#include "pcnt.h"
#include "arm_features.h"
-#define CALLER_SAVE_REGS 0x0007ffff
-
#define unused __attribute__((unused))
void do_memhandler_pre();
|| (*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);
return ptr + offset / 4;
}
+#if 0
// find where external branch is liked to using addr of it's stub:
// get address that the stub loads (dyna_linker arg1),
// treat it as a pointer to branch insn,
assert(0);
return NULL;
}
+#endif
// Allocate a specific ARM register.
static void alloc_arm_reg(struct regstat *cur,int i,signed char reg,int hr)
out += 4;
}
-static void output_w64(uint64_t dword)
-{
- *((uint64_t *)out) = dword;
- out+=8;
-}
-
-/*
-static u_int rm_rd(u_int rm, u_int rd)
-{
- assert(rm < 31);
- assert(rd < 31);
- return (rm << 16) | rd;
-}
-*/
-
static u_int rn_rd(u_int rn, u_int rd)
{
assert(rn < 31);
assem_debug("adds %s,%s,%s\n",regname64[rt],regname64[rs1],regname64[rs2]);
output_w32(0xab000000 | rm_rn_rd(rs2, rs1, rt));
}
+#define emit_adds_ptr emit_adds64
static void emit_neg(u_int rs, u_int rt)
{
else
abort();
}
+#define emit_readptr emit_readdword
static void emit_readshword(void *addr, u_int rt)
{
static void emit_loadreg(u_int r, u_int hr)
{
int is64 = 0;
- assert(r < 64);
if (r == 0)
emit_zeroreg(hr);
else {
- void *addr = &psxRegs.GPR.r[r];
+ void *addr;
switch (r) {
//case HIREG: addr = &hi; break;
//case LOREG: addr = &lo; break;
case CCREG: addr = &cycle_count; break;
case CSREG: addr = &Status; break;
case INVCP: addr = &invc_ptr; is64 = 1; break;
- default: assert(r < 34); break;
+ case ROREG: addr = &ram_offset; is64 = 1; break;
+ default:
+ assert(r < 34);
+ addr = &psxRegs.GPR.r[r];
+ break;
}
if (is64)
emit_readdword(addr, hr);
emit_addimm_s(1, 0, rt, imm, rt);
}
-static void emit_addimm_no_flags(u_int imm,u_int rt)
-{
- emit_addimm(rt,imm,rt);
-}
-
static void emit_logicop_imm(u_int op, u_int rs, u_int imm, u_int rt)
{
const char *names[] = { "and", "orr", "eor", "ands" };
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]);
output_w32(0x6b000000 | rm_rn_rd(rt, rs, WZR));
}
+static void emit_cmpcs(u_int rs,u_int rt)
+{
+ assem_debug("ccmp %s,%s,#0,cs\n",regname[rs],regname[rt]);
+ output_w32(0x7a400000 | (COND_CS << 12) | rm_rn_rd(rt, rs, 0));
+}
+
static void emit_set_gz32(u_int rs, u_int rt)
{
//assem_debug("set_gz32\n");
output_w32(0x34000000 | is64 | isnz | imm19_rt(offset, r));
}
-static void emit_cbz(const void *a, u_int r)
+static unused void emit_cbz(const void *a, u_int r)
{
emit_cb(0, 0, a, r);
}
assem_debug("ldr %s, [%s,%s, uxtw #3]\n",regname64[rt],regname64[rs1],regname[rs2]);
output_w32(0xf8605800 | rm_rn_rd(rs2, rs1, rt));
}
+#define emit_readptr_dualindexedx_ptrlen emit_readdword_dualindexedx8
static void emit_ldrb_dualindexed(u_int rs1, u_int rs2, u_int rt)
{
emit_bic(rs1, rs2, rt);
}
-static void emit_loadlp_ofs(u_int ofs, u_int rt)
-{
- output_w32(0x58000000 | imm19_rt(ofs, rt));
-}
-
static void emit_ldst(int is_st, int is64, u_int rt, u_int rn, u_int ofs)
{
u_int op = 0xb9000000;
}
// parsed by get_pointer, find_extjump_insn
-static void emit_extjump2(u_char *addr, u_int target, void *linker)
+static void emit_extjump(u_char *addr, u_int target)
{
assert(((addr[3]&0xfc)==0x14) || ((addr[3]&0xff)==0x54)); // b or b.cond
// addr is in the current recompiled block (max 256k)
// offset shouldn't exceed +/-1MB
emit_adr(addr, 1);
- emit_far_jump(linker);
+ emit_far_jump(dyna_linker);
}
static void check_extjump2(void *src)
|| is_rotated_mask(v1 ^ v2);
}
-// trashes r2
+static void emit_movimm_from64(u_int rs_val, u_int rs, uintptr_t rt_val, u_int rt)
+{
+ if (rt_val < 0x100000000ull) {
+ emit_movimm_from(rs_val, rs, rt_val, rt);
+ return;
+ }
+ // just move the whole thing. At least on Linux all addresses
+ // seem to be 48bit, so 3 insns - not great not terrible
+ assem_debug("movz %s,#%#lx\n", regname64[rt], rt_val & 0xffff);
+ output_w32(0xd2800000 | imm16_rd(rt_val & 0xffff, rt));
+ assem_debug("movk %s,#%#lx,lsl #16\n", regname64[rt], (rt_val >> 16) & 0xffff);
