X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fassem_arm64.c;h=d35ad451eb11860d574f0209851af016eea51b7c;hb=a5cd72d0e598f037fd9d9f23948af5b2fb06e2eb;hp=ff0d1a6c293b0841953a44708eee201cc6863441;hpb=3039c914c0ac33ee89127db3bd4c53eb3c25cafd;p=pcsx_rearmed.git

diff --git a/libpcsxcore/new_dynarec/assem_arm64.c b/libpcsxcore/new_dynarec/assem_arm64.c
index ff0d1a6c..d35ad451 100644
--- a/libpcsxcore/new_dynarec/assem_arm64.c
+++ b/libpcsxcore/new_dynarec/assem_arm64.c
@@ -23,19 +23,15 @@
 #include "pcnt.h"
 #include "arm_features.h"
 
-#define unused __attribute__((unused))
-
 void do_memhandler_pre();
 void do_memhandler_post();
 
 /* Linker */
 static void set_jump_target(void *addr, void *target)
 {
-  u_int *ptr = addr;
+  u_int *ptr = NDRC_WRITE_OFFSET(addr);
   intptr_t offset = (u_char *)target - (u_char *)addr;
 
-  ptr += ndrc_write_ofs / sizeof(ptr[0]);
-
   if ((*ptr&0xFC000000) == 0x14000000) { // b
     assert(offset>=-134217728LL&&offset<134217728LL);
     *ptr=(*ptr&0xFC000000)|((offset>>2)&0x3ffffff);
@@ -47,7 +43,7 @@ static void set_jump_target(void *addr, void *target)
     // 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);
+    *ptr=(*ptr&0xFF00001F)|(((offset>>2)&0x7ffff)<<5);
   }
   else if((*ptr&0x9f000000)==0x10000000) { // adr
     // generated by do_miniht_insert
@@ -144,7 +140,7 @@ static unused const char *condname[16] = {
 
 static void output_w32(u_int word)
 {
-  *((u_int *)(out + ndrc_write_ofs)) = word;
+  *((u_int *)NDRC_WRITE_OFFSET(out)) = word;
   out += 4;
 }
 
@@ -306,6 +302,12 @@ static void emit_add(u_int rs1, u_int rs2, u_int rt)
   output_w32(0x0b000000 | rm_rn_rd(rs2, rs1, rt));
 }
 
+static void emit_adds(u_int rs1, u_int rs2, u_int rt)
+{
+  assem_debug("adds %s,%s,%s\n", regname[rt], regname[rs1], regname[rs2]);
+  output_w32(0x2b000000 | rm_rn_rd(rs2, rs1, rt));
+}
+
 static void emit_add64(u_int rs1, u_int rs2, u_int rt)
 {
   assem_debug("add %s,%s,%s\n", regname64[rt], regname64[rs1], regname64[rs2]);
@@ -319,19 +321,37 @@ static void emit_adds64(u_int rs1, u_int rs2, u_int rt)
 }
 #define emit_adds_ptr emit_adds64
 
+static void emit_add_lsrimm(u_int rs1, u_int rs2, u_int shift, u_int rt)
+{
+  assem_debug("add %s,%s,%s,lsr #%u\n",regname[rt],regname[rs1],regname[rs2],shift);
+  output_w32(0x0b400000 | rm_imm6_rn_rd(rs2, shift, rs1, rt));
+}
+
 static void emit_neg(u_int rs, u_int rt)
 {
   assem_debug("neg %s,%s\n",regname[rt],regname[rs]);
   output_w32(0x4b000000 | rm_rn_rd(rs, WZR, rt));
 }
 
+static void emit_negs(u_int rs, u_int rt)
+{
+  assem_debug("negs %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0x6b000000 | rm_rn_rd(rs, WZR, rt));
+}
+
 static void emit_sub(u_int rs1, u_int rs2, u_int rt)
 {
   assem_debug("sub %s,%s,%s\n", regname[rt], regname[rs1], regname[rs2]);
   output_w32(0x4b000000 | rm_imm6_rn_rd(rs2, 0, rs1, rt));
 }
 
