#include "pcnt.h"
#include "arm_features.h"
-#ifndef __MACH__
-#define CALLER_SAVE_REGS 0x100f
-#else
-#define CALLER_SAVE_REGS 0x120f
-#endif
-
-#define unused __attribute__((unused))
-
#ifdef DRC_DBG
#pragma GCC diagnostic ignored "-Wunused-function"
#pragma GCC diagnostic ignored "-Wunused-variable"
0,
0};
-static u_int needs_clear_cache[1<<(TARGET_SIZE_2-17)];
-
/* Linker */
static void set_jump_target(void *addr, void *target_)
// get address that insn one after stub loads (dyna_linker arg1),
// treat it as a pointer to branch insn,
// return addr where that branch jumps to
+#if 0
static void *get_pointer(void *stub)
{
//printf("get_pointer(%x)\n",(int)stub);
assert((*i_ptr&0x0f000000)==0x0a000000); // b
return (u_char *)i_ptr+((*i_ptr<<8)>>6)+8;
}
-
-// Find the "clean" entry point from a "dirty" entry point
-// by skipping past the call to verify_code
-static void *get_clean_addr(void *addr)
-{
- signed int *ptr = addr;
- #ifndef HAVE_ARMV7
- ptr+=4;
- #else
- ptr+=6;
- #endif
- if((*ptr&0xFF000000)!=0xeb000000) ptr++;
- assert((*ptr&0xFF000000)==0xeb000000); // bl instruction
- ptr++;
- if((*ptr&0xFF000000)==0xea000000) {
- return (char *)ptr+((*ptr<<8)>>6)+8; // follow jump
- }
- return ptr;
-}
-
-static int verify_dirty(const u_int *ptr)
-{
- #ifndef HAVE_ARMV7
- u_int offset;
- // get from literal pool
- assert((*ptr&0xFFFF0000)==0xe59f0000);
- offset=*ptr&0xfff;
- u_int source=*(u_int*)((void *)ptr+offset+8);
- ptr++;
- assert((*ptr&0xFFFF0000)==0xe59f0000);
- offset=*ptr&0xfff;
- u_int copy=*(u_int*)((void *)ptr+offset+8);
- ptr++;
- assert((*ptr&0xFFFF0000)==0xe59f0000);
- offset=*ptr&0xfff;
- u_int len=*(u_int*)((void *)ptr+offset+8);
- ptr++;
- ptr++;
- #else
- // ARMv7 movw/movt
- assert((*ptr&0xFFF00000)==0xe3000000);
- u_int source=(ptr[0]&0xFFF)+((ptr[0]>>4)&0xF000)+((ptr[2]<<16)&0xFFF0000)+((ptr[2]<<12)&0xF0000000);
- u_int copy=(ptr[1]&0xFFF)+((ptr[1]>>4)&0xF000)+((ptr[3]<<16)&0xFFF0000)+((ptr[3]<<12)&0xF0000000);
- u_int len=(ptr[4]&0xFFF)+((ptr[4]>>4)&0xF000);
- ptr+=6;
- #endif
- if((*ptr&0xFF000000)!=0xeb000000) ptr++;
- assert((*ptr&0xFF000000)==0xeb000000); // bl instruction
- //printf("verify_dirty: %x %x %x\n",source,copy,len);
- return !memcmp((void *)source,(void *)copy,len);
-}
-
-// This doesn't necessarily find all clean entry points, just
-// guarantees that it's not dirty
-static int isclean(void *addr)
-{
- #ifndef HAVE_ARMV7
- u_int *ptr=((u_int *)addr)+4;
- #else
- u_int *ptr=((u_int *)addr)+6;
- #endif
- if((*ptr&0xFF000000)!=0xeb000000) ptr++;
- if((*ptr&0xFF000000)!=0xeb000000) return 1; // bl instruction
- if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code) return 0;
- if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code_ds) 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)
-{
- u_int *ptr = addr;
- #ifndef HAVE_ARMV7
- u_int offset;
- // get from literal pool
- assert((*ptr&0xFFFF0000)==0xe59f0000);
- offset=*ptr&0xfff;
