#include <stdlib.h>
#include <stdint.h> //include for uint64_t
#include <assert.h>
+#include <errno.h>
+#include <sys/mman.h>
+#ifdef __MACH__
+#include <libkern/OSCacheControl.h>
+#endif
+#ifdef _3DS
+#include <3ds_utils.h>
+#endif
-#include "emu_if.h" //emulator interface
+#include "new_dynarec_config.h"
+#include "../psxhle.h"
+#include "../psxinterpreter.h"
+#include "../gte.h"
+#include "emu_if.h" // emulator interface
+#include "arm_features.h"
-#include <sys/mman.h>
+#define noinline __attribute__((noinline,noclone))
+#ifndef ARRAY_SIZE
+#define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
+#endif
+#ifndef min
+#define min(a, b) ((b) < (a) ? (b) : (a))
+#endif
+#ifndef max
+#define max(a, b) ((b) > (a) ? (b) : (a))
+#endif
+
+//#define DISASM
+//#define ASSEM_PRINT
+
+#ifdef ASSEM_PRINT
+#define assem_debug printf
+#else
+#define assem_debug(...)
+#endif
+//#define inv_debug printf
+#define inv_debug(...)
#ifdef __i386__
#include "assem_x86.h"
#ifdef __arm__
#include "assem_arm.h"
#endif
+#ifdef __aarch64__
+#include "assem_arm64.h"
+#endif
+#define RAM_SIZE 0x200000
#define MAXBLOCK 4096
#define MAX_OUTPUT_BLOCK_SIZE 262144
-#define CLOCK_DIVIDER 2
+#ifdef VITA
+// apparently Vita has a 16MB limit, so either we cut tc in half,
+// or use this hack (it's a hack because tc size was designed to be power-of-2)
+#define TC_REDUCE_BYTES 4096
+#else
+#define TC_REDUCE_BYTES 0
+#endif
+
+struct ndrc_mem
+{
+ u_char translation_cache[(1 << TARGET_SIZE_2) - TC_REDUCE_BYTES];
+ struct
+ {
+ struct tramp_insns ops[2048 / sizeof(struct tramp_insns)];
+ const void *f[2048 / sizeof(void *)];
+ } tramp;
+};
+
+#ifdef BASE_ADDR_DYNAMIC
+static struct ndrc_mem *ndrc;
+#else
+static struct ndrc_mem ndrc_ __attribute__((aligned(4096)));
+static struct ndrc_mem *ndrc = &ndrc_;
+#endif
+
+// stubs
+enum stub_type {
+ CC_STUB = 1,
+ FP_STUB = 2,
+ LOADB_STUB = 3,
+ LOADH_STUB = 4,
+ LOADW_STUB = 5,
+ LOADD_STUB = 6,
+ LOADBU_STUB = 7,
+ LOADHU_STUB = 8,
+ STOREB_STUB = 9,
+ STOREH_STUB = 10,
+ STOREW_STUB = 11,
+ STORED_STUB = 12,
+ STORELR_STUB = 13,
+ INVCODE_STUB = 14,
+};
+
+// regmap_pre[i] - regs before [i] insn starts; dirty things here that
+// don't match .regmap will be written back
+// [i].regmap_entry - regs that must be set up if someone jumps here
+// [i].regmap - regs [i] insn will read/(over)write
+// branch_regs[i].* - same as above but for branches, takes delay slot into account
struct regstat
{
signed char regmap_entry[HOST_REGS];
signed char regmap[HOST_REGS];
- uint64_t was32;
- uint64_t is32;
uint64_t wasdirty;
uint64_t dirty;
uint64_t u;
- uint64_t uu;
- u_int wasconst;
- u_int isconst;
- uint64_t constmap[HOST_REGS];
+ u_int wasconst; // before; for example 'lw r2, (r2)' wasconst is true
+ u_int isconst; // ... but isconst is false when r2 is known
+ u_int loadedconst; // host regs that have constants loaded
+ u_int waswritten; // MIPS regs that were used as store base before
};
+// 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];
- u_char opcode2[MAXBLOCK];
- u_char bt[MAXBLOCK];
- u_char rs1[MAXBLOCK];
- u_char rs2[MAXBLOCK];
- u_char rt1[MAXBLOCK];
- u_char rt2[MAXBLOCK];
- u_char us1[MAXBLOCK];
- u_char us2[MAXBLOCK];
- u_char dep1[MAXBLOCK];
- u_char dep2[MAXBLOCK];
- u_char lt1[MAXBLOCK];
- int imm[MAXBLOCK];
- u_int ba[MAXBLOCK];
- char likely[MAXBLOCK];
- char is_ds[MAXBLOCK];
- char ooo[MAXBLOCK];
- uint64_t unneeded_reg[MAXBLOCK];
- uint64_t unneeded_reg_upper[MAXBLOCK];
- uint64_t branch_unneeded_reg[MAXBLOCK];
- uint64_t branch_unneeded_reg_upper[MAXBLOCK];
- uint64_t p32[MAXBLOCK];
- uint64_t pr32[MAXBLOCK];
- signed char regmap_pre[MAXBLOCK][HOST_REGS];
- signed char regmap[MAXBLOCK][HOST_REGS];
- signed char regmap_entry[MAXBLOCK][HOST_REGS];
- uint64_t constmap[MAXBLOCK][HOST_REGS];
- struct regstat regs[MAXBLOCK];
- struct regstat branch_regs[MAXBLOCK];
- signed char minimum_free_regs[MAXBLOCK];
- u_int needed_reg[MAXBLOCK];
- uint64_t requires_32bit[MAXBLOCK];
- u_int wont_dirty[MAXBLOCK];
- u_int will_dirty[MAXBLOCK];
- int ccadj[MAXBLOCK];
- int slen;
- u_int instr_addr[MAXBLOCK];
- u_int link_addr[MAXBLOCK][3];
- int linkcount;
- u_int stubs[MAXBLOCK*3][8];
- int stubcount;
- u_int literals[1024][2];
- int literalcount;
- int is_delayslot;
- int cop1_usable;
+struct ht_entry
+{
+ u_int vaddr[2];
+ void *tcaddr[2];
+};
+
+struct code_stub
+{
+ enum stub_type type;
+ void *addr;
+ void *retaddr;
+ u_int a;
+ uintptr_t b;
+ uintptr_t c;
+ u_int d;
+ u_int e;
+};
+
+struct link_entry
+{
+ void *addr;
+ u_int target;
+ 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 use_lt1:1;
+ 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 ll_entry *jump_in[4096];
- struct ll_entry *jump_out[4096];
+ struct ht_entry hash_table[65536] __attribute__((aligned(16)));
+ struct ll_entry *jump_in[4096] __attribute__((aligned(16)));
struct ll_entry *jump_dirty[4096];
- u_int hash_table[65536][4] __attribute__((aligned(16)));
- char shadow[1048576] __attribute__((aligned(16)));
- void *copy;
- int expirep;
-#ifndef PCSX
- u_int using_tlb;
-#else
- static const u_int using_tlb=0;
-#endif
- static u_int sp_in_mirror;
- u_int stop_after_jal;
+
+ static struct ll_entry *jump_out[4096];
+ static u_int start;
+ static u_int *source;
+ static uint64_t gte_rs[MAXBLOCK]; // gte: 32 data and 32 ctl regs
+ static uint64_t gte_rt[MAXBLOCK];
+ static uint64_t gte_unneeded[MAXBLOCK];
+ static u_int smrv[32]; // speculated MIPS register values
+ static u_int smrv_strong; // mask or regs that are likely to have correct values
+ static u_int smrv_weak; // same, but somewhat less likely
+ static u_int smrv_strong_next; // same, but after current insn executes
+ static u_int smrv_weak_next;
+ static int imm[MAXBLOCK];
+ static u_int ba[MAXBLOCK];
+ static uint64_t unneeded_reg[MAXBLOCK];
+ static uint64_t branch_unneeded_reg[MAXBLOCK];
+ // see 'struct regstat' for a description
+ static signed char regmap_pre[MAXBLOCK][HOST_REGS];
+ // contains 'real' consts at [i] insn, but may differ from what's actually
+ // loaded in host reg as 'final' value is always loaded, see get_final_value()
+ static uint32_t current_constmap[HOST_REGS];
+ static uint32_t constmap[MAXBLOCK][HOST_REGS];
+ static struct regstat regs[MAXBLOCK];
+ static struct regstat branch_regs[MAXBLOCK];
+ static signed char minimum_free_regs[MAXBLOCK];
+ static int ccadj[MAXBLOCK];
+ static int slen;
+ static void *instr_addr[MAXBLOCK];
+ static struct link_entry link_addr[MAXBLOCK];
+ static int linkcount;
+ static struct code_stub stubs[MAXBLOCK*3];
+ static int stubcount;
+ static u_int literals[1024][2];
+ static int literalcount;
+ static int is_delayslot;
+ static char shadow[1048576] __attribute__((aligned(16)));
+ static void *copy;
+ static int expirep;
+ static u_int stop_after_jal;
+ static u_int f1_hack;
+
+ int new_dynarec_hacks;
+ int new_dynarec_hacks_pergame;
+ int new_dynarec_hacks_old;
+ int new_dynarec_did_compile;
+
+ #define HACK_ENABLED(x) ((new_dynarec_hacks | new_dynarec_hacks_pergame) & (x))
+
+ extern int cycle_count; // ... until end of the timeslice, counts -N -> 0
+ extern int last_count; // last absolute target, often = next_interupt
+ 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];
- extern int cycle_count;
/* registers that may be allocated */
/* 1-31 gpr */
-#define HIREG 32 // hi
-#define LOREG 33 // lo
-#define FSREG 34 // FPU status (FCSR)
+#define LOREG 32 // lo
+#define HIREG 33 // hi
+//#define FSREG 34 // FPU status (FCSR)
#define CSREG 35 // Coprocessor status
#define CCREG 36 // Cycle count
#define INVCP 37 // Pointer to invalid_code
-#define MMREG 38 // Pointer to memory_map
+//#define MMREG 38 // Pointer to memory_map
#define ROREG 39 // ram offset (if rdram!=0x80000000)
#define TEMPREG 40
#define FTEMP 40 // FPU temporary register
#define PTEMP 41 // Prefetch temporary register
-#define TLREG 42 // TLB mapping offset
+//#define TLREG 42 // TLB mapping offset
#define RHASH 43 // Return address hash
#define RHTBL 44 // Return address hash table address
#define RTEMP 45 // JR/JALR address register
#define MAXREG 45
#define AGEN1 46 // Address generation temporary register
-#define AGEN2 47 // Address generation temporary register
-#define MGEN1 48 // Maptable address generation temporary register
-#define MGEN2 49 // Maptable address generation temporary register
+//#define AGEN2 47 // Address generation temporary register
+//#define MGEN1 48 // Maptable address generation temporary register
+//#define MGEN2 49 // Maptable address generation temporary register
#define BTREG 50 // Branch target temporary register
/* instruction types */
#define 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
#define COP0 15 // Coprocessor 0
#define COP1 16 // Coprocessor 1
#define C1LS 17 // Coprocessor 1 load/store
-#define FJUMP 18 // Conditional branch (floating point)
-#define FLOAT 19 // Floating point unit
-#define FCONV 20 // Convert integer to float
-#define FCOMP 21 // Floating point compare (sets FSREG)
-#define SYSCALL 22// SYSCALL
+//#define FJUMP 18 // Conditional branch (floating point)
+//#define FLOAT 19 // Floating point unit
+//#define FCONV 20 // Convert integer to float
+//#define FCOMP 21 // Floating point compare (sets FSREG)
+#define SYSCALL 22// SYSCALL,BREAK
#define OTHER 23 // Other
#define SPAN 24 // Branch/delay slot spans 2 pages
#define NI 25 // Not implemented
#define C2OP 29 // Coprocessor 2 operation
#define INTCALL 30// Call interpreter to handle rare corner cases
- /* stubs */
-#define CC_STUB 1
-#define FP_STUB 2
-#define LOADB_STUB 3
-#define LOADH_STUB 4
-#define LOADW_STUB 5
-#define LOADD_STUB 6
-#define LOADBU_STUB 7
-#define LOADHU_STUB 8
-#define STOREB_STUB 9
-#define STOREH_STUB 10
-#define STOREW_STUB 11
-#define STORED_STUB 12
-#define STORELR_STUB 13
-#define INVCODE_STUB 14
-
/* branch codes */
#define TAKEN 1
#define NOTTAKEN 2
#define NULLDS 3
+#define DJT_1 (void *)1l // no function, just a label in assem_debug log
+#define DJT_2 (void *)2l
+
// asm linkage
-int new_recompile_block(int addr);
+int new_recompile_block(u_int addr);
void *get_addr_ht(u_int vaddr);
void invalidate_block(u_int block);
void invalidate_addr(u_int addr);
void remove_hash(int vaddr);
-void jump_vaddr();
void dyna_linker();
void dyna_linker_ds();
void verify_code();
-void verify_code_vm();
void verify_code_ds();
void cc_interrupt();
void fp_exception();
void fp_exception_ds();
-void jump_syscall();
-void jump_syscall_hle();
-void jump_eret();
-void jump_hlecall();
-void jump_intcall();
+void jump_syscall (u_int u0, u_int u1, u_int pc);
+void jump_syscall_ds(u_int u0, u_int u1, u_int pc);
+void jump_break (u_int u0, u_int u1, u_int pc);
+void jump_break_ds(u_int u0, u_int u1, u_int pc);
+void jump_to_new_pc();
+void call_gteStall();
+void clean_blocks(u_int page);
+void add_jump_out(u_int vaddr, void *src);
void new_dyna_leave();
-// TLB
-void TLBWI_new();
-void TLBWR_new();
-void read_nomem_new();
-void read_nomemb_new();
-void read_nomemh_new();
-void read_nomemd_new();
-void write_nomem_new();
-void write_nomemb_new();
-void write_nomemh_new();
-void write_nomemd_new();
-void write_rdram_new();
-void write_rdramb_new();
-void write_rdramh_new();
-void write_rdramd_new();
-extern u_int memory_map[1048576];
-
// Needed by assembler
-void wb_register(signed char r,signed char regmap[],uint64_t dirty,uint64_t is32);
-void wb_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty);
-void wb_needed_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr);
-void load_all_regs(signed char i_regmap[]);
-void load_needed_regs(signed char i_regmap[],signed char next_regmap[]);
-void load_regs_entry(int t);
-void load_all_consts(signed char regmap[],int is32,u_int dirty,int i);
+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(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);
+static int get_final_value(int hr, int i, int *value);
+static void add_stub(enum stub_type type, void *addr, void *retaddr,
+ u_int a, uintptr_t b, uintptr_t c, u_int d, u_int e);
+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, 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);
+static void pass_args(int a0, int a1);
+static void emit_far_jump(const void *f);
+static void emit_far_call(const void *f);
+
+#ifdef VITA
+#include <psp2/kernel/sysmem.h>
+static int sceBlock;
+// note: this interacts with RetroArch's Vita bootstrap code: bootstrap/vita/sbrk.c
+extern int getVMBlock();
+int _newlib_vm_size_user = sizeof(*ndrc);
+#endif
+
+static void mprotect_w_x(void *start, void *end, int is_x)
+{
+#ifdef NO_WRITE_EXEC
+ #if defined(VITA)
+ // *Open* enables write on all memory that was
+ // allocated by sceKernelAllocMemBlockForVM()?
+ if (is_x)
+ sceKernelCloseVMDomain();
+ else
+ sceKernelOpenVMDomain();
+ #else
+ u_long mstart = (u_long)start & ~4095ul;
+ u_long mend = (u_long)end;
+ if (mprotect((void *)mstart, mend - mstart,
+ PROT_READ | (is_x ? PROT_EXEC : PROT_WRITE)) != 0)
+ SysPrintf("mprotect(%c) failed: %s\n", is_x ? 'x' : 'w', strerror(errno));
+ #endif
+#endif
+}
-int tracedebug=0;
+static void start_tcache_write(void *start, void *end)
+{
+ mprotect_w_x(start, end, 0);
+}
-//#define DEBUG_CYCLE_COUNT 1
+static void end_tcache_write(void *start, void *end)
+{
+#if defined(__arm__) || defined(__aarch64__)
+ size_t len = (char *)end - (char *)start;
+ #if defined(__BLACKBERRY_QNX__)
+ msync(start, len, MS_SYNC | MS_CACHE_ONLY | MS_INVALIDATE_ICACHE);
+ #elif defined(__MACH__)
+ sys_cache_control(kCacheFunctionPrepareForExecution, start, len);
+ #elif defined(VITA)
+ sceKernelSyncVMDomain(sceBlock, start, len);
+ #elif defined(_3DS)
+ ctr_flush_invalidate_cache();
+ #elif defined(__aarch64__)
+ // as of 2021, __clear_cache() is still broken on arm64
+ // so here is a custom one :(
+ clear_cache_arm64(start, end);
+ #else
+ __clear_cache(start, end);
+ #endif
+ (void)len;
+#endif
-void nullf() {}
-//#define assem_debug printf
-//#define inv_debug printf
-#define assem_debug nullf
-#define inv_debug nullf
+ mprotect_w_x(start, end, 1);
+}
+
+static void *start_block(void)
+{
+ u_char *end = out + MAX_OUTPUT_BLOCK_SIZE;
+ if (end > ndrc->translation_cache + sizeof(ndrc->translation_cache))
+ end = ndrc->translation_cache + sizeof(ndrc->translation_cache);
+ start_tcache_write(out, end);
+ return out;
+}
+
+static void end_block(void *start)
+{
+ end_tcache_write(start, out);
+}
-static void tlb_hacks()
+// also takes care of w^x mappings when patching code
+static u_int needs_clear_cache[1<<(TARGET_SIZE_2-17)];
+
+static void mark_clear_cache(void *target)
+{
+ uintptr_t offset = (u_char *)target - ndrc->translation_cache;
+ u_int mask = 1u << ((offset >> 12) & 31);
+ if (!(needs_clear_cache[offset >> 17] & mask)) {
+ char *start = (char *)((uintptr_t)target & ~4095l);
+ start_tcache_write(start, start + 4095);
+ needs_clear_cache[offset >> 17] |= mask;
+ }
+}
+
+// Clearing the cache is rather slow on ARM Linux, so mark the areas
+// that need to be cleared, and then only clear these areas once.
+static void do_clear_cache(void)
{
-#ifndef DISABLE_TLB
- // Goldeneye hack
- if (strncmp((char *) ROM_HEADER->nom, "GOLDENEYE",9) == 0)
+ int i, j;
+ for (i = 0; i < (1<<(TARGET_SIZE_2-17)); i++)
{
- u_int addr;
- int n;
- switch (ROM_HEADER->Country_code&0xFF)
+ u_int bitmap = needs_clear_cache[i];
+ if (!bitmap)
+ continue;
+ for (j = 0; j < 32; j++)
{
- case 0x45: // U
- addr=0x34b30;
- break;
- case 0x4A: // J
- addr=0x34b70;
- break;
- case 0x50: // E
- addr=0x329f0;
- break;
- default:
- // Unknown country code
- addr=0;
- break;
- }
- u_int rom_addr=(u_int)rom;
- #ifdef ROM_COPY
- // Since memory_map is 32-bit, on 64-bit systems the rom needs to be
- // in the lower 4G of memory to use this hack. Copy it if necessary.
- if((void *)rom>(void *)0xffffffff) {
- munmap(ROM_COPY, 67108864);
- if(mmap(ROM_COPY, 12582912,
- PROT_READ | PROT_WRITE,
- MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
- -1, 0) <= 0) {printf("mmap() failed\n");}
- memcpy(ROM_COPY,rom,12582912);
- rom_addr=(u_int)ROM_COPY;
- }
- #endif
- if(addr) {
- for(n=0x7F000;n<0x80000;n++) {
- memory_map[n]=(((u_int)(rom_addr+addr-0x7F000000))>>2)|0x40000000;
+ u_char *start, *end;
+ if (!(bitmap & (1<<j)))
+ continue;
+
+ start = ndrc->translation_cache + i*131072 + j*4096;
+ end = start + 4095;
+ for (j++; j < 32; j++) {
+ if (!(bitmap & (1<<j)))
+ break;
+ end += 4096;
}
+ end_tcache_write(start, end);
}
+ needs_clear_cache[i] = 0;
}
-#endif
+}
+
+//#define DEBUG_CYCLE_COUNT 1
+
+#define NO_CYCLE_PENALTY_THR 12
+
+int cycle_multiplier = CYCLE_MULT_DEFAULT; // 100 for 1.0
+int cycle_multiplier_override;
+int cycle_multiplier_old;
+static int cycle_multiplier_active;
+
+static int CLOCK_ADJUST(int x)
+{
+ int m = cycle_multiplier_active;
+ int s = (x >> 31) | 1;
+ return (x * m + s * 50) / 100;
+}
+
+static int ds_writes_rjump_rs(int i)
+{
+ 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)
{
-#ifndef PCSX
- u_int page=(vaddr^0x80000000)>>12;
-#else
u_int page=vaddr&~0xe0000000;
if (page < 0x1000000)
page &= ~0x0e00000; // RAM mirrors
page>>=12;
-#endif
-#ifndef DISABLE_TLB
- if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[vaddr>>12]^0x80000000)>>12;
-#endif
if(page>2048) page=2048+(page&2047);
return page;
}
+// no virtual mem in PCSX
static u_int get_vpage(u_int vaddr)
{
- u_int vpage=(vaddr^0x80000000)>>12;
-#ifndef DISABLE_TLB
- if(vpage>262143&&tlb_LUT_r[vaddr>>12]) vpage&=2047; // jump_dirty uses a hash of the virtual address instead
-#endif
- if(vpage>2048) vpage=2048+(vpage&2047);
- return vpage;
+ return get_page(vaddr);
+}
+
+static struct ht_entry *hash_table_get(u_int vaddr)
+{
+ return &hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+}
+
+static void hash_table_add(struct ht_entry *ht_bin, u_int vaddr, void *tcaddr)
+{
+ ht_bin->vaddr[1] = ht_bin->vaddr[0];
+ ht_bin->tcaddr[1] = ht_bin->tcaddr[0];
+ ht_bin->vaddr[0] = vaddr;
+ ht_bin->tcaddr[0] = tcaddr;
+}
+
+// some messy ari64's code, seems to rely on unsigned 32bit overflow
+static int doesnt_expire_soon(void *tcaddr)
+{
+ u_int diff = (u_int)((u_char *)tcaddr - out) << (32-TARGET_SIZE_2);
+ return diff > (u_int)(0x60000000 + (MAX_OUTPUT_BLOCK_SIZE << (32-TARGET_SIZE_2)));
}
// Get address from virtual address
// This is called from the recompiled JR/JALR instructions
-void *get_addr(u_int vaddr)
+void noinline *get_addr(u_int vaddr)
{
u_int page=get_page(vaddr);
u_int vpage=get_vpage(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) {
- //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];
- ht_bin[2]=ht_bin[0];
- ht_bin[1]=(int)head->addr;
- ht_bin[0]=vaddr;
+ if(head->vaddr==vaddr) {
+ //printf("TRACE: count=%d next=%d (get_addr match %x: %p)\n",Count,next_interupt,vaddr,head->addr);
+ hash_table_add(hash_table_get(vaddr), vaddr, head->addr);
return head->addr;
}
head=head->next;
}
head=jump_dirty[vpage];
while(head!=NULL) {
- if(head->vaddr==vaddr&&head->reg32==0) {
- //printf("TRACE: count=%d next=%d (get_addr match dirty %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr);
+ if(head->vaddr==vaddr) {
+ //printf("TRACE: count=%d next=%d (get_addr match dirty %x: %p)\n",Count,next_interupt,vaddr,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)) {
+ if (doesnt_expire_soon(head->addr))
+ 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);
- int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
- if(ht_bin[0]==vaddr) {
- ht_bin[1]=(int)head->addr; // Replace existing entry
- }
+ struct ht_entry *ht_bin = hash_table_get(vaddr);
+ if (ht_bin->vaddr[0] == vaddr)
+ ht_bin->tcaddr[0] = head->addr; // Replace existing entry
else
- {
- ht_bin[3]=ht_bin[1];
- ht_bin[2]=ht_bin[0];
- ht_bin[1]=(int)head->addr;
- ht_bin[0]=vaddr;
- }
+ hash_table_add(ht_bin, vaddr, head->addr);
+
return head->addr;
}
}
//printf("TRACE: count=%d next=%d (get_addr no-match %x)\n",Count,next_interupt,vaddr);
int r=new_recompile_block(vaddr);
if(r==0) return get_addr(vaddr);
- // Execute in unmapped page, generate pagefault execption
+ // generate an address error
Status|=2;
- Cause=(vaddr<<31)|0x8;
+ Cause=(vaddr<<31)|(4<<2);
EPC=(vaddr&1)?vaddr-5:vaddr;
BadVAddr=(vaddr&~1);
- Context=(Context&0xFF80000F)|((BadVAddr>>9)&0x007FFFF0);
- EntryHi=BadVAddr&0xFFFFE000;
- return get_addr_ht(0x80000000);
+ return get_addr_ht(0x80000080);
}
// Look up address in hash table first
void *get_addr_ht(u_int vaddr)
{
//printf("TRACE: count=%d next=%d (get_addr_ht %x)\n",Count,next_interupt,vaddr);
- 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];
+ const struct ht_entry *ht_bin = hash_table_get(vaddr);
+ if (ht_bin->vaddr[0] == vaddr) return ht_bin->tcaddr[0];
+ if (ht_bin->vaddr[1] == vaddr) return ht_bin->tcaddr[1];
return get_addr(vaddr);
}
-void *get_addr_32(u_int vaddr,u_int flags)
+static void clear_all_regs(signed char regmap[])
{
-#ifdef FORCE32
- return get_addr(vaddr);
+ memset(regmap, -1, sizeof(regmap[0]) * HOST_REGS);
+}
+
+// get_reg: get allocated host reg from mips reg
+// returns -1 if no such mips reg was allocated
+#if defined(__arm__) && defined(HAVE_ARMV6) && HOST_REGS == 13 && EXCLUDE_REG == 11
+
+extern signed char get_reg(const signed char regmap[], signed char r);
+
#else
- //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;
+
+static signed char get_reg(const signed char regmap[], signed char r)
+{
+ int hr;
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ if (hr == EXCLUDE_REG)
+ continue;
+ if (regmap[hr] == r)
+ return hr;
}
- //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
+ return -1;
}
-void clear_all_regs(signed char regmap[])
+#endif
+
+// get reg as mask bit (1 << hr)
+static u_int get_regm(const signed char regmap[], signed char r)
{
- int hr;
- for (hr=0;hr<HOST_REGS;hr++) regmap[hr]=-1;
+ return (1u << (get_reg(regmap, r) & 31)) & ~(1u << 31);
}
-signed char get_reg(signed char regmap[],int r)
+static signed char get_reg_temp(const signed char regmap[])
{
int hr;
- for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&®map[hr]==r) return hr;
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ if (hr == EXCLUDE_REG)
+ continue;
+ if (regmap[hr] == (signed char)-1)
+ return hr;
+ }
return -1;
}
// Find a register that is available for two consecutive cycles
-signed char get_reg2(signed char regmap1[],signed char regmap2[],int r)
+static signed char get_reg2(signed char regmap1[], const signed char regmap2[], int r)
{
int hr;
for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&®map1[hr]==r&®map2[hr]==r) return hr;
return -1;
}
-int count_free_regs(signed char regmap[])
+// reverse reg map: mips -> host
+#define RRMAP_SIZE 64
+static void make_rregs(const signed char regmap[], signed char rrmap[RRMAP_SIZE],
+ u_int *regs_can_change)
+{
+ u_int r, hr, hr_can_change = 0;
+ memset(rrmap, -1, RRMAP_SIZE);
+ for (hr = 0; hr < HOST_REGS; )
+ {
+ r = regmap[hr];
+ rrmap[r & (RRMAP_SIZE - 1)] = hr;
+ // only add mips $1-$31+$lo, others shifted out
+ hr_can_change |= (uint64_t)1 << (hr + ((r - 1) & 32));
+ hr++;
+ if (hr == EXCLUDE_REG)
+ hr++;
+ }
+ hr_can_change |= 1u << (rrmap[33] & 31);
+ hr_can_change |= 1u << (rrmap[CCREG] & 31);
+ hr_can_change &= ~(1u << 31);
+ *regs_can_change = hr_can_change;
+}
+
+// same as get_reg, but takes rrmap
+static signed char get_rreg(signed char rrmap[RRMAP_SIZE], signed char r)
+{
+ assert(0 <= r && r < RRMAP_SIZE);
+ return rrmap[r];
+}
+
+static int count_free_regs(const signed char regmap[])
{
int count=0;
int hr;
return count;
}
-void dirty_reg(struct regstat *cur,signed char reg)
+static void dirty_reg(struct regstat *cur, signed char reg)
{
int hr;
- if(!reg) return;
- for (hr=0;hr<HOST_REGS;hr++) {
- if((cur->regmap[hr]&63)==reg) {
- cur->dirty|=1<<hr;
- }
- }
-}
-
-// If we dirty the lower half of a 64 bit register which is now being
-// sign-extended, we need to dump the upper half.
-// Note: Do this only after completion of the instruction, because
-// some instructions may need to read the full 64-bit value even if
-// overwriting it (eg SLTI, DSRA32).
-static void flush_dirty_uppers(struct regstat *cur)
-{
- int hr,reg;
- for (hr=0;hr<HOST_REGS;hr++) {
- if((cur->dirty>>hr)&1) {
- reg=cur->regmap[hr];
- if(reg>=64)
- if((cur->is32>>(reg&63))&1) cur->regmap[hr]=-1;
- }
- }
+ if (!reg) return;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0)
+ cur->dirty |= 1<<hr;
}
-void set_const(struct regstat *cur,signed char reg,uint64_t value)
+static void set_const(struct regstat *cur, signed char reg, uint32_t value)
{
int hr;
- if(!reg) return;
- for (hr=0;hr<HOST_REGS;hr++) {
- if(cur->regmap[hr]==reg) {
- cur->isconst|=1<<hr;
- cur->constmap[hr]=value;
- }
- else if((cur->regmap[hr]^64)==reg) {
- cur->isconst|=1<<hr;
- cur->constmap[hr]=value>>32;
- }
+ if (!reg) return;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0) {
+ cur->isconst |= 1<<hr;
+ current_constmap[hr] = value;
}
}
-void clear_const(struct regstat *cur,signed char reg)
+static void clear_const(struct regstat *cur, signed char reg)
{
int hr;
- if(!reg) return;
- for (hr=0;hr<HOST_REGS;hr++) {
- if((cur->regmap[hr]&63)==reg) {
- cur->isconst&=~(1<<hr);
- }
- }
+ if (!reg) return;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0)
+ cur->isconst &= ~(1<<hr);
}
-int is_const(struct regstat *cur,signed char reg)
+static int is_const(const struct regstat *cur, signed char reg)
{
int hr;
- if(reg<0) return 0;
- if(!reg) return 1;
- for (hr=0;hr<HOST_REGS;hr++) {
- if((cur->regmap[hr]&63)==reg) {
- return (cur->isconst>>hr)&1;
- }
- }
+ if (reg < 0) return 0;
+ if (!reg) return 1;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0)
+ return (cur->isconst>>hr)&1;
return 0;
}
-uint64_t get_const(struct regstat *cur,signed char reg)
+
+static uint32_t get_const(const struct regstat *cur, signed char reg)
{
int hr;
- if(!reg) return 0;
- for (hr=0;hr<HOST_REGS;hr++) {
- if(cur->regmap[hr]==reg) {
- return cur->constmap[hr];
- }
- }
- printf("Unknown constant in r%d\n",reg);
- exit(1);
+ if (!reg) return 0;
+ hr = get_reg(cur->regmap, reg);
+ if (hr >= 0)
+ return current_constmap[hr];
+
+ SysPrintf("Unknown constant in r%d\n", reg);
+ abort();
}
// Least soon needed registers
// Look at the next ten instructions and see which registers
// will be used. Try not to reallocate these.
-void lsn(u_char hsn[], int i, int *preferred_reg)
+static void lsn(u_char hsn[], int i, int *preferred_reg)
{
int j;
int b=-1;
j=slen-i-1;
break;
}
- if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+ if (dops[i+j].is_ujump)
{
// Don't go past an unconditonal jump
j++;
}
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||itype[i+j]==FJUMP))
+ if(i+j>=0&&(dops[i+j].itype==UJUMP||dops[i+j].itype==CJUMP||dops[i+j].itype==SJUMP))
{
hsn[CCREG]=j;
b=j;
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&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)) {
- 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 TLB registers either
- if(itype[i]==LOAD || itype[i]==LOADLR || itype[i]==STORE || itype[i]==STORELR || itype[i]==C1LS || itype[i]==C2LS) {
- hsn[TLREG]=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;
}
// We only want to allocate registers if we're going to use them again soon
-int needed_again(int r, int i)
+static 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 (i > 0 && dops[i-1].is_ujump)
{
if(ba[i-1]<start || ba[i-1]>start+slen*4-4)
return 0; // Don't need any registers if exiting the block
j=slen-i-1;
break;
}
- if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+ 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||itype[i+j]==FJUMP))
+ 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;
}
// Try to match register allocations at the end of a loop with those
// at the beginning
-int loop_reg(int i, int r, int hr)
+static int loop_reg(int i, int r, int hr)
{
int j,k;
for(j=0;j<9;j++)
j=slen-i-1;
break;
}
- if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+ if (dops[i+j].is_ujump)
{
// Don't go past an unconditonal jump
j++;
}
k=0;
if(i>0){
- if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)
+ if(dops[i-1].itype==UJUMP||dops[i-1].itype==CJUMP||dops[i-1].itype==SJUMP)
k--;
}
for(;k<j;k++)
{
- if(r<64&&((unneeded_reg[i+k]>>r)&1)) return hr;
- if(r>64&&((unneeded_reg_upper[i+k]>>r)&1)) return hr;
- if(i+k>=0&&(itype[i+k]==UJUMP||itype[i+k]==CJUMP||itype[i+k]==SJUMP||itype[i+k]==FJUMP))
+ assert(r < 64);
+ if((unneeded_reg[i+k]>>r)&1) return hr;
+ 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))
{
// Allocate every register, preserving source/target regs
-void alloc_all(struct regstat *cur,int i)
+static void alloc_all(struct regstat *cur,int i)
{
int hr;
-
+
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG) {
- if(((cur->regmap[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]!=dops[i].rs1)&&(cur->regmap[hr]!=dops[i].rs2)&&
+ (cur->regmap[hr]!=dops[i].rt1)&&(cur->regmap[hr]!=dops[i].rt2))
{
cur->regmap[hr]=-1;
cur->dirty&=~(1<<hr);
}
// Don't need zeros
- if((cur->regmap[hr]&63)==0)
+ if(cur->regmap[hr]==0)
{
cur->regmap[hr]=-1;
cur->dirty&=~(1<<hr);
}
}
+#ifndef NDEBUG
+static int host_tempreg_in_use;
+
+static void host_tempreg_acquire(void)
+{
+ assert(!host_tempreg_in_use);
+ host_tempreg_in_use = 1;
+}
+
+static void host_tempreg_release(void)
+{
+ host_tempreg_in_use = 0;
+}
+#else
+static void host_tempreg_acquire(void) {}
+static void host_tempreg_release(void) {}
+#endif
+
+#ifdef ASSEM_PRINT
+extern void gen_interupt();
+extern void do_insn_cmp();
+#define FUNCNAME(f) { f, " " #f }
+static const struct {
+ void *addr;
+ const char *name;
+} function_names[] = {
+ FUNCNAME(cc_interrupt),
+ FUNCNAME(gen_interupt),
+ FUNCNAME(get_addr_ht),
+ FUNCNAME(get_addr),
+ FUNCNAME(jump_handler_read8),
+ FUNCNAME(jump_handler_read16),
+ FUNCNAME(jump_handler_read32),
+ FUNCNAME(jump_handler_write8),
+ FUNCNAME(jump_handler_write16),
+ FUNCNAME(jump_handler_write32),
+ FUNCNAME(invalidate_addr),
+ FUNCNAME(jump_to_new_pc),
+ FUNCNAME(jump_break),
+ FUNCNAME(jump_break_ds),
+ FUNCNAME(jump_syscall),
+ FUNCNAME(jump_syscall_ds),
+ FUNCNAME(call_gteStall),
+ FUNCNAME(clean_blocks),
+ FUNCNAME(new_dyna_leave),
+ FUNCNAME(pcsx_mtc0),
+ FUNCNAME(pcsx_mtc0_ds),
+#ifdef DRC_DBG
+ FUNCNAME(do_insn_cmp),
+#endif
+#ifdef __arm__
+ FUNCNAME(verify_code),
+#endif
+};
-void div64(int64_t dividend,int64_t divisor)
-{
- lo=dividend/divisor;
- hi=dividend%divisor;
- //printf("TRACE: ddiv %8x%8x %8x%8x\n" ,(int)reg[HIREG],(int)(reg[HIREG]>>32)
- // ,(int)reg[LOREG],(int)(reg[LOREG]>>32));
-}
-void divu64(uint64_t dividend,uint64_t divisor)
-{
- lo=dividend/divisor;
- hi=dividend%divisor;
- //printf("TRACE: ddivu %8x%8x %8x%8x\n",(int)reg[HIREG],(int)(reg[HIREG]>>32)
- // ,(int)reg[LOREG],(int)(reg[LOREG]>>32));
-}
-
-void mult64(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;
- int sign = 0;
-
- if (m1 < 0)
- {
- op2 = -m1;
- sign = 1 - sign;
- }
- else op2 = m1;
- if (m2 < 0)
- {
- op4 = -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)
- {
- hi = ~hi;
- if (!lo) hi++;
- else lo = ~lo + 1;
- }
-}
-
-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));
-}
-
-uint64_t ldl_merge(uint64_t original,uint64_t loaded,u_int bits)
-{
- if(bits) {
- original<<=64-bits;
- original>>=64-bits;
- loaded<<=bits;
- original|=loaded;
- }
- else original=loaded;
- return original;
-}
-uint64_t ldr_merge(uint64_t original,uint64_t loaded,u_int bits)
-{
- if(bits^56) {
- original>>=64-(bits^56);
- original<<=64-(bits^56);
- loaded>>=bits^56;
- original|=loaded;
- }
- else original=loaded;
- return original;
+static const char *func_name(const void *a)
+{
+ int i;
+ for (i = 0; i < sizeof(function_names)/sizeof(function_names[0]); i++)
+ if (function_names[i].addr == a)
+ return function_names[i].name;
+ return "";
}
+#else
+#define func_name(x) ""
+#endif
#ifdef __i386__
#include "assem_x86.c"
#ifdef __arm__
#include "assem_arm.c"
#endif
+#ifdef __aarch64__
+#include "assem_arm64.c"
+#endif
+
+static void *get_trampoline(const void *f)
+{
+ size_t i;
+
+ for (i = 0; i < ARRAY_SIZE(ndrc->tramp.f); i++) {
+ if (ndrc->tramp.f[i] == f || ndrc->tramp.f[i] == NULL)
+ break;
+ }
+ if (i == ARRAY_SIZE(ndrc->tramp.f)) {
+ SysPrintf("trampoline table is full, last func %p\n", f);
+ abort();
+ }
+ if (ndrc->tramp.f[i] == NULL) {
+ start_tcache_write(&ndrc->tramp.f[i], &ndrc->tramp.f[i + 1]);
+ ndrc->tramp.f[i] = f;
+ end_tcache_write(&ndrc->tramp.f[i], &ndrc->tramp.f[i + 1]);
+ }
+ return &ndrc->tramp.ops[i];
+}
+
+static void emit_far_jump(const void *f)
+{
+ if (can_jump_or_call(f)) {
+ emit_jmp(f);
+ return;
+ }
+
+ f = get_trampoline(f);
+ emit_jmp(f);
+}
+
+static void emit_far_call(const void *f)
+{
+ if (can_jump_or_call(f)) {
+ emit_call(f);
+ return;
+ }
+
+ f = get_trampoline(f);
+ emit_call(f);
+}
// Add virtual address mapping to linked list
-void ll_add(struct ll_entry **head,int vaddr,void *addr)
+static void ll_add(struct ll_entry **head,int vaddr,void *addr)
{
struct ll_entry *new_entry;
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)
+static 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
// but don't return addresses which are about to expire from the cache
-void *check_addr(u_int vaddr)
+static void *check_addr(u_int vaddr)
{
- u_int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
- if(ht_bin[0]==vaddr) {
- if(((ht_bin[1]-MAX_OUTPUT_BLOCK_SIZE-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
- if(isclean(ht_bin[1])) return (void *)ht_bin[1];
- }
- if(ht_bin[2]==vaddr) {
- if(((ht_bin[3]-MAX_OUTPUT_BLOCK_SIZE-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
- if(isclean(ht_bin[3])) return (void *)ht_bin[3];
+ struct ht_entry *ht_bin = hash_table_get(vaddr);
+ size_t i;
+ for (i = 0; i < ARRAY_SIZE(ht_bin->vaddr); i++) {
+ if (ht_bin->vaddr[i] == vaddr)
+ if (doesnt_expire_soon((u_char *)ht_bin->tcaddr[i] - MAX_OUTPUT_BLOCK_SIZE))
+ if (isclean(ht_bin->tcaddr[i]))
+ return ht_bin->tcaddr[i];
}
u_int page=get_page(vaddr);
struct ll_entry *head;
head=jump_in[page];
- while(head!=NULL) {
- if(head->vaddr==vaddr&&head->reg32==0) {
- if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) {
+ while (head != NULL) {
+ if (head->vaddr == vaddr) {
+ if (doesnt_expire_soon(head->addr)) {
// Update existing entry with current address
- if(ht_bin[0]==vaddr) {
- ht_bin[1]=(int)head->addr;
+ if (ht_bin->vaddr[0] == vaddr) {
+ ht_bin->tcaddr[0] = head->addr;
return head->addr;
}
- if(ht_bin[2]==vaddr) {
- ht_bin[3]=(int)head->addr;
+ if (ht_bin->vaddr[1] == vaddr) {
+ ht_bin->tcaddr[1] = head->addr;
return head->addr;
}
// Insert into hash table with low priority.
// Don't evict existing entries, as they are probably
// addresses that are being accessed frequently.
- 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;
+ if (ht_bin->vaddr[0] == -1) {
+ ht_bin->vaddr[0] = vaddr;
+ ht_bin->tcaddr[0] = head->addr;
+ }
+ else if (ht_bin->vaddr[1] == -1) {
+ ht_bin->vaddr[1] = vaddr;
+ ht_bin->tcaddr[1] = head->addr;
}
return head->addr;
}
void remove_hash(int vaddr)
{
//printf("remove hash: %x\n",vaddr);
- int *ht_bin=hash_table[(((vaddr)>>16)^vaddr)&0xFFFF];
- if(ht_bin[2]==vaddr) {
- ht_bin[2]=ht_bin[3]=-1;
+ struct ht_entry *ht_bin = hash_table_get(vaddr);
+ if (ht_bin->vaddr[1] == vaddr) {
+ ht_bin->vaddr[1] = -1;
+ ht_bin->tcaddr[1] = NULL;
}
- if(ht_bin[0]==vaddr) {
- ht_bin[0]=ht_bin[2];
- ht_bin[1]=ht_bin[3];
- ht_bin[2]=ht_bin[3]=-1;
+ if (ht_bin->vaddr[0] == vaddr) {
+ ht_bin->vaddr[0] = ht_bin->vaddr[1];
+ ht_bin->tcaddr[0] = ht_bin->tcaddr[1];
+ ht_bin->vaddr[1] = -1;
+ ht_bin->tcaddr[1] = NULL;
}
}
-void ll_remove_matching_addrs(struct ll_entry **head,int addr,int shift)
+static void ll_remove_matching_addrs(struct ll_entry **head,
+ uintptr_t base_offs_s, int shift)
{
struct ll_entry *next;
while(*head) {
- if(((u_int)((*head)->addr)>>shift)==(addr>>shift) ||
- ((u_int)((*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 %x (%x)\n",(int)(*head)->addr,(*head)->vaddr);
+ inv_debug("EXP: Remove pointer to %p (%x)\n",(*head)->addr,(*head)->vaddr);
remove_hash((*head)->vaddr);
next=(*head)->next;
free(*head);
}
// Remove all entries from linked list
-void ll_clear(struct ll_entry **head)
+static void ll_clear(struct ll_entry **head)
{
struct ll_entry *cur;
struct ll_entry *next;
- if(cur=*head) {
+ if((cur=*head)) {
*head=0;
while(cur) {
next=cur->next;
}
// Dereference the pointers and remove if it matches
-void ll_kill_pointers(struct ll_entry *head,int addr,int shift)
+static void ll_kill_pointers(struct ll_entry *head,
+ uintptr_t base_offs_s, int shift)
{
while(head) {
- int ptr=get_pointer(head->addr);
- inv_debug("EXP: Lookup pointer to %x at %x (%x)\n",(int)ptr,(int)head->addr,head->vaddr);
- if(((ptr>>shift)==(addr>>shift)) ||
- (((ptr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift)))
+ 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 %x (%x)\n",(int)head->addr,head->vaddr);
- u_int host_addr=(u_int)kill_pointer(head->addr);
- #ifdef __arm__
- needs_clear_cache[(host_addr-(u_int)BASE_ADDR)>>17]|=1<<(((host_addr-(u_int)BASE_ADDR)>>12)&31);
- #endif
+ inv_debug("EXP: Kill pointer at %p (%x)\n",head->addr,head->vaddr);
+ void *host_addr=find_extjump_insn(head->addr);
+ mark_clear_cache(host_addr);
+ set_jump_target(host_addr, head->addr);
}
head=head->next;
}
}
// This is called when we write to a compiled block (see do_invstub)
-void invalidate_page(u_int page)
+static void invalidate_page(u_int page)
{
struct ll_entry *head;
struct ll_entry *next;
head=jump_out[page];
jump_out[page]=0;
while(head!=NULL) {
- inv_debug("INVALIDATE: kill pointer to %x (%x)\n",head->vaddr,(int)head->addr);
- u_int host_addr=(u_int)kill_pointer(head->addr);
- #ifdef __arm__
- needs_clear_cache[(host_addr-(u_int)BASE_ADDR)>>17]|=1<<(((host_addr-(u_int)BASE_ADDR)>>12)&31);
- #endif
+ 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); // point back to dyna_linker
next=head->next;
free(head);
head=next;
for(first=page+1;first<last;first++) {
invalidate_page(first);
}
- #ifdef __arm__
- do_clear_cache();
- #endif
-
+ do_clear_cache();
+
// Don't trap writes
invalid_code[block]=1;
-#ifndef DISABLE_TLB
- // If there is a valid TLB entry for this page, remove write protect
- if(tlb_LUT_w[block]) {
- assert(tlb_LUT_r[block]==tlb_LUT_w[block]);
- // CHECK: Is this right?
- memory_map[block]=((tlb_LUT_w[block]&0xFFFFF000)-(block<<12)+(unsigned int)rdram-0x80000000)>>2;
- u_int real_block=tlb_LUT_w[block]>>12;
- invalid_code[real_block]=1;
- if(real_block>=0x80000&&real_block<0x80800) memory_map[real_block]=((u_int)rdram-0x80000000)>>2;
- }
- else if(block>=0x80000&&block<0x80800) memory_map[block]=((u_int)rdram-0x80000000)>>2;
-#endif
#ifdef USE_MINI_HT
memset(mini_ht,-1,sizeof(mini_ht));
head=jump_dirty[vpage];
//printf("page=%d vpage=%d\n",page,vpage);
while(head!=NULL) {
- u_int start,end;
if(vpage>2047||(head->vaddr>>12)==block) { // Ignore vaddr hash collision
- get_bounds((int)head->addr,&start,&end);
- //printf("start: %x end: %x\n",start,end);
- if(page<2048&&start>=0x80000000&&end<0x80000000+RAM_SIZE) {
- if(((start-(u_int)rdram)>>12)<=page&&((end-1-(u_int)rdram)>>12)>=page) {
- if((((start-(u_int)rdram)>>12)&2047)<first) first=((start-(u_int)rdram)>>12)&2047;
- if((((end-1-(u_int)rdram)>>12)&2047)>last) last=((end-1-(u_int)rdram)>>12)&2047;
- }
- }
-#ifndef DISABLE_TLB
- if(page<2048&&(signed int)start>=(signed int)0xC0000000&&(signed int)end>=(signed int)0xC0000000) {
- if(((start+memory_map[start>>12]-(u_int)rdram)>>12)<=page&&((end-1+memory_map[(end-1)>>12]-(u_int)rdram)>>12)>=page) {
- if((((start+memory_map[start>>12]-(u_int)rdram)>>12)&2047)<first) first=((start+memory_map[start>>12]-(u_int)rdram)>>12)&2047;
- if((((end-1+memory_map[(end-1)>>12]-(u_int)rdram)>>12)&2047)>last) last=((end-1+memory_map[(end-1)>>12]-(u_int)rdram)>>12)&2047;
+ u_char *start, *end;
+ get_bounds(head->addr, &start, &end);
+ //printf("start: %p end: %p\n", start, end);
+ if (page < 2048 && start >= rdram && end < rdram+RAM_SIZE) {
+ if (((start-rdram)>>12) <= page && ((end-1-rdram)>>12) >= page) {
+ if ((((start-rdram)>>12)&2047) < first) first = ((start-rdram)>>12)&2047;
+ if ((((end-1-rdram)>>12)&2047) > last) last = ((end-1-rdram)>>12)&2047;
}
}
-#endif
}
head=head->next;
}
void invalidate_addr(u_int addr)
{
-#ifdef PCSX
//static int rhits;
// this check is done by the caller
//if (inv_code_start<=addr&&addr<=inv_code_end) { rhits++; return; }
- u_int page=get_page(addr);
+ u_int page=get_vpage(addr);
if(page<2048) { // RAM
struct ll_entry *head;
u_int addr_min=~0, addr_max=0;
- int mask=RAM_SIZE-1;
+ u_int mask=RAM_SIZE-1;
+ u_int addr_main=0x80000000|(addr&mask);
int pg1;
- inv_code_start=addr&~0xfff;
- inv_code_end=addr|0xfff;
+ inv_code_start=addr_main&~0xfff;
+ inv_code_end=addr_main|0xfff;
pg1=page;
if (pg1>0) {
// must check previous page too because of spans..
}
for(;pg1<=page;pg1++) {
for(head=jump_dirty[pg1];head!=NULL;head=head->next) {
- u_int start,end;
- get_bounds((int)head->addr,&start,&end);
- if((start&mask)<=(addr&mask)&&(addr&mask)<(end&mask)) {
+ u_char *start_h, *end_h;
+ u_int start, end;
+ get_bounds(head->addr, &start_h, &end_h);
+ start = (uintptr_t)start_h - ram_offset;
+ end = (uintptr_t)end_h - ram_offset;
+ if(start<=addr_main&&addr_main<end) {
if(start<addr_min) addr_min=start;
if(end>addr_max) addr_max=end;
}
- else if(addr<start) {
+ else if(addr_main<start) {
if(start<inv_code_end)
inv_code_end=start-1;
}
return;
}
else {
- inv_debug("INV ADDR: %08x miss, inv %08x-%08x, sk %d\n", addr, inv_code_start, inv_code_end, 0);//rhits);
- }
- //rhits=0;
- if(page!=0) // FIXME: don't know what's up with page 0 (Klonoa)
+ inv_code_start=(addr&~mask)|(inv_code_start&mask);
+ inv_code_end=(addr&~mask)|(inv_code_end&mask);
+ inv_debug("INV ADDR: %08x miss, inv %08x-%08x, sk %d\n", addr, inv_code_start, inv_code_end, 0);
return;
+ }
}
-#endif
invalidate_block(addr>>12);
}
// This is called when loading a save state.
// Anything could have changed, so invalidate everything.
-void invalidate_all_pages()
+void invalidate_all_pages(void)
{
- u_int page,n;
+ u_int page;
for(page=0;page<4096;page++)
invalidate_page(page);
for(page=0;page<1048576;page++)
restore_candidate[(page&2047)>>3]|=1<<(page&7);
restore_candidate[((page&2047)>>3)+256]|=1<<(page&7);
}
- #ifdef __arm__
- __clear_cache((void *)BASE_ADDR,(void *)BASE_ADDR+(1<<TARGET_SIZE_2));
- #endif
#ifdef USE_MINI_HT
memset(mini_ht,-1,sizeof(mini_ht));
#endif
- #ifndef DISABLE_TLB
- // TLB
- for(page=0;page<0x100000;page++) {
- if(tlb_LUT_r[page]) {
- memory_map[page]=((tlb_LUT_r[page]&0xFFFFF000)-(page<<12)+(unsigned int)rdram-0x80000000)>>2;
- if(!tlb_LUT_w[page]||!invalid_code[page])
- memory_map[page]|=0x40000000; // Write protect
- }
- else memory_map[page]=-1;
- if(page==0x80000) page=0xC0000;
- }
- tlb_hacks();
- #endif
+ do_clear_cache();
+}
+
+static void do_invstub(int n)
+{
+ literal_pool(20);
+ u_int reglist=stubs[n].a;
+ set_jump_target(stubs[n].addr, out);
+ save_regs(reglist);
+ if(stubs[n].b!=0) emit_mov(stubs[n].b,0);
+ emit_far_call(invalidate_addr);
+ restore_regs(reglist);
+ emit_jmp(stubs[n].retaddr); // return address
}
// Add an entry to jump_out after making a link
-void add_link(u_int vaddr,void *src)
+// src should point to code by emit_extjump2()
+void add_jump_out(u_int vaddr,void *src)
{
u_int page=get_page(vaddr);
- inv_debug("add_link: %x -> %x (%d)\n",(int)src,vaddr,page);
- int *ptr=(int *)(src+4);
- assert((*ptr&0x0fff0000)==0x059f0000);
+ inv_debug("add_jump_out: %p -> %x (%d)\n",src,vaddr,page);
+ check_extjump2(src);
ll_add(jump_out+page,vaddr,src);
- //int ptr=get_pointer(src);
- //inv_debug("add_link: Pointer is to %x\n",(int)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
while(head!=NULL) {
if(!invalid_code[head->vaddr>>12]) {
// 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))) {
- u_int start,end;
- if(verify_dirty((int)head->addr)) {
- //printf("Possibly Restore %x (%x)\n",head->vaddr, (int)head->addr);
+ if (doesnt_expire_soon(head->addr)) {
+ if(verify_dirty(head->addr)) {
+ u_char *start, *end;
+ //printf("Possibly Restore %x (%p)\n",head->vaddr, head->addr);
u_int i;
u_int inv=0;
- get_bounds((int)head->addr,&start,&end);
- if(start-(u_int)rdram<RAM_SIZE) {
- for(i=(start-(u_int)rdram+0x80000000)>>12;i<=(end-1-(u_int)rdram+0x80000000)>>12;i++) {
+ get_bounds(head->addr, &start, &end);
+ if (start - rdram < RAM_SIZE) {
+ for (i = (start-rdram+0x80000000)>>12; i <= (end-1-rdram+0x80000000)>>12; i++) {
inv|=invalid_code[i];
}
}
- if((signed int)head->vaddr>=(signed int)0xC0000000) {
- u_int addr = (head->vaddr+(memory_map[head->vaddr>>12]<<2));
- //printf("addr=%x start=%x end=%x\n",addr,start,end);
- if(addr<start||addr>=end) inv=1;
- }
else if((signed int)head->vaddr>=(signed int)0x80000000+RAM_SIZE) {
inv=1;
}
if(!inv) {
- void * clean_addr=(void *)get_clean_addr((int)head->addr);
- if((((u_int)clean_addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) {
+ void *clean_addr = get_clean_addr(head->addr);
+ if (doesnt_expire_soon(clean_addr)) {
u_int ppage=page;
-#ifndef DISABLE_TLB
- if(page<2048&&tlb_LUT_r[head->vaddr>>12]) ppage=(tlb_LUT_r[head->vaddr>>12]^0x80000000)>>12;
-#endif
- inv_debug("INV: Restored %x (%x/%x)\n",head->vaddr, (int)head->addr, (int)clean_addr);
+ inv_debug("INV: Restored %x (%p/%p)\n",head->vaddr, head->addr, 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);
- 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
- }
- }
+ ll_add_flags(jump_in+ppage,head->vaddr,head->reg_sv_flags,clean_addr);
+ struct ht_entry *ht_bin = hash_table_get(head->vaddr);
+ if (ht_bin->vaddr[0] == head->vaddr)
+ ht_bin->tcaddr[0] = clean_addr; // Replace existing entry
+ if (ht_bin->vaddr[1] == head->vaddr)
+ ht_bin->tcaddr[1] = clean_addr; // Replace existing entry
}
}
}
}
}
+/* Register allocation */
-void mov_alloc(struct regstat *current,int i)
+// Note: registers are allocated clean (unmodified state)
+// if you intend to modify the register, you must call dirty_reg().
+static void alloc_reg(struct regstat *cur,int i,signed char reg)
{
- // Note: Don't need to actually alloc the source registers
- if((~current->is32>>rs1[i])&1) {
- //alloc_reg64(current,i,rs1[i]);
- alloc_reg64(current,i,rt1[i]);
- current->is32&=~(1LL<<rt1[i]);
- } else {
- //alloc_reg(current,i,rs1[i]);
- alloc_reg(current,i,rt1[i]);
- current->is32|=(1LL<<rt1[i]);
+ int r,hr;
+ 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);
+ assert(reg >= 0);
+
+ // Don't allocate unused registers
+ if((cur->u>>reg)&1) return;
+
+ // see if it's already allocated
+ if (get_reg(cur->regmap, reg) >= 0)
+ return;
+
+ // Keep the same mapping if the register was already allocated in a loop
+ preferred_reg = loop_reg(i,reg,preferred_reg);
+
+ // Try to allocate the preferred register
+ if(cur->regmap[preferred_reg]==-1) {
+ cur->regmap[preferred_reg]=reg;
+ cur->dirty&=~(1<<preferred_reg);
+ cur->isconst&=~(1<<preferred_reg);
+ return;
+ }
+ r=cur->regmap[preferred_reg];
+ assert(r < 64);
+ if((cur->u>>r)&1) {
+ cur->regmap[preferred_reg]=reg;
+ cur->dirty&=~(1<<preferred_reg);
+ cur->isconst&=~(1<<preferred_reg);
+ return;
+ }
+
+ // Clear any unneeded registers
+ // We try to keep the mapping consistent, if possible, because it
+ // makes branches easier (especially loops). So we try to allocate
+ // first (see above) before removing old mappings. If this is not
+ // possible then go ahead and clear out the registers that are no
+ // longer needed.
+ for(hr=0;hr<HOST_REGS;hr++)
+ {
+ r=cur->regmap[hr];
+ if(r>=0) {
+ assert(r < 64);
+ 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 = 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<<hr);
+ cur->isconst&=~(1<<hr);
+ return;
+ }
+ }
+ hr++;
+ if (hr == EXCLUDE_REG)
+ hr++;
+ if (hr == HOST_REGS)
+ hr = 0;
+ if (hr == PREFERRED_REG_FIRST)
+ break;
+ }
+ }
+
+ // Try to allocate any available register
+ for (hr = PREFERRED_REG_FIRST; ; ) {
+ if (cur->regmap[hr] < 0) {
+ cur->regmap[hr]=reg;
+ cur->dirty&=~(1<<hr);
+ cur->isconst&=~(1<<hr);
+ return;
+ }
+ hr++;
+ if (hr == EXCLUDE_REG)
+ hr++;
+ if (hr == HOST_REGS)
+ hr = 0;
+ if (hr == PREFERRED_REG_FIRST)
+ break;
+ }
+
+ // Ok, now we have to evict someone
+ // Pick a register we hopefully won't need soon
+ u_char hsn[MAXREG+1];
+ memset(hsn,10,sizeof(hsn));
+ int j;
+ lsn(hsn,i,&preferred_reg);
+ //printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",cur->regmap[0],cur->regmap[1],cur->regmap[2],cur->regmap[3],cur->regmap[5],cur->regmap[6],cur->regmap[7]);
+ //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+ if(i>0) {
+ // Don't evict the cycle count at entry points, otherwise the entry
+ // stub will have to write it.
