notaz.gp2x.de
/
pcsx_rearmed.git
/ blobdiff
commit
grep
author
committer
pickaxe
?
search:
re
summary
|
shortlog
|
log
|
commit
|
commitdiff
|
tree
raw
|
inline
| side by side
drc: drop 'likely' branch support, simplify
[pcsx_rearmed.git]
/
libpcsxcore
/
new_dynarec
/
assem_arm64.c
diff --git
a/libpcsxcore/new_dynarec/assem_arm64.c
b/libpcsxcore/new_dynarec/assem_arm64.c
index
6380b27
..
ea1b8a3
100644
(file)
--- a/
libpcsxcore/new_dynarec/assem_arm64.c
+++ b/
libpcsxcore/new_dynarec/assem_arm64.c
@@
-44,7
+44,7
@@
static void set_jump_target(void *addr, void *target)
|| (*ptr&0x7e000000) == 0x34000000) { // cbz/cbnz
// Conditional branch are limited to +/- 1MB
// block max size is 256k so branching beyond the +/- 1MB limit
|| (*ptr&0x7e000000) == 0x34000000) { // cbz/cbnz
// Conditional branch are limited to +/- 1MB
// block max size is 256k so branching beyond the +/- 1MB limit
- // should only happen when jumping to an already compiled block (see add_
link
)
+ // should only happen when jumping to an already compiled block (see add_
jump_out
)
// a workaround would be to do a trampoline jump via a stub at the end of the block
assert(-1048576 <= offset && offset < 1048576);
*ptr=(*ptr&0xFF00000F)|(((offset>>2)&0x7ffff)<<5);
// a workaround would be to do a trampoline jump via a stub at the end of the block
assert(-1048576 <= offset && offset < 1048576);
*ptr=(*ptr&0xFF00000F)|(((offset>>2)&0x7ffff)<<5);
@@
-813,6
+813,12
@@
static void emit_cmovl_reg(u_int rs,u_int rt)
output_w32(0x1a800000 | (COND_LT << 12) | rm_rn_rd(rt, rs, rt));
}
output_w32(0x1a800000 | (COND_LT << 12) | rm_rn_rd(rt, rs, rt));
}
+static void emit_cmovb_reg(u_int rs,u_int rt)
+{
+ assem_debug("csel %s,%s,%s,cc\n",regname[rt],regname[rs],regname[rt]);
+ output_w32(0x1a800000 | (COND_CC << 12) | rm_rn_rd(rt, rs, rt));
+}
+
static void emit_cmovs_reg(u_int rs,u_int rt)
{
assem_debug("csel %s,%s,%s,mi\n",regname[rt],regname[rs],regname[rt]);
static void emit_cmovs_reg(u_int rs,u_int rt)
{
assem_debug("csel %s,%s,%s,mi\n",regname[rt],regname[rs],regname[rt]);
@@
-1374,10
+1380,10
@@
static void do_readstub(int n)
u_int reglist = stubs[n].e;
const signed char *i_regmap = i_regs->regmap;
int rt;
u_int reglist = stubs[n].e;
const signed char *i_regmap = i_regs->regmap;
int rt;
- if(
itype[i]==C1LS||itype[i]==C2LS||itype[i]
==LOADLR) {
+ if(
dops[i].itype==C1LS||dops[i].itype==C2LS||dops[i].itype
==LOADLR) {
rt=get_reg(i_regmap,FTEMP);
}else{
rt=get_reg(i_regmap,FTEMP);
}else{
- rt=get_reg(i_regmap,
rt1[i]
);
+ rt=get_reg(i_regmap,
dops[i].rt1
);
}
assert(rs>=0);
int r,temp=-1,temp2=HOST_TEMPREG,regs_saved=0;
}
assert(rs>=0);
int r,temp=-1,temp2=HOST_TEMPREG,regs_saved=0;
@@
-1389,7
+1395,7
@@
static void do_readstub(int n)
break;
}
}
break;
}
}
- if(rt>=0&&
rt1[i]
!=0)
+ if(rt>=0&&
dops[i].rt1
!=0)
reglist&=~(1<<rt);
if(temp==-1) {
save_regs(reglist);
reglist&=~(1<<rt);
if(temp==-1) {
save_regs(reglist);
@@
-1404,7
+1410,7
@@
static void do_readstub(int n)
emit_adds64(temp2,temp2,temp2);
handler_jump=out;
emit_jc(0);
emit_adds64(temp2,temp2,temp2);
handler_jump=out;
emit_jc(0);
- if(
itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]
!=0)) {
+ if(
dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1
!=0)) {
switch(type) {
case LOADB_STUB: emit_ldrsb_dualindexed(temp2,rs,rt); break;
case LOADBU_STUB: emit_ldrb_dualindexed(temp2,rs,rt); break;
switch(type) {
case LOADB_STUB: emit_ldrsb_dualindexed(temp2,rs,rt); break;
case LOADBU_STUB: emit_ldrb_dualindexed(temp2,rs,rt); break;
@@
-1439,7
+1445,7
@@
static void do_readstub(int n)
emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2);
emit_far_call(handler);
// (no cycle reload after read)
emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2);
emit_far_call(handler);
// (no cycle reload after read)
- if(
itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]
!=0)) {
+ if(
dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1
!=0)) {
loadstore_extend(type,0,rt);
}
if(restore_jump)
loadstore_extend(type,0,rt);
}
if(restore_jump)
@@
-1448,7
+1454,8
@@
static void do_readstub(int n)
emit_jmp(stubs[n].retaddr);
}
emit_jmp(stubs[n].retaddr);
}
-static void inline_readstub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+static void inline_readstub(enum stub_type type, int i, u_int addr,
+ const signed char regmap[], int target, int adj, u_int reglist)
{
int rs=get_reg(regmap,target);
int rt=get_reg(regmap,target);
{
int rs=get_reg(regmap,target);
int rt=get_reg(regmap,target);
@@
-1462,7
+1469,7
@@
static void inline_readstub(enum stub_type type, int i, u_int addr, signed char
// return;
handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr);
if (handler == NULL) {
// return;
handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr);
if (handler == NULL) {
- if(rt<0||
rt1[i]
==0)
+ if(rt<0||
dops[i].rt1
==0)
return;
if (addr != host_addr) {
if (host_addr >= 0x100000000ull)
return;
if (addr != host_addr) {
if (host_addr >= 0x100000000ull)
@@
-1490,7
+1497,7
@@
static void inline_readstub(enum stub_type type, int i, u_int addr, signed char
}
// call a memhandler
}
// call a memhandler
- if(rt>=0&&
rt1[i]
!=0)
+ if(rt>=0&&
dops[i].rt1
!=0)
reglist&=~(1<<rt);
save_regs(reglist);
if(target==0)
reglist&=~(1<<rt);
save_regs(reglist);
if(target==0)
@@
-1511,7
+1518,7
@@
static void inline_readstub(enum stub_type type, int i, u_int addr, signed char
emit_far_call(handler);
// (no cycle reload after read)
emit_far_call(handler);
// (no cycle reload after read)
- if(rt>=0&&
rt1[i]
!=0)
+ if(rt>=0&&
dops[i].rt1
!=0)
loadstore_extend(type, 0, rt);
restore_regs(reglist);
}
loadstore_extend(type, 0, rt);
restore_regs(reglist);
}
@@
-1527,10
+1534,10
@@
static void do_writestub(int n)
u_int reglist=stubs[n].e;
signed char *i_regmap=i_regs->regmap;
int rt,r;
u_int reglist=stubs[n].e;
signed char *i_regmap=i_regs->regmap;
int rt,r;
- if(
itype[i]==C1LS||itype[i]
==C2LS) {
+ if(
dops[i].itype==C1LS||dops[i].itype
==C2LS) {
rt=get_reg(i_regmap,r=FTEMP);
}else{
rt=get_reg(i_regmap,r=FTEMP);
}else{
- rt=get_reg(i_regmap,r=
rs2[i]
);
+ rt=get_reg(i_regmap,r=
dops[i].rs2
);
}
assert(rs>=0);
assert(rt>=0);
}
assert(rs>=0);
assert(rt>=0);
@@
-1607,7
+1614,8
@@
static void do_writestub(int n)
