optimizations, bugfixes, uae works (but with timing glitches?)

[picodrive.git] / cpu / Cyclone / OpMove.cpp

\r
#include "app.h"\r
\r
\r
// Pack our flags into r1, in SR/CCR register format\r
// trashes r0,r2\r
void OpFlagsToReg(int high)\r
{\r
  ot("  ldr r0,[r7,#0x4c]   ;@ X bit\n");\r
  ot("  mov r1,r9,lsr #28   ;@ ____NZCV\n");\r
  ot("  eor r2,r1,r1,ror #1 ;@ Bit 0=C^V\n");\r
  ot("  tst r2,#1           ;@ 1 if C!=V\n");\r
  ot("  eorne r1,r1,#3      ;@ ____NZVC\n");\r
  ot("\n");\r
  if (high) ot("  ldrb r2,[r7,#0x44]  ;@ Include SR high\n");\r
  ot("  and r0,r0,#0x20000000\n");\r
  ot("  orr r1,r1,r0,lsr #25 ;@ ___XNZVC\n");\r
  if (high) ot("  orr r1,r1,r2,lsl #8\n");\r
  ot("\n");\r
}\r
\r
// Convert SR/CRR register in r0 to our flags\r
// trashes r0,r1\r
void OpRegToFlags(int high)\r
{\r
  ot("  eor r1,r0,r0,ror #1 ;@ Bit 0=C^V\n");\r
  ot("  mov r2,r0,lsl #25\n");\r
  ot("  tst r1,#1           ;@ 1 if C!=V\n");\r
  ot("  eorne r0,r0,#3      ;@ ___XNZCV\n");\r
  ot("  str r2,[r7,#0x4c]   ;@ Store X bit\n");\r
  ot("  mov r9,r0,lsl #28   ;@ r9=NZCV...\n");\r
\r
  if (high)\r
  {\r
    ot("  mov r0,r0,ror #8\n");\r
    ot("  and r0,r0,#0xa7 ;@ only take defined bits\n");\r
    ot("  strb r0,[r7,#0x44] ;@ Store SR high\n");\r
  }\r
  ot("\n");\r
}\r
\r
void SuperEnd(void)\r
{\r
  ot(";@ ----------\n");\r
  ot(";@ tried execute privileged instruction in user mode\n");\r
  ot("WrongPrivilegeMode%s\n",ms?"":":");\r
  ot("  sub r4,r4,#2 ;@ last opcode wasn't executed - go back\n");\r
  ot("  mov r0,#0x20 ;@ privilege violation\n");\r
  ot("  bl Exception\n");\r
  Cycles=34;\r
  OpEnd(0);\r
}\r
\r
// does OSP and A7 swapping if needed\r
// new or old SR (not the one already in [r7,#0x44]) should be passed in r11\r
// trashes r0,r11\r
void SuperChange(int op,int load_srh)\r
{\r
  ot(";@ A7 <-> OSP?\n");\r
  if (load_srh)\r
    ot("  ldr r0,[r7,#0x44] ;@ Get other SR high\n");\r
  ot("  eor r0,r0,r11\n");\r
  ot("  tst r0,#0x20\n");\r
  ot("  beq no_sp_swap%.4x\n",op);\r
  ot(" ;@ swap OSP and A7:\n");\r
  ot("  ldr r11,[r7,#0x3C] ;@ Get A7\n");\r
  ot("  ldr r0, [r7,#0x48] ;@ Get OSP\n");\r
  ot("  str r11,[r7,#0x48]\n");\r
  ot("  str r0, [r7,#0x3C]\n");\r
  ot("no_sp_swap%.4x%s\n", op, ms?"":":");\r
}\r
\r
\r
\r
// --------------------- Opcodes 0x1000+ ---------------------\r
// Emit a Move opcode, 00xxdddd ddssssss\r
int OpMove(int op)\r
{\r
  int sea=0,tea=0;\r
  int size=0,use=0;\r
  int movea=0;\r
\r
  // Get source and target EA\r
  sea = op&0x003f;\r
  tea =(op&0x01c0)>>3;\r
  tea|=(op&0x0e00)>>9;\r
\r
  if (tea>=8 && tea<0x10) movea=1;\r
\r
  // Find size extension\r
  switch (op&0x3000)\r