- u_int source=*(u_int*)((void *)ptr+offset+8);
- ptr++;
- //assert((*ptr&0xFFFF0000)==0xe59f0000);
- //offset=*ptr&0xfff;
- //u_int copy=*(u_int*)((void *)ptr+offset+8);
- ptr++;
- assert((*ptr&0xFFFF0000)==0xe59f0000);
- offset=*ptr&0xfff;
- u_int len=*(u_int*)((void *)ptr+offset+8);
- ptr++;
- ptr++;
- #else
- // ARMv7 movw/movt
- assert((*ptr&0xFFF00000)==0xe3000000);
- u_int source=(ptr[0]&0xFFF)+((ptr[0]>>4)&0xF000)+((ptr[2]<<16)&0xFFF0000)+((ptr[2]<<12)&0xF0000000);
- //u_int copy=(ptr[1]&0xFFF)+((ptr[1]>>4)&0xF000)+((ptr[3]<<16)&0xFFF0000)+((ptr[3]<<12)&0xF0000000);
- u_int len=(ptr[4]&0xFFF)+((ptr[4]>>4)&0xF000);
- ptr+=6;
- #endif
- if((*ptr&0xFF000000)!=0xeb000000) ptr++;
- assert((*ptr&0xFF000000)==0xeb000000); // bl instruction
- *start=(u_char *)source;
- *end=(u_char *)source+len;
-}
+#endif
// Allocate a specific ARM register.
static void alloc_arm_reg(struct regstat *cur,int i,signed char reg,int hr)
output_w32(0xe0800000|rd_rn_rm(rt,rs1,rs2));
}
+static void emit_adds(int rs1,int rs2,int rt)
+{
+ assem_debug("adds %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe0900000|rd_rn_rm(rt,rs1,rs2));
+}
+#define emit_adds_ptr emit_adds
+
static void emit_adcs(int rs1,int rs2,int rt)
{
assem_debug("adcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
output_w32(0xe5900000|rd_rn_rm(rt,15,0));
}
+#ifdef HAVE_ARMV7
static void emit_movw(u_int imm,u_int rt)
{
assert(imm<65536);
assem_debug("movt %s,#%d (0x%x)\n",regname[rt],imm&0xffff0000,imm&0xffff0000);
output_w32(0xe3400000|rd_rn_rm(rt,0,0)|((imm>>16)&0xfff)|((imm>>12)&0xf0000));
}
+#endif
static void emit_movimm(u_int imm,u_int rt)
{
static void emit_loadreg(int r, int hr)
{
- if(r&64) {
- SysPrintf("64bit load in 32bit mode!\n");
- assert(0);
- return;
- }
- if((r&63)==0)
+ assert(hr != EXCLUDE_REG);
+ if (r == 0)
emit_zeroreg(hr);
else {
- int addr = (int)&psxRegs.GPR.r[r];
+ void *addr;
switch (r) {
//case HIREG: addr = &hi; break;
//case LOREG: addr = &lo; break;
- case CCREG: addr = (int)&cycle_count; break;
- case CSREG: addr = (int)&Status; break;
- case INVCP: addr = (int)&invc_ptr; break;
- default: assert(r < 34); break;
+ case CCREG: addr = &cycle_count; break;
+ case CSREG: addr = &Status; break;
+ case INVCP: addr = &invc_ptr; break;
+ case ROREG: addr = &ram_offset; break;
+ default:
+ assert(r < 34);
+ addr = &psxRegs.GPR.r[r];
+ break;
}
- u_int offset = addr-(u_int)&dynarec_local;
+ u_int offset = (u_char *)addr - (u_char *)&dynarec_local;
assert(offset<4096);
- assem_debug("ldr %s,fp+%d\n",regname[hr],offset);
+ assem_debug("ldr %s,fp+%d # r%d\n",regname[hr],offset,r);
output_w32(0xe5900000|rd_rn_rm(hr,FP,0)|offset);
}
}
static void emit_storereg(int r, int hr)
{
- if(r&64) {
- SysPrintf("64bit store in 32bit mode!\n");
- assert(0);
- return;
- }
+ assert(hr != EXCLUDE_REG);
int addr = (int)&psxRegs.GPR.r[r];
switch (r) {
//case HIREG: addr = &hi; break;
}
u_int offset = addr-(u_int)&dynarec_local;
