#endif
#include "new_dynarec_config.h"
-#include "../psxhle.h" //emulator interface
+#include "../psxhle.h"
+#include "../psxinterpreter.h"
#include "emu_if.h" //emulator interface
+#define noinline __attribute__((noinline,noclone))
#ifndef ARRAY_SIZE
#define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
#endif
#define MAXBLOCK 4096
#define MAX_OUTPUT_BLOCK_SIZE 262144
+struct ndrc_mem
+{
+ u_char translation_cache[1 << TARGET_SIZE_2];
+ struct
+ {
+ struct tramp_insns ops[2048 / sizeof(struct tramp_insns)];
+ const void *f[2048 / sizeof(void *)];
+ } tramp;
+};
+
+#ifdef BASE_ADDR_DYNAMIC
+static struct ndrc_mem *ndrc;
+#else
+static struct ndrc_mem ndrc_ __attribute__((aligned(4096)));
+static struct ndrc_mem *ndrc = &ndrc_;
+#endif
+
// stubs
enum stub_type {
CC_STUB = 1,
static char ooo[MAXBLOCK];
static uint64_t unneeded_reg[MAXBLOCK];
static uint64_t branch_unneeded_reg[MAXBLOCK];
- static signed char regmap_pre[MAXBLOCK][HOST_REGS];
+ static signed char regmap_pre[MAXBLOCK][HOST_REGS]; // pre-instruction i?
static uint64_t current_constmap[HOST_REGS];
static uint64_t constmap[MAXBLOCK][HOST_REGS];
static struct regstat regs[MAXBLOCK];
extern int pcaddr;
extern int pending_exception;
extern int branch_target;
- extern u_int mini_ht[32][2];
+ extern uintptr_t mini_ht[32][2];
extern u_char restore_candidate[512];
/* registers that may be allocated */
#define DJT_2 (void *)2l
// asm linkage
-int new_recompile_block(int addr);
+int new_recompile_block(u_int addr);
void *get_addr_ht(u_int vaddr);
void invalidate_block(u_int block);
void invalidate_addr(u_int addr);
void cc_interrupt();
void fp_exception();
void fp_exception_ds();
-void jump_syscall_hle();
-void jump_hlecall();
-void jump_intcall();
+void jump_to_new_pc();
void new_dyna_leave();
// Needed by assembler
static void load_regs_entry(int t);
static void load_all_consts(signed char regmap[],u_int dirty,int i);
-static int verify_dirty(u_int *ptr);
+static int verify_dirty(const u_int *ptr);
static int get_final_value(int hr, int i, int *value);
static void add_stub(enum stub_type type, void *addr, void *retaddr,
u_int a, uintptr_t b, uintptr_t c, u_int d, u_int e);
static void *get_direct_memhandler(void *table, u_int addr,
enum stub_type type, uintptr_t *addr_host);
static void pass_args(int a0, int a1);
+static void emit_far_jump(const void *f);
+static void emit_far_call(const void *f);
static void mprotect_w_x(void *start, void *end, int is_x)
{
static void *start_block(void)
{
u_char *end = out + MAX_OUTPUT_BLOCK_SIZE;
- if (end > translation_cache + (1<<TARGET_SIZE_2))
- end = translation_cache + (1<<TARGET_SIZE_2);
+ if (end > ndrc->translation_cache + sizeof(ndrc->translation_cache))
+ end = ndrc->translation_cache + sizeof(ndrc->translation_cache);
start_tcache_write(out, end);
return out;
}
// Get address from virtual address
// This is called from the recompiled JR/JALR instructions
-void *get_addr(u_int vaddr)
+void noinline *get_addr(u_int vaddr)
{
u_int page=get_page(vaddr);
u_int vpage=get_vpage(vaddr);
for (hr=0;hr<HOST_REGS;hr++) regmap[hr]=-1;
}
-signed char get_reg(signed char regmap[],int r)
+static signed char get_reg(const signed char regmap[],int r)
{
int hr;
for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&®map[hr]==r) return hr;
}
// Find a register that is available for two consecutive cycles
-signed char get_reg2(signed char regmap1[],signed char regmap2[],int r)
+static signed char get_reg2(signed char regmap1[], const signed char regmap2[], int r)
{
int hr;
for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&®map1[hr]==r&®map2[hr]==r) return hr;
}
}
+#ifndef NDEBUG
+static int host_tempreg_in_use;
+
+static void host_tempreg_acquire(void)
+{
+ assert(!host_tempreg_in_use);
+ host_tempreg_in_use = 1;
+}
+
+static void host_tempreg_release(void)
+{
+ host_tempreg_in_use = 0;
+}
+#else
+static void host_tempreg_acquire(void) {}
+static void host_tempreg_release(void) {}
+#endif
+
#ifdef DRC_DBG
extern void gen_interupt();
extern void do_insn_cmp();
-#define FUNCNAME(f) { (intptr_t)f, " " #f }
+#define FUNCNAME(f) { f, " " #f }
static const struct {
- intptr_t addr;
+ void *addr;
const char *name;
} function_names[] = {
FUNCNAME(cc_interrupt),
FUNCNAME(jump_handler_write16),
FUNCNAME(jump_handler_write32),
FUNCNAME(invalidate_addr),
- FUNCNAME(verify_code),
- FUNCNAME(jump_hlecall),
- FUNCNAME(jump_syscall_hle),
+ FUNCNAME(jump_to_new_pc),
FUNCNAME(new_dyna_leave),
FUNCNAME(pcsx_mtc0),
FUNCNAME(pcsx_mtc0_ds),
FUNCNAME(do_insn_cmp),
+#ifdef __arm__
+ FUNCNAME(verify_code),
+#endif
};
-static const char *func_name(intptr_t a)
+static const char *func_name(const void *a)
{
int i;
for (i = 0; i < sizeof(function_names)/sizeof(function_names[0]); i++)