+ 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
+ if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
+ for(hr=0;hr<HOST_REGS;hr++) {
+ // Evict both parts of a 64-bit register
+ if(cur->regmap[hr]==r) {
+ cur->regmap[hr]=-1;
+ cur->dirty&=~(1<<hr);
+ cur->isconst&=~(1<<hr);
+ }
+ }
+ cur->regmap[preferred_reg]=reg;
+ return;
+ }
+ for(r=1;r<=MAXREG;r++)
+ {
+ 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;hr<HOST_REGS;hr++) {
+ if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+ if(cur->regmap[hr]==r) {
+ cur->regmap[hr]=reg;
+ cur->dirty&=~(1<<hr);
+ cur->isconst&=~(1<<hr);
+ return;
+ }
+ }
+ }
+ }
+ }
+ }
+ }
+ for(j=10;j>=0;j--)
+ {
+ for(r=1;r<=MAXREG;r++)
+ {
+ if(hsn[r]==j) {
+ for(hr=0;hr<HOST_REGS;hr++) {
+ if(cur->regmap[hr]==r) {
+ cur->regmap[hr]=reg;
+ cur->dirty&=~(1<<hr);
+ cur->isconst&=~(1<<hr);
+ return;
+ }
+ }
+ }
+ }
}
- clear_const(current,rs1[i]);
- clear_const(current,rt1[i]);
- dirty_reg(current,rt1[i]);
+ SysPrintf("This shouldn't happen (alloc_reg)");abort();
}
-void shiftimm_alloc(struct regstat *current,int i)
+// Allocate a temporary register. This is done without regard to
+// dirty status or whether the register we request is on the unneeded list
+// Note: This will only allocate one register, even if called multiple times
+static void alloc_reg_temp(struct regstat *cur,int i,signed char reg)
{
- clear_const(current,rs1[i]);
- clear_const(current,rt1[i]);
- if(opcode2[i]<=0x3) // SLL/SRL/SRA
+ int r,hr;
+ int preferred_reg = -1;
+
+ // see if it's already allocated
+ for(hr=0;hr<HOST_REGS;hr++)
{
- 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]);
- current->is32|=1LL<<rt1[i];
- dirty_reg(current,rt1[i]);
+ if(hr!=EXCLUDE_REG&&cur->regmap[hr]==reg) return;
+ }
+
+ // Try to allocate any available register
+ for(hr=HOST_REGS-1;hr>=0;hr--) {
+ if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+ cur->regmap[hr]=reg;
+ cur->dirty&=~(1<<hr);
+ cur->isconst&=~(1<<hr);
+ return;
}
}
- if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA
+
+ // Find an unneeded register
+ for(hr=HOST_REGS-1;hr>=0;hr--)
{
- if(rt1[i]) {
- if(rs1[i]) alloc_reg64(current,i,rs1[i]);
- alloc_reg64(current,i,rt1[i]);
- current->is32&=~(1LL<<rt1[i]);
- dirty_reg(current,rt1[i]);
+ r=cur->regmap[hr];
+ if(r>=0) {
+ assert(r < 64);
+ if((cur->u>>r)&1) {
+ if(i==0||((unneeded_reg[i-1]>>r)&1)) {
+ cur->regmap[hr]=reg;
+ cur->dirty&=~(1<<hr);
+ cur->isconst&=~(1<<hr);
+ return;
+ }
+ }
}
}
- if(opcode2[i]==0x3c) // DSLL32
- {
- if(rt1[i]) {
- if(rs1[i]) alloc_reg(current,i,rs1[i]);
- alloc_reg64(current,i,rt1[i]);
- current->is32&=~(1LL<<rt1[i]);
- dirty_reg(current,rt1[i]);
+
+ // Ok, now we have to evict someone
+ // Pick a register we hopefully won't need soon
+ // TODO: we might want to follow unconditional jumps here
+ // TODO: get rid of dupe code and make this into a function
+ u_char hsn[MAXREG+1];
+ memset(hsn,10,sizeof(hsn));
+ int j;
+ lsn(hsn,i,&preferred_reg);
+ //printf("hsn: %d %d %d %d %d %d %d\n",hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+ if(i>0) {
+ // Don't evict the cycle count at entry points, otherwise the entry
+ // stub will have to write it.
+ 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!=dops[i-1].rs1&&r!=dops[i-1].rs2&&r!=dops[i-1].rt1&&r!=dops[i-1].rt2) {
+ for(hr=0;hr<HOST_REGS;hr++) {
+ if(hr!=HOST_CCREG||hsn[CCREG]>2) {
+ if(cur->regmap[hr]==r) {
+ cur->regmap[hr]=reg;
+ cur->dirty&=~(1<<hr);
+ cur->isconst&=~(1<<hr);
+ return;
+ }
+ }
+ }
+ }
+ }
}
}
- if(opcode2[i]==0x3e) // DSRL32
+ for(j=10;j>=0;j--)
{
- if(rt1[i]) {
- alloc_reg64(current,i,rs1[i]);
- if(imm[i]==32) {
- alloc_reg64(current,i,rt1[i]);
- current->is32&=~(1LL<<rt1[i]);
- } else {
- alloc_reg(current,i,rt1[i]);
- current->is32|=1LL<<rt1[i];
+ for(r=1;r<=MAXREG;r++)
+ {
+ if(hsn[r]==j) {
+ for(hr=0;hr<HOST_REGS;hr++) {
+ if(cur->regmap[hr]==r) {
+ cur->regmap[hr]=reg;
+ cur->dirty&=~(1<<hr);
+ cur->isconst&=~(1<<hr);
+ return;
+ }
+ }
}
- dirty_reg(current,rt1[i]);
}
}
- if(opcode2[i]==0x3f) // DSRA32
+ SysPrintf("This shouldn't happen");abort();
+}
+
+static void mov_alloc(struct regstat *current,int i)
+{
+ if (dops[i].rs1 == HIREG || dops[i].rs1 == LOREG) {
+ alloc_cc(current,i); // for stalls
+ dirty_reg(current,CCREG);
+ }
+
+ // Note: Don't need to actually alloc the source registers
+ //alloc_reg(current,i,dops[i].rs1);
+ alloc_reg(current,i,dops[i].rt1);
+
+ 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(dops[i].opcode2<=0x3) // SLL/SRL/SRA
{
- if(rt1[i]) {
- alloc_reg64(current,i,rs1[i]);
- alloc_reg(current,i,rt1[i]);
- current->is32|=1LL<<rt1[i];
- dirty_reg(current,rt1[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].use_lt1=!!dops[i].rs1;
+ alloc_reg(current,i,dops[i].rt1);
+ dirty_reg(current,dops[i].rt1);
+ if(is_const(current,dops[i].rs1)) {
+ 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==0x02) set_const(current,dops[i].rt1,(u_int)v>>imm[i]);
+ if(dops[i].opcode2==0x03) set_const(current,dops[i].rt1,v>>imm[i]);
+ }
+ else clear_const(current,dops[i].rt1);
}
}
+ else
+ {
+ clear_const(current,dops[i].rs1);
+ clear_const(current,dops[i].rt1);
+ }
+
+ if(dops[i].opcode2>=0x38&&dops[i].opcode2<=0x3b) // DSLL/DSRL/DSRA
+ {
+ assert(0);
+ }
+ if(dops[i].opcode2==0x3c) // DSLL32
+ {
+ assert(0);
+ }
+ if(dops[i].opcode2==0x3e) // DSRL32
+ {
+ assert(0);
+ }
+ if(dops[i].opcode2==0x3f) // DSRA32
+ {
+ assert(0);
+ }
}
-void shift_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;
}
- current->is32|=1LL<<rt1[i];
} else { // DSLLV/DSRLV/DSRAV
- if(rs1[i]) alloc_reg64(current,i,rs1[i]);
- if(rs2[i]) alloc_reg(current,i,rs2[i]);
- alloc_reg64(current,i,rt1[i]);
- current->is32&=~(1LL<<rt1[i]);
- if(opcode2[i]==0x16||opcode2[i]==0x17) // DSRLV and DSRAV need a temporary register
- {
- alloc_reg_temp(current,i,-1);
- minimum_free_regs[i]=1;
- }
+ 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);
}
}
-void alu_alloc(struct regstat *current,int i)
+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);
}
- current->is32|=1LL<<rt1[i];
}
- if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU
- if(rt1[i]) {
- if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
- {
- alloc_reg64(current,i,rs1[i]);
- alloc_reg64(current,i,rs2[i]);
- alloc_reg(current,i,rt1[i]);
- } else {
- 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);
}
- current->is32|=1LL<<rt1[i];
}
- 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]);
- }
- alloc_reg(current,i,rt1[i]);
- if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
- {
- if(!((current->uu>>rt1[i])&1)) {
- alloc_reg64(current,i,rt1[i]);
- }
- if(get_reg(current->regmap,rt1[i]|64)>=0) {
- if(rs1[i]&&rs2[i]) {
- alloc_reg64(current,i,rs1[i]);
- alloc_reg64(current,i,rs2[i]);
- }
- else
- {
- // Is is really worth it to keep 64-bit values in registers?
- #ifdef NATIVE_64BIT
- if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg64(current,i,rs1[i]);
- if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg64(current,i,rs2[i]);
- #endif
- }
- }
- current->is32&=~(1LL<<rt1[i]);
- } else {
- current->is32|=1LL<<rt1[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,dops[i].rt1);
}
}
- if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU
- if(rt1[i]) {
- if(rs1[i]&&rs2[i]) {
- if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
- alloc_reg64(current,i,rs1[i]);
- alloc_reg64(current,i,rs2[i]);
- alloc_reg64(current,i,rt1[i]);
- } else {
- alloc_reg(current,i,rs1[i]);
- alloc_reg(current,i,rs2[i]);
- alloc_reg(current,i,rt1[i]);
- }
- }
- else {
- alloc_reg(current,i,rt1[i]);
- if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
- // DADD used as move, or zeroing
- // If we have a 64-bit source, then make the target 64 bits too
- if(rs1[i]&&!((current->is32>>rs1[i])&1)) {
- if(get_reg(current->regmap,rs1[i])>=0) alloc_reg64(current,i,rs1[i]);
- alloc_reg64(current,i,rt1[i]);
- } else if(rs2[i]&&!((current->is32>>rs2[i])&1)) {
- if(get_reg(current->regmap,rs2[i])>=0) alloc_reg64(current,i,rs2[i]);
- alloc_reg64(current,i,rt1[i]);
- }
- if(opcode2[i]>=0x2e&&rs2[i]) {
- // DSUB used as negation - 64-bit result
- // If we have a 32-bit register, extend it to 64 bits
- if(get_reg(current->regmap,rs2[i])>=0) alloc_reg64(current,i,rs2[i]);
- alloc_reg64(current,i,rt1[i]);
- }
- }
- }
- if(rs1[i]&&rs2[i]) {
- current->is32&=~(1LL<<rt1[i]);
- } else if(rs1[i]) {
- current->is32&=~(1LL<<rt1[i]);
- if((current->is32>>rs1[i])&1)
- current->is32|=1LL<<rt1[i];
- } else if(rs2[i]) {
- current->is32&=~(1LL<<rt1[i]);
- if((current->is32>>rs2[i])&1)
- current->is32|=1LL<<rt1[i];
- } else {
- current->is32|=1LL<<rt1[i];
- }
- }
+ 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);
}
-void imm16_alloc(struct regstat *current,int i)
+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
- current->is32&=~(1LL<<rt1[i]);
- if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
- // TODO: Could preserve the 32-bit flag if the immediate is zero
- alloc_reg64(current,i,rt1[i]);
- alloc_reg64(current,i,rs1[i]);
- }
- clear_const(current,rs1[i]);
- clear_const(current,rt1[i]);
+ if(dops[i].rs1&&needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1);
+ else dops[i].use_lt1=!!dops[i].rs1;
+ if(dops[i].rt1) alloc_reg(current,i,dops[i].rt1);
+ if(dops[i].opcode==0x18||dops[i].opcode==0x19) { // DADDI/DADDIU
+ assert(0);
}
- else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU
- if((~current->is32>>rs1[i])&1) alloc_reg64(current,i,rs1[i]);
- current->is32|=1LL<<rt1[i];
- 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(((~current->is32>>rs1[i])&1)&&opcode[i]>0x0c) {
- if(rs1[i]!=rt1[i]) {
- if(needed_again(rs1[i],i)) alloc_reg64(current,i,rs1[i]);
- alloc_reg64(current,i,rt1[i]);
- current->is32&=~(1LL<<rt1[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 current->is32|=1LL<<rt1[i]; // ANDI clears upper bits
- 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 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]);
- current->is32|=1LL<<rt1[i];
+ else clear_const(current,dops[i].rt1);
}
else {
- set_const(current,rt1[i],((long long)((short)imm[i]))<<16); // LUI
- current->is32|=1LL<<rt1[i];
+ set_const(current,dops[i].rt1,imm[i]<<16); // LUI
}
- dirty_reg(current,rt1[i]);
+ dirty_reg(current,dops[i].rt1);
}
-void load_alloc(struct regstat *current,int i)
+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
{
- current->is32&=~(1LL<<rt1[i]);
- alloc_reg64(current,i,rt1[i]);
+ assert(0);
}
- else if(opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR
+ else if(dops[i].opcode==0x1A||dops[i].opcode==0x1B) // LDL/LDR
{
- current->is32&=~(1LL<<rt1[i]);
- alloc_reg64(current,i,rt1[i]);
- alloc_all(current,i);
- alloc_reg64(current,i,FTEMP);
- minimum_free_regs[i]=HOST_REGS;
+ assert(0);
}
- else current->is32|=1LL<<rt1[i];
- dirty_reg(current,rt1[i]);
- // If using TLB, need a register for pointer to the mapping table
- if(using_tlb) alloc_reg(current,i,TLREG);
+ 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);
{
// 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
}
- // If using TLB, need a register for pointer to the mapping table
- if(using_tlb) alloc_reg(current,i,TLREG);
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
{
- alloc_all(current,i);
- alloc_reg64(current,i,FTEMP);
- minimum_free_regs[i]=HOST_REGS;
+ assert(0);
}
}
}
-void store_alloc(struct regstat *current,int i)
+static 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
- alloc_reg64(current,i,rs2[i]);
- if(rs2[i]) alloc_reg(current,i,FTEMP);
- }
- // If using TLB, need a register for pointer to the mapping table
- if(using_tlb) alloc_reg(current,i,TLREG);
+ 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
minimum_free_regs[i]=1;
}
-void c1ls_alloc(struct regstat *current,int i)
+static 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
- alloc_reg64(current,i,FTEMP);
- }
- // If using TLB, need a register for pointer to the mapping table
- if(using_tlb) alloc_reg(current,i,TLREG);
- #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)
+static 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 using TLB, need a register for pointer to the mapping table
- if(using_tlb) alloc_reg(current,i,TLREG);
+ 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 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
}
#ifndef multdiv_alloc
-void multdiv_alloc(struct regstat *current,int i)
+static void multdiv_alloc(struct regstat *current,int i)
{
// case 0x18: MULT
// case 0x19: MULTU
// case 0x1D: DMULTU
// case 0x1E: DDIV
// case 0x1F: DDIVU
- clear_const(current,rs1[i]);
- clear_const(current,rs2[i]);
- if(rs1[i]&&rs2[i])
+ clear_const(current,dops[i].rs1);
+ clear_const(current,dops[i].rs2);
+ alloc_cc(current,i); // for stalls
+ 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<<HIREG);
current->u&=~(1LL<<LOREG);
alloc_reg(current,i,HIREG);
alloc_reg(current,i,LOREG);
- alloc_reg(current,i,rs1[i]);
- alloc_reg(current,i,rs2[i]);
- current->is32|=1LL<<HIREG;
- current->is32|=1LL<<LOREG;
+ alloc_reg(current,i,dops[i].rs1);
+ alloc_reg(current,i,dops[i].rs2);
dirty_reg(current,HIREG);
dirty_reg(current,LOREG);
}
else // 64-bit
{
- current->u&=~(1LL<<HIREG);
- current->u&=~(1LL<<LOREG);
- current->uu&=~(1LL<<HIREG);
- current->uu&=~(1LL<<LOREG);
- alloc_reg64(current,i,HIREG);
- //if(HOST_REGS>10) alloc_reg64(current,i,LOREG);
- alloc_reg64(current,i,rs1[i]);
- alloc_reg64(current,i,rs2[i]);
- alloc_all(current,i);
- current->is32&=~(1LL<<HIREG);
- current->is32&=~(1LL<<LOREG);
- dirty_reg(current,HIREG);
- dirty_reg(current,LOREG);
- minimum_free_regs[i]=HOST_REGS;
+ assert(0);
}
}
else
// The result is undefined, we return zero.
alloc_reg(current,i,HIREG);
alloc_reg(current,i,LOREG);
- current->is32|=1LL<<HIREG;
- current->is32|=1LL<<LOREG;
dirty_reg(current,HIREG);
dirty_reg(current,LOREG);
}
}
#endif
-void cop0_alloc(struct regstat *current,int i)
+static void cop0_alloc(struct regstat *current,int i)
{
- if(opcode2[i]==0) // MFC0
+ if(dops[i].opcode2==0) // MFC0
{
- if(rt1[i]) {
- clear_const(current,rt1[i]);
+ if(dops[i].rt1) {
+ clear_const(current,dops[i].rt1);
alloc_all(current,i);
- alloc_reg(current,i,rt1[i]);
- current->is32|=1LL<<rt1[i];
- dirty_reg(current,rt1[i]);
+ alloc_reg(current,i,dops[i].rt1);
+ dirty_reg(current,dops[i].rt1);
}
}
- else if(opcode2[i]==4) // MTC0
+ else if(dops[i].opcode2==4) // MTC0
{
- 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);
alloc_all(current,i);
}
else {
else
{
// TLBR/TLBWI/TLBWR/TLBP/ERET
- assert(opcode2[i]==0x10);
+ assert(dops[i].opcode2==0x10);
alloc_all(current,i);
}
minimum_free_regs[i]=HOST_REGS;
}
-void cop1_alloc(struct regstat *current,int i)
+static void cop2_alloc(struct regstat *current,int i)
{
- alloc_reg(current,i,CSREG); // Load status
- if(opcode2[i]<3) // MFC1/DMFC1/CFC1
+ if (dops[i].opcode2 < 3) // MFC2/CFC2
{
- if(rt1[i]){
- clear_const(current,rt1[i]);
- if(opcode2[i]==1) {
- alloc_reg64(current,i,rt1[i]); // DMFC1
- current->is32&=~(1LL<<rt1[i]);
- }else{
- alloc_reg(current,i,rt1[i]); // MFC1/CFC1
- current->is32|=1LL<<rt1[i];
- }
- dirty_reg(current,rt1[i]);
+ alloc_cc(current,i); // for stalls
+ dirty_reg(current,CCREG);
+ if(dops[i].rt1){
+ clear_const(current,dops[i].rt1);
+ alloc_reg(current,i,dops[i].rt1);
+ dirty_reg(current,dops[i].rt1);
}
- alloc_reg_temp(current,i,-1);
}
- else if(opcode2[i]>3) // MTC1/DMTC1/CTC1
+ else if (dops[i].opcode2 > 3) // MTC2/CTC2
{
- if(rs1[i]){
- clear_const(current,rs1[i]);
- if(opcode2[i]==5)
- alloc_reg64(current,i,rs1[i]); // DMTC1
- else
- alloc_reg(current,i,rs1[i]); // MTC1/CTC1
- alloc_reg_temp(current,i,-1);
+ if(dops[i].rs1){
+ clear_const(current,dops[i].rs1);
+ alloc_reg(current,i,dops[i].rs1);
}
else {
current->u&=~1LL;
alloc_reg(current,i,0);
- alloc_reg_temp(current,i,-1);
}
}
- minimum_free_regs[i]=1;
-}
-void fconv_alloc(struct regstat *current,int i)
-{
- alloc_reg(current,i,CSREG); // Load status
- alloc_reg_temp(current,i,-1);
- minimum_free_regs[i]=1;
-}
-void float_alloc(struct regstat *current,int i)
-{
- alloc_reg(current,i,CSREG); // Load status
alloc_reg_temp(current,i,-1);
minimum_free_regs[i]=1;
}
-void c2op_alloc(struct regstat *current,int i)
-{
- alloc_reg_temp(current,i,-1);
-}
-void fcomp_alloc(struct regstat *current,int i)
+
+static void c2op_alloc(struct regstat *current,int i)
{
- alloc_reg(current,i,CSREG); // Load status
- alloc_reg(current,i,FSREG); // Load flags
- dirty_reg(current,FSREG); // Flag will be modified
+ alloc_cc(current,i); // for stalls
+ dirty_reg(current,CCREG);
alloc_reg_temp(current,i,-1);
- minimum_free_regs[i]=1;
}
-void syscall_alloc(struct regstat *current,int i)
+static void syscall_alloc(struct regstat *current,int i)
{
alloc_cc(current,i);
dirty_reg(current,CCREG);
current->isconst=0;
}
-void delayslot_alloc(struct regstat *current,int i)
+static void delayslot_alloc(struct regstat *current,int i)
{
- switch(itype[i]) {
+ switch(dops[i].itype) {
case UJUMP:
case CJUMP:
case SJUMP:
case RJUMP:
- case FJUMP:
case SYSCALL:
case HLECALL:
case SPAN:
- assem_debug("jump in the delay slot. this shouldn't happen.\n");//exit(1);
- printf("Disabled speculative precompilation\n");
+ assem_debug("jump in the delay slot. this shouldn't happen.\n");//abort();
+ SysPrintf("Disabled speculative precompilation\n");
stop_after_jal=1;
break;
case IMM16:
cop0_alloc(current,i);
break;
case COP1:
+ break;
case COP2:
- cop1_alloc(current,i);
+ cop2_alloc(current,i);
break;
case C1LS:
c1ls_alloc(current,i);
case C2LS:
c2ls_alloc(current,i);
break;
- case FCONV:
- fconv_alloc(current,i);
- break;
- case FLOAT:
- float_alloc(current,i);
- break;
- case FCOMP:
- fcomp_alloc(current,i);
- break;
case C2OP:
c2op_alloc(current,i);
break;
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(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
- {
- if(rs1[i]) alloc_reg64(current,i,rs1[i]);
- if(rs2[i]) alloc_reg64(current,i,rs2[i]);
- }
- }
- else
- if((opcode[i]&0x2E)==6) // BLEZ/BGTZ/BLEZL/BGTZL
- {
- if(rs1[i]) alloc_reg(current,i,rs1[i]);
- if(!((current->is32>>rs1[i])&1))
- {
- if(rs1[i]) alloc_reg64(current,i,rs1[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]==0x11) // BC1
+ if((dops[i].opcode&0x2E)==6) // BLEZ/BGTZ/BLEZL/BGTZL
{
- alloc_reg(current,i,FSREG);
- alloc_reg(current,i,CSREG);
+ if(dops[i].rs1) alloc_reg(current,i,dops[i].rs1);
}
//else ...
}
-add_stub(int type,int addr,int retaddr,int a,int b,int c,int d,int e)
+static void add_stub(enum stub_type type, void *addr, void *retaddr,
+ u_int a, uintptr_t b, uintptr_t c, u_int d, u_int e)
{
- stubs[stubcount][0]=type;
- stubs[stubcount][1]=addr;
- stubs[stubcount][2]=retaddr;
- stubs[stubcount][3]=a;
- stubs[stubcount][4]=b;
- stubs[stubcount][5]=c;
- stubs[stubcount][6]=d;
- stubs[stubcount][7]=e;
+ assert(stubcount < ARRAY_SIZE(stubs));
+ stubs[stubcount].type = type;
+ stubs[stubcount].addr = addr;
+ stubs[stubcount].retaddr = retaddr;
+ stubs[stubcount].a = a;
+ stubs[stubcount].b = b;
+ stubs[stubcount].c = c;
+ stubs[stubcount].d = d;
+ stubs[stubcount].e = e;
stubcount++;
}
+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)
+{
+ add_stub(type, addr, retaddr, i, addr_reg, (uintptr_t)i_regs, ccadj, reglist);
+}
+
// Write out a single register
-void wb_register(signed char r,signed char regmap[],uint64_t dirty,uint64_t is32)
+static void wb_register(signed char r, const signed char regmap[], uint64_t dirty)
{
int hr;
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG) {
- if((regmap[hr]&63)==r) {
+ if(regmap[hr]==r) {
if((dirty>>hr)&1) {
- if(regmap[hr]<64) {
- emit_storereg(r,hr);
-#ifndef FORCE32
- if((is32>>regmap[hr])&1) {
- emit_sarimm(hr,31,hr);
- emit_storereg(r|64,hr);
- }
-#endif
- }else{
- emit_storereg(r|64,hr);
- }
+ assert(regmap[hr]<64);
+ emit_storereg(r,hr);
}
}
}
}
}
-int mchecksum()
+static void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty,uint64_t u)
{
- //if(!tracedebug) return 0;
- int i;
- int sum=0;
- for(i=0;i<2097152;i++) {
- unsigned int temp=sum;
- sum<<=1;
- sum|=(~temp)>>31;
- sum^=((u_int *)rdram)[i];
+ //if(dirty_pre==dirty) return;
+ int hr, r;
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ r = pre[hr];
+ if (r < 1 || r > 33 || ((u >> r) & 1))
+ continue;
+ if (((dirty_pre & ~dirty) >> hr) & 1)
+ emit_storereg(r, hr);
}
- return sum;
-}
-int rchecksum()
-{
- int i;
- int sum=0;
- for(i=0;i<64;i++)
- sum^=((u_int *)reg)[i];
- return sum;
-}
-void rlist()
-{
- int i;
- printf("TRACE: ");
- for(i=0;i<32;i++)
- printf("r%d:%8x%8x ",i,((int *)(reg+i))[1],((int *)(reg+i))[0]);
- printf("\n");
-#ifndef DISABLE_COP1
- printf("TRACE: ");
- for(i=0;i<32;i++)
- printf("f%d:%8x%8x ",i,((int*)reg_cop1_simple[i])[1],*((int*)reg_cop1_simple[i]));
- printf("\n");
-#endif
-}
-
-void enabletrace()
-{
- tracedebug=1;
}
-void memdebug(int i)
+// trashes r2
+static void pass_args(int a0, int a1)
{
- //printf("TRACE: count=%d next=%d (checksum %x) lo=%8x%8x\n",Count,next_interupt,mchecksum(),(int)(reg[LOREG]>>32),(int)reg[LOREG]);
- //printf("TRACE: count=%d next=%d (rchecksum %x)\n",Count,next_interupt,rchecksum());
- //rlist();
- //if(tracedebug) {
- //if(Count>=-2084597794) {
- if((signed int)Count>=-2084597794&&(signed int)Count<0) {
- //if(0) {
- printf("TRACE: count=%d next=%d (checksum %x)\n",Count,next_interupt,mchecksum());
- //printf("TRACE: count=%d next=%d (checksum %x) Status=%x\n",Count,next_interupt,mchecksum(),Status);
- //printf("TRACE: count=%d next=%d (checksum %x) hi=%8x%8x\n",Count,next_interupt,mchecksum(),(int)(reg[HIREG]>>32),(int)reg[HIREG]);
- rlist();
- #ifdef __i386__
- printf("TRACE: %x\n",(&i)[-1]);
- #endif
- #ifdef __arm__
- int j;
- printf("TRACE: %x \n",(&j)[10]);
- printf("TRACE: %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x\n",(&j)[1],(&j)[2],(&j)[3],(&j)[4],(&j)[5],(&j)[6],(&j)[7],(&j)[8],(&j)[9],(&j)[10],(&j)[11],(&j)[12],(&j)[13],(&j)[14],(&j)[15],(&j)[16],(&j)[17],(&j)[18],(&j)[19],(&j)[20]);
- #endif
- //fflush(stdout);
+ if(a0==1&&a1==0) {
+ // must swap
+ emit_mov(a0,2); emit_mov(a1,1); emit_mov(2,0);
+ }
+ else if(a0!=0&&a1==0) {
+ emit_mov(a1,1);
+ if (a0>=0) emit_mov(a0,0);
+ }
+ else {
+ if(a0>=0&&a0!=0) emit_mov(a0,0);
+ if(a1>=0&&a1!=1) emit_mov(a1,1);
}
- //printf("TRACE: %x\n",(&i)[-1]);
-}
-
-void tlb_debug(u_int cause, u_int addr, u_int iaddr)
-{
- printf("TLB Exception: instruction=%x addr=%x cause=%x\n",iaddr, addr, cause);
}
-void alu_assemble(int i,struct regstat *i_regs)
+static void alu_assemble(int i, const struct regstat *i_regs)
{
- if(opcode2[i]>=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(rt1[i]) {
- signed char s1l,s2l,s1h,s2h,tl,th;
- tl=get_reg(i_regs->regmap,rt1[i]);
- th=get_reg(i_regs->regmap,rt1[i]|64);
- if(tl>=0) {
- s1l=get_reg(i_regs->regmap,rs1[i]);
- s2l=get_reg(i_regs->regmap,rs2[i]);
- s1h=get_reg(i_regs->regmap,rs1[i]|64);
- s2h=get_reg(i_regs->regmap,rs2[i]|64);
- if(rs1[i]&&rs2[i]) {
- assert(s1l>=0);
- assert(s2l>=0);
- if(opcode2[i]&2) emit_subs(s1l,s2l,tl);
- else emit_adds(s1l,s2l,tl);
- if(th>=0) {
- #ifdef INVERTED_CARRY
- if(opcode2[i]&2) {if(s1h!=th) emit_mov(s1h,th);emit_sbb(th,s2h);}
- #else
- if(opcode2[i]&2) emit_sbc(s1h,s2h,th);
- #endif
- else emit_add(s1h,s2h,th);
- }
- }
- else if(rs1[i]) {
- if(s1l>=0) emit_mov(s1l,tl);
- else emit_loadreg(rs1[i],tl);
- if(th>=0) {
- if(s1h>=0) emit_mov(s1h,th);
- else emit_loadreg(rs1[i]|64,th);
- }
- }
- else if(rs2[i]) {
- if(s2l>=0) {
- if(opcode2[i]&2) emit_negs(s2l,tl);
- else emit_mov(s2l,tl);
- }
- else {
- emit_loadreg(rs2[i],tl);
- if(opcode2[i]&2) emit_negs(tl,tl);
- }
- if(th>=0) {
- #ifdef INVERTED_CARRY
- if(s2h>=0) emit_mov(s2h,th);
- else emit_loadreg(rs2[i]|64,th);
- if(opcode2[i]&2) {
- emit_adcimm(-1,th); // x86 has inverted carry flag
- emit_not(th,th);
- }
- #else
- if(opcode2[i]&2) {
- if(s2h>=0) emit_rscimm(s2h,0,th);
- else {
- emit_loadreg(rs2[i]|64,th);
- emit_rscimm(th,0,th);
- }
- }else{
- if(s2h>=0) emit_mov(s2h,th);
- else emit_loadreg(rs2[i]|64,th);
- }
- #endif
- }
- }
- else {
- emit_zeroreg(tl);
- if(th>=0) emit_zeroreg(th);
- }
- }
- }
+ 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]) {
- signed char s1l,s1h,s2l,s2h,t;
- if(!((i_regs->was32>>rs1[i])&(i_regs->was32>>rs2[i])&1))
+ 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]);
- //assert(t>=0);
- if(t>=0) {
- s1l=get_reg(i_regs->regmap,rs1[i]);
- s1h=get_reg(i_regs->regmap,rs1[i]|64);
- s2l=get_reg(i_regs->regmap,rs2[i]);
- s2h=get_reg(i_regs->regmap,rs2[i]|64);
- if(rs2[i]==0) // rx<r0
- {
- assert(s1h>=0);
- if(opcode2[i]==0x2a) // SLT
- emit_shrimm(s1h,31,t);
- else // SLTU (unsigned can not be less than zero)
- emit_zeroreg(t);
- }
- else if(rs1[i]==0) // r0<rx
- {
- assert(s2h>=0);
- if(opcode2[i]==0x2a) // SLT
- emit_set_gz64_32(s2h,s2l,t);
- else // SLTU (set if not zero)
- emit_set_nz64_32(s2h,s2l,t);
- }
- else {
- assert(s1l>=0);assert(s1h>=0);
- assert(s2l>=0);assert(s2h>=0);
- if(opcode2[i]==0x2a) // SLT
- emit_set_if_less64_32(s1h,s1l,s2h,s2l,t);
- else // SLTU
- emit_set_if_carry64_32(s1h,s1l,s2h,s2l,t);
- }
- }
- } else {
- 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) // rx<r0
+ s1l=get_reg(i_regs->regmap,dops[i].rs1);
+ s2l=get_reg(i_regs->regmap,dops[i].rs2);
+ if(dops[i].rs2==0) // rx<r0
{
- assert(s1l>=0);
- if(opcode2[i]==0x2a) // SLT
+ if(dops[i].opcode2==0x2a&&dops[i].rs1!=0) { // SLT
+ assert(s1l>=0);
emit_shrimm(s1l,31,t);
- else // SLTU (unsigned can not be less than zero)
+ }
+ else // SLTU (unsigned can not be less than zero, 0<0)
emit_zeroreg(t);
}
- else if(rs1[i]==0) // r0<rx
+ else if(dops[i].rs1==0) // r0<rx
{
assert(s2l>=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);
}
}
}
- if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR
- if(rt1[i]) {
- signed char s1l,s1h,s2l,s2h,th,tl;
- tl=get_reg(i_regs->regmap,rt1[i]);
- th=get_reg(i_regs->regmap,rt1[i]|64);
- if(!((i_regs->was32>>rs1[i])&(i_regs->was32>>rs2[i])&1)&&th>=0)
+ 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,dops[i].rt1);
{
- assert(tl>=0);
if(tl>=0) {
- s1l=get_reg(i_regs->regmap,rs1[i]);
- s1h=get_reg(i_regs->regmap,rs1[i]|64);
- s2l=get_reg(i_regs->regmap,rs2[i]);
- s2h=get_reg(i_regs->regmap,rs2[i]|64);
- if(rs1[i]&&rs2[i]) {
- assert(s1l>=0);assert(s1h>=0);
- assert(s2l>=0);assert(s2h>=0);
- if(opcode2[i]==0x24) { // AND
+ 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(dops[i].opcode2==0x24) { // AND
emit_and(s1l,s2l,tl);
- emit_and(s1h,s2h,th);
} else
- if(opcode2[i]==0x25) { // OR
+ if(dops[i].opcode2==0x25) { // OR
emit_or(s1l,s2l,tl);
- emit_or(s1h,s2h,th);
} else
- if(opcode2[i]==0x26) { // XOR
+ if(dops[i].opcode2==0x26) { // XOR
emit_xor(s1l,s2l,tl);
- emit_xor(s1h,s2h,th);
} else
- if(opcode2[i]==0x27) { // NOR
+ if(dops[i].opcode2==0x27) { // NOR
emit_or(s1l,s2l,tl);
- emit_or(s1h,s2h,th);
emit_not(tl,tl);
- emit_not(th,th);
}
}
else
{
- if(opcode2[i]==0x24) { // AND
+ if(dops[i].opcode2==0x24) { // AND
emit_zeroreg(tl);
- emit_zeroreg(th);
} 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);
- if(s1h>=0) emit_mov(s1h,th);
- else emit_loadreg(rs1[i]|64,th);
+ 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);
- if(s2h>=0) emit_mov(s2h,th);
- else emit_loadreg(rs2[i]|64,th);
- }
- else{
- emit_zeroreg(tl);
- emit_zeroreg(th);
+ 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);
+ else {
+ emit_loadreg(dops[i].rs1,tl);
emit_not(tl,tl);
}
- if(s1h>=0) emit_not(s1h,th);
- else{
- emit_loadreg(rs1[i]|64,th);
- emit_not(th,th);
- }
}
else
- if(rs2[i]){
+ if(dops[i].rs2){
if(s2l>=0) emit_not(s2l,tl);
- else{
- emit_loadreg(rs2[i],tl);
+ else {
+ emit_loadreg(dops[i].rs2,tl);
emit_not(tl,tl);
}
- if(s2h>=0) emit_not(s2h,th);
- else{
- emit_loadreg(rs2[i]|64,th);
- emit_not(th,th);
- }
- }
- else {
- emit_movimm(-1,tl);
- emit_movimm(-1,th);
}
- }
- }
- }
- }
- else
- {
- // 32 bit
- if(tl>=0) {
- s1l=get_reg(i_regs->regmap,rs1[i]);
- s2l=get_reg(i_regs->regmap,rs2[i]);
- if(rs1[i]&&rs2[i]) {
- assert(s1l>=0);
- assert(s2l>=0);
- if(opcode2[i]==0x24) { // AND
- emit_and(s1l,s2l,tl);
- } else
- if(opcode2[i]==0x25) { // OR
- emit_or(s1l,s2l,tl);
- } else
- if(opcode2[i]==0x26) { // XOR
- emit_xor(s1l,s2l,tl);
- } else
- if(opcode2[i]==0x27) { // NOR
- emit_or(s1l,s2l,tl);
- emit_not(tl,tl);
- }
- }
- else
- {
- if(opcode2[i]==0x24) { // AND
- emit_zeroreg(tl);
- } else
- if(opcode2[i]==0x25||opcode2[i]==0x26) { // OR/XOR
- if(rs1[i]){
- if(s1l>=0) emit_mov(s1l,tl);
- else emit_loadreg(rs1[i],tl); // CHECK: regmap_entry?
- }
- else
- if(rs2[i]){
- if(s2l>=0) emit_mov(s2l,tl);
- else emit_loadreg(rs2[i],tl); // CHECK: regmap_entry?
- }
- else emit_zeroreg(tl);
- } else
- if(opcode2[i]==0x27) { // NOR
- if(rs1[i]){
- if(s1l>=0) emit_not(s1l,tl);
- else {
- emit_loadreg(rs1[i],tl);
- emit_not(tl,tl);
- }
- }
- else
- if(rs2[i]){
- if(s2l>=0) emit_not(s2l,tl);
- else {
- emit_loadreg(rs2[i],tl);
- emit_not(tl,tl);
- }
- }
- else emit_movimm(-1,tl);
+ else emit_movimm(-1,tl);
}
}
}
}
}
-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))
}
}
}
- 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))
}
}
}
- if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU
- if(rt1[i]) {
- signed char sh,sl,th,tl;
- th=get_reg(i_regs->regmap,rt1[i]|64);
- tl=get_reg(i_regs->regmap,rt1[i]);
- sh=get_reg(i_regs->regmap,rs1[i]|64);
- sl=get_reg(i_regs->regmap,rs1[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,dops[i].rt1);
+ sl=get_reg(i_regs->regmap,dops[i].rs1);
if(tl>=0) {
- if(rs1[i]) {
- assert(sh>=0);
+ if(dops[i].rs1) {
assert(sl>=0);
- if(th>=0) {
- emit_addimm64_32(sh,sl,imm[i],th,tl);
- }
- else {
- emit_addimm(sl,imm[i],tl);
- }
+ emit_addimm(sl,imm[i],tl);
} else {
emit_movimm(imm[i],tl);
- if(th>=0) emit_movimm(((signed int)imm[i])>>31,th);
}
}
}
}
- 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
- signed char sh,sl,t;
- t=get_reg(i_regs->regmap,rt1[i]);
- sh=get_reg(i_regs->regmap,rs1[i]|64);
- sl=get_reg(i_regs->regmap,rs1[i]);
+ 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,dops[i].rt1);
+ sl=get_reg(i_regs->regmap,dops[i].rs1);
//assert(t>=0);
if(t>=0) {
- if(rs1[i]>0) {
- if(sh<0) assert((i_regs->was32>>rs1[i])&1);
- if(sh<0||((i_regs->was32>>rs1[i])&1)) {
- 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);
}
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);
}
}
- }else{ // 64-bit
- assert(sl>=0);
- if(opcode[i]==0x0a) // SLTI
- emit_slti64_32(sh,sl,imm[i],t);
- else // SLTIU
- emit_sltiu64_32(sh,sl,imm[i],t);
- }
}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<imm[i]) emit_movimm(1,t);
else emit_zeroreg(t);
else // SLTIU
}
}
}
- else if(opcode[i]>=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI
- if(rt1[i]) {
- signed char sh,sl,th,tl;
- th=get_reg(i_regs->regmap,rt1[i]|64);
- tl=get_reg(i_regs->regmap,rt1[i]);
- sh=get_reg(i_regs->regmap,rs1[i]|64);
- sl=get_reg(i_regs->regmap,rs1[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,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))
}
else
emit_zeroreg(tl);
- if(th>=0) emit_zeroreg(th);
}
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(th>=0) {
- if(sh<0) {
- emit_loadreg(rs1[i]|64,th);
+ if(dops[i].opcode==0x0d) { // ORI
+ if(sl<0) {
+ emit_orimm(tl,imm[i],tl);
}else{
- emit_mov(sh,th);
+ if(!((i_regs->wasconst>>sl)&1))
+ emit_orimm(sl,imm[i],tl);
+ else
+ emit_movimm(constmap[i][sl]|imm[i],tl);
}
}
- if(opcode[i]==0x0d) //ORI
- if(sl<0) {
- emit_orimm(tl,imm[i],tl);
- }else{
- if(!((i_regs->wasconst>>sl)&1))
- emit_orimm(sl,imm[i],tl);
- else
- emit_movimm(constmap[i][sl]|imm[i],tl);
- }
- if(opcode[i]==0x0e) //XORI
- if(sl<0) {
- emit_xorimm(tl,imm[i],tl);
- }else{
- if(!((i_regs->wasconst>>sl)&1))
- emit_xorimm(sl,imm[i],tl);
- else
- emit_movimm(constmap[i][sl]^imm[i],tl);
+ if(dops[i].opcode==0x0e) { // XORI
+ if(sl<0) {
+ emit_xorimm(tl,imm[i],tl);
+ }else{
+ if(!((i_regs->wasconst>>sl)&1))
+ emit_xorimm(sl,imm[i],tl);
+ else
+ emit_movimm(constmap[i][sl]^imm[i],tl);
+ }
}
}
else {
emit_movimm(imm[i],tl);
- if(th>=0) emit_zeroreg(th);
}
}
}
}
}
-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){
- if(rs1[i]==0)
+ if(t>=0&&!((i_regs->isconst>>t)&1)){
+ 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);
}
}
}
}
- //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
{
- if(rt1[i]) {
- signed char sh,sl,th,tl;
- th=get_reg(i_regs->regmap,rt1[i]|64);
- tl=get_reg(i_regs->regmap,rt1[i]);
- sh=get_reg(i_regs->regmap,rs1[i]|64);
- sl=get_reg(i_regs->regmap,rs1[i]);
- if(tl>=0) {
- if(rs1[i]==0)
- {
- emit_zeroreg(tl);
- if(th>=0) emit_zeroreg(th);
- }
- else
- {
- assert(sl>=0);
- assert(sh>=0);
- if(imm[i]) {
- if(opcode2[i]==0x38) // DSLL
- {
- if(th>=0) emit_shldimm(sh,sl,imm[i],th);
- emit_shlimm(sl,imm[i],tl);
- }
- if(opcode2[i]==0x3a) // DSRL
- {
- emit_shrdimm(sl,sh,imm[i],tl);
- if(th>=0) emit_shrimm(sh,imm[i],th);
- }
- if(opcode2[i]==0x3b) // DSRA
- {
- emit_shrdimm(sl,sh,imm[i],tl);
- if(th>=0) emit_sarimm(sh,imm[i],th);
- }
- }else{
- // Shift by zero
- if(sl!=tl) emit_mov(sl,tl);
- if(th>=0&&sh!=th) emit_mov(sh,th);
- }
- }
- }
- }
+ assert(0);
}
- if(opcode2[i]==0x3c) // DSLL32
+ if(dops[i].opcode2==0x3c) // DSLL32
{
- if(rt1[i]) {
- signed char sl,tl,th;
- tl=get_reg(i_regs->regmap,rt1[i]);
- th=get_reg(i_regs->regmap,rt1[i]|64);
- sl=get_reg(i_regs->regmap,rs1[i]);
- if(th>=0||tl>=0){
- assert(tl>=0);
- assert(th>=0);
- assert(sl>=0);
- emit_mov(sl,th);
- emit_zeroreg(tl);
- if(imm[i]>32)
- {
- emit_shlimm(th,imm[i]&31,th);
- }
- }
- }
+ assert(0);
}
- if(opcode2[i]==0x3e) // DSRL32
+ if(dops[i].opcode2==0x3e) // DSRL32
{
- if(rt1[i]) {
- signed char sh,tl,th;
- tl=get_reg(i_regs->regmap,rt1[i]);
- th=get_reg(i_regs->regmap,rt1[i]|64);
- sh=get_reg(i_regs->regmap,rs1[i]|64);
- if(tl>=0){
- assert(sh>=0);
- emit_mov(sh,tl);
- if(th>=0) emit_zeroreg(th);
- if(imm[i]>32)
- {
- emit_shrimm(tl,imm[i]&31,tl);
- }
- }
- }
+ assert(0);
}
- if(opcode2[i]==0x3f) // DSRA32
+ if(dops[i].opcode2==0x3f) // DSRA32
{
- if(rt1[i]) {
- signed char sh,tl;
- tl=get_reg(i_regs->regmap,rt1[i]);
- sh=get_reg(i_regs->regmap,rs1[i]|64);
- if(tl>=0){
- assert(sh>=0);
- emit_mov(sh,tl);
- if(imm[i]>32)
- {
- emit_sarimm(tl,imm[i]&31,tl);
- }
- }
- }
+ assert(0);
}
}
#ifndef shift_assemble
-void shift_assemble(int i,struct regstat *i_regs)
+static void shift_assemble(int i, const struct regstat *i_regs)
{
- printf("Need shift_assemble for this architecture.\n");
- exit(1);
+ signed char s,t,shift;
+ if (dops[i].rt1 == 0)
+ return;
+ 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(dops[i].rs1==0)
+ emit_zeroreg(t);
+ 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(dops[i].opcode2) {
+ case 4: // SLLV
+ emit_shl(s,HOST_TEMPREG,t);
+ break;
+ case 6: // SRLV
+ emit_shr(s,HOST_TEMPREG,t);
+ break;
+ case 7: // SRAV
+ emit_sar(s,HOST_TEMPREG,t);
+ break;
+ default:
+ assert(0);
+ }
+ host_tempreg_release();
+ }
}
+
#endif
-void load_assemble(int i,struct regstat *i_regs)
+enum {
+ MTYPE_8000 = 0,
+ MTYPE_8020,
+ MTYPE_0000,
+ MTYPE_A000,
+ MTYPE_1F80,
+};
+
+static int get_ptr_mem_type(u_int a)
+{
+ if(a < 0x00200000) {
+ if(a<0x1000&&((start>>20)==0xbfc||(start>>24)==0xa0))
+ // return wrong, must use memhandler for BIOS self-test to pass
+ // 007 does similar stuff from a00 mirror, weird stuff
+ return MTYPE_8000;
+ return MTYPE_0000;
+ }
+ if(0x1f800000 <= a && a < 0x1f801000)
+ return MTYPE_1F80;
+ if(0x80200000 <= a && a < 0x80800000)
+ return MTYPE_8020;
+ if(0xa0000000 <= a && a < 0xa0200000)
+ return MTYPE_A000;
+ return MTYPE_8000;
+}
+
+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 = 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);
+ }
+ else {
+ // use the mirror we are running on
+ type=get_ptr_mem_type(start);
+ //printf("set nospec @%08x r%d %d\n", start+i*4, mr, type);
+ }
+
+ if(type==MTYPE_8020) { // RAM 80200000+ mirror
+ host_tempreg_acquire();
+ emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
+ addr=*addr_reg_override=HOST_TEMPREG;
+ type=0;
+ }
+ else if(type==MTYPE_0000) { // RAM 0 mirror
+ host_tempreg_acquire();
+ emit_orimm(addr,0x80000000,HOST_TEMPREG);
+ addr=*addr_reg_override=HOST_TEMPREG;
+ type=0;
+ }
+ else if(type==MTYPE_A000) { // RAM A mirror
+ host_tempreg_acquire();
+ emit_andimm(addr,~0x20000000,HOST_TEMPREG);
+ addr=*addr_reg_override=HOST_TEMPREG;
+ type=0;
+ }
+ else if(type==MTYPE_1F80) { // scratchpad
+ if (psxH == (void *)0x1f800000) {
+ host_tempreg_acquire();
+ emit_xorimm(addr,0x1f800000,HOST_TEMPREG);
+ emit_cmpimm(HOST_TEMPREG,0x1000);
+ host_tempreg_release();
+ jaddr=out;
+ emit_jc(0);
+ }
+ else {
+ // do the usual RAM check, jump will go to the right handler
+ type=0;
+ }
+ }
+
+ if (type == 0) // need ram check
+ {
+ emit_cmpimm(addr,RAM_SIZE);
+ jaddr = out;
+ #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+ // Hint to branch predictor that the branch is unlikely to be taken
+ if (dops[i].rs1 >= 28)
+ emit_jno_unlikely(0);
+ else
+ #endif
+ emit_jno(0);
+ if (ram_offset != 0)
+ *offset_reg = get_ro_reg(i_regs, 0);
+ }
+
+ return jaddr;
+}
+
+// return memhandler, or get directly accessable address and return 0
+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 & msb)) {
+ uintptr_t v = l1 << 1;
+ *addr_host = v + addr;
+ return NULL;
+ }
+ else {
+ l1 <<= 1;
+ 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];
+ else
+ l2 = ((uintptr_t *)l1)[(addr&0xfff)/4];
+ if (!(l2 & msb)) {
+ uintptr_t v = l2 << 1;
+ *addr_host = v + (addr&0xfff);
+ return NULL;
+ }
+ return (void *)(l2 << 1);
+ }
+}
+
+static u_int get_host_reglist(const signed char *regmap)
+{
+ u_int reglist = 0, hr;
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ if (hr != EXCLUDE_REG && regmap[hr] >= 0)
+ reglist |= 1 << hr;
+ }
+ return reglist;
+}
+
+static u_int reglist_exclude(u_int reglist, int r1, int r2)
{
- int s,th,tl,addr,map=-1;
+ if (r1 >= 0)
+ reglist &= ~(1u << r1);
+ if (r2 >= 0)
+ reglist &= ~(1u << r2);
+ return reglist;
+}
+
+// find a temp caller-saved register not in reglist (so assumed to be free)
+static int reglist_find_free(u_int reglist)
+{
+ u_int free_regs = ~reglist & CALLER_SAVE_REGS;
+ if (free_regs == 0)
+ return -1;
+ return __builtin_ctz(free_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;
- int jaddr=0;
+ void *jaddr=0;
int memtarget=0,c=0;
- int fastload_reg_override=0;
- u_int hr,reglist=0;
- th=get_reg(i_regs->regmap,rt1[i]|64);
- tl=get_reg(i_regs->regmap,rt1[i]);
- s=get_reg(i_regs->regmap,rs1[i]);
+ int offset_reg = -1;
+ int fastio_reg_override = -1;
+ u_int reglist=get_host_reglist(i_regs->regmap);
+ tl=get_reg(i_regs->regmap,dops[i].rt1);
+ s=get_reg(i_regs->regmap,dops[i].rs1);
offset=imm[i];
- for(hr=0;hr<HOST_REGS;hr++) {
- if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
- }
if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
if(s>=0) {
c=(i_regs->wasconst>>s)&1;
if (c) {
memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
- if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
}
}
//printf("load_assemble: c=%d\n",c);
- //if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset);
+ //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...
