* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
#ifdef PCSX
+#include "../gte.h"
+#define FLAGLESS
+#include "../gte.h"
+#undef FLAGLESS
#include "../gte_arm.h"
#include "../gte_neon.h"
#include "pcnt.h"
#endif
+#if !BASE_ADDR_FIXED
+char translation_cache[1 << TARGET_SIZE_2] __attribute__((aligned(4096)));
+#endif
+
extern int cycle_count;
extern int last_count;
extern int pcaddr;
return 1;
}
+// get source that block at addr was compiled from (host pointers)
void get_bounds(int addr,u_int *start,u_int *end)
{
u_int *ptr=(u_int *)addr;
assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],imm);
output_w32(0xe2800000|rd_rn_rm(rt,rs,0)|armval);
}else if(genimm(-imm,&armval)) {
- assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],imm);
+ assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],-imm);
output_w32(0xe2400000|rd_rn_rm(rt,rs,0)|armval);
}else if(imm<0) {
assert(imm>-65536);
// put rt_val into rt, potentially making use of rs with value rs_val
static void emit_movimm_from(u_int rs_val,int rs,u_int rt_val,int rt)
{
- u_int xor=rs_val^rt_val;
+ u_int armval;
+ int diff;
+ if(genimm(rt_val,&armval)) {
+ assem_debug("mov %s,#%d\n",regname[rt],rt_val);
+ output_w32(0xe3a00000|rd_rn_rm(rt,0,0)|armval);
+ return;
+ }
+ if(genimm(~rt_val,&armval)) {
+ assem_debug("mvn %s,#%d\n",regname[rt],rt_val);
+ output_w32(0xe3e00000|rd_rn_rm(rt,0,0)|armval);
+ return;
+ }
+ diff=rt_val-rs_val;
+ if(genimm(diff,&armval)) {
+ assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],diff);
+ output_w32(0xe2800000|rd_rn_rm(rt,rs,0)|armval);
+ return;
+ }else if(genimm(-diff,&armval)) {
+ assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],-diff);
+ output_w32(0xe2400000|rd_rn_rm(rt,rs,0)|armval);
+ return;
+ }
+ emit_movimm(rt_val,rt);
+}
+
+// return 1 if above function can do it's job cheaply
+static int is_similar_value(u_int v1,u_int v2)
+{
u_int xs;
- for(xs=xor;xs!=0&&(xs&3)==0;xs>>=2)
+ int diff;
+ if(v1==v2) return 1;
+ diff=v2-v1;
+ for(xs=diff;xs!=0&&(xs&3)==0;xs>>=2)
;
- if(xs<0x100)
- emit_xorimm(rs,xor,rt);
- else
- emit_movimm(rt_val,rt);
+ if(xs<0x100) return 1;
+ for(xs=-diff;xs!=0&&(xs&3)==0;xs>>=2)
+ ;
+ if(xs<0x100) return 1;
+ return 0;
}
// trashes r2
temp=r; break;
}
}
- if(rt>=0)
+ if(rt>=0&&rt1[i]!=0)
reglist&=~(1<<rt);
if(temp==-1) {
save_regs(reglist);
return;
handler=get_direct_memhandler(mem_rtab,addr,type,&host_addr);
if (handler==0) {
- if(rt<0)
+ if(rt<0||rt1[i]==0)
return;
if(addr!=host_addr)
emit_movimm_from(addr,rs,host_addr,rs);
}
// call a memhandler
- if(rt>=0)
+ if(rt>=0&&rt1[i]!=0)
reglist&=~(1<<rt);
save_regs(reglist);
if(target==0)
else
emit_call(handler);
- if(rt>=0) {
+ if(rt>=0&&rt1[i]!=0) {
switch(type) {
case LOADB_STUB: emit_signextend8(0,rt); break;
case LOADBU_STUB: emit_andimm(0,0xff,rt); break;
set_jump_target(restore_jump,(int)out);
restore_regs(reglist);
ra=stubs[n][2];
- if(!restore_jump) ra+=4*3; // skip invcode check
emit_jmp(ra);
#else // if !PCSX
