X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fassem_arm64.c;h=0b49221137710b369ca693706acb71af5aaddd26;hp=e7df2b025af204c000bc794e4701624761f397fe;hb=2330734fa3064bf3a159c3c56f9a2e005598360e;hpb=81dbbf4cbb16fc6c9a82a5b91e102c8005c5726a

diff --git a/libpcsxcore/new_dynarec/assem_arm64.c b/libpcsxcore/new_dynarec/assem_arm64.c
index e7df2b02..0b492211 100644
--- a/libpcsxcore/new_dynarec/assem_arm64.c
+++ b/libpcsxcore/new_dynarec/assem_arm64.c
@@ -23,8 +23,6 @@
 #include "pcnt.h"
 #include "arm_features.h"
 
-#define CALLER_SAVE_REGS 0x0007ffff
-
 #define unused __attribute__((unused))
 
 void do_memhandler_pre();
@@ -44,7 +42,7 @@ static void set_jump_target(void *addr, void *target)
         || (*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);
@@ -330,6 +328,7 @@ static void emit_adds64(u_int rs1, u_int rs2, u_int rt)
   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)
 {
@@ -434,6 +433,7 @@ static void emit_readdword(void *addr, u_int rt)
   else
     abort();
 }
+#define emit_readptr emit_readdword
 
 static void emit_readshword(void *addr, u_int rt)
 {
@@ -460,6 +460,7 @@ static void emit_loadreg(u_int r, u_int hr)
     case CCREG: addr = &cycle_count; break;
     case CSREG: addr = &Status; break;
     case INVCP: addr = &invc_ptr; is64 = 1; break;
+    case ROREG: addr = &ram_offset; is64 = 1; break;
     default: assert(r < 34); break;
     }
     if (is64)
@@ -621,11 +622,6 @@ static void emit_addimm_and_set_flags(int imm, u_int rt)
   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" };
@@ -813,6 +809,12 @@ static void emit_cmovl_reg(u_int rs,u_int rt)
   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]);
@@ -1041,6 +1043,7 @@ static void emit_readdword_dualindexedx8(u_int rs1, u_int rs2, u_int rt)
   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)
 {
@@ -1328,7 +1331,27 @@ static int is_similar_value(u_int v1, u_int v2)
     || 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) {
@@ -1374,10 +1397,10 @@ static void do_readstub(int n)
   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;
@@ -1389,7 +1412,7 @@ static void do_readstub(int n)
       break;
     }
   }
-  if(rt>=0&&rt1[i]!=0)
+  if(rt>=0&&dops[i].rt1!=0)
     reglist&=~(1<<rt);
   if(temp==-1) {
     save_regs(reglist);
@@ -1404,7 +1427,7 @@ static void do_readstub(int n)
   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;
@@ -1436,10 +1459,10 @@ static void do_readstub(int n)
   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)
@@ -1459,17 +1482,14 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
   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;
@@ -1480,8 +1500,8 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
     }
     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)
@@ -1491,7 +1511,7 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
   }
 
   // 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)
@@ -1500,7 +1520,7 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
     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);
@@ -1512,7 +1532,7 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
   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);
 }
@@ -1528,10 +1548,10 @@ static void do_writestub(int n)
   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);
@@ -1577,7 +1597,6 @@ static void do_writestub(int n)
     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;
@@ -1596,10 +1615,10 @@ static void do_writestub(int n)
   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)
@@ -1618,11 +1637,8 @@ static void inline_writestub(enum stub_type type, int i, u_int addr,
   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;
@@ -1640,12 +1656,12 @@ static void inline_writestub(enum stub_type type, int i, u_int addr,
   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);
@@ -1659,12 +1675,12 @@ static int verify_code_arm64(const void *source, const void *copy, u_int size)
 }
 
 // this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
-static void do_dirty_stub_base(u_int vaddr)
+static void do_dirty_stub_base(u_int vaddr, u_int source_len)
 {
-  assert(slen <= MAXBLOCK);
+  assert(source_len <= MAXBLOCK*4);
   emit_loadlp_ofs(0, 0); // ldr x1, source
   emit_loadlp_ofs(0, 1); // ldr x2, copy
-  emit_movz(slen*4, 2);
+  emit_movz(source_len, 2);
   emit_far_call(verify_code_arm64);
   void *jmp = out;
   emit_cbz(0, 0);
@@ -1679,7 +1695,7 @@ 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[2] & 0xffe0001f) == 0x52800002); // movz w2, #source_len
   assert( ptr[8]               == 0xd61f0000); // br x0
 }
 
@@ -1700,11 +1716,11 @@ static void do_dirty_stub_emit_literals(u_int *loadlps)
   output_w64((uintptr_t)copy);
 }
 
-static void *do_dirty_stub(int i)
+static void *do_dirty_stub(int i, u_int source_len)
 {
   assem_debug("do_dirty_stub %x\n",start+i*4);
   u_int *loadlps = (void *)out;
-  do_dirty_stub_base(start + i*4);
+  do_dirty_stub_base(start + i*4, source_len);
   void *entry = out;
   load_regs_entry(i);
   if (entry == out)
@@ -1714,10 +1730,10 @@ static void *do_dirty_stub(int i)
   return entry;
 }
 
-static void do_dirty_stub_ds(void)
+static void do_dirty_stub_ds(u_int source_len)
 {
   u_int *loadlps = (void *)out;
-  do_dirty_stub_base(start + 1);
+  do_dirty_stub_base(start + 1, source_len);
   void *lit_jumpover = out;
   emit_jmp(out + 8*2);
   do_dirty_stub_emit_literals(loadlps);
@@ -1754,7 +1770,7 @@ static int verify_dirty(const u_int *ptr)
   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
+  len = get_from_movz(&ptr[2]);                   // movz w3, #source_len
   return !memcmp(source, copy, len);
 }
 
@@ -1774,7 +1790,7 @@ 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
+  *end = *start + get_from_movz(&ptr[2]);           // movz w3, #source_len
 }
 
 /* Special assem */
@@ -1782,7 +1798,7 @@ static void get_bounds(void *addr, u_char **start, u_char **end)
 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);
@@ -1873,21 +1889,21 @@ static void c2op_mfc2_29_assemble(signed char tl, signed char temp)
     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);
@@ -1895,7 +1911,7 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
         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);
@@ -1907,8 +1923,8 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
     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);
@@ -1916,7 +1932,7 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
         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);
@@ -1924,7 +1940,7 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
 
         // 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);
@@ -1941,15 +1957,15 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
   {
     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);