/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* Mupen64plus/PCSX - assem_arm.c *
* Copyright (C) 2009-2011 Ari64 *
- * Copyright (C) 2010-2011 Gražvydas "notaz" Ignotas *
+ * Copyright (C) 2010-2021 Gražvydas "notaz" Ignotas *
* *
* This program is free software; you can redistribute it and/or modify *
* it under the terms of the GNU General Public License as published by *
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. *
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
-#include "../gte.h"
#define FLAGLESS
#include "../gte.h"
#undef FLAGLESS
#include "pcnt.h"
#include "arm_features.h"
-#if defined(BASE_ADDR_FIXED)
-#elif defined(BASE_ADDR_DYNAMIC)
-u_char *translation_cache;
-#else
-u_char translation_cache[1 << TARGET_SIZE_2] __attribute__((aligned(4096)));
-#endif
-
#ifndef __MACH__
#define CALLER_SAVE_REGS 0x100f
#else
emit_cmovb_imm(1,rt);
}
+static int can_jump_or_call(const void *a)
+{
+ intptr_t offset = (u_char *)a - out - 8;
+ return (-33554432 <= offset && offset < 33554432);
+}
+
static void emit_call(const void *a_)
{
int a = (int)a_;
emit_loadlp(target,0);
emit_loadlp((u_int)addr,1);
- assert(addr>=translation_cache&&addr<(translation_cache+(1<<TARGET_SIZE_2)));
+ assert(addr>=ndrc->translation_cache&&addr<(ndrc->translation_cache+(1<<TARGET_SIZE_2)));
//assert((target>=0x80000000&&target<0x80800000)||(target>0xA4000000&&target<0xA4001000));
//DEBUG >
#ifdef DEBUG_CYCLE_COUNT
emit_writeword(ECX,&last_count);
#endif
//DEBUG <
- emit_jmp(linker);
+ emit_far_jump(linker);
}
static void check_extjump2(void *src)
enum stub_type type=stubs[n].type;
int i=stubs[n].a;
int rs=stubs[n].b;
- struct regstat *i_regs=(struct regstat *)stubs[n].c;
+ const struct regstat *i_regs=(struct regstat *)stubs[n].c;
u_int reglist=stubs[n].e;
- signed char *i_regmap=i_regs->regmap;
+ const signed char *i_regmap=i_regs->regmap;
int rt;
if(itype[i]==C1LS||itype[i]==C2LS||itype[i]==LOADLR) {
rt=get_reg(i_regmap,FTEMP);
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_call(handler);
+ emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2);
+ emit_far_call(handler);
if(itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]!=0)) {
mov_loadtype_adj(type,0,rt);
}
emit_jmp(stubs[n].retaddr); // return address
}
-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);
if(rs<0) rs=get_reg(regmap,-1);
assert(rs>=0);
- u_int is_dynamic,far_call=0;
+ u_int is_dynamic;
uintptr_t host_addr = 0;
void *handler;
int cc=get_reg(regmap,CCREG);
- if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj+1),cc,target?rs:-1,rt))
+ if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj),cc,target?rs:-1,rt))
return;
handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr);
if (handler == NULL) {
emit_movimm(addr,0);
else if(rs!=0)
emit_mov(rs,0);
- int offset=(u_char *)handler-out-8;
- if(offset<-33554432||offset>=33554432) {
- // unreachable memhandler, a plugin func perhaps
- emit_movimm((u_int)handler,12);
- far_call=1;
- }
if(cc<0)
emit_loadreg(CCREG,2);
if(is_dynamic) {
emit_movimm(((u_int *)mem_rtab)[addr>>12]<<1,1);
- emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2);
+ emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2);
}
else {
emit_readword(&last_count,3);
- emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2);
+ emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2);
emit_add(2,3,2);
emit_writeword(2,&Count);
}
- if(far_call)
- emit_callreg(12);
- else
- emit_call(handler);
+ emit_far_call(handler);
if(rt>=0&&rt1[i]!=0) {
switch(type) {
enum stub_type type=stubs[n].type;
int i=stubs[n].a;
int rs=stubs[n].b;
- struct regstat *i_regs=(struct regstat *)stubs[n].c;