#include "assem_arm64.c"
#endif
+static void *get_trampoline(const void *f)
+{
+ size_t i;
+
+ for (i = 0; i < ARRAY_SIZE(ndrc->tramp.f); i++) {
+ if (ndrc->tramp.f[i] == f || ndrc->tramp.f[i] == NULL)
+ break;
+ }
+ if (i == ARRAY_SIZE(ndrc->tramp.f)) {
+ SysPrintf("trampoline table is full, last func %p\n", f);
+ abort();
+ }
+ if (ndrc->tramp.f[i] == NULL) {
+ start_tcache_write(&ndrc->tramp.f[i], &ndrc->tramp.f[i + 1]);
+ ndrc->tramp.f[i] = f;
+ end_tcache_write(&ndrc->tramp.f[i], &ndrc->tramp.f[i + 1]);
+ }
+ return &ndrc->tramp.ops[i];
+}
+
+static void emit_far_jump(const void *f)
+{
+ if (can_jump_or_call(f)) {
+ emit_jmp(f);
+ return;
+ }
+
+ f = get_trampoline(f);
+ emit_jmp(f);
+}
+
+static void emit_far_call(const void *f)
+{
+ if (can_jump_or_call(f)) {
+ emit_call(f);
+ return;
+ }
+
+ f = get_trampoline(f);
+ emit_call(f);
+}
+
// Add virtual address mapping to linked list
void ll_add(struct ll_entry **head,int vaddr,void *addr)
{
}
// This is called when we write to a compiled block (see do_invstub)
-void invalidate_page(u_int page)
+static void invalidate_page(u_int page)
{
struct ll_entry *head;
struct ll_entry *next;
#endif
}
+static void do_invstub(int n)
+{
+ literal_pool(20);
+ u_int reglist=stubs[n].a;
+ set_jump_target(stubs[n].addr, out);
+ save_regs(reglist);
+ if(stubs[n].b!=0) emit_mov(stubs[n].b,0);
+ emit_far_call(invalidate_addr);
+ restore_regs(reglist);
+ emit_jmp(stubs[n].retaddr); // return address
+}
+
// Add an entry to jump_out after making a link
+// src should point to code by emit_extjump2()
void add_link(u_int vaddr,void *src)
{
u_int page=get_page(vaddr);
inv_debug("add_link: %p -> %x (%d)\n",src,vaddr,page);
- int *ptr=(int *)(src+4);
- assert((*ptr&0x0fff0000)==0x059f0000);
- (void)ptr;
+ check_extjump2(src);
ll_add(jump_out+page,vaddr,src);
//void *ptr=get_pointer(src);
//inv_debug("add_link: Pointer is to %p\n",ptr);
static void add_stub(enum stub_type type, void *addr, void *retaddr,
u_int a, uintptr_t b, uintptr_t c, u_int d, u_int e)
{
- assert(a < ARRAY_SIZE(stubs));
+ assert(stubcount < ARRAY_SIZE(stubs));
stubs[stubcount].type = type;
stubs[stubcount].addr = addr;
stubs[stubcount].retaddr = retaddr;
}
#ifndef shift_assemble
-void shift_assemble(int i,struct regstat *i_regs)
+static void shift_assemble(int i,struct regstat *i_regs)
{
- printf("Need shift_assemble for this architecture.\n");
- abort();
+ signed char s,t,shift;
+ if (rt1[i] == 0)
+ return;
+ assert(opcode2[i]<=0x07); // SLLV/SRLV/SRAV
+ t = get_reg(i_regs->regmap, rt1[i]);
+ s = get_reg(i_regs->regmap, rs1[i]);
+ shift = get_reg(i_regs->regmap, rs2[i]);
+ if (t < 0)
+ return;
+
+ if(rs1[i]==0)
+ emit_zeroreg(t);
+ else if(rs2[i]==0) {
+ assert(s>=0);
+ if(s!=t) emit_mov(s,t);
+ }
+ else {
+ host_tempreg_acquire();
+ emit_andimm(shift,31,HOST_TEMPREG);
+ switch(opcode2[i]) {
+ case 4: // SLLV
+ emit_shl(s,HOST_TEMPREG,t);
+ break;
+ case 6: // SRLV
+ emit_shr(s,HOST_TEMPREG,t);
+ break;
+ case 7: // SRAV
+ emit_sar(s,HOST_TEMPREG,t);
+ break;
+ default:
+ assert(0);
+ }
+ host_tempreg_release();
+ }
}
+
#endif
enum {
}
if(type==MTYPE_8020) { // RAM 80200000+ mirror
+ host_tempreg_acquire();
emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
addr=*addr_reg_override=HOST_TEMPREG;
type=0;
}
else if(type==MTYPE_0000) { // RAM 0 mirror
+ host_tempreg_acquire();
emit_orimm(addr,0x80000000,HOST_TEMPREG);
addr=*addr_reg_override=HOST_TEMPREG;
type=0;
}
else if(type==MTYPE_A000) { // RAM A mirror
+ host_tempreg_acquire();
emit_andimm(addr,~0x20000000,HOST_TEMPREG);
addr=*addr_reg_override=HOST_TEMPREG;
type=0;
}
else if(type==MTYPE_1F80) { // scratchpad
if (psxH == (void *)0x1f800000) {
- emit_addimm(addr,-0x1f800000,HOST_TEMPREG);
+ host_tempreg_acquire();
+ emit_xorimm(addr,0x1f800000,HOST_TEMPREG);
emit_cmpimm(HOST_TEMPREG,0x1000);
+ host_tempreg_release();
jaddr=out;
emit_jc(0);
}
#endif
emit_jno(0);
if(ram_offset!=0) {
+ host_tempreg_acquire();
emit_addimm(addr,ram_offset,HOST_TEMPREG);
addr=*addr_reg_override=HOST_TEMPREG;
}
int offset;
void *jaddr=0;
int memtarget=0,c=0;
- int fastload_reg_override=0;
+ int fastio_reg_override=-1;
u_int hr,reglist=0;
tl=get_reg(i_regs->regmap,rt1[i]);
s=get_reg(i_regs->regmap,rs1[i]);
if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
#endif
{
- jaddr=emit_fastpath_cmp_jump(i,addr,&fastload_reg_override);
+ jaddr=emit_fastpath_cmp_jump(i,addr,&fastio_reg_override);
}
}
else if(ram_offset&&memtarget) {
+ host_tempreg_acquire();
emit_addimm(addr,ram_offset,HOST_TEMPREG);
- fastload_reg_override=HOST_TEMPREG;
+ fastio_reg_override=HOST_TEMPREG;
}
int dummy=(rt1[i]==0)||(tl!=get_reg(i_regs->regmap,rt1[i])); // ignore loads to r0 and unneeded reg
if (opcode[i]==0x20) { // LB