-static void emit_sub_asrimm(u_int rs1, u_int rs2, u_int shift, u_int rt)
+static void emit_subs(u_int rs1, u_int rs2, u_int rt)
+{
+  assem_debug("subs %s,%s,%s\n", regname[rt], regname[rs1], regname[rs2]);
+  output_w32(0x6b000000 | rm_imm6_rn_rd(rs2, 0, rs1, rt));
+}
+
+static unused void emit_sub_asrimm(u_int rs1, u_int rs2, u_int shift, u_int rt)
 {
   assem_debug("sub %s,%s,%s,asr #%u\n",regname[rt],regname[rs1],regname[rs2],shift);
   output_w32(0x4b800000 | rm_imm6_rn_rd(rs2, shift, rs1, rt));
@@ -401,11 +421,32 @@ static void emit_movimm(u_int imm, u_int rt)
   }
 }
 
+static void emit_movimm64(uint64_t imm, u_int rt)
+{
+  u_int shift, op, imm16, insns = 0;
+  for (shift = 0; shift < 4; shift++) {
+    imm16 = (imm >> shift * 16) & 0xffff;
+    if (!imm16)
+      continue;
+    op = insns ? 0xf2800000 : 0xd2800000;
+    assem_debug("mov%c %s,#%#x", insns ? 'k' : 'z', regname64[rt], imm16);
+    if (shift)
+      assem_debug(",lsl #%u", shift * 16);
+    assem_debug("\n");
+    output_w32(op | (shift << 21) | imm16_rd(imm16, rt));
+    insns++;
+  }
+  if (!insns) {
+    assem_debug("movz %s,#0\n", regname64[rt]);
+    output_w32(0xd2800000 | imm16_rd(0, rt));
+  }
+}
+
 static void emit_readword(void *addr, u_int rt)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   if (!(offset & 3) && offset <= 16380) {
-    assem_debug("ldr %s,[x%d+%#lx]\n", regname[rt], FP, offset);
+    assem_debug("ldr %s,[x%d+%#lx]%s\n", regname[rt], FP, offset, fpofs_name(offset));
     output_w32(0xb9400000 | imm12_rn_rd(offset >> 2, FP, rt));
   }
   else
@@ -416,7 +457,7 @@ static void emit_readdword(void *addr, u_int rt)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   if (!(offset & 7) && offset <= 32760) {
-    assem_debug("ldr %s,[x%d+%#lx]\n", regname64[rt], FP, offset);
+    assem_debug("ldr %s,[x%d+%#lx]%s\n", regname64[rt], FP, offset, fpofs_name(offset));
     output_w32(0xf9400000 | imm12_rn_rd(offset >> 3, FP, rt));
   }
   else
@@ -446,7 +487,7 @@ static void emit_loadreg(u_int r, u_int hr)
     //case HIREG: addr = &hi; break;
     //case LOREG: addr = &lo; break;
     case CCREG: addr = &cycle_count; break;
-    case CSREG: addr = &Status; break;
+    case CSREG: addr = &psxRegs.CP0.n.SR; break;
     case INVCP: addr = &invc_ptr; is64 = 1; break;
     case ROREG: addr = &ram_offset; is64 = 1; break;
     default:
@@ -465,7 +506,7 @@ static void emit_writeword(u_int rt, void *addr)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   if (!(offset & 3) && offset <= 16380) {
-    assem_debug("str %s,[x%d+%#lx]\n", regname[rt], FP, offset);
+    assem_debug("str %s,[x%d+%#lx]%s\n", regname[rt], FP, offset, fpofs_name(offset));
     output_w32(0xb9000000 | imm12_rn_rd(offset >> 2, FP, rt));
   }
   else
@@ -476,7 +517,7 @@ static void emit_writedword(u_int rt, void *addr)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   if (!(offset & 7) && offset <= 32760) {
-    assem_debug("str %s,[x%d+%#lx]\n", regname64[rt], FP, offset);
+    assem_debug("str %s,[x%d+%#lx]%s\n", regname64[rt], FP, offset, fpofs_name(offset));
     output_w32(0xf9000000 | imm12_rn_rd(offset >> 3, FP, rt));
   }
   else
@@ -547,6 +588,12 @@ static void emit_orrshr_imm(u_int rs,u_int imm,u_int rt)
   output_w32(0x2a400000 | rm_imm6_rn_rd(rs, imm, rt, rt));
 }
 
