(u_int addr) for (i=slen-1;i>=0;i--) { int hr; - if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP) + if(dops[i].is_jump) { if(ba[i]=(start+slen*4)) { @@ -8289,7 +8315,7 @@ int new_recompile_block(u_int addr) } } // Conditional branch may need registers for following instructions - if (!is_ujump(i)) + if (!dops[i].is_ujump) { if(i0&&!bt[i]&&((regs[i].wasdirty>>hr)&1)) { + 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(rt1[i-1]==(regmap_pre[i][hr]&63)) nr|=1<0&&!((unneeded_reg[i]>>regs[i].regmap_entry[hr])&1))) { - if(rt1[i-1]==(regs[i].regmap_entry[hr]&63)) nr|=1<>hr)&1)) { if(regs[i].regmap_entry[hr]!=CCREG) regs[i].regmap_entry[hr]=-1; - if((regs[i].regmap[hr]&63)!=rs1[i] && (regs[i].regmap[hr]&63)!=rs2[i] && - (regs[i].regmap[hr]&63)!=rt1[i] && (regs[i].regmap[hr]&63)!=rt2[i] && - (regs[i].regmap[hr]&63)!=PTEMP && (regs[i].regmap[hr]&63)!=CCREG) + if(dops[i].is_jump) { - if (!is_ujump(i)) - { - if(likely[i]) { - regs[i].regmap[hr]=-1; - regs[i].isconst&=~(1<0) { - int map=-1,temp=-1; - if(itype[i]==STORE || itype[i]==STORELR || - (opcode[i]&0x3b)==0x39 || (opcode[i]&0x3b)==0x3a) { // SWC1/SDC1 || SWC2/SDC2 - map=INVCP; - } - if(itype[i]==LOADLR || itype[i]==STORELR || - itype[i]==C1LS || itype[i]==C2LS) - temp=FTEMP; - if((regs[i].regmap[hr]&63)!=rt1[i] && (regs[i].regmap[hr]&63)!=rt2[i] && - regs[i].regmap[hr]!=rs1[i] && regs[i].regmap[hr]!=rs2[i] && - (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=map && - (itype[i]!=SPAN||regs[i].regmap[hr]!=CCREG)) + int map1 = -1, map2 = -1, temp=-1; + if (dops[i].is_load || dops[i].is_store) + map1 = ROREG; + if (dops[i].is_store) + 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 && + 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 && + //(dops[i].itype!=SPAN||regs[i].regmap[hr]!=CCREG) + regs[i].regmap[hr] != CCREG) { - if(i0) if(regmap_pre[i+1][hr]!=regs[i].regmap[hr]) @@ -8493,18 +8506,18 @@ int new_recompile_block(u_int addr) clear_all_regs(f_regmap); for(i=0;i=start && ba[i]<(start+i*4)) - if(itype[i+1]==NOP||itype[i+1]==MOV||itype[i+1]==ALU - ||itype[i+1]==SHIFTIMM||itype[i+1]==IMM16||itype[i+1]==LOAD - ||itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS - ||itype[i+1]==SHIFT||itype[i+1]==COP1 - ||itype[i+1]==COP2||itype[i+1]==C2LS||itype[i+1]==C2OP) + if(dops[i+1].itype==NOP||dops[i+1].itype==MOV||dops[i+1].itype==ALU + ||dops[i+1].itype==SHIFTIMM||dops[i+1].itype==IMM16||dops[i+1].itype==LOAD + ||dops[i+1].itype==STORE||dops[i+1].itype==STORELR||dops[i+1].itype==C1LS + ||dops[i+1].itype==SHIFT||dops[i+1].itype==COP1 + ||dops[i+1].itype==COP2||dops[i+1].itype==C2LS||dops[i+1].itype==C2OP) { int t=(ba[i]-start)>>2; - if(t>0&&(itype[t-1]!=UJUMP&&itype[t-1]!=RJUMP&&itype[t-1]!=CJUMP&&itype[t-1]!=SJUMP)) // loop_preload can't handle jumps into delay slots - if(t<2||(itype[t-2]!=UJUMP&&itype[t-2]!=RJUMP)||rt1[t-2]!=31) // call/ret assumes no registers allocated + if(t > 0 && !dops[t-1].is_jump) // loop_preload can't handle jumps into delay slots + if(t<2||(dops[t-2].itype!=UJUMP&&dops[t-2].itype!=RJUMP)||dops[t-2].rt1!