#ifdef _3DS
#include <3ds_utils.h>
#endif
+#ifdef HAVE_LIBNX
+#include <switch.h>
+static Jit g_jit;
+#endif
#include "new_dynarec_config.h"
#include "../psxhle.h"
#include "../psxinterpreter.h"
#include "../gte.h"
#include "emu_if.h" // emulator interface
+#include "linkage_offsets.h"
+#include "compiler_features.h"
#include "arm_features.h"
-#define noinline __attribute__((noinline,noclone))
#ifndef ARRAY_SIZE
#define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
#endif
//#define DISASM
//#define ASSEM_PRINT
+//#define REGMAP_PRINT // with DISASM only
+//#define INV_DEBUG_W
//#define STAT_PRINT
#ifdef ASSEM_PRINT
#define TC_REDUCE_BYTES 0
#endif
+struct ndrc_tramp
+{
+ struct tramp_insns ops[2048 / sizeof(struct tramp_insns)];
+ const void *f[2048 / sizeof(void *)];
+};
+
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;
+ struct ndrc_tramp tramp;
};
#ifdef BASE_ADDR_DYNAMIC
static struct ndrc_mem ndrc_ __attribute__((aligned(4096)));
static struct ndrc_mem *ndrc = &ndrc_;
#endif
+#ifdef TC_WRITE_OFFSET
+# ifdef __GLIBC__
+# include <sys/types.h>
+# include <sys/stat.h>
+# include <fcntl.h>
+# include <unistd.h>
+# endif
+static long ndrc_write_ofs;
+#define NDRC_WRITE_OFFSET(x) (void *)((char *)(x) + ndrc_write_ofs)
+#else
+#define NDRC_WRITE_OFFSET(x) (x)
+#endif
// stubs
enum stub_type {
CC_STUB = 1,
- FP_STUB = 2,
+ //FP_STUB = 2,
LOADB_STUB = 3,
LOADH_STUB = 4,
LOADW_STUB = 5,
- LOADD_STUB = 6,
+ //LOADD_STUB = 6,
LOADBU_STUB = 7,
LOADHU_STUB = 8,
STOREB_STUB = 9,
STOREH_STUB = 10,
STOREW_STUB = 11,
- STORED_STUB = 12,
+ //STORED_STUB = 12,
STORELR_STUB = 13,
INVCODE_STUB = 14,
+ OVERFLOW_STUB = 15,
+ ALIGNMENT_STUB = 16,
};
// regmap_pre[i] - regs before [i] insn starts; dirty things here that
uint64_t dirty;
uint64_t u;
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 isconst; // ... but isconst is false when r2 is known (hr)
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;
- void *addr;
- struct ll_entry *next;
+ //u_int waswritten; // MIPS regs that were used as store base before
};
struct ht_entry
u_int tc_offs;
//u_int tc_len;
u_int reg_sv_flags;
- u_short is_dirty;
+ u_char is_dirty;
+ u_char inv_near_misses;
u_short jump_in_cnt;
struct {
u_int vaddr;
} jump_in[0];
};
+struct jump_info
+{
+ int alloc;
+ int count;
+ struct {
+ u_int target_vaddr;
+ void *stub;
+ } e[0];
+};
+
static struct decoded_insn
{
u_char itype;
- u_char opcode;
- u_char opcode2;
+ u_char opcode; // bits 31-26
+ u_char opcode2; // (depends on opcode)
u_char rs1;
u_char rs2;
u_char rt1;
u_char is_ujump:1;
u_char is_load:1;
u_char is_store:1;
+ u_char is_delay_load:1; // is_load + MFC/CFC
+ u_char is_exception:1; // unconditional, also interp. fallback
+ u_char may_except:1; // might generate an exception
} dops[MAXBLOCK];
+static struct compile_info
+{
+ int imm;
+ u_int ba;
+ int ccadj;
+ signed char min_free_regs;
+ signed char addr;
+ signed char reserved[2];
+} cinfo[MAXBLOCK];
+
static u_char *out;
+ static char invalid_code[0x100000];
static struct ht_entry hash_table[65536];
static struct block_info *blocks[PAGE_COUNT];
- static struct ll_entry *jump_out[PAGE_COUNT];
+ static struct jump_info *jumps[PAGE_COUNT];
static u_int start;
static u_int *source;
static uint64_t gte_rs[MAXBLOCK]; // gte: 32 data and 32 ctl regs
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 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];
#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 AGEN1 46 // Address generation temporary register (pass5b_preallocate2)
//#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 (hi/lo only)
#define ALU 6 // Arithmetic/logic
#define MULTDIV 7 // Multiply/divide
#define SHIFT 8 // Shift by register
#define CJUMP 13 // Conditional branch (BEQ/BNE/BGTZ/BLEZ)
#define SJUMP 14 // Conditional branch (regimm format)
#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 RFE 16
#define SYSCALL 22// SYSCALL,BREAK
-#define OTHER 23 // Other
-//#define SPAN 24 // Branch/delay slot spans 2 pages
-#define NI 25 // Not implemented
+#define OTHER 23 // Other/unknown - do nothing
#define HLECALL 26// PCSX fake opcodes for HLE
#define COP2 27 // Coprocessor 2 move
#define C2LS 28 // Coprocessor 2 load/store
// asm linkage
void dyna_linker();
void cc_interrupt();
-void fp_exception();
-void fp_exception_ds();
void jump_syscall (u_int u0, u_int u1, u_int pc);
void jump_syscall_ds(u_int u0, u_int u1, u_int pc);
void jump_break (u_int u0, u_int u1, u_int pc);
void jump_break_ds(u_int u0, u_int u1, u_int pc);
+void jump_overflow (u_int u0, u_int u1, u_int pc);
+void jump_overflow_ds(u_int u0, u_int u1, u_int pc);
+void jump_addrerror (u_int cause, u_int addr, u_int pc);
+void jump_addrerror_ds(u_int cause, u_int addr, u_int pc);
void jump_to_new_pc();
void call_gteStall();
void new_dyna_leave();
void *ndrc_get_addr_ht_param(u_int vaddr, int can_compile);
void *ndrc_get_addr_ht(u_int vaddr);
-void ndrc_invalidate_addr(u_int addr);
void ndrc_add_jump_out(u_int vaddr, void *src);
+void ndrc_write_invalidate_one(u_int addr);
+static void ndrc_write_invalidate_many(u_int addr, u_int end);
static int new_recompile_block(u_int addr);
static void invalidate_block(struct block_info *block);
+static void exception_assemble(int i, const struct regstat *i_regs, int ccadj_);
// Needed by assembler
static void wb_register(signed char r, const signed char regmap[], uint64_t dirty);
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);
sceKernelCloseVMDomain();
else
sceKernelOpenVMDomain();
+ #elif defined(HAVE_LIBNX)
+ Result rc;
+ // check to avoid the full flush in jitTransitionToExecutable()
+ if (g_jit.type != JitType_CodeMemory) {
+ if (is_x)
+ rc = jitTransitionToExecutable(&g_jit);
+ else
+ rc = jitTransitionToWritable(&g_jit);
+ if (R_FAILED(rc))
+ ;//SysPrintf("jitTransition %d %08x\n", is_x, rc);
+ }
+ #elif defined(TC_WRITE_OFFSET)
+ // separated rx and rw areas are always available
#else
u_long mstart = (u_long)start & ~4095ul;
u_long mend = (u_long)end;
sceKernelSyncVMDomain(sceBlock, start, len);
#elif defined(_3DS)
ctr_flush_invalidate_cache();
+ #elif defined(HAVE_LIBNX)
+ if (g_jit.type == JitType_CodeMemory) {
+ armDCacheClean(start, len);
+ armICacheInvalidate((char *)start - ndrc_write_ofs, len);
+ // as of v4.2.1 libnx lacks isb
+ __asm__ volatile("isb" ::: "memory");
+ }
#elif defined(__aarch64__)
// as of 2021, __clear_cache() is still broken on arm64
// so here is a custom one :(
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);
+ start_tcache_write(NDRC_WRITE_OFFSET(out), NDRC_WRITE_OFFSET(end));
return out;
}
static void end_block(void *start)
{
- end_tcache_write(start, out);
+ end_tcache_write(NDRC_WRITE_OFFSET(start), NDRC_WRITE_OFFSET(out));
+}
+
+#ifdef NDRC_CACHE_FLUSH_ALL
+
+static int needs_clear_cache;
+
+static void mark_clear_cache(void *target)
+{
+ if (!needs_clear_cache) {
+ start_tcache_write(NDRC_WRITE_OFFSET(ndrc), NDRC_WRITE_OFFSET(ndrc + 1));
+ needs_clear_cache = 1;
+ }
+}
+
+static void do_clear_cache(void)
+{
+ if (needs_clear_cache) {
+ end_tcache_write(NDRC_WRITE_OFFSET(ndrc), NDRC_WRITE_OFFSET(ndrc + 1));
+ needs_clear_cache = 0;
+ }
}
+#else
+
// also takes care of w^x mappings when patching code
static u_int needs_clear_cache[1<<(TARGET_SIZE_2-17)];
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);
+ char *start = (char *)NDRC_WRITE_OFFSET((uintptr_t)target & ~4095l);
start_tcache_write(start, start + 4095);
needs_clear_cache[offset >> 17] |= mask;
}
break;
end += 4096;
}
- end_tcache_write(start, end);
+ end_tcache_write(NDRC_WRITE_OFFSET(start), NDRC_WRITE_OFFSET(end));
}
needs_clear_cache[i] = 0;
}
}
-//#define DEBUG_CYCLE_COUNT 1
+#endif // NDRC_CACHE_FLUSH_ALL
#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 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);
+ return dops[i].rs1 != 0
+ && (dops[i].rs1 == dops[i+1].rt1 || dops[i].rs1 == dops[i+1].rt2
+ || dops[i].rs1 == dops[i].rt1); // overwrites itself - same effect
}
// psx addr mirror masking (for invalidation)
static void mark_invalid_code(u_int vaddr, u_int len, char invalid)
{
+ u_int vaddr_m = vaddr & 0x1fffffff;
u_int i, j;
- vaddr &= 0x1fffffff;
- for (i = vaddr & ~0xfff; i < vaddr + len; i += 0x1000) {
+ for (i = vaddr_m & ~0xfff; i < vaddr_m + len; i += 0x1000) {
// ram mirrors, but should not hurt bios
for (j = 0; j < 0x800000; j += 0x200000) {
invalid_code[(i|j) >> 12] =
return diff > EXPIRITY_OFFSET + MAX_OUTPUT_BLOCK_SIZE;
}
+static unused void check_for_block_changes(u_int start, u_int end)
+{
+ u_int start_page = get_page_prev(start);
+ u_int end_page = get_page(end - 1);
+ u_int page;
+
+ for (page = start_page; page <= end_page; page++) {
+ struct block_info *block;
+ for (block = blocks[page]; block != NULL; block = block->next) {
+ if (block->is_dirty)
+ continue;
+ if (memcmp(block->source, block->copy, block->len)) {
+ printf("bad block %08x-%08x %016llx %016llx @%08x\n",
+ block->start, block->start + block->len,
+ *(long long *)block->source, *(long long *)block->copy, psxRegs.pc);
+ fflush(stdout);
+ abort();
+ }
+ }
+ }
+}
+
static void *try_restore_block(u_int vaddr, u_int start_page, u_int end_page)
{
void *found_clean = NULL;
if (memcmp(block->source, block->copy, block->len))
continue;
- block->is_dirty = 0;
+ block->is_dirty = block->inv_near_misses = 0;
found_clean = block->jump_in[i].addr;
hash_table_add(vaddr, found_clean);
mark_invalid_code(block->start, block->len, 0);
return ndrc_get_addr_ht(vaddr);
// generate an address error
- Status|=2;
- Cause=(vaddr<<31)|(4<<2);
- EPC=(vaddr&1)?vaddr-5:vaddr;
- BadVAddr=(vaddr&~1);
+#ifdef DRC_DBG
+ last_count -= 2;
+#endif
+ psxRegs.CP0.n.Cause &= 0x300;
+ psxRegs.CP0.n.EPC = vaddr;
+ if (vaddr & 3) {
+ psxRegs.CP0.n.Cause |= R3000E_AdEL << 2;
+ psxRegs.CP0.n.BadVAddr = vaddr;
+ } else
+ psxRegs.CP0.n.Cause |= R3000E_IBE << 2;
+ psxRegs.pc = 0x80000080;
return ndrc_get_addr_ht(0x80000080);
}
// Look up address in hash table first
void *ndrc_get_addr_ht_param(u_int vaddr, int can_compile)
{
+ //check_for_block_changes(vaddr, vaddr + MAXBLOCK);
const struct ht_entry *ht_bin = hash_table_get(vaddr);
+ u_int vaddr_a = vaddr & ~3;
stat_inc(stat_ht_lookups);
- if (ht_bin->vaddr[0] == vaddr) return ht_bin->tcaddr[0];
- if (ht_bin->vaddr[1] == vaddr) return ht_bin->tcaddr[1];
+ if (ht_bin->vaddr[0] == vaddr_a) return ht_bin->tcaddr[0];
+ if (ht_bin->vaddr[1] == vaddr_a) return ht_bin->tcaddr[1];
return get_addr(vaddr, can_compile);
}
#endif
+// get reg suitable for writing
+static signed char get_reg_w(const signed char regmap[], signed char r)
+{
+ return r == 0 ? -1 : get_reg(regmap, r);
+}
+
// get reg as mask bit (1 << hr)
static u_int get_regm(const signed char regmap[], signed char r)
{
}
if(b>=0)
{
- if(ba[i+b]>=start && ba[i+b]<(start+slen*4))
+ if(cinfo[i+b].ba>=start && cinfo[i+b].ba<(start+slen*4))
{
// Follow first branch
- int t=(ba[i+b]-start)>>2;
+ int t=(cinfo[i+b].ba-start)>>2;
j=7-b;if(t+j>=slen) j=slen-t-1;
for(;j>=0;j--)
{
if (i > 0 && dops[i-1].is_ujump)
{
- if(ba[i-1]<start || ba[i-1]>start+slen*4-4)
+ if(cinfo[i-1].ba<start || cinfo[i-1].ba>start+slen*4-4)
return 0; // Don't need any registers if exiting the block
}
for(j=0;j<9;j++)
j++;
break;
}
- if(dops[i+j].itype==SYSCALL||dops[i+j].itype==HLECALL||dops[i+j].itype==INTCALL||((source[i+j]&0xfc00003f)==0x0d))
+ if (dops[i+j].is_exception)
{
break;
}
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))
+ if(cinfo[i+k].ba>=start && cinfo[i+k].ba<(start+i*4))
{
- int t=(ba[i+k]-start)>>2;
+ int t=(cinfo[i+k].ba-start)>>2;
int reg=get_reg(regs[t].regmap_entry,r);
if(reg>=0) return reg;
//reg=get_reg(regs[t+1].regmap_entry,r);
FUNCNAME(jump_handler_write8),
FUNCNAME(jump_handler_write16),
FUNCNAME(jump_handler_write32),
- FUNCNAME(ndrc_invalidate_addr),
+ FUNCNAME(ndrc_write_invalidate_one),
+ FUNCNAME(ndrc_write_invalidate_many),
FUNCNAME(jump_to_new_pc),
FUNCNAME(jump_break),
FUNCNAME(jump_break_ds),
FUNCNAME(jump_syscall),
FUNCNAME(jump_syscall_ds),
+ FUNCNAME(jump_overflow),
+ FUNCNAME(jump_overflow_ds),
+ FUNCNAME(jump_addrerror),
+ FUNCNAME(jump_addrerror_ds),
FUNCNAME(call_gteStall),
FUNCNAME(new_dyna_leave),
FUNCNAME(pcsx_mtc0),
FUNCNAME(pcsx_mtc0_ds),
+ FUNCNAME(execI),
+#ifdef __aarch64__
+ FUNCNAME(do_memhandler_pre),
+ FUNCNAME(do_memhandler_post),
+#endif
#ifdef DRC_DBG
FUNCNAME(do_insn_cmp),
#endif
return function_names[i].name;
return "";
}
+
+static const char *fpofs_name(u_int ofs)
+{
+ u_int *p = (u_int *)&dynarec_local + ofs/sizeof(u_int);
+ static char buf[64];
+ switch (ofs) {
+ #define ofscase(x) case LO_##x: return " ; " #x
+ ofscase(next_interupt);
+ ofscase(last_count);
+ ofscase(pending_exception);
+ ofscase(stop);
+ ofscase(address);
+ ofscase(lo);
+ ofscase(hi);
+ ofscase(PC);
+ ofscase(cycle);
+ ofscase(mem_rtab);
+ ofscase(mem_wtab);
+ ofscase(psxH_ptr);
+ ofscase(invc_ptr);
+ ofscase(ram_offset);
+ #undef ofscase
+ }
+ buf[0] = 0;
+ if (psxRegs.GPR.r <= p && p < &psxRegs.GPR.r[32])
+ snprintf(buf, sizeof(buf), " ; r%d", (int)(p - psxRegs.GPR.r));
+ else if (psxRegs.CP0.r <= p && p < &psxRegs.CP0.r[32])
+ snprintf(buf, sizeof(buf), " ; cp0 $%d", (int)(p - psxRegs.CP0.r));
+ else if (psxRegs.CP2D.r <= p && p < &psxRegs.CP2D.r[32])
+ snprintf(buf, sizeof(buf), " ; cp2d $%d", (int)(p - psxRegs.CP2D.r));
+ else if (psxRegs.CP2C.r <= p && p < &psxRegs.CP2C.r[32])
+ snprintf(buf, sizeof(buf), " ; cp2c $%d", (int)(p - psxRegs.CP2C.r));
+ return buf;
+}
#else
#define func_name(x) ""
+#define fpofs_name(x) ""
#endif
#ifdef __i386__
static void *get_trampoline(const void *f)
{
+ struct ndrc_tramp *tramp = NDRC_WRITE_OFFSET(&ndrc->tramp);
size_t i;
- for (i = 0; i < ARRAY_SIZE(ndrc->tramp.f); i++) {
- if (ndrc->tramp.f[i] == f || ndrc->tramp.f[i] == NULL)
+ for (i = 0; i < ARRAY_SIZE(tramp->f); i++) {
+ if (tramp->f[i] == f || tramp->f[i] == NULL)
break;
}
- if (i == ARRAY_SIZE(ndrc->tramp.f)) {
+ if (i == ARRAY_SIZE(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]);
+ if (tramp->f[i] == NULL) {
+ start_tcache_write(&tramp->f[i], &tramp->f[i + 1]);
+ tramp->f[i] = f;
+ end_tcache_write(&tramp->f[i], &tramp->f[i + 1]);
+#ifdef HAVE_LIBNX
+ // invalidate the RX mirror (unsure if necessary, but just in case...)