+ const struct regstat *i_regs=(struct regstat *)stubs[n].c;
u_int reglist=stubs[n].e;
- signed char *i_regmap=i_regs->regmap;
+ const signed char *i_regmap=i_regs->regmap;
int rt,r;
if(itype[i]==C1LS||itype[i]==C2LS) {
rt=get_reg(i_regmap,r=FTEMP);
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_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2);
// returns new cycle_count
- emit_call(handler);
- emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d+1),cc<0?2:cc);
+ emit_far_call(handler);
+ emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d),cc<0?2:cc);
if(cc<0)
emit_storereg(CCREG,2);
if(restore_jump)
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 cc=get_reg(regmap,CCREG);
if(cc<0)
emit_loadreg(CCREG,2);
- emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2);
+ emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2);
emit_movimm((u_int)handler,3);
// returns new cycle_count
- emit_call(jump_handler_write_h);
- emit_addimm(0,-CLOCK_ADJUST(adj+1),cc<0?2:cc);
+ emit_far_call(jump_handler_write_h);
+ emit_addimm(0,-CLOCK_ADJUST(adj),cc<0?2:cc);
if(cc<0)
emit_storereg(CCREG,2);
restore_regs(reglist);
{
assem_debug("do_dirty_stub %x\n",start+i*4);
do_dirty_stub_emit_args(start + i*4);
- emit_call(verify_code);
+ emit_far_call(verify_code);
void *entry = out;
load_regs_entry(i);
if (entry == out)
static void do_dirty_stub_ds()
{
do_dirty_stub_emit_args(start + 1);
- emit_call(verify_code_ds);
+ emit_far_call(verify_code_ds);
}
/* 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_regs_all(reglist);
+ cop2_call_stall_check(op, i, i_regs, 0);
#ifdef PCNT
- emit_movimm(op,0);
- emit_call(pcnt_gte_start);
+ emit_movimm(op, 0);
+ emit_far_call(pcnt_gte_start);
#endif
- emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); // cop2 regs
+ emit_addimm(FP, (u_char *)&psxRegs.CP2D.r[0] - (u_char *)&dynarec_local, 0); // cop2 regs
}
static void c2op_epilogue(u_int op,u_int reglist)
{
#ifdef PCNT
emit_movimm(op,0);
- emit_call(pcnt_gte_end);
+ emit_far_call(pcnt_gte_end);
#endif
restore_regs_all(reglist);
}
static void c2op_call_MACtoIR(int lm,int need_flags)
{
if(need_flags)
- emit_call(lm?gteMACtoIR_lm1:gteMACtoIR_lm0);
+ emit_far_call(lm?gteMACtoIR_lm1:gteMACtoIR_lm0);
else
- emit_call(lm?gteMACtoIR_lm1_nf:gteMACtoIR_lm0_nf);
+ emit_far_call(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(func);
+ emit_far_call(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(need_flags?gteMACtoRGB:gteMACtoRGB_nf);
+ emit_far_call(need_flags?gteMACtoRGB:gteMACtoRGB_nf);
}
-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;
- int need_flags,need_ir;
- for(hr=0;hr<HOST_REGS;hr++) {
- if(i_regs->regmap[hr]>=0) reglist_full|=1<<hr;
- }
- reglist=reglist_full&CALLER_SAVE_REGS;
+ u_int c2op = source[i] & 0x3f;
+ u_int reglist_full = get_host_reglist(i_regs->regmap);
+ u_int reglist = reglist_full & CALLER_SAVE_REGS;
+ int need_flags, need_ir;
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 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)
+ if(HACK_ENABLED(NDHACK_GTE_NO_FLAGS))
need_flags=0;
int shift = (source[i] >> 19) & 1;
int lm = (source[i] >> 10) & 1;
int cv = (source[i] >> 13) & 3;
int mx = (source[i] >> 17) & 3;
reglist=reglist_full&(CALLER_SAVE_REGS|0xf0); // +{r4-r7}
- c2op_prologue(c2op,reglist);
+ c2op_prologue(c2op,i,i_regs,reglist);
/* r4,r5 = VXYZ(v) packed; r6 = &MX11(mx); r7 = &CV1(cv) */
if(v<3)
emit_ldrd(v*8,0,4);
emit_readword(&zeromem_ptr,7);
#ifdef __ARM_NEON__
emit_movimm(source[i],1); // opcode
- emit_call(gteMVMVA_part_neon);
+ emit_far_call(gteMVMVA_part_neon);
if(need_flags) {
emit_movimm(lm,1);