-#ifdef PCSX
- if(tl<0&&(!c||(((u_int)constmap[i][s]+offset)>>16)==0x1f80)
- ||rt1[i]==0) {
+ if((tl<0&&(!c||(((u_int)constmap[i][s]+offset)>>16)==0x1f80))
+ ||dops[i].rt1==0) {
// could be FIFO, must perform the read
// ||dummy read
assem_debug("(forced read)\n");
- tl=get_reg(i_regs->regmap,-1);
+ tl=get_reg_temp(i_regs->regmap);
assert(tl>=0);
}
-#endif
if(offset||s<0||c) addr=tl;
else addr=s;
- //if(tl<0) tl=get_reg(i_regs->regmap,-1);
+ //if(tl<0) tl=get_reg_temp(i_regs->regmap);
if(tl>=0) {
//printf("load_assemble: c=%d\n",c);
- //if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset);
+ //if(c) printf("load_assemble: const=%lx\n",(long)constmap[i][s]+offset);
assert(tl>=0); // Even if the load is a NOP, we must check for pagefaults and I/O
reglist&=~(1<<tl);
- if(th>=0) reglist&=~(1<<th);
- if(!using_tlb) {
- if(!c) {
- #ifdef RAM_OFFSET
- map=get_reg(i_regs->regmap,ROREG);
- if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG);
- #endif
-//#define R29_HACK 1
- #ifdef R29_HACK
- // Strmnnrmn's speed hack
- if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
- #endif
- {
- #ifdef PCSX
- if(sp_in_mirror&&rs1[i]==29) {
- emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
- emit_cmpimm(HOST_TEMPREG,RAM_SIZE);
- fastload_reg_override=HOST_TEMPREG;
- }
- else
- #endif
- emit_cmpimm(addr,RAM_SIZE);
- jaddr=(int)out;
- #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
- // Hint to branch predictor that the branch is unlikely to be taken
- if(rs1[i]>=28)
- emit_jno_unlikely(0);
- else
- #endif
- emit_jno(0);
- }
+ if(!c) {
+ #ifdef R29_HACK
+ // Strmnnrmn's speed hack
+ if(dops[i].rs1!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
+ #endif
+ {
+ jaddr = emit_fastpath_cmp_jump(i, i_regs, addr,
+ &offset_reg, &fastio_reg_override);
}
- }else{ // using tlb
- int x=0;
- if (opcode[i]==0x20||opcode[i]==0x24) x=3; // LB/LBU
- if (opcode[i]==0x21||opcode[i]==0x25) x=2; // LH/LHU
- map=get_reg(i_regs->regmap,TLREG);
- assert(map>=0);
- reglist&=~(1<<map);
- map=do_tlb_r(addr,tl,map,x,-1,-1,c,constmap[i][s]+offset);
- do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr);
}
- int dummy=(rt1[i]==0)||(tl!=get_reg(i_regs->regmap,rt1[i])); // ignore loads to r0 and unneeded reg
- if (opcode[i]==0x20) { // LB
+ else if (ram_offset && memtarget) {
+ offset_reg = get_ro_reg(i_regs, 0);
+ }
+ 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) {
- #ifdef HOST_IMM_ADDR32
- if(c)
- emit_movsbl_tlb((constmap[i][s]+offset)^3,map,tl);
- else
- #endif
- {
- //emit_xorimm(addr,3,tl);
- //gen_tlb_addr_r(tl,map);
- //emit_movsbl_indexed((int)rdram-0x80000000,tl,tl);
- int x=0,a=tl;
-#ifdef BIG_ENDIAN_MIPS
- if(!c) emit_xorimm(addr,3,tl);
- else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
-#else
- if(!c) a=addr;
-#endif
- if(fastload_reg_override) a=fastload_reg_override;
+ int a = tl;
+ if (!c) a = addr;
+ if (fastio_reg_override >= 0)
+ a = fastio_reg_override;
- emit_movsbl_indexed_tlb(x,a,map,tl);
- }
+ if (offset_reg >= 0)
+ emit_ldrsb_dualindexed(offset_reg, a, tl);
+ else
+ emit_movsbl_indexed(0, a, tl);
}
if(jaddr)
- add_stub(LOADB_STUB,jaddr,(int)out,i,addr,(int)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) {
- #ifdef HOST_IMM_ADDR32
- if(c)
- emit_movswl_tlb((constmap[i][s]+offset)^2,map,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
- #endif
- {
- int x=0,a=tl;
-#ifdef BIG_ENDIAN_MIPS
- if(!c) emit_xorimm(addr,2,tl);
- else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
-#else
- if(!c) a=addr;
-#endif
- if(fastload_reg_override) a=fastload_reg_override;
- //#ifdef
- //emit_movswl_indexed_tlb(x,tl,map,tl);
- //else
- if(map>=0) {
- gen_tlb_addr_r(a,map);
- emit_movswl_indexed(x,a,tl);
- }else{
- #ifdef RAM_OFFSET
- emit_movswl_indexed(x,a,tl);
- #else
- emit_movswl_indexed((int)rdram-0x80000000+x,a,tl);
- #endif
- }
- }
+ emit_movswl_indexed(0, a, tl);
}
if(jaddr)
- add_stub(LOADH_STUB,jaddr,(int)out,i,addr,(int)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(fastload_reg_override) a=fastload_reg_override;
- //emit_readword_indexed((int)rdram-0x80000000,addr,tl);
- #ifdef HOST_IMM_ADDR32
- if(c)
- emit_readword_tlb(constmap[i][s]+offset,map,tl);
- else
- #endif
- emit_readword_indexed_tlb(0,a,map,tl);
+ int a = addr;
+ if (fastio_reg_override >= 0)
+ a = fastio_reg_override;
+ do_load_word(a, tl, offset_reg);
}
if(jaddr)
- add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)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) {
- #ifdef HOST_IMM_ADDR32
- if(c)
- emit_movzbl_tlb((constmap[i][s]+offset)^3,map,tl);
- else
- #endif
- {
- //emit_xorimm(addr,3,tl);
- //gen_tlb_addr_r(tl,map);
- //emit_movzbl_indexed((int)rdram-0x80000000,tl,tl);
- int x=0,a=tl;
-#ifdef BIG_ENDIAN_MIPS
- if(!c) emit_xorimm(addr,3,tl);
- else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
-#else
- if(!c) a=addr;
-#endif
- if(fastload_reg_override) a=fastload_reg_override;
+ int a = tl;
+ if (!c) a = addr;
+ if (fastio_reg_override >= 0)
+ a = fastio_reg_override;
- emit_movzbl_indexed_tlb(x,a,map,tl);
- }
+ if (offset_reg >= 0)
+ emit_ldrb_dualindexed(offset_reg, a, tl);
+ else
+ emit_movzbl_indexed(0, a, tl);
}
if(jaddr)
- add_stub(LOADBU_STUB,jaddr,(int)out,i,addr,(int)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) {
- #ifdef HOST_IMM_ADDR32
- if(c)
- emit_movzwl_tlb((constmap[i][s]+offset)^2,map,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
- #endif
- {
- int x=0,a=tl;
-#ifdef BIG_ENDIAN_MIPS
- if(!c) emit_xorimm(addr,2,tl);
- else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
-#else
- if(!c) a=addr;
-#endif
- if(fastload_reg_override) a=fastload_reg_override;
- //#ifdef
- //emit_movzwl_indexed_tlb(x,tl,map,tl);
- //#else
- if(map>=0) {
- gen_tlb_addr_r(a,map);
- emit_movzwl_indexed(x,a,tl);
- }else{
- #ifdef RAM_OFFSET
- emit_movzwl_indexed(x,a,tl);
- #else
- emit_movzwl_indexed((int)rdram-0x80000000+x,a,tl);
- #endif
- }
- }
+ emit_movzwl_indexed(0, a, tl);
}
if(jaddr)
- add_stub(LOADHU_STUB,jaddr,(int)out,i,addr,(int)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);
+ 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 (opcode[i]==0x27) { // LWU
- assert(th>=0);
- if(!c||memtarget) {
- if(!dummy) {
- int a=addr;
- if(fastload_reg_override) a=fastload_reg_override;
- //emit_readword_indexed((int)rdram-0x80000000,addr,tl);
- #ifdef HOST_IMM_ADDR32
- if(c)
- emit_readword_tlb(constmap[i][s]+offset,map,tl);
- else
- #endif
- emit_readword_indexed_tlb(0,a,map,tl);
- }
- if(jaddr)
- add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+ }
+ 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, int ccadj_)
+{
+ int s,tl,temp,temp2,addr;
+ int offset;
+ void *jaddr=0;
+ int memtarget=0,c=0;
+ int offset_reg = -1;
+ int fastio_reg_override = -1;
+ u_int reglist=get_host_reglist(i_regs->regmap);
+ tl=get_reg(i_regs->regmap,dops[i].rt1);
+ s=get_reg(i_regs->regmap,dops[i].rs1);
+ temp=get_reg_temp(i_regs->regmap);
+ temp2=get_reg(i_regs->regmap,FTEMP);
+ addr=get_reg(i_regs->regmap,AGEN1+(i&1));
+ assert(addr<0);
+ offset=imm[i];
+ reglist|=1<<temp;
+ if(offset||s<0||c) addr=temp2;
+ else addr=s;
+ if(s>=0) {
+ c=(i_regs->wasconst>>s)&1;
+ if(c) {
+ memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
}
- else {
- inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ }
+ if(!c) {
+ emit_shlimm(addr,3,temp);
+ 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, i_regs, temp2,
+ &offset_reg, &fastio_reg_override);
+ }
+ else {
+ if (ram_offset && memtarget) {
+ offset_reg = get_ro_reg(i_regs, 0);
+ }
+ 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
}
- emit_zeroreg(th);
}
- if (opcode[i]==0x37) { // LD
+ if (dops[i].opcode==0x22||dops[i].opcode==0x26) { // LWL/LWR
if(!c||memtarget) {
- if(!dummy) {
- int a=addr;
- if(fastload_reg_override) a=fastload_reg_override;
- //gen_tlb_addr_r(tl,map);
- //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,addr,th);
- //emit_readword_indexed((int)rdram-0x7FFFFFFC,addr,tl);
- #ifdef HOST_IMM_ADDR32
- if(c)
- emit_readdword_tlb(constmap[i][s]+offset,map,th,tl);
- else
- #endif
- emit_readdword_indexed_tlb(0,a,map,th,tl);
- }
- if(jaddr)
- add_stub(LOADD_STUB,jaddr,(int)out,i,addr,(int)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(LOADD_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+ 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 (dops[i].opcode==0x22) // LWL
+ emit_xorimm(temp,24,temp);
+ host_tempreg_acquire();
+ emit_movimm(-1,HOST_TEMPREG);
+ if (dops[i].opcode==0x26) {
+ emit_shr(temp2,temp,temp2);
+ emit_bic_lsr(tl,HOST_TEMPREG,temp,tl);
+ }else{
+ emit_shl(temp2,temp,temp2);
+ emit_bic_lsl(tl,HOST_TEMPREG,temp,tl);
+ }
+ host_tempreg_release();
+ emit_or(temp2,tl,tl);
+ }
+ //emit_storereg(dops[i].rt1,tl); // DEBUG
+ }
+ if (dops[i].opcode==0x1A||dops[i].opcode==0x1B) { // LDL/LDR
+ assert(0);
}
- }
- //emit_storereg(rt1[i],tl); // DEBUG
- //if(opcode[i]==0x23)
- //if(opcode[i]==0x24)
- //if(opcode[i]==0x23||opcode[i]==0x24)
- /*if(opcode[i]==0x21||opcode[i]==0x23||opcode[i]==0x24)
- {
- //emit_pusha();
- save_regs(0x100f);
- emit_readword((int)&last_count,ECX);
- #ifdef __i386__
- if(get_reg(i_regs->regmap,CCREG)<0)
- emit_loadreg(CCREG,HOST_CCREG);
- emit_add(HOST_CCREG,ECX,HOST_CCREG);
- emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
- emit_writeword(HOST_CCREG,(int)&Count);
- #endif
- #ifdef __arm__
- if(get_reg(i_regs->regmap,CCREG)<0)
- emit_loadreg(CCREG,0);
- else
- emit_mov(HOST_CCREG,0);
- emit_add(0,ECX,0);
- emit_addimm(0,2*ccadj[i],0);
- emit_writeword(0,(int)&Count);
- #endif
- emit_call((int)memdebug);
- //emit_popa();
- restore_regs(0x100f);
- }/**/
-}
-
-#ifndef loadlr_assemble
-void loadlr_assemble(int i,struct regstat *i_regs)
-{
- printf("Need loadlr_assemble for this architecture.\n");
- exit(1);
}
#endif
-void store_assemble(int i,struct regstat *i_regs)
+static void store_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
- int s,th,tl,map=-1;
+ int s,tl;
int addr,temp;
int offset;
- int jaddr=0,jaddr2,type;
+ void *jaddr=0;
+ enum stub_type type=0;
int memtarget=0,c=0;
int agr=AGEN1+(i&1);
- int faststore_reg_override=0;
- u_int hr,reglist=0;
- th=get_reg(i_regs->regmap,rs2[i]|64);
- tl=get_reg(i_regs->regmap,rs2[i]);
- s=get_reg(i_regs->regmap,rs1[i]);
+ int offset_reg = -1;
+ int fastio_reg_override = -1;
+ u_int reglist=get_host_reglist(i_regs->regmap);
+ tl=get_reg(i_regs->regmap,dops[i].rs2);
+ s=get_reg(i_regs->regmap,dops[i].rs1);
temp=get_reg(i_regs->regmap,agr);
- if(temp<0) temp=get_reg(i_regs->regmap,-1);
+ if(temp<0) temp=get_reg_temp(i_regs->regmap);
offset=imm[i];
if(s>=0) {
c=(i_regs->wasconst>>s)&1;
if(c) {
memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
- if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
}
}
assert(tl>=0);
assert(temp>=0);
- for(hr=0;hr<HOST_REGS;hr++) {
- if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
- }
if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
if(offset||s<0||c) addr=temp;
else addr=s;
- if(!using_tlb) {
- if(!c) {
- #ifdef PCSX
- if(sp_in_mirror&&rs1[i]==29) {
- emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
- emit_cmpimm(HOST_TEMPREG,RAM_SIZE);
- faststore_reg_override=HOST_TEMPREG;
- }
- else
- #endif
- #ifdef R29_HACK
- // Strmnnrmn's speed hack
- if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
- #endif
- emit_cmpimm(addr,RAM_SIZE);
- #ifdef DESTRUCTIVE_SHIFT
- if(s==addr) emit_mov(s,temp);
- #endif
- #ifdef R29_HACK
- memtarget=1;
- if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
- #endif
- {
- jaddr=(int)out;
- #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
- // Hint to branch predictor that the branch is unlikely to be taken
- if(rs1[i]>=28)
- emit_jno_unlikely(0);
- else
- #endif
- emit_jno(0);
- }
- }
- }else{ // using tlb
- int x=0;
- if (opcode[i]==0x28) x=3; // SB
- if (opcode[i]==0x29) x=2; // SH
- map=get_reg(i_regs->regmap,TLREG);
- assert(map>=0);
- reglist&=~(1<<map);
- map=do_tlb_w(addr,temp,map,x,c,constmap[i][s]+offset);
- do_tlb_w_branch(map,c,constmap[i][s]+offset,&jaddr);
+ if (!c) {
+ jaddr = emit_fastpath_cmp_jump(i, i_regs, addr,
+ &offset_reg, &fastio_reg_override);
}
-
- if (opcode[i]==0x28) { // SB
- if(!c||memtarget) {
- int x=0,a=temp;
-#ifdef BIG_ENDIAN_MIPS
- if(!c) emit_xorimm(addr,3,temp);
- else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
-#else
- if(!c) a=addr;
-#endif
- if(faststore_reg_override) a=faststore_reg_override;
- //gen_tlb_addr_w(temp,map);
- //emit_writebyte_indexed(tl,(int)rdram-0x80000000,temp);
- emit_writebyte_indexed_tlb(tl,x,a,map,a);
- }
- type=STOREB_STUB;
+ else if (ram_offset && memtarget) {
+ offset_reg = get_ro_reg(i_regs, 0);
}
- if (opcode[i]==0x29) { // SH
+
+ switch (dops[i].opcode) {
+ case 0x28: // SB
if(!c||memtarget) {
- int x=0,a=temp;
-#ifdef BIG_ENDIAN_MIPS
- if(!c) emit_xorimm(addr,2,temp);
- else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
-#else
- if(!c) a=addr;
-#endif
- if(faststore_reg_override) a=faststore_reg_override;
- //#ifdef
- //emit_writehword_indexed_tlb(tl,x,temp,map,temp);
- //#else
- if(map>=0) {
- gen_tlb_addr_w(a,map);
- emit_writehword_indexed(tl,x,a);
- }else
- emit_writehword_indexed(tl,(int)rdram-0x80000000+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_byte(a, tl, offset_reg);
+ }
+ type = STOREB_STUB;
+ break;
+ case 0x29: // SH
if(!c||memtarget) {
- int a=addr;
- if(faststore_reg_override) a=faststore_reg_override;
- //emit_writeword_indexed(tl,(int)rdram-0x80000000,addr);
- emit_writeword_indexed_tlb(tl,0,a,map,temp);
- }
- type=STOREW_STUB;
- }
- if (opcode[i]==0x3F) { // SD
+ 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(faststore_reg_override) a=faststore_reg_override;
- if(rs2[i]) {
- assert(th>=0);
- //emit_writeword_indexed(th,(int)rdram-0x80000000,addr);
- //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,addr);
- emit_writedword_indexed_tlb(th,tl,0,a,map,temp);
- }else{
- // Store zero
- //emit_writeword_indexed(tl,(int)rdram-0x80000000,temp);
- //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,temp);
- emit_writedword_indexed_tlb(tl,tl,0,a,map,temp);
- }
- }
- type=STORED_STUB;
+ 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);
+ }
+ 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<<addr;
+ add_stub_r(type,jaddr,out,i,addr,i_regs,ccadj_,reglist);
+ jaddr=0;
}
- if(!using_tlb) {
+ if(!(i_regs->waswritten&(1<<dops[i].rs1)) && !HACK_ENABLED(NDHACK_NO_SMC_CHECK)) {
if(!c||memtarget) {
#ifdef DESTRUCTIVE_SHIFT
// The x86 shift operation is 'destructive'; it overwrites the
assert(ir>=0);
emit_cmpmem_indexedsr12_reg(ir,addr,1);
#else
- emit_cmpmem_indexedsr12_imm((int)invalid_code,addr,1);
+ emit_cmpmem_indexedsr12_imm(invalid_code,addr,1);
#endif
#if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
emit_callne(invalidate_addr_reg[addr]);
#else
- jaddr2=(int)out;
+ void *jaddr2 = out;
emit_jne(0);
- add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),addr,0,0,0);
+ add_stub(INVCODE_STUB,jaddr2,out,reglist|(1<<HOST_CCREG),addr,0,0,0);
#endif
}
}
+ u_int addr_val=constmap[i][s]+offset;
if(jaddr) {
- add_stub(type,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+ add_stub_r(type,jaddr,out,i,addr,i_regs,ccadj_,reglist);
} else if(c&&!memtarget) {
- inline_writestub(type,i,constmap[i][s]+offset,i_regs->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
+ // (see Spyro2 title->attract mode)
+ if(c&&start+i*4<addr_val&&addr_val<start+slen*4) {
+ SysPrintf("write to %08x hits block %08x, pc=%08x\n",addr_val,start,start+i*4);
+ assert(i_regs->regmap==regs[i].regmap); // not delay slot
+ if(i_regs->regmap==regs[i].regmap) {
+ load_all_consts(regs[i].regmap_entry,regs[i].wasdirty,i);
+ wb_dirtys(regs[i].regmap_entry,regs[i].wasdirty);
+ emit_movimm(start+i*4+4,0);
+ emit_writeword(0,&pcaddr);
+ emit_addimm(HOST_CCREG,2,HOST_CCREG);
+ emit_far_call(get_addr_ht);
+ emit_jmpreg(0);
+ }
}
- //if(opcode[i]==0x2B || opcode[i]==0x3F)
- //if(opcode[i]==0x2B || opcode[i]==0x28)
- //if(opcode[i]==0x2B || opcode[i]==0x29)
- //if(opcode[i]==0x2B)
- /*if(opcode[i]==0x2B || opcode[i]==0x28 || opcode[i]==0x29 || opcode[i]==0x3F)
- {
- #ifdef __i386__
- emit_pusha();
- #endif
- #ifdef __arm__
- save_regs(0x100f);
- #endif
- emit_readword((int)&last_count,ECX);
- #ifdef __i386__
- if(get_reg(i_regs->regmap,CCREG)<0)
- emit_loadreg(CCREG,HOST_CCREG);
- emit_add(HOST_CCREG,ECX,HOST_CCREG);
- emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
- emit_writeword(HOST_CCREG,(int)&Count);
- #endif
- #ifdef __arm__
- if(get_reg(i_regs->regmap,CCREG)<0)
- emit_loadreg(CCREG,0);
- else
- emit_mov(HOST_CCREG,0);
- emit_add(0,ECX,0);
- emit_addimm(0,2*ccadj[i],0);
- emit_writeword(0,(int)&Count);
- #endif
- emit_call((int)memdebug);
- #ifdef __i386__
- emit_popa();
- #endif
- #ifdef __arm__
- restore_regs(0x100f);
- #endif
- }/**/
}
-void storelr_assemble(int i,struct regstat *i_regs)
+static void storelr_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
- int s,th,tl;
+ int s,tl;
int temp;
- int temp2;
int offset;
- int jaddr=0,jaddr2;
- int case1,case2,case3;
- int done0,done1,done2;
+ void *jaddr=0;
+ void *case1, *case23, *case3;
+ void *done0, *done1, *done2;
int memtarget=0,c=0;
int agr=AGEN1+(i&1);
- u_int hr,reglist=0;
- th=get_reg(i_regs->regmap,rs2[i]|64);
- tl=get_reg(i_regs->regmap,rs2[i]);
- s=get_reg(i_regs->regmap,rs1[i]);
+ int offset_reg = -1;
+ u_int reglist=get_host_reglist(i_regs->regmap);
+ tl=get_reg(i_regs->regmap,dops[i].rs2);
+ s=get_reg(i_regs->regmap,dops[i].rs1);
temp=get_reg(i_regs->regmap,agr);
- if(temp<0) temp=get_reg(i_regs->regmap,-1);
+ if(temp<0) temp=get_reg_temp(i_regs->regmap);
offset=imm[i];
if(s>=0) {
c=(i_regs->isconst>>s)&1;
if(c) {
memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
- if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
}
}
assert(tl>=0);
- for(hr=0;hr<HOST_REGS;hr++) {
- if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
- }
assert(temp>=0);
- if(!using_tlb) {
- if(!c) {
- emit_cmpimm(s<0||offset?temp:s,RAM_SIZE);
- if(!offset&&s!=temp) emit_mov(s,temp);
- jaddr=(int)out;
- emit_jno(0);
- }
- else
- {
- if(!memtarget||!rs1[i]) {
- jaddr=(int)out;
- emit_jmp(0);
- }
- }
- #ifdef RAM_OFFSET
- int map=get_reg(i_regs->regmap,ROREG);
- if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG);
- gen_tlb_addr_w(temp,map);
- #else
- if((u_int)rdram!=0x80000000)
- emit_addimm_no_flags((u_int)rdram-(u_int)0x80000000,temp);
- #endif
- }else{ // using tlb
- int map=get_reg(i_regs->regmap,TLREG);
- assert(map>=0);
- reglist&=~(1<<map);
- map=do_tlb_w(c||s<0||offset?temp:s,temp,map,0,c,constmap[i][s]+offset);
- if(!c&&!offset&&s>=0) emit_mov(s,temp);
- do_tlb_w_branch(map,c,constmap[i][s]+offset,&jaddr);
- if(!jaddr&&!memtarget) {
- jaddr=(int)out;
+ if(!c) {
+ emit_cmpimm(s<0||offset?temp:s,RAM_SIZE);
+ if(!offset&&s!=temp) emit_mov(s,temp);
+ jaddr=out;
+ emit_jno(0);
+ }
+ else
+ {
+ if(!memtarget||!dops[i].rs1) {
+ jaddr=out;
emit_jmp(0);
}
- gen_tlb_addr_w(temp,map);
}
+ if (ram_offset)
+ offset_reg = get_ro_reg(i_regs, 0);
- if (opcode[i]==0x2C||opcode[i]==0x2D) { // SDL/SDR
- temp2=get_reg(i_regs->regmap,FTEMP);
- if(!rs2[i]) temp2=th=tl;
+ if (dops[i].opcode==0x2C||dops[i].opcode==0x2D) { // SDL/SDR
+ assert(0);
}
-#ifndef BIG_ENDIAN_MIPS
- emit_xorimm(temp,3,temp);
-#endif
emit_testimm(temp,2);
- case2=(int)out;
+ case23=out;
emit_jne(0);
emit_testimm(temp,1);
- case1=(int)out;
+ case1=out;
emit_jne(0);
// 0
- if (opcode[i]==0x2A) { // SWL
- emit_writeword_indexed(tl,0,temp);
- }
- if (opcode[i]==0x2E) { // SWR
- emit_writebyte_indexed(tl,3,temp);
- }
- if (opcode[i]==0x2C) { // SDL
- emit_writeword_indexed(th,0,temp);
- if(rs2[i]) emit_mov(tl,temp2);
+ if (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);
}
- if (opcode[i]==0x2D) { // SDR
- emit_writebyte_indexed(tl,3,temp);
- if(rs2[i]) emit_shldimm(th,tl,24,temp2);
+ else if (dops[i].opcode == 0x2E) { // SWR
+ // Write entire word
+ do_store_word(temp, 0, tl, offset_reg, 1);
}
- done0=(int)out;
+ done0 = out;
emit_jmp(0);
// 1
- set_jump_target(case1,(int)out);
- if (opcode[i]==0x2A) { // SWL
- // Write 3 msb into three least significant bytes
- if(rs2[i]) emit_rorimm(tl,8,tl);
- emit_writehword_indexed(tl,-1,temp);
- if(rs2[i]) emit_rorimm(tl,16,tl);
- emit_writebyte_indexed(tl,1,temp);
- if(rs2[i]) emit_rorimm(tl,8,tl);
- }
- if (opcode[i]==0x2E) { // SWR
- // Write two lsb into two most significant bytes
- emit_writehword_indexed(tl,1,temp);
+ set_jump_target(case1, out);
+ 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);
}
- if (opcode[i]==0x2C) { // SDL
- if(rs2[i]) emit_shrdimm(tl,th,8,temp2);
- // Write 3 msb into three least significant bytes
- if(rs2[i]) emit_rorimm(th,8,th);
- emit_writehword_indexed(th,-1,temp);
- if(rs2[i]) emit_rorimm(th,16,th);
- emit_writebyte_indexed(th,1,temp);
- if(rs2[i]) emit_rorimm(th,8,th);
- }
- if (opcode[i]==0x2D) { // SDR
- if(rs2[i]) emit_shldimm(th,tl,16,temp2);
- // Write two lsb into two most significant bytes
- emit_writehword_indexed(tl,1,temp);
+ 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=(int)out;
+ done1=out;
emit_jmp(0);
- // 2
- set_jump_target(case2,(int)out);
+ // 2,3
+ set_jump_target(case23, out);
emit_testimm(temp,1);
- case3=(int)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);
- }
- if (opcode[i]==0x2E) { // SWR
- // Write 3 lsb into three most significant bytes
- emit_writebyte_indexed(tl,-1,temp);
- if(rs2[i]) emit_rorimm(tl,8,tl);
- emit_writehword_indexed(tl,0,temp);
- if(rs2[i]) emit_rorimm(tl,24,tl);
- }
- if (opcode[i]==0x2C) { // SDL
- if(rs2[i]) emit_shrdimm(tl,th,16,temp2);
- // Write two msb into two least significant bytes
- if(rs2[i]) emit_rorimm(th,16,th);
- emit_writehword_indexed(th,-2,temp);
- if(rs2[i]) emit_rorimm(th,16,th);
+ // 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);
}
- if (opcode[i]==0x2D) { // SDR
- if(rs2[i]) emit_shldimm(th,tl,8,temp2);
- // Write 3 lsb into three most significant bytes
- emit_writebyte_indexed(tl,-1,temp);
- if(rs2[i]) emit_rorimm(tl,8,tl);
- emit_writehword_indexed(tl,0,temp);
- if(rs2[i]) emit_rorimm(tl,24,tl);
+ 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=(int)out;
+ done2 = out;
emit_jmp(0);
// 3
- set_jump_target(case3,(int)out);
- if (opcode[i]==0x2A) { // SWL
- // Write msb into least significant byte
- if(rs2[i]) emit_rorimm(tl,24,tl);
- emit_writebyte_indexed(tl,-3,temp);
- if(rs2[i]) emit_rorimm(tl,8,tl);
- }
- if (opcode[i]==0x2E) { // SWR
- // Write entire word
- emit_writeword_indexed(tl,-3,temp);
- }
- if (opcode[i]==0x2C) { // SDL
- if(rs2[i]) emit_shrdimm(tl,th,24,temp2);
- // Write msb into least significant byte
- if(rs2[i]) emit_rorimm(th,24,th);
- emit_writebyte_indexed(th,-3,temp);
- if(rs2[i]) emit_rorimm(th,8,th);
- }
- if (opcode[i]==0x2D) { // SDR
- if(rs2[i]) emit_mov(th,temp2);
- // Write entire word
- emit_writeword_indexed(tl,-3,temp);
- }
- set_jump_target(done0,(int)out);
- set_jump_target(done1,(int)out);
- set_jump_target(done2,(int)out);
- if (opcode[i]==0x2C) { // SDL
- emit_testimm(temp,4);
- done0=(int)out;
- emit_jne(0);
- emit_andimm(temp,~3,temp);
- emit_writeword_indexed(temp2,4,temp);
- set_jump_target(done0,(int)out);
- }
- if (opcode[i]==0x2D) { // SDR
- emit_testimm(temp,4);
- done0=(int)out;
- emit_jeq(0);
- emit_andimm(temp,~3,temp);
- emit_writeword_indexed(temp2,-4,temp);
- set_jump_target(done0,(int)out);
- }
+ set_jump_target(case3, out);
+ if (dops[i].opcode == 0x2A) { // SWL
+ do_store_word(temp, -3, tl, offset_reg, 0);
+ }
+ 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(STORELR_STUB,jaddr,(int)out,i,(int)i_regs,temp,ccadj[i],reglist);
- if(!using_tlb) {
- #ifdef RAM_OFFSET
- int map=get_reg(i_regs->regmap,ROREG);
- if(map<0) map=HOST_TEMPREG;
- gen_orig_addr_w(temp,map);
- #else
- emit_addimm_no_flags((u_int)0x80000000-(u_int)rdram,temp);
- #endif
+ add_stub_r(STORELR_STUB,jaddr,out,i,temp,i_regs,ccadj_,reglist);
+ if(!(i_regs->waswritten&(1<<dops[i].rs1)) && !HACK_ENABLED(NDHACK_NO_SMC_CHECK)) {
#if defined(HOST_IMM8)
int ir=get_reg(i_regs->regmap,INVCP);
assert(ir>=0);
emit_cmpmem_indexedsr12_reg(ir,temp,1);
#else
- emit_cmpmem_indexedsr12_imm((int)invalid_code,temp,1);
+ emit_cmpmem_indexedsr12_imm(invalid_code,temp,1);
#endif
#if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
emit_callne(invalidate_addr_reg[temp]);
#else
- jaddr2=(int)out;
+ void *jaddr2 = out;
emit_jne(0);
- add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),temp,0,0,0);
+ add_stub(INVCODE_STUB,jaddr2,out,reglist|(1<<HOST_CCREG),temp,0,0,0);
#endif
}
- /*
- emit_pusha();
- //save_regs(0x100f);
- emit_readword((int)&last_count,ECX);
- if(get_reg(i_regs->regmap,CCREG)<0)
- emit_loadreg(CCREG,HOST_CCREG);
- emit_add(HOST_CCREG,ECX,HOST_CCREG);
- emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
- emit_writeword(HOST_CCREG,(int)&Count);
- emit_call((int)memdebug);
- emit_popa();
- //restore_regs(0x100f);
- /**/
}
-void c1ls_assemble(int i,struct regstat *i_regs)
+static void cop0_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
-#ifndef DISABLE_COP1
- int s,th,tl;
- int temp,ar;
- int map=-1;
- int offset;
- int c=0;
- int jaddr,jaddr2=0,jaddr3,type;
- int agr=AGEN1+(i&1);
- u_int hr,reglist=0;
- th=get_reg(i_regs->regmap,FTEMP|64);
- tl=get_reg(i_regs->regmap,FTEMP);
- s=get_reg(i_regs->regmap,rs1[i]);
- temp=get_reg(i_regs->regmap,agr);
- if(temp<0) temp=get_reg(i_regs->regmap,-1);
- offset=imm[i];
- assert(tl>=0);
- assert(rs1[i]>0);
- assert(temp>=0);
- for(hr=0;hr<HOST_REGS;hr++) {
- if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
- }
- if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
- if (opcode[i]==0x31||opcode[i]==0x35) // LWC1/LDC1
+ if(dops[i].opcode2==0) // MFC0
{
- // Loads use a temporary register which we need to save
- reglist|=1<<temp;
- }
- if (opcode[i]==0x39||opcode[i]==0x3D) // SWC1/SDC1
- ar=temp;
- else // LWC1/LDC1
- ar=tl;
- //if(s<0) emit_loadreg(rs1[i],ar); //address_generation does this now
- //else c=(i_regs->wasconst>>s)&1;
- if(s>=0) c=(i_regs->wasconst>>s)&1;
- // Check cop1 unusable
- if(!cop1_usable) {
- signed char rs=get_reg(i_regs->regmap,CSREG);
- assert(rs>=0);
- emit_testimm(rs,0x20000000);
- jaddr=(int)out;
- emit_jeq(0);
- add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0);
- cop1_usable=1;
- }
- if (opcode[i]==0x39) { // SWC1 (get float address)
- emit_readword((int)®_cop1_simple[(source[i]>>16)&0x1f],tl);
- }
- if (opcode[i]==0x3D) { // SDC1 (get double address)
- emit_readword((int)®_cop1_double[(source[i]>>16)&0x1f],tl);
+ 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&&dops[i].rt1!=0) {
+ emit_readword(®_cop0[copr],t);
+ }
}
- // Generate address + offset
- if(!using_tlb) {
- if(!c)
- emit_cmpimm(offset||c||s<0?ar:s,RAM_SIZE);
+ else if(dops[i].opcode2==4) // MTC0
+ {
+ signed char s=get_reg(i_regs->regmap,dops[i].rs1);
+ char copr=(source[i]>>11)&0x1f;
+ assert(s>=0);
+ 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,ccadj_,HOST_CCREG);
+ emit_writeword(HOST_CCREG,&Count);
+ }
+ // What a mess. The status register (12) can enable interrupts,
+ // so needs a special case to handle a pending interrupt.
+ // The interrupt must be taken immediately, because a subsequent
+ // instruction might disable interrupts again.
+ if(copr==12||copr==13) {
+ if (is_delayslot) {
+ // burn cycles to cause cc_interrupt, which will
+ // reschedule next_interupt. Relies on CCREG from above.
+ assem_debug("MTC0 DS %d\n", copr);
+ emit_writeword(HOST_CCREG,&last_count);
+ emit_movimm(0,HOST_CCREG);
+ emit_storereg(CCREG,HOST_CCREG);
+ emit_loadreg(dops[i].rs1,1);
+ emit_movimm(copr,0);
+ emit_far_call(pcsx_mtc0_ds);
+ emit_loadreg(dops[i].rs1,s);
+ return;
+ }
+ emit_movimm(start+i*4+4,HOST_TEMPREG);
+ emit_writeword(HOST_TEMPREG,&pcaddr);
+ emit_movimm(0,HOST_TEMPREG);
+ emit_writeword(HOST_TEMPREG,&pending_exception);
+ }
+ if(s==HOST_CCREG)
+ emit_loadreg(dops[i].rs1,1);
+ else if(s!=1)
+ emit_mov(s,1);
+ emit_movimm(copr,0);
+ emit_far_call(pcsx_mtc0);
+ if(copr==9||copr==11||copr==12||copr==13) {
+ emit_readword(&Count,HOST_CCREG);
+ emit_readword(&next_interupt,HOST_TEMPREG);
+ 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);
+ }
+ if(copr==12||copr==13) {
+ assert(!is_delayslot);
+ emit_readword(&pending_exception,14);
+ emit_test(14,14);
+ void *jaddr = out;
+ emit_jeq(0);
+ emit_readword(&pcaddr, 0);
+ emit_addimm(HOST_CCREG,2,HOST_CCREG);
+ emit_far_call(get_addr_ht);
+ emit_jmpreg(0);
+ set_jump_target(jaddr, out);
+ }
+ emit_loadreg(dops[i].rs1,s);
}
else
{
- map=get_reg(i_regs->regmap,TLREG);
- assert(map>=0);
- reglist&=~(1<<map);
- if (opcode[i]==0x31||opcode[i]==0x35) { // LWC1/LDC1
- map=do_tlb_r(offset||c||s<0?ar:s,ar,map,0,-1,-1,c,constmap[i][s]+offset);
- }
- if (opcode[i]==0x39||opcode[i]==0x3D) { // SWC1/SDC1
- map=do_tlb_w(offset||c||s<0?ar:s,ar,map,0,c,constmap[i][s]+offset);
+ assert(dops[i].opcode2==0x10);
+ //if((source[i]&0x3f)==0x10) // RFE
+ {
+ emit_readword(&Status,0);
+ emit_andimm(0,0x3c,1);
+ emit_andimm(0,~0xf,0);
+ emit_orrshr_imm(1,2,0);
+ emit_writeword(0,&Status);
}
}
- if (opcode[i]==0x39) { // SWC1 (read float)
- emit_readword_indexed(0,tl,tl);
+}
+
+static void cop1_unusable(int i, const struct regstat *i_regs)
+{
+ // XXX: should just just do the exception instead
+ //if(!cop1_usable)
+ {
+ void *jaddr=out;
+ emit_jmp(0);
+ add_stub_r(FP_STUB,jaddr,out,i,0,i_regs,is_delayslot,0);
}
- if (opcode[i]==0x3D) { // SDC1 (read double)
- emit_readword_indexed(4,tl,th);
- emit_readword_indexed(0,tl,tl);
+}
+
+static void cop1_assemble(int i, const struct regstat *i_regs)
+{
+ cop1_unusable(i, i_regs);
+}
+
+static void c1ls_assemble(int i, const struct regstat *i_regs)
+{
+ cop1_unusable(i, i_regs);
+}
+
+// FP_STUB
+static void do_cop1stub(int n)
+{
+ literal_pool(256);
+ assem_debug("do_cop1stub %x\n",start+stubs[n].a*4);
+ set_jump_target(stubs[n].addr, out);
+ int i=stubs[n].a;
+// int rs=stubs[n].b;
+ struct regstat *i_regs=(struct regstat *)stubs[n].c;
+ int ds=stubs[n].d;
+ if(!ds) {
+ load_all_consts(regs[i].regmap_entry,regs[i].wasdirty,i);
+ //if(i_regs!=®s[i]) printf("oops: regs[i]=%x i_regs=%x",(int)®s[i],(int)i_regs);
+ }
+ //else {printf("fp exception in delay slot\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,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 (dops[i].opcode == 0x3a) { // SWC2
+ *cycles = 0;
+ return 1;
}
- if (opcode[i]==0x31) { // LWC1 (get target address)
- emit_readword((int)®_cop1_simple[(source[i]>>16)&0x1f],temp);
+ if (dops[i].itype == COP2 && (dops[i].opcode2 == 0 || dops[i].opcode2 == 2)) { // MFC2/CFC2
+ *cycles = 0;
+ return 1;
}
- if (opcode[i]==0x35) { // LDC1 (get target address)
- emit_readword((int)®_cop1_double[(source[i]>>16)&0x1f],temp);
+ if (dops[i].itype == C2OP) {
+ *cycles = gte_cycletab[source[i] & 0x3f];
+ return 1;
}
- if(!using_tlb) {
- if(!c) {
- jaddr2=(int)out;
- emit_jno(0);
- }
- else if(((signed int)(constmap[i][s]+offset))>=(signed int)0x80000000+RAM_SIZE) {
- jaddr2=(int)out;
- emit_jmp(0); // inline_readstub/inline_writestub? Very rare case
- }
- #ifdef DESTRUCTIVE_SHIFT
- if (opcode[i]==0x39||opcode[i]==0x3D) { // SWC1/SDC1
- if(!offset&&!c&&s>=0) emit_mov(s,ar);
- }
- #endif
- }else{
- if (opcode[i]==0x31||opcode[i]==0x35) { // LWC1/LDC1
- do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr2);
- }
- if (opcode[i]==0x39||opcode[i]==0x3D) { // SWC1/SDC1
- do_tlb_w_branch(map,c,constmap[i][s]+offset,&jaddr2);
+ // ... what about MTC2/CTC2/LWC2?
+ return 0;
+}
+
+#if 0
+static void log_gte_stall(int stall, u_int cycle)
+{
+ if ((u_int)stall <= 44)
+ printf("x stall %2d %u\n", stall, cycle + last_count);
+}
+
+static void emit_log_gte_stall(int i, int stall, u_int reglist)
+{
+ save_regs(reglist);
+ if (stall > 0)
+ emit_movimm(stall, 0);
+ else
+ emit_mov(HOST_TEMPREG, 0);
+ emit_addimm(HOST_CCREG, ccadj[i], 1);
+ emit_far_call(log_gte_stall);
+ restore_regs(reglist);
+}
+#endif
+
+static void cop2_do_stall_check(u_int op, int i, const struct regstat *i_regs, u_int reglist)
+{
+ int j = i, other_gte_op_cycles = -1, stall = -MAXBLOCK, cycles_passed;
+ int rtmp = reglist_find_free(reglist);
+
+ if (HACK_ENABLED(NDHACK_NO_STALLS))
+ return;
+ if (get_reg(i_regs->regmap, CCREG) != HOST_CCREG) {
+ // happens occasionally... cc evicted? Don't bother then
+ //printf("no cc %08x\n", start + i*4);
+ return;
+ }
+ if (!dops[i].bt) {
+ for (j = i - 1; j >= 0; 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 = ccadj[i] - ccadj[j];
+ if (other_gte_op_cycles >= 0)
+ stall = other_gte_op_cycles - cycles_passed;
+ else if (cycles_passed >= 44)
+ stall = 0; // can't stall
+ if (stall == -MAXBLOCK && rtmp >= 0) {
+ // unknown stall, do the expensive runtime check
+ assem_debug("; cop2_do_stall_check\n");
+#if 0 // too slow
+ save_regs(reglist);
+ emit_movimm(gte_cycletab[op], 0);
+ 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, -ccadj[i], rtmp);
+ emit_sub(rtmp, HOST_CCREG, HOST_TEMPREG);
+ emit_cmpimm(HOST_TEMPREG, 44);
+ emit_cmovb_reg(rtmp, HOST_CCREG);
+ //emit_log_gte_stall(i, 0, reglist);
+ host_tempreg_release();
+#endif
}
- if (opcode[i]==0x31) { // LWC1
- //if(s>=0&&!c&&!offset) emit_mov(s,tl);
- //gen_tlb_addr_r(ar,map);
- //emit_readword_indexed((int)rdram-0x80000000,tl,tl);
- #ifdef HOST_IMM_ADDR32
- if(c) emit_readword_tlb(constmap[i][s]+offset,map,tl);
- else
- #endif
- emit_readword_indexed_tlb(0,offset||c||s<0?tl:s,map,tl);
- type=LOADW_STUB;
- }
- if (opcode[i]==0x35) { // LDC1
- assert(th>=0);
- //if(s>=0&&!c&&!offset) emit_mov(s,tl);
- //gen_tlb_addr_r(ar,map);
- //emit_readword_indexed((int)rdram-0x80000000,tl,th);
- //emit_readword_indexed((int)rdram-0x7FFFFFFC,tl,tl);
- #ifdef HOST_IMM_ADDR32
- if(c) emit_readdword_tlb(constmap[i][s]+offset,map,th,tl);
- else
- #endif
- emit_readdword_indexed_tlb(0,offset||c||s<0?tl:s,map,th,tl);
- type=LOADD_STUB;
- }
- if (opcode[i]==0x39) { // SWC1
- //emit_writeword_indexed(tl,(int)rdram-0x80000000,temp);
- emit_writeword_indexed_tlb(tl,0,offset||c||s<0?temp:s,map,temp);
- type=STOREW_STUB;
+ else if (stall > 0) {
+ //emit_log_gte_stall(i, stall, reglist);
+ emit_addimm(HOST_CCREG, stall, HOST_CCREG);
}
- if (opcode[i]==0x3D) { // SDC1
- assert(th>=0);
- //emit_writeword_indexed(th,(int)rdram-0x80000000,temp);
- //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,temp);
- emit_writedword_indexed_tlb(th,tl,0,offset||c||s<0?temp:s,map,temp);
- type=STORED_STUB;
- }
- if(!using_tlb) {
- if (opcode[i]==0x39||opcode[i]==0x3D) { // SWC1/SDC1
- #ifndef DESTRUCTIVE_SHIFT
- temp=offset||c||s<0?ar:s;
- #endif
- #if defined(HOST_IMM8)
- int ir=get_reg(i_regs->regmap,INVCP);
- assert(ir>=0);
- emit_cmpmem_indexedsr12_reg(ir,temp,1);
- #else
- emit_cmpmem_indexedsr12_imm((int)invalid_code,temp,1);
- #endif
- #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
- emit_callne(invalidate_addr_reg[temp]);
- #else
- jaddr3=(int)out;
- emit_jne(0);
- add_stub(INVCODE_STUB,jaddr3,(int)out,reglist|(1<<HOST_CCREG),temp,0,0,0);
- #endif
+
+ // save gteBusyCycle, if needed
+ if (gte_cycletab[op] == 0)
+ return;
+ other_gte_op_cycles = -1;
+ for (j = i + 1; j < slen; j++) {
+ if (cop2_is_stalling_op(j, &other_gte_op_cycles))
+ break;
+ if (dops[j].is_jump) {
+ // check ds
+ if (j + 1 < slen && cop2_is_stalling_op(j + 1, &other_gte_op_cycles))
+ j++;
+ break;
}
}
- if(jaddr2) add_stub(type,jaddr2,(int)out,i,offset||c||s<0?ar:s,(int)i_regs,ccadj[i],reglist);
- if (opcode[i]==0x31) { // LWC1 (write float)
- emit_writeword_indexed(tl,0,temp);
+ if (other_gte_op_cycles >= 0)
+ // will handle stall when assembling that op
+ return;
+ cycles_passed = ccadj[min(j, slen -1)] - ccadj[i];
+ if (cycles_passed >= 44)
+ return;
+ assem_debug("; save gteBusyCycle\n");
+ host_tempreg_acquire();
+#if 0
+ emit_readword(&last_count, HOST_TEMPREG);
+ emit_add(HOST_TEMPREG, HOST_CCREG, 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, ccadj[i] + gte_cycletab[op], HOST_TEMPREG);
+ emit_writeword(HOST_TEMPREG, &psxRegs.gteBusyCycle);
+#endif
+ host_tempreg_release();
+}
+
+static int is_mflohi(int i)
+{
+ return (dops[i].itype == MOV && (dops[i].rs1 == HIREG || dops[i].rs1 == LOREG));
+}
+
+static int check_multdiv(int i, int *cycles)
+{
+ if (dops[i].itype != MULTDIV)
+ return 0;
+ 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, int ccadj_)
+{
+ int j, found = 0, c = 0;
+ if (HACK_ENABLED(NDHACK_NO_STALLS))
+ return;
+ if (get_reg(i_regs->regmap, CCREG) != HOST_CCREG) {
+ // happens occasionally... cc evicted? Don't bother then
+ return;
+ }
+ for (j = i + 1; j < slen; j++) {
+ if (dops[j].bt)
+ break;
+ if ((found = is_mflohi(j)))
+ break;
+ if (dops[j].is_jump) {
+ // check ds
+ if (j + 1 < slen && (found = is_mflohi(j + 1)))
+ j++;
+ break;
+ }
}
- if (opcode[i]==0x35) { // LDC1 (write double)
- emit_writeword_indexed(th,4,temp);
- emit_writeword_indexed(tl,0,temp);
+ if (found)
+ // handle all in multdiv_do_stall()
+ return;
+ check_multdiv(i, &c);
+ assert(c > 0);
+ assem_debug("; muldiv prepare stall %d\n", c);
+ host_tempreg_acquire();
+ emit_addimm(HOST_CCREG, ccadj_ + c, HOST_TEMPREG);
+ emit_writeword(HOST_TEMPREG, &psxRegs.muldivBusyCycle);
+ host_tempreg_release();
+}
+
+static void multdiv_do_stall(int i, const struct regstat *i_regs)
+{
+ int j, known_cycles = 0;
+ u_int reglist = get_host_reglist(i_regs->regmap);
+ int rtmp = get_reg_temp(i_regs->regmap);
+ if (rtmp < 0)
+ rtmp = reglist_find_free(reglist);
+ if (HACK_ENABLED(NDHACK_NO_STALLS))
+ return;
+ if (get_reg(i_regs->regmap, CCREG) != HOST_CCREG || rtmp < 0) {
+ // happens occasionally... cc evicted? Don't bother then
+ //printf("no cc/rtmp %08x\n", start + i*4);
+ return;
+ }
+ if (!dops[i].bt) {
+ for (j = i - 1; j >= 0; 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 -= 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);
+ return;
+ }
+ assem_debug("; muldiv stall unresolved\n");
+ host_tempreg_acquire();
+ emit_readword(&psxRegs.muldivBusyCycle, 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);
+ //emit_log_gte_stall(i, 0, reglist);
+ host_tempreg_release();
+}
+
+static void cop2_get_dreg(u_int copr,signed char tl,signed char temp)
+{
+ switch (copr) {
+ case 1:
+ case 3:
+ case 5:
+ case 8:
+ case 9:
+ case 10:
+ case 11:
+ emit_readword(®_cop2d[copr],tl);
+ emit_signextend16(tl,tl);
+ emit_writeword(tl,®_cop2d[copr]); // hmh
+ break;
+ case 7:
+ case 16:
+ case 17:
+ case 18:
+ case 19:
+ emit_readword(®_cop2d[copr],tl);
+ emit_andimm(tl,0xffff,tl);
+ emit_writeword(tl,®_cop2d[copr]);
+ break;
+ case 15:
+ emit_readword(®_cop2d[14],tl); // SXY2
+ emit_writeword(tl,®_cop2d[copr]);
+ break;
+ case 28:
+ case 29:
+ c2op_mfc2_29_assemble(tl,temp);
+ break;
+ default:
+ emit_readword(®_cop2d[copr],tl);
+ break;
}
- //if(opcode[i]==0x39)
- /*if(opcode[i]==0x39||opcode[i]==0x31)
- {
- emit_pusha();
- emit_readword((int)&last_count,ECX);
- if(get_reg(i_regs->regmap,CCREG)<0)
- emit_loadreg(CCREG,HOST_CCREG);
- emit_add(HOST_CCREG,ECX,HOST_CCREG);
- emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
- emit_writeword(HOST_CCREG,(int)&Count);
- emit_call((int)memdebug);
- emit_popa();
- }/**/
+}
+
+static void cop2_put_dreg(u_int copr,signed char sl,signed char temp)
+{
+ switch (copr) {
+ case 15:
+ emit_readword(®_cop2d[13],temp); // SXY1
+ emit_writeword(sl,®_cop2d[copr]);
+ emit_writeword(temp,®_cop2d[12]); // SXY0
+ emit_readword(®_cop2d[14],temp); // SXY2
+ emit_writeword(sl,®_cop2d[14]);
+ emit_writeword(temp,®_cop2d[13]); // SXY1
+ break;
+ case 28:
+ emit_andimm(sl,0x001f,temp);
+ emit_shlimm(temp,7,temp);
+ emit_writeword(temp,®_cop2d[9]);
+ emit_andimm(sl,0x03e0,temp);
+ emit_shlimm(temp,2,temp);
+ emit_writeword(temp,®_cop2d[10]);
+ emit_andimm(sl,0x7c00,temp);
+ emit_shrimm(temp,3,temp);
+ emit_writeword(temp,®_cop2d[11]);
+ emit_writeword(sl,®_cop2d[28]);
+ break;
+ case 30:
+ emit_xorsar_imm(sl,sl,31,temp);
+#if defined(HAVE_ARMV5) || defined(__aarch64__)
+ emit_clz(temp,temp);
#else
- cop1_unusable(i, i_regs);
+ emit_movs(temp,HOST_TEMPREG);
+ emit_movimm(0,temp);
+ emit_jeq((int)out+4*4);
+ emit_addpl_imm(temp,1,temp);
+ emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG);
+ emit_jns((int)out-2*4);
#endif
+ emit_writeword(sl,®_cop2d[30]);
+ emit_writeword(temp,®_cop2d[31]);
+ break;
+ case 31:
+ break;
+ default:
+ emit_writeword(sl,®_cop2d[copr]);
+ break;
+ }
}
-void c2ls_assemble(int i,struct regstat *i_regs)
+static void c2ls_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
int s,tl;
int ar;
int offset;
int memtarget=0,c=0;
- int jaddr2=0,jaddr3,type;
+ void *jaddr2=NULL;
+ enum stub_type type;
int agr=AGEN1+(i&1);
- u_int hr,reglist=0;
+ 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);
- assert(!using_tlb);
- for(hr=0;hr<HOST_REGS;hr++) {
- if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
- }
if(i_regs->regmap[HOST_CCREG]==CCREG)
reglist&=~(1<<HOST_CCREG);
// get the address
- if (opcode[i]==0x3a) { // SWC2
+ if (dops[i].opcode==0x3a) { // SWC2
ar=get_reg(i_regs->regmap,agr);
- if(ar<0) ar=get_reg(i_regs->regmap,-1);
+ if(ar<0) ar=get_reg_temp(i_regs->regmap);
reglist|=1<<ar;
} else { // LWC2
ar=tl;
if (!offset&&!c&&s>=0) ar=s;
assert(ar>=0);
- if (opcode[i]==0x3a) { // SWC2
- cop2_get_dreg(copr,tl,HOST_TEMPREG);
+ cop2_do_stall_check(0, i, i_regs, reglist);
+
+ if (dops[i].opcode==0x3a) { // SWC2
+ cop2_get_dreg(copr,tl,-1);
type=STOREW_STUB;
}
else
type=LOADW_STUB;
if(c&&!memtarget) {
- jaddr2=(int)out;
+ jaddr2=out;
emit_jmp(0); // inline_readstub/inline_writestub?
}
else {
if(!c) {
- emit_cmpimm(offset||c||s<0?ar:s,RAM_SIZE);
- jaddr2=(int)out;
- emit_jno(0);
- }
- if (opcode[i]==0x32) { // LWC2
- #ifdef HOST_IMM_ADDR32
- if(c) emit_readword_tlb(constmap[i][s]+offset,-1,tl);
- else
- #endif
- emit_readword_indexed(0,ar,tl);
+ 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
- emit_writeword_indexed(tl,0,ar);
+ 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 || offset_reg == HOST_TEMPREG)
+ host_tempreg_release();
if(jaddr2)
- add_stub(type,jaddr2,(int)out,i,ar,(int)i_regs,ccadj[i],reglist);
- if (opcode[i]==0x3a) { // SWC2
+ add_stub_r(type,jaddr2,out,i,ar,i_regs,ccadj_,reglist);
+ if(dops[i].opcode==0x3a) // SWC2
+ if(!(i_regs->waswritten&(1<<dops[i].rs1)) && !HACK_ENABLED(NDHACK_NO_SMC_CHECK)) {
#if defined(HOST_IMM8)
int ir=get_reg(i_regs->regmap,INVCP);
assert(ir>=0);
emit_cmpmem_indexedsr12_reg(ir,ar,1);
#else
- emit_cmpmem_indexedsr12_imm((int)invalid_code,ar,1);
+ emit_cmpmem_indexedsr12_imm(invalid_code,ar,1);
#endif
#if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
emit_callne(invalidate_addr_reg[ar]);
#else
- jaddr3=(int)out;
+ void *jaddr3 = out;
emit_jne(0);
- add_stub(INVCODE_STUB,jaddr3,(int)out,reglist|(1<<HOST_CCREG),ar,0,0,0);
+ add_stub(INVCODE_STUB,jaddr3,out,reglist|(1<<HOST_CCREG),ar,0,0,0);
#endif
}
- if (opcode[i]==0x32) { // LWC2
+ if (dops[i].opcode==0x32) { // LWC2
+ host_tempreg_acquire();
cop2_put_dreg(copr,tl,HOST_TEMPREG);
+ host_tempreg_release();
}
}
-#ifndef multdiv_assemble
-void multdiv_assemble(int i,struct regstat *i_regs)
+static void cop2_assemble(int i, const struct regstat *i_regs)
{
- printf("Need multdiv_assemble for this architecture.\n");
- exit(1);
-}
-#endif
+ u_int copr = (source[i]>>11) & 0x1f;
+ signed char temp = get_reg_temp(i_regs->regmap);
-void mov_assemble(int i,struct regstat *i_regs)
-{
- //if(opcode2[i]==0x10||opcode2[i]==0x12) { // MFHI/MFLO
- //if(opcode2[i]==0x11||opcode2[i]==0x13) { // MTHI/MTLO
- if(rt1[i]) {
- signed char sh,sl,th,tl;
- th=get_reg(i_regs->regmap,rt1[i]|64);
- tl=get_reg(i_regs->regmap,rt1[i]);
- //assert(tl>=0);
- if(tl>=0) {
- sh=get_reg(i_regs->regmap,rs1[i]|64);
- sl=get_reg(i_regs->regmap,rs1[i]);
- if(sl>=0) emit_mov(sl,tl);
- else emit_loadreg(rs1[i],tl);
- if(th>=0) {
- if(sh>=0) emit_mov(sh,th);
- else emit_loadreg(rs1[i]|64,th);
- }
+ if (!HACK_ENABLED(NDHACK_NO_STALLS)) {
+ u_int reglist = reglist_exclude(get_host_reglist(i_regs->regmap), temp, -1);
+ 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);
}
-}
-
-#ifndef fconv_assemble
-void fconv_assemble(int i,struct regstat *i_regs)
-{
- printf("Need fconv_assemble for this architecture.\n");
- exit(1);
-}
-#endif
-
-#if 0
-void float_assemble(int i,struct regstat *i_regs)
-{
- printf("Need float_assemble for this architecture.\n");
- exit(1);
-}
-#endif
+ 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 (dops[i].opcode2==4) { // MTC2
+ signed char sl=get_reg(i_regs->regmap,dops[i].rs1);
+ cop2_put_dreg(copr,sl,temp);
+ }
+ else if (dops[i].opcode2==2) // CFC2
+ {
+ 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 (dops[i].opcode2==6) // CTC2
+ {
+ signed char sl=get_reg(i_regs->regmap,dops[i].rs1);
+ switch(copr) {
+ case 4:
+ case 12:
+ case 20:
+ case 26:
+ case 27:
+ case 29:
+ case 30:
+ emit_signextend16(sl,temp);
+ break;
+ case 31:
+ c2op_ctc2_31_assemble(sl,temp);
+ break;
+ default:
+ temp=sl;
+ break;
+ }
+ emit_writeword(temp,®_cop2c[copr]);
+ assert(sl>=0);
+ }
+}
-void syscall_assemble(int i,struct regstat *i_regs)
+static void do_unalignedwritestub(int n)
{
- signed char ccreg=get_reg(i_regs->regmap,CCREG);
- assert(ccreg==HOST_CCREG);
- assert(!is_delayslot);
- emit_movimm(start+i*4,EAX); // Get PC
- emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); // CHECK: is this right? There should probably be an extra cycle...
- emit_jmp((int)jump_syscall_hle); // XXX
+ assem_debug("do_unalignedwritestub %x\n",start+stubs[n].a*4);
+ literal_pool(256);
+ set_jump_target(stubs[n].addr, out);
+
+ int i=stubs[n].a;
+ struct regstat *i_regs=(struct regstat *)stubs[n].c;
+ int addr=stubs[n].b;
+ u_int reglist=stubs[n].e;
+ signed char *i_regmap=i_regs->regmap;
+ int temp2=get_reg(i_regmap,FTEMP);
+ int rt;
+ rt=get_reg(i_regmap,dops[i].rs2);
+ assert(rt>=0);
+ assert(addr>=0);
+ assert(dops[i].opcode==0x2a||dops[i].opcode==0x2e); // SWL/SWR only implemented
+ reglist|=(1<<addr);
+ reglist&=~(1<<temp2);
+
+ // don't bother with it and call write handler
+ save_regs(reglist);
+ pass_args(addr,rt);
+ int cc=get_reg(i_regmap,CCREG);
+ if(cc<0)
+ emit_loadreg(CCREG,2);
+ emit_addimm(cc<0?2:cc,(int)stubs[n].d+1,2);
+ emit_far_call((dops[i].opcode==0x2a?jump_handle_swl:jump_handle_swr));
+ emit_addimm(0,-((int)stubs[n].d+1),cc<0?2:cc);
+ if(cc<0)
+ emit_storereg(CCREG,2);
+ restore_regs(reglist);
+ emit_jmp(stubs[n].retaddr); // return address
}
-void hlecall_assemble(int i,struct regstat *i_regs)
+#ifndef multdiv_assemble
+void multdiv_assemble(int i,struct regstat *i_regs)
{
- signed char ccreg=get_reg(i_regs->regmap,CCREG);
- assert(ccreg==HOST_CCREG);
- assert(!is_delayslot);
- emit_movimm(start+i*4+4,0); // Get PC
- emit_movimm((int)psxHLEt[source[i]&7],1);
- emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); // XXX
- emit_jmp((int)jump_hlecall);
+ printf("Need multdiv_assemble for this architecture.\n");
+ abort();
+}
+#endif
+
+static void mov_assemble(int i, const struct regstat *i_regs)
+{
+ //if(dops[i].opcode2==0x10||dops[i].opcode2==0x12) { // MFHI/MFLO
+ //if(dops[i].opcode2==0x11||dops[i].opcode2==0x13) { // MTHI/MTLO
+ if(dops[i].rt1) {
+ signed char sl,tl;
+ tl=get_reg(i_regs->regmap,dops[i].rt1);
+ //assert(tl>=0);
+ if(tl>=0) {
+ sl=get_reg(i_regs->regmap,dops[i].rs1);
+ if(sl>=0) emit_mov(sl,tl);
+ else emit_loadreg(dops[i].rs1,tl);
+ }
+ }
+ if (dops[i].rs1 == HIREG || dops[i].rs1 == LOREG) // MFHI/MFLO
+ multdiv_do_stall(i, i_regs);
}
-void intcall_assemble(int i,struct regstat *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, int ccadj_, u_int pc, void *func)
{
signed char ccreg=get_reg(i_regs->regmap,CCREG);
assert(ccreg==HOST_CCREG);
assert(!is_delayslot);
- emit_movimm(start+i*4,0); // Get PC
- emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
- emit_jmp((int)jump_intcall);
+ (void)ccreg;
+
+ emit_movimm(pc,3); // Get PC
+ emit_readword(&last_count,2);
+ emit_writeword(3,&psxRegs.pc);
+ 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, const struct regstat *i_regs, int ccadj_)
+{
+ // 'break' tends to be littered around to catch things like
+ // division by 0 and is almost never executed, so don't emit much code here
+ void *func = (dops[i].opcode2 == 0x0C)
+ ? (is_delayslot ? jump_syscall_ds : jump_syscall)
+ : (is_delayslot ? jump_break_ds : jump_break);
+ assert(get_reg(i_regs->regmap, CCREG) == HOST_CCREG);
+ emit_movimm(start + i*4, 2); // pc
+ emit_addimm(HOST_CCREG, ccadj_ + CLOCK_ADJUST(1), HOST_CCREG);
+ emit_far_jump(func);
+}
+
+static void hlecall_assemble(int i, const struct regstat *i_regs, int ccadj_)
+{
+ void *hlefunc = psxNULL;
+ uint32_t hleCode = source[i] & 0x03ffffff;
+ if (hleCode < ARRAY_SIZE(psxHLEt))
+ hlefunc = psxHLEt[hleCode];
+
+ call_c_cpu_handler(i, i_regs, ccadj_, start + i*4+4, hlefunc);
+}
+
+static void intcall_assemble(int i, const struct regstat *i_regs, int ccadj_)
+{
+ call_c_cpu_handler(i, i_regs, ccadj_, start + i*4, execI);
+}
+
+static void speculate_mov(int rs,int rt)
+{
+ if(rt!=0) {
+ smrv_strong_next|=1<<rt;
+ smrv[rt]=smrv[rs];
+ }
+}
+
+static void speculate_mov_weak(int rs,int rt)
+{
+ if(rt!=0) {
+ smrv_weak_next|=1<<rt;
+ smrv[rt]=smrv[rs];
+ }
+}
+
+static void speculate_register_values(int i)
+{
+ if(i==0) {
+ memcpy(smrv,psxRegs.GPR.r,sizeof(smrv));
+ // gp,sp are likely to stay the same throughout the block
+ smrv_strong_next=(1<<28)|(1<<29)|(1<<30);
+ smrv_weak_next=~smrv_strong_next;
+ //printf(" llr %08x\n", smrv[4]);
+ }
+ smrv_strong=smrv_strong_next;
+ smrv_weak=smrv_weak_next;
+ switch(dops[i].itype) {
+ case ALU:
+ 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<<dops[i].rt1);
+ smrv_weak_next&=~(1<<dops[i].rt1);
+ }
+ break;
+ case SHIFTIMM:
+ smrv_strong_next&=~(1<<dops[i].rt1);
+ smrv_weak_next&=~(1<<dops[i].rt1);
+ // fallthrough
+ case IMM16:
+ if(dops[i].rt1&&is_const(®s[i],dops[i].rt1)) {
+ int value,hr=get_reg(regs[i].regmap,dops[i].rt1);
+ if(hr>=0) {
+ if(get_final_value(hr,i,&value))
+ smrv[dops[i].rt1]=value;
+ else smrv[dops[i].rt1]=constmap[i][hr];
+ smrv_strong_next|=1<<dops[i].rt1;
+ }
+ }
+ else {
+ 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&&(dops[i].rt1==26||(smrv[dops[i].rt1]>>24)==0xa0)) {
+ // special case for BIOS
+ smrv[dops[i].rt1]=0xa0000000;
+ smrv_strong_next|=1<<dops[i].rt1;
+ break;
+ }
+ // fallthrough
+ case SHIFT:
+ case LOADLR:
+ case MOV:
+ smrv_strong_next&=~(1<<dops[i].rt1);
+ smrv_weak_next&=~(1<<dops[i].rt1);
+ break;
+ case COP0:
+ case COP2:
+ if(dops[i].opcode2==0||dops[i].opcode2==2) { // MFC/CFC
+ smrv_strong_next&=~(1<<dops[i].rt1);
+ smrv_weak_next&=~(1<<dops[i].rt1);
+ }
+ break;
+ case C2LS:
+ if (dops[i].opcode==0x32) { // LWC2
+ smrv_strong_next&=~(1<<dops[i].rt1);
+ smrv_weak_next&=~(1<<dops[i].rt1);
+ }
+ break;
+ }
+#if 0
+ int r=4;
+ printf("x %08x %08x %d %d c %08x %08x\n",smrv[r],start+i*4,
+ ((smrv_strong>>r)&1),(smrv_weak>>r)&1,regs[i].isconst,regs[i].wasconst);
+#endif
}
-void ds_assemble(int i,struct regstat *i_regs)
+static void ujump_assemble(int i, const struct regstat *i_regs);
+static void rjump_assemble(int i, const struct regstat *i_regs);
+static void cjump_assemble(int i, const struct regstat *i_regs);
+static void sjump_assemble(int i, const struct regstat *i_regs);
+static void pagespan_assemble(int i, const struct regstat *i_regs);
+
+static int assemble(int i, const struct regstat *i_regs, int ccadj_)
{
- is_delayslot=1;
- switch(itype[i]) {
+ int ds = 0;
+ switch (dops[i].itype) {
case ALU:
- alu_assemble(i,i_regs);break;
+ alu_assemble(i, i_regs);
+ break;
case IMM16:
- imm16_assemble(i,i_regs);break;
+ imm16_assemble(i, i_regs);
+ break;
case SHIFT:
- shift_assemble(i,i_regs);break;
+ shift_assemble(i, i_regs);
+ break;
case SHIFTIMM:
- shiftimm_assemble(i,i_regs);break;
+ shiftimm_assemble(i, i_regs);
+ break;
case LOAD:
- load_assemble(i,i_regs);break;
+ load_assemble(i, i_regs, ccadj_);
+ break;
case LOADLR:
- loadlr_assemble(i,i_regs);break;
+ loadlr_assemble(i, i_regs, ccadj_);
+ break;
case STORE:
- store_assemble(i,i_regs);break;
+ store_assemble(i, i_regs, ccadj_);
+ break;
case STORELR:
- storelr_assemble(i,i_regs);break;
+ storelr_assemble(i, i_regs, ccadj_);
+ break;
case COP0:
- cop0_assemble(i,i_regs);break;
+ cop0_assemble(i, i_regs, ccadj_);
+ break;
case COP1:
- cop1_assemble(i,i_regs);break;
+ cop1_assemble(i, i_regs);
+ break;
case C1LS:
- c1ls_assemble(i,i_regs);break;
+ c1ls_assemble(i, i_regs);
+ break;
case COP2:
- cop2_assemble(i,i_regs);break;
+ cop2_assemble(i, i_regs);
+ break;
case C2LS:
- c2ls_assemble(i,i_regs);break;
+ c2ls_assemble(i, i_regs, ccadj_);
+ break;
case C2OP:
- c2op_assemble(i,i_regs);break;
- case FCONV:
- fconv_assemble(i,i_regs);break;
- case FLOAT:
- float_assemble(i,i_regs);break;
- case FCOMP:
- fcomp_assemble(i,i_regs);break;
+ c2op_assemble(i, i_regs);
+ break;
case MULTDIV:
- multdiv_assemble(i,i_regs);break;
+ multdiv_assemble(i, i_regs);
+ multdiv_prepare_stall(i, i_regs, ccadj_);
+ break;
case MOV:
- mov_assemble(i,i_regs);break;
+ mov_assemble(i, i_regs);
+ break;
+ case SYSCALL:
+ syscall_assemble(i, i_regs, ccadj_);
+ break;
+ case HLECALL:
+ hlecall_assemble(i, i_regs, ccadj_);
+ break;
+ case INTCALL:
+ intcall_assemble(i, i_regs, ccadj_);
+ break;
+ case UJUMP:
+ ujump_assemble(i, i_regs);
+ ds = 1;
+ break;
+ case RJUMP:
+ rjump_assemble(i, i_regs);
+ ds = 1;
+ break;
+ case CJUMP:
+ cjump_assemble(i, i_regs);
+ ds = 1;
+ break;
+ case SJUMP:
+ sjump_assemble(i, i_regs);
+ ds = 1;
+ break;
+ case SPAN:
+ pagespan_assemble(i, i_regs);
+ break;
+ case NOP:
+ case OTHER:
+ case NI:
+ // not handled, just skip
+ break;
+ default:
+ assert(0);
+ }
+ return ds;
+}
+
+static void ds_assemble(int i, const struct regstat *i_regs)
+{
+ speculate_register_values(i);
+ is_delayslot = 1;
+ switch (dops[i].itype) {
case SYSCALL:
case HLECALL:
case INTCALL:
case RJUMP:
case CJUMP:
case SJUMP:
- case FJUMP:
- printf("Jump in the delay slot. This is probably a bug.\n");
+ SysPrintf("Jump in the delay slot. This is probably a bug.\n");
+ break;
+ default:
+ assemble(i, i_regs, ccadj[i]);
}
- is_delayslot=0;
+ is_delayslot = 0;
}
// Is the branch target a valid internal jump?