  {\r
    default: return 1;\r
    case 0x1000: size=0; break;\r
    case 0x3000: size=1; break;\r
    case 0x2000: size=2; break;\r
  }\r
\r
  if (size<1 && (movea || EaAn(sea))) return 1; // move.b An,* and movea.b * are invalid\r
\r
  // See if we can do this opcode:\r
  if (EaCanRead (sea,size)==0) return 1;\r
  if (EaCanWrite(tea     )==0) return 1;\r
\r
  use=OpBase(op,size);\r
  if (tea<0x38) use&=~0x0e00; // Use same handler for register ?0-7\r
  \r
  if (tea==0x1f || tea==0x27) use|=0x0e00; // Specific handler for (a7)+ and -(a7)\r
\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  OpStart(op,sea,tea); Cycles=4;\r
\r
  EaCalcRead(-1,1,sea,size,0x003f);\r
\r
  if (movea==0)\r
  {\r
    ot("  adds r1,r1,#0 ;@ Defines NZ, clears CV\n");\r
    ot("  mrs r9,cpsr ;@ r9=NZCV flags\n");\r
    ot("\n");\r
  }\r
\r
  if (movea) size=2; // movea always expands to 32-bits\r
\r
#if SPLIT_MOVEL_PD\r
  if ((tea&0x38)==0x20 && size==2) { // -(An)\r
    EaCalc (10,0x0e00,tea,size,0,0);\r
    ot("  mov r11,r1\n");\r
    ot("  add r0,r10,#2\n");\r
    EaWrite(0,     1,tea,1,0x0e00,0,0);\r
    EaWrite(10,   11,tea,1,0x0e00,1);\r
  }\r
  else\r
#endif\r
  {\r
    EaCalc (0,0x0e00,tea,size,0,0);\r
    EaWrite(0,     1,tea,size,0x0e00,0,0);\r
  }\r
\r
#if CYCLONE_FOR_GENESIS && !MEMHANDLERS_CHANGE_CYCLES\r
  // this is a bit hacky\r
  if ((tea==0x39||(tea&0x38)==0x10)&&size>=1)\r
    ot("  ldr r5,[r7,#0x5c] ;@ Load Cycles\n");\r
#endif\r
\r
  if((tea&0x38)==0x20) Cycles-=2; // less cycles when dest is -(An)\r
\r
  OpEnd(sea,tea);\r
  return 0;\r
}\r
\r
// --------------------- Opcodes 0x41c0+ ---------------------\r
// Emit an Lea opcode, 0100nnn1 11aaaaaa\r
int OpLea(int op)\r
{\r
  int use=0;\r
  int sea=0,tea=0;\r
\r
  sea= op&0x003f;\r
  tea=(op&0x0e00)>>9; tea|=8;\r
\r
  if (EaCanRead(sea,-1)==0) return 1; // See if we can do this opcode\r
\r
  use=OpBase(op,0);\r
  use&=~0x0e00; // Also use 1 handler for target ?0-7\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  OpStart(op,sea,tea);\r
\r
  EaCalc (1,0x003f,sea,0); // Lea\r
  EaCalc (0,0x0e00,tea,2);\r
  EaWrite(0,     1,tea,2,0x0e00);\r
\r
  Cycles=Ea_add_ns(g_lea_cycle_table,sea);\r
\r
  OpEnd(sea,tea);\r
\r
  return 0;\r
}\r
\r
// --------------------- Opcodes 0x40c0+ ---------------------\r
// Move SR opcode, 01000tt0 11aaaaaa move SR\r
int OpMoveSr(int op)\r
{\r
  int type=0,ea=0;\r
  int use=0,size=1;\r
\r
  type=(op>>9)&3; // from SR, from CCR, to CCR, to SR\r
  ea=op&0x3f;\r
\r
  if(EaAn(ea)) return 1; // can't use An regs\r
\r
  switch(type)\r
  {\r
    case 0:\r
      if (EaCanWrite(ea)==0) return 1; // See if we can do this opcode:\r
