X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fnew_dynarec.c;h=1c0ab56e1b9077b50711fb7f27636cfb4e388c6f;hb=1a5fd79401ac52789fad34c6b852b947200a6334;hp=84370089f7625e2d699a912c6fd0f9962843f0ce;hpb=19414bd6e0f0febe9c04bc3c876f400945861a95;p=pcsx_rearmed.git diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c index 84370089..1c0ab56e 100644 --- a/libpcsxcore/new_dynarec/new_dynarec.c +++ b/libpcsxcore/new_dynarec/new_dynarec.c @@ -21,6 +21,7 @@ #include #include //include for uint64_t #include +#include #include #include "emu_if.h" //emulator interface @@ -52,6 +53,16 @@ static void __clear_cache(void *start, void *end) { sys_dcache_flush(start, len); sys_icache_invalidate(start, len); } +#elif defined(_3DS) +#include "3ds_utils.h" +#define __clear_cache(start,end) svcFlushProcessDataCache(0xFFFF8001, start, (u32)(end)-(u32)(start)) +#elif defined(VITA) +#define __clear_cache vita_clear_cache +static void __clear_cache(void *start, void *end) { + size_t len = (char *)end - (char *)start; + int block = sceKernelFindMemBlockByAddr(start,len); + sceKernelSyncVMDomain(block, start, len); +} #endif #define MAXBLOCK 4096 @@ -73,17 +84,17 @@ struct regstat u_int waswritten; // MIPS regs that were used as store base before }; +// note: asm depends on this layout struct ll_entry { u_int vaddr; - u_int reg32; + u_int reg_sv_flags; void *addr; struct ll_entry *next; }; u_int start; u_int *source; - u_int pagelimit; char insn[MAXBLOCK][10]; u_char itype[MAXBLOCK]; u_char opcode[MAXBLOCK]; @@ -130,7 +141,7 @@ struct ll_entry int ccadj[MAXBLOCK]; int slen; u_int instr_addr[MAXBLOCK]; - u_int link_addr[MAXBLOCK][3]; + static u_int link_addr[MAXBLOCK][3]; int linkcount; u_int stubs[MAXBLOCK*3][8]; int stubcount; @@ -139,7 +150,7 @@ struct ll_entry int is_delayslot; int cop1_usable; u_char *out; - struct ll_entry *jump_in[4096]; + struct ll_entry *jump_in[4096] __attribute__((aligned(16))); struct ll_entry *jump_out[4096]; struct ll_entry *jump_dirty[4096]; u_int hash_table[65536][4] __attribute__((aligned(16))); @@ -192,7 +203,7 @@ struct ll_entry #define STORE 2 // Store #define LOADLR 3 // Unaligned load #define STORELR 4 // Unaligned store -#define MOV 5 // Move +#define MOV 5 // Move #define ALU 6 // Arithmetic/logic #define MULTDIV 7 // Multiply/divide #define SHIFT 8 // Shift by register @@ -310,18 +321,18 @@ static void tlb_hacks() { u_int addr; int n; - switch (ROM_HEADER->Country_code&0xFF) + switch (ROM_HEADER->Country_code&0xFF) { case 0x45: // U addr=0x34b30; - break; - case 0x4A: // J - addr=0x34b70; - break; - case 0x50: // E + break; + case 0x4A: // J + addr=0x34b70; + break; + case 0x50: // E addr=0x329f0; - break; - default: + break; + default: // Unknown country code addr=0; break; @@ -394,7 +405,7 @@ void *get_addr(u_int vaddr) //printf("TRACE: count=%d next=%d (get_addr %x,page %d)\n",Count,next_interupt,vaddr,page); head=jump_in[page]; while(head!=NULL) { - if(head->vaddr==vaddr&&head->reg32==0) { + if(head->vaddr==vaddr) { //printf("TRACE: count=%d next=%d (get_addr match %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr); int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF]; ht_bin[3]=ht_bin[1]; @@ -407,7 +418,7 @@ void *get_addr(u_int vaddr) } head=jump_dirty[vpage]; while(head!