-int internal_branch(uint64_t i_is32,int addr)
+static int internal_branch(int addr)
{
if(addr&1) return 0; // Indirect (register) jump
if(addr>=start && addr<start+slen*4-4)
{
- int t=(addr-start)>>2;
- // Delay slots are not valid branch targets
- //if(t>0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP||itype[t-1]==FJUMP)) return 0;
- // 64 -> 32 bit transition requires a recompile
- /*if(is32[t]&~unneeded_reg_upper[t]&~i_is32)
- {
- if(requires_32bit[t]&~i_is32) printf("optimizable: no\n");
- else printf("optimizable: yes\n");
- }*/
- //if(is32[t]&~unneeded_reg_upper[t]&~i_is32) return 0;
-#ifndef FORCE32
- if(requires_32bit[t]&~i_is32) return 0;
- else
-#endif
- return 1;
+ return 1;
}
return 0;
}
-#ifndef wb_invalidate
-void wb_invalidate(signed char pre[],signed char entry[],uint64_t dirty,uint64_t is32,
- uint64_t u,uint64_t uu)
+static void wb_invalidate(signed char pre[],signed char entry[],uint64_t dirty,uint64_t u)
{
int hr;
for(hr=0;hr<HOST_REGS;hr++) {
if(pre[hr]>=0) {
if((dirty>>hr)&1) {
if(get_reg(entry,pre[hr])<0) {
- if(pre[hr]<64) {
- if(!((u>>pre[hr])&1)) {
- emit_storereg(pre[hr],hr);
- if( ((is32>>pre[hr])&1) && !((uu>>pre[hr])&1) ) {
- emit_sarimm(hr,31,hr);
- emit_storereg(pre[hr]|64,hr);
- }
- }
- }else{
- if(!((uu>>(pre[hr]&63))&1) && !((is32>>(pre[hr]&63))&1)) {
- emit_storereg(pre[hr],hr);
- }
- }
+ assert(pre[hr]<64);
+ if(!((u>>pre[hr])&1))
+ emit_storereg(pre[hr],hr);
}
}
}
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG) {
if(pre[hr]!=entry[hr]) {
- if(pre[hr]>=0&&(pre[hr]&63)<TEMPREG) {
+ if(pre[hr]>=0&&pre[hr]<TEMPREG) {
int nr;
if((nr=get_reg(entry,pre[hr]))>=0) {
emit_mov(hr,nr);
}
}
}
-#endif
// Load the specified registers
// This only loads the registers given as arguments because
// we don't want to load things that will be overwritten
-void load_regs(signed char entry[],signed char regmap[],int is32,int rs1,int rs2)
+static inline void load_reg(signed char entry[], signed char regmap[], int rs)
{
- int hr;
- // Load 32-bit regs
- for(hr=0;hr<HOST_REGS;hr++) {
- if(hr!=EXCLUDE_REG&®map[hr]>=0) {
- if(entry[hr]!=regmap[hr]) {
- if(regmap[hr]==rs1||regmap[hr]==rs2)
- {
- if(regmap[hr]==0) {
- emit_zeroreg(hr);
- }
- else
- {
- emit_loadreg(regmap[hr],hr);
- }
- }
- }
- }
- }
- //Load 64-bit regs
- for(hr=0;hr<HOST_REGS;hr++) {
- if(hr!=EXCLUDE_REG&®map[hr]>=0) {
- if(entry[hr]!=regmap[hr]) {
- if(regmap[hr]-64==rs1||regmap[hr]-64==rs2)
- {
- assert(regmap[hr]!=64);
- if((is32>>(regmap[hr]&63))&1) {
- int lr=get_reg(regmap,regmap[hr]-64);
- if(lr>=0)
- emit_sarimm(lr,31,hr);
- else
- emit_loadreg(regmap[hr],hr);
- }
- else
- {
- emit_loadreg(regmap[hr],hr);
- }
- }
- }
- }
- }
+ int hr = get_reg(regmap, rs);
+ if (hr >= 0 && entry[hr] != regmap[hr])
+ emit_loadreg(regmap[hr], hr);
+}
+
+static void load_regs(signed char entry[], signed char regmap[], int rs1, int rs2)
+{
+ load_reg(entry, regmap, rs1);
+ if (rs1 != rs2)
+ load_reg(entry, regmap, rs2);
}
// Load registers prior to the start of a loop
static void loop_preload(signed char pre[],signed char entry[])
{
int hr;
- for(hr=0;hr<HOST_REGS;hr++) {
- if(hr!=EXCLUDE_REG) {
- if(pre[hr]!=entry[hr]) {
- if(entry[hr]>=0) {
- if(get_reg(pre,entry[hr])<0) {
- assem_debug("loop preload:\n");
- //printf("loop preload: %d\n",hr);
- if(entry[hr]==0) {
- emit_zeroreg(hr);
- }
- else if(entry[hr]<TEMPREG)
- {
- emit_loadreg(entry[hr],hr);
- }
- else if(entry[hr]-64<TEMPREG)
- {
- emit_loadreg(entry[hr],hr);
- }
- }
- }
- }
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ int r = entry[hr];
+ if (r >= 0 && pre[hr] != r && get_reg(pre, r) < 0) {
+ assem_debug("loop preload:\n");
+ if (r < TEMPREG)
+ emit_loadreg(r, hr);
}
}
}
// Generate address for load/store instruction
// goes to AGEN for writes, FTEMP for LOADLR and cop1/2 loads
-void address_generation(int i,struct regstat *i_regs,signed char entry[])
+static void address_generation(int i, const struct regstat *i_regs, signed char entry[])
{
- if(itype[i]==LOAD||itype[i]==LOADLR||itype[i]==STORE||itype[i]==STORELR||itype[i]==C1LS||itype[i]==C2LS) {
+ if (dops[i].is_load || dops[i].is_store) {
int ra=-1;
int agr=AGEN1+(i&1);
- 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(dops[i].itype==LOAD) {
+ ra=get_reg(i_regs->regmap,dops[i].rt1);
+ if(ra<0) ra=get_reg_temp(i_regs->regmap);
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(ra<0) ra=get_reg_temp(i_regs->regmap);
}
- 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);
+ if(ra<0) ra=get_reg_temp(i_regs->regmap);
}
}
- int rs=get_reg(i_regs->regmap,rs1[i]);
- int rm=get_reg(i_regs->regmap,TLREG);
+ 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(rm>=0) {
- if(!entry||entry[rm]!=mgr) {
- generate_map_const(offset,rm);
- } // else did it in the previous cycle
- }*/
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);
} // 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(rm>=0) {
- if(!entry||entry[rm]!=mgr) {
- if(itype[i]==STORE||itype[i]==STORELR||(opcode[i]&0x3b)==0x39||(opcode[i]&0x3b)==0x3a) {
- // Stores to memory go thru the mapper to detect self-modifying
- // code, loads don't.
- if((unsigned int)(constmap[i][rs]+offset)>=0xC0000000 ||
- (unsigned int)(constmap[i][rs]+offset)<0x80000000+RAM_SIZE )
- generate_map_const(constmap[i][rs]+offset,rm);
- }else{
- if((signed int)(constmap[i][rs]+offset)>=(signed int)0xC0000000)
- generate_map_const(constmap[i][rs]+offset,rm);
- }
- }
- }
- 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{
- #ifdef HOST_IMM_ADDR32
- if((itype[i]!=LOAD&&(opcode[i]&0x3b)!=0x31&&(opcode[i]&0x3b)!=0x32) || // LWC1/LDC1/LWC2/LDC2
- (using_tlb&&((signed int)constmap[i][rs]+offset)>=(signed int)0xC0000000))
- #endif
emit_movimm(constmap[i][rs]+offset,ra);
+ regs[i].loadedconst|=1<<ra;
}
} // 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{
}
}
// 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;
- #ifndef HOST_IMM_ADDR32
- // Mapper entry
- agr=MGEN1+((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 offset=imm[i+1];
- int c=(regs[i+1].wasconst>>rs)&1;
- if(c) {
- if(itype[i+1]==STORE||itype[i+1]==STORELR
- ||(opcode[i+1]&0x3b)==0x39||(opcode[i+1]&0x3b)==0x3a) { // SWC1/SDC1, SWC2/SDC2
- // Stores to memory go thru the mapper to detect self-modifying
- // code, loads don't.
- if((unsigned int)(constmap[i+1][rs]+offset)>=0xC0000000 ||
- (unsigned int)(constmap[i+1][rs]+offset)<0x80000000+RAM_SIZE )
- generate_map_const(constmap[i+1][rs]+offset,ra);
- }else{
- if((signed int)(constmap[i+1][rs]+offset)>=(signed int)0xC0000000)
- generate_map_const(constmap[i+1][rs]+offset,ra);
- }
- }
- /*else if(rs1[i]==0) {
- generate_map_const(offset,ra);
- }*/
- }
- #endif
// 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{
- #ifdef HOST_IMM_ADDR32
- if((itype[i+1]!=LOAD&&(opcode[i+1]&0x3b)!=0x31&&(opcode[i+1]&0x3b)!=0x32) || // LWC1/LDC1/LWC2/LDC2
- (using_tlb&&((signed int)constmap[i+1][rs]+offset)>=(signed int)0xC0000000))
- #endif
emit_movimm(constmap[i+1][rs]+offset,ra);
+ regs[i+1].loadedconst|=1<<ra;
}
}
- else if(rs1[i+1]==0) {
+ else if(dops[i+1].rs1==0) {
// Using r0 as a base address
- if (opcode[i+1]==0x22||opcode[i+1]==0x26) {
+ if (dops[i+1].opcode==0x22||dops[i+1].opcode==0x26) {
emit_movimm(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(offset&0xFFFFFFF8,ra); // LDL/LDR
}else{
emit_movimm(offset,ra);
}
}
-int get_final_value(int hr, int i, int *value)
+static int get_final_value(int hr, int i, int *value)
{
int reg=regs[i].regmap[hr];
while(i<slen-1) {
if(regs[i+1].regmap[hr]!=reg) break;
if(!((regs[i+1].isconst>>hr)&1)) break;
- if(bt[i+1]) break;
+ if(dops[i+1].bt) break;
i++;
}
if(i<slen-1) {
- if(itype[i]==UJUMP||itype[i]==RJUMP||itype[i]==CJUMP||itype[i]==SJUMP) {
+ if (dops[i].is_jump) {
*value=constmap[i][hr];
return 1;
}
- if(!bt[i+1]) {
- if(itype[i+1]==UJUMP||itype[i+1]==RJUMP||itype[i+1]==CJUMP||itype[i+1]==SJUMP) {
+ if(!dops[i+1].bt) {
+ if (dops[i+1].is_jump) {
// Load in delay slot, out-of-order execution
- if(itype[i+2]==LOAD&&rs1[i+2]==reg&&rt1[i+2]==reg&&((regs[i+1].wasconst>>hr)&1))
+ if(dops[i+2].itype==LOAD&&dops[i+2].rs1==reg&&dops[i+2].rt1==reg&&((regs[i+1].wasconst>>hr)&1))
{
- #ifdef HOST_IMM_ADDR32
- if(!using_tlb||((signed int)constmap[i][hr]+imm[i+2])<(signed int)0xC0000000) return 0;
- #endif
// 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)
{
- #ifdef HOST_IMM_ADDR32
- if(!using_tlb||((signed int)constmap[i][hr]+imm[i+1])<(signed int)0xC0000000) return 0;
- #endif
// Precompute load address
*value=constmap[i][hr]+imm[i+1];
- //printf("c=%x imm=%x\n",(int)constmap[i][hr],imm[i+1]);
+ //printf("c=%x imm=%lx\n",(long)constmap[i][hr],imm[i+1]);
return 1;
}
}
}
*value=constmap[i][hr];
- //printf("c=%x\n",(int)constmap[i][hr]);
+ //printf("c=%lx\n",(long)constmap[i][hr]);
if(i==slen-1) return 1;
- if(reg<64) {
- return !((unneeded_reg[i+1]>>reg)&1);
- }else{
- return !((unneeded_reg_upper[i+1]>>reg)&1);
- }
+ assert(reg < 64);
+ return !((unneeded_reg[i+1]>>reg)&1);
}
// Load registers with known constants
-void load_consts(signed char pre[],signed char regmap[],int is32,int i)
+static void load_consts(signed char pre[],signed char regmap[],int i)
{
- int hr;
+ int hr,hr2;
+ // propagate loaded constant flags
+ if(i==0||dops[i].bt)
+ regs[i].loadedconst=0;
+ else {
+ for(hr=0;hr<HOST_REGS;hr++) {
+ if(hr!=EXCLUDE_REG&®map[hr]>=0&&((regs[i-1].isconst>>hr)&1)&&pre[hr]==regmap[hr]
+ &®map[hr]==regs[i-1].regmap[hr]&&((regs[i-1].loadedconst>>hr)&1))
+ {
+ regs[i].loadedconst|=1<<hr;
+ }
+ }
+ }
// Load 32-bit regs
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG&®map[hr]>=0) {
//if(entry[hr]!=regmap[hr]) {
- if(i==0||!((regs[i-1].isconst>>hr)&1)||pre[hr]!=regmap[hr]||bt[i]) {
- if(((regs[i].isconst>>hr)&1)&®map[hr]<64&®map[hr]>0) {
- int value;
+ if(!((regs[i].loadedconst>>hr)&1)) {
+ assert(regmap[hr]<64);
+ if(((regs[i].isconst>>hr)&1)&®map[hr]>0) {
+ int value,similar=0;
if(get_final_value(hr,i,&value)) {
- if(value==0) {
+ // see if some other register has similar value
+ for(hr2=0;hr2<HOST_REGS;hr2++) {
+ if(hr2!=EXCLUDE_REG&&((regs[i].loadedconst>>hr2)&1)) {
+ if(is_similar_value(value,constmap[i][hr2])) {
+ similar=1;
+ break;
+ }
+ }
+ }
+ if(similar) {
+ int value2;
+ if(get_final_value(hr2,i,&value2)) // is this needed?
+ emit_movimm_from(value2,hr2,value,hr);
+ else
+ emit_movimm(value,hr);
+ }
+ else if(value==0) {
emit_zeroreg(hr);
}
else {
emit_movimm(value,hr);
}
}
- }
- }
- }
- }
- // Load 64-bit regs
- for(hr=0;hr<HOST_REGS;hr++) {
- if(hr!=EXCLUDE_REG&®map[hr]>=0) {
- //if(entry[hr]!=regmap[hr]) {
- if(i==0||!((regs[i-1].isconst>>hr)&1)||pre[hr]!=regmap[hr]||bt[i]) {
- if(((regs[i].isconst>>hr)&1)&®map[hr]>64) {
- if((is32>>(regmap[hr]&63))&1) {
- int lr=get_reg(regmap,regmap[hr]-64);
- assert(lr>=0);
- emit_sarimm(lr,31,hr);
- }
- else
- {
- int value;
- if(get_final_value(hr,i,&value)) {
- if(value==0) {
- emit_zeroreg(hr);
- }
- else {
- emit_movimm(value,hr);
- }
- }
- }
+ regs[i].loadedconst|=1<<hr;
}
}
}
}
}
-void load_all_consts(signed char regmap[],int is32,u_int dirty,int i)
+
+static void load_all_consts(const signed char regmap[], u_int dirty, int i)
{
int hr;
// Load 32-bit regs
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG&®map[hr]>=0&&((dirty>>hr)&1)) {
- if(((regs[i].isconst>>hr)&1)&®map[hr]<64&®map[hr]>0) {
+ assert(regmap[hr] < 64);
+ if(((regs[i].isconst>>hr)&1)&®map[hr]>0) {
int value=constmap[i][hr];
if(value==0) {
emit_zeroreg(hr);
}
}
}
- // Load 64-bit regs
- for(hr=0;hr<HOST_REGS;hr++) {
- if(hr!=EXCLUDE_REG&®map[hr]>=0&&((dirty>>hr)&1)) {
- if(((regs[i].isconst>>hr)&1)&®map[hr]>64) {
- if((is32>>(regmap[hr]&63))&1) {
- int lr=get_reg(regmap,regmap[hr]-64);
- assert(lr>=0);
- emit_sarimm(lr,31,hr);
- }
- else
- {
- int value=constmap[i][hr];
- if(value==0) {
- emit_zeroreg(hr);
- }
- else {
- emit_movimm(value,hr);
- }
- }
- }
- }
- }
}
// Write out all dirty registers (except cycle count)
-void wb_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty)
+static void wb_dirtys(const signed char i_regmap[], uint64_t i_dirty)
{
int hr;
for(hr=0;hr<HOST_REGS;hr++) {
if(i_regmap[hr]>0) {
if(i_regmap[hr]!=CCREG) {
if((i_dirty>>hr)&1) {
- if(i_regmap[hr]<64) {
- emit_storereg(i_regmap[hr],hr);
-#ifndef FORCE32
- if( ((i_is32>>i_regmap[hr])&1) ) {
- #ifdef DESTRUCTIVE_WRITEBACK
- emit_sarimm(hr,31,hr);
- emit_storereg(i_regmap[hr]|64,hr);
- #else
- emit_sarimm(hr,31,HOST_TEMPREG);
- emit_storereg(i_regmap[hr]|64,HOST_TEMPREG);
- #endif
- }
-#endif
- }else{
- if( !((i_is32>>(i_regmap[hr]&63))&1) ) {
- emit_storereg(i_regmap[hr],hr);
- }
- }
+ assert(i_regmap[hr]<64);
+ emit_storereg(i_regmap[hr],hr);
}
}
}
}
}
}
+
// Write out dirty registers that we need to reload (pair with load_needed_regs)
// This writes the registers not written by store_regs_bt
-void wb_needed_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+static void wb_needed_dirtys(const signed char i_regmap[], uint64_t i_dirty, int addr)
{
int hr;
int t=(addr-start)>>2;
if(hr!=EXCLUDE_REG) {
if(i_regmap[hr]>0) {
if(i_regmap[hr]!=CCREG) {
- if(i_regmap[hr]==regs[t].regmap_entry[hr] && ((regs[t].dirty>>hr)&1) && !(((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
+ if(i_regmap[hr]==regs[t].regmap_entry[hr] && ((regs[t].dirty>>hr)&1)) {
if((i_dirty>>hr)&1) {
- if(i_regmap[hr]<64) {
- emit_storereg(i_regmap[hr],hr);
-#ifndef FORCE32
- if( ((i_is32>>i_regmap[hr])&1) ) {
- #ifdef DESTRUCTIVE_WRITEBACK
- emit_sarimm(hr,31,hr);
- emit_storereg(i_regmap[hr]|64,hr);
- #else
- emit_sarimm(hr,31,HOST_TEMPREG);
- emit_storereg(i_regmap[hr]|64,HOST_TEMPREG);
- #endif
- }
-#endif
- }else{
- if( !((i_is32>>(i_regmap[hr]&63))&1) ) {
- emit_storereg(i_regmap[hr],hr);
- }
- }
+ assert(i_regmap[hr]<64);
+ emit_storereg(i_regmap[hr],hr);
}
}
}
}
// 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;hr<HOST_REGS;hr++) {
emit_zeroreg(hr);
}
else
- if(i_regmap[hr]>0 && (i_regmap[hr]&63)<TEMPREG && i_regmap[hr]!=CCREG)
+ if(i_regmap[hr]>0 && i_regmap[hr]<TEMPREG && i_regmap[hr]!=CCREG)
{
emit_loadreg(i_regmap[hr],hr);
}
}
// Load all current registers also needed by next instruction
-void load_needed_regs(signed char i_regmap[],signed char next_regmap[])
+static void load_needed_regs(const signed char i_regmap[], const signed char next_regmap[])
{
int hr;
for(hr=0;hr<HOST_REGS;hr++) {
emit_zeroreg(hr);
}
else
- if(i_regmap[hr]>0 && (i_regmap[hr]&63)<TEMPREG && i_regmap[hr]!=CCREG)
+ if(i_regmap[hr]>0 && i_regmap[hr]<TEMPREG && i_regmap[hr]!=CCREG)
{
emit_loadreg(i_regmap[hr],hr);
}
}
// Load all regs, storing cycle count if necessary
-void load_regs_entry(int t)
+static void load_regs_entry(int t)
{
int hr;
- if(is_ds[t]) emit_addimm(HOST_CCREG,CLOCK_DIVIDER,HOST_CCREG);
- else if(ccadj[t]) emit_addimm(HOST_CCREG,-ccadj[t]*CLOCK_DIVIDER,HOST_CCREG);
+ if(dops[t].is_ds) emit_addimm(HOST_CCREG,CLOCK_ADJUST(1),HOST_CCREG);
+ else if(ccadj[t]) emit_addimm(HOST_CCREG,-ccadj[t],HOST_CCREG);
if(regs[t].regmap_entry[HOST_CCREG]!=CCREG) {
emit_storereg(CCREG,HOST_CCREG);
}
}
}
}
- // Load 64-bit regs
- for(hr=0;hr<HOST_REGS;hr++) {
- if(regs[t].regmap_entry[hr]>=64&®s[t].regmap_entry[hr]<TEMPREG+64) {
- assert(regs[t].regmap_entry[hr]!=64);
- if((regs[t].was32>>(regs[t].regmap_entry[hr]&63))&1) {
- int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
- if(lr<0) {
- emit_loadreg(regs[t].regmap_entry[hr],hr);
- }
- else
- {
- emit_sarimm(lr,31,hr);
- }
- }
- else
- {
- emit_loadreg(regs[t].regmap_entry[hr],hr);
- }
- }
- }
}
// Store dirty registers prior to branch
-void store_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+static void store_regs_bt(signed char i_regmap[],uint64_t i_dirty,int addr)
{
- if(internal_branch(i_is32,addr))
+ if(internal_branch(addr))
{
int t=(addr-start)>>2;
int hr;
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG) {
if(i_regmap[hr]>0 && i_regmap[hr]!=CCREG) {
- if(i_regmap[hr]!=regs[t].regmap_entry[hr] || !((regs[t].dirty>>hr)&1) || (((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
+ if(i_regmap[hr]!=regs[t].regmap_entry[hr] || !((regs[t].dirty>>hr)&1)) {
if((i_dirty>>hr)&1) {
- if(i_regmap[hr]<64) {
- if(!((unneeded_reg[t]>>i_regmap[hr])&1)) {
- emit_storereg(i_regmap[hr],hr);
- if( ((i_is32>>i_regmap[hr])&1) && !((unneeded_reg_upper[t]>>i_regmap[hr])&1) ) {
- #ifdef DESTRUCTIVE_WRITEBACK
- emit_sarimm(hr,31,hr);
- emit_storereg(i_regmap[hr]|64,hr);
- #else
- emit_sarimm(hr,31,HOST_TEMPREG);
- emit_storereg(i_regmap[hr]|64,HOST_TEMPREG);
- #endif
- }
- }
- }else{
- if( !((i_is32>>(i_regmap[hr]&63))&1) && !((unneeded_reg_upper[t]>>(i_regmap[hr]&63))&1) ) {
- emit_storereg(i_regmap[hr],hr);
- }
- }
+ assert(i_regmap[hr]<64);
+ if(!((unneeded_reg[t]>>i_regmap[hr])&1))
+ emit_storereg(i_regmap[hr],hr);
}
}
}
else
{
// Branch out of this block, write out all dirty regs
- wb_dirtys(i_regmap,i_is32,i_dirty);
+ wb_dirtys(i_regmap,i_dirty);
}
}
// Load all needed registers for branch target
-void load_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+static void load_regs_bt(signed char i_regmap[],uint64_t i_dirty,int addr)
{
//if(addr>=start && addr<(start+slen*4))
- if(internal_branch(i_is32,addr))
+ if(internal_branch(addr))
{
int t=(addr-start)>>2;
int hr;
// Load 32-bit regs
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG&®s[t].regmap_entry[hr]>=0&®s[t].regmap_entry[hr]<TEMPREG) {
- #ifdef DESTRUCTIVE_WRITEBACK
- if(i_regmap[hr]!=regs[t].regmap_entry[hr] || ( !((regs[t].dirty>>hr)&1) && ((i_dirty>>hr)&1) && (((i_is32&~unneeded_reg_upper[t])>>i_regmap[hr])&1) ) || (((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
- #else
- if(i_regmap[hr]!=regs[t].regmap_entry[hr] ) {
- #endif
+ if(i_regmap[hr]!=regs[t].regmap_entry[hr]) {
if(regs[t].regmap_entry[hr]==0) {
emit_zeroreg(hr);
}
}
}
}
- //Load 64-bit regs
- for(hr=0;hr<HOST_REGS;hr++) {
- if(hr!=EXCLUDE_REG&®s[t].regmap_entry[hr]>=64&®s[t].regmap_entry[hr]<TEMPREG+64) {
- if(i_regmap[hr]!=regs[t].regmap_entry[hr]) {
- assert(regs[t].regmap_entry[hr]!=64);
- if((i_is32>>(regs[t].regmap_entry[hr]&63))&1) {
- int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
- if(lr<0) {
- emit_loadreg(regs[t].regmap_entry[hr],hr);
- }
- else
- {
- emit_sarimm(lr,31,hr);
- }
- }
- else
- {
- emit_loadreg(regs[t].regmap_entry[hr],hr);
- }
- }
- else if((i_is32>>(regs[t].regmap_entry[hr]&63))&1) {
- int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
- assert(lr>=0);
- emit_sarimm(lr,31,hr);
- }
- }
- }
}
}
-int match_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+static int match_bt(signed char i_regmap[],uint64_t i_dirty,int addr)
{
if(addr>=start && addr<start+slen*4-4)
{
{
return 0;
}
- else
+ else
if((i_dirty>>hr)&1)
{
if(i_regmap[hr]<TEMPREG)
}
else if(i_regmap[hr]>=64&&i_regmap[hr]<TEMPREG+64)
{
- if(!((unneeded_reg_upper[t]>>(i_regmap[hr]&63))&1))
- return 0;
+ assert(0);
}
}
}
}
}
}
- if((((regs[t].was32^i_is32)&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)
- {
- //printf("%x: is32 no match\n",addr);
- return 0;
- }
}
}
}
- //if(is32[t]&~unneeded_reg_upper[t]&~i_is32) return 0;
-#ifndef FORCE32
- if(requires_32bit[t]&~i_is32) return 0;
-#endif
// Delay slots are not valid branch targets
- //if(t>0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP||itype[t-1]==FJUMP)) return 0;
+ //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
{
return 1;
}
+#ifdef DRC_DBG
+static void drc_dbg_emit_do_cmp(int i, int ccadj_)
+{
+ extern void do_insn_cmp();
+ //extern int cycle;
+ u_int hr, reglist = get_host_reglist(regs[i].regmap);
+
+ assem_debug("//do_insn_cmp %08x\n", start+i*4);
+ save_regs(reglist);
+ // write out changed consts to match the interpreter
+ if (i > 0 && !dops[i].bt) {
+ for (hr = 0; hr < HOST_REGS; 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))
+ continue;
+ if (i > 1 && reg == regs[i-2].regmap[hr] && constmap[i-1][hr] == constmap[i-2][hr])
+ continue;
+ emit_movimm(constmap[i-1][hr],0);
+ emit_storereg(reg, 0);
+ }
+ }
+ 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);
+ //emit_writeword(0,&cycle);
+ (void)get_reg2;
+ restore_regs(reglist);
+ assem_debug("\\\\do_insn_cmp\n");
+}
+#else
+#define drc_dbg_emit_do_cmp(x,y)
+#endif
+
// Used when a branch jumps into the delay slot of another branch
-void ds_assemble_entry(int i)
+static void ds_assemble_entry(int i)
{
- int t=(ba[i]-start)>>2;
- if(!instr_addr[t]) instr_addr[t]=(u_int)out;
+ 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, 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,regs[t].was32);
- load_regs(regs[t].regmap_entry,regs[t].regmap,regs[t].was32,rs1[t],rs2[t]);
+ wb_register(CCREG,regs[t].regmap_entry,regs[t].wasdirty);
+ 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)
- load_regs(regs[t].regmap_entry,regs[t].regmap,regs[t].was32,INVCP,INVCP);
- cop1_usable=0;
+ if (ram_offset && (dops[t].is_load || dops[t].is_store))
+ load_reg(regs[t].regmap_entry,regs[t].regmap,ROREG);
+ if (dops[t].is_store)
+ load_reg(regs[t].regmap_entry,regs[t].regmap,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 FCONV:
- fconv_assemble(t,®s[t]);break;
- case FLOAT:
- float_assemble(t,®s[t]);break;
- case FCOMP:
- fcomp_assemble(t,®s[t]);break;
- case MULTDIV:
- multdiv_assemble(t,®s[t]);break;
- case MOV:
- mov_assemble(t,®s[t]);break;
+ switch (dops[t].itype) {
case SYSCALL:
case HLECALL:
case INTCALL:
case RJUMP:
case CJUMP:
case SJUMP:
- case FJUMP:
- printf("Jump in the delay slot. This is probably a bug.\n");
+ 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].is32,regs[t].dirty,ba[i]+4);
- load_regs_bt(regs[t].regmap,regs[t].is32,regs[t].dirty,ba[i]+4);
- if(internal_branch(regs[t].is32,ba[i]+4))
+ store_regs_bt(regs[t].regmap,regs[t].dirty,ba[i]+4);
+ load_regs_bt(regs[t].regmap,regs[t].dirty,ba[i]+4);
+ if(internal_branch(ba[i]+4))
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- assert(internal_branch(regs[t].is32,ba[i]+4));
- add_to_linker((int)out,ba[i]+4,internal_branch(regs[t].is32,ba[i]+4));
+ assert(internal_branch(ba[i]+4));
+ add_to_linker(out,ba[i]+4,internal_branch(ba[i]+4));
emit_jmp(0);
}
-void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert)
+static void emit_extjump(void *addr, u_int target)
+{
+ emit_extjump2(addr, target, dyna_linker);
+}
+
+static void emit_extjump_ds(void *addr, u_int target)
+{
+ emit_extjump2(addr, target, dyna_linker_ds);
+}
+
+// Load 2 immediates optimizing for small code size
+static void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2)
+{
+ emit_movimm(imm1,rt1);
+ emit_movimm_from(imm1,rt1,imm2,rt2);
+}
+
+static void do_cc(int i, const signed char i_regmap[], int *adj,
+ int addr, int taken, int invert)
{
- int count;
- int jaddr;
- int idle=0;
- if(itype[i]==RJUMP)
+ int count, count_plus2;
+ void *jaddr;
+ void *idle=NULL;
+ int t=0;
+ if(dops[i].itype==RJUMP)
{
*adj=0;
}
//if(ba[i]>=start && ba[i]<(start+slen*4))
- if(internal_branch(branch_regs[i].is32,ba[i]))
+ if(internal_branch(ba[i]))
{
- int t=(ba[i]-start)>>2;
- if(is_ds[t]) *adj=-1; // Branch into delay slot adds an extra cycle
+ t=(ba[i]-start)>>2;
+ 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);
- idle=(int)out;
+ idle=out;
//emit_subfrommem(&idlecount,HOST_CCREG); // Count idle cycles
emit_andimm(HOST_CCREG,3,HOST_CCREG);
- jaddr=(int)out;
+ jaddr=out;
emit_jmp(0);
}
else if(*adj==0||invert) {
- emit_addimm_and_set_flags(CLOCK_DIVIDER*(count+2),HOST_CCREG);
- jaddr=(int)out;
+ int cycles = count_plus2;
+ // faster loop HACK
+#if 0
+ if (t&&*adj) {
+ int rel=t-i;
+ if(-NO_CYCLE_PENALTY_THR<rel&&rel<0)
+ cycles=*adj+count+2-*adj;
+ }
+#endif
+ emit_addimm_and_set_flags(cycles, HOST_CCREG);
+ jaddr = out;
emit_jns(0);
}
else
{
- emit_cmpimm(HOST_CCREG,-CLOCK_DIVIDER*(count+2));
- jaddr=(int)out;
+ emit_cmpimm(HOST_CCREG, -count_plus2);
+ jaddr = out;
emit_jns(0);
}
- add_stub(CC_STUB,jaddr,idle?idle:(int)out,(*adj==0||invert||idle)?0:(count+2),i,addr,taken,0);
+ add_stub(CC_STUB,jaddr,idle?idle:out,(*adj==0||invert||idle)?0:count_plus2,i,addr,taken,0);
}
-void do_ccstub(int n)
+static void do_ccstub(int n)
{
literal_pool(256);
- assem_debug("do_ccstub %x\n",start+stubs[n][4]*4);
- set_jump_target(stubs[n][1],(int)out);
- int i=stubs[n][4];
- if(stubs[n][6]==NULLDS) {
+ assem_debug("do_ccstub %x\n",start+(u_int)stubs[n].b*4);
+ set_jump_target(stubs[n].addr, out);
+ int i=stubs[n].b;
+ if(stubs[n].d==NULLDS) {
// Delay slot instruction is nullified ("likely" branch)
- wb_dirtys(regs[i].regmap,regs[i].is32,regs[i].dirty);
+ wb_dirtys(regs[i].regmap,regs[i].dirty);
}
- else if(stubs[n][6]!=TAKEN) {
- wb_dirtys(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty);
+ else if(stubs[n].d!=TAKEN) {
+ wb_dirtys(branch_regs[i].regmap,branch_regs[i].dirty);
}
else {
- if(internal_branch(branch_regs[i].is32,ba[i]))
- wb_needed_dirtys(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ if(internal_branch(ba[i]))
+ wb_needed_dirtys(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
}
- if(stubs[n][5]!=-1)
+ if(stubs[n].c!=-1)
{
// Save PC as return address
- emit_movimm(stubs[n][5],EAX);
- emit_writeword(EAX,(int)&pcaddr);
+ emit_movimm(stubs[n].c,EAX);
+ emit_writeword(EAX,&pcaddr);
}
else
{
// Return address depends on which way the branch goes
- if(itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+ if(dops[i].itype==CJUMP||dops[i].itype==SJUMP)
{
- int s1l=get_reg(branch_regs[i].regmap,rs1[i]);
- int s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
- int s2l=get_reg(branch_regs[i].regmap,rs2[i]);
- int s2h=get_reg(branch_regs[i].regmap,rs2[i]|64);
- if(rs1[i]==0)
+ int s1l=get_reg(branch_regs[i].regmap,dops[i].rs1);
+ int s2l=get_reg(branch_regs[i].regmap,dops[i].rs2);
+ if(dops[i].rs1==0)
{
- s1l=s2l;s1h=s2h;
- s2l=s2h=-1;
+ s1l=s2l;
+ s2l=-1;
}
- else if(rs2[i]==0)
+ else if(dops[i].rs2==0)
{
- s2l=s2h=-1;
- }
- if((branch_regs[i].is32>>rs1[i])&(branch_regs[i].is32>>rs2[i])&1) {
- s1h=s2h=-1;
+ s2l=-1;
}
assert(s1l>=0);
#ifdef DESTRUCTIVE_WRITEBACK
- if(rs1[i]) {
- if((branch_regs[i].dirty>>s1l)&(branch_regs[i].is32>>rs1[i])&1)
- emit_loadreg(rs1[i],s1l);
- }
+ if(dops[i].rs1) {
+ if((branch_regs[i].dirty>>s1l)&&1)
+ emit_loadreg(dops[i].rs1,s1l);
+ }
else {
- if((branch_regs[i].dirty>>s1l)&(branch_regs[i].is32>>rs2[i])&1)
- emit_loadreg(rs2[i],s1l);
+ if((branch_regs[i].dirty>>s1l)&1)
+ emit_loadreg(dops[i].rs2,s1l);
}
if(s2l>=0)
- if((branch_regs[i].dirty>>s2l)&(branch_regs[i].is32>>rs2[i])&1)
- emit_loadreg(rs2[i],s2l);
+ if((branch_regs[i].dirty>>s2l)&1)
+ emit_loadreg(dops[i].rs2,s2l);
#endif
int hr=0;
int addr=-1,alt=-1,ntaddr=-1;
while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
- (branch_regs[i].regmap[hr]&63)!=rs1[i] &&
- (branch_regs[i].regmap[hr]&63)!=rs2[i] )
+ branch_regs[i].regmap[hr]!=dops[i].rs1 &&
+ branch_regs[i].regmap[hr]!=dops[i].rs2 )
{
addr=hr++;break;
}
while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
- (branch_regs[i].regmap[hr]&63)!=rs1[i] &&
- (branch_regs[i].regmap[hr]&63)!=rs2[i] )
+ branch_regs[i].regmap[hr]!=dops[i].rs1 &&
+ branch_regs[i].regmap[hr]!=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(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
- (branch_regs[i].regmap[hr]&63)!=rs1[i] &&
- (branch_regs[i].regmap[hr]&63)!=rs2[i] )
+ branch_regs[i].regmap[hr]!=dops[i].rs1 &&
+ branch_regs[i].regmap[hr]!=dops[i].rs2 )
{
ntaddr=hr;break;
}
}
assert(hr<HOST_REGS);
}
- if((opcode[i]&0x2f)==4) // BEQ
+ if((dops[i].opcode&0x2f)==4) // BEQ
{
#ifdef HAVE_CMOV_IMM
- if(s1h<0) {
- if(s2l>=0) emit_cmp(s1l,s2l);
- else emit_test(s1l,s1l);
- emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
- }
- else
+ if(s2l>=0) emit_cmp(s1l,s2l);
+ else emit_test(s1l,s1l);
+ emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
+ #else
+ emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
+ if(s2l>=0) emit_cmp(s1l,s2l);
+ else emit_test(s1l,s1l);
+ emit_cmovne_reg(alt,addr);
#endif
- {
- emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
- if(s1h>=0) {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- emit_cmovne_reg(alt,addr);
- }
- if(s2l>=0) emit_cmp(s1l,s2l);
- else emit_test(s1l,s1l);
- emit_cmovne_reg(alt,addr);
- }
}
- if((opcode[i]&0x2f)==5) // BNE
+ if((dops[i].opcode&0x2f)==5) // BNE
{
#ifdef HAVE_CMOV_IMM
- if(s1h<0) {
- if(s2l>=0) emit_cmp(s1l,s2l);
- else emit_test(s1l,s1l);
- emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
- }
- else
+ if(s2l>=0) emit_cmp(s1l,s2l);
+ else emit_test(s1l,s1l);
+ emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
+ #else
+ emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
+ if(s2l>=0) emit_cmp(s1l,s2l);
+ else emit_test(s1l,s1l);
+ emit_cmovne_reg(alt,addr);
#endif
- {
- emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
- if(s1h>=0) {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- emit_cmovne_reg(alt,addr);
- }
- if(s2l>=0) emit_cmp(s1l,s2l);
- else emit_test(s1l,s1l);
- emit_cmovne_reg(alt,addr);
- }
}
- 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);
emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
emit_cmpimm(s1l,1);
- if(s1h>=0) emit_mov(addr,ntaddr);
emit_cmovl_reg(alt,addr);
- if(s1h>=0) {
- emit_test(s1h,s1h);
- emit_cmovne_reg(ntaddr,addr);
- emit_cmovs_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);
emit_mov2imm_compact(ba[i],addr,start+i*4+8,ntaddr);
emit_cmpimm(s1l,1);
- if(s1h>=0) emit_mov(addr,alt);
emit_cmovl_reg(ntaddr,addr);
- if(s1h>=0) {
- emit_test(s1h,s1h);
- emit_cmovne_reg(alt,addr);
- emit_cmovs_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);
emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
- if(s1h>=0) emit_test(s1h,s1h);
- else emit_test(s1l,s1l);
+ 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);
emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
- if(s1h>=0) emit_test(s1h,s1h);
- else emit_test(s1l,s1l);
+ 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);
emit_cmovne_reg(alt,addr);
}
}
- emit_writeword(addr,(int)&pcaddr);
+ 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,(int)&pcaddr);
+ emit_writeword(r,&pcaddr);
}
- else {printf("Unknown branch type in do_ccstub\n");exit(1);}
+ else {SysPrintf("Unknown branch type in do_ccstub\n");abort();}
}
// Update cycle count
assert(branch_regs[i].regmap[HOST_CCREG]==CCREG||branch_regs[i].regmap[HOST_CCREG]==-1);
- if(stubs[n][3]) emit_addimm(HOST_CCREG,CLOCK_DIVIDER*stubs[n][3],HOST_CCREG);
- emit_call((int)cc_interrupt);
- if(stubs[n][3]) emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*stubs[n][3],HOST_CCREG);
- if(stubs[n][6]==TAKEN) {
- if(internal_branch(branch_regs[i].is32,ba[i]))
+ 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,-(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((int)&pcaddr,get_reg(branch_regs[i].regmap,RTEMP));
+ 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][6]==NOTTAKEN) {
+ }else if(stubs[n].d==NOTTAKEN) {
if(i<slen-2) load_needed_regs(branch_regs[i].regmap,regmap_pre[i+2]);
else load_all_regs(branch_regs[i].regmap);
- }else if(stubs[n][6]==NULLDS) {
+ }else if(stubs[n].d==NULLDS) {
// Delay slot instruction is nullified ("likely" branch)
if(i<slen-2) load_needed_regs(regs[i].regmap,regmap_pre[i+2]);
else load_all_regs(regs[i].regmap);
}else{
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);
- emit_writeword(EAX,(int)&Count);
- emit_call((int)gen_interupt);
- emit_readword((int)&Count,HOST_CCREG);
- emit_readword((int)&next_interupt,EAX);
- emit_readword((int)&pending_exception,EBX);
- emit_writeword(EAX,(int)&last_count);
- emit_sub(HOST_CCREG,EAX,HOST_CCREG);
- emit_test(EBX,EBX);
- int jne_instr=(int)out;
- emit_jne(0);
- if(stubs[n][3]) emit_addimm(HOST_CCREG,-2*stubs[n][3],HOST_CCREG);
- load_all_regs(branch_regs[i].regmap);
- emit_jmp(stubs[n][2]); // return address
- set_jump_target(jne_instr,(int)out);
- emit_readword((int)&pcaddr,EAX);
- // Call get_addr_ht instead of doing the hash table here.
- // This code is executed infrequently and takes up a lot of space
- // so smaller is better.
- emit_storereg(CCREG,HOST_CCREG);
- emit_pushreg(EAX);
- emit_call((int)get_addr_ht);
- emit_loadreg(CCREG,HOST_CCREG);
- emit_addimm(ESP,4,ESP);
- emit_jmpreg(EAX);*/
-}
-
-add_to_linker(int addr,int target,int ext)
-{
- link_addr[linkcount][0]=addr;
- link_addr[linkcount][1]=target;
- link_addr[linkcount][2]=ext;
+ if (stubs[n].retaddr)
+ emit_jmp(stubs[n].retaddr);
+ else
+ do_jump_vaddr(stubs[n].e);
+}
+
+static void add_to_linker(void *addr, u_int target, int ext)
+{
+ assert(linkcount < ARRAY_SIZE(link_addr));
+ link_addr[linkcount].addr = addr;
+ link_addr[linkcount].target = target;
+ link_addr[linkcount].ext = ext;
linkcount++;
}
return_address=start+i*4+8;
if(rt>=0) {
#ifdef USE_MINI_HT
- if(internal_branch(branch_regs[i].is32,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;
#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);
+ if(i_regmap[temp]!=PTEMP) emit_movimm((uintptr_t)hash_table_get(return_address),temp);
}
#endif
emit_movimm(return_address,rt); // PC into link register
#ifdef IMM_PREFETCH
- emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+ emit_prefetch(hash_table_get(return_address));
#endif
}
}
}
-void ujump_assemble(int i,struct regstat *i_regs)
+static void ujump_assemble(int i, const struct regstat *i_regs)
{
- signed char *i_regmap=i_regs->regmap;
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;
- 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);
+ if(get_reg(branch_regs[i].regmap,31)>0)
+ 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;
- uint64_t bc_unneeded_upper=branch_regs[i].uu;
- bc_unneeded|=1|(1LL<<rt1[i]);
- bc_unneeded_upper|=1|(1LL<<rt1[i]);
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- bc_unneeded,bc_unneeded_upper);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
- if(!ra_done&&rt1[i]==31)
+ bc_unneeded|=1|(1LL<<dops[i].rt1);
+ wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,bc_unneeded);
+ load_reg(regs[i].regmap,branch_regs[i].regmap,CCREG);
+ if(!ra_done&&dops[i].rt1==31)
ujump_assemble_write_ra(i);
int cc,adj;
cc=get_reg(branch_regs[i].regmap,CCREG);
assert(cc==HOST_CCREG);
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
#ifdef REG_PREFETCH
- if(rt1[i]==31&&temp>=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_DIVIDER*(ccadj[i]+2-adj),cc);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
- if(internal_branch(branch_regs[i].is32,ba[i]))
+ 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(branch_regs[i].is32,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 {
- add_to_linker((int)out,ba[i],internal_branch(branch_regs[i].is32,ba[i]));
+ add_to_linker(out,ba[i],internal_branch(ba[i]));
emit_jmp(0);
}
}
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;
#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);
+ if(i_regmap[temp]!=PTEMP) emit_movimm((uintptr_t)hash_table_get(return_address),temp);
}
#endif
emit_movimm(return_address,rt); // PC into link register
#ifdef IMM_PREFETCH
- emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+ emit_prefetch(hash_table_get(return_address));
#endif
}
-void rjump_assemble(int i,struct regstat *i_regs)
+static void rjump_assemble(int i, const struct regstat *i_regs)
{
- signed char *i_regmap=i_regs->regmap;
int temp;
- int rs,cc,adj;
+ 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);
}
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;
int return_address=start+i*4+8;
- if(i_regmap[temp]==PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+ if(i_regmap[temp]==PTEMP) emit_movimm((uintptr_t)hash_table_get(return_address),temp);
}
}
#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;
- uint64_t bc_unneeded_upper=branch_regs[i].uu;
- bc_unneeded|=1|(1LL<<rt1[i]);
- bc_unneeded_upper|=1|(1LL<<rt1[i]);
- bc_unneeded&=~(1LL<<rs1[i]);
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- bc_unneeded,bc_unneeded_upper);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],CCREG);
- if(!ra_done&&rt1[i]!=0)
+ bc_unneeded|=1|(1LL<<dops[i].rt1);
+ bc_unneeded&=~(1LL<<dops[i].rs1);
+ wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,bc_unneeded);
+ load_regs(regs[i].regmap,branch_regs[i].regmap,dops[i].rs1,CCREG);
+ if(!ra_done&&dops[i].rt1!=0)
rjump_assemble_write_ra(i);
cc=get_reg(branch_regs[i].regmap,CCREG);
assert(cc==HOST_CCREG);
+ (void)cc;
#ifdef USE_MINI_HT
int rh=get_reg(branch_regs[i].regmap,RHASH);
int ht=get_reg(branch_regs[i].regmap,RHTBL);
- if(rs1[i]==31) {
+ if(dops[i].rs1==31) {
if(regs[i].regmap[rh]!=RHASH) do_preload_rhash(rh);
do_preload_rhtbl(ht);
do_rhash(rs,rh);
}
#endif
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,-1);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,-1);
#ifdef DESTRUCTIVE_WRITEBACK
- if((branch_regs[i].dirty>>rs)&(branch_regs[i].is32>>rs1[i])&1) {
- if(rs1[i]!=rt1[i+1]&&rs1[i]!=rt2[i+1]) {
- emit_loadreg(rs1[i],rs);
+ if((branch_regs[i].dirty>>rs)&1) {
+ 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_DIVIDER*(ccadj[i]+2),HOST_CCREG);
- add_stub(CC_STUB,(int)out,jump_vaddr_reg[rs],0,i,-1,TAKEN,0);
- emit_jns(0);
- //load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,-1);
+ emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG);
+ add_stub(CC_STUB,out,NULL,0,i,-1,TAKEN,rs);
+ 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
#endif
{
- //if(rs!=EAX) emit_mov(rs,EAX);
- //emit_jmp((int)jump_vaddr_eax);
- emit_jmp(jump_vaddr_reg[rs]);
- }
- /* Check hash table
- temp=!rs;
- emit_mov(rs,temp);
- emit_shrimm(rs,16,rs);
- emit_xor(temp,rs,rs);
- emit_movzwl_reg(rs,rs);
- emit_shlimm(rs,4,rs);
- emit_cmpmem_indexed((int)hash_table,rs,temp);
- emit_jne((int)out+14);
- emit_readword_indexed((int)hash_table+4,rs,rs);
- emit_jmpreg(rs);
- emit_cmpmem_indexed((int)hash_table+8,rs,temp);
- emit_addimm_no_flags(8,rs);
- emit_jeq((int)out-17);
- // No hit on hash table, call compiler
- emit_pushreg(temp);
-//DEBUG >
-#ifdef DEBUG_CYCLE_COUNT
- emit_readword((int)&last_count,ECX);
- emit_add(HOST_CCREG,ECX,HOST_CCREG);
- emit_readword((int)&next_interupt,ECX);
- emit_writeword(HOST_CCREG,(int)&Count);
- emit_sub(HOST_CCREG,ECX,HOST_CCREG);
- emit_writeword(ECX,(int)&last_count);
-#endif
-//DEBUG <
- emit_storereg(CCREG,HOST_CCREG);
- emit_call((int)get_addr);
- emit_loadreg(CCREG,HOST_CCREG);
- emit_addimm(ESP,4,ESP);
- emit_jmpreg(EAX);*/
+ do_jump_vaddr(rs);
+ }
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
- if(rt1[i]!=31&&i<slen-2&&(((u_int)out)&7)) emit_mov(13,13);
+ if(dops[i].rt1!=31&&i<slen-2&&(((u_int)out)&7)) emit_mov(13,13);
#endif
}
-void cjump_assemble(int i,struct regstat *i_regs)
+static void cjump_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].is32,branch_regs[i].dirty,ba[i]);
+ match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
assem_debug("match=%d\n",match);
- int s1h,s1l,s2h,s2l;
- int prev_cop1_usable=cop1_usable;
+ int s1l,s2l;
int unconditional=0,nop=0;
- int only32=0;
int invert=0;
- int internal=internal_branch(branch_regs[i].is32,ba[i]);
+ int internal=internal_branch(ba[i]);
if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
if(!match) invert=1;
#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);
- s2l=get_reg(branch_regs[i].regmap,rs2[i]);
- s2h=get_reg(branch_regs[i].regmap,rs2[i]|64);
+ #ifdef __aarch64__
+ invert=1; // because of near cond. branches
+ #endif
+
+ 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]);
- s1h=get_reg(i_regmap,rs1[i]|64);
- s2l=get_reg(i_regmap,rs2[i]);
- s2h=get_reg(i_regmap,rs2[i]|64);
+ 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;s1h=s2h;
- s2l=s2h=-1;
- only32=(regs[i].was32>>rs2[i])&1;
+ s1l=s2l;
+ s2l=-1;
}
- else if(rs2[i]==0)
+ else if(dops[i].rs2==0)
{
- s2l=s2h=-1;
- only32=(regs[i].was32>>rs1[i])&1;
- }
- else {
- only32=(regs[i].was32>>rs1[i])&(regs[i].was32>>rs2[i])&1;
+ 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;
- uint64_t bc_unneeded_upper=branch_regs[i].uu;
- bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
- bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
+ bc_unneeded&=~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
bc_unneeded|=1;
- bc_unneeded_upper|=1;
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- bc_unneeded,bc_unneeded_upper);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs2[i]);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+ wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,bc_unneeded);
+ load_regs(regs[i].regmap,branch_regs[i].regmap,dops[i].rs1,dops[i].rs2);
+ load_reg(regs[i].regmap,branch_regs[i].regmap,CCREG);
cc=get_reg(branch_regs[i].regmap,CCREG);
assert(cc==HOST_CCREG);
- if(unconditional)
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ if(unconditional)
+ store_regs_bt(branch_regs[i].regmap,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");
if(unconditional) {
do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0);
if(i!=(ba[i]-start)>>2 || source[i+1]!=0) {
- if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ 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 {
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jmp(0);
}
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
}
}
else if(nop) {
- emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
- int jaddr=(int)out;
+ emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc);
+ void *jaddr=out;
emit_jns(0);
- add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+ add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
}
else {
- int taken=0,nottaken=0,nottaken1=0;
+ 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_DIVIDER*(ccadj[i]+2-adj),cc);
- if(!only32)
- {
- assert(s1h>=0);
- if(opcode[i]==4) // BEQ
- {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- nottaken1=(int)out;
- emit_jne(1);
- }
- if(opcode[i]==5) // BNE
- {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- if(invert) taken=(int)out;
- else add_to_linker((int)out,ba[i],internal);
- emit_jne(0);
- }
- if(opcode[i]==6) // BLEZ
- {
- emit_test(s1h,s1h);
- if(invert) taken=(int)out;
- else add_to_linker((int)out,ba[i],internal);
- emit_js(0);
- nottaken1=(int)out;
- emit_jne(1);
- }
- if(opcode[i]==7) // BGTZ
- {
- emit_test(s1h,s1h);
- nottaken1=(int)out;
- emit_js(1);
- if(invert) taken=(int)out;
- else add_to_linker((int)out,ba[i],internal);
- emit_jne(0);
- }
- } // if(!only32)
-
+ 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);
if(invert){
- nottaken=(int)out;
- emit_jne(1);
+ nottaken=out;
+ emit_jne(DJT_1);
}else{
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
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);
if(invert){
- nottaken=(int)out;
- emit_jeq(1);
+ nottaken=out;
+ emit_jeq(DJT_1);
}else{
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jne(0);
}
}
- if(opcode[i]==6) // BLEZ
+ if(dops[i].opcode==6) // BLEZ
{
emit_cmpimm(s1l,1);
if(invert){
- nottaken=(int)out;
- emit_jge(1);
+ nottaken=out;
+ emit_jge(DJT_1);
}else{
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jl(0);
}
}
- if(opcode[i]==7) // BGTZ
+ if(dops[i].opcode==7) // BGTZ
{
emit_cmpimm(s1l,1);
if(invert){
- nottaken=(int)out;
- emit_jl(1);
+ nottaken=out;
+ emit_jl(DJT_1);
}else{
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jge(0);
}
}
if(invert) {
- if(taken) set_jump_target(taken,(int)out);
+ 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_DIVIDER*adj,cc);
- add_to_linker((int)out,ba[i],internal);
+ emit_addimm(cc,-adj,cc);
+ add_to_linker(out,ba[i],internal);
}else{
emit_addnop(13);
- add_to_linker((int)out,ba[i],internal*2);
+ add_to_linker(out,ba[i],internal*2);
}
emit_jmp(0);
}else
#endif
{
- if(adj) emit_addimm(cc,-CLOCK_DIVIDER*adj,cc);
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ 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 {
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jmp(0);
}
}
- set_jump_target(nottaken,(int)out);
+ set_jump_target(nottaken, out);
}
- if(nottaken1) set_jump_target(nottaken1,(int)out);
+ if(nottaken1) set_jump_target(nottaken1, out);
if(adj) {
- if(!invert) emit_addimm(cc,CLOCK_DIVIDER*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");
- int taken=0,nottaken=0,nottaken1=0;
+ void *taken = NULL, *nottaken = NULL, *nottaken1 = NULL;
if(!unconditional&&!nop) {
- if(!only32)
- {
- assert(s1h>=0);
- if((opcode[i]&0x2f)==4) // BEQ
- {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- nottaken1=(int)out;
- emit_jne(2);
- }
- if((opcode[i]&0x2f)==5) // BNE
- {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- taken=(int)out;
- emit_jne(1);
- }
- if((opcode[i]&0x2f)==6) // BLEZ
- {
- emit_test(s1h,s1h);
- taken=(int)out;
- emit_js(1);
- nottaken1=(int)out;
- emit_jne(2);
- }
- if((opcode[i]&0x2f)==7) // BGTZ
- {
- emit_test(s1h,s1h);
- nottaken1=(int)out;
- emit_js(2);
- taken=(int)out;
- 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
+ if((dops[i].opcode&0x2f)==4) // BEQ
{
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
- nottaken=(int)out;
- emit_jne(2);
+ 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=(int)out;
- emit_jeq(2);
+ nottaken=out;
+ emit_jeq(DJT_2);
}
- if((opcode[i]&0x2f)==6) // BLEZ
+ if((dops[i].opcode&0x2f)==6) // BLEZ
{
emit_cmpimm(s1l,1);
- nottaken=(int)out;
- emit_jge(2);
+ nottaken=out;
+ emit_jge(DJT_2);
}
- if((opcode[i]&0x2f)==7) // BGTZ
+ if((dops[i].opcode&0x2f)==7) // BGTZ
{
emit_cmpimm(s1l,1);
- nottaken=(int)out;
- emit_jl(2);
+ nottaken=out;
+ emit_jl(DJT_2);
}
} // if(!unconditional)
int adj;
uint64_t ds_unneeded=branch_regs[i].u;
- uint64_t ds_unneeded_upper=branch_regs[i].uu;
- ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
- ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
+ ds_unneeded&=~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
ds_unneeded|=1;
- ds_unneeded_upper|=1;
// branch taken
if(!nop) {
- if(taken) set_jump_target(taken,(int)out);
+ if(taken) set_jump_target(taken, out);
assem_debug("1:\n");
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- ds_unneeded,ds_unneeded_upper);
+ 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,regs[i].was32,rs1[i+1],rs2[i+1]);
+ 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,regs[i].was32,CCREG,INVCP);
+ if (ram_offset)
+ load_reg(regs[i].regmap,branch_regs[i].regmap,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) {
// CHECK: Is the following instruction (fall thru) allocated ok?