      break;\r
\r
    case 1:\r
      return 1; // no such op in 68000\r
\r
    case 2: case 3:\r
      if (EaCanRead(ea,size)==0) return 1; // See if we can do this opcode:\r
      break;\r
  }\r
\r
  use=OpBase(op,size);\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  // 68000 model allows reading whole SR in user mode (but newer models don't)\r
  OpStart(op,ea,0,0,type==3);\r
  Cycles=12;\r
  if (type==0) Cycles=(ea>=8)?8:6;\r
\r
  if (type==0 || type==1)\r
  {\r
    OpFlagsToReg(type==0);\r
    EaCalc (0,0x003f,ea,size,0,0);\r
    EaWrite(0,     1,ea,size,0x003f,0,0);\r
  }\r
\r
  if (type==2 || type==3)\r
  {\r
    EaCalcReadNoSE(-1,0,ea,size,0x003f);\r
    OpRegToFlags(type==3);\r
    if (type==3) {\r
      SuperChange(op,0);\r
    }\r
  }\r
\r
  OpEnd(ea,0,0,type==3);\r
\r
  return 0;\r
}\r
\r
\r
// Ori/Andi/Eori $nnnn,sr 0000t0t0 01111100\r
int OpArithSr(int op)\r
{\r
  int type=0,ea=0;\r
  int use=0,size=0;\r
\r
  type=(op>>9)&5; if (type==4) return 1;\r
  size=(op>>6)&1; // ccr or sr?\r
  ea=0x3c;\r
\r
  use=OpBase(op,size);\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  OpStart(op,ea,0,0,size!=0); Cycles=16;\r
\r
  EaCalc(10,0x003f,ea,size);\r
  EaRead(10,    10,ea,size,0x003f);\r
\r
  OpFlagsToReg(size);\r
  if (type==0) ot("  orr r0,r1,r10\n");\r
  if (type==1) ot("  and r0,r1,r10\n");\r
  if (type==5) ot("  eor r0,r1,r10\n");\r
  OpRegToFlags(size);\r
  if (size && type!=0) { // we can't enter supervisor mode, nor unmask irqs just by using OR\r
    SuperChange(op,0);\r
  }\r
\r
  OpEnd(ea,0,0,size!=0 && type!=0);\r
\r
  return 0;\r
}\r
\r
// --------------------- Opcodes 0x4850+ ---------------------\r
// Emit an Pea opcode, 01001000 01aaaaaa\r
int OpPea(int op)\r
{\r
  int use=0;\r
  int ea=0;\r
\r
  ea=op&0x003f; if (ea<0x10) return 1; // Swap opcode\r
  if (EaCanRead(ea,-1)==0) return 1; // See if we can do this opcode:\r
\r
  use=OpBase(op,0);\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  OpStart(op,ea);\r
\r
  ot("  ldr r10,[r7,#0x3c]\n");\r
  EaCalc (1,0x003f, ea,0);\r
  ot("\n");\r
  ot("  sub r0,r10,#4 ;@ Predecrement A7\n");\r
  ot("  str r0,[r7,#0x3c] ;@ Save A7\n");\r
  ot("\n");\r
  MemHandler(1,2); // Write 32-bit\r
  ot("\n");\r
\r
  Cycles=6+Ea_add_ns(g_pea_cycle_table,ea);\r
\r
  OpEnd(ea);\r
\r
  return 0;\r
}\r
\r
// --------------------- Opcodes 0x4880+ ---------------------\r
// Emit a Movem opcode, 01001d00 1xeeeeee regmask\r
int OpMovem(int op)\r
{\r
  int size=0,ea=0,cea=0,dir=0;\r
  int use=0,decr=0,change=0;\r
\r
  size=((op>>6)&1)+1; // word, long\r
  ea=op&0x003f;\r
  dir=(op>>10)&1; // Direction (1==ea2reg)\r
\r
  if (dir) {\r
