}
// Alloc cycle count into dedicated register
-static void alloc_cc(struct regstat *cur,int i)
+static void alloc_cc(struct regstat *cur, int i)
{
- alloc_arm_reg(cur,i,CCREG,HOST_CCREG);
+ alloc_arm_reg(cur, i, CCREG, HOST_CCREG);
+}
+
+static void alloc_cc_optional(struct regstat *cur, int i)
+{
+ if (cur->regmap[HOST_CCREG] < 0) {
+ alloc_arm_reg(cur, i, CCREG, HOST_CCREG);
+ cur->noevict &= ~(1u << HOST_CCREG);
+ }
}
/* Special alloc */
int i = stubs[n].a;
int rs = stubs[n].b;
const struct regstat *i_regs = (void *)stubs[n].c;
+ int adj = (int)stubs[n].d;
u_int reglist = stubs[n].e;
const signed char *i_regmap = i_regs->regmap;
int rt;
handler=jump_handler_read32;
assert(handler);
pass_args64(rs,temp2);
- int cc=get_reg(i_regmap,CCREG);
- if(cc<0)
- emit_loadreg(CCREG,2);
- emit_addimm(cc<0?2:cc,(int)stubs[n].d,2);
+ int cc, cc_use;
+ cc = cc_use = get_reg(i_regmap, CCREG);
+ if (cc < 0)
+ emit_loadreg(CCREG, (cc_use = 2));
+ emit_addimm(cc_use, adj, 2);
+
emit_far_call(handler);
- // (no cycle reload after read)
+
+#if 0
+ // cycle reload for read32 only (value in w2 both in and out)
+ if (type == LOADW_STUB) {
+ emit_addimm(2, -adj, cc_use);
+ if (cc < 0)
+ emit_storereg(CCREG, cc_use);
+ }
+#endif
if(dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
loadstore_extend(type,0,rt);
}
u_int is_dynamic=0;
uintptr_t host_addr = 0;
void *handler;
- int cc=get_reg(regmap,CCREG);
+ int cc, cc_use;
+ cc = cc_use = get_reg(regmap, CCREG);
//if(pcsx_direct_read(type,addr,adj,cc,target?ra:-1,rt))
// return;
handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr);
emit_movimm(addr,0);
else if(ra!=0)
emit_mov(ra,0);
- if(cc<0)
- emit_loadreg(CCREG,2);
- emit_addimm(cc<0?2:cc,adj,2);
+ if (cc < 0)
+ emit_loadreg(CCREG, (cc_use = 2));
+ emit_addimm(cc_use, adj, 2);
if(is_dynamic) {
uintptr_t l1 = ((uintptr_t *)mem_rtab)[addr>>12] << 1;
intptr_t offset = (l1 & ~0xfffl) - ((intptr_t)out & ~0xfffl);
emit_far_call(handler);
- // (no cycle reload after read)
+#if 0
+ // cycle reload for read32 only (value in w2 both in and out)
+ if (type == LOADW_STUB) {
+ if (!is_dynamic)
+ emit_far_call(do_memhandler_post);
+ emit_addimm(2, -adj, cc_use);
+ if (cc < 0)
+ emit_storereg(CCREG, cc_use);
+ }
+#endif
if(rt>=0&&dops[i].rt1!=0)
loadstore_extend(type, 0, rt);
restore_regs(reglist);
int i=stubs[n].a;
int rs=stubs[n].b;
struct regstat *i_regs=(struct regstat *)stubs[n].c;
+ int adj = (int)stubs[n].d;
u_int reglist=stubs[n].e;
signed char *i_regmap=i_regs->regmap;
int rt,r;
emit_mov64(temp2,3);
host_tempreg_release();
}
- int cc=get_reg(i_regmap,CCREG);
- if(cc<0)
- emit_loadreg(CCREG,2);
- emit_addimm(cc<0?2:cc,(int)stubs[n].d,2);
- // returns new cycle_count
+ int cc, cc_use;
+ cc = cc_use = get_reg(i_regmap, CCREG);
+ if (cc < 0)
+ emit_loadreg(CCREG, (cc_use = 2));
+ emit_addimm(cc_use, adj, 2);
+
emit_far_call(handler);
- emit_addimm(0,-(int)stubs[n].d,cc<0?2:cc);
- if(cc<0)
- emit_storereg(CCREG,2);
- if(restore_jump)
+
+ // new cycle_count returned in x2
+ emit_addimm(2, -adj, cc_use);
+ if (cc < 0)
+ emit_storereg(CCREG, cc_use);
+ if (restore_jump)
set_jump_target(restore_jump, out);
restore_regs(reglist);
emit_jmp(stubs[n].retaddr);
emit_far_call(do_memhandler_pre);
emit_far_call(handler);
emit_far_call(do_memhandler_post);
- emit_addimm(0, -adj, cc_use);
+ emit_addimm(2, -adj, cc_use);
if (cc < 0)
emit_storereg(CCREG, cc_use);
restore_regs(reglist);
}
#define multdiv_assemble multdiv_assemble_arm64
+// wb_dirtys making use of stp when possible
+static void wb_dirtys(const signed char i_regmap[], u_int i_dirty)
+{
+ signed char mregs[34+1];
+ int r, hr;
+ memset(mregs, -1, sizeof(mregs));
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ r = i_regmap[hr];
+ if (hr == EXCLUDE_REG || r <= 0 || r == CCREG)
+ continue;
+ if (!((i_dirty >> hr) & 1))
+ continue;
+ assert(r < 34u);
+ mregs[r] = hr;
+ }
+ for (r = 1; r < 34; r++) {
+ if (mregs[r] < 0)
+ continue;
+ if (mregs[r+1] >= 0) {
+ uintptr_t offset = (u_char *)&psxRegs.GPR.r[r] - (u_char *)&dynarec_local;
+ emit_ldstp(1, 0, mregs[r], mregs[r+1], FP, offset);
+ r++;
+ }
+ else
+ emit_storereg(r, mregs[r]);
+ }
+}
+#define wb_dirtys wb_dirtys
+
+static void load_all_regs(const signed char i_regmap[])
+{
+ signed char mregs[34+1];
+ int r, hr;
+ memset(mregs, -1, sizeof(mregs));
+ for (hr = 0; hr < HOST_REGS; hr++) {
+ r = i_regmap[hr];
+ if (hr == EXCLUDE_REG || r < 0 || r == CCREG)
+ continue;
+ if ((u_int)r < 34u)
+ mregs[r] = hr;
+ else if (r < TEMPREG)
+ emit_loadreg(r, hr);
+ }
+ if (mregs[0] >= 0)
+ emit_zeroreg(mregs[0]); // we could use arm64's ZR instead of reg alloc
+ for (r = 1; r < 34; r++) {
+ if (mregs[r] < 0)
+ continue;
+ if (mregs[r+1] >= 0) {
+ uintptr_t offset = (u_char *)&psxRegs.GPR.r[r] - (u_char *)&dynarec_local;
+ emit_ldstp(0, 0, mregs[r], mregs[r+1], FP, offset);
+ r++;
+ }
+ else
+ emit_loadreg(r, mregs[r]);
+ }
+}
+#define load_all_regs load_all_regs
+
static void do_jump_vaddr(u_int rs)
{
if (rs != 0)