emit_jmp(stubs[n].retaddr);
}
emit_jmp(stubs[n].retaddr);
}
-static void inline_writestub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+static void inline_writestub(enum stub_type type, int i, u_int addr,
+ const signed char regmap[], int target, int adj, u_int reglist)
{
int rs = get_reg(regmap,-1);
int rt = get_reg(regmap,target);
{
int rs = get_reg(regmap,-1);
int rt = get_reg(regmap,target);
@@
-1657,12
+1665,12
@@
static int verify_code_arm64(const void *source, const void *copy, u_int size)
}
// this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
}
// this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
-static void do_dirty_stub_base(u_int vaddr)
+static void do_dirty_stub_base(u_int vaddr
, u_int source_len
)
{
{
- assert(s
len <= MAXBLOCK
);
+ assert(s
ource_len <= MAXBLOCK*4
);
emit_loadlp_ofs(0, 0); // ldr x1, source
emit_loadlp_ofs(0, 1); // ldr x2, copy
emit_loadlp_ofs(0, 0); // ldr x1, source
emit_loadlp_ofs(0, 1); // ldr x2, copy
- emit_movz(s
len*4
, 2);
+ emit_movz(s
ource_len
, 2);
emit_far_call(verify_code_arm64);
void *jmp = out;
emit_cbz(0, 0);
emit_far_call(verify_code_arm64);
void *jmp = out;
emit_cbz(0, 0);
@@
-1677,7
+1685,7
@@
static void assert_dirty_stub(const u_int *ptr)
{
assert((ptr[0] & 0xff00001f) == 0x58000000); // ldr x0, source
assert((ptr[1] & 0xff00001f) == 0x58000001); // ldr x1, copy
{
assert((ptr[0] & 0xff00001f) == 0x58000000); // ldr x0, source
assert((ptr[1] & 0xff00001f) == 0x58000001); // ldr x1, copy
- assert((ptr[2] & 0xffe0001f) == 0x52800002); // movz w2, #s
len*4
+ assert((ptr[2] & 0xffe0001f) == 0x52800002); // movz w2, #s
ource_len
assert( ptr[8] == 0xd61f0000); // br x0
}
assert( ptr[8] == 0xd61f0000); // br x0
}
@@
-1698,11
+1706,11
@@
static void do_dirty_stub_emit_literals(u_int *loadlps)
output_w64((uintptr_t)copy);
}
output_w64((uintptr_t)copy);
}
-static void *do_dirty_stub(int i)
+static void *do_dirty_stub(int i
, u_int source_len
)
{
assem_debug("do_dirty_stub %x\n",start+i*4);
u_int *loadlps = (void *)out;
{
assem_debug("do_dirty_stub %x\n",start+i*4);
u_int *loadlps = (void *)out;
- do_dirty_stub_base(start + i*4);
+ do_dirty_stub_base(start + i*4
, source_len
);
void *entry = out;
load_regs_entry(i);
if (entry == out)
void *entry = out;
load_regs_entry(i);
if (entry == out)
@@
-1712,10
+1720,10
@@
static void *do_dirty_stub(int i)
return entry;
}
return entry;
}
-static void do_dirty_stub_ds(
void
)
+static void do_dirty_stub_ds(
u_int source_len
)
{
u_int *loadlps = (void *)out;
{
u_int *loadlps = (void *)out;
- do_dirty_stub_base(start + 1);
+ do_dirty_stub_base(start + 1
, source_len
);
void *lit_jumpover = out;
emit_jmp(out + 8*2);
do_dirty_stub_emit_literals(loadlps);
void *lit_jumpover = out;
emit_jmp(out + 8*2);
do_dirty_stub_emit_literals(loadlps);
@@
-1752,7
+1760,7
@@
static int verify_dirty(const u_int *ptr)
assert_dirty_stub(ptr);
source = (void *)get_from_ldr_literal(&ptr[0]); // ldr x1, source
copy = (void *)get_from_ldr_literal(&ptr[1]); // ldr x1, copy
assert_dirty_stub(ptr);
source = (void *)get_from_ldr_literal(&ptr[0]); // ldr x1, source