=NULL) { - if(head->vaddr==vaddr&&head->reg32==0) { + if(head->vaddr==vaddr) { //printf("TRACE: count=%d next=%d (get_addr match dirty %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr); // Don't restore blocks which are about to expire from the cache if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) @@ -466,94 +477,6 @@ void *get_addr_ht(u_int vaddr) return get_addr(vaddr); } -void *get_addr_32(u_int vaddr,u_int flags) -{ -#ifdef FORCE32 - return get_addr(vaddr); -#else - //printf("TRACE: count=%d next=%d (get_addr_32 %x,flags %x)\n",Count,next_interupt,vaddr,flags); - int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF]; - if(ht_bin[0]==vaddr) return (void *)ht_bin[1]; - if(ht_bin[2]==vaddr) return (void *)ht_bin[3]; - u_int page=get_page(vaddr); - u_int vpage=get_vpage(vaddr); - struct ll_entry *head; - head=jump_in[page]; - while(head!=NULL) { - if(head->vaddr==vaddr&&(head->reg32&flags)==0) { - //printf("TRACE: count=%d next=%d (get_addr_32 match %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr); - if(head->reg32==0) { - int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF]; - if(ht_bin[0]==-1) { - ht_bin[1]=(int)head->addr; - ht_bin[0]=vaddr; - }else if(ht_bin[2]==-1) { - ht_bin[3]=(int)head->addr; - ht_bin[2]=vaddr; - } - //ht_bin[3]=ht_bin[1]; - //ht_bin[2]=ht_bin[0]; - //ht_bin[1]=(int)head->addr; - //ht_bin[0]=vaddr; - } - return head->addr; - } - head=head->next; - } - head=jump_dirty[vpage]; - while(head!=NULL) { - if(head->vaddr==vaddr&&(head->reg32&flags)==0) { - //printf("TRACE: count=%d next=%d (get_addr_32 match dirty %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr); - // Don't restore blocks which are about to expire from the cache - if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) - if(verify_dirty(head->addr)) { - //printf("restore candidate: %x (%d) d=%d\n",vaddr,page,invalid_code[vaddr>>12]); - invalid_code[vaddr>>12]=0; - inv_code_start=inv_code_end=~0; - memory_map[vaddr>>12]|=0x40000000; - if(vpage<2048) { -#ifndef DISABLE_TLB - if(tlb_LUT_r[vaddr>>12]) { - invalid_code[tlb_LUT_r[vaddr>>12]>>12]=0; - memory_map[tlb_LUT_r[vaddr>>12]>>12]|=0x40000000; - } -#endif - restore_candidate[vpage>>3]|=1<<(vpage&7); - } - else restore_candidate[page>>3]|=1<<(page&7); - if(head->reg32==0) { - int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF]; - if(ht_bin[0]==-1) { - ht_bin[1]=(int)head->addr; - ht_bin[0]=vaddr; - }else if(ht_bin[2]==-1) { - ht_bin[3]=(int)head->addr; - ht_bin[2]=vaddr; - } - //ht_bin[3]=ht_bin[1]; - //ht_bin[2]=ht_bin[0]; - //ht_bin[1]=(int)head->addr; - //ht_bin[0]=vaddr; - } - return head->addr; - } - } - head=head->next; - } - //printf("TRACE: count=%d next=%d (get_addr_32 no-match %x,flags %x)\n",Count,next_interupt,vaddr,flags); - int r=new_recompile_block(vaddr); - if(r==0) return get_addr(vaddr); - // Execute in unmapped page, generate pagefault execption - Status|=2; - Cause=(vaddr<<31)|0x8; - EPC=(vaddr&1)?vaddr-5:vaddr; - BadVAddr=(vaddr&~1); - Context=(Context&0xFF80000F)|((BadVAddr>>9)&0x007FFFF0); - EntryHi=BadVAddr&0xFFFFE000; - return get_addr_ht(0x80000000); -#endif -} - void clear_all_regs(signed char regmap[]) { int hr; @@ -610,7 +533,7 @@ static void flush_dirty_uppers(struct regstat *cur) for (hr=0;hrdirty>>hr)&1) { reg=cur->regmap[hr]; - if(reg>=64) + if(reg>=64) if((cur->is32>>(reg&63))&1) cur->regmap[hr]=-1; } } @@ -767,7 +690,7 @@ int needed_again(int r, int i) int j; int b=-1; int rn=10; - + if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)) { if(ba[i-1]start+slen*4-4) @@ -870,7 +793,7 @@ int loop_reg(int i, int r, int hr) void alloc_all(struct regstat *cur,int i) { int hr; - + for(hr=0;hrregmap[hr]&63)!=rs1[i])&&((cur->regmap[hr]&63)!