{
int x=0,a=tl;
if(!c) a=addr;
- if(fastload_reg_override) a=fastload_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_movsbl_indexed(x,a,tl);
}
if(!dummy) {
int x=0,a=tl;
if(!c) a=addr;
- if(fastload_reg_override) a=fastload_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_movswl_indexed(x,a,tl);
}
if(jaddr)
if(!c||memtarget) {
if(!dummy) {
int a=addr;
- if(fastload_reg_override) a=fastload_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_readword_indexed(0,a,tl);
}
if(jaddr)
if(!dummy) {
int x=0,a=tl;
if(!c) a=addr;
- if(fastload_reg_override) a=fastload_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_movzbl_indexed(x,a,tl);
}
if(!dummy) {
int x=0,a=tl;
if(!c) a=addr;
- if(fastload_reg_override) a=fastload_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_movzwl_indexed(x,a,tl);
}
if(jaddr)
assert(0);
}
}
+ if (fastio_reg_override == HOST_TEMPREG)
+ host_tempreg_release();
}
#ifndef loadlr_assemble
-void loadlr_assemble(int i,struct regstat *i_regs)
+static void loadlr_assemble(int i,struct regstat *i_regs)
{
- printf("Need loadlr_assemble for this architecture.\n");
- abort();
+ int s,tl,temp,temp2,addr;
+ int offset;
+ void *jaddr=0;
+ int memtarget=0,c=0;
+ int fastio_reg_override=-1;
+ u_int hr,reglist=0;
+ tl=get_reg(i_regs->regmap,rt1[i]);
+ s=get_reg(i_regs->regmap,rs1[i]);
+ temp=get_reg(i_regs->regmap,-1);
+ temp2=get_reg(i_regs->regmap,FTEMP);
+ addr=get_reg(i_regs->regmap,AGEN1+(i&1));
+ assert(addr<0);
+ offset=imm[i];
+ for(hr=0;hr<HOST_REGS;hr++) {
+ if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+ }
+ reglist|=1<<temp;
+ if(offset||s<0||c) addr=temp2;
+ else addr=s;
+ if(s>=0) {
+ c=(i_regs->wasconst>>s)&1;
+ if(c) {
+ memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
+ }
+ }
+ if(!c) {
+ emit_shlimm(addr,3,temp);
+ if (opcode[i]==0x22||opcode[i]==0x26) {
+ emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR
+ }else{
+ emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR
+ }
+ jaddr=emit_fastpath_cmp_jump(i,temp2,&fastio_reg_override);
+ }
+ else {
+ if(ram_offset&&memtarget) {
+ host_tempreg_acquire();
+ emit_addimm(temp2,ram_offset,HOST_TEMPREG);
+ fastio_reg_override=HOST_TEMPREG;
+ }
+ if (opcode[i]==0x22||opcode[i]==0x26) {
+ emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
+ }else{
+ emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR
+ }
+ }
+ if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR
+ if(!c||memtarget) {
+ int a=temp2;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
+ emit_readword_indexed(0,a,temp2);
+ if(fastio_reg_override==HOST_TEMPREG) host_tempreg_release();
+ if(jaddr) add_stub_r(LOADW_STUB,jaddr,out,i,temp2,i_regs,ccadj[i],reglist);
+ }
+ else
+ inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist);
+ if(rt1[i]) {
+ assert(tl>=0);
+ emit_andimm(temp,24,temp);
+ if (opcode[i]==0x22) // LWL
+ emit_xorimm(temp,24,temp);
+ host_tempreg_acquire();
+ emit_movimm(-1,HOST_TEMPREG);
+ if (opcode[i]==0x26) {
+ emit_shr(temp2,temp,temp2);
+ emit_bic_lsr(tl,HOST_TEMPREG,temp,tl);
+ }else{
+ emit_shl(temp2,temp,temp2);
+ emit_bic_lsl(tl,HOST_TEMPREG,temp,tl);
+ }
+ host_tempreg_release();
+ emit_or(temp2,tl,tl);
+ }
+ //emit_storereg(rt1[i],tl); // DEBUG
+ }
+ if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR
+ assert(0);
+ }
}
#endif
enum stub_type type;
int memtarget=0,c=0;
int agr=AGEN1+(i&1);
- int faststore_reg_override=0;
+ int fastio_reg_override=-1;
u_int hr,reglist=0;
tl=get_reg(i_regs->regmap,rs2[i]);
s=get_reg(i_regs->regmap,rs1[i]);
if(offset||s<0||c) addr=temp;
else addr=s;
if(!c) {
- jaddr=emit_fastpath_cmp_jump(i,addr,&faststore_reg_override);
+ jaddr=emit_fastpath_cmp_jump(i,addr,&fastio_reg_override);
}
else if(ram_offset&&memtarget) {
+ host_tempreg_acquire();
emit_addimm(addr,ram_offset,HOST_TEMPREG);
- faststore_reg_override=HOST_TEMPREG;
+ fastio_reg_override=HOST_TEMPREG;
}
if (opcode[i]==0x28) { // SB
if(!c||memtarget) {
int x=0,a=temp;
if(!c) a=addr;
- if(faststore_reg_override) a=faststore_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_writebyte_indexed(tl,x,a);
}
type=STOREB_STUB;
if(!c||memtarget) {
int x=0,a=temp;
if(!c) a=addr;
- if(faststore_reg_override) a=faststore_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_writehword_indexed(tl,x,a);
}
type=STOREH_STUB;
if (opcode[i]==0x2B) { // SW
if(!c||memtarget) {
int a=addr;
- if(faststore_reg_override) a=faststore_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_writeword_indexed(tl,0,a);
}
type=STOREW_STUB;
assert(0);
type=STORED_STUB;
}
+ if(fastio_reg_override==HOST_TEMPREG)
+ host_tempreg_release();
if(jaddr) {
// PCSX store handlers don't check invcode again
reglist|=1<<addr;
}
// basic current block modification detection..