=31) // call/ret assumes no registers allocated for(hr=0;hr=0) { @@ -8531,7 +8544,7 @@ int new_recompile_block(u_int addr) f_regmap[hr]=branch_regs[i].regmap[hr]; } } - if(ooo[i]) { + if(dops[i].ooo) { if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i+1]) f_regmap[hr]=branch_regs[i].regmap[hr]; }else{ @@ -8576,12 +8589,12 @@ int new_recompile_block(u_int addr) //printf("no-match due to different register\n"); break; } - if(itype[k-2]==UJUMP||itype[k-2]==RJUMP||itype[k-2]==CJUMP||itype[k-2]==SJUMP) { + if (dops[k-2].is_jump) { //printf("no-match due to branch\n"); break; } // call/ret fast path assumes no registers allocated - if(k>2&&(itype[k-3]==UJUMP||itype[k-3]==RJUMP)&&rt1[k-3]==31) { + if(k>2&&(dops[k-3].itype==UJUMP||dops[k-3].itype==RJUMP)&&dops[k-3].rt1==31) { break; } assert(r < 64); @@ -8625,7 +8638,7 @@ int new_recompile_block(u_int addr) branch_regs[i].dirty|=(1<=0) + if(dops[i+1].rs1) { + if((hr=get_reg(regs[i+1].regmap,dops[i+1].rs1))>=0) { if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0) { @@ -8807,8 +8820,8 @@ int new_recompile_block(u_int addr) } } } - if(rs2[i+1]) { - if((hr=get_reg(regs[i+1].regmap,rs2[i+1]))>=0) + if(dops[i+1].rs2) { + if((hr=get_reg(regs[i+1].regmap,dops[i+1].rs2))>=0) { if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0) { @@ -8824,14 +8837,14 @@ int new_recompile_block(u_int addr) } } // Preload target address for load instruction (non-constant) - if(itype[i+1]==LOAD&&rs1[i+1]&&get_reg(regs[i+1].regmap,rs1[i+1])<0) { - if((hr=get_reg(regs[i+1].regmap,rt1[i+1]))>=0) + if(dops[i+1].itype==LOAD&&dops[i+1].rs1&&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) { - regs[i].regmap[hr]=rs1[i+1]; - regmap_pre[i+1][hr]=rs1[i+1]; - regs[i+1].regmap_entry[hr]=rs1[i+1]; + regs[i].regmap[hr]=dops[i+1].rs1; + regmap_pre[i+1][hr]=dops[i+1].rs1; + regs[i+1].regmap_entry[hr]=dops[i+1].rs1; regs[i].isconst&=~(1<=0) + if(dops[i+1].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) { - regs[i].regmap[hr]=rs1[i+1]; - regmap_pre[i+1][hr]=rs1[i+1]; - regs[i+1].regmap_entry[hr]=rs1[i+1]; + regs[i].regmap[hr]=dops[i+1].rs1; + regmap_pre[i+1][hr]=dops[i+1].rs1; + regs[i+1].regmap_entry[hr]=dops[i+1].rs1; regs[i].isconst&=~(1<=0); if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0) { - regs[i].regmap[hr]=rs1[i+1]; - regmap_pre[i+1][hr]=rs1[i+1]; - regs[i+1].regmap_entry[hr]=rs1[i+1]; + regs[i].regmap[hr]=dops[i+1].rs1; + regmap_pre[i+1][hr]=dops[i+1].rs1; + regs[i+1].regmap_entry[hr]=dops[i+1].rs1; regs[i].isconst&=~(1<=0); if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0) { - regs[i].regmap[hr]=rs1[i+1]; - regmap_pre[i+1][hr]=rs1[i+1]; - regs[i+1].regmap_entry[hr]=rs1[i+1]; + regs[i].regmap[hr]=dops[i+1].rs1; + regmap_pre[i+1][hr]=dops[i+1].rs1; + regs[i+1].regmap_entry[hr]=dops[i+1].rs1; regs[i].isconst&=~(1<=0&®s[i].regmap[hr]<0) { - int rs=get_reg(regs[i+1].regmap,rs1[i+1]); + int rs=get_reg(regs[i+1].regmap,dops[i+1].rs1); if(rs>=0&&((regs[i+1].wasconst>>rs)&1)) { regs[i].regmap[hr]=AGEN1+((i+1)&1); regmap_pre[i+1][hr]=AGEN1+((i+1)&1); @@ -8945,19 +8958,19 @@ int new_recompile_block(u_int addr) /* Pass 7 - Identify 32-bit registers */ for (i=slen-1;i>=0;i--) { - if(itype[i]==CJUMP||itype[i]==SJUMP) + if(dops[i].itype==CJUMP||dops[i].itype==SJUMP) { // Conditional branch if((source[i]>>16)!=0x1000&&i\n"); - drc_dbg_emit_do_cmp(i); + drc_dbg_emit_do_cmp(i, ccadj[i]); // load regs if(regs[i].regmap_entry[HOST_CCREG]==CCREG&®s[i].regmap[HOST_CCREG]!