assert(offset<4096);
- assem_debug("str %s,fp+%d\n",regname[hr],offset);
+ assem_debug("str %s,fp+%d # r%d\n",regname[hr],offset,r);
output_w32(0xe5800000|rd_rn_rm(hr,FP,0)|offset);
}
}
}
-static void emit_addimm_no_flags(u_int imm,u_int rt)
-{
- emit_addimm(rt,imm,rt);
-}
-
static void emit_addnop(u_int r)
{
assert(r<16);
output_w32(0x23a00000|rd_rn_rm(rt,0,0)|armval);
}
+static void emit_cmovs_imm(int imm,int rt)
+{
+ assem_debug("movmi %s,#%d\n",regname[rt],imm);
+ u_int armval;
+ genimm_checked(imm,&armval);
+ output_w32(0x43a00000|rd_rn_rm(rt,0,0)|armval);
+}
+
static void emit_cmovne_reg(int rs,int rt)
{
assem_debug("movne %s,%s\n",regname[rt],regname[rs]);
output_w32(0xe1500000|rd_rn_rm(0,rs,rt));
}
+static void emit_cmpcs(int rs,int rt)
+{
+ assem_debug("cmpcs %s,%s\n",regname[rs],regname[rt]);
+ output_w32(0x21500000|rd_rn_rm(0,rs,rt));
+}
+
static void emit_set_gz32(int rs, int rt)
{
//assem_debug("set_gz32\n");
output_w32(0x3a000000|offset);
}
+static void *emit_cbz(int rs, const void *a)
+{
+ void *ret;
+ emit_test(rs, rs);
+ ret = out;
+ emit_jeq(a);
+ return ret;
+}
+
static unused void emit_callreg(u_int r)
{
assert(r<15);
assem_debug("ldr %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]);
output_w32(0xe7900000|rd_rn_rm(rt,rs1,rs2)|0x100);
}
+#define emit_readptr_dualindexedx_ptrlen emit_readword_dualindexedx4
+
+static void emit_ldr_dualindexed(int rs1, int rs2, int rt)
+{
+ assem_debug("ldr %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe7900000|rd_rn_rm(rt,rs1,rs2));
+}
static void emit_ldrcc_dualindexed(int rs1, int rs2, int rt)
{
output_w32(0x37900000|rd_rn_rm(rt,rs1,rs2));
}
+static void emit_ldrb_dualindexed(int rs1, int rs2, int rt)
+{
+ assem_debug("ldrb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe7d00000|rd_rn_rm(rt,rs1,rs2));
+}
+
static void emit_ldrccb_dualindexed(int rs1, int rs2, int rt)
{
assem_debug("ldrccb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
output_w32(0x37d00000|rd_rn_rm(rt,rs1,rs2));
}
+static void emit_ldrsb_dualindexed(int rs1, int rs2, int rt)
+{
+ assem_debug("ldrsb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe19000d0|rd_rn_rm(rt,rs1,rs2));
+}
+
static void emit_ldrccsb_dualindexed(int rs1, int rs2, int rt)
{
assem_debug("ldrccsb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
output_w32(0x319000d0|rd_rn_rm(rt,rs1,rs2));
}
+static void emit_ldrh_dualindexed(int rs1, int rs2, int rt)
+{
+ assem_debug("ldrh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe19000b0|rd_rn_rm(rt,rs1,rs2));
+}
+
static void emit_ldrcch_dualindexed(int rs1, int rs2, int rt)
{
assem_debug("ldrcch %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
output_w32(0x319000b0|rd_rn_rm(rt,rs1,rs2));
}
+static void emit_ldrsh_dualindexed(int rs1, int rs2, int rt)
+{
+ assem_debug("ldrsh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe19000f0|rd_rn_rm(rt,rs1,rs2));
+}
+
static void emit_ldrccsh_dualindexed(int rs1, int rs2, int rt)