=rs2[i])&& @@ -911,7 +834,7 @@ void mult64(uint64_t m1,uint64_t m2) unsigned long long int result1, result2, result3, result4; unsigned long long int temp1, temp2, temp3, temp4; int sign = 0; - + if (m1 < 0) { op2 = -m1; @@ -924,22 +847,22 @@ void mult64(uint64_t m1,uint64_t m2) sign = 1 - sign; } else op4 = m2; - + op1 = op2 & 0xFFFFFFFF; op2 = (op2 >> 32) & 0xFFFFFFFF; op3 = op4 & 0xFFFFFFFF; op4 = (op4 >> 32) & 0xFFFFFFFF; - + temp1 = op1 * op3; temp2 = (temp1 >> 32) + op1 * op4; temp3 = op2 * op3; temp4 = (temp3 >> 32) + op2 * op4; - + result1 = temp1 & 0xFFFFFFFF; result2 = temp2 + (temp3 & 0xFFFFFFFF); result3 = (result2 >> 32) + temp4; result4 = (result3 >> 32); - + lo = result1 | (result2 << 32); hi = (result3 & 0xFFFFFFFF) | (result4 << 32); if (sign) @@ -955,25 +878,25 @@ void multu64(uint64_t m1,uint64_t m2) unsigned long long int op1, op2, op3, op4; unsigned long long int result1, result2, result3, result4; unsigned long long int temp1, temp2, temp3, temp4; - + op1 = m1 & 0xFFFFFFFF; op2 = (m1 >> 32) & 0xFFFFFFFF; op3 = m2 & 0xFFFFFFFF; op4 = (m2 >> 32) & 0xFFFFFFFF; - + temp1 = op1 * op3; temp2 = (temp1 >> 32) + op1 * op4; temp3 = op2 * op3; temp4 = (temp3 >> 32) + op2 * op4; - + result1 = temp1 & 0xFFFFFFFF; result2 = temp2 + (temp3 & 0xFFFFFFFF); result3 = (result2 >> 32) + temp4; result4 = (result3 >> 32); - + lo = result1 | (result2 << 32); hi = (result3 & 0xFFFFFFFF) | (result4 << 32); - + //printf("TRACE: dmultu %8x%8x %8x%8x\n",(int)reg[HIREG],(int)(reg[HIREG]>>32) // ,(int)reg[LOREG],(int)(reg[LOREG]>>32)); } @@ -1019,19 +942,16 @@ void ll_add(struct ll_entry **head,int vaddr,void *addr) new_entry=malloc(sizeof(struct ll_entry)); assert(new_entry!=NULL); new_entry->vaddr=vaddr; - new_entry->reg32=0; + new_entry->reg_sv_flags=0; new_entry->addr=addr; new_entry->next=*head; *head=new_entry; } -// Add virtual address mapping for 32-bit compiled block -void ll_add_32(struct ll_entry **head,int vaddr,u_int reg32,void *addr) +void ll_add_flags(struct ll_entry **head,int vaddr,u_int reg_sv_flags,void *addr) { ll_add(head,vaddr,addr); -#ifndef FORCE32 - (*head)->reg32=reg32; -#endif + (*head)->reg_sv_flags=reg_sv_flags; } // Check if an address is already compiled @@ -1051,7 +971,7 @@ void *check_addr(u_int vaddr) struct ll_entry *head; head=jump_in[page]; while(head!=NULL) { - if(head->vaddr==vaddr&&head->reg32==0) { + if(head->vaddr==vaddr) { if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) { // Update existing entry with current address if(ht_bin[0]==vaddr) { @@ -1098,7 +1018,7 @@ void ll_remove_matching_addrs(struct ll_entry **head,int addr,int shift) { struct ll_entry *next; while(*head) { - if(((u_int)((*head)->addr)>>shift)==(addr>>shift) || + if(((u_int)((*head)->addr)>>shift)==(addr>>shift) || ((u_int)((*head)->addr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift)) { inv_debug("EXP: Remove pointer to %x (%x)\n",(int)(*head)->addr,(*head)->vaddr); @@ -1194,7 +1114,7 @@ static void invalidate_block_range(u_int block, u_int first, u_int last) #ifdef __arm__ do_clear_cache(); #endif - + // Don't trap writes invalid_code[block]=1; #ifndef DISABLE_TLB @@ -1400,15 +1320,13 @@ void clean_blocks(u_int page) inv_debug("INV: Restored %x (%x/%x)\n",head->vaddr, (int)head->addr, (int)clean_addr); //printf("page=%x, addr=%x\n",page,head->vaddr); //assert(head->vaddr>>12==(page|0x80000)); - ll_add_32(jump_in+ppage,head->vaddr,head->reg32,clean_addr); + ll_add_flags(jump_in+ppage,head->vaddr,head->reg_sv_flags,clean_addr); int *ht_bin=hash_table[((head->vaddr>>16)^head->vaddr)&0xFFFF]; - if(!head->reg32) { - if(ht_bin[0]==head->vaddr) { - ht_bin[1]=(int)clean_addr; // Replace existing entry - } - if(ht_bin[2]==head->vaddr) { - ht_bin[3]=(int)clean_addr; // Replace existing entry - } + if(ht_bin[0]==head->vaddr) { + ht_bin[1]=(int)clean_addr; // Replace existing entry + } + if(ht_bin[2]==head->vaddr) { + ht_bin[3]=(int)clean_addr; // Replace existing entry } } } @@ -3475,7 +3393,7 @@ void storelr_assemble(int i,struct regstat *i_regs) if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG); gen_tlb_addr_w(temp,map); #else - if((u_int)rdram!