// not looking back as that should be in mips cache already
+ // (see Spyro2 title->attract mode)
if(c&&start+i*4<addr_val&&addr_val<start+slen*4) {
SysPrintf("write to %08x hits block %08x, pc=%08x\n",addr_val,start,start+i*4);
assert(i_regs->regmap==regs[i].regmap); // not delay slot
wb_dirtys(regs[i].regmap_entry,regs[i].wasdirty);
emit_movimm(start+i*4+4,0);
emit_writeword(0,&pcaddr);
- emit_jmp(do_interrupt);
+ emit_addimm(HOST_CCREG,2,HOST_CCREG);
+ emit_far_call(get_addr_ht);
+ emit_jmpreg(0);
}
}
}
-void storelr_assemble(int i,struct regstat *i_regs)
+static void storelr_assemble(int i,struct regstat *i_regs)
{
int s,tl;
int temp;
emit_jmp(0);
}
}
- emit_addimm_no_flags(ram_offset,temp);
+ if(ram_offset)
+ emit_addimm_no_flags(ram_offset,temp);
if (opcode[i]==0x2C||opcode[i]==0x2D) { // SDL/SDR
assert(0);
if (opcode[i]==0x2A) { // SWL
emit_writeword_indexed(tl,0,temp);
}
- if (opcode[i]==0x2E) { // SWR
+ else if (opcode[i]==0x2E) { // SWR
emit_writebyte_indexed(tl,3,temp);
}
- if (opcode[i]==0x2C) { // SDL
- assert(0);
- }
- if (opcode[i]==0x2D) { // SDR
+ else
assert(0);
- }
done0=out;
emit_jmp(0);
// 1
emit_writebyte_indexed(tl,1,temp);
if(rs2[i]) emit_rorimm(tl,8,tl);
}
- if (opcode[i]==0x2E) { // SWR
+ else if (opcode[i]==0x2E) { // SWR
// Write two lsb into two most significant bytes
emit_writehword_indexed(tl,1,temp);
}
- if (opcode[i]==0x2C) { // SDL
- assert(0);
- }
- if (opcode[i]==0x2D) { // SDR
- assert(0);
- }
done1=out;
emit_jmp(0);
// 2
emit_writehword_indexed(tl,-2,temp);
if(rs2[i]) emit_rorimm(tl,16,tl);
}
- if (opcode[i]==0x2E) { // SWR
+ else if (opcode[i]==0x2E) { // SWR
// Write 3 lsb into three most significant bytes
emit_writebyte_indexed(tl,-1,temp);
if(rs2[i]) emit_rorimm(tl,8,tl);
emit_writehword_indexed(tl,0,temp);
if(rs2[i]) emit_rorimm(tl,24,tl);
}
- if (opcode[i]==0x2C) { // SDL
- assert(0);
- }
- if (opcode[i]==0x2D) { // SDR
- assert(0);
- }
done2=out;
emit_jmp(0);
// 3
emit_writebyte_indexed(tl,-3,temp);
if(rs2[i]) emit_rorimm(tl,8,tl);
}
- if (opcode[i]==0x2E) { // SWR
+ else if (opcode[i]==0x2E) { // SWR
// Write entire word
emit_writeword_indexed(tl,-3,temp);
}
- if (opcode[i]==0x2C) { // SDL
- assert(0);
- }
- if (opcode[i]==0x2D) { // SDR
- assert(0);
- }
set_jump_target(done0, out);
set_jump_target(done1, out);
set_jump_target(done2, out);
- if (opcode[i]==0x2C) { // SDL
- assert(0);
- }
- if (opcode[i]==0x2D) { // SDR
- assert(0);
- }
if(!c||!memtarget)
add_stub_r(STORELR_STUB,jaddr,out,i,temp,i_regs,ccadj[i],reglist);
if(!(i_regs->waswritten&(1<<rs1[i]))&&!(new_dynarec_hacks&NDHACK_NO_SMC_CHECK)) {
emit_storereg(CCREG,HOST_CCREG);
emit_loadreg(rs1[i],1);
emit_movimm(copr,0);
- emit_call(pcsx_mtc0_ds);
+ emit_far_call(pcsx_mtc0_ds);
emit_loadreg(rs1[i],s);
return;
}
emit_movimm(0,HOST_TEMPREG);
emit_writeword(HOST_TEMPREG,&pending_exception);
}
- //else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12);
- //else
if(s==HOST_CCREG)
emit_loadreg(rs1[i],1);
else if(s!=1)
emit_mov(s,1);
emit_movimm(copr,0);
- emit_call(pcsx_mtc0);
+ emit_far_call(pcsx_mtc0);
if(copr==9||copr==11||copr==12||copr==13) {
emit_readword(&Count,HOST_CCREG);
emit_readword(&next_interupt,HOST_TEMPREG);
assert(!is_delayslot);
emit_readword(&pending_exception,14);
emit_test(14,14);
- emit_jne(&do_interrupt);
+ void *jaddr = out;
+ emit_jeq(0);
+ emit_readword(&pcaddr, 0);
+ emit_addimm(HOST_CCREG,2,HOST_CCREG);
+ emit_far_call(get_addr_ht);
+ emit_jmpreg(0);
+ set_jump_target(jaddr, out);
}
emit_loadreg(rs1[i],s);
}
if(regs[i].regmap_entry[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
emit_movimm(start+(i-ds)*4,EAX); // Get PC
emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right? There should probably be an extra cycle...