+ armDCacheFlush(&ndrc->tramp.f[i], sizeof(ndrc->tramp.f[i]));
+#endif
}
return &ndrc->tramp.ops[i];
}
emit_call(f);
}
-// Add virtual address mapping to linked list
-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->addr=addr;
- new_entry->next=*head;
- *head=new_entry;
-}
-
// Check if an address is already compiled
// but don't return addresses which are about to expire from the cache
static void *check_addr(u_int vaddr)
return NULL;
}
-static void ll_remove_matching_addrs(struct ll_entry **head, u_int base_offs, int shift)
-{
- struct ll_entry *next;
- while (*head) {
- u_int tc_offs = (u_char *)((*head)->addr) - ndrc->translation_cache;
- if (((tc_offs ^ base_offs) >> shift) == 0) {
- inv_debug("EXP: rm link from tc_offs %x)\n", tc_offs);
- next = (*head)->next;
- free(*head);
- *head = next;
- }
- else
- {
- head = &((*head)->next);
- }
- }
-}
-
-// Remove all entries from linked list
-static void ll_clear(struct ll_entry **head)
-{
- struct ll_entry *cur;
- struct ll_entry *next;
- if((cur=*head)) {
- *head=0;
- while(cur) {
- next=cur->next;
- free(cur);
- cur=next;
- }
- }
-}
-
static void blocks_clear(struct block_info **head)
{
struct block_info *cur, *next;
int hit = 0;
while (*head) {
if ((((*head)->tc_offs ^ base_offs) >> shift) == 0) {
- inv_debug("EXP: rm block %08x (tc_offs %zx)\n", (*head)->start, (*head)->tc_offs);
+ inv_debug("EXP: rm block %08x (tc_offs %x)\n", (*head)->start, (*head)->tc_offs);
invalidate_block(*head);
next = (*head)->next;
free(*head);
}
// This is called when we write to a compiled block (see do_invstub)
-static void unlink_jumps_range(u_int start, u_int end)
+static void unlink_jumps_vaddr_range(u_int start, u_int end)
{
u_int page, start_page = get_page(start), end_page = get_page(end - 1);
- struct ll_entry **head, *next;
+ int i;
for (page = start_page; page <= end_page; page++) {
- for (head = &jump_out[page]; *head; ) {
- if ((*head)->vaddr < start || (*head)->vaddr >= end) {
- head = &((*head)->next);
+ struct jump_info *ji = jumps[page];
+ if (ji == NULL)
+ continue;
+ for (i = 0; i < ji->count; ) {
+ if (ji->e[i].target_vaddr < start || ji->e[i].target_vaddr >= end) {
+ i++;
continue;
}
- inv_debug("INV: rm link to %08x (tc_offs %zx)\n",
- (*head)->vaddr, (u_char *)((*head)->addr) - ndrc->translation_cache);
- void *host_addr = find_extjump_insn((*head)->addr);
+
+ inv_debug("INV: rm link to %08x (tc_offs %zx)\n", ji->e[i].target_vaddr,
+ (u_char *)ji->e[i].stub - ndrc->translation_cache);
+ void *host_addr = find_extjump_insn(ji->e[i].stub);
mark_clear_cache(host_addr);
- set_jump_target(host_addr, (*head)->addr); // point back to dyna_linker stub
+ set_jump_target(host_addr, ji->e[i].stub); // point back to dyna_linker stub
- next = (*head)->next;
- free(*head);
- *head = next;
stat_dec(stat_links);
+ ji->count--;
+ if (i < ji->count) {
+ ji->e[i] = ji->e[ji->count];
+ continue;
+ }
+ i++;
}
}
}
+static void unlink_jumps_tc_range(struct jump_info *ji, u_int base_offs, int shift)
+{
+ int i;
+ if (ji == NULL)
+ return;
+ for (i = 0; i < ji->count; ) {
+ u_int tc_offs = (u_char *)ji->e[i].stub - ndrc->translation_cache;
+ if (((tc_offs ^ base_offs) >> shift) != 0) {
+ i++;
+ continue;
+ }
+
+ inv_debug("EXP: rm link to %08x (tc_offs %x)\n", ji->e[i].target_vaddr, tc_offs);
+ stat_dec(stat_links);
+ ji->count--;
+ if (i < ji->count) {
+ ji->e[i] = ji->e[ji->count];
+ continue;
+ }
+ i++;
+ }
+}
+
static void invalidate_block(struct block_info *block)
{
u_int i;
block->is_dirty = 1;
- unlink_jumps_range(block->start, block->start + block->len);
+ unlink_jumps_vaddr_range(block->start, block->start + block->len);
for (i = 0; i < block->jump_in_cnt; i++)
hash_table_remove(block->jump_in[i].vaddr);
}
static int invalidate_range(u_int start, u_int end,
u32 *inv_start_ret, u32 *inv_end_ret)
{
+ struct block_info *last_block = NULL;
u_int start_page = get_page_prev(start);
u_int end_page = get_page(end - 1);
u_int start_m = pmmask(start);
- u_int end_m = pmmask(end);
+ u_int end_m = pmmask(end - 1);
u_int inv_start, inv_end;
u_int blk_start_m, blk_end_m;
u_int page;
int hit = 0;
// additional area without code (to supplement invalid_code[]), [start, end)
- // avoids excessive ndrc_invalidate_addr() calls
+ // avoids excessive ndrc_write_invalidate*() calls
inv_start = start_m & ~0xfff;
inv_end = end_m | 0xfff;
for (block = blocks[page]; block != NULL; block = block->next) {
if (block->is_dirty)
continue;
+ last_block = block;
blk_end_m = pmmask(block->start + block->len);
if (blk_end_m <= start_m) {
inv_start = max(inv_start, blk_end_m);
}
}
+ if (!hit && last_block && last_block->source) {
+ // could be some leftover unused block, uselessly trapping writes
+ last_block->inv_near_misses++;
+ if (last_block->inv_near_misses > 128) {
+ invalidate_block(last_block);
+ stat_inc(stat_inv_hits);
+ hit++;
+ }
+ }
if (hit) {
do_clear_cache();
#ifdef USE_MINI_HT
memset(mini_ht, -1, sizeof(mini_ht));
#endif
}
+
if (inv_start <= (start_m & ~0xfff) && inv_end >= (start_m | 0xfff))
// the whole page is empty now
mark_invalid_code(start, 1, 1);
invalidate_range(start, end, NULL, NULL);
}
-void ndrc_invalidate_addr(u_int addr)
+static void ndrc_write_invalidate_many(u_int start, u_int end)
{
// this check is done by the caller
//if (inv_code_start<=addr&&addr<=inv_code_end) { rhits++; return; }
- int ret = invalidate_range(addr, addr + 4, &inv_code_start, &inv_code_end);
+ int ret = invalidate_range(start, end, &inv_code_start, &inv_code_end);
+#ifdef INV_DEBUG_W
+ int invc = invalid_code[start >> 12];
+ u_int len = end - start;
if (ret)
- inv_debug("INV ADDR: %08x hit %d blocks\n", addr, ret);
+ printf("INV ADDR: %08x/%02x hit %d blocks\n", start, len, ret);
else
- inv_debug("INV ADDR: %08x miss, inv %08x-%08x\n", addr, inv_code_start, inv_code_end);
+ printf("INV ADDR: %08x/%02x miss, inv %08x-%08x invc %d->%d\n", start, len,
+ inv_code_start, inv_code_end, invc, invalid_code[start >> 12]);
+ check_for_block_changes(start, end);
+#endif
stat_inc(stat_inv_addr_calls);
+ (void)ret;
+}
+
+void ndrc_write_invalidate_one(u_int addr)
+{
+ ndrc_write_invalidate_many(addr, addr + 4);
}
// This is called when loading a save state.
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_readword(&inv_code_start, 1);
- emit_readword(&inv_code_end, 2);
- emit_cmp(0, 1);
- emit_cmpcs(2, 0);
- void *jaddr = out;
- emit_jc(0);
- emit_far_call(ndrc_invalidate_addr);
- set_jump_target(jaddr, out);
- restore_regs(reglist);
- emit_jmp(stubs[n].retaddr); // return address
-}
-
// Add an entry to jump_out after making a link
// src should point to code by emit_extjump()
-void ndrc_add_jump_out(u_int vaddr,void *src)
+void ndrc_add_jump_out(u_int vaddr, void *src)
{
- u_int page=get_page(vaddr);
- inv_debug("ndrc_add_jump_out: %p -> %x (%d)\n",src,vaddr,page);
- check_extjump2(src);
- ll_add(jump_out+page,vaddr,src);
- //inv_debug("ndrc_add_jump_out: to %p\n",get_pointer(src));
+ inv_debug("ndrc_add_jump_out: %p -> %x\n", src, vaddr);
+ u_int page = get_page(vaddr);
+ struct jump_info *ji;
+
stat_inc(stat_links);
+ check_extjump2(src);
+ ji = jumps[page];
+ if (ji == NULL) {
+ ji = malloc(sizeof(*ji) + sizeof(ji->e[0]) * 16);
+ ji->alloc = 16;
+ ji->count = 0;
+ }
+ else if (ji->count >= ji->alloc) {
+ ji->alloc += 16;
+ ji = realloc(ji, sizeof(*ji) + sizeof(ji->e[0]) * ji->alloc);
+ }
+ jumps[page] = ji;
+ ji->e[ji->count].target_vaddr = vaddr;
+ ji->e[ji->count].stub = src;
+ ji->count++;
}
/* Register allocation */
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]);
+ if(dops[i].opcode2==0x00) set_const(current,dops[i].rt1,v<<cinfo[i].imm);
+ if(dops[i].opcode2==0x02) set_const(current,dops[i].rt1,(u_int)v>>cinfo[i].imm);
+ if(dops[i].opcode2==0x03) set_const(current,dops[i].rt1,v>>cinfo[i].imm);
}
else clear_const(current,dops[i].rt1);
}
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;
+ cinfo[i].min_free_regs=1;
}
} else { // DSLLV/DSRLV/DSRAV
assert(0);
}
alloc_reg(current,i,dops[i].rt1);
}
+ if (dops[i].may_except) {
+ alloc_cc(current, i); // for exceptions
+ alloc_reg_temp(current, i, -1);
+ cinfo[i].min_free_regs = 1;
+ }
}
- if(dops[i].opcode2==0x2a||dops[i].opcode2==0x2b) { // SLT/SLTU
+ else if(dops[i].opcode2==0x2a||dops[i].opcode2==0x2b) { // SLT/SLTU
if(dops[i].rt1) {
alloc_reg(current,i,dops[i].rs1);
alloc_reg(current,i,dops[i].rs2);
alloc_reg(current,i,dops[i].rt1);
}
}
- if(dops[i].opcode2>=0x24&&dops[i].opcode2<=0x27) { // AND/OR/XOR/NOR
+ else 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].rt1);
}
}
- if(dops[i].opcode2>=0x2c&&dops[i].opcode2<=0x2f) { // DADD/DADDU/DSUB/DSUBU
- assert(0);
- }
clear_const(current,dops[i].rs1);
clear_const(current,dops[i].rs2);
clear_const(current,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;
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(dops[i].opcode==0x0a||dops[i].opcode==0x0b) { // SLTI/SLTIU
+ 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(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]);
+ if(dops[i].opcode==0x0c) set_const(current,dops[i].rt1,v&cinfo[i].imm);
+ if(dops[i].opcode==0x0d) set_const(current,dops[i].rt1,v|cinfo[i].imm);
+ if(dops[i].opcode==0x0e) set_const(current,dops[i].rt1,v^cinfo[i].imm);
}
else clear_const(current,dops[i].rt1);
}
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]);
+ set_const(current,dops[i].rt1,v+cinfo[i].imm);
}
else clear_const(current,dops[i].rt1);
+ if (dops[i].may_except) {
+ alloc_cc(current, i); // for exceptions
+ alloc_reg_temp(current, i, -1);
+ cinfo[i].min_free_regs = 1;
+ }
}
else {
- set_const(current,dops[i].rt1,imm[i]<<16); // LUI
+ set_const(current,dops[i].rt1,cinfo[i].imm<<16); // LUI
}
dirty_reg(current,dops[i].rt1);
}
static void load_alloc(struct regstat *current,int i)
{
+ int need_temp = 0;
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
alloc_reg(current, i, dops[i].rs1);
if (ram_offset)
alloc_reg(current, i, ROREG);
+ if (dops[i].may_except) {
+ alloc_cc(current, i); // for exceptions
+ dirty_reg(current, CCREG);
+ need_temp = 1;
+ }
if(dops[i].rt1&&!((current->u>>dops[i].rt1)&1)) {
alloc_reg(current,i,dops[i].rt1);
- assert(get_reg(current->regmap,dops[i].rt1)>=0);
- if(dops[i].opcode==0x27||dops[i].opcode==0x37) // LWU/LD
- {
- assert(0);
- }
- else if(dops[i].opcode==0x1A||dops[i].opcode==0x1B) // LDL/LDR
- {
- assert(0);
- }
+ assert(get_reg_w(current->regmap, dops[i].rt1)>=0);
dirty_reg(current,dops[i].rt1);
// LWL/LWR need a temporary register for the old value
if(dops[i].opcode==0x22||dops[i].opcode==0x26)
{
alloc_reg(current,i,FTEMP);
- alloc_reg_temp(current,i,-1);
- minimum_free_regs[i]=1;
+ need_temp = 1;
}
}
else
// Load to r0 or unneeded register (dummy load)
// but we still need a register to calculate the address
if(dops[i].opcode==0x22||dops[i].opcode==0x26)
- {
alloc_reg(current,i,FTEMP); // LWL/LWR need another temporary
- }
- alloc_reg_temp(current,i,-1);
- minimum_free_regs[i]=1;
- if(dops[i].opcode==0x1A||dops[i].opcode==0x1B) // LDL/LDR