}
assert(cc==HOST_CCREG);
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
assem_debug("cycle count (adj)\n");
- if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ 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 {
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jmp(0);
}
}
// branch not taken
- cop1_usable=prev_cop1_usable;
if(!unconditional) {
- if(nottaken1) set_jump_target(nottaken1,(int)out);
- set_jump_target(nottaken,(int)out);
+ 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,regs[i].is32,
- ds_unneeded,ds_unneeded_upper);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
- address_generation(i+1,&branch_regs[i],0);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,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_reg(regs[i].regmap,branch_regs[i].regmap,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_DIVIDER*(ccadj[i]+2),HOST_CCREG);
- int jaddr=(int)out;
+ emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG);
+ void *jaddr=out;
emit_jns(0);
- add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+ add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
emit_storereg(CCREG,HOST_CCREG);
}
else{
cc=get_reg(i_regmap,CCREG);
assert(cc==HOST_CCREG);
- emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
- int jaddr=(int)out;
+ emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc);
+ void *jaddr=out;
emit_jns(0);
- add_stub(CC_STUB,jaddr,(int)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);
}
}
}
}
-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].is32,branch_regs[i].dirty,ba[i]);
- assem_debug("smatch=%d\n",match);
- int s1h,s1l;
- int prev_cop1_usable=cop1_usable;
+ match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+ assem_debug("smatch=%d ooo=%d\n", match, dops[i].ooo);
+ int s1l;
int unconditional=0,nevertaken=0;
- int only32=0;
int invert=0;
- int internal=internal_branch(branch_regs[i].is32,ba[i]);
+ int internal=internal_branch(ba[i]);
if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
if(!match) invert=1;
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
if(i>(ba[i]-start)>>2) invert=1;
#endif
+ #ifdef __aarch64__
+ 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]);
- s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
+ if(dops[i].ooo) {
+ s1l=get_reg(branch_regs[i].regmap,dops[i].rs1);
}
else {
- s1l=get_reg(i_regmap,rs1[i]);
- s1h=get_reg(i_regmap,rs1[i]|64);
+ 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);
- }
- else {
- only32=(regs[i].was32>>rs1[i])&1;
+ //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;
- uint64_t bc_unneeded_upper=branch_regs[i].uu;
- bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
- bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
+ bc_unneeded&=~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
bc_unneeded|=1;
- bc_unneeded_upper|=1;
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- bc_unneeded,bc_unneeded_upper);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs1[i]);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
- if(rt1[i]==31) {
+ wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,bc_unneeded);
+ load_regs(regs[i].regmap,branch_regs[i].regmap,dops[i].rs1,dops[i].rs1);
+ load_reg(regs[i].regmap,branch_regs[i].regmap,CCREG);
+ if(dops[i].rt1==31) {
int rt,return_address;
rt=get_reg(branch_regs[i].regmap,31);
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]);
return_address=start+i*4+8;
emit_movimm(return_address,rt); // PC into link register
#ifdef IMM_PREFETCH
- if(!nevertaken) emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+ if(!nevertaken) emit_prefetch(hash_table_get(return_address));
#endif
}
}
cc=get_reg(branch_regs[i].regmap,CCREG);
assert(cc==HOST_CCREG);
- if(unconditional)
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ if(unconditional)
+ store_regs_bt(branch_regs[i].regmap,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");
if(unconditional) {
do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0);
if(i!=(ba[i]-start)>>2 || source[i+1]!=0) {
- if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ 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 {
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jmp(0);
}
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
}
}
else if(nevertaken) {
- emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
- int jaddr=(int)out;
+ emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc);
+ void *jaddr=out;
emit_jns(0);
- add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+ add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
}
else {
- int nottaken=0;
+ void *nottaken = NULL;
do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
- if(adj&&!invert) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
- if(!only32)
- {
- assert(s1h>=0);
- if((opcode2[i]&0xf)==0) // BLTZ/BLTZAL
- {
- emit_test(s1h,s1h);
- if(invert){
- nottaken=(int)out;
- emit_jns(1);
- }else{
- add_to_linker((int)out,ba[i],internal);
- emit_js(0);
- }
- }
- if((opcode2[i]&0xf)==1) // BGEZ/BLTZAL
- {
- emit_test(s1h,s1h);
- if(invert){
- nottaken=(int)out;
- emit_js(1);
- }else{
- add_to_linker((int)out,ba[i],internal);
- emit_jns(0);
- }
- }
- } // if(!only32)
- else
+ 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){
- nottaken=(int)out;
- emit_jns(1);
+ nottaken=out;
+ emit_jns(DJT_1);
}else{
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_js(0);
}
}
- if((opcode2[i]&0xf)==1) // BGEZ/BLTZAL
+ if((dops[i].opcode2&0xf)==1) // BGEZ/BLTZAL
{
emit_test(s1l,s1l);
if(invert){
- nottaken=(int)out;
- emit_js(1);
+ nottaken=out;
+ emit_js(DJT_1);
}else{
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jns(0);
}
}
- } // if(!only32)
-
+ }
+
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_DIVIDER*adj,cc);
- add_to_linker((int)out,ba[i],internal);
+ emit_addimm(cc,-adj,cc);
+ add_to_linker(out,ba[i],internal);
}else{
emit_addnop(13);
- add_to_linker((int)out,ba[i],internal*2);
+ add_to_linker(out,ba[i],internal*2);
}
emit_jmp(0);
}else
#endif
{
- if(adj) emit_addimm(cc,-CLOCK_DIVIDER*adj,cc);
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ 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 {
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jmp(0);
}
}
- set_jump_target(nottaken,(int)out);
+ set_jump_target(nottaken, out);
}
if(adj) {
- if(!invert) emit_addimm(cc,CLOCK_DIVIDER*adj,cc);
+ if(!invert) emit_addimm(cc,adj,cc);
}
} // (!unconditional)
} // if(ooo)
{
// In-order execution (branch first)
//printf("IOE\n");
- int nottaken=0;
- if(rt1[i]==31) {
+ void *nottaken = NULL;
+ if(dops[i].rt1==31) {
int rt,return_address;
rt=get_reg(branch_regs[i].regmap,31);
if(rt>=0) {
return_address=start+i*4+8;
emit_movimm(return_address,rt); // PC into link register
#ifdef IMM_PREFETCH
- emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+ emit_prefetch(hash_table_get(return_address));
#endif
}
}
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]);
- if(!only32)
- {
- assert(s1h>=0);
- if((opcode2[i]&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL
- {
- emit_test(s1h,s1h);
- nottaken=(int)out;
- emit_jns(1);
- }
- if((opcode2[i]&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL
- {
- emit_test(s1h,s1h);
- nottaken=(int)out;
- emit_js(1);
- }
- } // if(!only32)
- else
- {
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=(int)out;
- emit_jns(1);
+ 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=(int)out;
- emit_js(1);
+ nottaken=out;
+ emit_js(DJT_1);
}
- }
} // if(!unconditional)
int adj;
uint64_t ds_unneeded=branch_regs[i].u;
- uint64_t ds_unneeded_upper=branch_regs[i].uu;
- ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
- ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
+ ds_unneeded&=~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
ds_unneeded|=1;
- ds_unneeded_upper|=1;
// branch taken
if(!nevertaken) {
//assem_debug("1:\n");
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- ds_unneeded,ds_unneeded_upper);
+ 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,regs[i].was32,rs1[i+1],rs2[i+1]);
+ 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,regs[i].was32,CCREG,INVCP);
+ if (ram_offset)
+ load_reg(regs[i].regmap,branch_regs[i].regmap,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) {
// CHECK: Is the following instruction (fall thru) allocated ok?
}
assert(cc==HOST_CCREG);
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
assem_debug("cycle count (adj)\n");
- if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+ 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 {
- add_to_linker((int)out,ba[i],internal);
+ add_to_linker(out,ba[i],internal);
emit_jmp(0);
}
}
// branch not taken
- cop1_usable=prev_cop1_usable;
if(!unconditional) {
- set_jump_target(nottaken,(int)out);
+ set_jump_target(nottaken, out);
assem_debug("1:\n");
- if(!likely[i]) {
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- ds_unneeded,ds_unneeded_upper);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
- address_generation(i+1,&branch_regs[i],0);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,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);
+ if (ram_offset)
+ load_reg(regs[i].regmap,branch_regs[i].regmap,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_DIVIDER*(ccadj[i]+2),HOST_CCREG);
- int jaddr=(int)out;
+ emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG);
+ void *jaddr=out;
emit_jns(0);
- add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+ add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
emit_storereg(CCREG,HOST_CCREG);
}
else{
cc=get_reg(i_regmap,CCREG);
assert(cc==HOST_CCREG);
- emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
- int jaddr=(int)out;
+ emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc);
+ void *jaddr=out;
emit_jns(0);
- add_stub(CC_STUB,jaddr,(int)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);
}
}
}
}
-void fjump_assemble(int i,struct regstat *i_regs)
+static void pagespan_assemble(int i, const struct regstat *i_regs)
{
- signed char *i_regmap=i_regs->regmap;
- int cc;
- int match;
- match=match_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
- assem_debug("fmatch=%d\n",match);
- int fs,cs;
- int eaddr;
- int invert=0;
- int internal=internal_branch(branch_regs[i].is32,ba[i]);
- if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
- if(!match) invert=1;
- #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
- if(i>(ba[i]-start)>>2) invert=1;
- #endif
-
- if(ooo[i]) {
- fs=get_reg(branch_regs[i].regmap,FSREG);
- address_generation(i+1,i_regs,regs[i].regmap_entry); // Is this okay?
- }
- else {
- fs=get_reg(i_regmap,FSREG);
+ 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(dops[i].rs1==0)
+ {
+ s1l=s2l;
+ s2l=-1;
}
-
- // Check cop1 unusable
- if(!cop1_usable) {
- cs=get_reg(i_regmap,CSREG);
- assert(cs>=0);
- emit_testimm(cs,0x20000000);
- eaddr=(int)out;
- emit_jeq(0);
- add_stub(FP_STUB,eaddr,(int)out,i,cs,(int)i_regs,0,0);
- cop1_usable=1;
+ else if(dops[i].rs2==0)
+ {
+ s2l=-1;
}
-
- if(ooo[i]) {
- // Out of order execution (delay slot first)
- //printf("OOOE\n");
- ds_assemble(i+1,i_regs);
- int adj;
- uint64_t bc_unneeded=branch_regs[i].u;
- uint64_t bc_unneeded_upper=branch_regs[i].uu;
- bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
- bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
- bc_unneeded|=1;
- bc_unneeded_upper|=1;
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- bc_unneeded,bc_unneeded_upper);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs1[i]);
- 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);
- do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
- assem_debug("cycle count (adj)\n");
- if(1) {
- int nottaken=0;
- if(adj&&!invert) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
- if(1) {
- assert(fs>=0);
- emit_testimm(fs,0x800000);
- if(source[i]&0x10000) // BC1T
- {
- if(invert){
- nottaken=(int)out;
- emit_jeq(1);
- }else{
- add_to_linker((int)out,ba[i],internal);
- emit_jne(0);
- }
- }
- else // BC1F
- if(invert){
- nottaken=(int)out;
- emit_jne(1);
- }else{
- add_to_linker((int)out,ba[i],internal);
- emit_jeq(0);
- }
- {
- }
- } // if(!only32)
-
- if(invert) {
- if(adj) emit_addimm(cc,-CLOCK_DIVIDER*adj,cc);
- #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
- else if(match) emit_addnop(13);
- #endif
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,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]) {
- ds_assemble_entry(i);
- }
- else {
- add_to_linker((int)out,ba[i],internal);
- emit_jmp(0);
- }
- set_jump_target(nottaken,(int)out);
- }
-
- if(adj) {
- if(!invert) emit_addimm(cc,CLOCK_DIVIDER*adj,cc);
+ int hr=0;
+ int addr=-1,alt=-1,ntaddr=-1;
+ if(i_regs->regmap[HOST_BTREG]<0) {addr=HOST_BTREG;}
+ else {
+ while(hr<HOST_REGS)
+ {
+ if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
+ i_regs->regmap[hr]!=dops[i].rs1 &&
+ i_regs->regmap[hr]!=dops[i].rs2 )
+ {
+ addr=hr++;break;
}
- } // (!unconditional)
- } // if(ooo)
- else
- {
- // In-order execution (branch first)
- //printf("IOE\n");
- int nottaken=0;
- if(1) {
- //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]);
- if(1) {
- assert(fs>=0);
- emit_testimm(fs,0x800000);
- if(source[i]&0x10000) // BC1T
- {
- nottaken=(int)out;
- emit_jeq(1);
- }
- else // BC1F
- {
- nottaken=(int)out;
- emit_jne(1);
- }
- }
- } // if(!unconditional)
- int adj;
- uint64_t ds_unneeded=branch_regs[i].u;
- uint64_t ds_unneeded_upper=branch_regs[i].uu;
- ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
- ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
- ds_unneeded|=1;
- ds_unneeded_upper|=1;
- // branch taken
- //assem_debug("1:\n");
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- ds_unneeded,ds_unneeded_upper);
- // load regs
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
- address_generation(i+1,&branch_regs[i],0);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,INVCP);
- ds_assemble(i+1,&branch_regs[i]);
- cc=get_reg(branch_regs[i].regmap,CCREG);
- if(cc==-1) {
- emit_loadreg(CCREG,cc=HOST_CCREG);
- // CHECK: Is the following instruction (fall thru) allocated ok?
- }
- assert(cc==HOST_CCREG);
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
- do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
- assem_debug("cycle count (adj)\n");
- if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,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]) {
- ds_assemble_entry(i);
- }
- else {
- add_to_linker((int)out,ba[i],internal);
- emit_jmp(0);
- }
-
- // branch not taken
- if(1) { // <- FIXME (don't need this)
- set_jump_target(nottaken,(int)out);
- assem_debug("1:\n");
- if(!likely[i]) {
- wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
- ds_unneeded,ds_unneeded_upper);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
- address_generation(i+1,&branch_regs[i],0);
- load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
- ds_assemble(i+1,&branch_regs[i]);
- }
- cc=get_reg(branch_regs[i].regmap,CCREG);
- if(cc==-1&&!likely[i]) {
- // 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_DIVIDER*(ccadj[i]+2),HOST_CCREG);
- int jaddr=(int)out;
- emit_jns(0);
- add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
- emit_storereg(CCREG,HOST_CCREG);
- }
- else{
- cc=get_reg(i_regmap,CCREG);
- assert(cc==HOST_CCREG);
- emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
- int jaddr=(int)out;
- emit_jns(0);
- add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
- }
- }
- }
-}
-
-static void pagespan_assemble(int i,struct regstat *i_regs)
-{
- int s1l=get_reg(i_regs->regmap,rs1[i]);
- int s1h=get_reg(i_regs->regmap,rs1[i]|64);
- int s2l=get_reg(i_regs->regmap,rs2[i]);
- int s2h=get_reg(i_regs->regmap,rs2[i]|64);
- void *nt_branch=NULL;
- int taken=0;
- int nottaken=0;
- int unconditional=0;
- if(rs1[i]==0)
- {
- s1l=s2l;s1h=s2h;
- s2l=s2h=-1;
- }
- else if(rs2[i]==0)
- {
- s2l=s2h=-1;
- }
- if((i_regs->is32>>rs1[i])&(i_regs->is32>>rs2[i])&1) {
- s1h=s2h=-1;
- }
- int hr=0;
- int addr,alt,ntaddr;
- if(i_regs->regmap[HOST_BTREG]<0) {addr=HOST_BTREG;}
- else {
- while(hr<HOST_REGS)
- {
- if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
- (i_regs->regmap[hr]&63)!=rs1[i] &&
- (i_regs->regmap[hr]&63)!=rs2[i] )
- {
- addr=hr++;break;
- }
- hr++;
- }
- }
- while(hr<HOST_REGS)
+ hr++;
+ }
+ }
+ while(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG && hr!=HOST_BTREG &&
- (i_regs->regmap[hr]&63)!=rs1[i] &&
- (i_regs->regmap[hr]&63)!=rs2[i] )
+ i_regs->regmap[hr]!=dops[i].rs1 &&
+ i_regs->regmap[hr]!=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(hr<HOST_REGS)
{
if(hr!=EXCLUDE_REG && hr!=HOST_CCREG && hr!=HOST_BTREG &&
- (i_regs->regmap[hr]&63)!=rs1[i] &&
- (i_regs->regmap[hr]&63)!=rs2[i] )
+ i_regs->regmap[hr]!=dops[i].rs1 &&
+ i_regs->regmap[hr]!=dops[i].rs2 )
{
ntaddr=hr;break;
}
}
}
assert(hr<HOST_REGS);
- if((opcode[i]&0x2e)==4||opcode[i]==0x11) { // BEQ/BNE/BEQL/BNEL/BC1
- load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,CCREG,CCREG);
+ if((dops[i].opcode&0x2e)==4||dops[i].opcode==0x11) { // BEQ/BNE/BEQL/BNEL/BC1
+ load_reg(regs[i].regmap_entry,regs[i].regmap,CCREG);
}
- emit_addimm(HOST_CCREG,CLOCK_DIVIDER*(ccadj[i]+2),HOST_CCREG);
- if(opcode[i]==2) // J
+ emit_addimm(HOST_CCREG, ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG);
+ if(dops[i].opcode==2) // J
{
unconditional=1;
}
- if(opcode[i]==3) // JAL
+ if(dops[i].opcode==3) // JAL
{
// TODO: mini_ht
int rt=get_reg(i_regs->regmap,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;
}
else
#ifdef HAVE_CMOV_IMM
- if(s1h<0) {
+ if(1) {
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
{
assert(s1l>=0);
emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
- if(s1h>=0) {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- emit_cmovne_reg(alt,addr);
- }
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
emit_cmovne_reg(alt,addr);
}
}
- if((opcode[i]&0x3f)==5) // BNE
+ if((dops[i].opcode&0x3f)==5) // BNE
{
#ifdef HAVE_CMOV_IMM
- if(s1h<0) {
- if(s2l>=0) emit_cmp(s1l,s2l);
- else emit_test(s1l,s1l);
- emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
- }
- else
+ if(s2l>=0) emit_cmp(s1l,s2l);
+ else emit_test(s1l,s1l);
+ emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
+ #else
+ assert(s1l>=0);
+ emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
+ if(s2l>=0) emit_cmp(s1l,s2l);
+ else emit_test(s1l,s1l);
+ emit_cmovne_reg(alt,addr);
#endif
- {
- assert(s1l>=0);
- emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
- if(s1h>=0) {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- emit_cmovne_reg(alt,addr);
- }
- if(s2l>=0) emit_cmp(s1l,s2l);
- else emit_test(s1l,s1l);
- emit_cmovne_reg(alt,addr);
- }
}
- if((opcode[i]&0x3f)==0x14) // BEQL
+ if((dops[i].opcode&0x3f)==0x14) // BEQL
{
- if(s1h>=0) {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- nottaken=(int)out;
- emit_jne(0);
- }
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
- if(nottaken) set_jump_target(nottaken,(int)out);
- nottaken=(int)out;
+ if(nottaken) set_jump_target(nottaken, out);
+ nottaken=out;
emit_jne(0);
}
- if((opcode[i]&0x3f)==0x15) // BNEL
+ if((dops[i].opcode&0x3f)==0x15) // BNEL
{
- if(s1h>=0) {
- if(s2h>=0) emit_cmp(s1h,s2h);
- else emit_test(s1h,s1h);
- taken=(int)out;
- emit_jne(0);
- }
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
- nottaken=(int)out;
+ nottaken=out;
emit_jeq(0);
- if(taken) set_jump_target(taken,(int)out);
+ 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);
- if(s1h>=0) emit_mov(addr,ntaddr);
emit_cmovl_reg(alt,addr);
- if(s1h>=0) {
- emit_test(s1h,s1h);
- emit_cmovne_reg(ntaddr,addr);
- emit_cmovs_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);
- if(s1h>=0) emit_mov(addr,alt);
emit_cmovl_reg(ntaddr,addr);
- if(s1h>=0) {
- emit_test(s1h,s1h);
- emit_cmovne_reg(alt,addr);
- emit_cmovs_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);
if((source[i]&0x30000)==0x20000) // BC1FL
{
emit_testimm(s1l,0x800000);
- nottaken=(int)out;
+ nottaken=out;
emit_jne(0);
}
if((source[i]&0x30000)==0x30000) // BC1TL
{
emit_testimm(s1l,0x800000);
- nottaken=(int)out;
+ nottaken=out;
emit_jeq(0);
}
}
assert(i_regs->regmap[HOST_CCREG]==CCREG);
- wb_dirtys(regs[i].regmap,regs[i].is32,regs[i].dirty);
- if(likely[i]||unconditional)
+ wb_dirtys(regs[i].regmap,regs[i].dirty);
+ if(unconditional)
{
emit_movimm(ba[i],HOST_BTREG);
}
int target_addr=start+i*4+5;
void *stub=out;
void *compiled_target_addr=check_addr(target_addr);
- emit_extjump_ds((int)branch_addr,target_addr);
+ emit_extjump_ds(branch_addr, target_addr);
if(compiled_target_addr) {
- set_jump_target((int)branch_addr,(int)compiled_target_addr);
- add_link(target_addr,stub);
- }
- else set_jump_target((int)branch_addr,(int)stub);
- if(likely[i]) {
- // Not-taken path
- set_jump_target((int)nottaken,(int)out);
- wb_dirtys(regs[i].regmap,regs[i].is32,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((int)branch_addr,target_addr);
- if(compiled_target_addr) {
- set_jump_target((int)branch_addr,(int)compiled_target_addr);
- add_link(target_addr,stub);
- }
- else set_jump_target((int)branch_addr,(int)stub);
+ set_jump_target(branch_addr, compiled_target_addr);
+ add_jump_out(target_addr,stub);
}
+ else set_jump_target(branch_addr, stub);
}
// Assemble the delay slot for the above
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,regs[0].was32);
+ wb_register(CCREG,regs[0].regmap_entry,regs[0].wasdirty);
if(regs[0].regmap[HOST_BTREG]!=BTREG)
- emit_writeword(HOST_BTREG,(int)&branch_target);
- load_regs(regs[0].regmap_entry,regs[0].regmap,regs[0].was32,rs1[0],rs2[0]);
+ emit_writeword(HOST_BTREG,&branch_target);
+ 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)
- load_regs(regs[0].regmap_entry,regs[0].regmap,regs[0].was32,INVCP,INVCP);
- cop1_usable=0;
+ if (ram_offset && (dops[0].is_load || dops[0].is_store))
+ load_reg(regs[0].regmap_entry,regs[0].regmap,ROREG);
+ if (dops[0].is_store)
+ load_reg(regs[0].regmap_entry,regs[0].regmap,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 FCONV:
- fconv_assemble(0,®s[0]);break;
- case FLOAT:
- float_assemble(0,®s[0]);break;
- case FCOMP:
- fcomp_assemble(0,®s[0]);break;
- case MULTDIV:
- multdiv_assemble(0,®s[0]);break;
- case MOV:
- mov_assemble(0,®s[0]);break;
+ switch (dops[0].itype) {
case SYSCALL:
case HLECALL:
case INTCALL:
case RJUMP:
case CJUMP:
case SJUMP:
- case FJUMP:
- printf("Jump in the delay slot. This is probably a bug.\n");
+ 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) {
- btaddr=get_reg(regs[0].regmap,-1);
- emit_readword((int)&branch_target,btaddr);
+ btaddr=get_reg_temp(regs[0].regmap);
+ emit_readword(&branch_target,btaddr);
}
assert(btaddr!=HOST_CCREG);
if(regs[0].regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
#ifdef HOST_IMM8
+ host_tempreg_acquire();
emit_movimm(start+4,HOST_TEMPREG);
emit_cmp(btaddr,HOST_TEMPREG);
+ host_tempreg_release();
#else
emit_cmpimm(btaddr,start+4);
#endif
- int branch=(int)out;
+ void *branch = out;
emit_jeq(0);
- store_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,-1);
- emit_jmp(jump_vaddr_reg[btaddr]);
- set_jump_target(branch,(int)out);
- store_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,start+4);
- load_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,start+4);
+ store_regs_bt(regs[0].regmap,regs[0].dirty,-1);
+ do_jump_vaddr(btaddr);
+ set_jump_target(branch, out);
+ store_regs_bt(regs[0].regmap,regs[0].dirty,start+4);
+ load_regs_bt(regs[0].regmap,regs[0].dirty,start+4);
}
-// Basic liveness analysis for MIPS registers
-void unneeded_registers(int istart,int iend,int r)
+static void check_regmap(signed char *regmap)
{
- int i;
- uint64_t u,uu,b,bu;
- uint64_t temp_u,temp_uu;
- uint64_t tdep;
- if(iend==slen-1) {
- u=1;uu=1;
- }else{
- u=unneeded_reg[iend+1];
- uu=unneeded_reg_upper[iend+1];
- u=1;uu=1;
- }
- 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||itype[i]==FJUMP)
- {
- // If subroutine call, flag return address as a possible branch target
- if(rt1[i]==31 && i<slen-2) bt[i+2]=1;
-
- if(ba[i]<start || ba[i]>=(start+slen*4))
- {
- // Branch out of this block, flush all regs
- u=1;
- uu=1;
- /* Hexagon hack
- if(itype[i]==UJUMP&&rt1[i]==31)
- {
- uu=u=0x300C00F; // Discard at, v0-v1, t6-t9
- }
- if(itype[i]==RJUMP&&rs1[i]==31)
- {
- uu=u=0x300C0F3; // Discard at, a0-a3, t6-t9
- }
- if(start>0x80000400&&start<0x80000000+RAM_SIZE) {
- if(itype[i]==UJUMP&&rt1[i]==31)
- {
- //uu=u=0x30300FF0FLL; // Discard at, v0-v1, t0-t9, lo, hi
- uu=u=0x300FF0F; // Discard at, v0-v1, t0-t9
- }
- if(itype[i]==RJUMP&&rs1[i]==31)
- {
- //uu=u=0x30300FFF3LL; // Discard at, a0-a3, t0-t9, lo, hi
- uu=u=0x300FFF3; // Discard at, a0-a3, t0-t9
- }
- }*/
- branch_unneeded_reg[i]=u;
- branch_unneeded_reg_upper[i]=uu;
- // Merge in delay slot
- tdep=(~uu>>rt1[i+1])&1;
- u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
- uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
- u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
- uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
- u|=1;uu|=1;
- // If branch is "likely" (and conditional)
- // then we skip the delay slot on the fall-thru path
- if(likely[i]) {
- if(i<slen-1) {
- u&=unneeded_reg[i+2];
- uu&=unneeded_reg_upper[i+2];
- }
- else
- {
- u=1;
- uu=1;
- }
- }
- }
- else
- {
- // Internal branch, flag target
- bt[(ba[i]-start)>>2]=1;
- if(ba[i]<=start+i*4) {
- // Backward branch
- if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
- {
- // Unconditional branch
- temp_u=1;temp_uu=1;
- } else {
- // Conditional branch (not taken case)
- temp_u=unneeded_reg[i+2];
- temp_uu=unneeded_reg_upper[i+2];
- }
- // Merge in delay slot
- tdep=(~temp_uu>>rt1[i+1])&1;
- temp_u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
- temp_uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
- temp_u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
- temp_uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- temp_uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
- temp_u|=1;temp_uu|=1;
- // If branch is "likely" (and conditional)
- // then we skip the delay slot on the fall-thru path
- if(likely[i]) {
- if(i<slen-1) {
- temp_u&=unneeded_reg[i+2];
- temp_uu&=unneeded_reg_upper[i+2];
- }
- else
- {
- temp_u=1;
- temp_uu=1;
- }
- }
- tdep=(~temp_uu>>rt1[i])&1;
- temp_u|=(1LL<<rt1[i])|(1LL<<rt2[i]);
- temp_uu|=(1LL<<rt1[i])|(1LL<<rt2[i]);
- temp_u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
- temp_uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
- temp_uu&=~((tdep<<dep1[i])|(tdep<<dep2[i]));
- temp_u|=1;temp_uu|=1;
- unneeded_reg[i]=temp_u;
- unneeded_reg_upper[i]=temp_uu;
- // Only go three levels deep. This recursion can take an
- // excessive amount of time if there are a lot of nested loops.
- if(r<2) {
- unneeded_registers((ba[i]-start)>>2,i-1,r+1);
- }else{
- unneeded_reg[(ba[i]-start)>>2]=1;
- unneeded_reg_upper[(ba[i]-start)>>2]=1;
- }
- } /*else*/ if(1) {
- if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
- {
- // Unconditional branch
- u=unneeded_reg[(ba[i]-start)>>2];
- uu=unneeded_reg_upper[(ba[i]-start)>>2];
- branch_unneeded_reg[i]=u;
- branch_unneeded_reg_upper[i]=uu;
- //u=1;
- //uu=1;
- //branch_unneeded_reg[i]=u;
- //branch_unneeded_reg_upper[i]=uu;
- // Merge in delay slot
- tdep=(~uu>>rt1[i+1])&1;
- u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
- uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
- u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
- uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
- u|=1;uu|=1;
- } else {
- // Conditional branch
- b=unneeded_reg[(ba[i]-start)>>2];
- bu=unneeded_reg_upper[(ba[i]-start)>>2];
- branch_unneeded_reg[i]=b;
- branch_unneeded_reg_upper[i]=bu;
- //b=1;
- //bu=1;
- //branch_unneeded_reg[i]=b;
- //branch_unneeded_reg_upper[i]=bu;
- // Branch delay slot
- tdep=(~uu>>rt1[i+1])&1;
- b|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
- bu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
- b&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
- bu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- bu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
- b|=1;bu|=1;
- // If branch is "likely" then we skip the
- // delay slot on the fall-thru path
- if(likely[i]) {
- u=b;
- uu=bu;
- if(i<slen-1) {
- u&=unneeded_reg[i+2];
- uu&=unneeded_reg_upper[i+2];
- //u=1;
- //uu=1;
- }
- } else {
- u&=b;
- uu&=bu;
- //u=1;
- //uu=1;
- }
- if(i<slen-1) {
- branch_unneeded_reg[i]&=unneeded_reg[i+2];
- branch_unneeded_reg_upper[i]&=unneeded_reg_upper[i+2];
- //branch_unneeded_reg[i]=1;
- //branch_unneeded_reg_upper[i]=1;
- } else {
- branch_unneeded_reg[i]=1;
- branch_unneeded_reg_upper[i]=1;
- }
- }
- }
- }
- }
- else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
- {
- // SYSCALL instruction (software interrupt)
- u=1;
- uu=1;
- }
- else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
- {
- // ERET instruction (return from interrupt)
- u=1;
- uu=1;
- }
- //u=uu=1; // DEBUG
- tdep=(~uu>>rt1[i])&1;
- // Written registers are unneeded
- u|=1LL<<rt1[i];
- u|=1LL<<rt2[i];
- uu|=1LL<<rt1[i];
- uu|=1LL<<rt2[i];
- // Accessed registers are needed
- u&=~(1LL<<rs1[i]);
- u&=~(1LL<<rs2[i]);
- uu&=~(1LL<<us1[i]);
- uu&=~(1LL<<us2[i]);
- // Source-target dependencies
- uu&=~(tdep<<dep1[i]);
- uu&=~(tdep<<dep2[i]);
- // R0 is always unneeded
- u|=1;uu|=1;
- // Save it
- unneeded_reg[i]=u;
- unneeded_reg_upper[i]=uu;
- /*
- printf("ur (%d,%d) %x: ",istart,iend,start+i*4);
- printf("U:");
- int r;
- for(r=1;r<=CCREG;r++) {
- if((unneeded_reg[i]>>r)&1) {
- if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
- printf(" UU:");
- for(r=1;r<=CCREG;r++) {
- if(((unneeded_reg_upper[i]&~unneeded_reg[i])>>r)&1) {
- if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
- printf("\n");*/
- }
-#ifdef FORCE32
- for (i=iend;i>=istart;i--)
- {
- unneeded_reg_upper[i]=branch_unneeded_reg_upper[i]=-1LL;
+#ifndef NDEBUG
+ int i,j;
+ for (i = 0; i < HOST_REGS; i++) {
+ if (regmap[i] < 0)
+ continue;
+ for (j = i + 1; j < HOST_REGS; j++)
+ assert(regmap[i] != regmap[j]);
}
#endif
}
-// Identify registers which are likely to contain 32-bit values
-// This is used to predict whether any branches will jump to a
-// location with 64-bit values in registers.
-static void provisional_32bit()
+#ifdef DISASM
+#include <inttypes.h>
+static char insn[MAXBLOCK][10];
+
+#define set_mnemonic(i_, n_) \
+ strcpy(insn[i_], n_)
+
+void print_regmap(const char *name, const signed char *regmap)
{
- int i,j;
- uint64_t is32=1;
- uint64_t lastbranch=1;
-
- for(i=0;i<slen;i++)
- {
- if(i>0) {
- if(itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP) {
- if(i>1) is32=lastbranch;
- else is32=1;
- }
- }
- if(i>1)
- {
- if(itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP) {
- if(likely[i-2]) {
- if(i>2) is32=lastbranch;
- else is32=1;
- }
- }
- if((opcode[i-2]&0x2f)==0x05) // BNE/BNEL
- {
- if(rs1[i-2]==0||rs2[i-2]==0)
- {
- if(rs1[i-2]) {
- is32|=1LL<<rs1[i-2];
- }
- if(rs2[i-2]) {
- is32|=1LL<<rs2[i-2];
- }
- }
- }
- }
- // If something jumps here with 64-bit values
- // then promote those registers to 64 bits
- if(bt[i])
- {
- uint64_t temp_is32=is32;
- for(j=i-1;j>=0;j--)
- {
- if(ba[j]==start+i*4)
- //temp_is32&=branch_regs[j].is32;
- temp_is32&=p32[j];
- }
- for(j=i;j<slen;j++)
- {
- if(ba[j]==start+i*4)
- temp_is32=1;
- }
- is32=temp_is32;
- }
- int type=itype[i];
- int op=opcode[i];
- int op2=opcode2[i];
- int rt=rt1[i];
- int s1=rs1[i];
- int s2=rs2[i];
- if(type==UJUMP||type==RJUMP||type==CJUMP||type==SJUMP||type==FJUMP) {
- // Branches don't write registers, consider the delay slot instead.
- type=itype[i+1];
- op=opcode[i+1];
- op2=opcode2[i+1];
- rt=rt1[i+1];
- s1=rs1[i+1];
- s2=rs2[i+1];
- lastbranch=is32;
- }
- switch(type) {
- case LOAD:
- if(opcode[i]==0x27||opcode[i]==0x37|| // LWU/LD
- opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR
- is32&=~(1LL<<rt);
- else
- is32|=1LL<<rt;
- break;
- case STORE:
- case STORELR:
- break;
- case LOADLR:
- if(op==0x1a||op==0x1b) is32&=~(1LL<<rt); // LDR/LDL
- if(op==0x22) is32|=1LL<<rt; // LWL
- break;
- case IMM16:
- if (op==0x08||op==0x09|| // ADDI/ADDIU
- op==0x0a||op==0x0b|| // SLTI/SLTIU
- op==0x0c|| // ANDI
- op==0x0f) // LUI
- {
- is32|=1LL<<rt;
- }
- if(op==0x18||op==0x19) { // DADDI/DADDIU
- is32&=~(1LL<<rt);
- //if(imm[i]==0)
- // is32|=((is32>>s1)&1LL)<<rt;
- }
- if(op==0x0d||op==0x0e) { // ORI/XORI
- uint64_t sr=((is32>>s1)&1LL);
- is32&=~(1LL<<rt);
- is32|=sr<<rt;
- }
- break;
- case UJUMP:
- break;
- case RJUMP:
- break;
- case CJUMP:
- break;
- case SJUMP:
- break;
- case FJUMP:
- break;
- case ALU:
- if(op2>=0x20&&op2<=0x23) { // ADD/ADDU/SUB/SUBU
- is32|=1LL<<rt;
- }
- if(op2==0x2a||op2==0x2b) { // SLT/SLTU
- is32|=1LL<<rt;
- }
- else if(op2>=0x24&&op2<=0x27) { // AND/OR/XOR/NOR
- uint64_t sr=((is32>>s1)&(is32>>s2)&1LL);
- is32&=~(1LL<<rt);
- is32|=sr<<rt;
- }
- else if(op2>=0x2c&&op2<=0x2d) { // DADD/DADDU
- if(s1==0&&s2==0) {
- is32|=1LL<<rt;
- }
- else if(s2==0) {
- uint64_t sr=((is32>>s1)&1LL);
- is32&=~(1LL<<rt);
- is32|=sr<<rt;
- }
- else if(s1==0) {
- uint64_t sr=((is32>>s2)&1LL);
- is32&=~(1LL<<rt);
- is32|=sr<<rt;
- }
- else {
- is32&=~(1LL<<rt);
- }
- }
- else if(op2>=0x2e&&op2<=0x2f) { // DSUB/DSUBU
- if(s1==0&&s2==0) {
- is32|=1LL<<rt;
- }
- else if(s2==0) {
- uint64_t sr=((is32>>s1)&1LL);
- is32&=~(1LL<<rt);
- is32|=sr<<rt;
- }
- else {
- is32&=~(1LL<<rt);
- }
- }
- break;
- case MULTDIV:
- if (op2>=0x1c&&op2<=0x1f) { // DMULT/DMULTU/DDIV/DDIVU
- is32&=~((1LL<<HIREG)|(1LL<<LOREG));
- }
- else {
- is32|=(1LL<<HIREG)|(1LL<<LOREG);
- }
- break;
- case MOV:
- {
- uint64_t sr=((is32>>s1)&1LL);
- is32&=~(1LL<<rt);
- is32|=sr<<rt;
- }
- break;
- case SHIFT:
- if(op2>=0x14&&op2<=0x17) is32&=~(1LL<<rt); // DSLLV/DSRLV/DSRAV
- else is32|=1LL<<rt; // SLLV/SRLV/SRAV
- break;
- case SHIFTIMM:
- is32|=1LL<<rt;
- // DSLL/DSRL/DSRA/DSLL32/DSRL32 but not DSRA32 have 64-bit result
- if(op2>=0x38&&op2<0x3f) is32&=~(1LL<<rt);
- break;
- case COP0:
- if(op2==0) is32|=1LL<<rt; // MFC0
- break;
- case COP1:
- case COP2:
- if(op2==0) is32|=1LL<<rt; // MFC1
- if(op2==1) is32&=~(1LL<<rt); // DMFC1
- if(op2==2) is32|=1LL<<rt; // CFC1
- break;
- case C1LS:
- case C2LS:
- break;
- case FLOAT:
- case FCONV:
- break;
- case FCOMP:
- break;
- case C2OP:
- case SYSCALL:
- case HLECALL:
- break;
- default:
- break;
- }
- is32|=1;
- p32[i]=is32;
-
- if(i>0)
- {
- if(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)
- {
- if(rt1[i-1]==31) // JAL/JALR
- {
- // Subroutine call will return here, don't alloc any registers
- is32=1;
- }
- else if(i+1<slen)
- {
- // Internal branch will jump here, match registers to caller
- is32=0x3FFFFFFFFLL;
- }
- }
- }
- }
-}
-
-// Identify registers which may be assumed to contain 32-bit values
-// and where optimizations will rely on this.
-// This is used to determine whether backward branches can safely
-// jump to a location with 64-bit values in registers.
-static void provisional_r32()
-{
- u_int r32=0;
- int i;
-
- for (i=slen-1;i>=0;i--)
- {
- int hr;
- if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
- {
- if(ba[i]<start || ba[i]>=(start+slen*4))
- {
- // Branch out of this block, don't need anything
- r32=0;
- }
- else
- {
- // Internal branch
- // Need whatever matches the target
- // (and doesn't get overwritten by the delay slot instruction)
- r32=0;
- int t=(ba[i]-start)>>2;
- if(ba[i]>start+i*4) {
- // Forward branch
- //if(!(requires_32bit[t]&~regs[i].was32))
- // r32|=requires_32bit[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
- if(!(pr32[t]&~regs[i].was32))
- r32|=pr32[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
- }else{
- // Backward branch
- if(!(regs[t].was32&~unneeded_reg_upper[t]&~regs[i].was32))
- r32|=regs[t].was32&~unneeded_reg_upper[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
- }
- }
- // Conditional branch may need registers for following instructions
- if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
- {
- if(i<slen-2) {
- //r32|=requires_32bit[i+2];
- r32|=pr32[i+2];
- r32&=regs[i].was32;
- // Mark this address as a branch target since it may be called
- // upon return from interrupt
- //bt[i+2]=1;
- }
- }
- // Merge in delay slot
- if(!likely[i]) {
- // These are overwritten unless the branch is "likely"
- // and the delay slot is nullified if not taken
- r32&=~(1LL<<rt1[i+1]);
- r32&=~(1LL<<rt2[i+1]);
- }
- // Assume these are needed (delay slot)
- if(us1[i+1]>0)
- {
- if((regs[i].was32>>us1[i+1])&1) r32|=1LL<<us1[i+1];
- }
- if(us2[i+1]>0)
- {
- if((regs[i].was32>>us2[i+1])&1) r32|=1LL<<us2[i+1];
- }
- if(dep1[i+1]&&!((unneeded_reg_upper[i]>>dep1[i+1])&1))
- {
- if((regs[i].was32>>dep1[i+1])&1) r32|=1LL<<dep1[i+1];
- }
- if(dep2[i+1]&&!((unneeded_reg_upper[i]>>dep2[i+1])&1))
- {
- if((regs[i].was32>>dep2[i+1])&1) r32|=1LL<<dep2[i+1];
- }
- }
- else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
- {
- // SYSCALL instruction (software interrupt)
- r32=0;
- }
- else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
- {
- // ERET instruction (return from interrupt)
- r32=0;
- }
- // Check 32 bits
- r32&=~(1LL<<rt1[i]);
- r32&=~(1LL<<rt2[i]);
- if(us1[i]>0)
- {
- if((regs[i].was32>>us1[i])&1) r32|=1LL<<us1[i];
- }
- if(us2[i]>0)
- {
- if((regs[i].was32>>us2[i])&1) r32|=1LL<<us2[i];
- }
- if(dep1[i]&&!((unneeded_reg_upper[i]>>dep1[i])&1))
- {
- if((regs[i].was32>>dep1[i])&1) r32|=1LL<<dep1[i];
- }
- if(dep2[i]&&!((unneeded_reg_upper[i]>>dep2[i])&1))
- {
- if((regs[i].was32>>dep2[i])&1) r32|=1LL<<dep2[i];
- }
- //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;hr<HOST_REGS;hr++)
- {
- if(regs[i].regmap_entry[hr]>0&®s[i].regmap_entry[hr]<64) {
- if((regs[i].was32>>regs[i].regmap_entry[hr])&(regs[i].wasdirty>>hr)&1) {
- if(!((unneeded_reg_upper[i]>>regs[i].regmap_entry[hr])&1))
- pr32[i]|=1LL<<regs[i].regmap_entry[hr];
- //requires_32bit[i]|=1LL<<regs[i].regmap_entry[hr];
- }
- }
- }
- }
-}
-
-// Write back dirty registers as soon as we will no longer modify them,
-// so that we don't end up with lots of writes at the branches.
-void clean_registers(int istart,int iend,int wr)
-{
- int i;
- int r;
- u_int will_dirty_i,will_dirty_next,temp_will_dirty;
- u_int wont_dirty_i,wont_dirty_next,temp_wont_dirty;
- if(iend==slen-1) {
- will_dirty_i=will_dirty_next=0;
- wont_dirty_i=wont_dirty_next=0;
- }else{
- will_dirty_i=will_dirty_next=will_dirty[iend+1];
- wont_dirty_i=wont_dirty_next=wont_dirty[iend+1];
- }
- for (i=iend;i>=istart;i--)
- {
- if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
- {
- if(ba[i]<start || ba[i]>=(start+slen*4))
- {
- // Branch out of this block, flush all regs
- if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
- {
- // Unconditional branch
- will_dirty_i=0;
- wont_dirty_i=0;
- // Merge in delay slot (will dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((branch_regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- }
- }
- }
- else
- {
- // Conditional branch
- will_dirty_i=0;
- wont_dirty_i=wont_dirty_next;
- // Merge in delay slot (will dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if(!likely[i]) {
- // Might not dirty if likely branch is not taken
- if((branch_regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==0) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- //if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- //if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- }
- }
- }
- }
- // Merge in delay slot (wont dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) wont_dirty_i|=1<<r;
- if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) wont_dirty_i|=1<<r;
- if(branch_regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- }
- }
- if(wr) {
- #ifndef DESTRUCTIVE_WRITEBACK
- branch_regs[i].dirty&=wont_dirty_i;
- #endif
- branch_regs[i].dirty|=will_dirty_i;
- }
- }
- else
- {
- // Internal branch
- if(ba[i]<=start+i*4) {
- // Backward branch
- if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
- {
- // Unconditional branch
- temp_will_dirty=0;
- temp_wont_dirty=0;
- // Merge in delay slot (will dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((branch_regs[i].regmap[r]&63)==rt1[i]) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
- if(branch_regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i]) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i]) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
- if(regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
- }
- }
- } else {
- // Conditional branch (not taken case)
- temp_will_dirty=will_dirty_next;
- temp_wont_dirty=wont_dirty_next;
- // Merge in delay slot (will dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if(!likely[i]) {
- // Will not dirty if likely branch is not taken
- if((branch_regs[i].regmap[r]&63)==rt1[i]) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) temp_will_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
- if(branch_regs[i].regmap[r]==0) temp_will_dirty&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
- //if((regs[i].regmap[r]&63)==rt1[i]) temp_will_dirty|=1<<r;
- //if((regs[i].regmap[r]&63)==rt2[i]) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) temp_will_dirty|=1<<r;
- if((regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
- if(regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
- }
- }
- }
- }
- // Merge in delay slot (wont dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((regs[i].regmap[r]&63)==rt1[i]) temp_wont_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i]) temp_wont_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) temp_wont_dirty|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) temp_wont_dirty|=1<<r;
- if(regs[i].regmap[r]==CCREG) temp_wont_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i]) temp_wont_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) temp_wont_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) temp_wont_dirty|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) temp_wont_dirty|=1<<r;
- if(branch_regs[i].regmap[r]==CCREG) temp_wont_dirty|=1<<r;
- }
- }
- // Deal with changed mappings
- if(i<iend) {
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if(regs[i].regmap[r]!=regmap_pre[i][r]) {
- temp_will_dirty&=~(1<<r);
- temp_wont_dirty&=~(1<<r);
- if((regmap_pre[i][r]&63)>0 && (regmap_pre[i][r]&63)<34) {
- temp_will_dirty|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
- temp_wont_dirty|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
- } else {
- temp_will_dirty|=1<<r;
- temp_wont_dirty|=1<<r;
- }
- }
- }
- }
- }
- if(wr) {
- will_dirty[i]=temp_will_dirty;
- wont_dirty[i]=temp_wont_dirty;
- clean_registers((ba[i]-start)>>2,i-1,0);
- }else{
- // Limit recursion. It can take an excessive amount
- // of time if there are a lot of nested loops.
- will_dirty[(ba[i]-start)>>2]=0;
- wont_dirty[(ba[i]-start)>>2]=-1;
- }
- }
- /*else*/ if(1)
- {
- if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
- {
- // Unconditional branch
- will_dirty_i=0;
- wont_dirty_i=0;
- //if(ba[i]>start+i*4) { // Disable recursion (for debugging)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if(branch_regs[i].regmap[r]==regs[(ba[i]-start)>>2].regmap_entry[r]) {
- will_dirty_i|=will_dirty[(ba[i]-start)>>2]&(1<<r);
- wont_dirty_i|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
- }
- if(branch_regs[i].regmap[r]>=0) {
- will_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>(branch_regs[i].regmap[r]&63))&1)<<r;
- wont_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>(branch_regs[i].regmap[r]&63))&1)<<r;
- }
- }
- }
- //}
- // Merge in delay slot
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((branch_regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- }
- }
- } else {
- // Conditional branch
- will_dirty_i=will_dirty_next;
- wont_dirty_i=wont_dirty_next;
- //if(ba[i]>start+i*4) { // Disable recursion (for debugging)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- signed char target_reg=branch_regs[i].regmap[r];
- if(target_reg==regs[(ba[i]-start)>>2].regmap_entry[r]) {
- will_dirty_i&=will_dirty[(ba[i]-start)>>2]&(1<<r);
- wont_dirty_i|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
- }
- else if(target_reg>=0) {
- will_dirty_i&=((unneeded_reg[(ba[i]-start)>>2]>>(target_reg&63))&1)<<r;
- wont_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>(target_reg&63))&1)<<r;
- }
- // Treat delay slot as part of branch too
- /*if(regs[i+1].regmap[r]==regs[(ba[i]-start)>>2].regmap_entry[r]) {
- will_dirty[i+1]&=will_dirty[(ba[i]-start)>>2]&(1<<r);
- wont_dirty[i+1]|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
- }
- else
- {
- will_dirty[i+1]&=~(1<<r);
- }*/
- }
- }
- //}
- // Merge in delay slot
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if(!likely[i]) {
- // Might not dirty if likely branch is not taken
- if((branch_regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- //if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- //if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- }
- }
- }
- }
- // Merge in delay slot (won't dirty)
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i+1]) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i+1]) wont_dirty_i|=1<<r;
- if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt1[i+1]) wont_dirty_i|=1<<r;
- if((branch_regs[i].regmap[r]&63)==rt2[i+1]) wont_dirty_i|=1<<r;
- if(branch_regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- }
- }
- if(wr) {
- #ifndef DESTRUCTIVE_WRITEBACK
- branch_regs[i].dirty&=wont_dirty_i;
- #endif
- branch_regs[i].dirty|=will_dirty_i;
- }
- }
- }
- }
- else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
- {
- // SYSCALL instruction (software interrupt)
- will_dirty_i=0;
- wont_dirty_i=0;
- }
- else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
- {
- // ERET instruction (return from interrupt)
- will_dirty_i=0;
- wont_dirty_i=0;
- }
- will_dirty_next=will_dirty_i;
- wont_dirty_next=wont_dirty_i;
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
- if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
- if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
- if(i>istart) {
- 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.
- if((regs[i].regmap[r]&63)==rt1[i-1]) wont_dirty_i|=1<<r;
- if((regs[i].regmap[r]&63)==rt2[i-1]) wont_dirty_i|=1<<r;
- }
- }
- }
- }
- // Save it
- will_dirty[i]=will_dirty_i;
- wont_dirty[i]=wont_dirty_i;
- // Mark registers that won't be dirtied as not dirty
- if(wr) {
- /*printf("wr (%d,%d) %x will:",istart,iend,start+i*4);
- for(r=0;r<HOST_REGS;r++) {
- if((will_dirty_i>>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&&itype[i-1]!=FJUMP)) {
- 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||itype[i]==FJUMP)
- {
- if(i<iend-1&&itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000) {
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if(regs[i].regmap[r]==regmap_pre[i+2][r]) {
- regs[i+2].wasdirty&=wont_dirty_i|~(1<<r);
- }else {/*printf("i: %x (%d) mismatch(+2): %d\n",start+i*4,i,r);/*assert(!((wont_dirty_i>>r)&1));*/}
- }
- }
- }
- }
- else
- {
- if(i<iend) {
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- if(regs[i].regmap[r]==regmap_pre[i+1][r]) {
- regs[i+1].wasdirty&=wont_dirty_i|~(1<<r);
- }else {/*printf("i: %x (%d) mismatch(+1): %d\n",start+i*4,i,r);/*assert(!((wont_dirty_i>>r)&1));*/}
- }
- }
- }
- }
- #endif
- //}
- }
- // Deal with changed mappings
- temp_will_dirty=will_dirty_i;
- temp_wont_dirty=wont_dirty_i;
- for(r=0;r<HOST_REGS;r++) {
- if(r!=EXCLUDE_REG) {
- int nr;
- if(regs[i].regmap[r]==regmap_pre[i][r]) {
- if(wr) {
- #ifndef DESTRUCTIVE_WRITEBACK
- regs[i].wasdirty&=wont_dirty_i|~(1<<r);
- #endif
- regs[i].wasdirty|=will_dirty_i&(1<<r);
- }
- }
- else if(regmap_pre[i][r]>=0&&(nr=get_reg(regs[i].regmap,regmap_pre[i][r]))>=0) {
- // Register moved to a different register
- will_dirty_i&=~(1<<r);
- wont_dirty_i&=~(1<<r);
- will_dirty_i|=((temp_will_dirty>>nr)&1)<<r;
- wont_dirty_i|=((temp_wont_dirty>>nr)&1)<<r;
- if(wr) {
- #ifndef DESTRUCTIVE_WRITEBACK
- regs[i].wasdirty&=wont_dirty_i|~(1<<r);
- #endif
- regs[i].wasdirty|=will_dirty_i&(1<<r);
- }
- }
- else {
- will_dirty_i&=~(1<<r);
- wont_dirty_i&=~(1<<r);
- if((regmap_pre[i][r]&63)>0 && (regmap_pre[i][r]&63)<34) {
- will_dirty_i|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
- wont_dirty_i|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
- } else {
- wont_dirty_i|=1<<r;
- /*printf("i: %x (%d) mismatch: %d\n",start+i*4,i,r);/*assert(!((will_dirty>>r)&1));*/
- }
- }
- }
- }
- }
+ char buf[5];
+ int i, l;
+ fputs(name, stdout);
+ for (i = 0; i < HOST_REGS; i++) {
+ l = 0;
+ if (regmap[i] >= 0)
+ l = snprintf(buf, sizeof(buf), "$%d", regmap[i]);
+ for (; l < 3; l++)
+ buf[l] = ' ';
+ buf[l] = 0;
+ printf(" r%d=%s", i, buf);
+ }
+ fputs("\n", stdout);
}
/* disassembly */
void disassemble_inst(int i)
{
- if (bt[i]) printf("*"); else printf(" ");
- switch(itype[i]) {
+ if (dops[i].bt) printf("*"); else printf(" ");
+ switch(dops[i].itype) {
case UJUMP:
printf (" %x: %s %8x\n",start+i*4,insn[i],ba[i]);break;
case CJUMP:
- printf (" %x: %s r%d,r%d,%8x\n",start+i*4,insn[i],rs1[i],rs2[i],i?start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>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;
- case FJUMP:
- printf (" %x: %s %8x\n",start+i*4,insn[i],ba[i]);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]);
//printf (" %s %8x\n",insn[i],source[i]);
printf (" %x: %s\n",start+i*4,insn[i]);
}
+ return;
+ printf("D: %"PRIu64" WD: %"PRIu64" U: %"PRIu64"\n",
+ regs[i].dirty, regs[i].wasdirty, unneeded_reg[i]);
+ print_regmap("pre: ", regmap_pre[i]);
+ print_regmap("entry: ", regs[i].regmap_entry);
+ print_regmap("map: ", regs[i].regmap);
+ if (dops[i].is_jump) {
+ print_regmap("bentry:", branch_regs[i].regmap_entry);
+ print_regmap("bmap: ", branch_regs[i].regmap);
+ }
+}
+#else
+#define set_mnemonic(i_, n_)
+static void disassemble_inst(int i) {}
+#endif // DISASM
+
+#define DRC_TEST_VAL 0x74657374
+
+static void new_dynarec_test(void)
+{
+ int (*testfunc)(void);
+ void *beginning;
+ int ret[2];
+ size_t i;
+
+ // check structure linkage
+ if ((u_char *)rcnts - (u_char *)&psxRegs != sizeof(psxRegs))
+ {
+ SysPrintf("linkage_arm* miscompilation/breakage detected.\n");
+ }
+
+ SysPrintf("testing if we can run recompiled code @%p...\n", out);
+ ((volatile u_int *)out)[0]++; // make cache dirty
+
+ for (i = 0; i < ARRAY_SIZE(ret); i++) {
+ out = ndrc->translation_cache;
+ beginning = start_block();
+ emit_movimm(DRC_TEST_VAL + i, 0); // test
+ emit_ret();
+ literal_pool(0);
+ end_block(beginning);
+ testfunc = beginning;
+ ret[i] = testfunc();
+ }
+
+ if (ret[0] == DRC_TEST_VAL && ret[1] == DRC_TEST_VAL + 1)
+ SysPrintf("test passed.\n");
+ else
+ SysPrintf("test failed, will likely crash soon (r=%08x %08x)\n", ret[0], ret[1]);
+ out = ndrc->translation_cache;
}
// clear the state completely, instead of just marking
// things invalid like invalidate_all_pages() does
-void new_dynarec_clear_full()
+void new_dynarec_clear_full(void)
{
int n;
- out=(u_char *)BASE_ADDR;
+ out = ndrc->translation_cache;
memset(invalid_code,1,sizeof(invalid_code));
memset(hash_table,0xff,sizeof(hash_table));
memset(mini_ht,-1,sizeof(mini_ht));
literalcount=0;
stop_after_jal=0;
inv_code_start=inv_code_end=~0;
+ hack_addr=0;
+ f1_hack=0;
// TLB
-#ifndef DISABLE_TLB
- using_tlb=0;
-#endif
- sp_in_mirror=0;
- for(n=0;n<524288;n++) // 0 .. 0x7FFFFFFF
- memory_map[n]=-1;
- for(n=524288;n<526336;n++) // 0x80000000 .. 0x807FFFFF
- memory_map[n]=((u_int)rdram-0x80000000)>>2;
- for(n=526336;n<1048576;n++) // 0x80800000 .. 0xFFFFFFFF
- memory_map[n]=-1;
for(n=0;n<4096;n++) ll_clear(jump_in+n);
for(n=0;n<4096;n++) ll_clear(jump_out+n);
for(n=0;n<4096;n++) ll_clear(jump_dirty+n);
+
+ cycle_multiplier_old = cycle_multiplier;
+ new_dynarec_hacks_old = new_dynarec_hacks;
}
-void new_dynarec_init()
+void new_dynarec_init(void)
{
- printf("Init new dynarec\n");
- out=(u_char *)BASE_ADDR;
- if (mmap (out, 1<<TARGET_SIZE_2,
+ SysPrintf("Init new dynarec, ndrc size %x\n", (int)sizeof(*ndrc));
+
+#ifdef _3DS
+ check_rosalina();
+#endif
+#ifdef BASE_ADDR_DYNAMIC
+ #ifdef VITA
+ sceBlock = getVMBlock(); //sceKernelAllocMemBlockForVM("code", sizeof(*ndrc));
+ if (sceBlock <= 0)
+ SysPrintf("sceKernelAllocMemBlockForVM failed: %x\n", sceBlock);
+ int ret = sceKernelGetMemBlockBase(sceBlock, (void **)&ndrc);
+ if (ret < 0)
+ SysPrintf("sceKernelGetMemBlockBase failed: %x\n", ret);
+ sceKernelOpenVMDomain();
+ sceClibPrintf("translation_cache = 0x%08lx\n ", (long)ndrc->translation_cache);
+ #elif defined(_MSC_VER)
+ ndrc = VirtualAlloc(NULL, sizeof(*ndrc), MEM_COMMIT | MEM_RESERVE,
+ PAGE_EXECUTE_READWRITE);
+ #else
+ uintptr_t desired_addr = 0;
+ #ifdef __ELF__
+ extern char _end;
+ desired_addr = ((uintptr_t)&_end + 0xffffff) & ~0xffffffl;
+ #endif
+ ndrc = mmap((void *)desired_addr, sizeof(*ndrc),
PROT_READ | PROT_WRITE | PROT_EXEC,
- MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
- -1, 0) <= 0) {printf("mmap() failed\n");}
-#ifdef MUPEN64
- rdword=&readmem_dword;
- fake_pc.f.r.rs=&readmem_dword;
- fake_pc.f.r.rt=&readmem_dword;
- fake_pc.f.r.rd=&readmem_dword;
+ MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+ if (ndrc == MAP_FAILED) {
+ SysPrintf("mmap() failed: %s\n", strerror(errno));
+ abort();
+ }
+ #endif
+#else
+ #ifndef NO_WRITE_EXEC
+ // not all systems allow execute in data segment by default
+ // size must be 4K aligned for 3DS?