- emit_jmp(ds?fp_exception_ds:fp_exception);
+ emit_far_jump(ds?fp_exception_ds:fp_exception);
}
static void cop2_get_dreg(u_int copr,signed char tl,signed char temp)
break;
case 28:
case 29:
- emit_readword(®_cop2d[9],temp);
- emit_testimm(temp,0x8000); // do we need this?
- emit_andimm(temp,0xf80,temp);
- emit_andne_imm(temp,0,temp);
- emit_shrimm(temp,7,tl);
- emit_readword(®_cop2d[10],temp);
- emit_testimm(temp,0x8000);
- emit_andimm(temp,0xf80,temp);
- emit_andne_imm(temp,0,temp);
- emit_orrshr_imm(temp,2,tl);
- emit_readword(®_cop2d[11],temp);
- emit_testimm(temp,0x8000);
- emit_andimm(temp,0xf80,temp);
- emit_andne_imm(temp,0,temp);
- emit_orrshl_imm(temp,3,tl);
- emit_writeword(tl,®_cop2d[copr]);
+ c2op_mfc2_29_assemble(tl,temp);
break;
default:
emit_readword(®_cop2d[copr],tl);
emit_writeword(sl,®_cop2d[28]);
break;
case 30:
- emit_movs(sl,temp);
- emit_mvnmi(temp,temp);
+ emit_xorsar_imm(sl,sl,31,temp);
#if defined(HAVE_ARMV5) || defined(__aarch64__)
emit_clz(temp,temp);
#else
void *jaddr2=NULL;
enum stub_type type;
int agr=AGEN1+(i&1);
- int fastio_reg_override=0;
+ int fastio_reg_override=-1;
u_int hr,reglist=0;
u_int copr=(source[i]>>16)&0x1f;
s=get_reg(i_regs->regmap,rs1[i]);
assert(ar>=0);
if (opcode[i]==0x3a) { // SWC2
- cop2_get_dreg(copr,tl,HOST_TEMPREG);
+ cop2_get_dreg(copr,tl,-1);
type=STOREW_STUB;
}
else
jaddr2=emit_fastpath_cmp_jump(i,ar,&fastio_reg_override);
}
else if(ram_offset&&memtarget) {
+ host_tempreg_acquire();
emit_addimm(ar,ram_offset,HOST_TEMPREG);
fastio_reg_override=HOST_TEMPREG;
}
if (opcode[i]==0x32) { // LWC2
int a=ar;
- if(fastio_reg_override) a=fastio_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_readword_indexed(0,a,tl);
}
if (opcode[i]==0x3a) { // SWC2
if(!offset&&!c&&s>=0) emit_mov(s,ar);
#endif
int a=ar;
- if(fastio_reg_override) a=fastio_reg_override;
+ if(fastio_reg_override>=0) a=fastio_reg_override;
emit_writeword_indexed(tl,0,a);
}
}
+ if(fastio_reg_override==HOST_TEMPREG)
+ host_tempreg_release();
if(jaddr2)
add_stub_r(type,jaddr2,out,i,ar,i_regs,ccadj[i],reglist);
if(opcode[i]==0x3a) // SWC2
#endif
}
if (opcode[i]==0x32) { // LWC2
+ host_tempreg_acquire();
cop2_put_dreg(copr,tl,HOST_TEMPREG);
+ host_tempreg_release();
}
}
emit_signextend16(sl,temp);
break;
case 31:
- //value = value & 0x7ffff000;
- //if (value & 0x7f87e000) value |= 0x80000000;
- emit_shrimm(sl,12,temp);
- emit_shlimm(temp,12,temp);
- emit_testimm(temp,0x7f000000);
- emit_testeqimm(temp,0x00870000);
- emit_testeqimm(temp,0x0000e000);
- emit_orrne_imm(temp,0x80000000,temp);
+ c2op_ctc2_31_assemble(sl,temp);
break;
default:
temp=sl;
}
}
+static void do_unalignedwritestub(int n)
+{
+ assem_debug("do_unalignedwritestub %x\n",start+stubs[n].a*4);
+ literal_pool(256);
+ set_jump_target(stubs[n].addr, out);
+
+ int i=stubs[n].a;
+ struct regstat *i_regs=(struct regstat *)stubs[n].c;
+ int addr=stubs[n].b;
+ u_int reglist=stubs[n].e;
+ signed char *i_regmap=i_regs->regmap;
+ int temp2=get_reg(i_regmap,FTEMP);
+ int rt;
+ rt=get_reg(i_regmap,rs2[i]);
+ assert(rt>=0);
+ assert(addr>=0);
+ assert(opcode[i]==0x2a||opcode[i]==0x2e); // SWL/SWR only implemented
+ reglist|=(1<<addr);
+ reglist&=~(1<<temp2);
+
+#if 1
+ // don't bother with it and call write handler
+ save_regs(reglist);
+ pass_args(addr,rt);
+ int cc=get_reg(i_regmap,CCREG);
+ if(cc<0)
+ emit_loadreg(CCREG,2);
+ emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d+1),2);
+ emit_far_call((opcode[i]==0x2a?jump_handle_swl:jump_handle_swr));
+ emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d+1),cc<0?2:cc);
+ if(cc<0)
+ emit_storereg(CCREG,2);
+ restore_regs(reglist);
+ emit_jmp(stubs[n].retaddr); // return address
+#else
+ emit_andimm(addr,0xfffffffc,temp2);
+ emit_writeword(temp2,&address);
+
+ save_regs(reglist);
+ emit_shrimm(addr,16,1);
+ int cc=get_reg(i_regmap,CCREG);
+ if(cc<0) {
+ emit_loadreg(CCREG,2);