- {
- assert(0);
- }
+ need_temp = 1;
+ }
+ if (need_temp) {
+ alloc_reg_temp(current, i, -1);
+ cinfo[i].min_free_regs = 1;
}
}
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
alloc_reg(current, i, INVCP);
#endif
- if(dops[i].opcode==0x2a||dops[i].opcode==0x2e||dops[i].opcode==0x2c||dops[i].opcode==0x2d) { // SWL/SWL/SDL/SDR
+ if (dops[i].opcode == 0x2a || dops[i].opcode == 0x2e) { // SWL/SWL
alloc_reg(current,i,FTEMP);
}
+ if (dops[i].may_except) {
+ alloc_cc(current, i); // for exceptions
+ dirty_reg(current, CCREG);
+ }
// We need a temporary register for address generation
alloc_reg_temp(current,i,-1);
- minimum_free_regs[i]=1;
-}
-
-static void c1ls_alloc(struct regstat *current,int i)
-{
- clear_const(current,dops[i].rt1);
- alloc_reg(current,i,CSREG); // Status
+ cinfo[i].min_free_regs=1;
}
static void c2ls_alloc(struct regstat *current,int i)
if (dops[i].opcode == 0x3a) // SWC2
alloc_reg(current,i,INVCP);
#endif
+ if (dops[i].may_except) {
+ alloc_cc(current, i); // for exceptions
+ dirty_reg(current, CCREG);
+ }
// We need a temporary register for address generation
alloc_reg_temp(current,i,-1);
- minimum_free_regs[i]=1;
+ cinfo[i].min_free_regs=1;
}
#ifndef multdiv_alloc
{
if(dops[i].rt1) {
clear_const(current,dops[i].rt1);
- alloc_all(current,i);
alloc_reg(current,i,dops[i].rt1);
dirty_reg(current,dops[i].rt1);
}
current->u&=~1LL;
alloc_reg(current,i,0);
}
+ cinfo[i].min_free_regs = HOST_REGS;
}
- else
- {
- // TLBR/TLBWI/TLBWR/TLBP/ERET
- assert(dops[i].opcode2==0x10);
- alloc_all(current,i);
- }
- minimum_free_regs[i]=HOST_REGS;
+}
+
+static void rfe_alloc(struct regstat *current, int i)
+{
+ alloc_all(current, i);
+ cinfo[i].min_free_regs = HOST_REGS;
}
static void cop2_alloc(struct regstat *current,int i)
}
}
alloc_reg_temp(current,i,-1);
- minimum_free_regs[i]=1;
+ cinfo[i].min_free_regs=1;
}
static void c2op_alloc(struct regstat *current,int i)
alloc_cc(current,i);
dirty_reg(current,CCREG);
alloc_all(current,i);
- minimum_free_regs[i]=HOST_REGS;
+ cinfo[i].min_free_regs=HOST_REGS;
current->isconst=0;
}
case COP0:
cop0_alloc(current,i);
break;
- case COP1:
+ case RFE:
+ rfe_alloc(current,i);
break;
case COP2:
cop2_alloc(current,i);
break;
- case C1LS:
- c1ls_alloc(current,i);
- break;
case C2LS:
c2ls_alloc(current,i);
break;
}
}
-static void alu_assemble(int i, const struct regstat *i_regs)
+static void alu_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
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,dops[i].rt1);
- if(t>=0) {
- 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) {
+ int do_oflow = dops[i].may_except; // ADD/SUB with exceptions enabled
+ if (dops[i].rt1 || do_oflow) {
+ int do_exception_check = 0;
+ signed char s1, s2, t, tmp;
+ t = get_reg_w(i_regs->regmap, dops[i].rt1);
+ tmp = get_reg_temp(i_regs->regmap);
+ if (do_oflow)
+ assert(tmp >= 0);
+ //if (t < 0 && do_oflow) // broken s2
+ // t = tmp;
+ if (t >= 0) {
+ 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(dops[i].opcode2&2) emit_sub(s1,s2,t);
- else emit_add(s1,s2,t);
+ if (dops[i].opcode2 & 2) {
+ if (do_oflow) {
+ emit_subs(s1, s2, tmp);
+ do_exception_check = 1;
+ }
+ else
+ emit_sub(s1,s2,t);
+ }
+ else {
+ if (do_oflow) {
+ emit_adds(s1, s2, tmp);
+ do_exception_check = 1;
+ }
+ else
+ emit_add(s1,s2,t);
+ }
}
else if(dops[i].rs1) {
if(s1>=0) emit_mov(s1,t);
else emit_loadreg(dops[i].rs1,t);
}
else if(dops[i].rs2) {
- if(s2>=0) {
- if(dops[i].opcode2&2) emit_neg(s2,t);
- else emit_mov(s2,t);
+ if (s2 < 0) {
+ emit_loadreg(dops[i].rs2, t);
+ s2 = t;
}
- else {
- emit_loadreg(dops[i].rs2,t);
- if(dops[i].opcode2&2) emit_neg(t,t);
+ if (dops[i].opcode2 & 2) {
+ if (do_oflow) {
+ emit_negs(s2, tmp);
+ do_exception_check = 1;
+ }
+ else
+ emit_neg(s2, t);
}
+ else if (s2 != t)
+ emit_mov(s2, t);
}
- else emit_zeroreg(t);
+ else
+ emit_zeroreg(t);
+ }
+ if (do_exception_check) {
+ void *jaddr = out;
+ emit_jo(0);
+ if (t >= 0 && tmp != t)
+ emit_mov(tmp, t);
+ add_stub_r(OVERFLOW_STUB, jaddr, out, i, 0, i_regs, ccadj_, 0);
}
}
}
- if(dops[i].opcode2>=0x2c&&dops[i].opcode2<=0x2f) { // DADD/DADDU/DSUB/DSUBU
- assert(0);
- }
- if(dops[i].opcode2==0x2a||dops[i].opcode2==0x2b) { // SLT/SLTU
+ else 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,dops[i].rt1);
+ t=get_reg_w(i_regs->regmap, dops[i].rt1);
//assert(t>=0);
if(t>=0) {
s1l=get_reg(i_regs->regmap,dops[i].rs1);
}
}
}
- if(dops[i].opcode2>=0x24&&dops[i].opcode2<=0x27) { // AND/OR/XOR/NOR
+ else 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);
+ tl=get_reg_w(i_regs->regmap, dops[i].rt1);
{
if(tl>=0) {
s1l=get_reg(i_regs->regmap,dops[i].rs1);
}
}
-static void imm16_assemble(int i, const struct regstat *i_regs)
+static void imm16_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
if (dops[i].opcode==0x0f) { // LUI
if(dops[i].rt1) {
signed char t;
- t=get_reg(i_regs->regmap,dops[i].rt1);
+ t=get_reg_w(i_regs->regmap, dops[i].rt1);
//assert(t>=0);
if(t>=0) {
if(!((i_regs->isconst>>t)&1))
- emit_movimm(imm[i]<<16,t);
+ emit_movimm(cinfo[i].imm<<16,t);
}
}
}
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,dops[i].rt1);
+ int is_addi = dops[i].may_except;
+ if (dops[i].rt1 || is_addi) {
+ signed char s, t, tmp;
+ t=get_reg_w(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);
+ tmp = get_reg_temp(i_regs->regmap);
+ if (is_addi) {
+ assert(tmp >= 0);
+ if (t < 0) t = tmp;
+ }
if(t>=0) {
if(!((i_regs->isconst>>t)&1)) {
- if(s<0) {
+ int sum, do_exception_check = 0;
+ if (s < 0) {
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))
- emit_addimm(s,imm[i],t);
+ if (is_addi) {
+ emit_addimm_and_set_flags3(t, cinfo[i].imm, tmp);
+ do_exception_check = 1;
+ }
else
- emit_movimm(constmap[i][s]+imm[i],t);
+ emit_addimm(t, cinfo[i].imm, t);
+ } else {
+ if (!((i_regs->wasconst >> s) & 1)) {
+ if (is_addi) {
+ emit_addimm_and_set_flags3(s, cinfo[i].imm, tmp);
+ do_exception_check = 1;
+ }
+ else
+ emit_addimm(s, cinfo[i].imm, t);
+ }
+ else {
+ int oflow = add_overflow(constmap[i][s], cinfo[i].imm, sum);
+ if (is_addi && oflow)
+ do_exception_check = 2;
+ else
+ emit_movimm(sum, t);
+ }
+ }
+ if (do_exception_check) {
+ void *jaddr = out;
+ if (do_exception_check == 2)
+ emit_jmp(0);
+ else {
+ emit_jo(0);
+ if (tmp != t)
+ emit_mov(tmp, t);
+ }
+ add_stub_r(OVERFLOW_STUB, jaddr, out, i, 0, i_regs, ccadj_, 0);
}
}
}
} else {
if(t>=0) {
if(!((i_regs->isconst>>t)&1))
- emit_movimm(imm[i],t);
- }
- }
- }
- }
- 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(dops[i].rs1) {
- assert(sl>=0);
- emit_addimm(sl,imm[i],tl);
- } else {
- emit_movimm(imm[i],tl);
+ emit_movimm(cinfo[i].imm,t);
}
}
}
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);
+ t=get_reg_w(i_regs->regmap, dops[i].rt1);
sl=get_reg(i_regs->regmap,dops[i].rs1);
//assert(t>=0);
if(t>=0) {
if(dops[i].opcode==0x0a) { // SLTI
if(sl<0) {
if(i_regs->regmap_entry[t]!=dops[i].rs1) emit_loadreg(dops[i].rs1,t);
- emit_slti32(t,imm[i],t);
+ emit_slti32(t,cinfo[i].imm,t);
}else{
- emit_slti32(sl,imm[i],t);
+ emit_slti32(sl,cinfo[i].imm,t);
}
}
else { // SLTIU
if(sl<0) {
if(i_regs->regmap_entry[t]!=dops[i].rs1) emit_loadreg(dops[i].rs1,t);
- emit_sltiu32(t,imm[i],t);
+ emit_sltiu32(t,cinfo[i].imm,t);
}else{
- emit_sltiu32(sl,imm[i],t);
+ emit_sltiu32(sl,cinfo[i].imm,t);
}
}
}else{
// SLTI(U) with r0 is just stupid,
// nonetheless examples can be found
if(dops[i].opcode==0x0a) // SLTI
- if(0<imm[i]) emit_movimm(1,t);
+ if(0<cinfo[i].imm) emit_movimm(1,t);
else emit_zeroreg(t);
else // SLTIU
{
- if(imm[i]) emit_movimm(1,t);
+ if(cinfo[i].imm) emit_movimm(1,t);
else emit_zeroreg(t);
}
}
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);
+ tl=get_reg_w(i_regs->regmap, dops[i].rt1);
sl=get_reg(i_regs->regmap,dops[i].rs1);
if(tl>=0 && !((i_regs->isconst>>tl)&1)) {
if(dops[i].opcode==0x0c) //ANDI
if(dops[i].rs1) {
if(sl<0) {
if(i_regs->regmap_entry[tl]!=dops[i].rs1) emit_loadreg(dops[i].rs1,tl);
- emit_andimm(tl,imm[i],tl);
+ emit_andimm(tl,cinfo[i].imm,tl);
}else{
if(!((i_regs->wasconst>>sl)&1))
- emit_andimm(sl,imm[i],tl);
+ emit_andimm(sl,cinfo[i].imm,tl);
else
- emit_movimm(constmap[i][sl]&imm[i],tl);
+ emit_movimm(constmap[i][sl]&cinfo[i].imm,tl);
}
}
else
}
if(dops[i].opcode==0x0d) { // ORI
if(sl<0) {
- emit_orimm(tl,imm[i],tl);
+ emit_orimm(tl,cinfo[i].imm,tl);
}else{
if(!((i_regs->wasconst>>sl)&1))
- emit_orimm(sl,imm[i],tl);
+ emit_orimm(sl,cinfo[i].imm,tl);
else
- emit_movimm(constmap[i][sl]|imm[i],tl);
+ emit_movimm(constmap[i][sl]|cinfo[i].imm,tl);
}
}
if(dops[i].opcode==0x0e) { // XORI
if(sl<0) {
- emit_xorimm(tl,imm[i],tl);
+ emit_xorimm(tl,cinfo[i].imm,tl);
}else{
if(!((i_regs->wasconst>>sl)&1))
- emit_xorimm(sl,imm[i],tl);
+ emit_xorimm(sl,cinfo[i].imm,tl);
else
- emit_movimm(constmap[i][sl]^imm[i],tl);
+ emit_movimm(constmap[i][sl]^cinfo[i].imm,tl);
}
}
}
else {
- emit_movimm(imm[i],tl);
+ emit_movimm(cinfo[i].imm,tl);
}
}
}
{
if(dops[i].rt1) {
signed char s,t;
- t=get_reg(i_regs->regmap,dops[i].rt1);
+ t=get_reg_w(i_regs->regmap, dops[i].rt1);
s=get_reg(i_regs->regmap,dops[i].rs1);
//assert(t>=0);
if(t>=0&&!((i_regs->isconst>>t)&1)){
else
{
if(s<0&&i_regs->regmap_entry[t]!=dops[i].rs1) emit_loadreg(dops[i].rs1,t);
- if(imm[i]) {
+ if(cinfo[i].imm) {
if(dops[i].opcode2==0) // SLL
{
- emit_shlimm(s<0?t:s,imm[i],t);
+ emit_shlimm(s<0?t:s,cinfo[i].imm,t);
}
if(dops[i].opcode2==2) // SRL
{
- emit_shrimm(s<0?t:s,imm[i],t);
+ emit_shrimm(s<0?t:s,cinfo[i].imm,t);
}
if(dops[i].opcode2==3) // SRA
{
- emit_sarimm(s<0?t:s,imm[i],t);
+ emit_sarimm(s<0?t:s,cinfo[i].imm,t);
}
}else{
// Shift by zero
}
static void *emit_fastpath_cmp_jump(int i, const struct regstat *i_regs,
- int addr, int *offset_reg, int *addr_reg_override)
+ int addr, int *offset_reg, int *addr_reg_override, int ccadj_)
{
void *jaddr = NULL;
int type = 0;
int mr = dops[i].rs1;
+ assert(addr >= 0);
*offset_reg = -1;
if(((smrv_strong|smrv_weak)>>mr)&1) {
type=get_ptr_mem_type(smrv[mr]);
//printf("set nospec @%08x r%d %d\n", start+i*4, mr, type);
}
+ if (dops[i].may_except) {
+ // alignment check
+ u_int op = dops[i].opcode;
+ int mask = ((op & 0x37) == 0x21 || op == 0x25) ? 1 : 3; // LH/SH/LHU
+ void *jaddr;
+ emit_testimm(addr, mask);
+ jaddr = out;
+ emit_jne(0);
+ add_stub_r(ALIGNMENT_STUB, jaddr, out, i, addr, i_regs, ccadj_, 0);
+ }
+
if(type==MTYPE_8020) { // RAM 80200000+ mirror
host_tempreg_acquire();
emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
static void load_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
- int s,tl,addr;
+ int addr = cinfo[i].addr;
+ int s,tl;
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);
+ tl=get_reg_w(i_regs->regmap, dops[i].rt1);
s=get_reg(i_regs->regmap,dops[i].rs1);
- offset=imm[i];
+ offset=cinfo[i].imm;
if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
if(s>=0) {
c=(i_regs->wasconst>>s)&1;
}
//printf("load_assemble: c=%d\n",c);
//if(c) printf("load_assemble: const=%lx\n",(long)constmap[i][s]+offset);
- // FIXME: Even if the load is a NOP, we should check for pagefaults...