=0x80000000) + if((u_int)rdram!=0x80000000) emit_addimm_no_flags((u_int)rdram-(u_int)0x80000000,temp); #endif }else{ // using tlb @@ -4248,7 +4166,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[]) int mgr=MGEN1+(i&1); if(itype[i]==LOAD) { ra=get_reg(i_regs->regmap,rt1[i]); - if(ra<0) ra=get_reg(i_regs->regmap,-1); + if(ra<0) ra=get_reg(i_regs->regmap,-1); assert(ra>=0); } if(itype[i]==LOADLR) { @@ -4846,7 +4764,7 @@ int match_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr) { return 0; } - else + else if((i_dirty>>hr)&1) { if(i_regmap[hr]>s1l)&(branch_regs[i].is32>>rs1[i])&1) emit_loadreg(rs1[i],s1l); - } + } else { if((branch_regs[i].dirty>>s1l)&(branch_regs[i].is32>>rs2[i])&1) emit_loadreg(rs2[i],s1l); @@ -5283,7 +5201,7 @@ void do_ccstub(int n) load_all_regs(branch_regs[i].regmap); } emit_jmp(stubs[n][2]); // return address - + /* This works but uses a lot of memory... emit_readword((int)&last_count,ECX); emit_add(HOST_CCREG,ECX,EAX); @@ -5317,7 +5235,7 @@ add_to_linker(int addr,int target,int ext) { link_addr[linkcount][0]=addr; link_addr[linkcount][1]=target; - link_addr[linkcount][2]=ext; + link_addr[linkcount][2]=ext; linkcount++; } @@ -5343,7 +5261,7 @@ static void ujump_assemble_write_ra(int i) #endif { #ifdef REG_PREFETCH - if(temp>=0) + if(temp>=0) { if(i_regmap[temp]!=PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp); } @@ -5364,10 +5282,10 @@ void ujump_assemble(int i,struct regstat *i_regs) 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(rt1[i]==31&&temp>=0) { int return_address=start+i*4+8; - if(get_reg(branch_regs[i].regmap,31)>0) + if(get_reg(branch_regs[i].regmap,31)>0) if(i_regmap[temp]==PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp); } #endif @@ -5418,7 +5336,7 @@ static void rjump_assemble_write_ra(int i) assert(rt>=0); return_address=start+i*4+8; #ifdef REG_PREFETCH - if(temp>=0) + if(temp>=0) { if(i_regmap[temp]!=PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp); } @@ -5447,7 +5365,7 @@ 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(rt1[i]==31) { if((temp=get_reg(branch_regs[i].regmap,PTEMP))>=0) { int return_address=start+i*4+8; @@ -5581,7 +5499,7 @@ void cjump_assemble(int i,struct regstat *i_regs) #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK if(i>(ba[i]-start)>>2) invert=1; #endif - + if(ooo[i]) { s1l=get_reg(branch_regs[i].regmap,rs1[i]); s1h=get_reg(branch_regs[i].regmap,rs1[i]|64); @@ -5636,7 +5554,7 @@ void cjump_assemble(int i,struct regstat *i_regs) load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG); cc=get_reg(branch_regs[i].regmap,CCREG); assert(cc==HOST_CCREG); - if(unconditional) + if(unconditional) store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]); //do_cc(i,branch_regs[i].regmap,&adj,unconditional?ba[i]:-1,unconditional); //assem_debug("cycle count (adj)\n"); @@ -5708,7 +5626,7 @@ void cjump_assemble(int i,struct regstat *i_regs) emit_jne(0); } } // if(!only32) - + //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 @@ -5838,7 +5756,7 @@ void cjump_assemble(int i,struct regstat *i_regs) emit_jne(1); } } // if(!only32) - + //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 @@ -6021,7 +5939,7 @@ void sjump_assemble(int i,struct regstat *i_regs) } cc=get_reg(branch_regs[i].regmap,CCREG); assert(cc==HOST_CCREG); - if(unconditional) + if(unconditional) store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]); //do_cc(i,branch_regs[i].regmap,&adj,unconditional?ba[i]:-1,unconditional); assem_debug("cycle count (adj)\n"); @@ -6108,7 +6026,7 @@ void sjump_assemble(int i,struct regstat *i_regs) } } } // if(!only32) - + if(invert) { #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK if(match&&(!internal||!is_ds[(ba[i]-start)>>2])) { @@ -6356,7 +6274,7 @@ void fjump_assemble(int i,struct regstat *i_regs) { } } // if(!only32) - + if(invert) { if(adj) emit_addimm(cc,-CLOCK_ADJUST(adj),cc); #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK @@ -6864,14 +6782,14 @@ void unneeded_registers(int istart,int iend,int r) { // If subroutine call, flag return address as a possible branch target if(rt1[i]==31 && i=(start+slen*4)) { // Branch out of this block, flush all regs u=1; uu=1; gte_u=gte_u_unknown; - /* Hexagon hack + /* Hexagon hack if(itype[i]==UJUMP&&rt1[i]==31) { uu=u=0x300C00F; // Discard at, v0-v1, t6-t9 @@ -7134,7 +7052,7 @@ static void provisional_32bit() int i,j; uint64_t is32=1; uint64_t lastbranch=1; - + for(i=0;i0) { @@ -7171,13 +7089,13 @@ static void provisional_32bit() uint64_t temp_is32=is32; for(j=i-1;j>=0;j--) { - if(ba[j]==start+i*4) + if(ba[j]==start+i*4) //temp_is32&=branch_regs[j].is32; temp_is32&=p32[j]; } for(j=i;j=0;i--) { int hr; @@ -7462,7 +7380,7 @@ static void provisional_r32() } //requires_32bit[i]=r32; pr32[i]=r32; - + // Dirty registers which are 32-bit, require 32-bit input // as they will be written as 32-bit values for(hr=0;hristart) { - if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=FJUMP) + if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=FJUMP) { // Don't store a register immediately after writing it, // may prevent dual-issue. @@ -8005,6 +7923,28 @@ void disassemble_inst(int i) static void disassemble_inst(int i) {} #endif // DISASM +#define DRC_TEST_VAL 0x74657374 + +static int new_dynarec_test(void) +{ + int (*testfunc)(void) = (void *)out; + int ret; + emit_movimm(DRC_TEST_VAL,0); // test + emit_jmpreg(14); + literal_pool(0); +#ifdef __arm__ + __clear_cache((void *)testfunc, out); +#endif + SysPrintf("testing if we can run recompiled code..\n"); + ret = testfunc(); + if (ret == DRC_TEST_VAL) + SysPrintf("test passed.\n"); + else + SysPrintf("test failed: %08x\n", ret); + out=(u_char *)BASE_ADDR; + return ret == DRC_TEST_VAL; +} + // clear the state completely, instead of just marking // things invalid like invalidate_all_pages() does void new_dynarec_clear_full() @@ -8039,17 +7979,30 @@ void new_dynarec_clear_full() void new_dynarec_init() { - printf("Init new dynarec\n"); + SysPrintf("Init new dynarec\n"); out=(u_char *)BASE_ADDR; -#if BASE_ADDR_FIXED +#if defined(VITA) + if (mmap (out, 1<= 0x80000000 && addr < 0x80000000+RAM_SIZE) { + *limit = (addr & 0x80600000) + 0x00200000; + return (u_int *)((u_int)rdram + (addr&0x1fffff)); + } +} + +static u_int scan_for_ret(u_int addr) +{ + u_int limit = 0; + u_int *mem; + + mem = get_source_start(addr, &limit); + if (mem == NULL) + return addr; + + if (limit > addr + 0x1000) + limit = addr + 0x1000; + for (; addr < limit; addr += 4, mem++) { + if (*mem == 0x03e00008) // jr $ra + return addr + 8; + } +} + +struct savestate_block { + uint32_t addr; + uint32_t regflags; +}; + +static int addr_cmp(const void *p1_, const void *p2_) +{ + const struct savestate_block *p1 = p1_, *p2 = p2_; + return p1->addr - p2->addr; +} + +int new_dynarec_save_blocks(void *save, int size) +{ + struct savestate_block *blocks = save; + int maxcount = size / sizeof(blocks[0]); + struct