+ if (mprotect(ndrc, sizeof(*ndrc),
+ PROT_READ | PROT_WRITE | PROT_EXEC) != 0)
+ SysPrintf("mprotect() failed: %s\n", strerror(errno));
+ #endif
#endif
- int n;
+ out = ndrc->translation_cache;
+ cycle_multiplier=200;
new_dynarec_clear_full();
#ifdef HOST_IMM8
// Copy this into local area so we don't have to put it in every literal pool
invc_ptr=invalid_code;
#endif
-#ifdef MUPEN64
- for(n=0;n<0x8000;n++) { // 0 .. 0x7FFFFFFF
- writemem[n] = write_nomem_new;
- writememb[n] = write_nomemb_new;
- writememh[n] = write_nomemh_new;
-#ifndef FORCE32
- writememd[n] = write_nomemd_new;
-#endif
- readmem[n] = read_nomem_new;
- readmemb[n] = read_nomemb_new;
- readmemh[n] = read_nomemh_new;
-#ifndef FORCE32
- readmemd[n] = read_nomemd_new;
-#endif
- }
- for(n=0x8000;n<0x8080;n++) { // 0x80000000 .. 0x807FFFFF
- writemem[n] = write_rdram_new;
- writememb[n] = write_rdramb_new;
- writememh[n] = write_rdramh_new;
-#ifndef FORCE32
- writememd[n] = write_rdramd_new;
-#endif
- }
- for(n=0xC000;n<0x10000;n++) { // 0xC0000000 .. 0xFFFFFFFF
- writemem[n] = write_nomem_new;
- writememb[n] = write_nomemb_new;
- writememh[n] = write_nomemh_new;
-#ifndef FORCE32
- writememd[n] = write_nomemd_new;
-#endif
- readmem[n] = read_nomem_new;
- readmemb[n] = read_nomemb_new;
- readmemh[n] = read_nomemh_new;
-#ifndef FORCE32
- readmemd[n] = read_nomemd_new;
-#endif
- }
-#endif
- tlb_hacks();
arch_init();
+ new_dynarec_test();
+ ram_offset=(uintptr_t)rdram-0x80000000;
+ if (ram_offset!=0)
+ SysPrintf("warning: RAM is not directly mapped, performance will suffer\n");
}
-void new_dynarec_cleanup()
+void new_dynarec_cleanup(void)
{
int n;
- if (munmap ((void *)BASE_ADDR, 1<<TARGET_SIZE_2) < 0) {printf("munmap() failed\n");}
+#ifdef BASE_ADDR_DYNAMIC
+ #ifdef VITA
+ // sceBlock is managed by retroarch's bootstrap code
+ //sceKernelFreeMemBlock(sceBlock);
+ //sceBlock = -1;
+ #else
+ if (munmap(ndrc, sizeof(*ndrc)) < 0)
+ SysPrintf("munmap() failed\n");
+ #endif
+#endif
for(n=0;n<4096;n++) ll_clear(jump_in+n);
for(n=0;n<4096;n++) ll_clear(jump_out+n);
for(n=0;n<4096;n++) ll_clear(jump_dirty+n);
#ifdef ROM_COPY
- if (munmap (ROM_COPY, 67108864) < 0) {printf("munmap() failed\n");}
+ if (munmap (ROM_COPY, 67108864) < 0) {SysPrintf("munmap() failed\n");}
#endif
}
-int new_recompile_block(int addr)
-{
-/*
- if(addr==0x800cd050) {
- int block;
- for(block=0x80000;block<0x80800;block++) invalidate_block(block);
- int n;
- for(n=0;n<=2048;n++) ll_clear(jump_dirty+n);
- }
-*/
- //if(Count==365117028) tracedebug=1;
- 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)
- //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();
- start = (u_int)addr&~3;
- //assert(((u_int)addr&1)==0);
-#ifdef PCSX
- if(!sp_in_mirror&&(signed int)(psxRegs.GPR.n.sp&0xffe00000)>0x80200000&&
- 0x10000<=psxRegs.GPR.n.sp&&(psxRegs.GPR.n.sp&~0xe0e00000)<RAM_SIZE) {
- printf("SP hack enabled (%08x), @%08x\n", psxRegs.GPR.n.sp, psxRegs.pc);
- sp_in_mirror=1;
- }
- if (Config.HLE && start == 0x80001000) // hlecall
+static u_int *get_source_start(u_int addr, u_int *limit)
+{
+ if (addr < 0x00200000 ||
+ (0xa0000000 <= addr && addr < 0xa0200000))
{
- // XXX: is this enough? Maybe check hleSoftCall?
- u_int beginning=(u_int)out;
- u_int page=get_page(start);
- invalid_code[start>>12]=0;
- emit_movimm(start,0);
- emit_writeword(0,(int)&pcaddr);
- emit_jmp((int)new_dyna_leave);
- literal_pool(0);
-#ifdef __arm__
- __clear_cache((void *)beginning,out);
-#endif
- ll_add(jump_in+page,start,(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;
+ *limit = (addr&0xa0000000)|0x00200000;
+ return (u_int *)(rdram + (addr&0x1fffff));
}
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;
+ /* (0x9fc00000 <= addr && addr < 0x9fc80000) ||*/
+ (0xbfc00000 <= addr && addr < 0xbfc80000)))
+ {
+ // BIOS. The multiplier should be much higher as it's uncached 8bit mem,
+ // but timings in PCSX are too tied to the interpreter's BIAS
+ if (!HACK_ENABLED(NDHACK_OVERRIDE_CYCLE_M))
+ cycle_multiplier_active = 200;
+
+ *limit = (addr & 0xfff00000) | 0x80000;
+ return (u_int *)((u_char *)psxR + (addr&0x7ffff));
}
- else
-#endif
-#ifdef MUPEN64
- if ((int)addr >= 0xa4000000 && (int)addr < 0xa4001000) {
- source = (u_int *)((u_int)SP_DMEM+start-0xa4000000);
- pagelimit = 0xa4001000;
+ else if (addr >= 0x80000000 && addr < 0x80000000+RAM_SIZE) {
+ *limit = (addr & 0x80600000) + 0x00200000;
+ return (u_int *)(rdram + (addr&0x1fffff));
+ }
+ return NULL;
+}
+
+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;
+ }
+ return addr;
+}
+
+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 < ARRAY_SIZE(jump_in); 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;
}
- 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);
+
+ 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;
}
- 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
+
+ get_addr(blocks[b].addr);
+
+ for (f = blocks[b].regflags, i = 0; f; f >>= 1, i++) {
+ if (f & 1)
+ psxRegs.GPR.r[i] = 0x80000000;
}
- //printf("source= %x\n",(int)source);
- }
-#endif
- else {
- printf("Compile at bogus memory address: %x \n", (int)addr);
- exit(1);
}
- /* Pass 1: disassemble */
- /* Pass 2: register dependencies, branch targets */
- /* Pass 3: register allocation */
- /* Pass 4: branch dependencies */
- /* Pass 5: pre-alloc */
- /* Pass 6: optimize clean/dirty state */
- /* Pass 7: flag 32-bit registers */
- /* Pass 8: assembly */
- /* Pass 9: linker */
- /* Pass 10: garbage collection / free memory */
+ memcpy(&psxRegs.GPR, regs_save, sizeof(regs_save));
+}
- int i,j;
- int done=0;
- unsigned int type,op,op2;
+static int apply_hacks(void)
+{
+ int i;
+ if (HACK_ENABLED(NDHACK_NO_COMPAT_HACKS))
+ return 0;
+ /* special hack(s) */
+ for (i = 0; i < slen - 4; i++)
+ {
+ // lui a4, 0xf200; jal <rcnt_read>; addu a0, 2; slti v0, 28224
+ if (source[i] == 0x3c04f200 && dops[i+1].itype == UJUMP
+ && source[i+2] == 0x34840002 && dops[i+3].opcode == 0x0a
+ && imm[i+3] == 0x6e40 && dops[i+3].rs1 == 2)
+ {
+ SysPrintf("PE2 hack @%08x\n", start + (i+3)*4);
+ dops[i + 3].itype = NOP;
+ }
+ }
+ i = slen;
+ if (i > 10 && source[i-1] == 0 && source[i-2] == 0x03e00008
+ && source[i-4] == 0x8fbf0018 && source[i-6] == 0x00c0f809
+ && dops[i-7].itype == STORE)
+ {
+ i = i-8;
+ if (dops[i].itype == IMM16)
+ i--;
+ // swl r2, 15(r6); swr r2, 12(r6); sw r6, *; jalr r6
+ if (dops[i].itype == STORELR && dops[i].rs1 == 6
+ && dops[i-1].itype == STORELR && dops[i-1].rs1 == 6)
+ {
+ SysPrintf("F1 hack from %08x, old dst %08x\n", start, hack_addr);
+ f1_hack = 1;
+ return 1;
+ }
+ }
+ return 0;
+}
- //printf("addr = %x source = %x %x\n", addr,source,source[0]);
-
- /* Pass 1 disassembly */
+static noinline void pass1_disassemble(u_int pagelimit)
+{
+ int i, j, done = 0, ni_count = 0;
+ unsigned int type,op,op2;
- for(i=0;!done;i++) {
- bt[i]=0;likely[i]=0;ooo[i]=0;op2=0;
+ for (i = 0; !done; i++)
+ {
+ memset(&dops[i], 0, sizeof(dops[i]));
+ 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;
+ case 0x00: set_mnemonic(i, "special"); type=NI;
op2=source[i]&0x3f;
switch(op2)
{
- case 0x00: strcpy(insn[i],"SLL"); type=SHIFTIMM; break;
- case 0x02: strcpy(insn[i],"SRL"); type=SHIFTIMM; break;
- case 0x03: strcpy(insn[i],"SRA"); type=SHIFTIMM; break;
- case 0x04: strcpy(insn[i],"SLLV"); type=SHIFT; break;
- case 0x06: strcpy(insn[i],"SRLV"); type=SHIFT; break;
- case 0x07: strcpy(insn[i],"SRAV"); type=SHIFT; break;
- case 0x08: strcpy(insn[i],"JR"); type=RJUMP; break;
- case 0x09: strcpy(insn[i],"JALR"); type=RJUMP; break;
- case 0x0C: strcpy(insn[i],"SYSCALL"); type=SYSCALL; break;
- case 0x0D: strcpy(insn[i],"BREAK"); type=OTHER; break;
- case 0x0F: strcpy(insn[i],"SYNC"); type=OTHER; break;
- case 0x10: strcpy(insn[i],"MFHI"); type=MOV; break;
- case 0x11: strcpy(insn[i],"MTHI"); type=MOV; break;
- case 0x12: strcpy(insn[i],"MFLO"); type=MOV; break;
- case 0x13: strcpy(insn[i],"MTLO"); type=MOV; break;
- case 0x18: strcpy(insn[i],"MULT"); type=MULTDIV; break;
- case 0x19: strcpy(insn[i],"MULTU"); type=MULTDIV; break;
- case 0x1A: strcpy(insn[i],"DIV"); type=MULTDIV; break;
- case 0x1B: strcpy(insn[i],"DIVU"); type=MULTDIV; break;
- case 0x20: strcpy(insn[i],"ADD"); type=ALU; break;
- case 0x21: strcpy(insn[i],"ADDU"); type=ALU; break;
- case 0x22: strcpy(insn[i],"SUB"); type=ALU; break;
- case 0x23: strcpy(insn[i],"SUBU"); type=ALU; break;
- case 0x24: strcpy(insn[i],"AND"); type=ALU; break;
- case 0x25: strcpy(insn[i],"OR"); type=ALU; break;
- case 0x26: strcpy(insn[i],"XOR"); type=ALU; break;
- case 0x27: strcpy(insn[i],"NOR"); type=ALU; break;
- case 0x2A: strcpy(insn[i],"SLT"); type=ALU; break;
- case 0x2B: strcpy(insn[i],"SLTU"); type=ALU; break;
- case 0x30: strcpy(insn[i],"TGE"); type=NI; break;
- case 0x31: strcpy(insn[i],"TGEU"); type=NI; break;
- case 0x32: strcpy(insn[i],"TLT"); type=NI; break;
- case 0x33: strcpy(insn[i],"TLTU"); type=NI; break;
- case 0x34: strcpy(insn[i],"TEQ"); type=NI; break;
- case 0x36: strcpy(insn[i],"TNE"); type=NI; break;
-#ifndef FORCE32
- case 0x14: strcpy(insn[i],"DSLLV"); type=SHIFT; break;
- case 0x16: strcpy(insn[i],"DSRLV"); type=SHIFT; break;
- case 0x17: strcpy(insn[i],"DSRAV"); type=SHIFT; break;
- case 0x1C: strcpy(insn[i],"DMULT"); type=MULTDIV; break;
- case 0x1D: strcpy(insn[i],"DMULTU"); type=MULTDIV; break;
- case 0x1E: strcpy(insn[i],"DDIV"); type=MULTDIV; break;
- case 0x1F: strcpy(insn[i],"DDIVU"); type=MULTDIV; break;
- case 0x2C: strcpy(insn[i],"DADD"); type=ALU; break;
- case 0x2D: strcpy(insn[i],"DADDU"); type=ALU; break;
- case 0x2E: strcpy(insn[i],"DSUB"); type=ALU; break;
- case 0x2F: strcpy(insn[i],"DSUBU"); type=ALU; break;
- case 0x38: strcpy(insn[i],"DSLL"); type=SHIFTIMM; break;
- case 0x3A: strcpy(insn[i],"DSRL"); type=SHIFTIMM; break;
- case 0x3B: strcpy(insn[i],"DSRA"); type=SHIFTIMM; break;
- case 0x3C: strcpy(insn[i],"DSLL32"); type=SHIFTIMM; break;
- case 0x3E: strcpy(insn[i],"DSRL32"); type=SHIFTIMM; break;
- case 0x3F: strcpy(insn[i],"DSRA32"); type=SHIFTIMM; break;
+ case 0x00: set_mnemonic(i, "SLL"); type=SHIFTIMM; break;
+ case 0x02: set_mnemonic(i, "SRL"); type=SHIFTIMM; break;
+ case 0x03: set_mnemonic(i, "SRA"); type=SHIFTIMM; break;
+ case 0x04: set_mnemonic(i, "SLLV"); type=SHIFT; break;
+ case 0x06: set_mnemonic(i, "SRLV"); type=SHIFT; break;
+ case 0x07: set_mnemonic(i, "SRAV"); type=SHIFT; break;
+ case 0x08: set_mnemonic(i, "JR"); type=RJUMP; break;
+ case 0x09: set_mnemonic(i, "JALR"); type=RJUMP; break;
+ case 0x0C: set_mnemonic(i, "SYSCALL"); type=SYSCALL; break;
+ case 0x0D: set_mnemonic(i, "BREAK"); type=SYSCALL; break;
+ case 0x0F: set_mnemonic(i, "SYNC"); type=OTHER; break;
+ case 0x10: set_mnemonic(i, "MFHI"); type=MOV; break;
+ case 0x11: set_mnemonic(i, "MTHI"); type=MOV; break;
+ case 0x12: set_mnemonic(i, "MFLO"); type=MOV; break;
+ case 0x13: set_mnemonic(i, "MTLO"); type=MOV; break;
+ case 0x18: set_mnemonic(i, "MULT"); type=MULTDIV; break;
+ case 0x19: set_mnemonic(i, "MULTU"); type=MULTDIV; break;
+ case 0x1A: set_mnemonic(i, "DIV"); type=MULTDIV; break;
+ case 0x1B: set_mnemonic(i, "DIVU"); type=MULTDIV; break;
+ case 0x20: set_mnemonic(i, "ADD"); type=ALU; break;
+ case 0x21: set_mnemonic(i, "ADDU"); type=ALU; break;
+ case 0x22: set_mnemonic(i, "SUB"); type=ALU; break;
+ case 0x23: set_mnemonic(i, "SUBU"); type=ALU; break;
+ case 0x24: set_mnemonic(i, "AND"); type=ALU; break;
+ case 0x25: set_mnemonic(i, "OR"); type=ALU; break;
+ case 0x26: set_mnemonic(i, "XOR"); type=ALU; break;
+ case 0x27: set_mnemonic(i, "NOR"); type=ALU; break;
+ case 0x2A: set_mnemonic(i, "SLT"); type=ALU; break;
+ case 0x2B: set_mnemonic(i, "SLTU"); type=ALU; break;
+ case 0x30: set_mnemonic(i, "TGE"); type=NI; break;
+ case 0x31: set_mnemonic(i, "TGEU"); type=NI; break;
+ case 0x32: set_mnemonic(i, "TLT"); type=NI; break;
+ case 0x33: set_mnemonic(i, "TLTU"); type=NI; break;
+ case 0x34: set_mnemonic(i, "TEQ"); type=NI; break;
+ case 0x36: set_mnemonic(i, "TNE"); type=NI; break;
+#if 0
+ case 0x14: set_mnemonic(i, "DSLLV"); type=SHIFT; break;
+ case 0x16: set_mnemonic(i, "DSRLV"); type=SHIFT; break;
+ case 0x17: set_mnemonic(i, "DSRAV"); type=SHIFT; break;
+ case 0x1C: set_mnemonic(i, "DMULT"); type=MULTDIV; break;
+ case 0x1D: set_mnemonic(i, "DMULTU"); type=MULTDIV; break;
+ case 0x1E: set_mnemonic(i, "DDIV"); type=MULTDIV; break;
+ case 0x1F: set_mnemonic(i, "DDIVU"); type=MULTDIV; break;
+ case 0x2C: set_mnemonic(i, "DADD"); type=ALU; break;
+ case 0x2D: set_mnemonic(i, "DADDU"); type=ALU; break;
+ case 0x2E: set_mnemonic(i, "DSUB"); type=ALU; break;
+ case 0x2F: set_mnemonic(i, "DSUBU"); type=ALU; break;
+ case 0x38: set_mnemonic(i, "DSLL"); type=SHIFTIMM; break;
+ case 0x3A: set_mnemonic(i, "DSRL"); type=SHIFTIMM; break;
+ case 0x3B: set_mnemonic(i, "DSRA"); type=SHIFTIMM; break;
+ case 0x3C: set_mnemonic(i, "DSLL32"); type=SHIFTIMM; break;
+ case 0x3E: set_mnemonic(i, "DSRL32"); type=SHIFTIMM; break;
+ case 0x3F: set_mnemonic(i, "DSRA32"); type=SHIFTIMM; break;
#endif
}
break;
- case 0x01: strcpy(insn[i],"regimm"); type=NI;
+ case 0x01: set_mnemonic(i, "regimm"); type=NI;
op2=(source[i]>>16)&0x1f;
switch(op2)
{
- case 0x00: strcpy(insn[i],"BLTZ"); type=SJUMP; break;
- case 0x01: strcpy(insn[i],"BGEZ"); type=SJUMP; break;
- case 0x02: strcpy(insn[i],"BLTZL"); type=SJUMP; break;
- case 0x03: strcpy(insn[i],"BGEZL"); type=SJUMP; break;
- case 0x08: strcpy(insn[i],"TGEI"); type=NI; break;
- case 0x09: strcpy(insn[i],"TGEIU"); type=NI; break;
- case 0x0A: strcpy(insn[i],"TLTI"); type=NI; break;
- case 0x0B: strcpy(insn[i],"TLTIU"); type=NI; break;
- case 0x0C: strcpy(insn[i],"TEQI"); type=NI; break;
- case 0x0E: strcpy(insn[i],"TNEI"); type=NI; break;
- case 0x10: strcpy(insn[i],"BLTZAL"); type=SJUMP; break;
- case 0x11: strcpy(insn[i],"BGEZAL"); type=SJUMP; break;
- case 0x12: strcpy(insn[i],"BLTZALL"); type=SJUMP; break;
- case 0x13: strcpy(insn[i],"BGEZALL"); type=SJUMP; break;
+ case 0x00: set_mnemonic(i, "BLTZ"); type=SJUMP; break;
+ case 0x01: set_mnemonic(i, "BGEZ"); type=SJUMP; break;
+ //case 0x02: set_mnemonic(i, "BLTZL"); type=SJUMP; break;
+ //case 0x03: set_mnemonic(i, "BGEZL"); type=SJUMP; break;
+ //case 0x08: set_mnemonic(i, "TGEI"); type=NI; break;
+ //case 0x09: set_mnemonic(i, "TGEIU"); type=NI; break;
+ //case 0x0A: set_mnemonic(i, "TLTI"); type=NI; break;
+ //case 0x0B: set_mnemonic(i, "TLTIU"); type=NI; break;
+ //case 0x0C: set_mnemonic(i, "TEQI"); type=NI; break;
+ //case 0x0E: set_mnemonic(i, "TNEI"); type=NI; break;
+ case 0x10: set_mnemonic(i, "BLTZAL"); type=SJUMP; break;
+ case 0x11: set_mnemonic(i, "BGEZAL"); type=SJUMP; break;
+ //case 0x12: set_mnemonic(i, "BLTZALL"); type=SJUMP; break;
+ //case 0x13: set_mnemonic(i, "BGEZALL"); type=SJUMP; break;
}
break;
- case 0x02: strcpy(insn[i],"J"); type=UJUMP; break;
- case 0x03: strcpy(insn[i],"JAL"); type=UJUMP; break;
- case 0x04: strcpy(insn[i],"BEQ"); type=CJUMP; break;
- case 0x05: strcpy(insn[i],"BNE"); type=CJUMP; break;
- case 0x06: strcpy(insn[i],"BLEZ"); type=CJUMP; break;
- case 0x07: strcpy(insn[i],"BGTZ"); type=CJUMP; break;
- case 0x08: strcpy(insn[i],"ADDI"); type=IMM16; break;
- case 0x09: strcpy(insn[i],"ADDIU"); type=IMM16; break;
- case 0x0A: strcpy(insn[i],"SLTI"); type=IMM16; break;
- case 0x0B: strcpy(insn[i],"SLTIU"); type=IMM16; break;
- case 0x0C: strcpy(insn[i],"ANDI"); type=IMM16; break;
- case 0x0D: strcpy(insn[i],"ORI"); type=IMM16; break;
- case 0x0E: strcpy(insn[i],"XORI"); type=IMM16; break;
- case 0x0F: strcpy(insn[i],"LUI"); type=IMM16; break;
- case 0x10: strcpy(insn[i],"cop0"); type=NI;
+ case 0x02: set_mnemonic(i, "J"); type=UJUMP; break;
+ case 0x03: set_mnemonic(i, "JAL"); type=UJUMP; break;
+ case 0x04: set_mnemonic(i, "BEQ"); type=CJUMP; break;
+ case 0x05: set_mnemonic(i, "BNE"); type=CJUMP; break;
+ case 0x06: set_mnemonic(i, "BLEZ"); type=CJUMP; break;
+ case 0x07: set_mnemonic(i, "BGTZ"); type=CJUMP; break;
+ case 0x08: set_mnemonic(i, "ADDI"); type=IMM16; break;
+ case 0x09: set_mnemonic(i, "ADDIU"); type=IMM16; break;
+ case 0x0A: set_mnemonic(i, "SLTI"); type=IMM16; break;
+ case 0x0B: set_mnemonic(i, "SLTIU"); type=IMM16; break;
+ case 0x0C: set_mnemonic(i, "ANDI"); type=IMM16; break;
+ case 0x0D: set_mnemonic(i, "ORI"); type=IMM16; break;
+ case 0x0E: set_mnemonic(i, "XORI"); type=IMM16; break;
+ case 0x0F: set_mnemonic(i, "LUI"); type=IMM16; break;
+ case 0x10: set_mnemonic(i, "cop0"); type=NI;
op2=(source[i]>>21)&0x1f;
switch(op2)
{
- case 0x00: strcpy(insn[i],"MFC0"); type=COP0; break;
- case 0x04: strcpy(insn[i],"MTC0"); type=COP0; break;
- case 0x10: strcpy(insn[i],"tlb"); type=NI;
- switch(source[i]&0x3f)
- {
- case 0x01: strcpy(insn[i],"TLBR"); type=COP0; break;
- case 0x02: strcpy(insn[i],"TLBWI"); type=COP0; break;
- case 0x06: strcpy(insn[i],"TLBWR"); type=COP0; break;
- case 0x08: strcpy(insn[i],"TLBP"); type=COP0; break;
-#ifdef PCSX
- case 0x10: strcpy(insn[i],"RFE"); type=COP0; break;
-#else
- case 0x18: strcpy(insn[i],"ERET"); type=COP0; break;
-#endif
- }
+ case 0x00: set_mnemonic(i, "MFC0"); type=COP0; break;
+ case 0x02: set_mnemonic(i, "CFC0"); type=COP0; break;
+ case 0x04: set_mnemonic(i, "MTC0"); type=COP0; break;
+ case 0x06: set_mnemonic(i, "CTC0"); type=COP0; break;
+ case 0x10: set_mnemonic(i, "RFE"); type=COP0; break;
}
break;
- case 0x11: strcpy(insn[i],"cop1"); type=NI;
+ case 0x11: set_mnemonic(i, "cop1"); type=COP1;
op2=(source[i]>>21)&0x1f;
- switch(op2)
- {
- case 0x00: strcpy(insn[i],"MFC1"); type=COP1; break;
- case 0x01: strcpy(insn[i],"DMFC1"); type=COP1; break;
- case 0x02: strcpy(insn[i],"CFC1"); type=COP1; break;
- case 0x04: strcpy(insn[i],"MTC1"); type=COP1; break;
- case 0x05: strcpy(insn[i],"DMTC1"); type=COP1; break;
- case 0x06: strcpy(insn[i],"CTC1"); type=COP1; break;
- case 0x08: strcpy(insn[i],"BC1"); type=FJUMP;
- switch((source[i]>>16)&0x3)
- {
- case 0x00: strcpy(insn[i],"BC1F"); break;
- case 0x01: strcpy(insn[i],"BC1T"); break;
- case 0x02: strcpy(insn[i],"BC1FL"); break;
- case 0x03: strcpy(insn[i],"BC1TL"); break;
- }
- break;
- case 0x10: strcpy(insn[i],"C1.S"); type=NI;
- switch(source[i]&0x3f)
- {
- case 0x00: strcpy(insn[i],"ADD.S"); type=FLOAT; break;
- case 0x01: strcpy(insn[i],"SUB.S"); type=FLOAT; break;
- case 0x02: strcpy(insn[i],"MUL.S"); type=FLOAT; break;
- case 0x03: strcpy(insn[i],"DIV.S"); type=FLOAT; break;
- case 0x04: strcpy(insn[i],"SQRT.S"); type=FLOAT; break;
- case 0x05: strcpy(insn[i],"ABS.S"); type=FLOAT; break;
- case 0x06: strcpy(insn[i],"MOV.S"); type=FLOAT; break;
- case 0x07: strcpy(insn[i],"NEG.S"); type=FLOAT; break;
- case 0x08: strcpy(insn[i],"ROUND.L.S"); type=FCONV; break;
- case 0x09: strcpy(insn[i],"TRUNC.L.S"); type=FCONV; break;
- case 0x0A: strcpy(insn[i],"CEIL.L.S"); type=FCONV; break;
- case 0x0B: strcpy(insn[i],"FLOOR.L.S"); type=FCONV; break;
- case 0x0C: strcpy(insn[i],"ROUND.W.S"); type=FCONV; break;
- case 0x0D: strcpy(insn[i],"TRUNC.W.S"); type=FCONV; break;
- case 0x0E: strcpy(insn[i],"CEIL.W.S"); type=FCONV; break;
- case 0x0F: strcpy(insn[i],"FLOOR.W.S"); type=FCONV; break;
- case 0x21: strcpy(insn[i],"CVT.D.S"); type=FCONV; break;
- case 0x24: strcpy(insn[i],"CVT.W.S"); type=FCONV; break;
- case 0x25: strcpy(insn[i],"CVT.L.S"); type=FCONV; break;
- case 0x30: strcpy(insn[i],"C.F.S"); type=FCOMP; break;
- case 0x31: strcpy(insn[i],"C.UN.S"); type=FCOMP; break;
- case 0x32: strcpy(insn[i],"C.EQ.S"); type=FCOMP; break;
- case 0x33: strcpy(insn[i],"C.UEQ.S"); type=FCOMP; break;
- case 0x34: strcpy(insn[i],"C.OLT.S"); type=FCOMP; break;
- case 0x35: strcpy(insn[i],"C.ULT.S"); type=FCOMP; break;
- case 0x36: strcpy(insn[i],"C.OLE.S"); type=FCOMP; break;
- case 0x37: strcpy(insn[i],"C.ULE.S"); type=FCOMP; break;
- case 0x38: strcpy(insn[i],"C.SF.S"); type=FCOMP; break;
- case 0x39: strcpy(insn[i],"C.NGLE.S"); type=FCOMP; break;
- case 0x3A: strcpy(insn[i],"C.SEQ.S"); type=FCOMP; break;
- case 0x3B: strcpy(insn[i],"C.NGL.S"); type=FCOMP; break;
- case 0x3C: strcpy(insn[i],"C.LT.S"); type=FCOMP; break;
- case 0x3D: strcpy(insn[i],"C.NGE.S"); type=FCOMP; break;
- case 0x3E: strcpy(insn[i],"C.LE.S"); type=FCOMP; break;
- case 0x3F: strcpy(insn[i],"C.NGT.S"); type=FCOMP; break;
- }
- break;
- case 0x11: strcpy(insn[i],"C1.D"); type=NI;
- switch(source[i]&0x3f)
- {
- case 0x00: strcpy(insn[i],"ADD.D"); type=FLOAT; break;
- case 0x01: strcpy(insn[i],"SUB.D"); type=FLOAT; break;
- case 0x02: strcpy(insn[i],"MUL.D"); type=FLOAT; break;
- case 0x03: strcpy(insn[i],"DIV.D"); type=FLOAT; break;
- case 0x04: strcpy(insn[i],"SQRT.D"); type=FLOAT; break;
- case 0x05: strcpy(insn[i],"ABS.D"); type=FLOAT; break;
- case 0x06: strcpy(insn[i],"MOV.D"); type=FLOAT; break;
- case 0x07: strcpy(insn[i],"NEG.D"); type=FLOAT; break;
- case 0x08: strcpy(insn[i],"ROUND.L.D"); type=FCONV; break;
- case 0x09: strcpy(insn[i],"TRUNC.L.D"); type=FCONV; break;
- case 0x0A: strcpy(insn[i],"CEIL.L.D"); type=FCONV; break;
- case 0x0B: strcpy(insn[i],"FLOOR.L.D"); type=FCONV; break;
- case 0x0C: strcpy(insn[i],"ROUND.W.D"); type=FCONV; break;
- case 0x0D: strcpy(insn[i],"TRUNC.W.D"); type=FCONV; break;
- case 0x0E: strcpy(insn[i],"CEIL.W.D"); type=FCONV; break;
- case 0x0F: strcpy(insn[i],"FLOOR.W.D"); type=FCONV; break;
- case 0x20: strcpy(insn[i],"CVT.S.D"); type=FCONV; break;
- case 0x24: strcpy(insn[i],"CVT.W.D"); type=FCONV; break;
- case 0x25: strcpy(insn[i],"CVT.L.D"); type=FCONV; break;
- case 0x30: strcpy(insn[i],"C.F.D"); type=FCOMP; break;
- case 0x31: strcpy(insn[i],"C.UN.D"); type=FCOMP; break;
- case 0x32: strcpy(insn[i],"C.EQ.D"); type=FCOMP; break;
- case 0x33: strcpy(insn[i],"C.UEQ.D"); type=FCOMP; break;
- case 0x34: strcpy(insn[i],"C.OLT.D"); type=FCOMP; break;
- case 0x35: strcpy(insn[i],"C.ULT.D"); type=FCOMP; break;
- case 0x36: strcpy(insn[i],"C.OLE.D"); type=FCOMP; break;
- case 0x37: strcpy(insn[i],"C.ULE.D"); type=FCOMP; break;
- case 0x38: strcpy(insn[i],"C.SF.D"); type=FCOMP; break;
- case 0x39: strcpy(insn[i],"C.NGLE.D"); type=FCOMP; break;
- case 0x3A: strcpy(insn[i],"C.SEQ.D"); type=FCOMP; break;
- case 0x3B: strcpy(insn[i],"C.NGL.D"); type=FCOMP; break;
- case 0x3C: strcpy(insn[i],"C.LT.D"); type=FCOMP; break;
- case 0x3D: strcpy(insn[i],"C.NGE.D"); type=FCOMP; break;
- case 0x3E: strcpy(insn[i],"C.LE.D"); type=FCOMP; break;
- case 0x3F: strcpy(insn[i],"C.NGT.D"); type=FCOMP; break;
- }
- break;
- case 0x14: strcpy(insn[i],"C1.W"); type=NI;
- switch(source[i]&0x3f)
- {
- case 0x20: strcpy(insn[i],"CVT.S.W"); type=FCONV; break;
- case 0x21: strcpy(insn[i],"CVT.D.W"); type=FCONV; break;
- }
- break;
- case 0x15: strcpy(insn[i],"C1.L"); type=NI;
- switch(source[i]&0x3f)
- {
- case 0x20: strcpy(insn[i],"CVT.S.L"); type=FCONV; break;
- case 0x21: strcpy(insn[i],"CVT.D.L"); type=FCONV; break;
- }
- break;
- }
break;
-#ifndef FORCE32
- case 0x14: strcpy(insn[i],"BEQL"); type=CJUMP; break;
- case 0x15: strcpy(insn[i],"BNEL"); type=CJUMP; break;
- case 0x16: strcpy(insn[i],"BLEZL"); type=CJUMP; break;
- case 0x17: strcpy(insn[i],"BGTZL"); type=CJUMP; break;
- case 0x18: strcpy(insn[i],"DADDI"); type=IMM16; break;
- case 0x19: strcpy(insn[i],"DADDIU"); type=IMM16; break;
- case 0x1A: strcpy(insn[i],"LDL"); type=LOADLR; break;
- case 0x1B: strcpy(insn[i],"LDR"); type=LOADLR; break;
+#if 0
+ case 0x14: set_mnemonic(i, "BEQL"); type=CJUMP; break;
+ case 0x15: set_mnemonic(i, "BNEL"); type=CJUMP; break;
+ case 0x16: set_mnemonic(i, "BLEZL"); type=CJUMP; break;
+ case 0x17: set_mnemonic(i, "BGTZL"); type=CJUMP; break;
+ case 0x18: set_mnemonic(i, "DADDI"); type=IMM16; break;
+ case 0x19: set_mnemonic(i, "DADDIU"); type=IMM16; break;
+ case 0x1A: set_mnemonic(i, "LDL"); type=LOADLR; break;
+ case 0x1B: set_mnemonic(i, "LDR"); type=LOADLR; break;
#endif
- case 0x20: strcpy(insn[i],"LB"); type=LOAD; break;
- case 0x21: strcpy(insn[i],"LH"); type=LOAD; break;
- case 0x22: strcpy(insn[i],"LWL"); type=LOADLR; break;
- case 0x23: strcpy(insn[i],"LW"); type=LOAD; break;
- case 0x24: strcpy(insn[i],"LBU"); type=LOAD; break;
- case 0x25: strcpy(insn[i],"LHU"); type=LOAD; break;
- case 0x26: strcpy(insn[i],"LWR"); type=LOADLR; break;
-#ifndef FORCE32
- case 0x27: strcpy(insn[i],"LWU"); type=LOAD; break;
+ case 0x20: set_mnemonic(i, "LB"); type=LOAD; break;
+ case 0x21: set_mnemonic(i, "LH"); type=LOAD; break;
+ case 0x22: set_mnemonic(i, "LWL"); type=LOADLR; break;
+ case 0x23: set_mnemonic(i, "LW"); type=LOAD; break;
+ case 0x24: set_mnemonic(i, "LBU"); type=LOAD; break;
+ case 0x25: set_mnemonic(i, "LHU"); type=LOAD; break;
+ case 0x26: set_mnemonic(i, "LWR"); type=LOADLR; break;
+#if 0
+ case 0x27: set_mnemonic(i, "LWU"); type=LOAD; break;
#endif
- case 0x28: strcpy(insn[i],"SB"); type=STORE; break;
- case 0x29: strcpy(insn[i],"SH"); type=STORE; break;
- case 0x2A: strcpy(insn[i],"SWL"); type=STORELR; break;
- case 0x2B: strcpy(insn[i],"SW"); type=STORE; break;
-#ifndef FORCE32
- case 0x2C: strcpy(insn[i],"SDL"); type=STORELR; break;
- case 0x2D: strcpy(insn[i],"SDR"); type=STORELR; break;
+ case 0x28: set_mnemonic(i, "SB"); type=STORE; break;
+ case 0x29: set_mnemonic(i, "SH"); type=STORE; break;
+ case 0x2A: set_mnemonic(i, "SWL"); type=STORELR; break;
+ case 0x2B: set_mnemonic(i, "SW"); type=STORE; break;
+#if 0
+ case 0x2C: set_mnemonic(i, "SDL"); type=STORELR; break;
+ case 0x2D: set_mnemonic(i, "SDR"); type=STORELR; break;
#endif
- case 0x2E: strcpy(insn[i],"SWR"); type=STORELR; break;
- case 0x2F: strcpy(insn[i],"CACHE"); type=NOP; break;
- case 0x30: strcpy(insn[i],"LL"); type=NI; break;
- case 0x31: strcpy(insn[i],"LWC1"); type=C1LS; break;
-#ifndef FORCE32
- case 0x34: strcpy(insn[i],"LLD"); type=NI; break;
- case 0x35: strcpy(insn[i],"LDC1"); type=C1LS; break;
- case 0x37: strcpy(insn[i],"LD"); type=LOAD; break;
+ case 0x2E: set_mnemonic(i, "SWR"); type=STORELR; break;
+ case 0x2F: set_mnemonic(i, "CACHE"); type=NOP; break;
+ case 0x30: set_mnemonic(i, "LL"); type=NI; break;
+ case 0x31: set_mnemonic(i, "LWC1"); type=C1LS; break;
+#if 0
+ case 0x34: set_mnemonic(i, "LLD"); type=NI; break;
+ case 0x35: set_mnemonic(i, "LDC1"); type=C1LS; break;
+ case 0x37: set_mnemonic(i, "LD"); type=LOAD; break;
#endif
- case 0x38: strcpy(insn[i],"SC"); type=NI; break;
- case 0x39: strcpy(insn[i],"SWC1"); type=C1LS; break;
-#ifndef FORCE32
- case 0x3C: strcpy(insn[i],"SCD"); type=NI; break;
- case 0x3D: strcpy(insn[i],"SDC1"); type=C1LS; break;
- case 0x3F: strcpy(insn[i],"SD"); type=STORE; break;
+ case 0x38: set_mnemonic(i, "SC"); type=NI; break;
+ case 0x39: set_mnemonic(i, "SWC1"); type=C1LS; break;
+#if 0
+ case 0x3C: set_mnemonic(i, "SCD"); type=NI; break;
+ case 0x3D: set_mnemonic(i, "SDC1"); type=C1LS; break;
+ case 0x3F: set_mnemonic(i, "SD"); type=STORE; break;
#endif
-#ifdef PCSX
- case 0x12: strcpy(insn[i],"COP2"); type=NI;
- // note: COP MIPS-1 encoding differs from MIPS32
+ case 0x12: set_mnemonic(i, "COP2"); type=NI;
op2=(source[i]>>21)&0x1f;
- if (source[i]&0x3f) {
+ //if (op2 & 0x10)
+ if (source[i]&0x3f) { // use this hack to support old savestates with patched gte insns
if (gte_handlers[source[i]&0x3f]!=NULL) {
- snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
+#ifdef DISASM
+ if (gte_regnames[source[i]&0x3f]!=NULL)
+ strcpy(insn[i],gte_regnames[source[i]&0x3f]);
+ else
+ snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
+#endif
type=C2OP;
}
}
else switch(op2)
{
- case 0x00: strcpy(insn[i],"MFC2"); type=COP2; break;
- case 0x02: strcpy(insn[i],"CFC2"); type=COP2; break;
- case 0x04: strcpy(insn[i],"MTC2"); type=COP2; break;
- case 0x06: strcpy(insn[i],"CTC2"); type=COP2; break;
+ case 0x00: set_mnemonic(i, "MFC2"); type=COP2; break;
+ case 0x02: set_mnemonic(i, "CFC2"); type=COP2; break;
+ case 0x04: set_mnemonic(i, "MTC2"); type=COP2; break;
+ case 0x06: set_mnemonic(i, "CTC2"); type=COP2; break;
}
break;
- case 0x32: strcpy(insn[i],"LWC2"); type=C2LS; break;
- case 0x3A: strcpy(insn[i],"SWC2"); type=C2LS; break;
- case 0x3B: strcpy(insn[i],"HLECALL"); type=HLECALL; break;
-#endif
- default: strcpy(insn[i],"???"); type=NI;
- printf("NI %08x @%08x (%08x)\n", source[i], addr + i*4, addr);
+ case 0x32: set_mnemonic(i, "LWC2"); type=C2LS; break;
+ case 0x3A: set_mnemonic(i, "SWC2"); type=C2LS; break;
+ case 0x3B: set_mnemonic(i, "HLECALL"); type=HLECALL; break;
+ default: set_mnemonic(i, "???"); type=NI;
+ SysPrintf("NI %08x @%08x (%08x)\n", source[i], start + i*4, start);
break;
}
- itype[i]=type;
- opcode2[i]=op2;
+ dops[i].itype=type;
+ dops[i].opcode2=op2;
/* Get registers/immediates */
- lt1[i]=0;
- us1[i]=0;
- us2[i]=0;
- dep1[i]=0;
- dep2[i]=0;
+ dops[i].use_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];
- if(op==0x2c||op==0x2d||op==0x3f) us1[i]=rs2[i]; // 64-bit SDL/SDR/SD
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==0x1a||op==0x1b) us1[i]=rs2[i]; // LDR/LDL
- 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==0x18||op==0x19) us1[i]=rs1[i]; // DADDI/DADDIU
- if(op==0x0a||op==0x0b) us1[i]=rs1[i]; // SLTI/SLTIU
- 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;
}
- us1[i]=rs1[i];
- us2[i]=rs2[i];
- likely[i]=op>>4;
break;
case SJUMP:
- rs1[i]=(source[i]>>21)&0x1f;
- rs2[i]=CCREG;
- rt1[i]=0;
- rt2[i]=0;
- us1[i]=rs1[i];
+ 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 FJUMP:
- rs1[i]=FSREG;
- rs2[i]=CSREG;
- rt1[i]=0;
- rt2[i]=0;
- likely[i]=((source[i])>>17)&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==0x2a||op2==0x2b) { // SLT/SLTU
- us1[i]=rs1[i];us2[i]=rs2[i];
- }
- else 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;
- if (op2>=0x1c&&op2<=0x1f) { // DMULT/DMULTU/DDIV/DDIVU
- us1[i]=rs1[i];us2[i]=rs2[i];
- }
+ 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;
- // DSLLV/DSRLV/DSRAV are 64-bit
- if(op2>=0x14&&op2<=0x17) us1[i]=rs1[i];
+ 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;
- // DSLL/DSRL/DSRA/DSRA32/DSRL32 but not DSLL32 require 64-bit source
- if(op2>=0x38&&op2!=0x3c) us1[i]=rs1[i];
break;
case COP0:
- rs1[i]=0;
- rs2[i]=0;
- rt1[i]=0;
- rt2[i]=0;
- if(op2==0) rt1[i]=(source[i]>>16)&0x1F; // MFC0
- if(op2==4) rs1[i]=(source[i]>>16)&0x1F; // MTC0
- 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:
+ 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; // MFC1/DMFC1/CFC1
- if(op2>3) rs1[i]=(source[i]>>16)&0x1F; // MTC1/DMTC1/CTC1
- if(op2==5) us1[i]=rs1[i]; // DMTC1
- 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)
+ {
+ case 0x00: gte_rs[i]=1ll<<gr; break; // MFC2
+ case 0x04: gte_rt[i]=1ll<<gr; break; // MTC2
+ case 0x02: gte_rs[i]=1ll<<(gr+32); break; // CFC2
+ case 0x06: gte_rt[i]=1ll<<(gr+32); break; // CTC2
+ }
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 FLOAT:
- case FCONV:
- rs1[i]=0;
- rs2[i]=CSREG;
- rt1[i]=0;
- rt2[i]=0;
- break;
- case FCOMP:
- rs1[i]=FSREG;
- rs2[i]=CSREG;
- rt1[i]=FSREG;
- rt2[i]=0;
+ case C2OP:
+ 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
+ if((source[i]&0x3f)==GTE_MVMVA) {
+ int v = (source[i] >> 15) & 3;
+ gte_rs[i]&=~0xe3fll;
+ if(v==3) gte_rs[i]|=0xe00ll;
+ else gte_rs[i]|=3ll<<(v*2);
+ }
break;
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||type==FJUMP)
+ else if(type==CJUMP||type==SJUMP)
ba[i]=start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14);
else ba[i]=-1;
-#ifdef PCSX
- if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)) {
+
+ /* 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||type==FJUMP) {
+ if (dops[i].is_jump) {
// don't handle first branch and call interpreter if it's hit
- printf("branch in delay slot @%08x (%08x)\n", addr + i*4, addr);
+ SysPrintf("branch in delay slot @%08x (%08x)\n", start + i*4, start);
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
- printf("load delay @%08x (%08x)\n", addr + i*4, addr);
+ SysPrintf("load delay @%08x (%08x)\n", start + i*4, start);
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&&(itype[i-3]==RJUMP||itype[i-3]==UJUMP||itype[i-3]==CJUMP||itype[i-3]==SJUMP))) {
+ 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
- printf("v0 overwrite @%08x (%08x)\n", addr + i*4, addr);
+ SysPrintf("v0 overwrite @%08x (%08x)\n", start + i*4, start);
do_in_intrp=1;
}
}
- if(do_in_intrp) {
- rs1[i-1]=CCREG;
- rs2[i-1]=rt1[i-1]=rt2[i-1]=0;
- ba[i-1]=-1;
- itype[i-1]=INTCALL;
- done=2;
+ if (do_in_intrp) {
+ memset(&dops[i-1], 0, sizeof(dops[i-1]));
+ dops[i-1].itype = INTCALL;
+ dops[i-1].rs1 = CCREG;
+ ba[i-1] = -1;
+ done = 2;
i--; // don't compile the DS
}
}
-#endif
+
/* Is this the end of the block? */
- if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)) {
- 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 {
// 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(done==2) {
+ if (dops[i].itype == SYSCALL || dops[i].itype == HLECALL || dops[i].itype == INTCALL)
+ done = stop_after_jal ? 1 : 2;
+ if (done == 2) {
// Does the block continue due to a branch?
for(j=i-1;j>=0;j--)
{
assert(start+i*4<pagelimit);
if (i==MAXBLOCK-1) done=1;
// Stop if we're compiling junk
- if(itype[i]==NI&&opcode[i]==0x11) {
+ if(dops[i].itype == NI && (++ni_count > 8 || dops[i].opcode == 0x11)) {
done=stop_after_jal=1;
- printf("Disabled speculative precompilation\n");
+ SysPrintf("Disabled speculative precompilation\n");
}
}
slen=i;
- if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==RJUMP||itype[i-1]==FJUMP) {
+ if (dops[i-1].is_jump) {
if(start+i*4==pagelimit) {
- itype[i-1]=SPAN;
+ dops[i-1].itype=SPAN;
}
}
assert(slen>0);
+}
- /* Pass 2 - Register dependencies and branch targets */
-
- unneeded_registers(0,slen-1,0);
-
- /* Pass 3 - Register allocation */
-
- struct regstat current; // Current register allocations/status
- current.is32=1;
- current.dirty=0;
- current.u=unneeded_reg[0];
- current.uu=unneeded_reg_upper[0];
- clear_all_regs(current.regmap);
- alloc_reg(¤t,0,CCREG);
- dirty_reg(¤t,CCREG);
- current.isconst=0;
- current.wasconst=0;
- int ds=0;
- int cc=0;
- int hr=-1;
-
-#ifndef FORCE32
- provisional_32bit();
-#endif
- if((u_int)addr&1) {
- // First instruction is delay slot
- cc=-1;
- bt[1]=1;
- ds=1;
- unneeded_reg[0]=1;
- unneeded_reg_upper[0]=1;
- current.regmap[HOST_BTREG]=BTREG;
+// Basic liveness analysis for MIPS registers
+static noinline void pass2_unneeded_regs(int istart,int iend,int r)
+{
+ int i;
+ uint64_t u,gte_u,b,gte_b;
+ uint64_t temp_u,temp_gte_u=0;
+ uint64_t gte_u_unknown=0;
+ if (HACK_ENABLED(NDHACK_GTE_UNNEEDED))
+ gte_u_unknown=~0ll;
+ if(iend==slen-1) {
+ u=1;
+ gte_u=gte_u_unknown;
+ }else{
+ //u=unneeded_reg[iend+1];
+ u=1;
+ gte_u=gte_unneeded[iend+1];
}
-
- for(i=0;i<slen;i++)
+
+ for (i=iend;i>=istart;i--)
{
- if(bt[i])
+ //printf("unneeded registers i=%d (%d,%d) r=%d\n",i,istart,iend,r);
+ if(dops[i].is_jump)
{
- int hr;
- for(hr=0;hr<HOST_REGS;hr++)
+ // If subroutine call, flag return address as a possible branch target
+ if(dops[i].rt1==31 && i<slen-2) dops[i+2].bt=1;
+
+ if(ba[i]<start || ba[i]>=(start+slen*4))
{
- // Is this really necessary?
- if(current.regmap[hr]==0) current.regmap[hr]=-1;
+ // Branch out of this block, flush all regs
+ u=1;
+ gte_u=gte_u_unknown;
+ branch_unneeded_reg[i]=u;
+ // Merge in delay slot
+ u|=(1LL<<dops[i+1].rt1)|(1LL<<dops[i+1].rt2);
+ u&=~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
+ u|=1;
+ gte_u|=gte_rt[i+1];
+ gte_u&=~gte_rs[i+1];
}
- current.isconst=0;
- }
- if(i>1)
- {
- if((opcode[i-2]&0x2f)==0x05) // BNE/BNEL
+ else
{
- if(rs1[i-2]==0||rs2[i-2]==0)
- {
- if(rs1[i-2]) {
- current.is32|=1LL<<rs1[i-2];
- int hr=get_reg(current.regmap,rs1[i-2]|64);
- if(hr>=0) current.regmap[hr]=-1;
+ // Internal branch, flag target
+ dops[(ba[i]-start)>>2].bt=1;
+ if(ba[i]<=start+i*4) {
+ // Backward branch
+ if(dops[i].is_ujump)
+ {
+ // Unconditional branch
+ temp_u=1;
+ temp_gte_u=0;
+ } else {
+ // Conditional branch (not taken case)
+ temp_u=unneeded_reg[i+2];
+ temp_gte_u&=gte_unneeded[i+2];
}
- if(rs2[i-2]) {
- current.is32|=1LL<<rs2[i-2];
- int hr=get_reg(current.regmap,rs2[i-2]|64);
- if(hr>=0) current.regmap[hr]=-1;
+ // Merge in delay slot
+ temp_u|=(1LL<<dops[i+1].rt1)|(1LL<<dops[i+1].rt2);
+ temp_u&=~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
+ temp_u|=1;
+ temp_gte_u|=gte_rt[i+1];
+ temp_gte_u&=~gte_rs[i+1];
+ temp_u|=(1LL<<dops[i].rt1)|(1LL<<dops[i].rt2);
+ temp_u&=~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
+ temp_u|=1;
+ temp_gte_u|=gte_rt[i];
+ temp_gte_u&=~gte_rs[i];
+ unneeded_reg[i]=temp_u;
+ gte_unneeded[i]=temp_gte_u;
+ // Only go three levels deep. This recursion can take an
+ // excessive amount of time if there are a lot of nested loops.
+ if(r<2) {
+ pass2_unneeded_regs((ba[i]-start)>>2,i-1,r+1);
+ }else{
+ unneeded_reg[(ba[i]-start)>>2]=1;
+ gte_unneeded[(ba[i]-start)>>2]=gte_u_unknown;
}
- }
- }
- }
-#ifndef FORCE32
- // If something jumps here with 64-bit values
- // then promote those registers to 64 bits
- if(bt[i])
- {
- uint64_t temp_is32=current.is32;
- for(j=i-1;j>=0;j--)
- {
- if(ba[j]==start+i*4)
- temp_is32&=branch_regs[j].is32;
- }
- for(j=i;j<slen;j++)
- {
- if(ba[j]==start+i*4)
- //temp_is32=1;
- temp_is32&=p32[j];
- }
- if(temp_is32!=current.is32) {
- //printf("dumping 32-bit regs (%x)\n",start+i*4);
- #ifndef DESTRUCTIVE_WRITEBACK
- if(ds)
- #endif
- for(hr=0;hr<HOST_REGS;hr++)
- {
- int r=current.regmap[hr];
- if(r>0&&r<64)
+ } /*else*/ if(1) {
+ if (dops[i].is_ujump)
{
- if((current.dirty>>hr)&((current.is32&~temp_is32)>>r)&1) {
- temp_is32|=1LL<<r;
- //printf("restore %d\n",r);
+ // 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<<dops[i+1].rt1)|(1LL<<dops[i+1].rt2);
+ u&=~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
+ u|=1;
+ gte_u|=gte_rt[i+1];
+ gte_u&=~gte_rs[i+1];
+ } else {
+ // Conditional branch
+ b=unneeded_reg[(ba[i]-start)>>2];
+ gte_b=gte_unneeded[(ba[i]-start)>>2];
+ branch_unneeded_reg[i]=b;
+ // Branch delay slot
+ b|=(1LL<<dops[i+1].rt1)|(1LL<<dops[i+1].rt2);
+ b&=~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
+ b|=1;
+ gte_b|=gte_rt[i+1];
+ gte_b&=~gte_rs[i+1];
+ u&=b;
+ gte_u&=gte_b;
+ if(i<slen-1) {
+ branch_unneeded_reg[i]&=unneeded_reg[i+2];
+ } else {
+ branch_unneeded_reg[i]=1;
}
}
}
- current.is32=temp_is32;
}
}
-#else
- current.is32=-1LL;
-#endif
-
- memcpy(regmap_pre[i],current.regmap,sizeof(current.regmap));
- regs[i].wasconst=current.isconst;
- regs[i].was32=current.is32;
- regs[i].wasdirty=current.dirty;
- #if defined(DESTRUCTIVE_WRITEBACK) && !defined(FORCE32)
- // To change a dirty register from 32 to 64 bits, we must write
- // it out during the previous cycle (for branches, 2 cycles)
- if(i<slen-1&&bt[i+1]&&itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP)
+ else if(dops[i].itype==SYSCALL||dops[i].itype==HLECALL||dops[i].itype==INTCALL)
{
- uint64_t temp_is32=current.is32;
- for(j=i-1;j>=0;j--)
- {
- if(ba[j]==start+i*4+4)
- temp_is32&=branch_regs[j].is32;
- }
- for(j=i;j<slen;j++)
- {
- if(ba[j]==start+i*4+4)
- //temp_is32=1;
- temp_is32&=p32[j];
- }
- if(temp_is32!=current.is32) {
- //printf("pre-dumping 32-bit regs (%x)\n",start+i*4);
- for(hr=0;hr<HOST_REGS;hr++)
- {
- int r=current.regmap[hr];
- if(r>0)
- {
- if((current.dirty>>hr)&((current.is32&~temp_is32)>>(r&63))&1) {
- if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP&&itype[i]!=FJUMP)
- {
- if(rs1[i]!=(r&63)&&rs2[i]!=(r&63))
- {
- //printf("dump %d/r%d\n",hr,r);
- current.regmap[hr]=-1;
- if(get_reg(current.regmap,r|64)>=0)
- current.regmap[get_reg(current.regmap,r|64)]=-1;
- }
- }
- }
- }
- }
- }
+ // SYSCALL instruction (software interrupt)
+ u=1;
}
- else if(i<slen-2&&bt[i+2]&&(source[i-1]>>16)!=0x1000&&(itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP))
+ else if(dops[i].itype==COP0 && (source[i]&0x3f)==0x18)
{
- uint64_t temp_is32=current.is32;
- for(j=i-1;j>=0;j--)
- {
- if(ba[j]==start+i*4+8)
- temp_is32&=branch_regs[j].is32;
- }
- for(j=i;j<slen;j++)
- {
- if(ba[j]==start+i*4+8)
- //temp_is32=1;
- temp_is32&=p32[j];
+ // ERET instruction (return from interrupt)
+ u=1;
+ }
+ //u=1; // DEBUG
+ // Written registers are unneeded
+ u|=1LL<<dops[i].rt1;
+ u|=1LL<<dops[i].rt2;
+ gte_u|=gte_rt[i];
+ // Accessed registers are needed
+ u&=~(1LL<<dops[i].rs1);
+ u&=~(1LL<<dops[i].rs2);
+ gte_u&=~gte_rs[i];
+ if(gte_rs[i]&&dops[i].rt1&&(unneeded_reg[i+1]&(1ll<<dops[i].rt1)))
+ gte_u|=gte_rs[i]>e_unneeded[i+1]; // MFC2/CFC2 to dead register, unneeded
+ // Source-target dependencies
+ // R0 is always unneeded
+ u|=1;
+ // Save it
+ unneeded_reg[i]=u;
+ gte_unneeded[i]=gte_u;
+ /*
+ printf("ur (%d,%d) %x: ",istart,iend,start+i*4);
+ printf("U:");
+ int r;
+ for(r=1;r<=CCREG;r++) {
+ if((unneeded_reg[i]>>r)&1) {
+ if(r==HIREG) printf(" HI");
+ else if(r==LOREG) printf(" LO");
+ else printf(" r%d",r);
}
- if(temp_is32!=current.is32) {
- //printf("pre-dumping 32-bit regs (%x)\n",start+i*4);
- for(hr=0;hr<HOST_REGS;hr++)
- {
- int r=current.regmap[hr];
- if(r>0)
- {
- if((current.dirty>>hr)&((current.is32&~temp_is32)>>(r&63))&1) {
- if(rs1[i]!=(r&63)&&rs2[i]!=(r&63)&&rs1[i+1]!=(r&63)&&rs2[i+1]!=(r&63))
- {
- //printf("dump %d/r%d\n",hr,r);
- current.regmap[hr]=-1;
- if(get_reg(current.regmap,r|64)>=0)
- current.regmap[get_reg(current.regmap,r|64)]=-1;
- }
- }
- }
- }
+ }
+ printf("\n");
+ */
+ }
+}
+
+static noinline void pass3_register_alloc(u_int addr)
+{
+ struct regstat current; // Current register allocations/status
+ clear_all_regs(current.regmap_entry);
+ clear_all_regs(current.regmap);
+ current.wasdirty = current.dirty = 0;
+ current.u = unneeded_reg[0];
+ alloc_reg(¤t, 0, CCREG);
+ dirty_reg(¤t, CCREG);
+ current.wasconst = 0;
+ current.isconst = 0;
+ current.loadedconst = 0;
+ current.waswritten = 0;
+ int ds=0;
+ int cc=0;
+ int hr;
+ int i, j;
+
+ if (addr & 1) {
+ // First instruction is delay slot
+ cc=-1;
+ dops[1].bt=1;
+ ds=1;
+ unneeded_reg[0]=1;
+ current.regmap[HOST_BTREG]=BTREG;
+ }
+
+ for(i=0;i<slen;i++)
+ {
+ if(dops[i].bt)
+ {
+ for(hr=0;hr<HOST_REGS;hr++)
+ {
+ // Is this really necessary?