- if((tl<0&&(!c||(((u_int)constmap[i][s]+offset)>>16)==0x1f80))
- ||dops[i].rt1==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_temp(i_regs->regmap);
+ tl = get_reg_temp(i_regs->regmap); // may be == addr
assert(tl>=0);
}
- if(offset||s<0||c) addr=tl;
- else addr=s;
- //if(tl<0) tl=get_reg_temp(i_regs->regmap);
+ assert(addr >= 0);
if(tl>=0) {
//printf("load_assemble: c=%d\n",c);
//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(!c) {
#ifdef R29_HACK
#endif
{
jaddr = emit_fastpath_cmp_jump(i, i_regs, addr,
- &offset_reg, &fastio_reg_override);
+ &offset_reg, &fastio_reg_override, ccadj_);
}
}
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
+ int dummy=(dops[i].rt1==0)||(tl!=get_reg_w(i_regs->regmap, dops[i].rt1)); // ignore loads to r0 and unneeded reg
switch (dops[i].opcode) {
case 0x20: // LB
if(!c||memtarget) {
if(!dummy) {
- int a = tl;
- if (!c) a = addr;
+ int a = addr;
if (fastio_reg_override >= 0)
a = fastio_reg_override;
case 0x21: // LH
if(!c||memtarget) {
if(!dummy) {
- int a = tl;
- if (!c) a = addr;
+ int a = addr;
if (fastio_reg_override >= 0)
a = fastio_reg_override;
if (offset_reg >= 0)
case 0x24: // LBU
if(!c||memtarget) {
if(!dummy) {
- int a = tl;
- if (!c) a = addr;
+ int a = addr;
if (fastio_reg_override >= 0)
a = fastio_reg_override;
case 0x25: // LHU
if(!c||memtarget) {
if(!dummy) {
- int a = tl;
- if(!c) a = addr;
+ int a = addr;
if (fastio_reg_override >= 0)
a = fastio_reg_override;
if (offset_reg >= 0)
else
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);
}
- }
+ } // tl >= 0
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 addr = cinfo[i].addr;
+ int s,tl,temp,temp2;
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);
+ tl=get_reg_w(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];
+ offset=cinfo[i].imm;
reglist|=1<<temp;
- if(offset||s<0||c) addr=temp2;
- else addr=s;
+ assert(addr >= 0);
if(s>=0) {
c=(i_regs->wasconst>>s)&1;
if(c) {
emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR
}
jaddr = emit_fastpath_cmp_jump(i, i_regs, temp2,
- &offset_reg, &fastio_reg_override);
+ &offset_reg, &fastio_reg_override, ccadj_);
}
else {
if (ram_offset && memtarget) {
}
#endif
+static void do_invstub(int n)
+{
+ literal_pool(20);
+ assem_debug("do_invstub\n");
+ u_int reglist = stubs[n].a;
+ u_int addrr = stubs[n].b;
+ int ofs_start = stubs[n].c;
+ int ofs_end = stubs[n].d;
+ int len = ofs_end - ofs_start;
+ u_int rightr = 0;
+
+ set_jump_target(stubs[n].addr, out);
+ save_regs(reglist);
+ if (addrr != 0 || ofs_start != 0)
+ emit_addimm(addrr, ofs_start, 0);
+ emit_readword(&inv_code_start, 2);
+ emit_readword(&inv_code_end, 3);
+ if (len != 0)
+ emit_addimm(0, len + 4, (rightr = 1));
+ emit_cmp(0, 2);
+ emit_cmpcs(3, rightr);
+ void *jaddr = out;
+ emit_jc(0);
+ void *func = (len != 0)
+ ? (void *)ndrc_write_invalidate_many
+ : (void *)ndrc_write_invalidate_one;
+ emit_far_call(func);
+ set_jump_target(jaddr, out);
+ restore_regs(reglist);
+ emit_jmp(stubs[n].retaddr);
+}
+
+static void do_store_smc_check(int i, const struct regstat *i_regs, u_int reglist, int addr)
+{
+ if (HACK_ENABLED(NDHACK_NO_SMC_CHECK))
+ return;
+ // this can't be used any more since we started to check exact
+ // block boundaries in invalidate_range()
+ //if (i_regs->waswritten & (1<<dops[i].rs1))
+ // return;
+ // (naively) assume nobody will run code from stack
+ if (dops[i].rs1 == 29)
+ return;
+
+ int j, imm_maxdiff = 32, imm_min = cinfo[i].imm, imm_max = cinfo[i].imm, count = 1;
+ if (i < slen - 1 && dops[i+1].is_store && dops[i+1].rs1 == dops[i].rs1
+ && abs(cinfo[i+1].imm - cinfo[i].imm) <= imm_maxdiff)
+ return;
+ for (j = i - 1; j >= 0; j--) {
+ if (!dops[j].is_store || dops[j].rs1 != dops[i].rs1
+ || abs(cinfo[j].imm - cinfo[j+1].imm) > imm_maxdiff)
+ break;
+ count++;
+ if (imm_min > cinfo[j].imm)
+ imm_min = cinfo[j].imm;
+ if (imm_max < cinfo[j].imm)
+ imm_max = cinfo[j].imm;
+ }
+#if defined(HOST_IMM8)
+ int ir = get_reg(i_regs->regmap, INVCP);
+ assert(ir >= 0);
+ host_tempreg_acquire();
+ emit_ldrb_indexedsr12_reg(ir, addr, HOST_TEMPREG);
+#else
+ emit_cmpmem_indexedsr12_imm(invalid_code, addr, 1);
+ #error not handled
+#endif
+#ifdef INVALIDATE_USE_COND_CALL
+ if (count == 1) {
+ emit_cmpimm(HOST_TEMPREG, 1);
+ emit_callne(invalidate_addr_reg[addr]);
+ host_tempreg_release();
+ return;
+ }
+#endif
+ void *jaddr = emit_cbz(HOST_TEMPREG, 0);
+ host_tempreg_release();
+ imm_min -= cinfo[i].imm;
+ imm_max -= cinfo[i].imm;
+ add_stub(INVCODE_STUB, jaddr, out, reglist|(1<<HOST_CCREG),
+ addr, imm_min, imm_max, 0);
+}
+
static void store_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
int s,tl;
- int addr,temp;
+ int addr = cinfo[i].addr;
int offset;
void *jaddr=0;
enum stub_type type=0;
int memtarget=0,c=0;
- int agr=AGEN1+(i&1);
int offset_reg = -1;
int fastio_reg_override = -1;
u_int reglist=get_host_reglist(i_regs->regmap);
tl=get_reg(i_regs->regmap,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_temp(i_regs->regmap);
- offset=imm[i];
+ offset=cinfo[i].imm;
if(s>=0) {
c=(i_regs->wasconst>>s)&1;
if(c) {
}
}
assert(tl>=0);
- assert(temp>=0);
+ assert(addr >= 0);
if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
- if(offset||s<0||c) addr=temp;
- else addr=s;
if (!c) {
jaddr = emit_fastpath_cmp_jump(i, i_regs, addr,
- &offset_reg, &fastio_reg_override);
+ &offset_reg, &fastio_reg_override, ccadj_);
}
else if (ram_offset && memtarget) {
offset_reg = get_ro_reg(i_regs, 0);
switch (dops[i].opcode) {
case 0x28: // SB
if(!c||memtarget) {
- int a = temp;
- if (!c) a = addr;
+ int a = addr;
if (fastio_reg_override >= 0)
a = fastio_reg_override;
do_store_byte(a, tl, offset_reg);
break;
case 0x29: // SH
if(!c||memtarget) {
- int a = temp;
- if (!c) a = addr;
+ int a = addr;
if (fastio_reg_override >= 0)
a = fastio_reg_override;
do_store_hword(a, 0, tl, offset_reg, 1);
}
type = STOREW_STUB;
break;
- case 0x3F: // SD
default:
assert(0);
}
add_stub_r(type,jaddr,out,i,addr,i_regs,ccadj_,reglist);
jaddr=0;
}
- 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
- // source register, so we need to make a copy first and use that.
- addr=temp;
- #endif
- #if defined(HOST_IMM8)
- int ir=get_reg(i_regs->regmap,INVCP);
- assert(ir>=0);
- emit_cmpmem_indexedsr12_reg(ir,addr,1);
- #else
- emit_cmpmem_indexedsr12_imm(invalid_code,addr,1);
- #endif
- #ifdef INVALIDATE_USE_COND_CALL
- emit_callne(invalidate_addr_reg[addr]);
- #else
- void *jaddr2 = out;
- emit_jne(0);
- add_stub(INVCODE_STUB,jaddr2,out,reglist|(1<<HOST_CCREG),addr,0,0,0);
- #endif
+ do_store_smc_check(i, i_regs, reglist, addr);
}
}
u_int addr_val=constmap[i][s]+offset;
static void storelr_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
+ int addr = cinfo[i].addr;
int s,tl;
- int temp;
int offset;
void *jaddr=0;
void *case1, *case23, *case3;
void *done0, *done1, *done2;
int memtarget=0,c=0;
- int agr=AGEN1+(i&1);
int offset_reg = -1;
u_int reglist=get_host_reglist(i_regs->regmap);
tl=get_reg(i_regs->regmap,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_temp(i_regs->regmap);
- offset=imm[i];
+ offset=cinfo[i].imm;
if(s>=0) {
c=(i_regs->isconst>>s)&1;
if(c) {
}
}
assert(tl>=0);
- assert(temp>=0);
+ assert(addr >= 0);
if(!c) {
- emit_cmpimm(s<0||offset?temp:s,RAM_SIZE);
- if(!offset&&s!=temp) emit_mov(s,temp);
+ emit_cmpimm(addr, RAM_SIZE);
+ if (!offset && s != addr) emit_mov(s, addr);
jaddr=out;
emit_jno(0);
}
assert(0);
}
- emit_testimm(temp,2);
+ emit_testimm(addr,2);
case23=out;
emit_jne(0);
- emit_testimm(temp,1);
+ emit_testimm(addr,1);
case1=out;
emit_jne(0);
// 0
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);
+ do_store_byte(addr, tl, offset_reg);
if (dops[i].rs2) emit_rorimm(tl, 8, tl);
}
else if (dops[i].opcode == 0x2E) { // SWR
// Write entire word
- do_store_word(temp, 0, tl, offset_reg, 1);
+ do_store_word(addr, 0, tl, offset_reg, 1);
}
done0 = out;
emit_jmp(0);
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);
+ do_store_hword(addr, -1, tl, offset_reg, 0);
if (dops[i].rs2) emit_rorimm(tl, 16, tl);
}
else if (dops[i].opcode == 0x2E) { // SWR
// Write 3 lsb into three most significant bytes
- do_store_byte(temp, tl, offset_reg);
+ do_store_byte(addr, tl, offset_reg);
if (dops[i].rs2) emit_rorimm(tl, 8, tl);
- do_store_hword(temp, 1, tl, offset_reg, 0);
+ do_store_hword(addr, 1, tl, offset_reg, 0);
if (dops[i].rs2) emit_rorimm(tl, 24, tl);
}
done1=out;
emit_jmp(0);
// 2,3
set_jump_target(case23, out);
- emit_testimm(temp,1);
+ emit_testimm(addr,1);
case3 = out;
emit_jne(0);
// 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);
+ do_store_hword(addr, -2, tl, offset_reg, 1);
if (dops[i].rs2) emit_rorimm(tl, 16, tl);
- do_store_byte(temp, tl, offset_reg);
+ do_store_byte(addr, tl, offset_reg);
if (dops[i].rs2) emit_rorimm(tl, 8, 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);
+ do_store_hword(addr, 0, tl, offset_reg, 1);
}
done2 = out;
emit_jmp(0);
// 3
set_jump_target(case3, out);
if (dops[i].opcode == 0x2A) { // SWL
- do_store_word(temp, -3, tl, offset_reg, 0);
+ do_store_word(addr, -3, tl, offset_reg, 0);
}
else if (dops[i].opcode == 0x2E) { // SWR
- do_store_byte(temp, tl, offset_reg);
+ do_store_byte(addr, tl, offset_reg);
}
set_jump_target(done0, out);
set_jump_target(done1, out);
if (offset_reg == HOST_TEMPREG)
host_tempreg_release();
if(!c||!memtarget)
- 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(invalid_code,temp,1);
- #endif
- #ifdef INVALIDATE_USE_COND_CALL
- emit_callne(invalidate_addr_reg[temp]);
- #else
- void *jaddr2 = out;
- emit_jne(0);
- add_stub(INVCODE_STUB,jaddr2,out,reglist|(1<<HOST_CCREG),temp,0,0,0);
- #endif
- }
+ add_stub_r(STORELR_STUB,jaddr,out,i,addr,i_regs,ccadj_,reglist);
+ do_store_smc_check(i, i_regs, reglist, addr);
}
static void cop0_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
if(dops[i].opcode2==0) // MFC0
{
- signed char t=get_reg(i_regs->regmap,dops[i].rt1);
+ signed char t=get_reg_w(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);
}
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);
+ emit_writeword(HOST_CCREG,&psxRegs.cycle);
}
// What a mess. The status register (12) can enable interrupts,
// so needs a special case to handle a pending interrupt.
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(&psxRegs.cycle,HOST_CCREG);
emit_readword(&next_interupt,HOST_TEMPREG);
emit_addimm(HOST_CCREG,-ccadj_,HOST_CCREG);
emit_sub(HOST_CCREG,HOST_TEMPREG,HOST_CCREG);
}
if(copr==12||copr==13) {
assert(!is_delayslot);
- emit_readword(&pending_exception,14);
- emit_test(14,14);
+ emit_readword(&pending_exception,HOST_TEMPREG);
+ emit_test(HOST_TEMPREG,HOST_TEMPREG);
void *jaddr = out;
emit_jeq(0);
emit_readword(&pcaddr, 0);
}
emit_loadreg(dops[i].rs1,s);
}
- else
- {
- 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);
- }
- }
-}
-
-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);
- }
-}
-
-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)
+static void rfe_assemble(int i, const struct regstat *i_regs)
{
- 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);
+ emit_readword(&psxRegs.CP0.n.SR, 0);
+ emit_andimm(0, 0x3c, 1);
+ emit_andimm(0, ~0xf, 0);
+ emit_orrshr_imm(1, 2, 0);
+ emit_writeword(0, &psxRegs.CP0.n.SR);
}
static int cop2_is_stalling_op(int i, int *cycles)
emit_movimm(stall, 0);
else
emit_mov(HOST_TEMPREG, 0);
- emit_addimm(HOST_CCREG, ccadj[i], 1);
+ emit_addimm(HOST_CCREG, cinfo[i].ccadj, 1);
emit_far_call(log_gte_stall);
restore_regs(reglist);
}
//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])
+ if (j > 0 && cinfo[j - 1].ccadj > cinfo[j].ccadj)
break;
}
j = max(j, 0);
}
- cycles_passed = ccadj[i] - ccadj[j];
+ cycles_passed = cinfo[i].ccadj - cinfo[j].ccadj;
if (other_gte_op_cycles >= 0)
stall = other_gte_op_cycles - cycles_passed;
else if (cycles_passed >= 44)
#if 0 // too slow
save_regs(reglist);
emit_movimm(gte_cycletab[op], 0);
- emit_addimm(HOST_CCREG, ccadj[i], 1);
+ emit_addimm(HOST_CCREG, cinfo[i].ccadj, 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_addimm(rtmp, -cinfo[i].ccadj, rtmp);
emit_sub(rtmp, HOST_CCREG, HOST_TEMPREG);
emit_cmpimm(HOST_TEMPREG, 44);
emit_cmovb_reg(rtmp, HOST_CCREG);
if (other_gte_op_cycles >= 0)
// will handle stall when assembling that op
return;
- cycles_passed = ccadj[min(j, slen -1)] - ccadj[i];
+ cycles_passed = cinfo[min(j, slen -1)].ccadj - cinfo[i].ccadj;
if (cycles_passed >= 44)
return;
assem_debug("; save gteBusyCycle\n");
#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, cinfo[i].ccadj, 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_addimm(HOST_CCREG, cinfo[i].ccadj + gte_cycletab[op], HOST_TEMPREG);
emit_writeword(HOST_TEMPREG, &psxRegs.gteBusyCycle);
#endif
host_tempreg_release();
if (is_mflohi(j))
// already handled by this op
return;
- if (dops[j].bt || (j > 0 && ccadj[j - 1] > ccadj[j]))
+ if (dops[j].bt || (j > 0 && cinfo[j - 1].ccadj > cinfo[j].ccadj))
break;
}
j = max(j, 0);
}
if (known_cycles > 0) {
- known_cycles -= ccadj[i] - ccadj[j];
+ known_cycles -= cinfo[i].ccadj - cinfo[j].ccadj;
assem_debug("; muldiv stall resolved %d\n", known_cycles);
if (known_cycles > 0)
emit_addimm(HOST_CCREG, known_cycles, HOST_CCREG);
assem_debug("; muldiv stall unresolved\n");
host_tempreg_acquire();
emit_readword(&psxRegs.muldivBusyCycle, rtmp);
- emit_addimm(rtmp, -ccadj[i], rtmp);
+ emit_addimm(rtmp, -cinfo[i].ccadj, rtmp);
emit_sub(rtmp, HOST_CCREG, HOST_TEMPREG);
emit_cmpimm(HOST_TEMPREG, 37);
emit_cmovb_reg(rtmp, HOST_CCREG);
int memtarget=0,c=0;
void *jaddr2=NULL;
enum stub_type type;
- int agr=AGEN1+(i&1);
int offset_reg = -1;
int fastio_reg_override = -1;
u_int reglist=get_host_reglist(i_regs->regmap);
u_int copr=(source[i]>>16)&0x1f;
s=get_reg(i_regs->regmap,dops[i].rs1);
tl=get_reg(i_regs->regmap,FTEMP);
- offset=imm[i];
+ offset=cinfo[i].imm;
assert(dops[i].rs1>0);
assert(tl>=0);
reglist&=~(1<<HOST_CCREG);
// get the address
+ ar = cinfo[i].addr;
+ assert(ar >= 0);
if (dops[i].opcode==0x3a) { // SWC2
- ar=get_reg(i_regs->regmap,agr);
- if(ar<0) ar=get_reg_temp(i_regs->regmap);
- reglist|=1<<ar;
- } else { // LWC2
- ar=tl;
+ reglist |= 1<<ar;
}
if(s>=0) c=(i_regs->wasconst>>s)&1;
memtarget=c&&(((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE);
- if (!offset&&!c&&s>=0) ar=s;
- assert(ar>=0);
cop2_do_stall_check(0, i, i_regs, reglist);
else {
if(!c) {
jaddr2 = emit_fastpath_cmp_jump(i, i_regs, ar,
- &offset_reg, &fastio_reg_override);
+ &offset_reg, &fastio_reg_override, ccadj_);
}
else if (ram_offset && memtarget) {
offset_reg = get_ro_reg(i_regs, 0);
if(jaddr2)
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(invalid_code,ar,1);
-#endif
- #ifdef INVALIDATE_USE_COND_CALL
- emit_callne(invalidate_addr_reg[ar]);
- #else
- void *jaddr3 = out;
- emit_jne(0);
- add_stub(INVCODE_STUB,jaddr3,out,reglist|(1<<HOST_CCREG),ar,0,0,0);
- #endif
- }
+ do_store_smc_check(i, i_regs, reglist, ar);
if (dops[i].opcode==0x32) { // LWC2
host_tempreg_acquire();
cop2_put_dreg(copr,tl,HOST_TEMPREG);
cop2_do_stall_check(0, i, i_regs, reglist);
}
if (dops[i].opcode2==0) { // MFC2
- signed char tl=get_reg(i_regs->regmap,dops[i].rt1);
+ signed char tl=get_reg_w(i_regs->regmap, dops[i].rt1);
if(tl>=0&&dops[i].rt1!=0)
cop2_get_dreg(copr,tl,temp);
}
}
else if (dops[i].opcode2==2) // CFC2
{
- signed char tl=get_reg(i_regs->regmap,dops[i].rt1);
+ signed char tl=get_reg_w(i_regs->regmap, dops[i].rt1);
if(tl>=0&&dops[i].rt1!=0)
emit_readword(®_cop2c[copr],tl);
}
emit_jmp(stubs[n].retaddr); // return address
}
+static void do_overflowstub(int n)
+{
+ assem_debug("do_overflowstub %x\n", start + (u_int)stubs[n].a * 4);
+ literal_pool(24);
+ int i = stubs[n].a;
+ struct regstat *i_regs = (struct regstat *)stubs[n].c;
+ int ccadj = stubs[n].d;
+ set_jump_target(stubs[n].addr, out);
+ wb_dirtys(regs[i].regmap, regs[i].dirty);
+ exception_assemble(i, i_regs, ccadj);
+}
+
+static void do_alignmentstub(int n)
+{
+ assem_debug("do_alignmentstub %x\n", start + (u_int)stubs[n].a * 4);
+ literal_pool(24);
+ int i = stubs[n].a;
+ struct regstat *i_regs = (struct regstat *)stubs[n].c;
+ int ccadj = stubs[n].d;
+ int is_store = dops[i].itype == STORE || dops[i].opcode == 0x3A; // SWC2
+ int cause = (dops[i].opcode & 3) << 28;
+ cause |= is_store ? (R3000E_AdES << 2) : (R3000E_AdEL << 2);
+ set_jump_target(stubs[n].addr, out);
+ wb_dirtys(regs[i].regmap, regs[i].dirty);
+ if (stubs[n].b != 1)
+ emit_mov(stubs[n].b, 1); // faulting address
+ emit_movimm(cause, 0);
+ exception_assemble(i, i_regs, ccadj);
+}
+
#ifndef multdiv_assemble
void multdiv_assemble(int i,struct regstat *i_regs)
{
//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);
+ tl=get_reg_w(i_regs->regmap, dops[i].rt1);
//assert(tl>=0);
if(tl>=0) {
sl=get_reg(i_regs->regmap,dops[i].rs1);
emit_addimm(HOST_CCREG,ccadj_,HOST_CCREG);
emit_add(2,HOST_CCREG,2);
emit_writeword(2,&psxRegs.cycle);