if(addr<0) addr=get_reg(i_regmap,-1);
else
#endif
emit_jno(0);
+ if(ram_offset!=0) {
+ emit_addimm(addr,ram_offset,HOST_TEMPREG);
+ addr=*addr_reg_override=HOST_TEMPREG;
+ }
}
return jaddr;
jaddr=emit_fastpath_cmp_jump(i,temp2,&fastload_reg_override);
}
else {
+ if(ram_offset&&memtarget) {
+ emit_addimm(temp2,ram_offset,HOST_TEMPREG);
+ fastload_reg_override=HOST_TEMPREG;
+ }
if (opcode[i]==0x22||opcode[i]==0x26) {
emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
}else{
emit_writeword(HOST_CCREG,(int)&last_count);
emit_movimm(0,HOST_CCREG);
emit_storereg(CCREG,HOST_CCREG);
- if(s!=1)
- emit_mov(s,1);
+ emit_loadreg(rs1[i],1);
emit_movimm(copr,0);
emit_call((int)pcsx_mtc0_ds);
+ emit_loadreg(rs1[i],s);
return;
}
#endif
//else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12);
//else
#ifdef PCSX
- if(s!=1)
+ if(s==HOST_CCREG)
+ emit_loadreg(rs1[i],1);
+ else if(s!=1)
emit_mov(s,1);
emit_movimm(copr,0);
emit_call((int)pcsx_mtc0);
#endif
if(copr==9||copr==11||copr==12||copr==13) {
emit_readword((int)&Count,HOST_CCREG);
- emit_readword((int)&next_interupt,ECX);
+ emit_readword((int)&next_interupt,HOST_TEMPREG);
emit_addimm(HOST_CCREG,-CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
- emit_sub(HOST_CCREG,ECX,HOST_CCREG);
- emit_writeword(ECX,(int)&last_count);
+ emit_sub(HOST_CCREG,HOST_TEMPREG,HOST_CCREG);
+ emit_writeword(HOST_TEMPREG,(int)&last_count);
emit_storereg(CCREG,HOST_CCREG);
}
if(copr==12||copr==13) {
assert(!is_delayslot);
emit_readword((int)&pending_exception,14);
+ emit_test(14,14);
+ emit_jne((int)&do_interrupt);
}
emit_loadreg(rs1[i],s);
if(get_reg(i_regs->regmap,rs1[i]|64)>=0)
emit_loadreg(rs1[i]|64,get_reg(i_regs->regmap,rs1[i]|64));
- if(copr==12||copr==13) {
- emit_test(14,14);
- emit_jne((int)&do_interrupt);
- }
cop1_usable=0;
}
else
static void c2op_prologue(u_int op,u_int reglist)
{
save_regs_all(reglist);
+#ifdef PCNT
+ emit_movimm(op,0);
+ emit_call((int)pcnt_gte_start);
+#endif
emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); // cop2 regs
}
static void c2op_epilogue(u_int op,u_int reglist)
{
+#ifdef PCNT
+ emit_movimm(op,0);
+ emit_call((int)pcnt_gte_end);
+#endif
restore_regs_all(reglist);
}
+static void c2op_call_MACtoIR(int lm,int need_flags)
+{
+ if(need_flags)
+ emit_call((int)(lm?gteMACtoIR_lm1:gteMACtoIR_lm0));
+ else
+ emit_call((int)(lm?gteMACtoIR_lm1_nf:gteMACtoIR_lm0_nf));
+}
+
+static void c2op_call_rgb_func(void *func,int lm,int need_ir,int need_flags)
+{
+ emit_call((int)func);
+ // func is C code and trashes r0
+ emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0);
+ if(need_flags||need_ir)
+ c2op_call_MACtoIR(lm,need_flags);
+ emit_call((int)(need_flags?gteMACtoRGB:gteMACtoRGB_nf));
+}
+
static void c2op_assemble(int i,struct regstat *i_regs)
{
signed char temp=get_reg(i_regs->regmap,-1);
u_int c2op=source[i]&0x3f;
- u_int hr,reglist=0;
+ u_int hr,reglist_full=0,reglist;
int need_flags,need_ir;
for(hr=0;hr<HOST_REGS;hr++) {