- emit_call(gteMACtoIR_flags_neon);
+ emit_far_call(gteMACtoIR_flags_neon);
}
#else
if(cv==3&&shift)
- emit_call((int)gteMVMVA_part_cv3sh12_arm);
+ emit_far_call((int)gteMVMVA_part_cv3sh12_arm);
else {
emit_movimm(shift,1);
- emit_call((int)(need_flags?gteMVMVA_part_arm:gteMVMVA_part_nf_arm));
+ emit_far_call((int)(need_flags?gteMVMVA_part_arm:gteMVMVA_part_nf_arm));
}
if(need_flags||need_ir)
c2op_call_MACtoIR(lm,need_flags);
#endif
#else /* if not HAVE_ARMV5 */
- c2op_prologue(c2op,reglist);
+ c2op_prologue(c2op,i,i_regs,reglist);
emit_movimm(source[i],1); // opcode
emit_writeword(1,&psxRegs.code);
- emit_call((int)(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]));
+ emit_far_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]);
#endif
break;
}
case GTE_OP:
- c2op_prologue(c2op,reglist);
- emit_call(shift?gteOP_part_shift:gteOP_part_noshift);
+ c2op_prologue(c2op,i,i_regs,reglist);
+ emit_far_call(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_prologue(c2op,i,i_regs,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_prologue(c2op,i,i_regs,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(shift?gteSQR_part_shift:gteSQR_part_noshift);
+ c2op_prologue(c2op,i,i_regs,reglist);
+ emit_far_call(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_prologue(c2op,i,i_regs,reglist);
c2op_call_rgb_func(gteDCPL_part,lm,need_ir,need_flags);
break;
case GTE_GPF:
- c2op_prologue(c2op,reglist);
+ c2op_prologue(c2op,i,i_regs,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_prologue(c2op,i,i_regs,reglist);
c2op_call_rgb_func(shift?gteGPL_part_shift:gteGPL_part_noshift,lm,need_ir,need_flags);
break;
#endif
default:
- c2op_prologue(c2op,reglist);
+ c2op_prologue(c2op,i,i_regs,reglist);
#ifdef DRC_DBG
emit_movimm(source[i],1); // opcode
emit_writeword(1,&psxRegs.code);
#endif
- emit_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]);
+ emit_far_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]);
break;
}
c2op_epilogue(c2op,reglist);
static void do_jump_vaddr(int rs)
{
- emit_jmp(jump_vaddr_reg[rs]);
+ emit_far_jump(jump_vaddr_reg[rs]);
}
static void do_preload_rhash(int r) {
#endif
}
-static void mark_clear_cache(void *target)
-{
- u_long offset = (u_char *)target - translation_cache;
- u_int mask = 1u << ((offset >> 12) & 31);
- if (!(needs_clear_cache[offset >> 17] & mask)) {
- char *start = (char *)((u_long)target & ~4095ul);
- start_tcache_write(start, start + 4096);
- needs_clear_cache[offset >> 17] |= mask;
- }
-}
-
-// Clearing the cache is rather slow on ARM Linux, so mark the areas
-// that need to be cleared, and then only clear these areas once.
-static void do_clear_cache()
-{
- int i,j;
- for (i=0;i<(1<<(TARGET_SIZE_2-17));i++)
- {
- u_int bitmap=needs_clear_cache[i];
- if(bitmap) {
- u_char *start, *end;
- for(j=0;j<32;j++)
- {
- if(bitmap&(1<<j)) {
- start=translation_cache+i*131072+j*4096;
- end=start+4095;
- j++;
- while(j<32) {
- if(bitmap&(1<<j)) {
- end+=4096;
- j++;
- }else{
- end_tcache_write(start, end);
- break;
- }
- }
- }
- }
- needs_clear_cache[i]=0;
- }
- }
-}
-
// CPU-architecture-specific initialization
-static void arch_init() {
+static void arch_init(void)
+{
+ uintptr_t diff = (u_char *)&ndrc->tramp.f - (u_char *)&ndrc->tramp.ops - 8;
+ struct tramp_insns *ops = ndrc->tramp.ops;
+ size_t i;
+ assert(!(diff & 3));
+ assert(diff < 0x1000);
+ start_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));
+ for (i = 0; i < ARRAY_SIZE(ndrc->tramp.ops); i++)
+ ops[i].ldrpc = 0xe5900000 | rd_rn_rm(15,15,0) | diff; // ldr pc, [=val]
+ end_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));
}
// vim:shiftwidth=2:expandtab