+ emit_addimm_ptr(FP,(u_char *)&psxRegs - (u_char *)&dynarec_local,0);
emit_far_call(func);
emit_far_jump(jump_to_new_pc);
}
-static void syscall_assemble(int i, const struct regstat *i_regs, int ccadj_)
+static void exception_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);
+ void *func;
+ if (dops[i].itype == ALU || dops[i].itype == IMM16)
+ func = is_delayslot ? jump_overflow_ds : jump_overflow;
+ else if (dops[i].itype == LOAD || dops[i].itype == STORE)
+ func = is_delayslot ? jump_addrerror_ds : jump_addrerror;
+ else if (dops[i].opcode2 == 0x0C)
+ func = is_delayslot ? jump_syscall_ds : jump_syscall;
+ else
+ func = is_delayslot ? jump_break_ds : jump_break;
+ if (get_reg(i_regs->regmap, CCREG) != HOST_CCREG) // evicted
+ emit_loadreg(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_bad()
+{
+ assert(0);
+}
+
static void hlecall_assemble(int i, const struct regstat *i_regs, int ccadj_)
{
- void *hlefunc = psxNULL;
+ void *hlefunc = hlecall_bad;
uint32_t hleCode = source[i] & 0x03ffffff;
if (hleCode < ARRAY_SIZE(psxHLEt))
hlefunc = psxHLEt[hleCode];
// 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);
+ int value,hr=get_reg_w(regs[i].regmap, dops[i].rt1);
if(hr>=0) {
if(get_final_value(hr,i,&value))
smrv[dops[i].rt1]=value;
int ds = 0;
switch (dops[i].itype) {
case ALU:
- alu_assemble(i, i_regs);
+ alu_assemble(i, i_regs, ccadj_);
break;
case IMM16:
- imm16_assemble(i, i_regs);
+ imm16_assemble(i, i_regs, ccadj_);
break;
case SHIFT:
shift_assemble(i, i_regs);
case COP0:
cop0_assemble(i, i_regs, ccadj_);
break;
- case COP1:
- cop1_assemble(i, i_regs);
- break;
- case C1LS:
- c1ls_assemble(i, i_regs);
+ case RFE:
+ rfe_assemble(i, i_regs);
break;
case COP2:
cop2_assemble(i, i_regs);
mov_assemble(i, i_regs);
break;
case SYSCALL:
- syscall_assemble(i, i_regs, ccadj_);
+ exception_assemble(i, i_regs, ccadj_);
break;
case HLECALL:
hlecall_assemble(i, i_regs, ccadj_);
break;
case NOP:
case OTHER:
- case NI:
// not handled, just skip
break;
default:
SysPrintf("Jump in the delay slot. This is probably a bug.\n");
break;
default:
- assemble(i, i_regs, ccadj[i]);
+ assemble(i, i_regs, cinfo[i].ccadj);
}
is_delayslot = 0;
}
}
// Generate address for load/store instruction
-// goes to AGEN for writes, FTEMP for LOADLR and cop1/2 loads
+// goes to AGEN (or temp) for writes, FTEMP for LOADLR and cop1/2 loads
+// AGEN is assigned by pass5b_preallocate2
static void address_generation(int i, const struct regstat *i_regs, signed char entry[])
{
if (dops[i].is_load || dops[i].is_store) {
- int ra=-1;
- int agr=AGEN1+(i&1);
+ int ra = -1;
+ int agr = AGEN1 + (i&1);
if(dops[i].itype==LOAD) {
- ra=get_reg(i_regs->regmap,dops[i].rt1);
- if(ra<0) ra=get_reg_temp(i_regs->regmap);
- assert(ra>=0);
+ if (!dops[i].may_except)
+ ra = get_reg_w(i_regs->regmap, dops[i].rt1); // reuse dest for agen
+ if (ra < 0)
+ ra = get_reg_temp(i_regs->regmap);
}
if(dops[i].itype==LOADLR) {
ra=get_reg(i_regs->regmap,FTEMP);
if(ra<0) ra=get_reg_temp(i_regs->regmap);
}
if(dops[i].itype==C2LS) {
- if ((dops[i].opcode&0x3b)==0x31||(dops[i].opcode&0x3b)==0x32) // LWC1/LDC1/LWC2/LDC2
+ if (dops[i].opcode == 0x32) // LWC2
ra=get_reg(i_regs->regmap,FTEMP);
- else { // SWC1/SDC1/SWC2/SDC2
+ else { // SWC2
ra=get_reg(i_regs->regmap,agr);
if(ra<0) ra=get_reg_temp(i_regs->regmap);
}
}
- int rs=get_reg(i_regs->regmap,dops[i].rs1);
- if(ra>=0) {
- int offset=imm[i];
+ int rs = get_reg(i_regs->regmap, dops[i].rs1);
+ //if(ra>=0)
+ {
+ int offset = cinfo[i].imm;
+ int add_offset = offset != 0;
int c=(i_regs->wasconst>>rs)&1;
if(dops[i].rs1==0) {
// Using r0 as a base address
+ assert(ra >= 0);
if(!entry||entry[ra]!=agr) {
if (dops[i].opcode==0x22||dops[i].opcode==0x26) {
emit_movimm(offset&0xFFFFFFFC,ra); // LWL/LWR
- }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]!=dops[i].rs1)
- emit_loadreg(dops[i].rs1,ra);
+ cinfo[i].addr = ra;
+ add_offset = 0;
+ }
+ else if (rs < 0) {
+ assert(ra >= 0);
+ if (!entry || entry[ra] != dops[i].rs1)
+ emit_loadreg(dops[i].rs1, ra);
+ cinfo[i].addr = ra;
//if(!entry||entry[ra]!=dops[i].rs1)
// printf("poor load scheduling!\n");
}
else if(c) {
if(dops[i].rs1!=dops[i].rt1||dops[i].itype!=LOAD) {
+ assert(ra >= 0);
if(!entry||entry[ra]!=agr) {
if (dops[i].opcode==0x22||dops[i].opcode==0x26) {
emit_movimm((constmap[i][rs]+offset)&0xFFFFFFFC,ra); // LWL/LWR
- }else if (dops[i].opcode==0x1a||dops[i].opcode==0x1b) {
- emit_movimm((constmap[i][rs]+offset)&0xFFFFFFF8,ra); // LDL/LDR
}else{
emit_movimm(constmap[i][rs]+offset,ra);
regs[i].loadedconst|=1<<ra;
}
} // else did it in the previous cycle
- } // else load_consts already did it
+ cinfo[i].addr = ra;
+ }
+ else // else load_consts already did it
+ cinfo[i].addr = rs;
+ add_offset = 0;
+ }
+ else if (dops[i].itype == STORELR) { // overwrites addr
+ assert(ra >= 0);
+ assert(rs != ra);
+ emit_mov(rs, ra);
+ cinfo[i].addr = ra;
}
- if(offset&&!c&&dops[i].rs1) {
+ else
+ cinfo[i].addr = rs;
+ if (add_offset) {
+ assert(ra >= 0);
if(rs>=0) {
emit_addimm(rs,offset,ra);
}else{
emit_addimm(ra,offset,ra);
}
+ cinfo[i].addr = ra;
}
}
+ assert(cinfo[i].addr >= 0);
}
// Preload constants for next instruction
if (dops[i+1].is_load || dops[i+1].is_store) {
ra=get_reg(i_regs->regmap,agr);
if(ra>=0) {
int rs=get_reg(regs[i+1].regmap,dops[i+1].rs1);
- int offset=imm[i+1];
+ int offset=cinfo[i+1].imm;
int c=(regs[i+1].wasconst>>rs)&1;
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) {
if(dops[i+2].itype==LOAD&&dops[i+2].rs1==reg&&dops[i+2].rt1==reg&&((regs[i+1].wasconst>>hr)&1))
{
// Precompute load address
- *value=constmap[i][hr]+imm[i+2];
+ *value=constmap[i][hr]+cinfo[i+2].imm;
return 1;
}
}
if(dops[i+1].itype==LOAD&&dops[i+1].rs1==reg&&dops[i+1].rt1==reg)
{
// Precompute load address
- *value=constmap[i][hr]+imm[i+1];
- //printf("c=%x imm=%lx\n",(long)constmap[i][hr],imm[i+1]);
+ *value=constmap[i][hr]+cinfo[i+1].imm;
+ //printf("c=%x imm=%lx\n",(long)constmap[i][hr],cinfo[i+1].imm);
return 1;
}
}
{
int hr;
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);
+ else if(cinfo[t].ccadj) emit_addimm(HOST_CCREG,-cinfo[t].ccadj,HOST_CCREG);
if(regs[t].regmap_entry[HOST_CCREG]!=CCREG) {
emit_storereg(CCREG,HOST_CCREG);
}
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)
+ if (hr == EXCLUDE_REG || reg <= 0)
continue;
if (!((regs[i-1].isconst >> hr) & 1))
continue;
// Used when a branch jumps into the delay slot of another branch
static void ds_assemble_entry(int i)
{
- int t = (ba[i] - start) >> 2;
+ int t = (cinfo[i].ba - 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("Assemble delay slot at %x\n",cinfo[i].ba);
assem_debug("<->\n");
drc_dbg_emit_do_cmp(t, ccadj_);
if(regs[t].regmap_entry[HOST_CCREG]==CCREG&®s[t].regmap[HOST_CCREG]!=CCREG)
default:
assemble(t, ®s[t], ccadj_);
}
- store_regs_bt(regs[t].regmap,regs[t].dirty,ba[i]+4);
- load_regs_bt(regs[t].regmap,regs[t].dirty,ba[i]+4);
- if(internal_branch(ba[i]+4))
+ store_regs_bt(regs[t].regmap,regs[t].dirty,cinfo[i].ba+4);
+ load_regs_bt(regs[t].regmap,regs[t].dirty,cinfo[i].ba+4);
+ if(internal_branch(cinfo[i].ba+4))
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- assert(internal_branch(ba[i]+4));
- add_to_linker(out,ba[i]+4,internal_branch(ba[i]+4));
+ assert(internal_branch(cinfo[i].ba+4));
+ add_to_linker(out,cinfo[i].ba+4,internal_branch(cinfo[i].ba+4));
emit_jmp(0);
}
{
*adj=0;
}
- //if(ba[i]>=start && ba[i]<(start+slen*4))
- if(internal_branch(ba[i]))
+ //if(cinfo[i].ba>=start && cinfo[i].ba<(start+slen*4))
+ if(internal_branch(cinfo[i].ba))
{
- t=(ba[i]-start)>>2;
+ t=(cinfo[i].ba-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=cinfo[t].ccadj;
}
else
{
*adj=0;
}
- count = ccadj[i];
+ count = cinfo[i].ccadj;
count_plus2 = count + CLOCK_ADJUST(2);
- if(taken==TAKEN && i==(ba[i]-start)>>2 && source[i+1]==0) {
+ if(taken==TAKEN && i==(cinfo[i].ba-start)>>2 && source[i+1]==0) {
// Idle loop
if(count&1) emit_addimm_and_set_flags(2*(count+2),HOST_CCREG);
idle=out;
wb_dirtys(branch_regs[i].regmap,branch_regs[i].dirty);
}
else {
- if(internal_branch(ba[i]))
- wb_needed_dirtys(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+ if(internal_branch(cinfo[i].ba))
+ wb_needed_dirtys(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
}
if(stubs[n].c!=-1)
{
// Save PC as return address
- emit_movimm(stubs[n].c,EAX);
- emit_writeword(EAX,&pcaddr);
+ emit_movimm(stubs[n].c,0);
+ emit_writeword(0,&pcaddr);
}
else
{
}
hr++;
}
- if((dops[i].opcode&0x2E)==6) // BLEZ/BGTZ needs another register
+ if ((dops[i].opcode & 0x3e) == 6) // BLEZ/BGTZ needs another register
{
while(hr<HOST_REGS)
{
}
assert(hr<HOST_REGS);
}
- if((dops[i].opcode&0x2f)==4) // BEQ
+ if (dops[i].opcode == 4) // BEQ
{
#ifdef HAVE_CMOV_IMM
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
- emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
+ emit_cmov2imm_e_ne_compact(cinfo[i].ba,start+i*4+8,addr);
#else
- emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
+ emit_mov2imm_compact(cinfo[i].ba,addr,start+i*4+8,alt);
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
emit_cmovne_reg(alt,addr);
#endif
}
- if((dops[i].opcode&0x2f)==5) // BNE
+ else if (dops[i].opcode == 5) // BNE
{
#ifdef HAVE_CMOV_IMM
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
- emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
+ emit_cmov2imm_e_ne_compact(start+i*4+8,cinfo[i].ba,addr);
#else
- emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
+ emit_mov2imm_compact(start+i*4+8,addr,cinfo[i].ba,alt);
if(s2l>=0) emit_cmp(s1l,s2l);
else emit_test(s1l,s1l);
emit_cmovne_reg(alt,addr);
#endif
}
- if((dops[i].opcode&0x2f)==6) // BLEZ
+ else if (dops[i].opcode == 6) // BLEZ
{
- //emit_movimm(ba[i],alt);
+ //emit_movimm(cinfo[i].ba,alt);
//emit_movimm(start+i*4+8,addr);
- emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
+ emit_mov2imm_compact(cinfo[i].ba,alt,start+i*4+8,addr);
emit_cmpimm(s1l,1);
emit_cmovl_reg(alt,addr);
}
- if((dops[i].opcode&0x2f)==7) // BGTZ
+ else if (dops[i].opcode == 7) // BGTZ
{
- //emit_movimm(ba[i],addr);
+ //emit_movimm(cinfo[i].ba,addr);
//emit_movimm(start+i*4+8,ntaddr);
- emit_mov2imm_compact(ba[i],addr,start+i*4+8,ntaddr);
+ emit_mov2imm_compact(cinfo[i].ba,addr,start+i*4+8,ntaddr);
emit_cmpimm(s1l,1);
emit_cmovl_reg(ntaddr,addr);
}
- if((dops[i].opcode==1)&&(dops[i].opcode2&0x2D)==0) // BLTZ
+ else if (dops[i].itype == SJUMP) // BLTZ/BGEZ
{
- //emit_movimm(ba[i],alt);
+ //emit_movimm(cinfo[i].ba,alt);
//emit_movimm(start+i*4+8,addr);
- emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
+ emit_mov2imm_compact(cinfo[i].ba,
+ (dops[i].opcode2 & 1) ? addr : alt, start + i*4 + 8,
+ (dops[i].opcode2 & 1) ? alt : addr);
emit_test(s1l,s1l);
emit_cmovs_reg(alt,addr);
}
- 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);
- emit_test(s1l,s1l);
- emit_cmovs_reg(alt,addr);
- }
- if(dops[i].opcode==0x11 && dops[i].opcode2==0x08 ) {
- if(source[i]&0x10000) // BC1T
- {
- //emit_movimm(ba[i],alt);
- //emit_movimm(start+i*4+8,addr);
- emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
- emit_testimm(s1l,0x800000);
- emit_cmovne_reg(alt,addr);
- }
- else // BC1F
- {
- //emit_movimm(ba[i],addr);
- //emit_movimm(start+i*4+8,alt);
- emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
- emit_testimm(s1l,0x800000);
- emit_cmovne_reg(alt,addr);
- }
- }
- emit_writeword(addr,&pcaddr);
+ emit_writeword(addr, &pcaddr);
}
else
if(dops[i].itype==RJUMP)
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);
+ if(internal_branch(cinfo[i].ba))
+ load_needed_regs(branch_regs[i].regmap,regs[(cinfo[i].ba-start)>>2].regmap_entry);
else if(dops[i].itype==RJUMP) {
if(get_reg(branch_regs[i].regmap,RTEMP)>=0)
emit_readword(&pcaddr,get_reg(branch_regs[i].regmap,RTEMP));
static void ujump_assemble(int i, const struct regstat *i_regs)
{
int ra_done=0;
- if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
+ if(i==(cinfo[i].ba-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);
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].dirty,ba[i]);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
#ifdef REG_PREFETCH
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, 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]))
+ do_cc(i,branch_regs[i].regmap,&adj,cinfo[i].ba,TAKEN,0);
+ if(adj) emit_addimm(cc, cinfo[i].ccadj + CLOCK_ADJUST(2) - adj, cc);
+ load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
+ if(internal_branch(cinfo[i].ba))
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- if (internal_branch(ba[i]) && dops[(ba[i]-start)>>2].is_ds) {
+ if (internal_branch(cinfo[i].ba) && dops[(cinfo[i].ba-start)>>2].is_ds) {
ds_assemble_entry(i);
}
else {
- add_to_linker(out,ba[i],internal_branch(ba[i]));
+ add_to_linker(out,cinfo[i].ba,internal_branch(cinfo[i].ba));
emit_jmp(0);
}
}
int rt,return_address;
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);
+ rt=get_reg_w(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;
}
#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
+ //if(adj) emit_addimm(cc,2*(cinfo[i].ccadj+2-adj),cc); // ??? - Shouldn't happen
//assert(adj==0);
- emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG);
+ emit_addimm_and_set_flags(cinfo[i].ccadj + 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)
+ if (dops[i+1].itype == RFE)
// special case for RFE
emit_jmp(0);
else
const signed char *i_regmap = i_regs->regmap;
int cc;
int match;
- match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+ match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
assem_debug("match=%d\n",match);
int s1l,s2l;
int unconditional=0,nop=0;
int invert=0;
- int internal=internal_branch(ba[i]);
- if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
+ int internal=internal_branch(cinfo[i].ba);
+ if(i==(cinfo[i].ba-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;
+ if(i>(cinfo[i].ba-start)>>2) invert=1;
#endif
#ifdef __aarch64__
invert=1; // because of near cond. branches
cc=get_reg(branch_regs[i].regmap,CCREG);
assert(cc==HOST_CCREG);
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);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
+ //do_cc(i,branch_regs[i].regmap,&adj,unconditional?cinfo[i].ba:-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, ccadj[i] + CLOCK_ADJUST(2) - adj, cc);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+ do_cc(i,branch_regs[i].regmap,&adj,cinfo[i].ba,TAKEN,0);
+ if(i!=(cinfo[i].ba-start)>>2 || source[i+1]!=0) {
+ if(adj) emit_addimm(cc, cinfo[i].ccadj + CLOCK_ADJUST(2) - adj, cc);
+ load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
if(internal)
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- if (internal && dops[(ba[i]-start)>>2].is_ds) {
+ if (internal && dops[(cinfo[i].ba-start)>>2].is_ds) {
ds_assemble_entry(i);
}
else {
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jmp(0);
}
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
}
}
else if(nop) {
- emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc);
+ emit_addimm_and_set_flags(cinfo[i].ccadj + CLOCK_ADJUST(2), cc);
void *jaddr=out;
emit_jns(0);
add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
else {
void *taken = NULL, *nottaken = NULL, *nottaken1 = NULL;
do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
- if(adj&&!invert) emit_addimm(cc, ccadj[i] + CLOCK_ADJUST(2) - adj, cc);
+ if(adj&&!invert) emit_addimm(cc, cinfo[i].ccadj + 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);
nottaken=out;
emit_jne(DJT_1);
}else{
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jeq(0);
}
}
nottaken=out;
emit_jeq(DJT_1);
}else{
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jne(0);
}
}
nottaken=out;
emit_jge(DJT_1);
}else{
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jl(0);
}
}
nottaken=out;
emit_jl(DJT_1);
}else{
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jge(0);
}
}
if(invert) {
if(taken) set_jump_target(taken, out);
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
- if (match && (!internal || !dops[(ba[i]-start)>>2].is_ds)) {
+ if (match && (!internal || !dops[(cinfo[i].ba-start)>>2].is_ds)) {
if(adj) {
emit_addimm(cc,-adj,cc);
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
}else{
emit_addnop(13);
- add_to_linker(out,ba[i],internal*2);
+ add_to_linker(out,cinfo[i].ba,internal*2);
}
emit_jmp(0);
}else
#endif
{
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]);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
+ load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
if(internal)
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- if (internal && dops[(ba[i] - start) >> 2].is_ds) {
+ if (internal && dops[(cinfo[i].ba - start) >> 2].is_ds) {
ds_assemble_entry(i);
}
else {
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jmp(0);
}
}
// CHECK: Is the following instruction (fall thru) allocated ok?