savestate_block tmp_blocks[1024]; + struct ll_entry *head; + int p, s, d, o, bcnt; + u_int addr; + + o = 0; + for (p = 0; p < sizeof(jump_in) / sizeof(jump_in[0]); p++) { + bcnt = 0; + for (head = jump_in[p]; head != NULL; head = head->next) { + tmp_blocks[bcnt].addr = head->vaddr; + tmp_blocks[bcnt].regflags = head->reg_sv_flags; + bcnt++; + } + if (bcnt < 1) + continue; + qsort(tmp_blocks, bcnt, sizeof(tmp_blocks[0]), addr_cmp); + + addr = tmp_blocks[0].addr; + for (s = d = 0; s < bcnt; s++) { + if (tmp_blocks[s].addr < addr) + continue; + if (d == 0 || tmp_blocks[d-1].addr != tmp_blocks[s].addr) + tmp_blocks[d++] = tmp_blocks[s]; + addr = scan_for_ret(tmp_blocks[s].addr); + } + + if (o + d > maxcount) + d = maxcount - o; + memcpy(&blocks[o], tmp_blocks, d * sizeof(blocks[0])); + o += d; + } + + return o * sizeof(blocks[0]); +} + +void new_dynarec_load_blocks(const void *save, int size) +{ + const struct savestate_block *blocks = save; + int count = size / sizeof(blocks[0]); + u_int regs_save[32]; + uint32_t f; + int i, b; + + get_addr(psxRegs.pc); + + // change GPRs for speculation to at least partially work.. + memcpy(regs_save, &psxRegs.GPR, sizeof(regs_save)); + for (i = 1; i < 32; i++) + psxRegs.GPR.r[i] = 0x80000000; + + for (b = 0; b < count; b++) { + for (f = blocks[b].regflags, i = 0; f; f >>= 1, i++) { + if (f & 1) + psxRegs.GPR.r[i] = 0x1f800000; + } + + get_addr(blocks[b].addr); + + for (f = blocks[b].regflags, i = 0; f; f >>= 1, i++) { + if (f & 1) + psxRegs.GPR.r[i] = 0x80000000; + } + } + + memcpy(&psxRegs.GPR, regs_save, sizeof(regs_save)); +} + +int new_recompile_block(int addr) +{ + u_int pagelimit = 0; + u_int state_rflags = 0; + int i; + assem_debug("NOTCOMPILED: addr = %x -> %x\n", (int)addr, (int)out); //printf("NOTCOMPILED: addr = %x -> %x\n", (int)addr, (int)out); //printf("TRACE: count=%d next=%d (compile %x)\n",Count,next_interupt,addr); - //if(debug) + //if(debug) //printf("TRACE: count=%d next=%d (checksum %x)\n",Count,next_interupt,mchecksum()); //printf("fpu mapping=%x enabled=%x\n",(Status & 0x04000000)>>26,(Status & 0x20000000)>>29); /*if(Count>=312978186) { rlist(); }*/ //rlist(); + + // this is just for speculation + for (i = 1; i < 32; i++) { + if ((psxRegs.GPR.r[i] & 0xffff0000) == 0x1f800000) + state_rflags |= 1 << i; + } + start = (u_int)addr&~3; //assert(((u_int)addr&1)==0); new_dynarec_did_compile=1; -#ifdef PCSX if (Config.HLE && start == 0x80001000) // hlecall { // XXX: is this enough? Maybe check hleSoftCall? @@ -8163,62 +8238,13 @@ int new_recompile_block(int addr) #ifdef __arm__ __clear_cache((void *)beginning,out); #endif - ll_add(jump_in+page,start,(void *)beginning); + ll_add_flags(jump_in+page,start,state_rflags,(void *)beginning); return 0; } - else if ((u_int)addr < 0x00200000 || - (0xa0000000 <= addr && addr < 0xa0200000)) { - // used for BIOS calls mostly? - source = (u_int *)((u_int)rdram+(start&0x1fffff)); - pagelimit = (addr&0xa0000000)|0x00200000; - } - else if (!Config.HLE && ( -/* (0x9fc00000 <= addr && addr < 0x9fc80000) ||*/ - (0xbfc00000 <= addr && addr < 0xbfc80000))) { - // BIOS - source = (u_int *)((u_int)psxR+(start&0x7ffff)); - pagelimit = (addr&0xfff00000)|0x80000; - } - else -#endif -#ifdef MUPEN64 - if ((int)addr >= 0xa4000000 && (int)addr < 0xa4001000) { - source = (u_int *)((u_int)SP_DMEM+start-0xa4000000); - pagelimit = 0xa4001000; - } - else -#endif - if ((int)addr >= 0x80000000 && (int)addr < 0x80000000+RAM_SIZE) { - source = (u_int *)((u_int)rdram+start-0x80000000); - pagelimit = 0x80000000+RAM_SIZE; - } -#ifndef DISABLE_TLB - else if ((signed