+ }
+ emit_movimm((u_int)readmem,0);
+ emit_addimm(cc<0?2:cc,2*stubs[n].d+2,2);
+ emit_call((int)&indirect_jump_indexed);
+ restore_regs(reglist);
+
+ emit_readword(&readmem_dword,temp2);
+ int temp=addr; //hmh
+ emit_shlimm(addr,3,temp);
+ emit_andimm(temp,24,temp);
+ if (opcode[i]==0x2a) // SWL
+ emit_xorimm(temp,24,temp);
+ emit_movimm(-1,HOST_TEMPREG);
+ if (opcode[i]==0x2a) { // SWL
+ emit_bic_lsr(temp2,HOST_TEMPREG,temp,temp2);
+ emit_orrshr(rt,temp,temp2);
+ }else{
+ emit_bic_lsl(temp2,HOST_TEMPREG,temp,temp2);
+ emit_orrshl(rt,temp,temp2);
+ }
+ emit_readword(&address,addr);
+ emit_writeword(temp2,&word);
+ //save_regs(reglist); // don't need to, no state changes
+ emit_shrimm(addr,16,1);
+ emit_movimm((u_int)writemem,0);
+ //emit_call((int)&indirect_jump_indexed);
+ emit_mov(15,14);
+ emit_readword_dualindexedx4(0,1,15);
+ emit_readword(&Count,HOST_TEMPREG);
+ emit_readword(&next_interupt,2);
+ emit_addimm(HOST_TEMPREG,-2*stubs[n].d-2,HOST_TEMPREG);
+ emit_writeword(2,&last_count);
+ emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc);
+ if(cc<0) {
+ emit_storereg(CCREG,HOST_TEMPREG);
+ }
+ restore_regs(reglist);
+ emit_jmp(stubs[n].retaddr); // return address
+#endif
+}
+
#ifndef multdiv_assemble
void multdiv_assemble(int i,struct regstat *i_regs)
{
}
}
-static void syscall_assemble(int i,struct regstat *i_regs)
+// call interpreter, exception handler, things that change pc/regs/cycles ...
+static void call_c_cpu_handler(int i, const struct regstat *i_regs, u_int pc, void *func)
{
signed char ccreg=get_reg(i_regs->regmap,CCREG);
assert(ccreg==HOST_CCREG);
assert(!is_delayslot);
(void)ccreg;
- emit_movimm(start+i*4,EAX); // Get PC
- emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right? There should probably be an extra cycle...
- emit_jmp(jump_syscall_hle); // XXX
+
+ emit_movimm(pc,3); // Get PC
+ emit_readword(&last_count,2);
+ emit_writeword(3,&psxRegs.pc);
+ emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // XXX
+ emit_add(2,HOST_CCREG,2);
+ emit_writeword(2,&psxRegs.cycle);
+ emit_far_call(func);
+ emit_far_jump(jump_to_new_pc);
+}
+
+static void syscall_assemble(int i,struct regstat *i_regs)
+{
+ emit_movimm(0x20,0); // cause code
+ emit_movimm(0,1); // not in delay slot
+ call_c_cpu_handler(i,i_regs,start+i*4,psxException);
}
static void hlecall_assemble(int i,struct regstat *i_regs)
{
- extern void psxNULL();
- signed char ccreg=get_reg(i_regs->regmap,CCREG);
- assert(ccreg==HOST_CCREG);
- assert(!is_delayslot);
- (void)ccreg;
- emit_movimm(start+i*4+4,0); // Get PC
+ void *hlefunc = psxNULL;
uint32_t hleCode = source[i] & 0x03ffffff;
- if (hleCode >= ARRAY_SIZE(psxHLEt))
- emit_movimm((uintptr_t)psxNULL,1);
- else
- emit_movimm((uintptr_t)psxHLEt[hleCode],1);
- emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // XXX
- emit_jmp(jump_hlecall);
+ if (hleCode < ARRAY_SIZE(psxHLEt))
+ hlefunc = psxHLEt[hleCode];
+
+ call_c_cpu_handler(i,i_regs,start+i*4+4,hlefunc);
}
static void intcall_assemble(int i,struct regstat *i_regs)
{
- signed char ccreg=get_reg(i_regs->regmap,CCREG);
- assert(ccreg==HOST_CCREG);
- assert(!is_delayslot);
- (void)ccreg;
- emit_movimm(start+i*4,0); // Get PC
- emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
- emit_jmp(jump_intcall);
+ call_c_cpu_handler(i,i_regs,start+i*4,execI);
}
static void speculate_mov(int rs,int rt)
static void drc_dbg_emit_do_cmp(int i)
{
extern void do_insn_cmp();
- extern int cycle;
+ //extern int cycle;
u_int hr,reglist=0;
for(hr=0;hr<HOST_REGS;hr++)
save_regs(reglist);
emit_movimm(start+i*4,0);
emit_writeword(0,&pcaddr);
- emit_call(do_insn_cmp);
+ emit_far_call(do_insn_cmp);
//emit_readword(&cycle,0);
//emit_addimm(0,2,0);
//emit_writeword(0,&cycle);
+ (void)get_reg2;
restore_regs(reglist);
}
#else
emit_extjump2(addr, target, dyna_linker_ds);
}
+// Load 2 immediates optimizing for small code size