}
assert(cc==HOST_CCREG);
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
- do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
+ do_cc(i,i_regmap,&adj,cinfo[i].ba,TAKEN,0);
assem_debug("cycle count (adj)\n");
- 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(adj) emit_addimm(cc, cinfo[i].ccadj + CLOCK_ADJUST(2) - adj, cc);
+ load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
if(internal)
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- if (internal && dops[(ba[i] - start) >> 2].is_ds) {
+ if (internal && dops[(cinfo[i].ba - start) >> 2].is_ds) {
ds_assemble_entry(i);
}
else {
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jmp(0);
}
}
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(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG);
+ emit_addimm_and_set_flags(cinfo[i].ccadj + CLOCK_ADJUST(2), HOST_CCREG);
void *jaddr=out;
emit_jns(0);
add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
else{
cc=get_reg(i_regmap,CCREG);
assert(cc==HOST_CCREG);
- emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc);
+ emit_addimm_and_set_flags(cinfo[i].ccadj + CLOCK_ADJUST(2), cc);
void *jaddr=out;
emit_jns(0);
add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
const signed char *i_regmap = i_regs->regmap;
int cc;
int match;
- match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+ match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
assem_debug("smatch=%d ooo=%d\n", match, dops[i].ooo);
int s1l;
int unconditional=0,nevertaken=0;
int invert=0;
- int internal=internal_branch(ba[i]);
- if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
+ int internal=internal_branch(cinfo[i].ba);
+ if(i==(cinfo[i].ba-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;
+ if(i>(cinfo[i].ba-start)>>2) invert=1;
#endif
#ifdef __aarch64__
invert=1; // because of near cond. branches
cc=get_reg(branch_regs[i].regmap,CCREG);
assert(cc==HOST_CCREG);
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);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
+ //do_cc(i,branch_regs[i].regmap,&adj,unconditional?cinfo[i].ba:-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, ccadj[i] + CLOCK_ADJUST(2) - adj, cc);
- load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+ do_cc(i,branch_regs[i].regmap,&adj,cinfo[i].ba,TAKEN,0);
+ if(i!=(cinfo[i].ba-start)>>2 || source[i+1]!=0) {
+ if(adj) emit_addimm(cc, cinfo[i].ccadj + CLOCK_ADJUST(2) - adj, cc);
+ load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
if(internal)
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- if (internal && dops[(ba[i] - start) >> 2].is_ds) {
+ if (internal && dops[(cinfo[i].ba - start) >> 2].is_ds) {
ds_assemble_entry(i);
}
else {
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jmp(0);
}
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
}
}
else if(nevertaken) {
- emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc);
+ emit_addimm_and_set_flags(cinfo[i].ccadj + CLOCK_ADJUST(2), cc);
void *jaddr=out;
emit_jns(0);
add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
else {
void *nottaken = NULL;
do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
- if(adj&&!invert) emit_addimm(cc, ccadj[i] + CLOCK_ADJUST(2) - adj, cc);
+ if(adj&&!invert) emit_addimm(cc, cinfo[i].ccadj + CLOCK_ADJUST(2) - adj, cc);
{
assert(s1l>=0);
- if((dops[i].opcode2&0xf)==0) // BLTZ/BLTZAL
+ if ((dops[i].opcode2 & 1) == 0) // BLTZ/BLTZAL
{
emit_test(s1l,s1l);
if(invert){
nottaken=out;
emit_jns(DJT_1);
}else{
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_js(0);
}
}
- if((dops[i].opcode2&0xf)==1) // BGEZ/BLTZAL
+ else // BGEZ/BGEZAL
{
emit_test(s1l,s1l);
if(invert){
nottaken=out;
emit_js(DJT_1);
}else{
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jns(0);
}
}
if(invert) {
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
- if (match && (!internal || !dops[(ba[i] - start) >> 2].is_ds)) {
+ if (match && (!internal || !dops[(cinfo[i].ba - start) >> 2].is_ds)) {
if(adj) {
emit_addimm(cc,-adj,cc);
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
}else{
emit_addnop(13);
- add_to_linker(out,ba[i],internal*2);
+ add_to_linker(out,cinfo[i].ba,internal*2);
}
emit_jmp(0);
}else
#endif
{
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]);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
+ load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
if(internal)
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- if (internal && dops[(ba[i] - start) >> 2].is_ds) {
+ if (internal && dops[(cinfo[i].ba - start) >> 2].is_ds) {
ds_assemble_entry(i);
}
else {
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jmp(0);
}
}
// In-order execution (branch first)
//printf("IOE\n");
void *nottaken = NULL;
- if(dops[i].rt1==31) {
- int rt,return_address;
- rt=get_reg(branch_regs[i].regmap,31);
- if(rt>=0) {
+ if (!unconditional) {
+ assert(s1l >= 0);
+ emit_test(s1l, s1l);
+ }
+ if (dops[i].rt1 == 31) {
+ int rt, return_address;
+ rt = get_reg(branch_regs[i].regmap,31);
+ if(rt >= 0) {
// Save the PC even if the branch is not taken
- return_address=start+i*4+8;
- emit_movimm(return_address,rt); // PC into link register
+ return_address = start + i*4+8;
+ emit_movimm(return_address, rt); // PC into link register
#ifdef IMM_PREFETCH
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]);
- assert(s1l>=0);
- if((dops[i].opcode2&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL
- {
- emit_test(s1l,s1l);
- nottaken=out;
- emit_jns(DJT_1);
- }
- if((dops[i].opcode2&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL
- {
- emit_test(s1l,s1l);
- nottaken=out;
- emit_js(DJT_1);
- }
- } // if(!unconditional)
+ if (!unconditional) {
+ nottaken = out;
+ if (!(dops[i].opcode2 & 1)) // BLTZ/BLTZAL
+ emit_jns(DJT_1);
+ else // BGEZ/BGEZAL
+ emit_js(DJT_1);
+ }
int adj;
uint64_t ds_unneeded=branch_regs[i].u;
ds_unneeded&=~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
// CHECK: Is the following instruction (fall thru) allocated ok?
}
assert(cc==HOST_CCREG);
- store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
- do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
+ store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
+ do_cc(i,i_regmap,&adj,cinfo[i].ba,TAKEN,0);
assem_debug("cycle count (adj)\n");
- 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(adj) emit_addimm(cc, cinfo[i].ccadj + CLOCK_ADJUST(2) - adj, cc);
+ load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,cinfo[i].ba);
if(internal)
assem_debug("branch: internal\n");
else
assem_debug("branch: external\n");
- if (internal && dops[(ba[i] - start) >> 2].is_ds) {
+ if (internal && dops[(cinfo[i].ba - start) >> 2].is_ds) {
ds_assemble_entry(i);
}
else {
- add_to_linker(out,ba[i],internal);
+ add_to_linker(out,cinfo[i].ba,internal);
emit_jmp(0);
}
}
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(ccadj[i] + CLOCK_ADJUST(2), HOST_CCREG);
+ emit_addimm_and_set_flags(cinfo[i].ccadj + CLOCK_ADJUST(2), HOST_CCREG);
void *jaddr=out;
emit_jns(0);
add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
else{
cc=get_reg(i_regmap,CCREG);
assert(cc==HOST_CCREG);
- emit_addimm_and_set_flags(ccadj[i] + CLOCK_ADJUST(2), cc);
+ emit_addimm_and_set_flags(cinfo[i].ccadj + CLOCK_ADJUST(2), cc);
void *jaddr=out;
emit_jns(0);
add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
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;
+ printf (" %x: %s %8x\n",start+i*4,insn[i],cinfo[i].ba);break;
case CJUMP:
- 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;
+ 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):cinfo[i].ba);break;
case SJUMP:
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 (dops[i].opcode==0x9&&dops[i].rt1!=31)
+ if (dops[i].opcode2 == 9 && 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],dops[i].rs1);
break;
case IMM16:
if(dops[i].opcode==0xf) //LUI
- printf (" %x: %s r%d,%4x0000\n",start+i*4,insn[i],dops[i].rt1,imm[i]&0xffff);
+ printf (" %x: %s r%d,%4x0000\n",start+i*4,insn[i],dops[i].rt1,cinfo[i].imm&0xffff);
else
- printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,imm[i]);
+ printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,cinfo[i].imm);
break;
case LOAD:
case LOADLR:
- printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,imm[i]);
+ printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,cinfo[i].imm);
break;
case STORE:
case STORELR:
- printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],dops[i].rs2,dops[i].rs1,imm[i]);
+ printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],dops[i].rs2,dops[i].rs1,cinfo[i].imm);
break;
case ALU:
case SHIFT:
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],dops[i].rt1,dops[i].rs1,imm[i]);
+ printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],dops[i].rt1,dops[i].rs1,cinfo[i].imm);
break;
case MOV:
if((dops[i].opcode2&0x1d)==0x10)
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(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(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
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,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,dops[i].rs1,imm[i]);
+ printf (" %x: %s cpr2[%d],r%d+%x\n",start+i*4,insn[i],(source[i]>>16)&0x1f,dops[i].rs1,cinfo[i].imm);
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]);
}
+ #ifndef REGMAP_PRINT
return;
- printf("D: %"PRIu64" WD: %"PRIu64" U: %"PRIu64"\n",
- regs[i].dirty, regs[i].wasdirty, unneeded_reg[i]);
+ #endif
+ printf("D: %"PRIx64" WD: %"PRIx64" U: %"PRIx64" hC: %x hWC: %x hLC: %x\n",
+ regs[i].dirty, regs[i].wasdirty, unneeded_reg[i],
+ regs[i].isconst, regs[i].wasconst, regs[i].loadedconst);
print_regmap("pre: ", regmap_pre[i]);
print_regmap("entry: ", regs[i].regmap_entry);
print_regmap("map: ", regs[i].regmap);
#define DRC_TEST_VAL 0x74657374
-static void new_dynarec_test(void)
+static noinline void new_dynarec_test(void)
{
int (*testfunc)(void);
void *beginning;
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
+ SysPrintf("(%p) testing if we can run recompiled code @%p...\n",
+ new_dynarec_test, out);
+ ((volatile u_int *)NDRC_WRITE_OFFSET(out))[0]++; // make the cache dirty
for (i = 0; i < ARRAY_SIZE(ret); i++) {
out = ndrc->translation_cache;
f1_hack=0;
for (n = 0; n < ARRAY_SIZE(blocks); n++)
blocks_clear(&blocks[n]);
- for (n = 0; n < ARRAY_SIZE(jump_out); n++)
- ll_clear(&jump_out[n]);
+ for (n = 0; n < ARRAY_SIZE(jumps); n++) {
+ free(jumps[n]);
+ jumps[n] = NULL;
+ }
stat_clear(stat_blocks);
stat_clear(stat_links);
- cycle_multiplier_old = cycle_multiplier;
+ cycle_multiplier_old = Config.cycle_multiplier;
new_dynarec_hacks_old = new_dynarec_hacks;
}
#elif defined(_MSC_VER)
ndrc = VirtualAlloc(NULL, sizeof(*ndrc), MEM_COMMIT | MEM_RESERVE,
PAGE_EXECUTE_READWRITE);
+ #elif defined(HAVE_LIBNX)
+ Result rc = jitCreate(&g_jit, sizeof(*ndrc));
+ if (R_FAILED(rc))
+ SysPrintf("jitCreate failed: %08x\n", rc);
+ SysPrintf("jitCreate: RX: %p RW: %p type: %d\n", g_jit.rx_addr, g_jit.rw_addr, g_jit.type);
+ jitTransitionToWritable(&g_jit);
+ ndrc = g_jit.rx_addr;
+ ndrc_write_ofs = (char *)g_jit.rw_addr - (char *)ndrc;
+ memset(NDRC_WRITE_OFFSET(&ndrc->tramp), 0, sizeof(ndrc->tramp));
#else
uintptr_t desired_addr = 0;
+ int prot = PROT_READ | PROT_WRITE | PROT_EXEC;
+ int flags = MAP_PRIVATE | MAP_ANONYMOUS;
+ int fd = -1;
#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_PRIVATE | MAP_ANONYMOUS, -1, 0);
+ #ifdef TC_WRITE_OFFSET
+ // mostly for testing
+ fd = open("/dev/shm/pcsxr", O_CREAT | O_RDWR, 0600);
+ ftruncate(fd, sizeof(*ndrc));
+ void *mw = mmap(NULL, sizeof(*ndrc), PROT_READ | PROT_WRITE,
+ (flags = MAP_SHARED), fd, 0);
+ assert(mw != MAP_FAILED);
+ prot = PROT_READ | PROT_EXEC;
+ #endif
+ ndrc = mmap((void *)desired_addr, sizeof(*ndrc), prot, flags, fd, 0);
if (ndrc == MAP_FAILED) {
SysPrintf("mmap() failed: %s\n", strerror(errno));
abort();
}
+ #ifdef TC_WRITE_OFFSET
+ ndrc_write_ofs = (char *)mw - (char *)ndrc;
+ #endif
#endif
#else
#ifndef NO_WRITE_EXEC
#endif
#endif
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
// sceBlock is managed by retroarch's bootstrap code
//sceKernelFreeMemBlock(sceBlock);
//sceBlock = -1;
+ #elif defined(HAVE_LIBNX)
+ jitClose(&g_jit);
+ ndrc = NULL;
#else
if (munmap(ndrc, sizeof(*ndrc)) < 0)
SysPrintf("munmap() failed\n");
+ ndrc = NULL;
#endif
#endif
for (n = 0; n < ARRAY_SIZE(blocks); n++)
blocks_clear(&blocks[n]);
- for (n = 0; n < ARRAY_SIZE(jump_out); n++)
- ll_clear(&jump_out[n]);
+ for (n = 0; n < ARRAY_SIZE(jumps); n++) {
+ free(jumps[n]);
+ jumps[n] = NULL;
+ }
stat_clear(stat_blocks);
stat_clear(stat_links);
- #ifdef ROM_COPY
- if (munmap (ROM_COPY, 67108864) < 0) {SysPrintf("munmap() failed\n");}
- #endif
new_dynarec_print_stats();
}
(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
+ // but timings in PCSX are too tied to the interpreter's 2-per-insn assumption
if (!HACK_ENABLED(NDHACK_OVERRIDE_CYCLE_M))
cycle_multiplier_active = 200;
// 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)
+ && cinfo[i+3].imm == 0x6e40 && dops[i+3].rs1 == 2)
{
SysPrintf("PE2 hack @%08x\n", start + (i+3)*4);
dops[i + 3].itype = NOP;
static noinline void pass1_disassemble(u_int pagelimit)
{
int i, j, done = 0, ni_count = 0;
- unsigned int type,op,op2;
+ unsigned int type,op,op2,op3;
for (i = 0; !done; i++)
{
+ int force_j_to_interpreter = 0;
memset(&dops[i], 0, sizeof(dops[i]));
- op2=0;
- minimum_free_regs[i]=0;