+static void emit_orn_asrimm(u_int rs1, u_int rs2, u_int shift, u_int rt)
+{
+  assem_debug("orn %s,%s,%s,asr #%u\n",regname[rt],regname[rs1],regname[rs2],shift);
+  output_w32(0x2aa00000 | rm_imm6_rn_rd(rs2, shift, rs1, rt));
+}
+
 static void emit_bicsar_imm(u_int rs,u_int imm,u_int rt)
 {
   assem_debug("bic %s,%s,%s,asr #%d\n",regname[rt],regname[rt],regname[rs],imm);
@@ -578,28 +625,43 @@ static void emit_addimm_s(u_int s, u_int is64, u_int rs, uintptr_t imm, u_int rt
     assem_debug("sub%s %s,%s,%#lx\n", st, regname[rt], regname[rs], -imm);
     output_w32(0x51000000 | is64 | s | imm12_rn_rd(-imm, rs, rt));
   }
-  else if (imm < 16777216) {
-    assem_debug("add %s,%s,#%#lx\n",regname[rt],regname[rt],imm&0xfff000);
-    output_w32(0x11400000 | is64 | imm12_rn_rd(imm >> 12, rs, rt));
-    if ((imm & 0xfff) || s) {
-      assem_debug("add%s %s,%s,#%#lx\n",st,regname[rt],regname[rs],imm&0xfff);
-      output_w32(0x11000000 | is64 | s | imm12_rn_rd(imm & 0xfff, rt, rt));
+  else if (imm < 16777216 && (!(imm & 0xfff) || !s)) {
+    assem_debug("add%s %s,%s,#%#lx\n", st, regname[rt], regname[rs], imm&0xfff000);
+    output_w32(0x11400000 | is64 | s | imm12_rn_rd(imm >> 12, rs, rt));
+    if (imm & 0xfff) {
+      assem_debug("add %s,%s,#%#lx\n", regname[rt], regname[rt], imm&0xfff);
+      output_w32(0x11000000 | is64 | imm12_rn_rd(imm & 0xfff, rt, rt));
     }
   }
-  else if (-imm < 16777216) {
-    assem_debug("sub %s,%s,#%#lx\n",regname[rt],regname[rt],-imm&0xfff000);
-    output_w32(0x51400000 | is64 | imm12_rn_rd(-imm >> 12, rs, rt));
-    if ((imm & 0xfff) || s) {
-      assem_debug("sub%s %s,%s,#%#lx\n",st,regname[rt],regname[rs],-imm&0xfff);
-      output_w32(0x51000000 | is64 | s | imm12_rn_rd(-imm & 0xfff, rt, rt));
+  else if (-imm < 16777216 && (!(-imm & 0xfff) || !s)) {
+    assem_debug("sub%s %s,%s,#%#lx\n", st, regname[rt], regname[rs], -imm&0xfff000);
+    output_w32(0x51400000 | is64 | s | imm12_rn_rd(-imm >> 12, rs, rt));
+    if (-imm & 0xfff) {
+      assem_debug("sub %s,%s,#%#lx\n", regname[rt], regname[rt], -imm&0xfff);
+      output_w32(0x51000000 | is64 | imm12_rn_rd(-imm & 0xfff, rt, rt));
     }
   }
-  else
-    abort();
+  else {
+    u_int tmp = rt;
+    assert(!is64);
+    if (rs == rt) {
+      host_tempreg_acquire();
+      tmp = HOST_TEMPREG;
+    }
+    emit_movimm(imm, tmp);
+    assem_debug("add%s %s,%s,%s\n", st, regname[rt], regname[rs], regname[tmp]);
+    output_w32(0x0b000000 | s | rm_rn_rd(rs, tmp, rt));
+    if (tmp == HOST_TEMPREG)
+      host_tempreg_release();
+  }
 }
 
 static void emit_addimm(u_int rs, uintptr_t imm, u_int rt)
 {
+  if (imm == 0) {
+    emit_mov(rs, rt);
+    return;
+  }
   emit_addimm_s(0, 0, rs, imm, rt);
 }
 
@@ -608,11 +670,21 @@ static void emit_addimm64(u_int rs, uintptr_t imm, u_int rt)
   emit_addimm_s(0, 1, rs, imm, rt);
 }
 
+static void emit_addimm_ptr(u_int rs, uintptr_t imm, u_int rt)
+{
+  emit_addimm64(rs, imm, rt);
+}
+
 static void emit_addimm_and_set_flags(int imm, u_int rt)
 {
   emit_addimm_s(1, 0, rt, imm, rt);
 }
 