int)addr >= (signed int)0xC0000000) { - //printf("addr=%x mm=%x\n",(u_int)addr,(memory_map[start>>12]<<2)); - //if(tlb_LUT_r[start>>12]) - //source = (u_int *)(((int)rdram)+(tlb_LUT_r[start>>12]&0xFFFFF000)+(((int)addr)&0xFFF)-0x80000000); - if((signed int)memory_map[start>>12]>=0) { - source = (u_int *)((u_int)(start+(memory_map[start>>12]<<2))); - pagelimit=(start+4096)&0xFFFFF000; - int map=memory_map[start>>12]; - int i; - for(i=0;i<5;i++) { - //printf("start: %x next: %x\n",map,memory_map[pagelimit>>12]); - if((map&0xBFFFFFFF)==(memory_map[pagelimit>>12]&0xBFFFFFFF)) pagelimit+=4096; - } - assem_debug("pagelimit=%x\n",pagelimit); - assem_debug("mapping=%x (%x)\n",memory_map[start>>12],(memory_map[start>>12]<<2)+start); - } - else { - assem_debug("Compile at unmapped memory address: %x \n", (int)addr); - //assem_debug("start: %x next: %x\n",memory_map[start>>12],memory_map[(start+4096)>>12]); - return -1; // Caller will invoke exception handler - } - //printf("source= %x\n",(int)source); - } -#endif - else { - SysPrintf("Compile at bogus memory address: %x \n", (int)addr); + + source = get_source_start(start, &pagelimit); + if (source == NULL) { + SysPrintf("Compile at bogus memory address: %08x\n", addr); exit(1); } @@ -8233,12 +8259,12 @@ int new_recompile_block(int addr) /* Pass 9: linker */ /* Pass 10: garbage collection / free memory */ - int i,j; + int j; int done=0; unsigned int type,op,op2; //printf("addr = %x source = %x %x\n", addr,source,source[0]); - + /* Pass 1 disassembly */ for(i=0;!done;i++) { @@ -8898,7 +8924,7 @@ int new_recompile_block(int addr) /* Pass 2 - Register dependencies and branch targets */ unneeded_registers(0,slen-1,0); - + /* Pass 3 - Register allocation */ struct regstat current; // Current register allocations/status @@ -8928,7 +8954,7 @@ int new_recompile_block(int addr) unneeded_reg_upper[0]=1; current.regmap[HOST_BTREG]=BTREG; } - + for(i=0;i=0;j--) { - if(ba[j]==start+i*4) + if(ba[j]==start+i*4) temp_is32&=branch_regs[j].is32; } for(j=i;j=0;j--) { - if(ba[j]==start+i*4+4) + if(ba[j]==start+i*4+4) temp_is32&=branch_regs[j].is32; } for(j=i;j=0) + if(get_reg(current.regmap,r|64)>=0) current.regmap[get_reg(current.regmap,r|64)]=-1; } } @@ -9051,12 +9077,12 @@ int new_recompile_block(int addr) uint64_t temp_is32=current.is32; for(j=i-1;j>=0;j--) { - if(ba[j]==start+i*4+8) + if(ba[j]==start+i*4+8) temp_is32&=branch_regs[j].is32; } for(j=i;j=0) + if(get_reg(current.regmap,r|64)>=0) current.regmap[get_reg(current.regmap,r|64)]=-1; } } @@ -9164,7 +9190,7 @@ int new_recompile_block(int addr) } } else { // First instruction expects CCREG to be allocated - if(i==0&&hr==HOST_CCREG) + if(i==0&&hr==HOST_CCREG) regs[i].regmap_entry[hr]=CCREG; else regs[i].regmap_entry[hr]=-1; @@ -9499,7 +9525,7 @@ int new_recompile_block(int addr) pagespan_alloc(¤t,i); break; } - + // Drop the upper half of registers that have become 32-bit current.uu|=current.is32&((1LL<1&&itype[i]==STORE&&itype[i-1]==STORE&&itype[i-2]==STORE&&!bt[i]) + { + cc+=4; + } else if(itype[i]==C2LS) { cc+=4; @@ -9907,11 +9937,11 @@ int new_recompile_block(int addr) if(current.regmap[HOST_BTREG]==BTREG) current.regmap[HOST_BTREG]=-1; regs[i].waswritten=current.waswritten; } - + /* Pass 4 - Cull unused host registers */ - + uint64_t nr=0; - + for (i=slen-1;i>=0;i--) { int hr; @@ -10053,7 +10083,7 @@ int new_recompile_block(int addr) } // Save it needed_reg[i]=nr; - + // Deallocate unneeded registers for(hr=0;hr=start && ba[i]<(start+i*4)) + if(ba[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 @@ -10243,10 +10273,10 @@ int new_recompile_block(int