+ if(current.regmap[hr]==0) current.regmap[hr]=-1;
}
+ current.isconst=0;
+ current.waswritten=0;
}
- #endif
- if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP&&itype[i]!=FJUMP) {
+
+ memcpy(regmap_pre[i],current.regmap,sizeof(current.regmap));
+ regs[i].wasconst=current.isconst;
+ regs[i].wasdirty=current.dirty;
+ regs[i].dirty=0;
+ regs[i].u=0;
+ regs[i].isconst=0;
+ regs[i].loadedconst=0;
+ if (!dops[i].is_jump) {
if(i+1<slen) {
- current.u=unneeded_reg[i+1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
- current.uu=unneeded_reg_upper[i+1]&~((1LL<<us1[i])|(1LL<<us2[i]));
- if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+ current.u=unneeded_reg[i+1]&~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
current.u|=1;
- current.uu|=1;
} else {
current.u=1;
- current.uu=1;
}
} else {
if(i+1<slen) {
- current.u=branch_unneeded_reg[i]&~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
- current.uu=branch_unneeded_reg_upper[i]&~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- if((~current.uu>>rt1[i+1])&1) current.uu&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
- current.u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
- current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
+ current.u=branch_unneeded_reg[i]&~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
+ current.u&=~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
current.u|=1;
- current.uu|=1;
- } else { printf("oops, branch at end of block with no delay slot\n");exit(1); }
+ } else {
+ SysPrintf("oops, branch at end of block with no delay slot @%08x\n", start + i*4);
+ abort();
+ }
}
- is_ds[i]=ds;
+ dops[i].is_ds=ds;
if(ds) {
ds=0; // Skip delay slot, already allocated as part of branch
// ...but we need to alloc it in case something jumps here
if(i+1<slen) {
current.u=branch_unneeded_reg[i-1]&unneeded_reg[i+1];
- current.uu=branch_unneeded_reg_upper[i-1]&unneeded_reg_upper[i+1];
}else{
current.u=branch_unneeded_reg[i-1];
- current.uu=branch_unneeded_reg_upper[i-1];
}
- current.u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
- current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
- if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+ current.u&=~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
current.u|=1;
- current.uu|=1;
struct regstat temp;
memcpy(&temp,¤t,sizeof(current));
temp.wasdirty=temp.dirty;
- temp.was32=temp.is32;
// TODO: Take into account unconditional branches, as below
delayslot_alloc(&temp,i);
memcpy(regs[i].regmap,temp.regmap,sizeof(temp.regmap));
regs[i].wasdirty=temp.wasdirty;
- regs[i].was32=temp.was32;
regs[i].dirty=temp.dirty;
- regs[i].is32=temp.is32;
regs[i].isconst=0;
regs[i].wasconst=0;
current.isconst=0;
}
else
{
- if(r<64){
+ assert(r < 64);
if((current.u>>r)&1) {
regs[i].regmap_entry[hr]=-1;
regs[i].regmap[hr]=-1;
//current.regmap[hr]=-1;
}else
regs[i].regmap_entry[hr]=r;
- }
- else {
- if((current.uu>>(r&63))&1) {
- regs[i].regmap_entry[hr]=-1;
- regs[i].regmap[hr]=-1;
- //Don't clear regs in the delay slot as the branch might need them
- //current.regmap[hr]=-1;
- }else
- regs[i].regmap_entry[hr]=r;
- }
}
} 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;
}
}
else { // Not delay slot
- switch(itype[i]) {
+ switch(dops[i].itype) {
case UJUMP:
//current.isconst=0; // DEBUG
//current.wasconst=0; // DEBUG
//regs[i].wasconst=0; // DEBUG
- clear_const(¤t,rt1[i]);
+ clear_const(¤t,dops[i].rt1);
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
- if (rt1[i]==31) {
+ if (dops[i].rt1==31) {
alloc_reg(¤t,i,31);
dirty_reg(¤t,31);
- //assert(rs1[i+1]!=31&&rs2[i+1]!=31);
- //assert(rt1[i+1]!=rt1[i]);
+ //assert(dops[i+1].rs1!=31&&dops[i+1].rs2!=31);
+ //assert(dops[i+1].rt1!=dops[i].rt1);
#ifdef REG_PREFETCH
alloc_reg(¤t,i,PTEMP);
#endif
- //current.is32|=1LL<<rt1[i];
}
- ooo[i]=1;
+ dops[i].ooo=1;
delayslot_alloc(¤t,i+1);
//current.isconst=0; // DEBUG
ds=1;
//current.isconst=0;
//current.wasconst=0;
//regs[i].wasconst=0;
- clear_const(¤t,rs1[i]);
- clear_const(¤t,rt1[i]);
+ clear_const(¤t,dops[i].rs1);
+ clear_const(¤t,dops[i].rt1);
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
- if(rs1[i]!=rt1[i+1]&&rs1[i]!=rt2[i+1]) {
- alloc_reg(¤t,i,rs1[i]);
- if (rt1[i]!=0) {
- alloc_reg(¤t,i,rt1[i]);
- dirty_reg(¤t,rt1[i]);
- assert(rs1[i+1]!=rt1[i]&&rs2[i+1]!=rt1[i]);
- assert(rt1[i+1]!=rt1[i]);
+ if (!ds_writes_rjump_rs(i)) {
+ alloc_reg(¤t,i,dops[i].rs1);
+ if (dops[i].rt1!=0) {
+ alloc_reg(¤t,i,dops[i].rt1);
+ dirty_reg(¤t,dops[i].rt1);
+ assert(dops[i+1].rs1!=dops[i].rt1&&dops[i+1].rs2!=dops[i].rt1);
+ assert(dops[i+1].rt1!=dops[i].rt1);
#ifdef REG_PREFETCH
alloc_reg(¤t,i,PTEMP);
#endif
}
#ifdef USE_MINI_HT
- if(rs1[i]==31) { // JALR
+ if(dops[i].rs1==31) { // JALR
alloc_reg(¤t,i,RHASH);
- #ifndef HOST_IMM_ADDR32
alloc_reg(¤t,i,RHTBL);
- #endif
}
#endif
delayslot_alloc(¤t,i+1);
alloc_reg(¤t,i,RTEMP);
}
//current.isconst=0; // DEBUG
- ooo[i]=1;
+ dops[i].ooo=1;
ds=1;
break;
case CJUMP:
//current.isconst=0;
//current.wasconst=0;
//regs[i].wasconst=0;
- clear_const(¤t,rs1[i]);
- clear_const(¤t,rs2[i]);
- if((opcode[i]&0x3E)==4) // BEQ/BNE
+ clear_const(¤t,dops[i].rs1);
+ clear_const(¤t,dops[i].rs2);
+ if((dops[i].opcode&0x3E)==4) // BEQ/BNE
{
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
- if(rs1[i]) alloc_reg(¤t,i,rs1[i]);
- if(rs2[i]) alloc_reg(¤t,i,rs2[i]);
- if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
- {
- if(rs1[i]) alloc_reg64(¤t,i,rs1[i]);
- if(rs2[i]) alloc_reg64(¤t,i,rs2[i]);
- }
- if((rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))||
- (rs2[i]&&(rs2[i]==rt1[i+1]||rs2[i]==rt2[i+1]))) {
+ if(dops[i].rs1) alloc_reg(¤t,i,dops[i].rs1);
+ if(dops[i].rs2) alloc_reg(¤t,i,dops[i].rs2);
+ if((dops[i].rs1&&(dops[i].rs1==dops[i+1].rt1||dops[i].rs1==dops[i+1].rt2))||
+ (dops[i].rs2&&(dops[i].rs2==dops[i+1].rt1||dops[i].rs2==dops[i+1].rt2))) {
// The delay slot overwrites one of our conditions.
// Allocate the branch condition registers instead.
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
- if(rs1[i]) alloc_reg(¤t,i,rs1[i]);
- if(rs2[i]) alloc_reg(¤t,i,rs2[i]);
- if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
- {
- if(rs1[i]) alloc_reg64(¤t,i,rs1[i]);
- if(rs2[i]) alloc_reg64(¤t,i,rs2[i]);
- }
+ if(dops[i].rs1) alloc_reg(¤t,i,dops[i].rs1);
+ if(dops[i].rs2) alloc_reg(¤t,i,dops[i].rs2);
}
else
{
- ooo[i]=1;
+ dops[i].ooo=1;
delayslot_alloc(¤t,i+1);
}
}
else
- if((opcode[i]&0x3E)==6) // BLEZ/BGTZ
+ if((dops[i].opcode&0x3E)==6) // BLEZ/BGTZ
{
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
- alloc_reg(¤t,i,rs1[i]);
- if(!(current.is32>>rs1[i]&1))
- {
- alloc_reg64(¤t,i,rs1[i]);
- }
- if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) {
+ alloc_reg(¤t,i,dops[i].rs1);
+ if(dops[i].rs1&&(dops[i].rs1==dops[i+1].rt1||dops[i].rs1==dops[i+1].rt2)) {
// The delay slot overwrites one of our conditions.
// Allocate the branch condition registers instead.
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
- if(rs1[i]) alloc_reg(¤t,i,rs1[i]);
- if(!((current.is32>>rs1[i])&1))
- {
- if(rs1[i]) alloc_reg64(¤t,i,rs1[i]);
- }
+ if(dops[i].rs1) alloc_reg(¤t,i,dops[i].rs1);
}
else
{
- ooo[i]=1;
+ dops[i].ooo=1;
delayslot_alloc(¤t,i+1);
}
}
else
// Don't alloc the delay slot yet because we might not execute it
- if((opcode[i]&0x3E)==0x14) // BEQL/BNEL
+ if((dops[i].opcode&0x3E)==0x14) // BEQL/BNEL
{
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
- alloc_reg(¤t,i,rs1[i]);
- alloc_reg(¤t,i,rs2[i]);
- if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
- {
- alloc_reg64(¤t,i,rs1[i]);
- alloc_reg64(¤t,i,rs2[i]);
- }
+ alloc_reg(¤t,i,dops[i].rs1);
+ alloc_reg(¤t,i,dops[i].rs2);
}
else
- if((opcode[i]&0x3E)==0x16) // BLEZL/BGTZL
+ if((dops[i].opcode&0x3E)==0x16) // BLEZL/BGTZL
{
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
- alloc_reg(¤t,i,rs1[i]);
- if(!(current.is32>>rs1[i]&1))
- {
- alloc_reg64(¤t,i,rs1[i]);
- }
+ alloc_reg(¤t,i,dops[i].rs1);
}
ds=1;
//current.isconst=0;
//current.isconst=0;
//current.wasconst=0;
//regs[i].wasconst=0;
- clear_const(¤t,rs1[i]);
- clear_const(¤t,rt1[i]);
- //if((opcode2[i]&0x1E)==0x0) // BLTZ/BGEZ
- if((opcode2[i]&0x0E)==0x0) // BLTZ/BGEZ
+ clear_const(¤t,dops[i].rs1);
+ clear_const(¤t,dops[i].rt1);
+ //if((dops[i].opcode2&0x1E)==0x0) // BLTZ/BGEZ
+ if((dops[i].opcode2&0x0E)==0x0) // BLTZ/BGEZ
{
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
- alloc_reg(¤t,i,rs1[i]);
- if(!(current.is32>>rs1[i]&1))
- {
- alloc_reg64(¤t,i,rs1[i]);
- }
- if (rt1[i]==31) { // BLTZAL/BGEZAL
+ alloc_reg(¤t,i,dops[i].rs1);
+ if (dops[i].rt1==31) { // BLTZAL/BGEZAL
alloc_reg(¤t,i,31);
dirty_reg(¤t,31);
//#ifdef REG_PREFETCH
//alloc_reg(¤t,i,PTEMP);
//#endif
- //current.is32|=1LL<<rt1[i];
}
- if((rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) // The delay slot overwrites the branch condition.
- ||(rt1[i]==31&&(rs1[i+1]==31||rs2[i+1]==31||rt1[i+1]==31||rt2[i+1]==31))) { // DS touches $ra
+ if((dops[i].rs1&&(dops[i].rs1==dops[i+1].rt1||dops[i].rs1==dops[i+1].rt2)) // The delay slot overwrites the branch condition.
+ ||(dops[i].rt1==31&&(dops[i+1].rs1==31||dops[i+1].rs2==31||dops[i+1].rt1==31||dops[i+1].rt2==31))) { // DS touches $ra
// Allocate the branch condition registers instead.
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
- if(rs1[i]) alloc_reg(¤t,i,rs1[i]);
- if(!((current.is32>>rs1[i])&1))
- {
- if(rs1[i]) alloc_reg64(¤t,i,rs1[i]);
- }
+ if(dops[i].rs1) alloc_reg(¤t,i,dops[i].rs1);
}
else
{
- ooo[i]=1;
+ dops[i].ooo=1;
delayslot_alloc(¤t,i+1);
}
}
else
// Don't alloc the delay slot yet because we might not execute it
- if((opcode2[i]&0x1E)==0x2) // BLTZL/BGEZL
+ if((dops[i].opcode2&0x1E)==0x2) // BLTZL/BGEZL
{
current.isconst=0;
current.wasconst=0;
regs[i].wasconst=0;
alloc_cc(¤t,i);
dirty_reg(¤t,CCREG);
- alloc_reg(¤t,i,rs1[i]);
- if(!(current.is32>>rs1[i]&1))
- {
- alloc_reg64(¤t,i,rs1[i]);
- }
+ alloc_reg(¤t,i,dops[i].rs1);
}
ds=1;
//current.isconst=0;
break;
- case FJUMP:
- current.isconst=0;
- current.wasconst=0;
- regs[i].wasconst=0;
- if(likely[i]==0) // BC1F/BC1T
- {
- // TODO: Theoretically we can run out of registers here on x86.
- // The delay slot can allocate up to six, and we need to check
- // CSREG before executing the delay slot. Possibly we can drop
- // the cycle count and then reload it after checking that the
- // FPU is in a usable state, or don't do out-of-order execution.
- alloc_cc(¤t,i);
- dirty_reg(¤t,CCREG);
- alloc_reg(¤t,i,FSREG);
- alloc_reg(¤t,i,CSREG);
- if(itype[i+1]==FCOMP) {
- // The delay slot overwrites the branch condition.
- // Allocate the branch condition registers instead.
- alloc_cc(¤t,i);
- dirty_reg(¤t,CCREG);
- alloc_reg(¤t,i,CSREG);
- alloc_reg(¤t,i,FSREG);
- }
- else {
- ooo[i]=1;
- delayslot_alloc(¤t,i+1);
- alloc_reg(¤t,i+1,CSREG);
- }
- }
- else
- // Don't alloc the delay slot yet because we might not execute it
- if(likely[i]) // BC1FL/BC1TL
- {
- alloc_cc(¤t,i);
- dirty_reg(¤t,CCREG);
- alloc_reg(¤t,i,CSREG);
- alloc_reg(¤t,i,FSREG);
- }
- ds=1;
- current.isconst=0;
- break;
case IMM16:
imm16_alloc(¤t,i);
break;
cop0_alloc(¤t,i);
break;
case COP1:
+ break;
case COP2:
- cop1_alloc(¤t,i);
+ cop2_alloc(¤t,i);
break;
case C1LS:
c1ls_alloc(¤t,i);
case C2OP:
c2op_alloc(¤t,i);
break;
- case FCONV:
- fconv_alloc(¤t,i);
- break;
- case FLOAT:
- float_alloc(¤t,i);
- break;
- case FCOMP:
- fcomp_alloc(¤t,i);
- break;
case SYSCALL:
case HLECALL:
case INTCALL:
pagespan_alloc(¤t,i);
break;
}
-
- // Drop the upper half of registers that have become 32-bit
- current.uu|=current.is32&((1LL<<rt1[i])|(1LL<<rt2[i]));
- if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP&&itype[i]!=FJUMP) {
- current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
- if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
- current.uu|=1;
- } else {
- current.uu|=current.is32&((1LL<<rt1[i+1])|(1LL<<rt2[i+1]));
- current.uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
- if((~current.uu>>rt1[i+1])&1) current.uu&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
- current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
- current.uu|=1;
- }
// Create entry (branch target) regmap
for(hr=0;hr<HOST_REGS;hr++)
{
- int r,or,er;
+ int r,or;
r=current.regmap[hr];
if(r>=0) {
if(r!=regmap_pre[i][hr]) {
// TODO: delay slot (?)
or=get_reg(regmap_pre[i],r); // Get old mapping for this register
- if(or<0||(r&63)>=TEMPREG){
+ if(or<0||r>=TEMPREG){
regs[i].regmap_entry[hr]=-1;
}
else
// Just move it to a different register
regs[i].regmap_entry[hr]=r;
// If it was dirty before, it's still dirty
- if((regs[i].wasdirty>>or)&1) dirty_reg(¤t,r&63);
+ if((regs[i].wasdirty>>or)&1) dirty_reg(¤t,r);
}
}
else
regs[i].regmap_entry[hr]=0;
}
else
- if(r<64){
+ {
+ assert(r<64);
if((current.u>>r)&1) {
regs[i].regmap_entry[hr]=-1;
//regs[i].regmap[hr]=-1;
}else
regs[i].regmap_entry[hr]=r;
}
- else {
- if((current.uu>>(r&63))&1) {
- regs[i].regmap_entry[hr]=-1;
- //regs[i].regmap[hr]=-1;
- current.regmap[hr]=-1;
- }else
- regs[i].regmap_entry[hr]=r;
- }
}
} else {
// Branches expect CCREG to be allocated at the target
- if(regmap_pre[i][hr]==CCREG)
+ if(regmap_pre[i][hr]==CCREG)
regs[i].regmap_entry[hr]=CCREG;
else
regs[i].regmap_entry[hr]=-1;
}
memcpy(regs[i].regmap,current.regmap,sizeof(current.regmap));
}
+
+ if(i>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<<dops[i-1].rs1;
+ current.waswritten&=~(1<<dops[i].rt1);
+ current.waswritten&=~(1<<dops[i].rt2);
+ if((dops[i].itype==STORE||dops[i].itype==STORELR||(dops[i].itype==C2LS&&dops[i].opcode==0x3a))&&(u_int)imm[i]>=0x800)
+ current.waswritten&=~(1<<dops[i].rs1);
+
/* Branch post-alloc */
if(i>0)
{
- current.was32=current.is32;
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<<rs1[i-1])|(1LL<<rs2[i-1]));
- branch_regs[i-1].uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
+ branch_regs[i-1].u=branch_unneeded_reg[i-1]&~((1LL<<dops[i-1].rs1)|(1LL<<dops[i-1].rs2));
alloc_cc(&branch_regs[i-1],i-1);
dirty_reg(&branch_regs[i-1],CCREG);
- if(rt1[i-1]==31) { // JAL
+ if(dops[i-1].rt1==31) { // JAL
alloc_reg(&branch_regs[i-1],i-1,31);
dirty_reg(&branch_regs[i-1],31);
- branch_regs[i-1].is32|=1LL<<31;
}
memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
- memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+ memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
break;
case RJUMP:
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<<rs1[i-1])|(1LL<<rs2[i-1]));
- branch_regs[i-1].uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
+ branch_regs[i-1].u=branch_unneeded_reg[i-1]&~((1LL<<dops[i-1].rs1)|(1LL<<dops[i-1].rs2));
alloc_cc(&branch_regs[i-1],i-1);
dirty_reg(&branch_regs[i-1],CCREG);
- alloc_reg(&branch_regs[i-1],i-1,rs1[i-1]);
- if(rt1[i-1]!=0) { // JALR
- alloc_reg(&branch_regs[i-1],i-1,rt1[i-1]);
- dirty_reg(&branch_regs[i-1],rt1[i-1]);
- branch_regs[i-1].is32|=1LL<<rt1[i-1];
+ alloc_reg(&branch_regs[i-1],i-1,dops[i-1].rs1);
+ if(dops[i-1].rt1!=0) { // JALR
+ alloc_reg(&branch_regs[i-1],i-1,dops[i-1].rt1);
+ dirty_reg(&branch_regs[i-1],dops[i-1].rt1);
}
#ifdef USE_MINI_HT
- if(rs1[i-1]==31) { // JALR
+ if(dops[i-1].rs1==31) { // JALR
alloc_reg(&branch_regs[i-1],i-1,RHASH);
- #ifndef HOST_IMM_ADDR32
alloc_reg(&branch_regs[i-1],i-1,RHTBL);
- #endif
}
#endif
memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
- memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+ memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
break;
case CJUMP:
- if((opcode[i-1]&0x3E)==4) // BEQ/BNE
+ if((dops[i-1].opcode&0x3E)==4) // BEQ/BNE
{
alloc_cc(¤t,i-1);
dirty_reg(¤t,CCREG);
- if((rs1[i-1]&&(rs1[i-1]==rt1[i]||rs1[i-1]==rt2[i]))||
- (rs2[i-1]&&(rs2[i-1]==rt1[i]||rs2[i-1]==rt2[i]))) {
+ if((dops[i-1].rs1&&(dops[i-1].rs1==dops[i].rt1||dops[i-1].rs1==dops[i].rt2))||
+ (dops[i-1].rs2&&(dops[i-1].rs2==dops[i].rt1||dops[i-1].rs2==dops[i].rt2))) {
// The delay slot overwrote one of our conditions
// Delay slot goes after the test (in order)
- current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
- current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
- if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+ current.u=branch_unneeded_reg[i-1]&~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
current.u|=1;
- current.uu|=1;
delayslot_alloc(¤t,i);
current.isconst=0;
}
else
{
- current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i-1])|(1LL<<rs2[i-1]));
- current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
+ current.u=branch_unneeded_reg[i-1]&~((1LL<<dops[i-1].rs1)|(1LL<<dops[i-1].rs2));
// Alloc the branch condition registers
- if(rs1[i-1]) alloc_reg(¤t,i-1,rs1[i-1]);
- if(rs2[i-1]) alloc_reg(¤t,i-1,rs2[i-1]);
- if(!((current.is32>>rs1[i-1])&(current.is32>>rs2[i-1])&1))
- {
- if(rs1[i-1]) alloc_reg64(¤t,i-1,rs1[i-1]);
- if(rs2[i-1]) alloc_reg64(¤t,i-1,rs2[i-1]);
- }
+ if(dops[i-1].rs1) alloc_reg(¤t,i-1,dops[i-1].rs1);
+ if(dops[i-1].rs2) alloc_reg(¤t,i-1,dops[i-1].rs2);
}
memcpy(&branch_regs[i-1],¤t,sizeof(current));
branch_regs[i-1].isconst=0;
branch_regs[i-1].wasconst=0;
memcpy(&branch_regs[i-1].regmap_entry,¤t.regmap,sizeof(current.regmap));
- memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+ memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
}
else
- if((opcode[i-1]&0x3E)==6) // BLEZ/BGTZ
+ if((dops[i-1].opcode&0x3E)==6) // BLEZ/BGTZ
{
alloc_cc(¤t,i-1);
dirty_reg(¤t,CCREG);
- if(rs1[i-1]==rt1[i]||rs1[i-1]==rt2[i]) {
+ if(dops[i-1].rs1==dops[i].rt1||dops[i-1].rs1==dops[i].rt2) {
// The delay slot overwrote the branch condition
// Delay slot goes after the test (in order)
- current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
- current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
- if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+ current.u=branch_unneeded_reg[i-1]&~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
current.u|=1;
- current.uu|=1;
delayslot_alloc(¤t,i);
current.isconst=0;
}
else
{
- current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
- current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
+ current.u=branch_unneeded_reg[i-1]&~(1LL<<dops[i-1].rs1);
// Alloc the branch condition register
- alloc_reg(¤t,i-1,rs1[i-1]);
- if(!(current.is32>>rs1[i-1]&1))
- {
- alloc_reg64(¤t,i-1,rs1[i-1]);
- }
+ alloc_reg(¤t,i-1,dops[i-1].rs1);
}
memcpy(&branch_regs[i-1],¤t,sizeof(current));
branch_regs[i-1].isconst=0;
branch_regs[i-1].wasconst=0;
memcpy(&branch_regs[i-1].regmap_entry,¤t.regmap,sizeof(current.regmap));
- memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+ memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
}
else
// Alloc the delay slot in case the branch is taken
- if((opcode[i-1]&0x3E)==0x14) // BEQL/BNEL
+ if((dops[i-1].opcode&0x3E)==0x14) // BEQL/BNEL
{
memcpy(&branch_regs[i-1],¤t,sizeof(current));
- branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
- branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
- if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
+ branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2)|(1LL<<dops[i].rt1)|(1LL<<dops[i].rt2)))|1;
alloc_cc(&branch_regs[i-1],i);
dirty_reg(&branch_regs[i-1],CCREG);
delayslot_alloc(&branch_regs[i-1],i);
memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
}
else
- if((opcode[i-1]&0x3E)==0x16) // BLEZL/BGTZL
+ if((dops[i-1].opcode&0x3E)==0x16) // BLEZL/BGTZL
{
memcpy(&branch_regs[i-1],¤t,sizeof(current));
- branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
- branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
- if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
+ branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2)|(1LL<<dops[i].rt1)|(1LL<<dops[i].rt2)))|1;
alloc_cc(&branch_regs[i-1],i);
dirty_reg(&branch_regs[i-1],CCREG);
delayslot_alloc(&branch_regs[i-1],i);
}
break;
case SJUMP:
- //if((opcode2[i-1]&0x1E)==0) // BLTZ/BGEZ
- if((opcode2[i-1]&0x0E)==0) // BLTZ/BGEZ
+ //if((dops[i-1].opcode2&0x1E)==0) // BLTZ/BGEZ
+ if((dops[i-1].opcode2&0x0E)==0) // BLTZ/BGEZ
{
alloc_cc(¤t,i-1);
dirty_reg(¤t,CCREG);
- if(rs1[i-1]==rt1[i]||rs1[i-1]==rt2[i]) {
+ if(dops[i-1].rs1==dops[i].rt1||dops[i-1].rs1==dops[i].rt2) {
// The delay slot overwrote the branch condition
// Delay slot goes after the test (in order)
- current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
- current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
- if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+ current.u=branch_unneeded_reg[i-1]&~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
current.u|=1;
- current.uu|=1;
delayslot_alloc(¤t,i);
current.isconst=0;
}
else
{
- current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
- current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
+ current.u=branch_unneeded_reg[i-1]&~(1LL<<dops[i-1].rs1);
// Alloc the branch condition register
- alloc_reg(¤t,i-1,rs1[i-1]);
- if(!(current.is32>>rs1[i-1]&1))
- {
- alloc_reg64(¤t,i-1,rs1[i-1]);
- }
+ alloc_reg(¤t,i-1,dops[i-1].rs1);
}
memcpy(&branch_regs[i-1],¤t,sizeof(current));
branch_regs[i-1].isconst=0;
branch_regs[i-1].wasconst=0;
memcpy(&branch_regs[i-1].regmap_entry,¤t.regmap,sizeof(current.regmap));
- memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+ memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
}
else
// Alloc the delay slot in case the branch is taken
- if((opcode2[i-1]&0x1E)==2) // BLTZL/BGEZL
+ if((dops[i-1].opcode2&0x1E)==2) // BLTZL/BGEZL
{
memcpy(&branch_regs[i-1],¤t,sizeof(current));
- branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
- branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
- if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
+ branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2)|(1LL<<dops[i].rt1)|(1LL<<dops[i].rt2)))|1;
alloc_cc(&branch_regs[i-1],i);
dirty_reg(&branch_regs[i-1],CCREG);
delayslot_alloc(&branch_regs[i-1],i);
memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
}
// FIXME: BLTZAL/BGEZAL
- if(opcode2[i-1]&0x10) { // BxxZAL
+ if(dops[i-1].opcode2&0x10) { // BxxZAL
alloc_reg(&branch_regs[i-1],i-1,31);
dirty_reg(&branch_regs[i-1],31);
- branch_regs[i-1].is32|=1LL<<31;
- }
- break;
- case FJUMP:
- if(likely[i-1]==0) // BC1F/BC1T
- {
- alloc_cc(¤t,i-1);
- dirty_reg(¤t,CCREG);
- if(itype[i]==FCOMP) {
- // The delay slot overwrote the branch condition
- // Delay slot goes after the test (in order)
- delayslot_alloc(¤t,i);
- current.isconst=0;
- }
- else
- {
- current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
- current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
- // Alloc the branch condition register
- alloc_reg(¤t,i-1,FSREG);
- }
- memcpy(&branch_regs[i-1],¤t,sizeof(current));
- memcpy(&branch_regs[i-1].regmap_entry,¤t.regmap,sizeof(current.regmap));
- }
- else // BC1FL/BC1TL
- {
- // Alloc the delay slot in case the branch is taken
- memcpy(&branch_regs[i-1],¤t,sizeof(current));
- branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
- branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
- if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
- alloc_cc(&branch_regs[i-1],i);
- dirty_reg(&branch_regs[i-1],CCREG);
- delayslot_alloc(&branch_regs[i-1],i);
- branch_regs[i-1].isconst=0;
- alloc_reg(¤t,i,CCREG); // Not taken path
- dirty_reg(¤t,CCREG);
- memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
}
break;
}
- if(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)
+ if (dops[i-1].is_ujump)
{
- if(rt1[i-1]==31) // JAL/JALR
+ if(dops[i-1].rt1==31) // JAL/JALR
{
// Subroutine call will return here, don't alloc any registers
- current.is32=1;
current.dirty=0;
clear_all_regs(current.regmap);
alloc_reg(¤t,i,CCREG);
else if(i+1<slen)
{
// Internal branch will jump here, match registers to caller
- current.is32=0x3FFFFFFFFLL;
current.dirty=0;
clear_all_regs(current.regmap);
alloc_reg(¤t,i,CCREG);
{
if(ba[j]==start+i*4+4) {
memcpy(current.regmap,branch_regs[j].regmap,sizeof(current.regmap));
- current.is32=branch_regs[j].is32;
current.dirty=branch_regs[j].dirty;
break;
}
if(current.regmap[hr]!=branch_regs[j].regmap[hr]) {
current.regmap[hr]=-1;
}
- current.is32&=branch_regs[j].is32;
current.dirty&=branch_regs[j].dirty;
}
}
}
// Count cycles in between branches
- ccadj[i]=cc;
- if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP||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;
}
-#ifdef PCSX
- else if(/*itype[i]==LOAD||*/itype[i]==STORE||itype[i]==C1LS) // load causes weird timing issues
+#if !defined(DRC_DBG)
+ else if(dops[i].itype==C2OP&>e_cycletab[source[i]&0x3f]>2)
{
- cc+=2; // 2 cycle penalty (after CLOCK_DIVIDER)
+ // 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(itype[i]==C2LS)
+ 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(dops[i].itype==C2LS)
+ {
+ // same as with C2OP
+ cc += HACK_ENABLED(NDHACK_NO_STALLS) ? 4 : 2;
+ }
#endif
else
{
cc++;
}
- flush_dirty_uppers(¤t);
- if(!is_ds[i]) {
- regs[i].is32=current.is32;
+ if(!dops[i].is_ds) {
regs[i].dirty=current.dirty;
regs[i].isconst=current.isconst;
- memcpy(constmap[i],current.constmap,sizeof(current.constmap));
+ memcpy(constmap[i],current_constmap,sizeof(constmap[i]));
}
for(hr=0;hr<HOST_REGS;hr++) {
if(hr!=EXCLUDE_REG&®s[i].regmap[hr]>=0) {
}
}
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;
-
+}
+
+static noinline void pass4_cull_unused_regs(void)
+{
+ u_int last_needed_regs[4] = {0,0,0,0};
+ u_int nr=0;
+ int i;
+
for (i=slen-1;i>=0;i--)
{
int hr;
- if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+ __builtin_prefetch(regs[i-2].regmap);
+ if(dops[i].is_jump)
{
if(ba[i]<start || ba[i]>=(start+slen*4))
{
}
}
// Conditional branch may need registers for following instructions
- if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+ if (!dops[i].is_ujump)
{
if(i<slen-2) {
- nr|=needed_reg[i+2];
+ nr |= last_needed_regs[(i+2) & 3];
for(hr=0;hr<HOST_REGS;hr++)
{
if(regmap_pre[i+2][hr]>=0&&get_reg(regs[i+2].regmap_entry,regmap_pre[i+2][hr])<0) nr&=~(1<<hr);
//if(regs[i].regmap[hr]!=regmap_pre[i][hr]) nr&=~(1<<hr);
//if(regs[i].regmap[hr]<0) nr&=~(1<<hr);
// Merge in delay slot
- for(hr=0;hr<HOST_REGS;hr++)
- {
- if(!likely[i]) {
- // These are overwritten unless the branch is "likely"
- // and the delay slot is nullified if not taken
- if(rt1[i+1]&&rt1[i+1]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
- if(rt2[i+1]&&rt2[i+1]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
- }
- if(us1[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(us2[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(rs1[i+1]==regmap_pre[i][hr]) nr|=1<<hr;
- if(rs2[i+1]==regmap_pre[i][hr]) nr|=1<<hr;
- if(us1[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- if(us2[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- if(rs1[i+1]==regs[i].regmap_entry[hr]) nr|=1<<hr;
- if(rs2[i+1]==regs[i].regmap_entry[hr]) nr|=1<<hr;
- if(dep1[i+1]&&!((unneeded_reg_upper[i]>>dep1[i+1])&1)) {
- if(dep1[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(dep2[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- }
- if(dep2[i+1]&&!((unneeded_reg_upper[i]>>dep2[i+1])&1)) {
- if(dep1[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- if(dep2[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- }
- if(itype[i+1]==STORE || itype[i+1]==STORELR || (opcode[i+1]&0x3b)==0x39 || (opcode[i+1]&0x3b)==0x3a) {
- if(regmap_pre[i][hr]==INVCP) nr|=1<<hr;
- if(regs[i].regmap_entry[hr]==INVCP) nr|=1<<hr;
- }
- }
- }
- else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
+ if (dops[i+1].rt1) nr &= ~get_regm(regs[i].regmap, dops[i+1].rt1);
+ if (dops[i+1].rt2) nr &= ~get_regm(regs[i].regmap, dops[i+1].rt2);
+ nr |= get_regm(regmap_pre[i], dops[i+1].rs1);
+ nr |= get_regm(regmap_pre[i], dops[i+1].rs2);
+ nr |= get_regm(regs[i].regmap_entry, dops[i+1].rs1);
+ nr |= get_regm(regs[i].regmap_entry, dops[i+1].rs2);
+ if (ram_offset && (dops[i+1].is_load || dops[i+1].is_store)) {
+ nr |= get_regm(regmap_pre[i], ROREG);
+ nr |= get_regm(regs[i].regmap_entry, ROREG);
+ }
+ if (dops[i+1].is_store) {
+ nr |= get_regm(regmap_pre[i], INVCP);
+ nr |= get_regm(regs[i].regmap_entry, INVCP);
+ }
+ }
+ else if(dops[i].itype==SYSCALL||dops[i].itype==HLECALL||dops[i].itype==INTCALL)
{
// SYSCALL instruction (software interrupt)
nr=0;
}
- else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
+ else if(dops[i].itype==COP0 && (source[i]&0x3f)==0x18)
{
// ERET instruction (return from interrupt)
nr=0;
}
}
}
+ // Overwritten registers are not needed
+ if (dops[i].rt1) nr &= ~get_regm(regs[i].regmap, dops[i].rt1);
+ if (dops[i].rt2) nr &= ~get_regm(regs[i].regmap, dops[i].rt2);
+ nr &= ~get_regm(regs[i].regmap, FTEMP);
+ // Source registers are needed
+ nr |= get_regm(regmap_pre[i], dops[i].rs1);
+ nr |= get_regm(regmap_pre[i], dops[i].rs2);
+ nr |= get_regm(regs[i].regmap_entry, dops[i].rs1);
+ nr |= get_regm(regs[i].regmap_entry, dops[i].rs2);
+ if (ram_offset && (dops[i].is_load || dops[i].is_store)) {
+ nr |= get_regm(regmap_pre[i], ROREG);
+ nr |= get_regm(regs[i].regmap_entry, ROREG);
+ }
+ if (dops[i].is_store) {
+ nr |= get_regm(regmap_pre[i], INVCP);
+ nr |= get_regm(regs[i].regmap_entry, INVCP);
+ }
+
+ if (i > 0 && !dops[i].bt && regs[i].wasdirty)
for(hr=0;hr<HOST_REGS;hr++)
{
- // Overwritten registers are not needed
- if(rt1[i]&&rt1[i]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
- if(rt2[i]&&rt2[i]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
- if(FTEMP==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
- // Source registers are needed
- if(us1[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(us2[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(rs1[i]==regmap_pre[i][hr]) nr|=1<<hr;
- if(rs2[i]==regmap_pre[i][hr]) nr|=1<<hr;
- if(us1[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- if(us2[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- if(rs1[i]==regs[i].regmap_entry[hr]) nr|=1<<hr;
- if(rs2[i]==regs[i].regmap_entry[hr]) nr|=1<<hr;
- if(dep1[i]&&!((unneeded_reg_upper[i]>>dep1[i])&1)) {
- if(dep1[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(dep1[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- }
- if(dep2[i]&&!((unneeded_reg_upper[i]>>dep2[i])&1)) {
- if(dep2[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(dep2[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- }
- if(itype[i]==STORE || itype[i]==STORELR || (opcode[i]&0x3b)==0x39 || (opcode[i]&0x3b)==0x3a) {
- if(regmap_pre[i][hr]==INVCP) nr|=1<<hr;
- if(regs[i].regmap_entry[hr]==INVCP) nr|=1<<hr;
- }
// Don't store a register immediately after writing it,
// may prevent dual-issue.
// But do so if this is a branch target, otherwise we
// might have to load the register before the branch.
- if(i>0&&!bt[i]&&((regs[i].wasdirty>>hr)&1)) {
- if((regmap_pre[i][hr]>0&®map_pre[i][hr]<64&&!((unneeded_reg[i]>>regmap_pre[i][hr])&1)) ||
- (regmap_pre[i][hr]>64&&!((unneeded_reg_upper[i]>>(regmap_pre[i][hr]&63))&1)) ) {
- if(rt1[i-1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
- if(rt2[i-1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+ if((regs[i].wasdirty>>hr)&1) {
+ if((regmap_pre[i][hr]>0&&!((unneeded_reg[i]>>regmap_pre[i][hr])&1))) {
+ if(dops[i-1].rt1==regmap_pre[i][hr]) nr|=1<<hr;
+ if(dops[i-1].rt2==regmap_pre[i][hr]) nr|=1<<hr;
}
- if((regs[i].regmap_entry[hr]>0&®s[i].regmap_entry[hr]<64&&!((unneeded_reg[i]>>regs[i].regmap_entry[hr])&1)) ||
- (regs[i].regmap_entry[hr]>64&&!((unneeded_reg_upper[i]>>(regs[i].regmap_entry[hr]&63))&1)) ) {
- if(rt1[i-1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
- if(rt2[i-1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+ if((regs[i].regmap_entry[hr]>0&&!((unneeded_reg[i]>>regs[i].regmap_entry[hr])&1))) {
+ if(dops[i-1].rt1==regs[i].regmap_entry[hr]) nr|=1<<hr;
+ if(dops[i-1].rt2==regs[i].regmap_entry[hr]) nr|=1<<hr;
}
}
}
// Cycle count is needed at branches. Assume it is needed at the target too.
- if(i==0||bt[i]||itype[i]==CJUMP||itype[i]==FJUMP||itype[i]==SPAN) {
+ if(i==0||dops[i].bt||dops[i].itype==CJUMP||dops[i].itype==SPAN) {
if(regmap_pre[i][HOST_CCREG]==CCREG) nr|=1<<HOST_CCREG;
if(regs[i].regmap_entry[HOST_CCREG]==CCREG) nr|=1<<HOST_CCREG;
}
// Save it
- needed_reg[i]=nr;
-
+ last_needed_regs[i & 3] = nr;
+
// Deallocate unneeded registers
for(hr=0;hr<HOST_REGS;hr++)
{
if(!((nr>>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(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
- {
- if(likely[i]) {
- regs[i].regmap[hr]=-1;
- regs[i].isconst&=~(1<<hr);
- if(i<slen-2) {
- regmap_pre[i+2][hr]=-1;
- regs[i+2].wasconst&=~(1<<hr);
- }
- }
- }
- }
- if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+ if(dops[i].is_jump)
{
- int d1=0,d2=0,map=0,temp=0;
- if(get_reg(regs[i].regmap,rt1[i+1]|64)>=0||get_reg(branch_regs[i].regmap,rt1[i+1]|64)>=0)
- {
- d1=dep1[i+1];
- d2=dep2[i+1];
- }
- if(using_tlb) {
- 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)
- map=TLREG;
- } else
- if(itype[i+1]==STORE || itype[i+1]==STORELR ||
- (opcode[i+1]&0x3b)==0x39 || (opcode[i+1]&0x3b)==0x3a) { // SWC1/SDC1 || SWC2/SDC2
- map=INVCP;
- }
- if(itype[i+1]==LOADLR || itype[i+1]==STORELR ||
- itype[i+1]==C1LS || itype[i+1]==C2LS)
- temp=FTEMP;
- 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)!=rt1[i+1] && (regs[i].regmap[hr]&63)!=rt2[i+1] &&
- (regs[i].regmap[hr]^64)!=us1[i+1] && (regs[i].regmap[hr]^64)!=us2[i+1] &&
- (regs[i].regmap[hr]^64)!=d1 && (regs[i].regmap[hr]^64)!=d2 &&
- regs[i].regmap[hr]!=rs1[i+1] && regs[i].regmap[hr]!=rs2[i+1] &&
- (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=PTEMP &&
+ int map1 = 0, map2 = 0, temp = 0; // or -1 ??
+ if (dops[i+1].is_load || dops[i+1].is_store)
+ map1 = ROREG;
+ if (dops[i+1].is_store)
+ map2 = INVCP;
+ if(dops[i+1].itype==LOADLR || dops[i+1].itype==STORELR || dops[i+1].itype==C2LS)
+ temp = FTEMP;
+ if(regs[i].regmap[hr]!=dops[i].rs1 && regs[i].regmap[hr]!=dops[i].rs2 &&
+ regs[i].regmap[hr]!=dops[i].rt1 && regs[i].regmap[hr]!=dops[i].rt2 &&
+ regs[i].regmap[hr]!=dops[i+1].rt1 && regs[i].regmap[hr]!=dops[i+1].rt2 &&
+ regs[i].regmap[hr]!=dops[i+1].rs1 && regs[i].regmap[hr]!=dops[i+1].rs2 &&
+ regs[i].regmap[hr]!=temp && regs[i].regmap[hr]!=PTEMP &&
regs[i].regmap[hr]!=RHASH && regs[i].regmap[hr]!=RHTBL &&
regs[i].regmap[hr]!=RTEMP && regs[i].regmap[hr]!=CCREG &&
- regs[i].regmap[hr]!=map )
+ regs[i].regmap[hr]!=map1 && regs[i].regmap[hr]!=map2)
{
regs[i].regmap[hr]=-1;
regs[i].isconst&=~(1<<hr);
- if((branch_regs[i].regmap[hr]&63)!=rs1[i] && (branch_regs[i].regmap[hr]&63)!=rs2[i] &&
- (branch_regs[i].regmap[hr]&63)!=rt1[i] && (branch_regs[i].regmap[hr]&63)!=rt2[i] &&
- (branch_regs[i].regmap[hr]&63)!=rt1[i+1] && (branch_regs[i].regmap[hr]&63)!=rt2[i+1] &&
- (branch_regs[i].regmap[hr]^64)!=us1[i+1] && (branch_regs[i].regmap[hr]^64)!=us2[i+1] &&
- (branch_regs[i].regmap[hr]^64)!=d1 && (branch_regs[i].regmap[hr]^64)!=d2 &&
- branch_regs[i].regmap[hr]!=rs1[i+1] && branch_regs[i].regmap[hr]!=rs2[i+1] &&
- (branch_regs[i].regmap[hr]&63)!=temp && branch_regs[i].regmap[hr]!=PTEMP &&
+ regs[i].dirty&=~(1<<hr);
+ regs[i+1].wasdirty&=~(1<<hr);
+ if(branch_regs[i].regmap[hr]!=dops[i].rs1 && branch_regs[i].regmap[hr]!=dops[i].rs2 &&
+ branch_regs[i].regmap[hr]!=dops[i].rt1 && branch_regs[i].regmap[hr]!=dops[i].rt2 &&
+ branch_regs[i].regmap[hr]!=dops[i+1].rt1 && branch_regs[i].regmap[hr]!=dops[i+1].rt2 &&
+ branch_regs[i].regmap[hr]!=dops[i+1].rs1 && branch_regs[i].regmap[hr]!=dops[i+1].rs2 &&
+ branch_regs[i].regmap[hr]!=temp && branch_regs[i].regmap[hr]!=PTEMP &&
branch_regs[i].regmap[hr]!=RHASH && branch_regs[i].regmap[hr]!=RHTBL &&
branch_regs[i].regmap[hr]!=RTEMP && branch_regs[i].regmap[hr]!=CCREG &&
- branch_regs[i].regmap[hr]!=map)
+ branch_regs[i].regmap[hr]!=map1 && branch_regs[i].regmap[hr]!=map2)
{
branch_regs[i].regmap[hr]=-1;
branch_regs[i].regmap_entry[hr]=-1;
- if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+ if (!dops[i].is_ujump)
{
- if(!likely[i]&&i<slen-2) {
+ if (i < slen-2) {
regmap_pre[i+2][hr]=-1;
regs[i+2].wasconst&=~(1<<hr);
}
// Non-branch
if(i>0)
{
- int d1=0,d2=0,map=-1,temp=-1;
- if(get_reg(regs[i].regmap,rt1[i]|64)>=0)
- {
- d1=dep1[i];
- d2=dep2[i];
- }
- if(using_tlb) {
- if(itype[i]==LOAD || itype[i]==LOADLR ||
- itype[i]==STORE || itype[i]==STORELR ||
- itype[i]==C1LS || itype[i]==C2LS)
- map=TLREG;
- } else 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]^64)!=us1[i] && (regs[i].regmap[hr]^64)!=us2[i] &&
- (regs[i].regmap[hr]^64)!=d1 && (regs[i].regmap[hr]^64)!=d2 &&
- 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]!=dops[i].rt1 && regs[i].regmap[hr]!=dops[i].rt2 &&
+ regs[i].regmap[hr]!=dops[i].rs1 && regs[i].regmap[hr]!=dops[i].rs2 &&
+ regs[i].regmap[hr]!=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(i<slen-1&&!is_ds[i]) {
- if(regmap_pre[i+1][hr]!=-1 || regs[i].regmap[hr]!=-1)
+ if(i<slen-1&&!dops[i].is_ds) {
+ assert(regs[i].regmap[hr]<64);
+ if(regmap_pre[i+1][hr]!=-1 || regs[i].regmap[hr]>0)
if(regmap_pre[i+1][hr]!=regs[i].regmap[hr])
- if(regs[i].regmap[hr]<64||!((regs[i].was32>>(regs[i].regmap[hr]&63))&1))
{
- printf("fail: %x (%d %d!=%d)\n",start+i*4,hr,regmap_pre[i+1][hr],regs[i].regmap[hr]);
+ SysPrintf("fail: %x (%d %d!=%d)\n",start+i*4,hr,regmap_pre[i+1][hr],regs[i].regmap[hr]);
assert(regmap_pre[i+1][hr]==regs[i].regmap[hr]);
}
regmap_pre[i+1][hr]=-1;
}
regs[i].regmap[hr]=-1;
regs[i].isconst&=~(1<<hr);
+ regs[i].dirty&=~(1<<hr);
+ regs[i+1].wasdirty&=~(1<<hr);
}
}
}
- }
- }
+ } // if needed
+ } // for hr
}
-
- /* Pass 5 - Pre-allocate registers */
-
- // If a register is allocated during a loop, try to allocate it for the
- // entire loop, if possible. This avoids loading/storing registers
- // inside of the loop.
-
+}
+
+// If a register is allocated during a loop, try to allocate it for the
+// entire loop, if possible. This avoids loading/storing registers
+// inside of the loop.
+static noinline void pass5a_preallocate1(void)
+{
+ int i, j, hr;
signed char f_regmap[HOST_REGS];
clear_all_regs(f_regmap);
for(i=0;i<slen-1;i++)
{
- if(itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+ if(dops[i].itype==UJUMP||dops[i].itype==CJUMP||dops[i].itype==SJUMP)
{
- 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
- ||itype[i+1]==SHIFT||itype[i+1]==COP1||itype[i+1]==FLOAT
- ||itype[i+1]==FCOMP||itype[i+1]==FCONV
- ||itype[i+1]==COP2||itype[i+1]==C2LS||itype[i+1]==C2OP)
+ if(ba[i]>=start && ba[i]<(start+i*4))
+ 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&&itype[t-1]!=FJUMP)) // 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<HOST_REGS;hr++)
{
- if(regs[i].regmap[hr]>64) {
- if(!((regs[i].dirty>>hr)&1))
- f_regmap[hr]=regs[i].regmap[hr];
- else f_regmap[hr]=-1;
- }
- else if(regs[i].regmap[hr]>=0) {
+ if(regs[i].regmap[hr]>=0) {
if(f_regmap[hr]!=regs[i].regmap[hr]) {
// dealloc old register
int n;
f_regmap[hr]=regs[i].regmap[hr];
}
}
- if(branch_regs[i].regmap[hr]>64) {
- if(!((branch_regs[i].dirty>>hr)&1))
- f_regmap[hr]=branch_regs[i].regmap[hr];
- else f_regmap[hr]=-1;
- }
- else if(branch_regs[i].regmap[hr]>=0) {
+ if(branch_regs[i].regmap[hr]>=0) {
if(f_regmap[hr]!=branch_regs[i].regmap[hr]) {
// dealloc old register
int n;
f_regmap[hr]=branch_regs[i].regmap[hr];
}
}
- if(ooo[i]) {
- if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i+1])
+ 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{
- 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
{
//printf("Test %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
if(r<34&&((unneeded_reg[j]>>r)&1)) break;
- if(r>63&&((unneeded_reg_upper[j]>>(r&63))&1)) break;
- if(r>63) {
- // NB This can exclude the case where the upper-half
- // register is lower numbered than the lower-half
- // register. Not sure if it's worth fixing...
- if(get_reg(regs[j].regmap,r&63)<0) break;
- if(get_reg(regs[j].regmap_entry,r&63)<0) break;
- if(regs[j].is32&(1LL<<(r&63))) break;
- }
- if(regs[j].regmap[hr]==f_regmap[hr]&&(f_regmap[hr]&63)<TEMPREG) {
+ assert(r < 64);
+ if(regs[j].regmap[hr]==f_regmap[hr]&&f_regmap[hr]<TEMPREG) {
//printf("Hit %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
int k;
if(regs[i].regmap[hr]==-1&&branch_regs[i].regmap[hr]==-1) {
+ if(get_reg(regs[i].regmap,f_regmap[hr])>=0) break;
if(get_reg(regs[i+2].regmap,f_regmap[hr])>=0) break;
- if(r>63) {
- if(get_reg(regs[i].regmap,r&63)<0) break;
- if(get_reg(branch_regs[i].regmap,r&63)<0) break;
- }
k=i;
while(k>1&®s[k-1].regmap[hr]==-1) {
if(count_free_regs(regs[k-1].regmap)<=minimum_free_regs[k-1]) {
//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||itype[k-2]==FJUMP) {
+ 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;
}
- if(r>63) {
- // NB This can exclude the case where the upper-half
- // register is lower numbered than the lower-half
- // register. Not sure if it's worth fixing...
- if(get_reg(regs[k-1].regmap,r&63)<0) break;
- if(regs[k-1].is32&(1LL<<(r&63))) break;
- }
k--;
}
- if(i<slen-1) {
- if((regs[k].is32&(1LL<<f_regmap[hr]))!=
- (regs[i+2].was32&(1LL<<f_regmap[hr]))) {
- //printf("bad match after branch\n");
- break;
- }
- }
if(regs[k-1].regmap[hr]==f_regmap[hr]&®map_pre[k][hr]==f_regmap[hr]) {
//printf("Extend r%d, %x ->\n",hr,start+k*4);
while(k<i) {
branch_regs[i].dirty|=(1<<hr)®s[i].dirty;
branch_regs[i].wasconst&=~(1<<hr);
branch_regs[i].isconst&=~(1<<hr);
- if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000) {
+ if (!dops[i].is_ujump) {
regmap_pre[i+2][hr]=f_regmap[hr];
regs[i+2].wasdirty&=~(1<<hr);
regs[i+2].wasdirty|=(1<<hr)®s[i].dirty;
- assert((branch_regs[i].is32&(1LL<<f_regmap[hr]))==
- (regs[i+2].was32&(1LL<<f_regmap[hr])));
}
}
}
regs[k].dirty&=~(1<<hr);
regs[k].wasconst&=~(1<<hr);
regs[k].isconst&=~(1<<hr);
- if(itype[k]==UJUMP||itype[k]==RJUMP||itype[k]==CJUMP||itype[k]==SJUMP||itype[k]==FJUMP) {
+ if (dops[k].is_jump) {
branch_regs[k].regmap_entry[hr]=f_regmap[hr];
branch_regs[k].regmap[hr]=f_regmap[hr];
branch_regs[k].dirty&=~(1<<hr);
branch_regs[k].wasconst&=~(1<<hr);
branch_regs[k].isconst&=~(1<<hr);
- if(itype[k]!=RJUMP&&itype[k]!=UJUMP&&(source[k]>>16)!=0x1000) {
+ if (!dops[k].is_ujump) {
regmap_pre[k+2][hr]=f_regmap[hr];
regs[k+2].wasdirty&=~(1<<hr);
- assert((branch_regs[k].is32&(1LL<<f_regmap[hr]))==
- (regs[k+2].was32&(1LL<<f_regmap[hr])));
}
}
else
//printf("no-match due to different register\n");
break;
}
- if((regs[j+1].is32&(1LL<<f_regmap[hr]))!=(regs[j].is32&(1LL<<f_regmap[hr]))) {
- //printf("32/64 mismatch %x %d\n",start+j*4,hr);
- break;
- }
- if(itype[j]==UJUMP||itype[j]==RJUMP||(source[j]>>16)==0x1000)
+ if (dops[j].is_ujump)
{
// Stop on unconditional branch
break;
}
- if(itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP)
+ if(dops[j].itype==CJUMP||dops[j].itype==SJUMP)
{
- if(ooo[j]) {
- if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j+1])
+ if(dops[j].ooo) {
+ 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) {
//printf("No free regs for store %x\n",start+j*4);
break;
}
- if(f_regmap[hr]>=64) {
- if(regs[j].is32&(1LL<<(f_regmap[hr]&63))) {
- break;
- }
- else
- {
- if(get_reg(regs[j].regmap,f_regmap[hr]&63)<0) {
- break;
- }
- }
- }
+ assert(f_regmap[hr]<64);
}
}
}
for(hr=0;hr<HOST_REGS;hr++)
{
if(hr!=EXCLUDE_REG) {
- if(regs[i].regmap[hr]>64) {
- if(!((regs[i].dirty>>hr)&1))
- f_regmap[hr]=regs[i].regmap[hr];
- }
- else if(regs[i].regmap[hr]>=0) {
+ if(regs[i].regmap[hr]>=0) {
if(f_regmap[hr]!=regs[i].regmap[hr]) {
// dealloc old register
int n;
}
}
// Try to restore cycle count at branch targets
- if(bt[i]) {
+ if(dops[i].bt) {
for(j=i;j<slen-1;j++) {
if(regs[j].regmap[HOST_CCREG]!=-1) break;
if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j]) {
regmap_pre[k+1][HOST_CCREG]=CCREG;
regs[k+1].wasdirty|=1<<HOST_CCREG;
regs[k].dirty|=1<<HOST_CCREG;
- regs[k].wasconst&=~(1<<HOST_CCREG);
- regs[k].isconst&=~(1<<HOST_CCREG);
- k++;
- }
- regs[j].regmap_entry[HOST_CCREG]=CCREG;
- }
- // Work backwards from the branch target
- if(j>i&&f_regmap[HOST_CCREG]==CCREG)
- {
- //printf("Extend backwards\n");
- int k;
- k=i;
- while(regs[k-1].regmap[HOST_CCREG]==-1) {
- if(count_free_regs(regs[k-1].regmap)<=minimum_free_regs[k-1]) {
- //printf("no free regs for store %x\n",start+(k-1)*4);
- break;
- }
- k--;
- }
- if(regs[k-1].regmap[HOST_CCREG]==CCREG) {
- //printf("Extend CC, %x ->\n",start+k*4);
- while(k<=i) {
- regs[k].regmap_entry[HOST_CCREG]=CCREG;
- regs[k].regmap[HOST_CCREG]=CCREG;
- regmap_pre[k+1][HOST_CCREG]=CCREG;
- regs[k+1].wasdirty|=1<<HOST_CCREG;
- regs[k].dirty|=1<<HOST_CCREG;
- regs[k].wasconst&=~(1<<HOST_CCREG);
- regs[k].isconst&=~(1<<HOST_CCREG);
- k++;
- }
- }
- else {
- //printf("Fail Extend CC, %x ->\n",start+k*4);
- }
- }
- }
- if(itype[i]!=STORE&&itype[i]!=STORELR&&itype[i]!=C1LS&&itype[i]!=SHIFT&&
- itype[i]!=NOP&&itype[i]!=MOV&&itype[i]!=ALU&&itype[i]!=SHIFTIMM&&
- itype[i]!=IMM16&&itype[i]!=LOAD&&itype[i]!=COP1&&itype[i]!=FLOAT&&
- itype[i]!=FCONV&&itype[i]!=FCOMP)
- {
- memcpy(f_regmap,regs[i].regmap,sizeof(f_regmap));
- }
- }
- }
-
- // Cache memory offset or tlb map pointer if a register is available
- #ifndef HOST_IMM_ADDR32
- #ifndef RAM_OFFSET
- if(using_tlb)
- #endif
- {
- int earliest_available[HOST_REGS];
- int loop_start[HOST_REGS];
- int score[HOST_REGS];
- int end[HOST_REGS];
- int reg=using_tlb?MMREG:ROREG;
-
- // Init
- for(hr=0;hr<HOST_REGS;hr++) {
- score[hr]=0;earliest_available[hr]=0;
- loop_start[hr]=MAXBLOCK;
- }
- for(i=0;i<slen-1;i++)
- {
- // Can't do anything if no registers are available
- if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i]) {
- for(hr=0;hr<HOST_REGS;hr++) {
- score[hr]=0;earliest_available[hr]=i+1;
- loop_start[hr]=MAXBLOCK;
- }
- }
- if(itype[i]==UJUMP||itype[i]==RJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP) {
- if(!ooo[i]) {
- if(count_free_regs(branch_regs[i].regmap)<=minimum_free_regs[i+1]) {
- for(hr=0;hr<HOST_REGS;hr++) {
- score[hr]=0;earliest_available[hr]=i+1;
- loop_start[hr]=MAXBLOCK;
- }
- }
- }else{
- if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i+1]) {
- for(hr=0;hr<HOST_REGS;hr++) {
- score[hr]=0;earliest_available[hr]=i+1;
- loop_start[hr]=MAXBLOCK;
- }
- }
- }
- }
- // Mark unavailable registers
- for(hr=0;hr<HOST_REGS;hr++) {
- if(regs[i].regmap[hr]>=0) {
- score[hr]=0;earliest_available[hr]=i+1;
- loop_start[hr]=MAXBLOCK;
- }
- if(itype[i]==UJUMP||itype[i]==RJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP) {
- if(branch_regs[i].regmap[hr]>=0) {
- score[hr]=0;earliest_available[hr]=i+2;
- loop_start[hr]=MAXBLOCK;
- }
- }
- }
- // No register allocations after unconditional jumps
- if(itype[i]==UJUMP||itype[i]==RJUMP||(source[i]>>16)==0x1000)
- {
- for(hr=0;hr<HOST_REGS;hr++) {
- score[hr]=0;earliest_available[hr]=i+2;
- loop_start[hr]=MAXBLOCK;
- }
- i++; // Skip delay slot too
- //printf("skip delay slot: %x\n",start+i*4);
- }
- else
- // Possible match
- if(itype[i]==LOAD||itype[i]==LOADLR||
- itype[i]==STORE||itype[i]==STORELR||itype[i]==C1LS) {
- for(hr=0;hr<HOST_REGS;hr++) {
- if(hr!=EXCLUDE_REG) {
- end[hr]=i-1;
- for(j=i;j<slen-1;j++) {
- if(regs[j].regmap[hr]>=0) break;
- if(itype[j]==UJUMP||itype[j]==RJUMP||itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP) {
- if(branch_regs[j].regmap[hr]>=0) break;
- if(ooo[j]) {
- if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j+1]) break;
- }else{
- if(count_free_regs(branch_regs[j].regmap)<=minimum_free_regs[j+1]) break;
- }
- }
- else if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j]) break;
- if(itype[j]==UJUMP||itype[j]==RJUMP||itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP) {
- int t=(ba[j]-start)>>2;
- if(t<j&&t>=earliest_available[hr]) {
- if(t==1||(t>1&&itype[t-2]!=UJUMP&&itype[t-2]!=RJUMP)||(t>1&&rt1[t-2]!=31)) { // call/ret assumes no registers allocated
- // Score a point for hoisting loop invariant
- if(t<loop_start[hr]) loop_start[hr]=t;
- //printf("set loop_start: i=%x j=%x (%x)\n",start+i*4,start+j*4,start+t*4);
- score[hr]++;
- end[hr]=j;
- }
- }
- else if(t<j) {
- if(regs[t].regmap[hr]==reg) {
- // Score a point if the branch target matches this register
- score[hr]++;
- end[hr]=j;
- }
- }
- if(itype[j+1]==LOAD||itype[j+1]==LOADLR||
- itype[j+1]==STORE||itype[j+1]==STORELR||itype[j+1]==C1LS) {
- score[hr]++;
- end[hr]=j;
- }
- }
- if(itype[j]==UJUMP||itype[j]==RJUMP||(source[j]>>16)==0x1000)
- {
- // Stop on unconditional branch
- break;
- }
- else
- if(itype[j]==LOAD||itype[j]==LOADLR||
- itype[j]==STORE||itype[j]==STORELR||itype[j]==C1LS) {
- score[hr]++;
- end[hr]=j;
- }
- }
- }
- }
- // Find highest score and allocate that register
- int maxscore=0;
- for(hr=0;hr<HOST_REGS;hr++) {
- if(hr!=EXCLUDE_REG) {
- if(score[hr]>score[maxscore]) {
- maxscore=hr;
- //printf("highest score: %d %d (%x->%x)\n",score[hr],hr,start+i*4,start+end[hr]*4);
- }
+ regs[k].wasconst&=~(1<<HOST_CCREG);
+ regs[k].isconst&=~(1<<HOST_CCREG);
+ k++;
}
+ regs[j].regmap_entry[HOST_CCREG]=CCREG;
}
- if(score[maxscore]>1)
+ // Work backwards from the branch target
+ if(j>i&&f_regmap[HOST_CCREG]==CCREG)
{
- if(i<loop_start[maxscore]) loop_start[maxscore]=i;
- for(j=loop_start[maxscore];j<slen&&j<=end[maxscore];j++) {
- //if(regs[j].regmap[maxscore]>=0) {printf("oops: %x %x was %d=%d\n",loop_start[maxscore]*4+start,j*4+start,maxscore,regs[j].regmap[maxscore]);}
- assert(regs[j].regmap[maxscore]<0);
- if(j>loop_start[maxscore]) regs[j].regmap_entry[maxscore]=reg;
- regs[j].regmap[maxscore]=reg;
- regs[j].dirty&=~(1<<maxscore);
- regs[j].wasconst&=~(1<<maxscore);
- regs[j].isconst&=~(1<<maxscore);
- if(itype[j]==UJUMP||itype[j]==RJUMP||itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP) {
- branch_regs[j].regmap[maxscore]=reg;
- branch_regs[j].wasdirty&=~(1<<maxscore);
- branch_regs[j].dirty&=~(1<<maxscore);
- branch_regs[j].wasconst&=~(1<<maxscore);
- branch_regs[j].isconst&=~(1<<maxscore);
- if(itype[j]!=RJUMP&&itype[j]!=UJUMP&&(source[j]>>16)!=0x1000) {
- regmap_pre[j+2][maxscore]=reg;
- regs[j+2].wasdirty&=~(1<<maxscore);
- }
- // loop optimization (loop_preload)
- int t=(ba[j]-start)>>2;
- if(t==loop_start[maxscore]) {
- if(t==1||(t>1&&itype[t-2]!=UJUMP&&itype[t-2]!=RJUMP)||(t>1&&rt1[t-2]!=31)) // call/ret assumes no registers allocated
- regs[t].regmap_entry[maxscore]=reg;
- }
+ //printf("Extend backwards\n");
+ int k;
+ k=i;
+ while(regs[k-1].regmap[HOST_CCREG]==-1) {
+ if(count_free_regs(regs[k-1].regmap)<=minimum_free_regs[k-1]) {
+ //printf("no free regs for store %x\n",start+(k-1)*4);
+ break;
}
- else
- {
- if(j<1||(itype[j-1]!=RJUMP&&itype[j-1]!=UJUMP&&itype[j-1]!=CJUMP&&itype[j-1]!=SJUMP&&itype[j-1]!=FJUMP)) {
- regmap_pre[j+1][maxscore]=reg;
- regs[j+1].wasdirty&=~(1<<maxscore);
- }
+ k--;
+ }
+ if(regs[k-1].regmap[HOST_CCREG]==CCREG) {
+ //printf("Extend CC, %x ->\n",start+k*4);
+ while(k<=i) {
+ regs[k].regmap_entry[HOST_CCREG]=CCREG;
+ regs[k].regmap[HOST_CCREG]=CCREG;
+ regmap_pre[k+1][HOST_CCREG]=CCREG;
+ regs[k+1].wasdirty|=1<<HOST_CCREG;
+ regs[k].dirty|=1<<HOST_CCREG;
+ regs[k].wasconst&=~(1<<HOST_CCREG);
+ regs[k].isconst&=~(1<<HOST_CCREG);
+ k++;
}
}
- i=j-1;
- if(itype[j-1]==RJUMP||itype[j-1]==UJUMP||itype[j-1]==CJUMP||itype[j-1]==SJUMP||itype[j-1]==FJUMP) i++; // skip delay slot
- for(hr=0;hr<HOST_REGS;hr++) {
- score[hr]=0;earliest_available[hr]=i+i;
- loop_start[hr]=MAXBLOCK;
+ else {
+ //printf("Fail Extend CC, %x ->\n",start+k*4);
}
}
}
+ if(dops[i].itype!=STORE&&dops[i].itype!=STORELR&&dops[i].itype!=C1LS&&dops[i].itype!=SHIFT&&
+ dops[i].itype!=NOP&&dops[i].itype!=MOV&&dops[i].itype!=ALU&&dops[i].itype!=SHIFTIMM&&
+ dops[i].itype!=IMM16&&dops[i].itype!=LOAD&&dops[i].itype!=COP1)
+ {
+ memcpy(f_regmap,regs[i].regmap,sizeof(f_regmap));
+ }
}
}
- #endif
-
- // This allocates registers (if possible) one instruction prior
- // to use, which can avoid a load-use penalty on certain CPUs.