+static void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2)
+{
+ emit_movimm(imm1,rt1);
+ emit_movimm_from(imm1,rt1,imm2,rt2);
+}
+
void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert)
{
int count;
static void do_ccstub(int n)
{
literal_pool(256);
- assem_debug("do_ccstub %lx\n",start+stubs[n].b*4);
+ assem_debug("do_ccstub %x\n",start+(u_int)stubs[n].b*4);
set_jump_target(stubs[n].addr, out);
int i=stubs[n].b;
if(stubs[n].d==NULLDS) {
// Update cycle count
assert(branch_regs[i].regmap[HOST_CCREG]==CCREG||branch_regs[i].regmap[HOST_CCREG]==-1);
if(stubs[n].a) emit_addimm(HOST_CCREG,CLOCK_ADJUST((signed int)stubs[n].a),HOST_CCREG);
- emit_call(cc_interrupt);
+ emit_far_call(cc_interrupt);
if(stubs[n].a) emit_addimm(HOST_CCREG,-CLOCK_ADJUST((signed int)stubs[n].a),HOST_CCREG);
if(stubs[n].d==TAKEN) {
if(internal_branch(ba[i]))
}else{
load_all_regs(branch_regs[i].regmap);
}
- emit_jmp(stubs[n].retaddr);
+ if (stubs[n].retaddr)
+ emit_jmp(stubs[n].retaddr);
+ else
+ do_jump_vaddr(stubs[n].e);
}
static void add_to_linker(void *addr, u_int target, int ext)
//if(adj) emit_addimm(cc,2*(ccadj[i]+2-adj),cc); // ??? - Shouldn't happen
//assert(adj==0);
emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG);
- add_stub(CC_STUB,out,jump_vaddr_reg[rs],0,i,-1,TAKEN,0);
+ add_stub(CC_STUB,out,NULL,0,i,-1,TAKEN,rs);
if(itype[i+1]==COP0&&(source[i+1]&0x3f)==0x10)
// special case for RFE
emit_jmp(0);
else
#endif
{
- emit_jmp(jump_vaddr_reg[rs]);
+ do_jump_vaddr(rs);
}
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
if(rt1[i]!=31&&i<slen-2&&(((u_int)out)&7)) emit_mov(13,13);
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
if(i>(ba[i]-start)>>2) invert=1;
#endif
+ #ifdef __aarch64__
+ invert=1; // because of near cond. branches
+ #endif
if(ooo[i]) {
s1l=get_reg(branch_regs[i].regmap,rs1[i]);
#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
if(i>(ba[i]-start)>>2) invert=1;
#endif
+ #ifdef __aarch64__
+ invert=1; // because of near cond. branches
+ #endif
//if(opcode2[i]>=0x10) return; // FIXME (BxxZAL)
//assert(opcode2[i]<0x10||rs1[i]==0); // FIXME (BxxZAL)
assert(btaddr!=HOST_CCREG);
if(regs[0].regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
#ifdef HOST_IMM8
+ host_tempreg_acquire();
emit_movimm(start+4,HOST_TEMPREG);
emit_cmp(btaddr,HOST_TEMPREG);
+ host_tempreg_release();
#else
emit_cmpimm(btaddr,start+4);
#endif
void *branch = out;
emit_jeq(0);
store_regs_bt(regs[0].regmap,regs[0].dirty,-1);
- emit_jmp(jump_vaddr_reg[btaddr]);
+ do_jump_vaddr(btaddr);
set_jump_target(branch, out);
store_regs_bt(regs[0].regmap,regs[0].dirty,start+4);
load_regs_bt(regs[0].regmap,regs[0].dirty,start+4);
((volatile u_int *)out)[0]++; // make cache dirty
for (i = 0; i < ARRAY_SIZE(ret); i++) {
- out = translation_cache;
+ out = ndrc->translation_cache;
beginning = start_block();
emit_movimm(DRC_TEST_VAL + i, 0); // test
emit_ret();
SysPrintf("test passed.\n");
else
SysPrintf("test failed, will likely crash soon (r=%08x %08x)\n", ret[0], ret[1]);
- out = translation_cache;
+ out = ndrc->translation_cache;
}
// clear the state completely, instead of just marking
void new_dynarec_clear_full()
{
int n;
- out = translation_cache;
+ out = ndrc->translation_cache;
memset(invalid_code,1,sizeof(invalid_code));
memset(hash_table,0xff,sizeof(hash_table));
memset(mini_ht,-1,sizeof(mini_ht));
{
SysPrintf("Init new dynarec\n");
- // allocate/prepare a buffer for translation cache
- // see assem_arm.h for some explanation
-#if defined(BASE_ADDR_FIXED)
- if (mmap(translation_cache, 1 << TARGET_SIZE_2,
- PROT_READ | PROT_WRITE | PROT_EXEC,
- MAP_PRIVATE | MAP_ANONYMOUS,
- -1, 0) != translation_cache) {
- SysPrintf("mmap() failed: %s\n", strerror(errno));
- SysPrintf("disable BASE_ADDR_FIXED and recompile\n");
- abort();
- }
-#elif defined(BASE_ADDR_DYNAMIC)
+#ifdef BASE_ADDR_DYNAMIC