    if (ea<0x10 || ea>0x3b || (ea&0x38)==0x20) return 1; // Invalid EA\r
  } else {\r
    if (ea<0x10 || ea>0x39 || (ea&0x38)==0x18) return 1;\r
  }\r
\r
  if ((ea&0x38)==0x18 || (ea&0x38)==0x20) change=1;\r
  if ((ea&0x38)==0x20) decr=1; // -(An), bitfield is decr\r
\r
  cea=ea; if (change) cea=0x10;\r
\r
  use=OpBase(op,size);\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  OpStart(op,ea,0,1);\r
\r
  ot("  ldrh r11,[r4],#2 ;@ r11=register mask\n");\r
\r
  ot(";@ r10=Register Index*4:\n");\r
  if (decr) ot("  mov r10,#0x40 ;@ order reversed for -(An)\n");\r
  else      ot("  mov r10,#-4\n");\r
  \r
  ot("\n");\r
  ot(";@ Get the address into r6:\n");\r
  EaCalc(6,0x003f,cea,size);\r
\r
  ot("\n");\r
  ot("  tst r11,r11\n");        // sanity check\r
  ot("  beq NoRegs%.4x\n",op);\r
\r
  ot("\n");\r
  ot("Movemloop%.4x%s\n",op, ms?"":":");\r
  ot("  add r10,r10,#%d ;@ r10=Next Register\n",decr?-4:4);\r
  ot("  movs r11,r11,lsr #1\n");\r
  ot("  bcc Movemloop%.4x\n",op);\r
  ot("\n");\r
\r
  if (decr) ot("  sub r6,r6,#%d ;@ Pre-decrement address\n",1<<size);\r
\r
  if (dir)\r
  {\r
    ot("  ;@ Copy memory to register:\n",1<<size);\r
    EaRead (6,0,ea,size,0x003f);\r
    ot("  str r0,[r7,r10] ;@ Save value into Dn/An\n");\r
  }\r
  else\r
  {\r
    // if (size == 2 && decr && SPLIT_MOVEL_PD) we should do 2xWrite16 here\r
    // (same as in movel.l ?, -(An)), but as this is not likely to be needed and\r
    // we do not want the performance hit, we do single Write32 instead.\r
    ot("  ;@ Copy register to memory:\n",1<<size);\r
    ot("  ldr r1,[r7,r10] ;@ Load value from Dn/An\n");\r
    EaWrite(6,1,ea,size,0x003f);\r
  }\r
\r
  if (decr==0) ot("  add r6,r6,#%d ;@ Post-increment address\n",1<<size);\r
\r
  ot("  sub r5,r5,#%d ;@ Take some cycles\n",2<<size);\r
  ot("  tst r11,r11\n");\r
  ot("  bne Movemloop%.4x\n",op);\r
  ot("\n");\r
\r
  if (change)\r
  {\r
    ot(";@ Write back address:\n");\r
    EaCalc (0,0x0007,8|(ea&7),2);\r
    EaWrite(0,     6,8|(ea&7),2,0x0007);\r
  }\r
\r
  ot("NoRegs%.4x%s\n",op, ms?"":":");\r
  ot("  ldr r6,=CycloneJumpTab ;@ restore Opcode Jump table\n");\r
  ot("\n");\r
\r
  if(dir) { // er\r
         if (ea==0x3a) Cycles=16; // ($nn,PC)\r
    else if (ea==0x3b) Cycles=18; // ($nn,pc,Rn)\r
    else Cycles=12;\r
  } else {\r
    Cycles=8;\r
  }\r
\r
  Cycles+=Ea_add_ns(g_movem_cycle_table,ea);\r
\r
  OpEnd(ea,0,1);\r
  ltorg();\r
  ot("\n");\r
\r
  return 0;\r
}\r
\r
// --------------------- Opcodes 0x4e60+ ---------------------\r
// Emit a Move USP opcode, 01001110 0110dnnn move An to/from USP\r
int OpMoveUsp(int op)\r
{\r
  int use=0,dir=0;\r
\r