{
assem_debug("ldrccsh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
output_w32(0x319000f0|rd_rn_rm(rt,rs1,rs2));
}
+static void emit_str_dualindexed(int rs1, int rs2, int rt)
+{
+ assem_debug("str %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe7800000|rd_rn_rm(rt,rs1,rs2));
+}
+
+static void emit_strb_dualindexed(int rs1, int rs2, int rt)
+{
+ assem_debug("strb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe7c00000|rd_rn_rm(rt,rs1,rs2));
+}
+
+static void emit_strh_dualindexed(int rs1, int rs2, int rt)
+{
+ assem_debug("strh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+ output_w32(0xe18000b0|rd_rn_rm(rt,rs1,rs2));
+}
+
static void emit_movsbl_indexed(int offset, int rs, int rt)
{
assert(offset>-256&&offset<256);
assem_debug("ldr %s,fp+%d\n",regname[rt],offset);
output_w32(0xe5900000|rd_rn_rm(rt,FP,0)|offset);
}
+#define emit_readptr emit_readword
static void emit_writeword_indexed(int rt, int offset, int rs)
{
}
// special case for checking invalid_code
-static void emit_cmpmem_indexedsr12_reg(int base,int r,int imm)
+static void emit_ldrb_indexedsr12_reg(int base, int r, int rt)
{
- assert(imm<128&&imm>=0);
- assert(r>=0&&r<16);
- assem_debug("ldrb lr,%s,%s lsr #12\n",regname[base],regname[r]);
- output_w32(0xe7d00000|rd_rn_rm(HOST_TEMPREG,base,r)|0x620);
- emit_cmpimm(HOST_TEMPREG,imm);
+ assem_debug("ldrb %s,%s,%s lsr #12\n",regname[rt],regname[base],regname[r]);
+ output_w32(0xe7d00000|rd_rn_rm(rt,base,r)|0x620);
}
static void emit_callne(int a)
output_w32(0x13800000|rd_rn_rm(rt,rs,0)|armval);
}
-static void emit_andne_imm(int rs,int imm,int rt)
-{
- u_int armval;
- genimm_checked(imm,&armval);
- assem_debug("andne %s,%s,#%d\n",regname[rt],regname[rs],imm);
- output_w32(0x12000000|rd_rn_rm(rt,rs,0)|armval);
-}
-
static unused void emit_addpl_imm(int rs,int imm,int rt)
{
u_int armval;
}
// 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)
{
u_char *ptr=(u_char *)addr;
assert((ptr[3]&0x0e)==0xa);
emit_loadlp(target,0);
emit_loadlp((u_int)addr,1);
- 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_readword(&last_count,ECX);
- emit_add(HOST_CCREG,ECX,HOST_CCREG);
- emit_readword(&next_interupt,ECX);
- emit_writeword(HOST_CCREG,&Count);
- emit_sub(HOST_CCREG,ECX,HOST_CCREG);
- emit_writeword(ECX,&last_count);
-#endif
-//DEBUG <
- emit_far_jump(linker);
+ assert(ndrc->translation_cache <= addr &&
+ addr < ndrc->translation_cache + sizeof(ndrc->translation_cache));
+ emit_far_jump(dyna_linker);
}
static void check_extjump2(void *src)
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);
if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
mov_loadtype_adj(type,0,rt);
{
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;
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) {
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),2);
+ emit_addimm(cc<0?2:cc,adj,2);
}
else {
emit_readword(&last_count,3);
- emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2);
+ emit_addimm(cc<0?2:cc,adj,2);
emit_add(2,3,2);
emit_writeword(2,&Count);
}
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);
int cc=get_reg(regmap,CCREG);
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);
emit_movimm((u_int)handler,3);
// returns new cycle_count
emit_far_call(jump_handler_write_h);
- emit_addimm(0,-CLOCK_ADJUST(adj),cc<0?2:cc);
+ emit_addimm(0,-adj,cc<0?2:cc);
if(cc<0)
emit_storereg(CCREG,2);
restore_regs(reglist);
}
-// this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
-static void do_dirty_stub_emit_args(u_int arg0, u_int source_len)
-{
- #ifndef HAVE_ARMV7
- emit_loadlp((int)source, 1);
- emit_loadlp((int)copy, 2);
- emit_loadlp(source_len, 3);
- #else
- emit_movw(((u_int)source)&0x0000FFFF, 1);
- emit_movw(((u_int)copy)&0x0000FFFF, 2);
- emit_movt(((u_int)source)&0xFFFF0000, 1);
- emit_movt(((u_int)copy)&0xFFFF0000, 2);
- emit_movw(source_len, 3);
- #endif
- emit_movimm(arg0, 0);
-}
-
-static void *do_dirty_stub(int i, u_int source_len)
-{
- assem_debug("do_dirty_stub %x\n",start+i*4);
- do_dirty_stub_emit_args(start + i*4, source_len);
- emit_far_call(verify_code);
- void *entry = out;
- load_regs_entry(i);
- if (entry == out)
- entry = instr_addr[i];
- emit_jmp(instr_addr[i]);
- return entry;
-}
-
-static void do_dirty_stub_ds(u_int source_len)
-{
- do_dirty_stub_emit_args(start + 1, source_len);
- emit_far_call(verify_code_ds);
-}
-
/* Special assem */
static void c2op_prologue(u_int op, int i, const struct regstat *i_regs, u_int reglist)
}
#else
if(cv==3&&shift)
- emit_far_call((int)gteMVMVA_part_cv3sh12_arm);
+ emit_far_call(gteMVMVA_part_cv3sh12_arm);
else {
emit_movimm(shift,1);
- emit_far_call((int)(need_flags?gteMVMVA_part_arm:gteMVMVA_part_nf_arm));
+ emit_far_call(need_flags?gteMVMVA_part_arm:gteMVMVA_part_nf_arm);
}
if(need_flags||need_ir)
c2op_call_MACtoIR(lm,need_flags);
static void do_mfc2_31_one(u_int copr,signed char temp)
{
emit_readword(®_cop2d[copr],temp);
- emit_testimm(temp,0x8000); // do we need this?
- emit_andne_imm(temp,0,temp);
- emit_cmpimm(temp,0xf80);
- emit_andimm(temp,0xf80,temp);
- emit_cmovae_imm(0xf80,temp);
+ emit_lsls_imm(temp,16,temp);
+ emit_cmovs_imm(0,temp);
+ emit_cmpimm(temp,0xf80<<16);
+ emit_andimm(temp,0xf80<<16,temp);
+ emit_cmovae_imm(0xf80<<16,temp);
}
static void c2op_mfc2_29_assemble(signed char tl, signed char temp)
temp = HOST_TEMPREG;
}
do_mfc2_31_one(9,temp);
- emit_shrimm(temp,7,tl);
+ emit_shrimm(temp,7+16,tl);
do_mfc2_31_one(10,temp);
- emit_orrshr_imm(temp,2,tl);
+ emit_orrshr_imm(temp,2+16,tl);
do_mfc2_31_one(11,temp);
- emit_orrshl_imm(temp,3,tl);
+ emit_orrshr_imm(temp,-3+16,tl);
emit_writeword(tl,®_cop2d[29]);
if (temp == HOST_TEMPREG)
host_tempreg_release();
}
-static void multdiv_assemble_arm(int i,struct regstat *i_regs)
+static void multdiv_assemble_arm(int i, const struct regstat *i_regs)
{
// case 0x18: MULT
// case 0x19: MULTU