addr) } } if(ooo[i]) { - if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i+1]) + if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i+1]) f_regmap[hr]=branch_regs[i].regmap[hr]; }else{ - if(count_free_regs(branch_regs[i].regmap)<=minimum_free_regs[i+1]) + if(count_free_regs(branch_regs[i].regmap)<=minimum_free_regs[i+1]) f_regmap[hr]=branch_regs[i].regmap[hr]; } // Avoid dirty->clean transition @@ -10416,10 +10446,10 @@ int new_recompile_block(int addr) if(itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP) { if(ooo[j]) { - if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j+1]) + if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j+1]) break; }else{ - if(count_free_regs(branch_regs[j].regmap)<=minimum_free_regs[j+1]) + if(count_free_regs(branch_regs[j].regmap)<=minimum_free_regs[j+1]) break; } if(get_reg(branch_regs[j].regmap,f_regmap[hr])>=0) { @@ -10492,7 +10522,7 @@ int new_recompile_block(int addr) regs[k].isconst&=~(1<i&&f_regmap[HOST_CCREG]==CCREG) @@ -10534,7 +10564,7 @@ int new_recompile_block(int addr) } } } - + // Cache memory offset or tlb map pointer if a register is available #ifndef HOST_IMM_ADDR32 #ifndef RAM_OFFSET @@ -10714,7 +10744,7 @@ int new_recompile_block(int addr) } } #endif - + // This allocates registers (if possible) one instruction prior // to use, which can avoid a load-use penalty on certain CPUs. for(i=0;i=0) { @@ -10889,7 +10919,7 @@ int new_recompile_block(int addr) } } 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(itype[i+1]==LOAD) + if(itype[i+1]==LOAD) hr=get_reg(regs[i+1].regmap,rt1[i+1]); if(itype[i+1]==LOADLR||(opcode[i+1]&0x3b)==0x31||(opcode[i+1]&0x3b)==0x32) // LWC1/LDC1, LWC2/LDC2 hr=get_reg(regs[i+1].regmap,FTEMP); @@ -10913,16 +10943,16 @@ int new_recompile_block(int addr) } } } - + /* Pass 6 - Optimize clean/dirty state */ clean_registers(0,slen-1,1); - + /* Pass 7 - Identify 32-bit registers */ #ifndef FORCE32 provisional_r32(); u_int r32=0; - + for (i=slen-1;i>=0;i--) { int hr; @@ -11018,7 +11048,7 @@ int new_recompile_block(int addr) if((regs[i].was32>>dep2[i])&1) r32|=1LL<>32); - assem_debug("%8x (%d) <- %8x\n",instr_addr[i],i,start+i*4); - assem_debug("jump_in: %x (restricted - %x)\n",start+i*4,r); - //int entry_point=(int)out; - ////assem_debug("entry_point: %x\n",entry_point); - //load_regs_entry(i); - //if(entry_point==(int)out) - // entry_point=instr_addr[i]; - //else - // emit_jmp(instr_addr[i]); - //ll_add_32(jump_in+page,vaddr,r,(void *)entry_point); - ll_add_32(jump_dirty+vpage,vaddr,r,(void *)out); - int entry_point=do_dirty_stub(i); - ll_add_32(jump_in+page,vaddr,r,(void *)entry_point); - } } } } @@ -11597,15 +11604,15 @@ int new_recompile_block(int addr) //printf("shadow buffer: %x-%x\n",(int)copy,(int)copy+slen*4); memcpy(copy,source,slen*4); copy+=slen*4; - + #ifdef __arm__ __clear_cache((void *)beginning,out); #endif - + // If we're within 256K of the end of the buffer, // start over from the beginning. (Is 256K enough?) if((u_int)out>(u_int)BASE_ADDR+(1<>12;i<=(start+slen*4)>>12;i++) { invalid_code[i]=0; @@ -11629,9 +11636,9 @@ int new_recompile_block(int addr) invalid_code[((u_int)0x80000000>>12)|(i&0x1ff)]= invalid_code[((u_int)0xa0000000>>12)|(i&0x1ff)]=0; #endif - + /* Pass 10 - Free memory by expiring oldest blocks */ - + int end=((((int)out-(int)BASE_ADDR)>>(TARGET_SIZE_2-16))+16384)&65535; while(expirep!=end) { @@ -11673,7 +11680,7 @@ int new_recompile_block(int addr) case 3: // Clear jump_out #ifdef __arm__ - if((expirep&2047)==0) + if((expirep&2047)==0) do_clear_cache(); #endif ll_remove_matching_addrs(jump_out+(expirep&2047),base,shift);