+static void emit_addimm_and_set_flags3(u_int rs, int imm, u_int rt)
+{
+  emit_addimm_s(1, 0, rs, 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" };
@@ -818,6 +890,12 @@ static void emit_csinvle_reg(u_int rs1,u_int rs2,u_int rt)
   output_w32(0x5a800000 | (COND_LE << 12) | rm_rn_rd(rs2, rs1, rt));
 }
 
+static void emit_csinvne_reg(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("csinv %s,%s,%s,ne\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x5a800000 | (COND_NE << 12) | rm_rn_rd(rs2, rs1, rt));
+}
+
 static void emit_slti32(u_int rs,int imm,u_int rt)
 {
   if(rs!=rt) emit_zeroreg(rt);
@@ -946,6 +1024,13 @@ static void emit_jge(const void *a)
   output_w32(0x54000000 | (offset << 5) | COND_GE);
 }
 
+static void emit_jo(const void *a)
+{
+  assem_debug("bvs %p\n", a);
+  u_int offset = genjmpcc(a);
+  output_w32(0x54000000 | (offset << 5) | COND_VS);
+}
+
 static void emit_jno(const void *a)
 {
   assem_debug("bvc %p\n", a);
@@ -969,9 +1054,11 @@ static void emit_cb(u_int isnz, u_int is64, const void *a, u_int r)
   output_w32(0x34000000 | is64 | isnz | imm19_rt(offset, r));
 }
 
-static unused void emit_cbz(const void *a, u_int r)
+static void *emit_cbz(u_int r, const void *a)
 {
+  void *ret = out;
   emit_cb(0, 0, a, r);
+  return ret;
 }
 
 static void emit_jmpreg(u_int r)
@@ -1179,14 +1266,11 @@ static void emit_clz(u_int rs, u_int rt)
 }
 
 // special case for checking invalid_code
-static void emit_cmpmem_indexedsr12_reg(u_int rbase, u_int r, u_int imm)
+static void emit_ldrb_indexedsr12_reg(u_int rbase, u_int r, u_int rt)
 {
-  host_tempreg_acquire();
-  emit_shrimm(r, 12, HOST_TEMPREG);
-  assem_debug("ldrb %s,[%s,%s,uxtw]\n",regname[HOST_TEMPREG],regname64[rbase],regname[HOST_TEMPREG]);
-  output_w32(0x38604800 | rm_rn_rd(HOST_TEMPREG, rbase, HOST_TEMPREG));
-  emit_cmpimm(HOST_TEMPREG, imm);
-  host_tempreg_release();
+  emit_shrimm(r, 12, rt);
+  assem_debug("ldrb %s,[%s,%s,uxtw]\n",regname[rt],regname64[rbase],regname[rt]);
+  output_w32(0x38604800 | rm_rn_rd(rt, rbase, rt));
 }
 
 // special for loadlr_assemble, rs2 is destroyed
@@ -1331,16 +1415,7 @@ static void emit_movimm_from64(u_int rs_val, u_int rs, uintptr_t rt_val, u_int r
   }
   // 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));
-  }
+  emit_movimm64(rt_val, rt);
 }
 
 // trashes x2
@@ -1389,7 +1464,7 @@ static void do_readstub(int n)
   u_int reglist = stubs[n].e;
   const signed char *i_regmap = i_regs->regmap;
   int rt;
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||dops[i].itype==LOADLR) {
+  if(dops[i].itype==C2LS||dops[i].itype==LOADLR) {
     rt=get_reg(i_regmap,FTEMP);
   }else{
     rt=get_reg(i_regmap,dops[i].rt1);
@@ -1419,7 +1494,7 @@ static void do_readstub(int n)
   emit_adds64(temp2,temp2,temp2);
   handler_jump=out;
   emit_jc(0);
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
+  if(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;
@@ -1454,7 +1529,7 @@ static void do_readstub(int n)
   emit_addimm(cc<0?2:cc,(int)stubs[n].d,2);
   emit_far_call(handler);
   // (no cycle reload after read)
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
+  if(dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
     loadstore_extend(type,0,rt);
   }
   if(restore_jump)
@@ -1515,8 +1590,13 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
   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);
-    emit_addimm64(1, l1 & 0xfff, 1);
+    intptr_t offset = (l1 & ~0xfffl) - ((intptr_t)out & ~0xfffl);
+    if (-4294967296l <= offset && offset < 4294967296l) {
+      emit_adrp((void *)l1, 1);
+      emit_addimm64(1, l1 & 0xfff, 1);
+    }
+    else
+      emit_movimm64(l1, 1);
   }
   else
     emit_far_call(do_memhandler_pre);
@@ -1540,7 +1620,7 @@ static void do_writestub(int n)
   u_int reglist=stubs[n].e;
   signed char *i_regmap=i_regs->regmap;
   int rt,r;
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS) {
+  if(dops[i].itype==C2LS) {
     rt=get_reg(i_regmap,r=FTEMP);
   }else{
     rt=get_reg(i_regmap,r=dops[i].rs2);
@@ -1806,8 +1886,10 @@ static void multdiv_assemble_arm64(int i, const struct regstat *i_regs)
 