+ output_w32(0xf2a00000 | imm16_rd((rt_val >> 16) & 0xffff, rt));
+ assem_debug("movk %s,#%#lx,lsl #32\n", regname64[rt], (rt_val >> 32) & 0xffff);
+ output_w32(0xf2c00000 | imm16_rd((rt_val >> 32) & 0xffff, rt));
+ if (rt_val >> 48) {
+ assem_debug("movk %s,#%#lx,lsl #48\n", regname64[rt], (rt_val >> 48) & 0xffff);
+ output_w32(0xf2e00000 | imm16_rd((rt_val >> 48) & 0xffff, rt));
+ }
+}
+
+// trashes x2
static void pass_args64(u_int a0, u_int a1)
{
if(a0==1&&a1==0) {
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,rt1[i]);
+ rt=get_reg(i_regmap,dops[i].rt1);
}
assert(rs>=0);
int r,temp=-1,temp2=HOST_TEMPREG,regs_saved=0;
break;
}
}
- if(rt>=0&&rt1[i]!=0)
+ if(rt>=0&&dops[i].rt1!=0)
reglist&=~(1<<rt);
if(temp==-1) {
save_regs(reglist);
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;
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),2);
+ emit_addimm(cc<0?2:cc,(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)
{
int rs=get_reg(regmap,target);
int rt=get_reg(regmap,target);
- if(rs<0) rs=get_reg(regmap,-1);
+ if(rs<0) rs=get_reg_temp(regmap);
assert(rs>=0);
u_int is_dynamic=0;
uintptr_t host_addr = 0;
void *handler;
int cc=get_reg(regmap,CCREG);
- //if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj),cc,target?rs:-1,rt))
+ //if(pcsx_direct_read(type,addr,adj,cc,target?rs:-1,rt))
// 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)
- abort(); // ROREG not implemented
- emit_movimm_from(addr, rs, host_addr, rs);
- }
+ if (addr != host_addr)
+ emit_movimm_from64(addr, rs, host_addr, rs);
switch(type) {
case LOADB_STUB: emit_movsbl_indexed(0,rs,rt); break;
case LOADBU_STUB: emit_movzbl_indexed(0,rs,rt); break;
}
return;
}
- is_dynamic=pcsxmem_is_handler_dynamic(addr);
- if(is_dynamic) {
+ is_dynamic = pcsxmem_is_handler_dynamic(addr);
+ if (is_dynamic) {
if(type==LOADB_STUB||type==LOADBU_STUB)
handler=jump_handler_read8;
if(type==LOADH_STUB||type==LOADHU_STUB)
}
// 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)
emit_mov(rs,0);
if(cc<0)
emit_loadreg(CCREG,2);
- emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2);
+ emit_addimm(cc<0?2:cc,adj,2);
if(is_dynamic) {
uintptr_t l1 = ((uintptr_t *)mem_rtab)[addr>>12] << 1;
emit_adrp((void *)l1, 1);
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);
}
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=rs2[i]);
+ rt=get_reg(i_regmap,r=dops[i].rs2);
}
assert(rs>=0);
assert(rt>=0);
emit_jmp(stubs[n].retaddr); // return address (invcode check)
set_jump_target(handler_jump, out);
- // TODO FIXME: regalloc should prefer callee-saved regs
if(!regs_saved)
save_regs(reglist);
void *handler=NULL;
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),2);
+ emit_addimm(cc<0?2:cc,(int)stubs[n].d,2);
// returns new cycle_count
emit_far_call(handler);
- emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d),cc<0?2:cc);
+ emit_addimm(0,-(int)stubs[n].d,cc<0?2:cc);
if(cc<0)
emit_storereg(CCREG,2);
if(restore_jump)
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 rs = get_reg_temp(regmap);
int rt = get_reg(regmap,target);
assert(rs >= 0);
assert(rt >= 0);
uintptr_t host_addr = 0;
void *handler = get_direct_memhandler(mem_wtab, addr, type, &host_addr);
if (handler == NULL) {
- if (addr != host_addr) {
- if (host_addr >= 0x100000000ull)
- abort(); // ROREG not implemented
- emit_movimm_from(addr, rs, host_addr, rs);
- }
+ if (addr != host_addr)
+ emit_movimm_from64(addr, rs, host_addr, rs);
switch (type) {
case STOREB_STUB: emit_writebyte_indexed(rt, 0, rs); break;
case STOREH_STUB: emit_writehword_indexed(rt, 0, rs); break;
cc = cc_use = get_reg(regmap, CCREG);
if (cc < 0)
emit_loadreg(CCREG, (cc_use = 2));
- emit_addimm(cc_use, CLOCK_ADJUST(adj), 2);
+ emit_addimm(cc_use, adj, 2);
emit_far_call(do_memhandler_pre);
emit_far_call(handler);
emit_far_call(do_memhandler_post);
- emit_addimm(0, -CLOCK_ADJUST(adj), cc_use);
+ emit_addimm(0, -adj, cc_use);
if (cc < 0)
emit_storereg(CCREG, cc_use);
restore_regs(reglist);
}
-static int verify_code_arm64(const void *source, const void *copy, u_int size)