+}
+
+// This allocates registers (if possible) one instruction prior
+// to use, which can avoid a load-use penalty on certain CPUs.
+static noinline void pass5b_preallocate2(void)
+{
+ int i, hr;
for(i=0;i<slen-1;i++)
{
- if(!i||(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP))
+ if (!i || !dops[i-1].is_jump)
{
- if(!bt[i+1])
+ if(!dops[i+1].bt)
{
- if(itype[i]==ALU||itype[i]==MOV||itype[i]==LOAD||itype[i]==SHIFTIMM||itype[i]==IMM16
- ||((itype[i]==COP1||itype[i]==COP2)&&opcode2[i]<3))
+ if(dops[i].itype==ALU||dops[i].itype==MOV||dops[i].itype==LOAD||dops[i].itype==SHIFTIMM||dops[i].itype==IMM16
+ ||((dops[i].itype==COP1||dops[i].itype==COP2)&&dops[i].opcode2<3))
{
- if(rs1[i+1]) {
- if((hr=get_reg(regs[i+1].regmap,rs1[i+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)
{
}
}
}
- 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)
{
}
}
// 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<<hr);
regs[i].isconst|=regs[i+1].isconst&(1<<hr);
constmap[i][hr]=constmap[i+1][hr];
}
}
}
- // Load source into target register
- if(lt1[i+1]&&get_reg(regs[i+1].regmap,rs1[i+1])<0) {
- if((hr=get_reg(regs[i+1].regmap,rt1[i+1]))>=0)
+ // Load source into target register
+ if(dops[i+1].use_lt1&&get_reg(regs[i+1].regmap,dops[i+1].rs1)<0) {
+ if((hr=get_reg(regs[i+1].regmap,dops[i+1].rt1))>=0)
{
if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0)
{
- 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<<hr);
regs[i].isconst|=regs[i+1].isconst&(1<<hr);
constmap[i][hr]=constmap[i+1][hr];
}
}
}
- // Preload map address
- #ifndef HOST_IMM_ADDR32
- 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) {
- hr=get_reg(regs[i+1].regmap,TLREG);
- if(hr>=0) {
- int sr=get_reg(regs[i+1].regmap,rs1[i+1]);
- if(sr>=0&&((regs[i+1].wasconst>>sr)&1)) {
- int nr;
- if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0)
- {
- regs[i].regmap[hr]=MGEN1+((i+1)&1);
- regmap_pre[i+1][hr]=MGEN1+((i+1)&1);
- regs[i+1].regmap_entry[hr]=MGEN1+((i+1)&1);
- regs[i].isconst&=~(1<<hr);
- regs[i].isconst|=regs[i+1].isconst&(1<<hr);
- constmap[i][hr]=constmap[i+1][hr];
- regs[i+1].wasdirty&=~(1<<hr);
- regs[i].dirty&=~(1<<hr);
- }
- else if((nr=get_reg2(regs[i].regmap,regs[i+1].regmap,-1))>=0)
- {
- // move it to another register
- regs[i+1].regmap[hr]=-1;
- regmap_pre[i+2][hr]=-1;
- regs[i+1].regmap[nr]=TLREG;
- regmap_pre[i+2][nr]=TLREG;
- regs[i].regmap[nr]=MGEN1+((i+1)&1);
- regmap_pre[i+1][nr]=MGEN1+((i+1)&1);
- regs[i+1].regmap_entry[nr]=MGEN1+((i+1)&1);
- regs[i].isconst&=~(1<<nr);
- regs[i+1].isconst&=~(1<<nr);
- regs[i].dirty&=~(1<<nr);
- regs[i+1].wasdirty&=~(1<<nr);
- regs[i+1].dirty&=~(1<<nr);
- regs[i+2].wasdirty&=~(1<<nr);
- }
- }
- }
- }
- #endif
// Address for store instruction (non-constant)
- if(itype[i+1]==STORE||itype[i+1]==STORELR
- ||(opcode[i+1]&0x3b)==0x39||(opcode[i+1]&0x3b)==0x3a) { // SB/SH/SW/SD/SWC1/SDC1/SWC2/SDC2
- if(get_reg(regs[i+1].regmap,rs1[i+1])<0) {
+ if(dops[i+1].itype==STORE||dops[i+1].itype==STORELR
+ ||(dops[i+1].opcode&0x3b)==0x39||(dops[i+1].opcode&0x3b)==0x3a) { // SB/SH/SW/SD/SWC1/SDC1/SWC2/SDC2
+ if(get_reg(regs[i+1].regmap,dops[i+1].rs1)<0) {
hr=get_reg2(regs[i].regmap,regs[i+1].regmap,-1);
- if(hr<0) hr=get_reg(regs[i+1].regmap,-1);
- else {regs[i+1].regmap[hr]=AGEN1+((i+1)&1);regs[i+1].isconst&=~(1<<hr);}
+ if(hr<0) hr=get_reg_temp(regs[i+1].regmap);
+ else {
+ regs[i+1].regmap[hr]=AGEN1+((i+1)&1);
+ regs[i+1].isconst&=~(1<<hr);
+ }
assert(hr>=0);
if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0)
{
- 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<<hr);
regs[i].isconst|=regs[i+1].isconst&(1<<hr);
constmap[i][hr]=constmap[i+1][hr];
}
}
}
- if(itype[i+1]==LOADLR||(opcode[i+1]&0x3b)==0x31||(opcode[i+1]&0x3b)==0x32) { // LWC1/LDC1, LWC2/LDC2
- if(get_reg(regs[i+1].regmap,rs1[i+1])<0) {
+ if(dops[i+1].itype==LOADLR||(dops[i+1].opcode&0x3b)==0x31||(dops[i+1].opcode&0x3b)==0x32) { // LWC1/LDC1, LWC2/LDC2
+ if(get_reg(regs[i+1].regmap,dops[i+1].rs1)<0) {
int nr;
hr=get_reg(regs[i+1].regmap,FTEMP);
assert(hr>=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<<hr);
regs[i].isconst|=regs[i+1].isconst&(1<<hr);
constmap[i][hr]=constmap[i+1][hr];
regmap_pre[i+2][hr]=-1;
regs[i+1].regmap[nr]=FTEMP;
regmap_pre[i+2][nr]=FTEMP;
- regs[i].regmap[nr]=rs1[i+1];
- regmap_pre[i+1][nr]=rs1[i+1];
- regs[i+1].regmap_entry[nr]=rs1[i+1];
+ regs[i].regmap[nr]=dops[i+1].rs1;
+ regmap_pre[i+1][nr]=dops[i+1].rs1;
+ regs[i+1].regmap_entry[nr]=dops[i+1].rs1;
regs[i].isconst&=~(1<<nr);
regs[i+1].isconst&=~(1<<nr);
regs[i].dirty&=~(1<<nr);
}
}
}
- 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)
- 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
+ if(dops[i+1].itype==LOAD||dops[i+1].itype==LOADLR||dops[i+1].itype==STORE||dops[i+1].itype==STORELR/*||dops[i+1].itype==C1LS||||dops[i+1].itype==C2LS*/) {
+ hr = -1;
+ if(dops[i+1].itype==LOAD)
+ hr=get_reg(regs[i+1].regmap,dops[i+1].rt1);
+ if(dops[i+1].itype==LOADLR||(dops[i+1].opcode&0x3b)==0x31||(dops[i+1].opcode&0x3b)==0x32) // LWC1/LDC1, LWC2/LDC2
hr=get_reg(regs[i+1].regmap,FTEMP);
- if(itype[i+1]==STORE||itype[i+1]==STORELR||(opcode[i+1]&0x3b)==0x39||(opcode[i+1]&0x3b)==0x3a) { // SWC1/SDC1/SWC2/SDC2
+ if(dops[i+1].itype==STORE||dops[i+1].itype==STORELR||(dops[i+1].opcode&0x3b)==0x39||(dops[i+1].opcode&0x3b)==0x3a) { // SWC1/SDC1/SWC2/SDC2
hr=get_reg(regs[i+1].regmap,AGEN1+((i+1)&1));
- if(hr<0) hr=get_reg(regs[i+1].regmap,-1);
+ if(hr<0) hr=get_reg_temp(regs[i+1].regmap);
}
if(hr>=0&®s[i].regmap[hr]<0) {
- int rs=get_reg(regs[i+1].regmap,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);
}
}
}
-
- /* 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--)
+// Write back dirty registers as soon as we will no longer modify them,
+// so that we don't end up with lots of writes at the branches.
+static noinline void pass6_clean_registers(int istart, int iend, int wr)
+{
+ static u_int wont_dirty[MAXBLOCK];
+ static u_int will_dirty[MAXBLOCK];
+ int i;
+ int r;
+ u_int will_dirty_i,will_dirty_next,temp_will_dirty;
+ u_int wont_dirty_i,wont_dirty_next,temp_wont_dirty;
+ if(iend==slen-1) {
+ will_dirty_i=will_dirty_next=0;
+ wont_dirty_i=wont_dirty_next=0;
+ }else{
+ will_dirty_i=will_dirty_next=will_dirty[iend+1];
+ wont_dirty_i=wont_dirty_next=wont_dirty[iend+1];
+ }
+ for (i=iend;i>=istart;i--)
{
- int hr;
- if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+ signed char rregmap_i[RRMAP_SIZE];
+ u_int hr_candirty = 0;
+ assert(HOST_REGS < 32);
+ make_rregs(regs[i].regmap, rregmap_i, &hr_candirty);
+ __builtin_prefetch(regs[i-1].regmap);
+ if(dops[i].is_jump)
{
+ signed char branch_rregmap_i[RRMAP_SIZE];
+ u_int branch_hr_candirty = 0;
+ make_rregs(branch_regs[i].regmap, branch_rregmap_i, &branch_hr_candirty);
if(ba[i]<start || ba[i]>=(start+slen*4))
{
- // Branch out of this block, don't need anything
- r32=0;
+ // Branch out of this block, flush all regs
+ will_dirty_i = 0;
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ will_dirty_i &= branch_hr_candirty;
+ if (dops[i].is_ujump)
+ {
+ // Unconditional branch
+ wont_dirty_i = 0;
+ // Merge in delay slot (will dirty)
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
+ }
+ else
+ {
+ // Conditional branch
+ wont_dirty_i = wont_dirty_next;
+ // Merge in delay slot (will dirty)
+ // (the original code had no explanation why these 2 are commented out)
+ //will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ //will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
+ }
+ // Merge in delay slot (wont dirty)
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ wont_dirty_i &= ~(1u << 31);
+ if(wr) {
+ #ifndef DESTRUCTIVE_WRITEBACK
+ branch_regs[i].dirty&=wont_dirty_i;
+ #endif
+ branch_regs[i].dirty|=will_dirty_i;
+ }
}
else
{
// Internal branch
- // Need whatever matches the target
- // (and doesn't get overwritten by the delay slot instruction)
- r32=0;
- int t=(ba[i]-start)>>2;
- if(ba[i]>start+i*4) {
- // Forward branch
- if(!(requires_32bit[t]&~regs[i].was32))
- r32|=requires_32bit[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
- }else{
+ if(ba[i]<=start+i*4) {
// Backward branch
- //if(!(regs[t].was32&~unneeded_reg_upper[t]&~regs[i].was32))
- // r32|=regs[t].was32&~unneeded_reg_upper[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
- if(!(pr32[t]&~regs[i].was32))
- r32|=pr32[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
+ if (dops[i].is_ujump)
+ {
+ // Unconditional branch
+ temp_will_dirty=0;
+ temp_wont_dirty=0;
+ // Merge in delay slot (will dirty)
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ temp_will_dirty &= branch_hr_candirty;
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ temp_will_dirty &= hr_candirty;
+ } else {
+ // Conditional branch (not taken case)
+ temp_will_dirty=will_dirty_next;
+ temp_wont_dirty=wont_dirty_next;
+ // Merge in delay slot (will dirty)
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ temp_will_dirty &= branch_hr_candirty;
+ //temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ //temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ temp_will_dirty |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ temp_will_dirty &= hr_candirty;
+ }
+ // Merge in delay slot (wont dirty)
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ temp_wont_dirty |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ temp_wont_dirty &= ~(1u << 31);
+ // Deal with changed mappings
+ if(i<iend) {
+ for(r=0;r<HOST_REGS;r++) {
+ if(r!=EXCLUDE_REG) {
+ if(regs[i].regmap[r]!=regmap_pre[i][r]) {
+ temp_will_dirty&=~(1<<r);
+ temp_wont_dirty&=~(1<<r);
+ if(regmap_pre[i][r]>0 && regmap_pre[i][r]<34) {
+ temp_will_dirty|=((unneeded_reg[i]>>regmap_pre[i][r])&1)<<r;
+ temp_wont_dirty|=((unneeded_reg[i]>>regmap_pre[i][r])&1)<<r;
+ } else {
+ temp_will_dirty|=1<<r;
+ temp_wont_dirty|=1<<r;
+ }
+ }
+ }
+ }
+ }
+ if(wr) {
+ will_dirty[i]=temp_will_dirty;
+ wont_dirty[i]=temp_wont_dirty;
+ pass6_clean_registers((ba[i]-start)>>2,i-1,0);
+ }else{
+ // Limit recursion. It can take an excessive amount
+ // of time if there are a lot of nested loops.
+ will_dirty[(ba[i]-start)>>2]=0;
+ wont_dirty[(ba[i]-start)>>2]=-1;
+ }
}
- }
- // Conditional branch may need registers for following instructions
- if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
- {
- if(i<slen-2) {
- r32|=requires_32bit[i+2];
- r32&=regs[i].was32;
- // Mark this address as a branch target since it may be called
- // upon return from interrupt
- bt[i+2]=1;
+ /*else*/ if(1)
+ {
+ if (dops[i].is_ujump)
+ {
+ // Unconditional branch
+ will_dirty_i=0;
+ wont_dirty_i=0;
+ //if(ba[i]>start+i*4) { // Disable recursion (for debugging)
+ for(r=0;r<HOST_REGS;r++) {
+ if(r!=EXCLUDE_REG) {
+ if(branch_regs[i].regmap[r]==regs[(ba[i]-start)>>2].regmap_entry[r]) {
+ will_dirty_i|=will_dirty[(ba[i]-start)>>2]&(1<<r);
+ wont_dirty_i|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
+ }
+ if(branch_regs[i].regmap[r]>=0) {
+ will_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>branch_regs[i].regmap[r])&1)<<r;
+ wont_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>branch_regs[i].regmap[r])&1)<<r;
+ }
+ }
+ }
+ //}
+ // Merge in delay slot
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ will_dirty_i &= branch_hr_candirty;
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
+ } else {
+ // Conditional branch
+ will_dirty_i=will_dirty_next;
+ wont_dirty_i=wont_dirty_next;
+ //if(ba[i]>start+i*4) // Disable recursion (for debugging)
+ for(r=0;r<HOST_REGS;r++) {
+ if(r!=EXCLUDE_REG) {
+ signed char target_reg=branch_regs[i].regmap[r];
+ if(target_reg==regs[(ba[i]-start)>>2].regmap_entry[r]) {
+ will_dirty_i&=will_dirty[(ba[i]-start)>>2]&(1<<r);
+ wont_dirty_i|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
+ }
+ else if(target_reg>=0) {
+ will_dirty_i&=((unneeded_reg[(ba[i]-start)>>2]>>target_reg)&1)<<r;
+ wont_dirty_i|=((unneeded_reg[(ba[i]-start)>>2]>>target_reg)&1)<<r;
+ }
+ }
+ }
+ // Merge in delay slot
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ will_dirty_i &= branch_hr_candirty;
+ //will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ //will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
+ }
+ // Merge in delay slot (won't dirty)
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i+1].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, dops[i+1].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(branch_rregmap_i, CCREG) & 31);
+ wont_dirty_i &= ~(1u << 31);
+ if(wr) {
+ #ifndef DESTRUCTIVE_WRITEBACK
+ branch_regs[i].dirty&=wont_dirty_i;
+ #endif
+ branch_regs[i].dirty|=will_dirty_i;
+ }
}
}
- // Merge in delay slot
- if(!likely[i]) {
- // These are overwritten unless the branch is "likely"
- // and the delay slot is nullified if not taken
- r32&=~(1LL<<rt1[i+1]);
- r32&=~(1LL<<rt2[i+1]);
- }
- // Assume these are needed (delay slot)
- if(us1[i+1]>0)
- {
- if((regs[i].was32>>us1[i+1])&1) r32|=1LL<<us1[i+1];
- }
- if(us2[i+1]>0)
- {
- if((regs[i].was32>>us2[i+1])&1) r32|=1LL<<us2[i+1];
- }
- if(dep1[i+1]&&!((unneeded_reg_upper[i]>>dep1[i+1])&1))
- {
- if((regs[i].was32>>dep1[i+1])&1) r32|=1LL<<dep1[i+1];
- }
- if(dep2[i+1]&&!((unneeded_reg_upper[i]>>dep2[i+1])&1))
- {
- if((regs[i].was32>>dep2[i+1])&1) r32|=1LL<<dep2[i+1];
+ }
+ else if(dops[i].itype==SYSCALL||dops[i].itype==HLECALL||dops[i].itype==INTCALL)
+ {
+ // SYSCALL instruction (software interrupt)
+ will_dirty_i=0;
+ wont_dirty_i=0;
+ }
+ else if(dops[i].itype==COP0 && (source[i]&0x3f)==0x18)
+ {
+ // ERET instruction (return from interrupt)
+ will_dirty_i=0;
+ wont_dirty_i=0;
+ }
+ will_dirty_next=will_dirty_i;
+ wont_dirty_next=wont_dirty_i;
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ will_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ will_dirty_i &= hr_candirty;
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i].rt2) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, CCREG) & 31);
+ wont_dirty_i &= ~(1u << 31);
+ if (i > istart && !dops[i].is_jump) {
+ // Don't store a register immediately after writing it,
+ // may prevent dual-issue.
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i-1].rt1) & 31);
+ wont_dirty_i |= 1u << (get_rreg(rregmap_i, dops[i-1].rt2) & 31);
+ }
+ // Save it
+ will_dirty[i]=will_dirty_i;
+ wont_dirty[i]=wont_dirty_i;
+ // Mark registers that won't be dirtied as not dirty
+ if(wr) {
+ regs[i].dirty|=will_dirty_i;
+ #ifndef DESTRUCTIVE_WRITEBACK
+ regs[i].dirty&=wont_dirty_i;
+ if(dops[i].is_jump)
+ {
+ if (i < iend-1 && !dops[i].is_ujump) {
+ for(r=0;r<HOST_REGS;r++) {
+ if(r!=EXCLUDE_REG) {
+ if(regs[i].regmap[r]==regmap_pre[i+2][r]) {
+ regs[i+2].wasdirty&=wont_dirty_i|~(1<<r);
+ }else {/*printf("i: %x (%d) mismatch(+2): %d\n",start+i*4,i,r);assert(!((wont_dirty_i>>r)&1));*/}
+ }
+ }
+ }
+ }
+ else
+ {
+ if(i<iend) {
+ for(r=0;r<HOST_REGS;r++) {
+ if(r!=EXCLUDE_REG) {
+ if(regs[i].regmap[r]==regmap_pre[i+1][r]) {
+ regs[i+1].wasdirty&=wont_dirty_i|~(1<<r);
+ }else {/*printf("i: %x (%d) mismatch(+1): %d\n",start+i*4,i,r);assert(!((wont_dirty_i>>r)&1));*/}
+ }
+ }
+ }
+ }
+ #endif
+ }
+ // Deal with changed mappings
+ temp_will_dirty=will_dirty_i;
+ temp_wont_dirty=wont_dirty_i;
+ for(r=0;r<HOST_REGS;r++) {
+ if(r!=EXCLUDE_REG) {
+ int nr;
+ if(regs[i].regmap[r]==regmap_pre[i][r]) {
+ if(wr) {
+ #ifndef DESTRUCTIVE_WRITEBACK
+ regs[i].wasdirty&=wont_dirty_i|~(1<<r);
+ #endif
+ regs[i].wasdirty|=will_dirty_i&(1<<r);
+ }
+ }
+ else if(regmap_pre[i][r]>=0&&(nr=get_rreg(rregmap_i,regmap_pre[i][r]))>=0) {
+ // Register moved to a different register
+ will_dirty_i&=~(1<<r);
+ wont_dirty_i&=~(1<<r);
+ will_dirty_i|=((temp_will_dirty>>nr)&1)<<r;
+ wont_dirty_i|=((temp_wont_dirty>>nr)&1)<<r;
+ if(wr) {
+ #ifndef DESTRUCTIVE_WRITEBACK
+ regs[i].wasdirty&=wont_dirty_i|~(1<<r);
+ #endif
+ regs[i].wasdirty|=will_dirty_i&(1<<r);
+ }
+ }
+ else {
+ will_dirty_i&=~(1<<r);
+ wont_dirty_i&=~(1<<r);
+ if(regmap_pre[i][r]>0 && regmap_pre[i][r]<34) {
+ will_dirty_i|=((unneeded_reg[i]>>regmap_pre[i][r])&1)<<r;
+ wont_dirty_i|=((unneeded_reg[i]>>regmap_pre[i][r])&1)<<r;
+ } else {
+ wont_dirty_i|=1<<r;
+ /*printf("i: %x (%d) mismatch: %d\n",start+i*4,i,r);assert(!((will_dirty>>r)&1));*/
+ }
+ }
}
}
- else if(itype[i]==SYSCALL||itype[i]==HLECALL||itype[i]==INTCALL)
- {
- // SYSCALL instruction (software interrupt)
- r32=0;
- }
- else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
- {
- // ERET instruction (return from interrupt)
- r32=0;
- }
- // Check 32 bits
- r32&=~(1LL<<rt1[i]);
- r32&=~(1LL<<rt2[i]);
- if(us1[i]>0)
- {
- if((regs[i].was32>>us1[i])&1) r32|=1LL<<us1[i];
- }
- if(us2[i]>0)
- {
- if((regs[i].was32>>us2[i])&1) r32|=1LL<<us2[i];
- }
- if(dep1[i]&&!((unneeded_reg_upper[i]>>dep1[i])&1))
- {
- if((regs[i].was32>>dep1[i])&1) r32|=1LL<<dep1[i];
- }
- if(dep2[i]&&!((unneeded_reg_upper[i]>>dep2[i])&1))
- {
- if((regs[i].was32>>dep2[i])&1) r32|=1LL<<dep2[i];
- }
- requires_32bit[i]=r32;
-
- // Dirty registers which are 32-bit, require 32-bit input
- // as they will be written as 32-bit values
- for(hr=0;hr<HOST_REGS;hr++)
+ }
+}
+
+static noinline void pass10_expire_blocks(void)
+{
+ int i, end;
+ end = (((out-ndrc->translation_cache)>>(TARGET_SIZE_2-16)) + 16384) & 65535;
+ while (expirep != end)
+ {
+ int shift=TARGET_SIZE_2-3; // Divide into 8 blocks
+ uintptr_t base_offs = ((uintptr_t)(expirep >> 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)
{
- if(regs[i].regmap_entry[hr]>0&®s[i].regmap_entry[hr]<64) {
- if((regs[i].was32>>regs[i].regmap_entry[hr])&(regs[i].wasdirty>>hr)&1) {
- if(!((unneeded_reg_upper[i]>>regs[i].regmap_entry[hr])&1))
- requires_32bit[i]|=1LL<<regs[i].regmap_entry[hr];
+ case 0:
+ // Clear jump_in and jump_dirty
+ 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_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];
+ 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;
+ }
+ 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];
+ ht_bin->vaddr[1] = -1;
+ ht_bin->tcaddr[1] = NULL;
+ }
}
- }
+ break;
+ case 3:
+ // Clear jump_out
+ if((expirep&2047)==0)
+ do_clear_cache();
+ 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;
}
- //requires_32bit[i]=is32[i]&~unneeded_reg_upper[i]; // DEBUG
+ expirep=(expirep+1)&65535;
}
-#else
+}
+
+int new_recompile_block(u_int addr)
+{
+ u_int pagelimit = 0;
+ u_int state_rflags = 0;
+ int i;
+
+ assem_debug("NOTCOMPILED: addr = %x -> %p\n", addr, out);
+
+ // 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); // start-in-delay-slot flag
+ new_dynarec_did_compile=1;
+ if (Config.HLE && start == 0x80001000) // hlecall
+ {
+ // XXX: is this enough? Maybe check hleSoftCall?
+ void *beginning=start_block();
+ u_int page=get_page(start);
+
+ invalid_code[start>>12]=0;
+ emit_movimm(start,0);
+ emit_writeword(0,&pcaddr);
+ emit_far_jump(new_dyna_leave);
+ literal_pool(0);
+ end_block(beginning);
+ ll_add_flags(jump_in+page,start,state_rflags,(void *)beginning);
+ return 0;
+ }
+ else if (f1_hack && hack_addr == 0) {
+ void *beginning = start_block();
+ u_int page = get_page(start);
+ emit_movimm(start, 0);
+ emit_writeword(0, &hack_addr);
+ emit_readword(&psxRegs.GPR.n.sp, 0);
+ emit_readptr(&mem_rtab, 1);
+ emit_shrimm(0, 12, 2);
+ 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);
+ return 0;
+ }
+
+ cycle_multiplier_active = cycle_multiplier_override && cycle_multiplier == CYCLE_MULT_DEFAULT
+ ? cycle_multiplier_override : cycle_multiplier;
+
+ source = get_source_start(start, &pagelimit);
+ if (source == NULL) {
+ if (addr != hack_addr) {
+ SysPrintf("Compile at bogus memory address: %08x\n", addr);
+ hack_addr = addr;
+ }
+ //abort();
+ return -1;
+ }
+
+ /* Pass 1: disassemble */
+ /* Pass 2: register dependencies, branch targets */
+ /* Pass 3: register allocation */
+ /* Pass 4: branch dependencies */
+ /* Pass 5: pre-alloc */
+ /* Pass 6: optimize clean/dirty state */
+ /* Pass 7: flag 32-bit registers */
+ /* Pass 8: assembly */
+ /* Pass 9: linker */
+ /* Pass 10: garbage collection / free memory */
+
+ /* Pass 1 disassembly */
+
+ pass1_disassemble(pagelimit);
+
+ int clear_hack_addr = apply_hacks();
+
+ /* Pass 2 - Register dependencies and branch targets */
+
+ pass2_unneeded_regs(0,slen-1,0);
+
+ /* Pass 3 - Register allocation */
+
+ pass3_register_alloc(addr);
+
+ /* Pass 4 - Cull unused host registers */
+
+ pass4_cull_unused_regs();
+
+ /* Pass 5 - Pre-allocate registers */
+
+ pass5a_preallocate1();
+ pass5b_preallocate2();
+
+ /* Pass 6 - Optimize clean/dirty state */
+ pass6_clean_registers(0, slen-1, 1);
+
+ /* Pass 7 - Identify 32-bit registers */
for (i=slen-1;i>=0;i--)
{
- if(itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+ if(dops[i].itype==CJUMP||dops[i].itype==SJUMP)
{
// Conditional branch
if((source[i]>>16)!=0x1000&&i<slen-2) {
// Mark this address as a branch target since it may be called
// upon return from interrupt
- bt[i+2]=1;
+ dops[i+2].bt=1;
}
}
}
-#endif
- if(itype[slen-1]==SPAN) {
- bt[slen-1]=1; // Mark as a branch target so instruction can restart after exception
- }
-
- /* Debug/disassembly */
- if((void*)assem_debug==(void*)printf)
- for(i=0;i<slen;i++)
- {
- printf("U:");
- int r;
- for(r=1;r<=CCREG;r++) {
- if((unneeded_reg[i]>>r)&1) {
- if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
-#ifndef FORCE32
- printf(" UU:");
- for(r=1;r<=CCREG;r++) {
- if(((unneeded_reg_upper[i]&~unneeded_reg[i])>>r)&1) {
- if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
- printf(" 32:");
- for(r=0;r<=CCREG;r++) {
- //if(((is32[i]>>r)&(~unneeded_reg[i]>>r))&1) {
- if((regs[i].was32>>r)&1) {
- if(r==CCREG) printf(" CC");
- else if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
-#endif
- printf("\n");
- #if defined(__i386__) || defined(__x86_64__)
- printf("pre: eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",regmap_pre[i][0],regmap_pre[i][1],regmap_pre[i][2],regmap_pre[i][3],regmap_pre[i][5],regmap_pre[i][6],regmap_pre[i][7]);
- #endif
- #ifdef __arm__
- printf("pre: r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d\n",regmap_pre[i][0],regmap_pre[i][1],regmap_pre[i][2],regmap_pre[i][3],regmap_pre[i][4],regmap_pre[i][5],regmap_pre[i][6],regmap_pre[i][7],regmap_pre[i][8],regmap_pre[i][9],regmap_pre[i][10],regmap_pre[i][12]);
- #endif
- printf("needs: ");
- if(needed_reg[i]&1) printf("eax ");
- if((needed_reg[i]>>1)&1) printf("ecx ");
- if((needed_reg[i]>>2)&1) printf("edx ");
- if((needed_reg[i]>>3)&1) printf("ebx ");
- if((needed_reg[i]>>5)&1) printf("ebp ");
- if((needed_reg[i]>>6)&1) printf("esi ");
- if((needed_reg[i]>>7)&1) printf("edi ");
- printf("r:");
- for(r=0;r<=CCREG;r++) {
- //if(((requires_32bit[i]>>r)&(~unneeded_reg[i]>>r))&1) {
- if((requires_32bit[i]>>r)&1) {
- if(r==CCREG) printf(" CC");
- else if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
- printf("\n");
- /*printf("pr:");
- for(r=0;r<=CCREG;r++) {
- //if(((requires_32bit[i]>>r)&(~unneeded_reg[i]>>r))&1) {
- if((pr32[i]>>r)&1) {
- if(r==CCREG) printf(" CC");
- else if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
- if(pr32[i]!=requires_32bit[i]) printf(" OOPS");
- printf("\n");*/
- #if defined(__i386__) || defined(__x86_64__)
- printf("entry: eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",regs[i].regmap_entry[0],regs[i].regmap_entry[1],regs[i].regmap_entry[2],regs[i].regmap_entry[3],regs[i].regmap_entry[5],regs[i].regmap_entry[6],regs[i].regmap_entry[7]);
- printf("dirty: ");
- if(regs[i].wasdirty&1) printf("eax ");
- if((regs[i].wasdirty>>1)&1) printf("ecx ");
- if((regs[i].wasdirty>>2)&1) printf("edx ");
- if((regs[i].wasdirty>>3)&1) printf("ebx ");
- if((regs[i].wasdirty>>5)&1) printf("ebp ");
- if((regs[i].wasdirty>>6)&1) printf("esi ");
- if((regs[i].wasdirty>>7)&1) printf("edi ");
- #endif
- #ifdef __arm__
- printf("entry: r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d\n",regs[i].regmap_entry[0],regs[i].regmap_entry[1],regs[i].regmap_entry[2],regs[i].regmap_entry[3],regs[i].regmap_entry[4],regs[i].regmap_entry[5],regs[i].regmap_entry[6],regs[i].regmap_entry[7],regs[i].regmap_entry[8],regs[i].regmap_entry[9],regs[i].regmap_entry[10],regs[i].regmap_entry[12]);
- printf("dirty: ");
- if(regs[i].wasdirty&1) printf("r0 ");
- if((regs[i].wasdirty>>1)&1) printf("r1 ");
- if((regs[i].wasdirty>>2)&1) printf("r2 ");
- if((regs[i].wasdirty>>3)&1) printf("r3 ");
- if((regs[i].wasdirty>>4)&1) printf("r4 ");
- if((regs[i].wasdirty>>5)&1) printf("r5 ");
- if((regs[i].wasdirty>>6)&1) printf("r6 ");
- if((regs[i].wasdirty>>7)&1) printf("r7 ");
- if((regs[i].wasdirty>>8)&1) printf("r8 ");
- if((regs[i].wasdirty>>9)&1) printf("r9 ");
- if((regs[i].wasdirty>>10)&1) printf("r10 ");
- if((regs[i].wasdirty>>12)&1) printf("r12 ");
- #endif
- printf("\n");
- disassemble_inst(i);
- //printf ("ccadj[%d] = %d\n",i,ccadj[i]);
- #if defined(__i386__) || defined(__x86_64__)
- printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d dirty: ",regs[i].regmap[0],regs[i].regmap[1],regs[i].regmap[2],regs[i].regmap[3],regs[i].regmap[5],regs[i].regmap[6],regs[i].regmap[7]);
- if(regs[i].dirty&1) printf("eax ");
- if((regs[i].dirty>>1)&1) printf("ecx ");
- if((regs[i].dirty>>2)&1) printf("edx ");
- if((regs[i].dirty>>3)&1) printf("ebx ");
- if((regs[i].dirty>>5)&1) printf("ebp ");
- if((regs[i].dirty>>6)&1) printf("esi ");
- if((regs[i].dirty>>7)&1) printf("edi ");
- #endif
- #ifdef __arm__
- printf("r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d dirty: ",regs[i].regmap[0],regs[i].regmap[1],regs[i].regmap[2],regs[i].regmap[3],regs[i].regmap[4],regs[i].regmap[5],regs[i].regmap[6],regs[i].regmap[7],regs[i].regmap[8],regs[i].regmap[9],regs[i].regmap[10],regs[i].regmap[12]);
- if(regs[i].dirty&1) printf("r0 ");
- if((regs[i].dirty>>1)&1) printf("r1 ");
- if((regs[i].dirty>>2)&1) printf("r2 ");
- if((regs[i].dirty>>3)&1) printf("r3 ");
- if((regs[i].dirty>>4)&1) printf("r4 ");
- if((regs[i].dirty>>5)&1) printf("r5 ");
- if((regs[i].dirty>>6)&1) printf("r6 ");
- if((regs[i].dirty>>7)&1) printf("r7 ");
- if((regs[i].dirty>>8)&1) printf("r8 ");
- if((regs[i].dirty>>9)&1) printf("r9 ");
- if((regs[i].dirty>>10)&1) printf("r10 ");
- if((regs[i].dirty>>12)&1) printf("r12 ");
- #endif
- printf("\n");
- if(regs[i].isconst) {
- printf("constants: ");
- #if defined(__i386__) || defined(__x86_64__)
- if(regs[i].isconst&1) printf("eax=%x ",(int)constmap[i][0]);
- if((regs[i].isconst>>1)&1) printf("ecx=%x ",(int)constmap[i][1]);
- if((regs[i].isconst>>2)&1) printf("edx=%x ",(int)constmap[i][2]);
- if((regs[i].isconst>>3)&1) printf("ebx=%x ",(int)constmap[i][3]);
- if((regs[i].isconst>>5)&1) printf("ebp=%x ",(int)constmap[i][5]);
- if((regs[i].isconst>>6)&1) printf("esi=%x ",(int)constmap[i][6]);
- if((regs[i].isconst>>7)&1) printf("edi=%x ",(int)constmap[i][7]);
- #endif
- #ifdef __arm__
- if(regs[i].isconst&1) printf("r0=%x ",(int)constmap[i][0]);
- if((regs[i].isconst>>1)&1) printf("r1=%x ",(int)constmap[i][1]);
- if((regs[i].isconst>>2)&1) printf("r2=%x ",(int)constmap[i][2]);
- if((regs[i].isconst>>3)&1) printf("r3=%x ",(int)constmap[i][3]);
- if((regs[i].isconst>>4)&1) printf("r4=%x ",(int)constmap[i][4]);
- if((regs[i].isconst>>5)&1) printf("r5=%x ",(int)constmap[i][5]);
- if((regs[i].isconst>>6)&1) printf("r6=%x ",(int)constmap[i][6]);
- if((regs[i].isconst>>7)&1) printf("r7=%x ",(int)constmap[i][7]);
- if((regs[i].isconst>>8)&1) printf("r8=%x ",(int)constmap[i][8]);
- if((regs[i].isconst>>9)&1) printf("r9=%x ",(int)constmap[i][9]);
- if((regs[i].isconst>>10)&1) printf("r10=%x ",(int)constmap[i][10]);
- if((regs[i].isconst>>12)&1) printf("r12=%x ",(int)constmap[i][12]);
- #endif
- printf("\n");
- }
-#ifndef FORCE32
- printf(" 32:");
- for(r=0;r<=CCREG;r++) {
- if((regs[i].is32>>r)&1) {
- if(r==CCREG) printf(" CC");
- else if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
- printf("\n");
-#endif
- /*printf(" p32:");
- for(r=0;r<=CCREG;r++) {
- if((p32[i]>>r)&1) {
- if(r==CCREG) printf(" CC");
- else if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
- if(p32[i]!=regs[i].is32) printf(" NO MATCH\n");
- else printf("\n");*/
- if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP) {
- #if defined(__i386__) || defined(__x86_64__)
- printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d dirty: ",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]);
- if(branch_regs[i].dirty&1) printf("eax ");
- if((branch_regs[i].dirty>>1)&1) printf("ecx ");
- if((branch_regs[i].dirty>>2)&1) printf("edx ");
- if((branch_regs[i].dirty>>3)&1) printf("ebx ");
- if((branch_regs[i].dirty>>5)&1) printf("ebp ");
- if((branch_regs[i].dirty>>6)&1) printf("esi ");
- if((branch_regs[i].dirty>>7)&1) printf("edi ");
- #endif
- #ifdef __arm__
- printf("branch(%d): r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d dirty: ",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[4],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7],branch_regs[i].regmap[8],branch_regs[i].regmap[9],branch_regs[i].regmap[10],branch_regs[i].regmap[12]);
- if(branch_regs[i].dirty&1) printf("r0 ");
- if((branch_regs[i].dirty>>1)&1) printf("r1 ");
- if((branch_regs[i].dirty>>2)&1) printf("r2 ");
- if((branch_regs[i].dirty>>3)&1) printf("r3 ");
- if((branch_regs[i].dirty>>4)&1) printf("r4 ");
- if((branch_regs[i].dirty>>5)&1) printf("r5 ");
- if((branch_regs[i].dirty>>6)&1) printf("r6 ");
- if((branch_regs[i].dirty>>7)&1) printf("r7 ");
- if((branch_regs[i].dirty>>8)&1) printf("r8 ");
- if((branch_regs[i].dirty>>9)&1) printf("r9 ");
- if((branch_regs[i].dirty>>10)&1) printf("r10 ");
- if((branch_regs[i].dirty>>12)&1) printf("r12 ");
- #endif
-#ifndef FORCE32
- printf(" 32:");
- for(r=0;r<=CCREG;r++) {
- if((branch_regs[i].is32>>r)&1) {
- if(r==CCREG) printf(" CC");
- else if(r==HIREG) printf(" HI");
- else if(r==LOREG) printf(" LO");
- else printf(" r%d",r);
- }
- }
- printf("\n");
-#endif
- }
+ if(dops[slen-1].itype==SPAN) {
+ dops[slen-1].bt=1; // Mark as a branch target so instruction can restart after exception
}
/* Pass 8 - Assembly */
linkcount=0;stubcount=0;
- ds=0;is_delayslot=0;
- cop1_usable=0;
- uint64_t is32_pre=0;
+ is_delayslot=0;
u_int dirty_pre=0;
- u_int beginning=(u_int)out;
+ void *beginning=start_block();
+ int ds = 0;
if((u_int)addr&1) {
ds=1;
pagespan_ds();
}
- u_int instr_addr0_override=0;
+ void *instr_addr0_override = NULL;
-#ifdef PCSX
if (start == 0x80030000) {
- // nasty hack for fastbios thing
+ // nasty hack for the fastbios thing
// override block entry to this code
- instr_addr0_override=(u_int)out;
+ instr_addr0_override = out;
emit_movimm(start,0);
// abuse io address var as a flag that we
// have already returned here once
- emit_readword((int)&address,1);
- emit_writeword(0,(int)&pcaddr);
- emit_writeword(0,(int)&address);
+ emit_readword(&address,1);
+ emit_writeword(0,&pcaddr);
+ emit_writeword(0,&address);
emit_cmp(0,1);
- emit_jne((int)new_dyna_leave);
+ #ifdef __aarch64__
+ emit_jeq(out + 4*2);
+ emit_far_jump(new_dyna_leave);
+ #else
+ emit_jne(new_dyna_leave);
+ #endif
}
-#endif
for(i=0;i<slen;i++)
{
+ __builtin_prefetch(regs[i+1].regmap);
+ check_regmap(regmap_pre[i]);
+ check_regmap(regs[i].regmap_entry);
+ check_regmap(regs[i].regmap);
//if(ds) printf("ds: ");
- if((void*)assem_debug==(void*)printf) disassemble_inst(i);
+ disassemble_inst(i);
if(ds) {
ds=0; // Skip delay slot
- if(bt[i]) assem_debug("OOPS - branch into delay slot\n");
- instr_addr[i]=0;
+ if(dops[i].bt) assem_debug("OOPS - branch into delay slot\n");
+ instr_addr[i] = NULL;
} else {
+ speculate_register_values(i);
#ifndef DESTRUCTIVE_WRITEBACK
- if(i<2||(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000))
+ if (i < 2 || !dops[i-2].is_ujump)
{
- wb_sx(regmap_pre[i],regs[i].regmap_entry,regs[i].wasdirty,is32_pre,regs[i].was32,
- unneeded_reg[i],unneeded_reg_upper[i]);
- wb_valid(regmap_pre[i],regs[i].regmap_entry,dirty_pre,regs[i].wasdirty,is32_pre,
- unneeded_reg[i],unneeded_reg_upper[i]);
+ wb_valid(regmap_pre[i],regs[i].regmap_entry,dirty_pre,regs[i].wasdirty,unneeded_reg[i]);
}
- if((itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)&&!likely[i]) {
- is32_pre=branch_regs[i].is32;
+ if((dops[i].itype==CJUMP||dops[i].itype==SJUMP)) {
dirty_pre=branch_regs[i].dirty;
}else{
- is32_pre=regs[i].is32;
dirty_pre=regs[i].dirty;
}
#endif
// write back
- if(i<2||(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000))
+ if (i < 2 || !dops[i-2].is_ujump)
{
- wb_invalidate(regmap_pre[i],regs[i].regmap_entry,regs[i].wasdirty,regs[i].was32,
- unneeded_reg[i],unneeded_reg_upper[i]);
+ wb_invalidate(regmap_pre[i],regs[i].regmap_entry,regs[i].wasdirty,unneeded_reg[i]);
loop_preload(regmap_pre[i],regs[i].regmap_entry);
}
// branch target entry point
- instr_addr[i]=(u_int)out;
+ instr_addr[i] = out;
assem_debug("<->\n");
+ drc_dbg_emit_do_cmp(i, ccadj[i]);
+ if (clear_hack_addr) {
+ emit_movimm(0, 0);
+ emit_writeword(0, &hack_addr);
+ clear_hack_addr = 0;
+ }
+
// load regs
if(regs[i].regmap_entry[HOST_CCREG]==CCREG&®s[i].regmap[HOST_CCREG]!=CCREG)
- wb_register(CCREG,regs[i].regmap_entry,regs[i].wasdirty,regs[i].was32);
- load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs1[i],rs2[i]);
+ wb_register(CCREG,regs[i].regmap_entry,regs[i].wasdirty);
+ 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,regs[i].was32,i);
- if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+ load_consts(regmap_pre[i],regs[i].regmap,i);
+ 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,regs[i].was32,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,regs[i].was32,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)
- load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,INVCP,INVCP);
+ 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_reg(regs[i].regmap_entry,regs[i].regmap,ROREG);
+ if (dops[i+1].is_store)
+ load_reg(regs[i].regmap_entry,regs[i].regmap,INVCP);
}
else if(i+1<slen)
{
// Preload registers for following instruction
- if(rs1[i+1]!=rs1[i]&&rs1[i+1]!=rs2[i])
- if(rs1[i+1]!=rt1[i]&&rs1[i+1]!=rt2[i])
- load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs1[i+1],rs1[i+1]);
- if(rs2[i+1]!=rs1[i+1]&&rs2[i+1]!=rs1[i]&&rs2[i+1]!=rs2[i])
- if(rs2[i+1]!=rt1[i]&&rs2[i+1]!=rt2[i])
- load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs2[i+1],rs2[i+1]);
+ if(dops[i+1].rs1!=dops[i].rs1&&dops[i+1].rs1!=dops[i].rs2)
+ if(dops[i+1].rs1!=dops[i].rt1&&dops[i+1].rs1!=dops[i].rt2)
+ 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)
+ if(dops[i+1].rs2!=dops[i].rt1&&dops[i+1].rs2!=dops[i].rt2)
+ load_regs(regs[i].regmap_entry,regs[i].regmap,dops[i+1].rs2,dops[i+1].rs2);
}
// TODO: if(is_ooo(i)) address_generation(i+1);
- if(itype[i]==CJUMP||itype[i]==FJUMP)
- load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,CCREG,CCREG);
- if(itype[i]==STORE||itype[i]==STORELR||(opcode[i]&0x3b)==0x39||(opcode[i]&0x3b)==0x3a)
- load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,INVCP,INVCP);
- if(bt[i]) cop1_usable=0;
- // assemble
- switch(itype[i]) {
- case ALU:
- alu_assemble(i,®s[i]);break;
- case IMM16:
- imm16_assemble(i,®s[i]);break;
- case SHIFT:
- shift_assemble(i,®s[i]);break;
- case SHIFTIMM:
- shiftimm_assemble(i,®s[i]);break;
- case LOAD:
- load_assemble(i,®s[i]);break;
- case LOADLR:
- loadlr_assemble(i,®s[i]);break;
- case STORE:
- store_assemble(i,®s[i]);break;
- case STORELR:
- storelr_assemble(i,®s[i]);break;
- case COP0:
- cop0_assemble(i,®s[i]);break;
- case COP1:
- cop1_assemble(i,®s[i]);break;
- case C1LS:
- c1ls_assemble(i,®s[i]);break;
- case COP2:
- cop2_assemble(i,®s[i]);break;
- case C2LS:
- c2ls_assemble(i,®s[i]);break;
- case C2OP:
- c2op_assemble(i,®s[i]);break;
- case FCONV:
- fconv_assemble(i,®s[i]);break;
- case FLOAT:
- float_assemble(i,®s[i]);break;
- case FCOMP:
- fcomp_assemble(i,®s[i]);break;
- case MULTDIV:
- multdiv_assemble(i,®s[i]);break;
- case MOV:
- mov_assemble(i,®s[i]);break;
- case SYSCALL:
- syscall_assemble(i,®s[i]);break;
- case HLECALL:
- hlecall_assemble(i,®s[i]);break;
- case INTCALL:
- intcall_assemble(i,®s[i]);break;
- case UJUMP:
- ujump_assemble(i,®s[i]);ds=1;break;
- case RJUMP:
- rjump_assemble(i,®s[i]);ds=1;break;
- case CJUMP:
- cjump_assemble(i,®s[i]);ds=1;break;
- case SJUMP:
- sjump_assemble(i,®s[i]);ds=1;break;
- case FJUMP:
- fjump_assemble(i,®s[i]);ds=1;break;
- case SPAN:
- pagespan_assemble(i,®s[i]);break;
- }
- if(itype[i]==UJUMP||itype[i]==RJUMP||(source[i]>>16)==0x1000)
+ if (!dops[i].is_jump || dops[i].itype == CJUMP)
+ load_reg(regs[i].regmap_entry,regs[i].regmap,CCREG);
+ if (ram_offset && (dops[i].is_load || dops[i].is_store))
+ load_reg(regs[i].regmap_entry,regs[i].regmap,ROREG);
+ if (dops[i].is_store)
+ load_reg(regs[i].regmap_entry,regs[i].regmap,INVCP);
+
+ ds = assemble(i, ®s[i], ccadj[i]);
+
+ if (dops[i].is_ujump)
literal_pool(1024);
else
literal_pool_jumpover(256);
}
}
- //assert(itype[i-2]==UJUMP||itype[i-2]==RJUMP||(source[i-2]>>16)==0x1000);
+
+ assert(slen > 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(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000&&itype[i-1]!=SPAN) {
- assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP);
+ 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&&itype[i-2]!=FJUMP) {
- store_regs_bt(regs[i-1].regmap,regs[i-1].is32,regs[i-1].dirty,start+i*4);
+ 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_DIVIDER*(ccadj[i-1]+1),HOST_CCREG);
- }
- else if(!likely[i-2])
- {
- store_regs_bt(branch_regs[i-2].regmap,branch_regs[i-2].is32,branch_regs[i-2].dirty,start+i*4);
- assert(branch_regs[i-2].regmap[HOST_CCREG]==CCREG);
+ emit_addimm(HOST_CCREG, ccadj[i-1] + CLOCK_ADJUST(1), HOST_CCREG);
}
else
{
- store_regs_bt(regs[i-2].regmap,regs[i-2].is32,regs[i-2].dirty,start+i*4);
- assert(regs[i-2].regmap[HOST_CCREG]==CCREG);
+ 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);
}
- add_to_linker((int)out,start+i*4,0);
+ add_to_linker(out,start+i*4,0);
emit_jmp(0);
}
}
else
{
assert(i>0);
- assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP);
- store_regs_bt(regs[i-1].regmap,regs[i-1].is32,regs[i-1].dirty,start+i*4);
+ 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_DIVIDER*(ccadj[i-1]+1),HOST_CCREG);
- add_to_linker((int)out,start+i*4,0);
+ emit_addimm(HOST_CCREG, ccadj[i-1] + CLOCK_ADJUST(1), HOST_CCREG);
+ add_to_linker(out,start+i*4,0);
emit_jmp(0);
}
// Stubs
for(i=0;i<stubcount;i++)
{
- switch(stubs[i][0])
+ switch(stubs[i].type)
{
case LOADB_STUB:
case LOADH_STUB:
/* Pass 9 - Linker */
for(i=0;i<linkcount;i++)
{
- assem_debug("%8x -> %8x\n",link_addr[i][0],link_addr[i][1]);
+ assem_debug("%p -> %8x\n",link_addr[i].addr,link_addr[i].target);
literal_pool(64);
- if(!link_addr[i][2])
+ if (!link_addr[i].ext)
{
- void *stub=out;
- void *addr=check_addr(link_addr[i][1]);
- emit_extjump(link_addr[i][0],link_addr[i][1]);
- if(addr) {
- set_jump_target(link_addr[i][0],(int)addr);
- add_link(link_addr[i][1],stub);
+ void *stub = out;
+ void *addr = check_addr(link_addr[i].target);
+ emit_extjump(link_addr[i].addr, link_addr[i].target);
+ if (addr) {
+ set_jump_target(link_addr[i].addr, addr);
+ add_jump_out(link_addr[i].target,stub);
}
- else set_jump_target(link_addr[i][0],(int)stub);
+ else
+ set_jump_target(link_addr[i].addr, stub);
}
else
{
// Internal branch
- int target=(link_addr[i][1]-start)>>2;
+ int target=(link_addr[i].target-start)>>2;
assert(target>=0&&target<slen);
assert(instr_addr[target]);
//#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
- //set_jump_target_fillslot(link_addr[i][0],instr_addr[target],link_addr[i][2]>>1);
+ //set_jump_target_fillslot(link_addr[i].addr,instr_addr[target],link_addr[i].ext>>1);
//#else
- set_jump_target(link_addr[i][0],instr_addr[target]);
+ set_jump_target(link_addr[i].addr, instr_addr[target]);
//#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;i<slen;i++)
{
- if(bt[i]||i==0)
+ if(dops[i].bt||i==0)
{
if(instr_addr[i]) // TODO - delay slots (=null)
{
u_int page=get_page(vaddr);
u_int vpage=get_vpage(vaddr);
literal_pool(256);
- //if(!(is32[i]&(~unneeded_reg_upper[i])&~(1LL<<CCREG)))
-#ifndef FORCE32
- if(!requires_32bit[i])
-#else
- if(1)
-#endif
{
- assem_debug("%8x (%d) <- %8x\n",instr_addr[i],i,start+i*4);
+ assem_debug("%p (%d) <- %8x\n",instr_addr[i],i,start+i*4);
assem_debug("jump_in: %x\n",start+i*4);
- ll_add(jump_dirty+vpage,vaddr,(void *)out);
- int entry_point=do_dirty_stub(i);
- ll_add(jump_in+page,vaddr,(void *)entry_point);
+ ll_add(jump_dirty+vpage,vaddr,out);
+ void *entry_point = do_dirty_stub(i, source_len);
+ ll_add_flags(jump_in+page,vaddr,state_rflags,entry_point);
// If there was an existing entry in the hash table,
// replace it with the new address.
// Don't add new entries. We'll insert the
// ones that actually get used in check_addr().
- int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
- if(ht_bin[0]==vaddr) {
- ht_bin[1]=entry_point;
- }
- if(ht_bin[2]==vaddr) {
- ht_bin[3]=entry_point;
- }
- }
- else
- {
- u_int r=requires_32bit[i]|!!(requires_32bit[i]>>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);
+ struct ht_entry *ht_bin = hash_table_get(vaddr);
+ if (ht_bin->vaddr[0] == vaddr)
+ ht_bin->tcaddr[0] = entry_point;
+ if (ht_bin->vaddr[1] == vaddr)
+ ht_bin->tcaddr[1] = entry_point;
}
}
}
// Align code
if(((u_int)out)&7) emit_addnop(13);
#endif
- assert((u_int)out-beginning<MAX_OUTPUT_BLOCK_SIZE);
- //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
-
+ assert(out - (u_char *)beginning < MAX_OUTPUT_BLOCK_SIZE);
+ //printf("shadow buffer: %p-%p\n",copy,(u_char *)copy+slen*4);
+ memcpy(copy, source, source_len);
+ copy += source_len;
+
+ end_block(beginning);
+
// If we're within 256K of the end of the buffer,
// start over from the beginning. (Is 256K enough?)
- if((int)out>BASE_ADDR+(1<<TARGET_SIZE_2)-MAX_OUTPUT_BLOCK_SIZE) out=(u_char *)BASE_ADDR;
-
+ if (out > ndrc->translation_cache + sizeof(ndrc->translation_cache) - MAX_OUTPUT_BLOCK_SIZE)
+ out = ndrc->translation_cache;
+
// Trap writes to any of the pages we compiled
for(i=start>>12;i<=(start+slen*4)>>12;i++) {
invalid_code[i]=0;
-#ifndef DISABLE_TLB
- memory_map[i]|=0x40000000;
- if((signed int)start>=(signed int)0xC0000000) {
- assert(using_tlb);
- j=(((u_int)i<<12)+(memory_map[i]<<2)-(u_int)rdram+(u_int)0x80000000)>>12;
- invalid_code[j]=0;
- memory_map[j]|=0x40000000;
- //printf("write protect physical page: %x (virtual %x)\n",j<<12,start);
- }
-#endif
}
inv_code_start=inv_code_end=~0;
-#ifdef PCSX
- // PCSX maps all RAM mirror invalid_code tests to 0x80000000..0x80000000+RAM_SIZE
+
+ // for PCSX we need to mark all mirrors too
if(get_page(start)<(RAM_SIZE>>12))
for(i=start>>12;i<=(start+slen*4)>>12;i++)
- invalid_code[((u_int)0x80000000>>12)|i]=0;
-#endif
-
+ invalid_code[((u_int)0x00000000>>12)|(i&0x1ff)]=
+ invalid_code[((u_int)0x80000000>>12)|(i&0x1ff)]=
+ invalid_code[((u_int)0xa0000000>>12)|(i&0x1ff)]=0;
+
/* Pass 10 - Free memory by expiring oldest blocks */
-
- int end=((((int)out-BASE_ADDR)>>(TARGET_SIZE_2-16))+16384)&65535;
- while(expirep!=end)
- {
- int shift=TARGET_SIZE_2-3; // Divide into 8 blocks
- int base=BASE_ADDR+((expirep>>13)<<shift); // Base address of this block
- 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);
- break;
- case 1:
- // Clear pointers
- ll_kill_pointers(jump_out[expirep&2047],base,shift);
- ll_kill_pointers(jump_out[(expirep&2047)+2048],base,shift);
- break;
- case 2:
- // Clear hash table
- for(i=0;i<32;i++) {
- int *ht_bin=hash_table[((expirep&2047)<<5)+i];
- if((ht_bin[3]>>shift)==(base>>shift) ||
- ((ht_bin[3]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) {
- inv_debug("EXP: Remove hash %x -> %x\n",ht_bin[2],ht_bin[3]);
- ht_bin[2]=ht_bin[3]=-1;
- }
- if((ht_bin[1]>>shift)==(base>>shift) ||
- ((ht_bin[1]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) {
- inv_debug("EXP: Remove hash %x -> %x\n",ht_bin[0],ht_bin[1]);
- ht_bin[0]=ht_bin[2];
- ht_bin[1]=ht_bin[3];
- ht_bin[2]=ht_bin[3]=-1;
- }
- }
- break;
- case 3:
- // Clear jump_out
- #ifdef __arm__
- if((expirep&2047)==0)
- do_clear_cache();
- #endif
- ll_remove_matching_addrs(jump_out+(expirep&2047),base,shift);
- ll_remove_matching_addrs(jump_out+2048+(expirep&2047),base,shift);
- break;
- }
- expirep=(expirep+1)&65535;
- }
+
+ pass10_expire_blocks();
+
+#ifdef ASSEM_PRINT
+ fflush(stdout);
+#endif
return 0;
}