#ifdef VITA
sceBlock = sceKernelAllocMemBlockForVM("code", 1 << TARGET_SIZE_2);
if (sceBlock < 0)
SysPrintf("sceKernelAllocMemBlockForVM failed\n");
- int ret = sceKernelGetMemBlockBase(sceBlock, (void **)&translation_cache);
+ int ret = sceKernelGetMemBlockBase(sceBlock, (void **)&ndrc);
if (ret < 0)
SysPrintf("sceKernelGetMemBlockBase failed\n");
#else
- translation_cache = mmap (NULL, 1 << TARGET_SIZE_2,
+ uintptr_t desired_addr = 0;
+ #ifdef __ELF__
+ extern char _end;
+ desired_addr = ((uintptr_t)&_end + 0xffffff) & ~0xffffffl;
+ #endif
+ ndrc = mmap((void *)desired_addr, sizeof(*ndrc),
PROT_READ | PROT_WRITE | PROT_EXEC,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
- if (translation_cache == MAP_FAILED) {
+ if (ndrc == MAP_FAILED) {
SysPrintf("mmap() failed: %s\n", strerror(errno));
abort();
}
#else
#ifndef NO_WRITE_EXEC
// not all systems allow execute in data segment by default
- if (mprotect(translation_cache, 1<<TARGET_SIZE_2, PROT_READ | PROT_WRITE | PROT_EXEC) != 0)
+ if (mprotect(ndrc, sizeof(ndrc->translation_cache) + sizeof(ndrc->tramp.ops),
+ PROT_READ | PROT_WRITE | PROT_EXEC) != 0)
SysPrintf("mprotect() failed: %s\n", strerror(errno));
#endif
#endif
- out = translation_cache;
+ out = ndrc->translation_cache;
cycle_multiplier=200;
new_dynarec_clear_full();
#ifdef HOST_IMM8
void new_dynarec_cleanup()
{
int n;
-#if defined(BASE_ADDR_FIXED) || defined(BASE_ADDR_DYNAMIC)
+#ifdef BASE_ADDR_DYNAMIC
#ifdef VITA
sceKernelFreeMemBlock(sceBlock);
sceBlock = -1;
#else
- if (munmap(translation_cache, 1<<TARGET_SIZE_2) < 0)
+ if (munmap(ndrc, sizeof(*ndrc)) < 0)
SysPrintf("munmap() failed\n");
#endif
#endif
memcpy(&psxRegs.GPR, regs_save, sizeof(regs_save));
}
-int new_recompile_block(int addr)
+int new_recompile_block(u_int addr)
{
u_int pagelimit = 0;
u_int state_rflags = 0;
invalid_code[start>>12]=0;
emit_movimm(start,0);
emit_writeword(0,&pcaddr);
- emit_jmp(new_dyna_leave);
+ emit_far_jump(new_dyna_leave);
literal_pool(0);
end_block(beginning);
ll_add_flags(jump_in+page,start,state_rflags,(void *)beginning);
{
if(i<slen-1&&!is_ds[i]) {
assert(regs[i].regmap[hr]<64);
- if(regmap_pre[i+1][hr]!=-1 || regs[i].regmap[hr]!=-1)
+ if(regmap_pre[i+1][hr]!=-1 || regs[i].regmap[hr]>0)
if(regmap_pre[i+1][hr]!=regs[i].regmap[hr])
{
SysPrintf("fail: %x (%d %d!=%d)\n",start+i*4,hr,regmap_pre[i+1][hr],regs[i].regmap[hr]);
}
}
}
- }
- }
+ } // if needed
+ } // for hr
}
/* Pass 5 - Pre-allocate registers */
void *instr_addr0_override = NULL;
if (start == 0x80030000) {
- // nasty hack for fastbios thing
+ // nasty hack for the fastbios thing
// override block entry to this code
instr_addr0_override = out;
emit_movimm(start,0);
emit_writeword(0,&pcaddr);
emit_writeword(0,&address);
emit_cmp(0,1);
+ #ifdef __aarch64__
+ emit_jeq(out + 4*2);
+ emit_far_jump(new_dyna_leave);
+ #else
emit_jne(new_dyna_leave);
+ #endif
}
for(i=0;i<slen;i++)
{
// If we're within 256K of the end of the buffer,
// start over from the beginning. (Is 256K enough?)
- if (out > translation_cache+(1<<TARGET_SIZE_2)-MAX_OUTPUT_BLOCK_SIZE)
- out = translation_cache;
+ if (out > ndrc->translation_cache + sizeof(ndrc->translation_cache) - MAX_OUTPUT_BLOCK_SIZE)
+ out = ndrc->translation_cache;
// Trap writes to any of the pages we compiled
for(i=start>>12;i<=(start+slen*4)>>12;i++) {
/* Pass 10 - Free memory by expiring oldest blocks */
- int end=(((out-translation_cache)>>(TARGET_SIZE_2-16))+16384)&65535;
+ int end=(((out-ndrc->translation_cache)>>(TARGET_SIZE_2-16))+16384)&65535;
while(expirep!=end)
{
int shift=TARGET_SIZE_2-3; // Divide into 8 blocks
- uintptr_t base=(uintptr_t)translation_cache+((expirep>>13)<<shift); // Base address of this block
+ uintptr_t base=(uintptr_t)ndrc->translation_cache+((expirep>>13)<<shift); // Base address of this block
inv_debug("EXP: Phase %d\n",expirep);
switch((expirep>>11)&3)
{