  dir=(op>>3)&1; // Direction\r
  use=op&~0x0007; // Use same opcode for all An\r
\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  OpStart(op,0,0,0,1); Cycles=4;\r
\r
  if (dir)\r
  {\r
    ot("  ldr r1,[r7,#0x48] ;@ Get from USP\n\n");\r
    EaCalc (0,0x000f,8,2,1);\r
    EaWrite(0,     1,8,2,0x000f,1);\r
  }\r
  else\r
  {\r
    EaCalc (0,0x000f,8,2,1);\r
    EaRead (0,     0,8,2,0x000f,1);\r
    ot("  str r0,[r7,#0x48] ;@ Put in USP\n\n");\r
  }\r
    \r
  OpEnd();\r
\r
  return 0;\r
}\r
\r
// --------------------- Opcodes 0x7000+ ---------------------\r
// Emit a Move Quick opcode, 0111nnn0 dddddddd  moveq #dd,Dn\r
int OpMoveq(int op)\r
{\r
  int use=0;\r
\r
  use=op&0xf100; // Use same opcode for all values\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  OpStart(op); Cycles=4;\r
\r
  ot("  movs r0,r8,asl #24\n");\r
  ot("  and r1,r8,#0x0e00\n");\r
  ot("  mov r0,r0,asr #24 ;@ Sign extended Quick value\n");\r
  ot("  mrs r9,cpsr ;@ r9=NZ flags\n");\r
  ot("  str r0,[r7,r1,lsr #7] ;@ Store into Dn\n");\r
  ot("\n");\r
\r
  OpEnd();\r
\r
  return 0;\r
}\r
\r
// --------------------- Opcodes 0xc140+ ---------------------\r
// Emit a Exchange opcode:\r
// 1100ttt1 01000sss  exg ds,dt\r
// 1100ttt1 01001sss  exg as,at\r
// 1100ttt1 10001sss  exg as,dt\r
int OpExg(int op)\r
{\r
  int use=0,type=0;\r
\r
  type=op&0xf8;\r
\r
  if (type!=0x40 && type!=0x48 && type!=0x88) return 1; // Not an exg opcode\r
\r
  use=op&0xf1f8; // Use same opcode for all values\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler\r
\r
  OpStart(op); Cycles=6;\r
\r
  ot("  and r10,r8,#0x0e00 ;@ Find T register\n");\r
  ot("  and r11,r8,#0x000f ;@ Find S register\n");\r
  if (type==0x48) ot("  orr r10,r10,#0x1000 ;@ T is an address register\n");\r
  ot("\n");\r
  ot("  ldr r0,[r7,r10,lsr #7] ;@ Get T\n");\r
  ot("  ldr r1,[r7,r11,lsl #2] ;@ Get S\n");\r
  ot("\n");\r
  ot("  str r0,[r7,r11,lsl #2] ;@ T->S\n");\r
  ot("  str r1,[r7,r10,lsr #7] ;@ S->T\n");  \r
  ot("\n");\r
\r
  OpEnd();\r
  \r
  return 0;\r
}\r
\r
// ------------------------- movep -------------------------------\r
// 0000ddd1 0z001sss\r
// 0000sss1 1z001ddd (to mem)\r
int OpMovep(int op)\r
{\r
  int ea=0,rea=0;\r
  int size=1,use=0,dir,aadd=0;\r
\r
  use=op&0xf1f8;\r
  if (op!=use) { OpUse(op,use); return 0; } // Use existing handler (for all dests, srcs)\r
\r
  // Get EA\r
  ea = (op&0x0007)|0x28;\r
  rea= (op&0x0e00)>>9;\r
  dir = (op>>7)&1;\r
\r
  // Find size extension\r
  if(op&0x0040) size=2;\r
\r
  OpStart(op,ea);\r
  \r
  if(dir) { // reg to mem\r
    EaCalcReadNoSE(-1,11,rea,size,0x0e00);\r
\r
    EaCalc(10,0x000f,ea,size);\r