=CCREG) wb_register(CCREG,regs[i].regmap_entry,regs[i].wasdirty); - load_regs(regs[i].regmap_entry,regs[i].regmap,rs1[i],rs2[i]); + load_regs(regs[i].regmap_entry,regs[i].regmap,dops[i].rs1,dops[i].rs2); address_generation(i,®s[i],regs[i].regmap_entry); load_consts(regmap_pre[i],regs[i].regmap,i); - if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP) + if(dops[i].is_jump) { // Load the delay slot registers if necessary - if(rs1[i+1]!=rs1[i]&&rs1[i+1]!=rs2[i]&&(rs1[i+1]!=rt1[i]||rt1[i]==0)) - load_regs(regs[i].regmap_entry,regs[i].regmap,rs1[i+1],rs1[i+1]); - if(rs2[i+1]!=rs1[i+1]&&rs2[i+1]!=rs1[i]&&rs2[i+1]!=rs2[i]&&(rs2[i+1]!=rt1[i]||rt1[i]==0)) - load_regs(regs[i].regmap_entry,regs[i].regmap,rs2[i+1],rs2[i+1]); - if(itype[i+1]==STORE||itype[i+1]==STORELR||(opcode[i+1]&0x3b)==0x39||(opcode[i+1]&0x3b)==0x3a) + if(dops[i+1].rs1!=dops[i].rs1&&dops[i+1].rs1!=dops[i].rs2&&(dops[i+1].rs1!=dops[i].rt1||dops[i].rt1==0)) + load_regs(regs[i].regmap_entry,regs[i].regmap,dops[i+1].rs1,dops[i+1].rs1); + if(dops[i+1].rs2!=dops[i+1].rs1&&dops[i+1].rs2!=dops[i].rs1&&dops[i+1].rs2!=dops[i].rs2&&(dops[i+1].rs2!=dops[i].rt1||dops[i].rt1==0)) + load_regs(regs[i].regmap_entry,regs[i].regmap,dops[i+1].rs2,dops[i+1].rs2); + if (ram_offset && (dops[i+1].is_load || dops[i+1].is_store)) + load_regs(regs[i].regmap_entry,regs[i].regmap,ROREG,ROREG); + if (dops[i+1].is_store) load_regs(regs[i].regmap_entry,regs[i].regmap,INVCP,INVCP); } else if(i+1 0); + if (slen > 0 && dops[slen-1].itype == INTCALL) { + // no ending needed for this block since INTCALL never returns + } // If the block did not end with an unconditional branch, // add a jump to the next instruction. - if(i>1) { - if(!is_ujump(i-2)&&itype[i-1]!=SPAN) { - assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP); + else if (i > 1) { + if (!dops[i-2].is_ujump && dops[i-1].itype != SPAN) { + assert(!dops[i-1].is_jump); assert(i==slen); - if(itype[i-2]!=CJUMP&&itype[i-2]!=SJUMP) { + if(dops[i-2].itype!=CJUMP&&dops[i-2].itype!=SJUMP) { store_regs_bt(regs[i-1].regmap,regs[i-1].dirty,start+i*4); if(regs[i-1].regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG); - emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i-1]+1),HOST_CCREG); + emit_addimm(HOST_CCREG, ccadj[i-1] + CLOCK_ADJUST(1), HOST_CCREG); } - else if(!likely[i-2]) + else { store_regs_bt(branch_regs[i-2].regmap,branch_regs[i-2].dirty,start+i*4); assert(branch_regs[i-2].regmap[HOST_CCREG]==CCREG); } - else - { - store_regs_bt(regs[i-2].regmap,regs[i-2].dirty,start+i*4); - assert(regs[i-2].regmap[HOST_CCREG]==CCREG); - } add_to_linker(out,start+i*4,0); emit_jmp(0); } @@ -9275,11 +9241,11 @@ int new_recompile_block(u_int addr) else { assert(i>0); - assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP); + assert(!dops[i-1].is_jump); store_regs_bt(regs[i-1].regmap,regs[i-1].dirty,start+i*4); if(regs[i-1].regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG); - emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i-1]+1),HOST_CCREG); + emit_addimm(HOST_CCREG, ccadj[i-1] + CLOCK_ADJUST(1), HOST_CCREG); add_to_linker(out,start+i*4,0); emit_jmp(0); } @@ -9328,7 +9294,7 @@ int new_recompile_block(u_int addr) emit_extjump(link_addr[i].addr, link_addr[i].target); if (addr) { set_jump_target(link_addr[i].addr, addr); - add_link(link_addr[i].target,stub); + add_jump_out(link_addr[i].target,stub); } else set_jump_target(link_addr[i].addr, stub); @@ -9346,11 +9312,20 @@ int new_recompile_block(u_int addr) //#endif } } + + u_int source_len = slen*4; + if (dops[slen-1].itype == INTCALL && source_len > 4) + // no need to treat the last instruction as compiled + // as interpreter fully handles it + source_len -= 4; + + if ((u_char *)copy + source_len > (u_char *)shadow + sizeof(shadow)) + copy = shadow; + // External Branch Targets (jump_in) - if(copy+slen*4>(void *)shadow+sizeof(shadow)) copy=shadow; for(i=0;itranslation_cache+((expirep>>13)<> 13) << shift); // Base offset of this block + uintptr_t base_offs_s = base_offs >> shift; inv_debug("EXP: Phase %d\n",expirep); switch((expirep>>11)&3) { case 0: // Clear jump_in and jump_dirty - ll_remove_matching_addrs(jump_in+(expirep&2047),base,shift); - ll_remove_matching_addrs(jump_dirty+(expirep&2047),base,shift); - ll_remove_matching_addrs(jump_in+2048+(expirep&2047),base,shift); - ll_remove_matching_addrs(jump_dirty+2048+(expirep&2047),base,shift); + ll_remove_matching_addrs(jump_in+(expirep&2047),base_offs_s,shift); + ll_remove_matching_addrs(jump_dirty+(expirep&2047),base_offs_s,shift); + ll_remove_matching_addrs(jump_in+2048+(expirep&2047),base_offs_s,shift); + ll_remove_matching_addrs(jump_dirty+2048+(expirep&2047),base_offs_s,shift); break; case 1: // Clear pointers - ll_kill_pointers(jump_out[expirep&2047],base,shift); - ll_kill_pointers(jump_out[(expirep&2047)+2048],base,shift); + ll_kill_pointers(jump_out[expirep&2047],base_offs_s,shift); + ll_kill_pointers(jump_out[(expirep&2047)+2048],base_offs_s,shift); break; case 2: // Clear hash table for(i=0;i<32;i++) { struct ht_entry *ht_bin = &hash_table[((expirep&2047)<<5)+i]; - if (((uintptr_t)ht_bin->tcaddr[1]>>shift) == (base>>shift) || - (((uintptr_t)ht_bin->tcaddr[1]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) { + uintptr_t o1 = (u_char *)ht_bin->tcaddr[1] - ndrc->translation_cache; + uintptr_t o2 = o1 - MAX_OUTPUT_BLOCK_SIZE; + if ((o1 >> shift) == base_offs_s || (o2 >> shift) == base_offs_s) { inv_debug("EXP: Remove hash %x -> %p\n",ht_bin->vaddr[1],ht_bin->tcaddr[1]); ht_bin->vaddr[1] = -1; ht_bin->tcaddr[1] = NULL; } - if (((uintptr_t)ht_bin->tcaddr[0]>>shift) == (base>>shift) || - (((uintptr_t)ht_bin->tcaddr[0]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) { + o1 = (u_char *)ht_bin->tcaddr[0] - ndrc->translation_cache; + o2 = o1 - MAX_OUTPUT_BLOCK_SIZE; + if ((o1 >> shift) == base_offs_s || (o2 >> shift) == base_offs_s) { inv_debug("EXP: Remove hash %x -> %p\n",ht_bin->vaddr[0],ht_bin->tcaddr[0]); ht_bin->vaddr[0] = ht_bin->vaddr[1]; ht_bin->tcaddr[0] = ht_bin->tcaddr[1]; @@ -9454,12 +9432,15 @@ int new_recompile_block(u_int addr) // Clear jump_out if((expirep&2047)==0) do_clear_cache(); - ll_remove_matching_addrs(jump_out+(expirep&2047),base,shift); - ll_remove_matching_addrs(jump_out+2048+(expirep&2047),base,shift); + ll_remove_matching_addrs(jump_out+(expirep&2047),base_offs_s,shift); + ll_remove_matching_addrs(jump_out+2048+(expirep&2047),base_offs_s,shift); break; } expirep=(expirep+1)&65535; } +#ifdef ASSEM_PRINT + fflush(stdout); +#endif return 0; }