-{
- int ret = memcmp(source, copy, size);
- //printf("%s %p,%#x = %d\n", __func__, source, size, ret);
- return ret;
-}
-
-// this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
-static void do_dirty_stub_base(u_int vaddr)
-{
- assert(slen <= MAXBLOCK);
- emit_loadlp_ofs(0, 0); // ldr x1, source
- emit_loadlp_ofs(0, 1); // ldr x2, copy
- emit_movz(slen*4, 2);
- emit_far_call(verify_code_arm64);
- void *jmp = out;
- emit_cbz(0, 0);
- emit_movz(vaddr & 0xffff, 0);
- emit_movk_lsl16(vaddr >> 16, 0);
- emit_far_call(get_addr);
- emit_jmpreg(0);
- set_jump_target(jmp, out);
-}
-
-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[2] & 0xffe0001f) == 0x52800002); // movz w2, #slen*4
- assert( ptr[8] == 0xd61f0000); // br x0
-}
-
-static void set_loadlp(u_int *loadl, void *lit)
-{
- uintptr_t ofs = (u_char *)lit - (u_char *)loadl;
- assert((*loadl & ~0x1f) == 0x58000000);
- assert((ofs & 3) == 0);
- assert(ofs < 0x100000);
- *loadl |= (ofs >> 2) << 5;
-}
-
-static void do_dirty_stub_emit_literals(u_int *loadlps)
-{
- set_loadlp(&loadlps[0], out);
- output_w64((uintptr_t)source);
- set_loadlp(&loadlps[1], out);
- output_w64((uintptr_t)copy);
-}
-
-static void *do_dirty_stub(int i)
-{
- assem_debug("do_dirty_stub %x\n",start+i*4);
- u_int *loadlps = (void *)out;
- do_dirty_stub_base(start + i*4);
- void *entry = out;
- load_regs_entry(i);
- if (entry == out)
- entry = instr_addr[i];
- emit_jmp(instr_addr[i]);
- do_dirty_stub_emit_literals(loadlps);
- return entry;
-}
-
-static void do_dirty_stub_ds(void)
-{
- u_int *loadlps = (void *)out;
- do_dirty_stub_base(start + 1);
- void *lit_jumpover = out;
- emit_jmp(out + 8*2);
- do_dirty_stub_emit_literals(loadlps);
- set_jump_target(lit_jumpover, out);
-}
-
-static uint64_t get_from_ldr_literal(const u_int *i)
-{
- signed int ofs;
- assert((i[0] & 0xff000000) == 0x58000000);
- ofs = i[0] << 8;
- ofs >>= 5+8;
- return *(uint64_t *)(i + ofs);
-}
-
-static uint64_t get_from_movz(const u_int *i)
-{
- assert((i[0] & 0x7fe00000) == 0x52800000);
- return (i[0] >> 5) & 0xffff;
-}
-
-// Find the "clean" entry point from a "dirty" entry point
-// by skipping past the call to verify_code
-static void *get_clean_addr(u_int *addr)
-{
- assert_dirty_stub(addr);
- return addr + 9;
-}
-
-static int verify_dirty(const u_int *ptr)
-{
- const void *source, *copy;
- u_int len;
- 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, #slen*4
- return !memcmp(source, copy, len);
-}
-
-static int isclean(void *addr)
-{
- const u_int *ptr = addr;
- if ((*ptr >> 24) == 0x58) { // the only place ldr (literal) is used
- assert_dirty_stub(ptr);
- return 0;
- }
- return 1;
-}
-
-// get source that block at addr was compiled from (host pointers)
-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
- *end = *start + get_from_movz(&ptr[2]); // movz w3, #slen*4
-}
-
/* Special assem */
static void c2op_prologue(u_int op, int i, const struct regstat *i_regs, u_int reglist)
{
save_load_regs_all(1, reglist);
- cop2_call_stall_check(op, i, i_regs, 0);
+ cop2_do_stall_check(op, i, i_regs, 0);
#ifdef PCNT
emit_movimm(op, 0);
emit_far_call(pcnt_gte_start);
host_tempreg_release();
}
-static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
+static void multdiv_assemble_arm64(int i, const struct regstat *i_regs)
{
// case 0x18: MULT
// 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
{
- 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);
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);
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);
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);
// 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);
{
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) {
- if (opcode2[i] == 0x1A) // DIV
+ if (dops[i].opcode2 == 0x1A) // DIV
emit_sub_asrimm(0,numerator,31,lr);
else
emit_movimm(~0,lr);
{
if (rs != 0)
emit_mov(rs, 0);
- emit_far_call(get_addr_ht);
+ emit_far_call(ndrc_get_addr_ht);
emit_jmpreg(0);
}
notaz.gp2x.de / pcsx_rearmed.git / blobdiff