- if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+ if(i_regs->regmap[hr]>=0) reglist_full|=1<<hr;
}
+ reglist=reglist_full&0x100f;
if (gte_handlers[c2op]!=NULL) {
need_flags=!(gte_unneeded[i+1]>>63); // +1 because of how liveness detection works
need_ir=(gte_unneeded[i+1]&0xe00)!=0xe00;
- assem_debug("gte unneeded %016llx, need_flags %d, need_ir %d\n",
- gte_unneeded[i+1],need_flags,need_ir);
-#ifdef ARMv5_ONLY
- // let's take more risk here
- need_flags=need_flags&>e_reads_flags;
-#endif
+ assem_debug("gte op %08x, unneeded %016llx, 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)
+ need_flags=0;
+ int shift = (source[i] >> 19) & 1;
+ int lm = (source[i] >> 10) & 1;
switch(c2op) {
+#ifndef DRC_DBG
case GTE_MVMVA: {
- int shift = (source[i] >> 19) & 1;
int v = (source[i] >> 15) & 3;
int cv = (source[i] >> 13) & 3;
int mx = (source[i] >> 17) & 3;
- int lm = (source[i] >> 10) & 1;
- reglist&=0x10ff; // +{r4-r7}
+ reglist=reglist_full&0x10ff; // +{r4-r7}
c2op_prologue(c2op,reglist);
/* r4,r5 = VXYZ(v) packed; r6 = &MX11(mx); r7 = &CV1(cv) */
if(v<3)
emit_movimm(shift,1);
emit_call((int)(need_flags?gteMVMVA_part_arm:gteMVMVA_part_nf_arm));
}
- if(need_flags||need_ir) {
- if(need_flags)
- emit_call((int)(lm?gteMACtoIR_lm1:gteMACtoIR_lm0));
- else
- emit_call((int)(lm?gteMACtoIR_lm1_nf:gteMACtoIR_lm0_nf)); // lm0 borked
- }
+ if(need_flags||need_ir)
+ c2op_call_MACtoIR(lm,need_flags);
#endif
break;
}
-
+ case GTE_OP:
+ c2op_prologue(c2op,reglist);
+ emit_call((int)(shift?gteOP_part_shift:gteOP_part_noshift));
+ if(need_flags||need_ir) {
+ emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0);
+ c2op_call_MACtoIR(lm,need_flags);
+ }
+ break;
+ case GTE_DPCS:
+ c2op_prologue(c2op,reglist);
+ c2op_call_rgb_func(shift?gteDPCS_part_shift:gteDPCS_part_noshift,lm,need_ir,need_flags);
+ break;
+ case GTE_INTPL:
+ c2op_prologue(c2op,reglist);
+ c2op_call_rgb_func(shift?gteINTPL_part_shift:gteINTPL_part_noshift,lm,need_ir,need_flags);
+ break;
+ case GTE_SQR:
+ c2op_prologue(c2op,reglist);
+ emit_call((int)(shift?gteSQR_part_shift:gteSQR_part_noshift));
+ if(need_flags||need_ir) {
+ emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0);
+ c2op_call_MACtoIR(lm,need_flags);
+ }
+ break;
+ case GTE_DCPL:
+ c2op_prologue(c2op,reglist);
+ c2op_call_rgb_func(gteDCPL_part,lm,need_ir,need_flags);
+ break;
+ case GTE_GPF:
+ c2op_prologue(c2op,reglist);
+ c2op_call_rgb_func(shift?gteGPF_part_shift:gteGPF_part_noshift,lm,need_ir,need_flags);
+ break;
+ case GTE_GPL:
+ c2op_prologue(c2op,reglist);
+ c2op_call_rgb_func(shift?gteGPL_part_shift:gteGPL_part_noshift,lm,need_ir,need_flags);
+ break;
+#endif
default:
- reglist&=0x100f;
c2op_prologue(c2op,reglist);
+#ifdef DRC_DBG
emit_movimm(source[i],1); // opcode
emit_writeword(1,(int)&psxRegs.code);
+#endif
emit_call((int)(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]));
break;
}
for(j=0;j<32;j++)
{
if(bitmap&(1<<j)) {
- start=BASE_ADDR+i*131072+j*4096;
+ start=(u_int)BASE_ADDR+i*131072+j*4096;
end=start+4095;
j++;
while(j<32) {