         // div 0 quotient (remainder is already correct)
         host_tempreg_acquire();
-        if (dops[i].opcode2 == 0x1A) // DIV
-          emit_sub_asrimm(0,numerator,31,HOST_TEMPREG);
+        if (dops[i].opcode2 == 0x1A) { // DIV
+          emit_add_lsrimm(WZR,numerator,31,HOST_TEMPREG);
+          emit_orn_asrimm(HOST_TEMPREG,numerator,31,HOST_TEMPREG);
+        }
         else
           emit_movimm(~0,HOST_TEMPREG);
         emit_test(denominator,denominator);
@@ -1831,8 +1913,10 @@ static void multdiv_assemble_arm64(int i, const struct regstat *i_regs)
         if (hr >= 0)
           emit_mov(numerator,hr);
         if (lr >= 0) {
-          if (dops[i].opcode2 == 0x1A) // DIV
-            emit_sub_asrimm(0,numerator,31,lr);
+          if (dops[i].opcode2 == 0x1A) { // DIV
+            emit_add_lsrimm(WZR,numerator,31,lr);
+            emit_orn_asrimm(lr,numerator,31,lr);
+          }
           else
             emit_movimm(~0,lr);
         }
@@ -1842,6 +1926,17 @@ static void multdiv_assemble_arm64(int i, const struct regstat *i_regs)
         if (lr >= 0) emit_movimm(~0,lr);
       }
     }
+    else if ((dops[i].opcode2==0x1A || dops[i].opcode2==0x1B) && dops[i].rs1==0)
+    {
+      signed char denominator = get_reg(i_regs->regmap, dops[i].rs2);
+      assert(denominator >= 0);
+      if (hr >= 0) emit_zeroreg(hr);
+      if (lr >= 0) {
+        emit_zeroreg(lr);
+        emit_test(denominator, denominator);
+        emit_csinvne_reg(lr, lr, lr);
+      }
+    }
     else
     {
       // Multiply by zero is zero.
@@ -1900,7 +1995,7 @@ static void do_miniht_insert(u_int return_address,u_int rt,int temp) {
   emit_writeword(rt,&mini_ht[(return_address&0xFF)>>3][0]);
 }
 
-static void clear_cache_arm64(char *start, char *end)
+static unused void clear_cache_arm64(char *start, char *end)
 {
   // Don't rely on GCC's __clear_cache implementation, as it caches
   // icache/dcache cache line sizes, that can vary between cores on
@@ -1945,13 +2040,13 @@ static void clear_cache_arm64(char *start, char *end)
 static void arch_init(void)
 {
   uintptr_t diff = (u_char *)&ndrc->tramp.f - (u_char *)&ndrc->tramp.ops;
-  struct tramp_insns *ops = ndrc->tramp.ops, *opsw;
+  struct tramp_insns *ops = NDRC_WRITE_OFFSET(ndrc->tramp.ops);
   size_t i;
   assert(!(diff & 3));
-  opsw = start_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));
+  start_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));
   for (i = 0; i < ARRAY_SIZE(ndrc->tramp.ops); i++) {
-    opsw[i].ldr = 0x58000000 | imm19_rt(diff >> 2, 17); // ldr x17, [=val]
-    opsw[i].br  = 0xd61f0000 | rm_rn_rd(0, 17, 0);      // br x17
+    ops[i].ldr = 0x58000000 | imm19_rt(diff >> 2, 17); // ldr x17, [=val]
+    ops[i].br  = 0xd61f0000 | rm_rn_rd(0, 17, 0);      // br x17
   }
   end_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));
 }