toir.c

The Valgrind distribution has multiple tools. The most popular is the memory checking tool (called M

      case 7:  return OFFB_XMM7;
      case 8:  return OFFB_XMM8;
      case 9:  return OFFB_XMM9;
      case 10: return OFFB_XMM10;
      case 11: return OFFB_XMM11;
      case 12: return OFFB_XMM12;
      case 13: return OFFB_XMM13;
      case 14: return OFFB_XMM14;
      case 15: return OFFB_XMM15;
      default: vpanic("xmmGuestRegOffset(amd64)");
   }
}

/* Lanes of vector registers are always numbered from zero being the
   least significant lane (rightmost in the register).  */

static Int xmmGuestRegLane16offset ( UInt xmmreg, Int laneno )
{
   /* Correct for little-endian host only. */
   vassert(!host_is_bigendian);
   vassert(laneno >= 0 && laneno < 8);
   return xmmGuestRegOffset( xmmreg ) + 2 * laneno;
}

static Int xmmGuestRegLane32offset ( UInt xmmreg, Int laneno )
{
   /* Correct for little-endian host only. */
   vassert(!host_is_bigendian);
   vassert(laneno >= 0 && laneno < 4);
   return xmmGuestRegOffset( xmmreg ) + 4 * laneno;
}

static Int xmmGuestRegLane64offset ( UInt xmmreg, Int laneno )
{
   /* Correct for little-endian host only. */
   vassert(!host_is_bigendian);
   vassert(laneno >= 0 && laneno < 2);
   return xmmGuestRegOffset( xmmreg ) + 8 * laneno;
}

//.. static IRExpr* getSReg ( UInt sreg )
//.. {
//..    return IRExpr_Get( segmentGuestRegOffset(sreg), Ity_I16 );
//.. }
//.. 
//.. static void putSReg ( UInt sreg, IRExpr* e )
//.. {
//..    vassert(typeOfIRExpr(irbb->tyenv,e) == Ity_I16);
//..    stmt( IRStmt_Put( segmentGuestRegOffset(sreg), e ) );
//.. }

static IRExpr* getXMMReg ( UInt xmmreg )
{
   return IRExpr_Get( xmmGuestRegOffset(xmmreg), Ity_V128 );
}

static IRExpr* getXMMRegLane64 ( UInt xmmreg, Int laneno )
{
   return IRExpr_Get( xmmGuestRegLane64offset(xmmreg,laneno), Ity_I64 );
}

static IRExpr* getXMMRegLane64F ( UInt xmmreg, Int laneno )
{
   return IRExpr_Get( xmmGuestRegLane64offset(xmmreg,laneno), Ity_F64 );
}

static IRExpr* getXMMRegLane32 ( UInt xmmreg, Int laneno )
{
   return IRExpr_Get( xmmGuestRegLane32offset(xmmreg,laneno), Ity_I32 );
}

static IRExpr* getXMMRegLane32F ( UInt xmmreg, Int laneno )
{
   return IRExpr_Get( xmmGuestRegLane32offset(xmmreg,laneno), Ity_F32 );
}

static void putXMMReg ( UInt xmmreg, IRExpr* e )
{
   vassert(typeOfIRExpr(irbb->tyenv,e) == Ity_V128);
   stmt( IRStmt_Put( xmmGuestRegOffset(xmmreg), e ) );
}

static void putXMMRegLane64 ( UInt xmmreg, Int laneno, IRExpr* e )
{
   vassert(typeOfIRExpr(irbb->tyenv,e) == Ity_I64);
   stmt( IRStmt_Put( xmmGuestRegLane64offset(xmmreg,laneno), e ) );
}

static void putXMMRegLane64F ( UInt xmmreg, Int laneno, IRExpr* e )
{
   vassert(typeOfIRExpr(irbb->tyenv,e) == Ity_F64);
   stmt( IRStmt_Put( xmmGuestRegLane64offset(xmmreg,laneno), e ) );
}

static void putXMMRegLane32F ( UInt xmmreg, Int laneno, IRExpr* e )
{
   vassert(typeOfIRExpr(irbb->tyenv,e) == Ity_F32);
   stmt( IRStmt_Put( xmmGuestRegLane32offset(xmmreg,laneno), e ) );
}

static void putXMMRegLane32 ( UInt xmmreg, Int laneno, IRExpr* e )
{
   vassert(typeOfIRExpr(irbb->tyenv,e) == Ity_I32);
   stmt( IRStmt_Put( xmmGuestRegLane32offset(xmmreg,laneno), e ) );
}

static void putXMMRegLane16 ( UInt xmmreg, Int laneno, IRExpr* e )
{
   vassert(typeOfIRExpr(irbb->tyenv,e) == Ity_I16);
   stmt( IRStmt_Put( xmmGuestRegLane16offset(xmmreg,laneno), e ) );
}

static IRExpr* mkV128 ( UShort mask )
{
   return IRExpr_Const(IRConst_V128(mask));
}

static IRExpr* mkAnd1 ( IRExpr* x, IRExpr* y )
{
   vassert(typeOfIRExpr(irbb->tyenv,x) == Ity_I1);
   vassert(typeOfIRExpr(irbb->tyenv,y) == Ity_I1);
   return unop(Iop_64to1, 
               binop(Iop_And64, 
                     unop(Iop_1Uto64,x), 
                     unop(Iop_1Uto64,y)));
}