- dops[i].opcode=op=source[i]>>26;
+ memset(&cinfo[i], 0, sizeof(cinfo[i]));
+ cinfo[i].ba = -1;
+ cinfo[i].addr = -1;
+ dops[i].opcode = op = source[i] >> 26;
+ op2 = 0;
+ type = INTCALL;
+ set_mnemonic(i, "???");
switch(op)
{
- case 0x00: set_mnemonic(i, "special"); type=NI;
+ case 0x00: set_mnemonic(i, "special");
op2=source[i]&0x3f;
switch(op2)
{
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: set_mnemonic(i, "regimm"); type=NI;
- op2=(source[i]>>16)&0x1f;
+ case 0x01: set_mnemonic(i, "regimm");
+ type = SJUMP;
+ op2 = (source[i] >> 16) & 0x1f;
switch(op2)
{
- 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;
+ case 0x10: set_mnemonic(i, "BLTZAL"); break;
+ case 0x11: set_mnemonic(i, "BGEZAL"); break;
+ default:
+ if (op2 & 1)
+ set_mnemonic(i, "BGEZ");
+ else
+ set_mnemonic(i, "BLTZ");
}
break;
case 0x02: set_mnemonic(i, "J"); type=UJUMP; 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;
+ case 0x10: set_mnemonic(i, "COP0");
+ op2 = (source[i]>>21) & 0x1f;
+ if (op2 & 0x10) {
+ op3 = source[i] & 0x1f;
+ switch (op3)
+ {
+ case 0x01: case 0x02: case 0x06: case 0x08: type = INTCALL; break;
+ case 0x10: set_mnemonic(i, "RFE"); type=RFE; break;
+ default: type = OTHER; break;
+ }
+ break;
+ }
switch(op2)
{
- case 0x00: set_mnemonic(i, "MFC0"); type=COP0; break;
- case 0x02: set_mnemonic(i, "CFC0"); type=COP0; break;
+ u32 rd;
+ case 0x00:
+ set_mnemonic(i, "MFC0");
+ rd = (source[i] >> 11) & 0x1F;
+ if (!(0x00000417u & (1u << rd)))
+ 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;
+ case 0x02:
+ case 0x06: type = INTCALL; break;
+ default: type = OTHER; break;
}
break;
- case 0x11: set_mnemonic(i, "cop1"); type=COP1;
+ case 0x11: set_mnemonic(i, "COP1");
op2=(source[i]>>21)&0x1f;
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: 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: 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: 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: 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
- case 0x12: set_mnemonic(i, "COP2"); type=NI;
+ case 0x12: set_mnemonic(i, "COP2");
op2=(source[i]>>21)&0x1f;
- //if (op2 & 0x10)
- if (source[i]&0x3f) { // use this hack to support old savestates with patched gte insns
+ if (op2 & 0x10) {
+ type = OTHER;
if (gte_handlers[source[i]&0x3f]!=NULL) {
#ifdef DISASM
if (gte_regnames[source[i]&0x3f]!=NULL)
else
snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
#endif
- type=C2OP;
+ type = C2OP;
}
}
else switch(op2)
case 0x06: set_mnemonic(i, "CTC2"); type=COP2; break;
}
break;
+ case 0x13: set_mnemonic(i, "COP3");
+ op2=(source[i]>>21)&0x1f;
+ 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;
+ 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;
+ case 0x2E: set_mnemonic(i, "SWR"); type=STORELR; break;
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);
+ case 0x3B:
+ if (Config.HLE && (source[i] & 0x03ffffff) < ARRAY_SIZE(psxHLEt)) {
+ set_mnemonic(i, "HLECALL");
+ type = HLECALL;
+ }
+ break;
+ default:
break;
}
+ if (type == INTCALL)
+ SysPrintf("NI %08x @%08x (%08x)\n", source[i], start + i*4, start);
dops[i].itype=type;
dops[i].opcode2=op2;
/* Get registers/immediates */
dops[i].use_lt1=0;
gte_rs[i]=gte_rt[i]=0;
+ dops[i].rs1 = 0;
+ dops[i].rs2 = 0;
+ dops[i].rt1 = 0;
+ dops[i].rt2 = 0;
switch(type) {
case LOAD:
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];
+ cinfo[i].imm=(short)source[i];
break;
case STORE:
case STORELR:
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];
+ cinfo[i].imm=(short)source[i];
break;
case LOADLR:
// LWL/LWR only load part of the register,
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];
+ cinfo[i].imm=(short)source[i];
break;
case IMM16:
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];
+ cinfo[i].imm=(unsigned short)source[i];
}else{
- imm[i]=(short)source[i];
+ cinfo[i].imm=(short)source[i];
}
break;
case UJUMP:
- 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) {
dops[i].rt1=31;
break;
case RJUMP:
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) {
dops[i].rt1=(source[i]>>11)&0x1f;
case CJUMP:
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
dops[i].rs2=0;
}
case SJUMP:
dops[i].rs1=(source[i]>>21)&0x1f;
dops[i].rs2=CCREG;
- dops[i].rt1=0;
- dops[i].rt2=0;
- if(op2&0x10) { // BxxAL
- dops[i].rt1=31;
+ if (op2 == 0x10 || op2 == 0x11) { // BxxAL
+ dops[i].rt1 = 31;
// NOTE: If the branch is not taken, r31 is still overwritten
}
break;
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:
dops[i].rs1=(source[i]>>21)&0x1f; // source
dops[i].rt2=LOREG;
break;
case MOV:
- 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
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:
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;
+ cinfo[i].imm=(source[i]>>6)&0x1f;
break;
case COP0:
- 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==0) dops[i].rt1=(source[i]>>16)&0x1F; // MFC0
+ if(op2==4) dops[i].rs1=(source[i]>>16)&0x1F; // MTC0
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:
- 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;
case 0x06: gte_rt[i]=1ll<<(gr+32); break; // CTC2
}
break;
- case C1LS:
- 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:
dops[i].rs1=(source[i]>>21)&0x1F;
- dops[i].rs2=0;
- dops[i].rt1=0;
- dops[i].rt2=0;
- imm[i]=(short)source[i];
+ cinfo[i].imm=(short)source[i];
if(op==0x32) gte_rt[i]=1ll<<((source[i]>>16)&0x1F); // LWC2
else gte_rs[i]=1ll<<((source[i]>>16)&0x1F); // SWC2
break;
case C2OP:
- 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
case HLECALL:
case INTCALL:
dops[i].rs1=CCREG;
- dops[i].rs2=0;
- dops[i].rt1=0;
- dops[i].rt2=0;
break;
default:
- dops[i].rs1=0;
- dops[i].rs2=0;
- dops[i].rt1=0;
- dops[i].rt2=0;
+ break;
}
/* Calculate branch target addresses */
if(type==UJUMP)
- ba[i]=((start+i*4+4)&0xF0000000)|(((unsigned int)source[i]<<6)>>4);
+ cinfo[i].ba=((start+i*4+4)&0xF0000000)|(((unsigned int)source[i]<<6)>>4);
else if(type==CJUMP&&dops[i].rs1==dops[i].rs2&&(op&1))
- ba[i]=start+i*4+8; // Ignore never taken branch
+ cinfo[i].ba=start+i*4+8; // Ignore never taken branch
else if(type==SJUMP&&dops[i].rs1==0&&!(op2&1))
- ba[i]=start+i*4+8; // Ignore never taken branch
+ cinfo[i].ba=start+i*4+8; // Ignore never taken branch
else if(type==CJUMP||type==SJUMP)
- ba[i]=start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14);
- else ba[i]=-1;
+ cinfo[i].ba=start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14);
/* simplify always (not)taken branches */
if (type == CJUMP && dops[i].rs1 == dops[i].rs2) {
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 */
+ dops[i].is_jump = type == RJUMP || type == UJUMP || type == CJUMP || type == SJUMP;
+ dops[i].is_ujump = type == RJUMP || type == UJUMP;
+ dops[i].is_load = type == LOAD || type == LOADLR || op == 0x32; // LWC2
+ dops[i].is_delay_load = (dops[i].is_load || (source[i] & 0xf3d00000) == 0x40000000); // MFC/CFC
+ dops[i].is_store = type == STORE || type == STORELR || op == 0x3a; // SWC2
+ dops[i].is_exception = type == SYSCALL || type == HLECALL || type == INTCALL;
+ dops[i].may_except = dops[i].is_exception || (type == ALU && (op2 == 0x20 || op2 == 0x22)) || op == 8;
+
+ if (((op & 0x37) == 0x21 || op == 0x25) // LH/SH/LHU
+ && ((cinfo[i].imm & 1) || Config.PreciseExceptions))
+ dops[i].may_except = 1;
+ if (((op & 0x37) == 0x23 || (op & 0x37) == 0x32) // LW/SW/LWC2/SWC2
+ && ((cinfo[i].imm & 3) || Config.PreciseExceptions))
+ dops[i].may_except = 1;
+
+ /* rare messy cases to just pass over to the interpreter */
if (i > 0 && dops[i-1].is_jump) {
- int do_in_intrp=0;
+ j = i - 1;
// branch in delay slot?
if (dops[i].is_jump) {
// don't handle first branch and call interpreter if it's hit
- SysPrintf("branch in delay slot @%08x (%08x)\n", start + i*4, start);
- do_in_intrp=1;
+ SysPrintf("branch in DS @%08x (%08x)\n", start + i*4, start);
+ force_j_to_interpreter = 1;
}
- // basic load delay detection
- else if((type==LOAD||type==LOADLR||type==COP0||type==COP2||type==C2LS)&&dops[i].rt1!=0) {
- int t=(ba[i-1]-start)/4;
+ // basic load delay detection through a branch
+ else if (dops[i].is_delay_load && dops[i].rt1 != 0) {
+ int t=(cinfo[i-1].ba-start)/4;
if(0 <= t && t < i &&(dops[i].rt1==dops[t].rs1||dops[i].rt1==dops[t].rs2)&&dops[t].itype!=CJUMP&&dops[t].itype!=SJUMP) {
// jump target wants DS result - potential load delay effect
- SysPrintf("load delay @%08x (%08x)\n", start + i*4, start);
- do_in_intrp=1;
+ SysPrintf("load delay in DS @%08x (%08x)\n", start + i*4, start);
+ force_j_to_interpreter = 1;
dops[t+1].bt=1; // expected return from interpreter
}
else if(i>=2&&dops[i-2].rt1==2&&dops[i].rt1==2&&dops[i].rs1!=2&&dops[i].rs2!=2&&dops[i-1].rs1!=2&&dops[i-1].rs2!=2&&
!(i>=3&&dops[i-3].is_jump)) {
// v0 overwrite like this is a sign of trouble, bail out
SysPrintf("v0 overwrite @%08x (%08x)\n", start + i*4, start);
- do_in_intrp=1;
+ force_j_to_interpreter = 1;
}
}
- 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
- }
+ }
+ else if (i > 0 && dops[i-1].is_delay_load && dops[i-1].rt1 != 0
+ && (dops[i].rs1 == dops[i-1].rt1 || dops[i].rs2 == dops[i-1].rt1)) {
+ SysPrintf("load delay @%08x (%08x)\n", start + i*4, start);
+ for (j = i - 1; j > 0 && dops[j-1].is_delay_load; j--)
+ if (dops[j-1].rt1 != dops[i-1].rt1)
+ break;
+ force_j_to_interpreter = 1;
+ }
+ if (force_j_to_interpreter) {
+ memset(&dops[j], 0, sizeof(dops[j]));
+ dops[j].itype = INTCALL;
+ dops[j].rs1 = CCREG;
+ cinfo[j].ba = -1;
+ done = 2;
+ i = j; // don't compile the problematic branch/load/etc
}
/* Is this the end of the block? */
if (i > 0 && dops[i-1].is_ujump) {
if (dops[i-1].rt1 == 0) { // not jal
- int found_bbranch = 0, t = (ba[i-1] - start) / 4;
+ int found_bbranch = 0, t = (cinfo[i-1].ba - start) / 4;
if ((u_int)(t - i) < 64 && start + (t+64)*4 < pagelimit) {
// scan for a branch back to i+1
for (j = t; j < t + 64; j++) {
// Don't get too close to the limit
if(i>MAXBLOCK/2) done=1;
}
- if (dops[i].itype == SYSCALL || dops[i].itype == HLECALL || dops[i].itype == INTCALL)
+ if (dops[i].itype == HLECALL)
+ stop = 1;
+ else if (dops[i].itype == INTCALL)
+ stop = 2;
+ else if (dops[i].is_exception)
done = stop_after_jal ? 1 : 2;
if (done == 2) {
// Does the block continue due to a branch?
for(j=i-1;j>=0;j--)
{
- if(ba[j]==start+i*4) done=j=0; // Branch into delay slot
- if(ba[j]==start+i*4+4) done=j=0;
- if(ba[j]==start+i*4+8) done=j=0;
+ if(cinfo[j].ba==start+i*4) done=j=0; // Branch into delay slot
+ if(cinfo[j].ba==start+i*4+4) done=j=0;
+ if(cinfo[j].ba==start+i*4+8) done=j=0;
}
}
//assert(i<MAXBLOCK-1);
assert(start+i*4<pagelimit);
if (i==MAXBLOCK-1) done=1;
// Stop if we're compiling junk
- if(dops[i].itype == NI && (++ni_count > 8 || dops[i].opcode == 0x11)) {
+ if (dops[i].itype == INTCALL && (++ni_count > 8 || dops[i].opcode == 0x11)) {
done=stop_after_jal=1;
SysPrintf("Disabled speculative precompilation\n");
}
// 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))
+ if(cinfo[i].ba<start || cinfo[i].ba>=(start+slen*4))
{
// Branch out of this block, flush all regs
u=1;
else
{
// Internal branch, flag target
- dops[(ba[i]-start)>>2].bt=1;
- if(ba[i]<=start+i*4) {
+ dops[(cinfo[i].ba-start)>>2].bt=1;
+ if(cinfo[i].ba<=start+i*4) {
// Backward branch
if(dops[i].is_ujump)
{
// 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);
+ pass2_unneeded_regs((cinfo[i].ba-start)>>2,i-1,r+1);
}else{
- unneeded_reg[(ba[i]-start)>>2]=1;
- gte_unneeded[(ba[i]-start)>>2]=gte_u_unknown;
+ unneeded_reg[(cinfo[i].ba-start)>>2]=1;
+ gte_unneeded[(cinfo[i].ba-start)>>2]=gte_u_unknown;
}
} /*else*/ if(1) {
if (dops[i].is_ujump)
{
// Unconditional branch
- u=unneeded_reg[(ba[i]-start)>>2];
- gte_u=gte_unneeded[(ba[i]-start)>>2];
+ u=unneeded_reg[(cinfo[i].ba-start)>>2];
+ gte_u=gte_unneeded[(cinfo[i].ba-start)>>2];
branch_unneeded_reg[i]=u;
// Merge in delay slot
u|=(1LL<<dops[i+1].rt1)|(1LL<<dops[i+1].rt2);
gte_u&=~gte_rs[i+1];
} else {
// Conditional branch
- b=unneeded_reg[(ba[i]-start)>>2];
- gte_b=gte_unneeded[(ba[i]-start)>>2];
+ b=unneeded_reg[(cinfo[i].ba-start)>>2];
+ gte_b=gte_unneeded[(cinfo[i].ba-start)>>2];
branch_unneeded_reg[i]=b;
// Branch delay slot
b|=(1LL<<dops[i+1].rt1)|(1LL<<dops[i+1].rt2);
}
}
}
- else if(dops[i].itype==SYSCALL||dops[i].itype==HLECALL||dops[i].itype==INTCALL)
+ else if(dops[i].may_except)
{
- // SYSCALL instruction (software interrupt)
+ // SYSCALL instruction, etc or conditional exception
u=1;
}
- else if(dops[i].itype==COP0 && (source[i]&0x3f)==0x18)
+ else if (dops[i].itype == RFE)
{
- // ERET instruction (return from interrupt)
u=1;
}
//u=1; // DEBUG
current.wasconst = 0;
current.isconst = 0;
current.loadedconst = 0;
- current.waswritten = 0;
+ //current.waswritten = 0;
int ds=0;
int cc=0;
int hr;
if(current.regmap[hr]==0) current.regmap[hr]=-1;
}
current.isconst=0;
- current.waswritten=0;
+ //current.waswritten=0;
}
memcpy(regmap_pre[i],current.regmap,sizeof(current.regmap));
delayslot_alloc(¤t,i+1);
//current.isconst=0; // DEBUG
ds=1;
- //printf("i=%d, isconst=%x\n",i,current.isconst);
break;
case RJUMP:
//current.isconst=0;
//current.isconst=0;
break;
case SJUMP:
- //current.isconst=0;
- //current.wasconst=0;
- //regs[i].wasconst=0;
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,dops[i].rs1);
- if (dops[i].rt1==31) { // BLTZAL/BGEZAL
+ 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
}
- 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.