copy = (void *)get_from_ldr_literal(&ptr[1]); // ldr x1, copy
- len = get_from_movz(&ptr[2]); // movz w3, #s
len*4
+ len = get_from_movz(&ptr[2]); // movz w3, #s
ource_len
return !memcmp(source, copy, len);
}
return !memcmp(source, copy, len);
}
@@
-1772,14
+1780,15
@@
static void get_bounds(void *addr, u_char **start, u_char **end)
const u_int *ptr = addr;
assert_dirty_stub(ptr);
*start = (u_char *)get_from_ldr_literal(&ptr[0]); // ldr x1, source
const u_int *ptr = addr;
assert_dirty_stub(ptr);
*start = (u_char *)get_from_ldr_literal(&ptr[0]); // ldr x1, source
- *end = *start + get_from_movz(&ptr[2]); // movz w3, #s
len*4
+ *end = *start + get_from_movz(&ptr[2]); // movz w3, #s
ource_len
}
/* Special assem */
}
/* Special assem */
-static void c2op_prologue(u_int op,u_int reglist)
+static void c2op_prologue(u_int op,
int i, const struct regstat *i_regs,
u_int reglist)
{
save_load_regs_all(1, reglist);
{
save_load_regs_all(1, reglist);
+ cop2_do_stall_check(op, i, i_regs, 0);
#ifdef PCNT
emit_movimm(op, 0);
emit_far_call(pcnt_gte_start);
#ifdef PCNT
emit_movimm(op, 0);
emit_far_call(pcnt_gte_start);
@@
-1797,7
+1806,7
@@
static void c2op_epilogue(u_int op,u_int reglist)
save_load_regs_all(0, reglist);
}
save_load_regs_all(0, reglist);
}
-static void c2op_assemble(int i,struct regstat *i_regs)
+static void c2op_assemble(int i,
const
struct regstat *i_regs)
{
u_int c2op=source[i]&0x3f;
u_int hr,reglist_full=0,reglist;
{
u_int c2op=source[i]&0x3f;
u_int hr,reglist_full=0,reglist;
@@
-1812,14
+1821,14
@@
static void c2op_assemble(int i,struct regstat *i_regs)
need_ir=(gte_unneeded[i+1]&0xe00)!=0xe00;
assem_debug("gte op %08x, unneeded %016lx, need_flags %d, need_ir %d\n",
source[i],gte_unneeded[i+1],need_flags,need_ir);
need_ir=(gte_unneeded[i+1]&0xe00)!=0xe00;
assem_debug("gte op %08x, unneeded %016lx, need_flags %d, need_ir %d\n",
source[i],gte_unneeded[i+1],need_flags,need_ir);
- if(
new_dynarec_hacks&NDHACK_GTE_NO_FLAGS
)
+ if(
HACK_ENABLED(NDHACK_GTE_NO_FLAGS)
)
need_flags=0;
//int shift = (source[i] >> 19) & 1;
//int lm = (source[i] >> 10) & 1;
switch(c2op) {
default:
(void)need_ir;
need_flags=0;
//int shift = (source[i] >> 19) & 1;
//int lm = (source[i] >> 10) & 1;
switch(c2op) {
default:
(void)need_ir;
- c2op_prologue(c2op,reglist);
+ c2op_prologue(c2op,
i, i_regs,
reglist);
emit_movimm(source[i],1); // opcode
emit_writeword(1,&psxRegs.code);
emit_far_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]);
emit_movimm(source[i],1); // opcode
emit_writeword(1,&psxRegs.code);
emit_far_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]);
@@
-1876,15
+1885,15
@@
static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
// case 0x19: MULTU
// case 0x1A: DIV
// case 0x1B: DIVU
// case 0x19: MULTU
// case 0x1A: DIV
// case 0x1B: DIVU
- if(
rs1[i]&&rs2[i]
)
+ if(
dops[i].rs1&&dops[i].rs2
)
{
{
- switch(
opcode2[i]
)
+ switch(
dops[i].opcode2
)
{
case 0x18: // MULT
case 0x19: // MULTU
{
{
case 0x18: // MULT
case 0x19: // MULTU
{
- signed char m1=get_reg(i_regs->regmap,
rs1[i]
);
- signed char m2=get_reg(i_regs->regmap,
rs2[i]
);
+ signed char m1=get_reg(i_regs->regmap,
dops[i].rs1