    if(size==2) { // if operand is long\r
      ot("  mov r1,r11,lsr #24 ;@ first byte\n");\r
      EaWrite(10,1,ea,0,0x000f); // store first byte\r
      ot("  add r0,r10,#%i\n",(aadd+=2));\r
      ot("  mov r1,r11,lsr #16 ;@ second byte\n");\r
      EaWrite(0,1,ea,0,0x000f); // store second byte\r
      ot("  add r0,r10,#%i\n",(aadd+=2));\r
    } else {\r
      ot("  mov r0,r10\n");\r
    }\r
    ot("  mov r1,r11,lsr #8 ;@ first or third byte\n");\r
    EaWrite(0,1,ea,0,0x000f);\r
    ot("  add r0,r10,#%i\n",(aadd+=2));\r
    ot("  and r1,r11,#0xff\n");\r
    EaWrite(0,1,ea,0,0x000f);\r
  } else { // mem to reg\r
    EaCalc(10,0x000f,ea,size,1);\r
    EaRead(10,11,ea,0,0x000f,1); // read first byte\r
    ot("  add r0,r10,#2\n");\r
    EaRead(0,1,ea,0,0x000f,1); // read second byte\r
    if(size==2) { // if operand is long\r
      ot("  orr r11,r11,r1,lsr #8 ;@ second byte\n");\r
      ot("  add r0,r10,#4\n");\r
      EaRead(0,1,ea,0,0x000f,1);\r
      ot("  orr r11,r11,r1,lsr #16 ;@ third byte\n");\r
      ot("  add r0,r10,#6\n");\r
      EaRead(0,1,ea,0,0x000f,1);\r
      ot("  orr r1,r11,r1,lsr #24 ;@ fourth byte\n");\r
    } else {\r
      ot("  orr r1,r11,r1,lsr #8 ;@ second byte\n");\r
    }\r
    // store the result\r
    EaCalc(11,0x0e00,rea,size,1);      // reg number -> r11\r
    EaWrite(11,1,rea,size,0x0e00,1);\r
  }\r
\r
  Cycles=(size==2)?24:16;\r
  OpEnd(ea);\r
\r
  return 0;\r
}\r
\r
// Emit a Stop/Reset opcodes, 01001110 011100t0 imm\r
int OpStopReset(int op)\r
{\r
  int type=(op>>1)&1; // stop/reset\r
\r
  OpStart(op,0,0,0,1);\r
\r
  if(type) {\r
    // copy immediate to SR, stop the CPU and eat all remaining cycles.\r
    ot("  ldrh r0,[r4],#2 ;@ Fetch the immediate\n");\r
    OpRegToFlags(1);\r
    SuperChange(op,0);\r
\r
    ot("\n");\r
\r
    ot("  mov r0,#1\n");\r
    ot("  str r0,[r7,#0x58] ;@ stopped\n");\r
    ot("\n");\r
\r
    ot("  mov r5,#0 ;@ eat cycles\n");\r
    Cycles = 4;\r
    ot("\n");\r
  }\r
  else\r
  {\r
    Cycles = 132;\r
#if USE_RESET_CALLBACK\r
    ot("  str r4,[r7,#0x40] ;@ Save PC\n");\r
    ot("  mov r1,r9,lsr #28\n");\r
    ot("  strb r1,[r7,#0x46] ;@ Save Flags (NZCV)\n");\r
    ot("  str r5,[r7,#0x5c] ;@ Save Cycles\n");\r
    ot("  ldr r11,[r7,#0x90] ;@ ResetCallback\n");\r
    ot("  tst r11,r11\n");\r
    ot("  movne lr,pc\n");\r
    ot("  bxne r11 ;@ call ResetCallback if it is defined\n");\r
    ot("  ldrb r9,[r7,#0x46] ;@ r9 = Load Flags (NZCV)\n");\r
    ot("  ldr r5,[r7,#0x5c] ;@ Load Cycles\n");\r
    ot("  ldr r4,[r7,#0x40] ;@ Load PC\n");\r
    ot("  mov r9,r9,lsl #28\n");\r
    ot("\n");\r
#endif\r
  }\r
\r
  OpEnd();\r
\r
  return 0;\r
}\r
\r