+ 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].rs1 == 31) // overwrites it's own 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;
delayslot_alloc(¤t,i+1);
}
}
- else
- // Don't alloc the delay slot yet because we might not execute it
- 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,dops[i].rs1);
- }
ds=1;
//current.isconst=0;
break;
case COP0:
cop0_alloc(¤t,i);
break;
- case COP1:
+ case RFE:
+ rfe_alloc(¤t,i);
break;
case COP2:
cop2_alloc(¤t,i);
break;
- case C1LS:
- c1ls_alloc(¤t,i);
- break;
case C2LS:
c2ls_alloc(¤t,i);
break;
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)
+#if 0 // see do_store_smc_check()
+ 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)cinfo[i-1].imm<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)
+ if((dops[i].itype==STORE||dops[i].itype==STORELR||(dops[i].itype==C2LS&&dops[i].opcode==0x3a))&&(u_int)cinfo[i].imm>=0x800)
current.waswritten&=~(1<<dops[i].rs1);
+#endif
/* Branch post-alloc */
if(i>0)
}
break;
case SJUMP:
- //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);
memcpy(&branch_regs[i-1].regmap_entry,¤t.regmap,sizeof(current.regmap));
memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
}
- else
- // Alloc the delay slot in case the branch is taken
- 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<<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);
- 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));
- }
// FIXME: BLTZAL/BGEZAL
- if(dops[i-1].opcode2&0x10) { // BxxZAL
+ if ((dops[i-1].opcode2 & 0x1e) == 0x10) { // BxxZAL
alloc_reg(&branch_regs[i-1],i-1,31);
dirty_reg(&branch_regs[i-1],31);
}
dirty_reg(¤t,CCREG);
for(j=i-1;j>=0;j--)
{
- if(ba[j]==start+i*4+4) {
+ if(cinfo[j].ba==start+i*4+4) {
memcpy(current.regmap,branch_regs[j].regmap,sizeof(current.regmap));
current.dirty=branch_regs[j].dirty;
break;
}
}
while(j>=0) {
- if(ba[j]==start+i*4+4) {
+ if(cinfo[j].ba==start+i*4+4) {
for(hr=0;hr<HOST_REGS;hr++) {
if(current.regmap[hr]!=branch_regs[j].regmap[hr]) {
current.regmap[hr]=-1;
}
// Count cycles in between branches
- ccadj[i] = CLOCK_ADJUST(cc);
- if (i > 0 && (dops[i-1].is_jump || dops[i].itype == SYSCALL || dops[i].itype == HLECALL))
+ cinfo[i].ccadj = CLOCK_ADJUST(cc);
+ if (i > 0 && (dops[i-1].is_jump || dops[i].is_exception))
{
cc=0;
}
}
}
if(current.regmap[HOST_BTREG]==BTREG) current.regmap[HOST_BTREG]=-1;
- regs[i].waswritten=current.waswritten;
+ //regs[i].waswritten=current.waswritten;
}
}
__builtin_prefetch(regs[i-2].regmap);
if(dops[i].is_jump)
{
- if(ba[i]<start || ba[i]>=(start+slen*4))
+ if(cinfo[i].ba<start || cinfo[i].ba>=(start+slen*4))
{
// Branch out of this block, don't need anything
nr=0;
// Internal branch
// Need whatever matches the target
nr=0;
- int t=(ba[i]-start)>>2;
+ int t=(cinfo[i].ba-start)>>2;
for(hr=0;hr<HOST_REGS;hr++)
{
if(regs[i].regmap_entry[hr]>=0) {
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(dops[i].itype==COP0 && (source[i]&0x3f)==0x18)
+ else if (dops[i].is_exception)
{
- // ERET instruction (return from interrupt)
+ // SYSCALL instruction, etc
nr=0;
}
else // Non-branch
{
if(dops[i].itype==UJUMP||dops[i].itype==CJUMP||dops[i].itype==SJUMP)
{
- if(ba[i]>=start && ba[i]<(start+i*4))
+ if(cinfo[i].ba>=start && cinfo[i].ba<(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==STORE||dops[i+1].itype==STORELR
+ ||dops[i+1].itype==SHIFT
||dops[i+1].itype==COP2||dops[i+1].itype==C2LS||dops[i+1].itype==C2OP)
{
- int t=(ba[i]-start)>>2;
+ int t=(cinfo[i].ba-start)>>2;
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(dops[i].ooo) {
- if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i+1])
+ if(count_free_regs(regs[i].regmap)<=cinfo[i+1].min_free_regs)
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)<=cinfo[i+1].min_free_regs)
f_regmap[hr]=branch_regs[i].regmap[hr];
}
// Avoid dirty->clean transition
int r=f_regmap[hr];
for(j=t;j<=i;j++)
{
- //printf("Test %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
+ //printf("Test %x -> %x, %x %d/%d\n",start+i*4,cinfo[i].ba,start+j*4,hr,r);
if(r<34&&((unneeded_reg[j]>>r)&1)) break;
assert(r < 64);
if(regs[j].regmap[hr]==f_regmap[hr]&&f_regmap[hr]<TEMPREG) {
- //printf("Hit %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
+ //printf("Hit %x -> %x, %x %d/%d\n",start+i*4,cinfo[i].ba,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;
k=i;
while(k>1&®s[k-1].regmap[hr]==-1) {
- if(count_free_regs(regs[k-1].regmap)<=minimum_free_regs[k-1]) {
+ if(count_free_regs(regs[k-1].regmap)<=cinfo[k-1].min_free_regs) {
//printf("no free regs for store %x\n",start+(k-1)*4);
break;
}
if(dops[j].itype==CJUMP||dops[j].itype==SJUMP)
{
if(dops[j].ooo) {
- if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j+1])
+ if(count_free_regs(regs[j].regmap)<=cinfo[j+1].min_free_regs)
break;
}else{
- if(count_free_regs(branch_regs[j].regmap)<=minimum_free_regs[j+1])
+ if(count_free_regs(branch_regs[j].regmap)<=cinfo[j+1].min_free_regs)
break;
}
if(get_reg(branch_regs[j].regmap,f_regmap[hr])>=0) {
break;
}
}
- if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j]) {
+ if(count_free_regs(regs[j].regmap)<=cinfo[j].min_free_regs) {
//printf("No free regs for store %x\n",start+j*4);
break;
}
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]) {
+ if(count_free_regs(regs[j].regmap)<=cinfo[j].min_free_regs) {
//printf("no free regs for store %x\n",start+j*4);
break;
}
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]) {
+ if(count_free_regs(regs[k-1].regmap)<=cinfo[k-1].min_free_regs) {
//printf("no free regs for store %x\n",start+(k-1)*4);
break;
}
}
}
}
- if(dops[i].itype!=STORE&&dops[i].itype!=STORELR&&dops[i].itype!=C1LS&&dops[i].itype!=SHIFT&&
+ if(dops[i].itype!=STORE&&dops[i].itype!=STORELR&&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)
+ dops[i].itype!=IMM16&&dops[i].itype!=LOAD)
{
memcpy(f_regmap,regs[i].regmap,sizeof(f_regmap));
}
{
if(!dops[i+1].bt)
{
+ int j, can_steal = 1;
+ for (j = i; j < i + 2; j++) {
+ int free_regs = 0;
+ if (cinfo[j].min_free_regs == 0)
+ continue;
+ for (hr = 0; hr < HOST_REGS; hr++)
+ if (hr != EXCLUDE_REG && regs[j].regmap[hr] < 0)
+ free_regs++;
+ if (free_regs <= cinfo[j].min_free_regs) {
+ can_steal = 0;
+ break;
+ }
+ }
+ if (!can_steal)
+ continue;
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))
+ ||(dops[i].itype==COP2&&dops[i].opcode2<3))
{
if(dops[i+1].rs1) {
if((hr=get_reg(regs[i+1].regmap,dops[i+1].rs1))>=0)
}
// Preload target address for load instruction (non-constant)
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((hr=get_reg_w(regs[i+1].regmap, dops[i+1].rt1))>=0)
{
if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<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((hr=get_reg_w(regs[i+1].regmap, dops[i+1].rt1))>=0)
{
if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0)
{
}
}
// Address for store instruction (non-constant)
- 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 (dops[i+1].is_store) { // SB/SH/SW/SWC2
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_temp(regs[i+1].regmap);
else {
regs[i+1].regmap[hr]=AGEN1+((i+1)&1);
regs[i+1].isconst&=~(1<<hr);
+ regs[i+1].dirty&=~(1<<hr);
+ regs[i+2].wasdirty&=~(1<<hr);
}
assert(hr>=0);
if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0)
}
}
}
- if(dops[i+1].itype==LOADLR||(dops[i+1].opcode&0x3b)==0x31||(dops[i+1].opcode&0x3b)==0x32) { // LWC1/LDC1, LWC2/LDC2
+ if (dops[i+1].itype == LOADLR || dops[i+1].opcode == 0x32) { // LWC2
if(get_reg(regs[i+1].regmap,dops[i+1].rs1)<0) {
int nr;
hr=get_reg(regs[i+1].regmap,FTEMP);
}
}
}
- 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*/) {
+ 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==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_w(regs[i+1].regmap, dops[i+1].rt1);
+ if (dops[i+1].itype == LOADLR || dops[i+1].opcode == 0x32) // LWC2
hr=get_reg(regs[i+1].regmap,FTEMP);
- if(dops[i+1].itype==STORE||dops[i+1].itype==STORELR||(dops[i+1].opcode&0x3b)==0x39||(dops[i+1].opcode&0x3b)==0x3a) { // SWC1/SDC1/SWC2/SDC2
+ if (dops[i+1].is_store) {
hr=get_reg(regs[i+1].regmap,AGEN1+((i+1)&1));
if(hr<0) hr=get_reg_temp(regs[i+1].regmap);
}
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))
+ if(cinfo[i].ba<start || cinfo[i].ba>=(start+slen*4))
{
// Branch out of this block, flush all regs
will_dirty_i = 0;
else
{
// Internal branch
- if(ba[i]<=start+i*4) {
+ if(cinfo[i].ba<=start+i*4) {
// Backward branch
if (dops[i].is_ujump)
{
if(wr) {
will_dirty[i]=temp_will_dirty;
wont_dirty[i]=temp_wont_dirty;
- pass6_clean_registers((ba[i]-start)>>2,i-1,0);
+ pass6_clean_registers((cinfo[i].ba-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;
+ will_dirty[(cinfo[i].ba-start)>>2]=0;
+ wont_dirty[(cinfo[i].ba-start)>>2]=-1;
}
}
/*else*/ if(1)
// Unconditional branch
will_dirty_i=0;
wont_dirty_i=0;
- //if(ba[i]>start+i*4) { // Disable recursion (for debugging)
+ //if(cinfo[i].ba>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]==regs[(cinfo[i].ba-start)>>2].regmap_entry[r]) {
+ will_dirty_i|=will_dirty[(cinfo[i].ba-start)>>2]&(1<<r);
+ wont_dirty_i|=wont_dirty[(cinfo[i].ba-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;
+ will_dirty_i|=((unneeded_reg[(cinfo[i].ba-start)>>2]>>branch_regs[i].regmap[r])&1)<<r;
+ wont_dirty_i|=((unneeded_reg[(cinfo[i].ba-start)>>2]>>branch_regs[i].regmap[r])&1)<<r;
}
}
}
// Conditional branch
will_dirty_i=will_dirty_next;
wont_dirty_i=wont_dirty_next;
- //if(ba[i]>start+i*4) // Disable recursion (for debugging)
+ //if(cinfo[i].ba>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);
+ if(target_reg==regs[(cinfo[i].ba-start)>>2].regmap_entry[r]) {
+ will_dirty_i&=will_dirty[(cinfo[i].ba-start)>>2]&(1<<r);
+ wont_dirty_i|=wont_dirty[(cinfo[i].ba-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;
+ will_dirty_i&=((unneeded_reg[(cinfo[i].ba-start)>>2]>>target_reg)&1)<<r;
+ wont_dirty_i|=((unneeded_reg[(cinfo[i].ba-start)>>2]>>target_reg)&1)<<r;
}
}
}
}
}
}
- 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)
+ else if (dops[i].is_exception)
{
- // ERET instruction (return from interrupt)
+ // SYSCALL instruction, etc
will_dirty_i=0;
wont_dirty_i=0;
}
u_int block_i = expirep / step & (PAGE_COUNT - 1);
u_int phase = (expirep >> (base_shift - 1)) & 1u;
if (!(expirep & (MAX_OUTPUT_BLOCK_SIZE / 2 - 1))) {
- inv_debug("EXP: base_offs %x/%x phase %u\n", base_offs,
- out - ndrc->translation_cache phase);
+ inv_debug("EXP: base_offs %x/%lx phase %u\n", base_offs,
+ (long)(out - ndrc->translation_cache), phase);
}
if (!phase) {
}
}
else
- ll_remove_matching_addrs(&jump_out[block_i], base_offs, base_shift);
+ unlink_jumps_tc_range(jumps[block_i], base_offs, base_shift);
}
}
block->tc_offs = beginning - ndrc->translation_cache;
//block->tc_len = out - beginning;
block->is_dirty = 0;
+ block->inv_near_misses = 0;
block->jump_in_cnt = jump_in_count;
// insert sorted by start mirror-unmasked vaddr
assem_debug("NOTCOMPILED: addr = %x -> %p\n", addr, out);
+ if (addr & 3) {
+ if (addr != hack_addr) {
+ SysPrintf("game crash @%08x, ra=%08x\n", addr, psxRegs.GPR.n.ra);
+ hack_addr = addr;
+ }
+ return -1;
+ }
+
// this is just for speculation
for (i = 1; i < 32; i++) {
if ((psxRegs.GPR.r[i] & 0xffff0000) == 0x1f800000)
state_rflags |= 1 << i;
}
- assert(!(addr & 3));
- start = addr & ~3;
+ start = addr;
new_dynarec_did_compile=1;
if (Config.HLE && start == 0x80001000) // hlecall
{
return 0;
}
- cycle_multiplier_active = cycle_multiplier_override && cycle_multiplier == CYCLE_MULT_DEFAULT
- ? cycle_multiplier_override : cycle_multiplier;
+ cycle_multiplier_active = Config.cycle_multiplier_override && Config.cycle_multiplier == CYCLE_MULT_DEFAULT
+ ? Config.cycle_multiplier_override : Config.cycle_multiplier;
source = get_source_start(start, &pagelimit);
if (source == NULL) {
// branch target entry point
instr_addr[i] = out;
assem_debug("<->\n");
- drc_dbg_emit_do_cmp(i, ccadj[i]);
+ drc_dbg_emit_do_cmp(i, cinfo[i].ccadj);
if (clear_hack_addr) {
emit_movimm(0, 0);
emit_writeword(0, &hack_addr);
if (dops[i].is_store)
load_reg(regs[i].regmap_entry,regs[i].regmap,INVCP);
- ds = assemble(i, ®s[i], ccadj[i]);
+ ds = assemble(i, ®s[i], cinfo[i].ccadj);
if (dops[i].is_ujump)
literal_pool(1024);
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, ccadj[i-1] + CLOCK_ADJUST(1), HOST_CCREG);
+ emit_addimm(HOST_CCREG, cinfo[i-1].ccadj + CLOCK_ADJUST(1), HOST_CCREG);
}
else
{
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, ccadj[i-1] + CLOCK_ADJUST(1), HOST_CCREG);
+ emit_addimm(HOST_CCREG, cinfo[i-1].ccadj + CLOCK_ADJUST(1), HOST_CCREG);
add_to_linker(out,start+i*4,0);
emit_jmp(0);
}
- // TODO: delay slot stubs?
// Stubs
- for(i=0;i<stubcount;i++)
+ for(i = 0; i < stubcount; i++)
{
switch(stubs[i].type)
{
case LOADB_STUB:
case LOADH_STUB:
case LOADW_STUB:
- case LOADD_STUB:
case LOADBU_STUB:
case LOADHU_STUB:
do_readstub(i);break;
case STOREB_STUB:
case STOREH_STUB:
case STOREW_STUB:
- case STORED_STUB:
do_writestub(i);break;
case CC_STUB:
do_ccstub(i);break;
case INVCODE_STUB:
do_invstub(i);break;
- case FP_STUB:
- do_cop1stub(i);break;
case STORELR_STUB:
do_unalignedwritestub(i);break;
+ case OVERFLOW_STUB:
+ do_overflowstub(i); break;
+ case ALIGNMENT_STUB:
+ do_alignmentstub(i); break;
+ default:
+ assert(0);
}
}
notaz.gp2x.de / pcsx_rearmed.git / blobdiff