);
+ signed char m2=get_reg(i_regs->regmap,
dops[i].rs2
);
signed char hi=get_reg(i_regs->regmap,HIREG);
signed char lo=get_reg(i_regs->regmap,LOREG);
assert(m1>=0);
signed char hi=get_reg(i_regs->regmap,HIREG);
signed char lo=get_reg(i_regs->regmap,LOREG);
assert(m1>=0);
@@
-1892,7
+1901,7
@@
static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
assert(hi>=0);
assert(lo>=0);
assert(hi>=0);
assert(lo>=0);
- if(
opcode2[i]
==0x18) // MULT
+ if(
dops[i].opcode2
==0x18) // MULT
emit_smull(m1,m2,hi);
else // MULTU
emit_umull(m1,m2,hi);
emit_smull(m1,m2,hi);
else // MULTU
emit_umull(m1,m2,hi);
@@
-1904,8
+1913,8
@@
static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
case 0x1A: // DIV
case 0x1B: // DIVU
{
case 0x1A: // DIV
case 0x1B: // DIVU
{
- signed char numerator=get_reg(i_regs->regmap,
rs1[i]
);
- signed char denominator=get_reg(i_regs->regmap,
rs2[i]
);
+ signed char numerator=get_reg(i_regs->regmap,
dops[i].rs1
);
+ signed char denominator=get_reg(i_regs->regmap,
dops[i].rs2
);
signed char quotient=get_reg(i_regs->regmap,LOREG);
signed char remainder=get_reg(i_regs->regmap,HIREG);
assert(numerator>=0);
signed char quotient=get_reg(i_regs->regmap,LOREG);
signed char remainder=get_reg(i_regs->regmap,HIREG);
assert(numerator>=0);
@@
-1913,7
+1922,7
@@
static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
assert(quotient>=0);
assert(remainder>=0);
assert(quotient>=0);
assert(remainder>=0);
- if (
opcode2[i]
== 0x1A) // DIV
+ if (
dops[i].opcode2
== 0x1A) // DIV
emit_sdiv(numerator,denominator,quotient);
else // DIVU
emit_udiv(numerator,denominator,quotient);
emit_sdiv(numerator,denominator,quotient);
else // DIVU
emit_udiv(numerator,denominator,quotient);
@@
-1921,7
+1930,7
@@
static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
// div 0 quotient (remainder is already correct)
host_tempreg_acquire();
// div 0 quotient (remainder is already correct)
host_tempreg_acquire();
- if (
opcode2[i]
== 0x1A) // DIV
+ if (
dops[i].opcode2
== 0x1A) // DIV
emit_sub_asrimm(0,numerator,31,HOST_TEMPREG);
else
emit_movimm(~0,HOST_TEMPREG);
emit_sub_asrimm(0,numerator,31,HOST_TEMPREG);
else
emit_movimm(~0,HOST_TEMPREG);
@@
-1938,15
+1947,15
@@
static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
{
signed char hr=get_reg(i_regs->regmap,HIREG);
signed char lr=get_reg(i_regs->regmap,LOREG);
{
signed char hr=get_reg(i_regs->regmap,HIREG);
signed char lr=get_reg(i_regs->regmap,LOREG);
- if ((
opcode2[i]==0x1A || opcode2[i]==0x1B) && rs2[i]
==0) // div 0
+ if ((
dops[i].opcode2==0x1A || dops[i].opcode2==0x1B) && dops[i].rs2
==0) // div 0
{
{
- if (
rs1[i]
) {
- signed char numerator = get_reg(i_regs->regmap,
rs1[i]
);
+ if (
dops[i].rs1
) {
+ signed char numerator = get_reg(i_regs->regmap,
dops[i].rs1
);
assert(numerator >= 0);
if (hr >= 0)
emit_mov(numerator,hr);
if (lr >= 0) {
assert(numerator >= 0);
if (hr >= 0)
emit_mov(numerator,hr);
if (lr >= 0) {
- if (
opcode2[i]
== 0x1A) // DIV
+ if (
dops[i].opcode2
== 0x1A) // DIV
emit_sub_asrimm(0,numerator,31,lr);
else
emit_movimm(~0,lr);
emit_sub_asrimm(0,numerator,31,lr);
else
emit_movimm(~0,lr);