/*------------------------------------------------------------*/
/*--- Helpers for %rflags.                                 ---*/
/*------------------------------------------------------------*/

/* -------------- Evaluating the flags-thunk. -------------- */

/* Build IR to calculate all the eflags from stored
   CC_OP/CC_DEP1/CC_DEP2/CC_NDEP.  Returns an expression ::
   Ity_I64. */
static IRExpr* mk_amd64g_calculate_rflags_all ( void )
{
   IRExpr** args
      = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP,   Ity_I64),
                       IRExpr_Get(OFFB_CC_DEP1, Ity_I64),
                       IRExpr_Get(OFFB_CC_DEP2, Ity_I64),
                       IRExpr_Get(OFFB_CC_NDEP, Ity_I64) );
   IRExpr* call
      = mkIRExprCCall(
           Ity_I64,
           0/*regparm*/, 
           "amd64g_calculate_rflags_all", &amd64g_calculate_rflags_all,
           args
        );
   /* Exclude OP and NDEP from definedness checking.  We're only
      interested in DEP1 and DEP2. */
   call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3);
   return call;
}

/* Build IR to calculate some particular condition from stored
   CC_OP/CC_DEP1/CC_DEP2/CC_NDEP.  Returns an expression ::
   Ity_Bit. */
static IRExpr* mk_amd64g_calculate_condition ( AMD64Condcode cond )
{
   IRExpr** args
      = mkIRExprVec_5( mkU64(cond),
                       IRExpr_Get(OFFB_CC_OP,   Ity_I64),
                       IRExpr_Get(OFFB_CC_DEP1, Ity_I64),
                       IRExpr_Get(OFFB_CC_DEP2, Ity_I64),
                       IRExpr_Get(OFFB_CC_NDEP, Ity_I64) );
   IRExpr* call
      = mkIRExprCCall(
           Ity_I64,
           0/*regparm*/, 
           "amd64g_calculate_condition", &amd64g_calculate_condition,
           args
        );
   /* Exclude the requested condition, OP and NDEP from definedness
      checking.  We're only interested in DEP1 and DEP2. */
   call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<1) | (1<<4);
   return unop(Iop_64to1, call);
}

/* Build IR to calculate just the carry flag from stored
   CC_OP/CC_DEP1/CC_DEP2/CC_NDEP.  Returns an expression :: Ity_I64. */
static IRExpr* mk_amd64g_calculate_rflags_c ( void )
{
   IRExpr** args
      = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP,   Ity_I64),
                       IRExpr_Get(OFFB_CC_DEP1, Ity_I64),
                       IRExpr_Get(OFFB_CC_DEP2, Ity_I64),
                       IRExpr_Get(OFFB_CC_NDEP, Ity_I64) );
   IRExpr* call
      = mkIRExprCCall(
           Ity_I64,
           0/*regparm*/, 
           "amd64g_calculate_rflags_c", &amd64g_calculate_rflags_c,
           args
        );
   /* Exclude OP and NDEP from definedness checking.  We're only
      interested in DEP1 and DEP2. */
   call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3);
   return call;
}


/* -------------- Building the flags-thunk. -------------- */

/* The machinery in this section builds the flag-thunk following a
   flag-setting operation.  Hence the various setFlags_* functions.
*/

static Bool isAddSub ( IROp op8 )
{
   return toBool(op8 == Iop_Add8 || op8 == Iop_Sub8);
}

static Bool isLogic ( IROp op8 )
{
   return toBool(op8 == Iop_And8 || op8 == Iop_Or8 || op8 == Iop_Xor8);
}

/* U-widen 8/16/32/64 bit int expr to 64. */
static IRExpr* widenUto64 ( IRExpr* e )
{
   switch (typeOfIRExpr(irbb->tyenv,e)) {
      case Ity_I64: return e;
      case Ity_I32: return unop(Iop_32Uto64, e);
      case Ity_I16: return unop(Iop_16Uto64, e);
      case Ity_I8:  return unop(Iop_8Uto64, e);
      default: vpanic("widenUto64");
   }
}

/* S-widen 8/16/32/64 bit int expr to 32. */
static IRExpr* widenSto64 ( IRExpr* e )
{
   switch (typeOfIRExpr(irbb->tyenv,e)) {
      case Ity_I64: return e;
      case Ity_I32: return unop(Iop_32Sto64, e);
      case Ity_I16: return unop(Iop_16Sto64, e);
      case Ity_I8:  return unop(Iop_8Sto64, e);
      default: vpanic("widenSto64");
   }
}

/* Narrow 8/16/32/64 bit int expr to 8/16/32/64.  Clearly only some
   of these combinations make sense. */
static IRExpr* narrowTo ( IRType dst_ty, IRExpr* e )
{
   IRType src_ty = typeOfIRExpr(irbb->tyenv,e);
   if (src_ty == dst_ty)
      return e;
   if (src_ty == Ity_I32 && dst_ty == Ity_I16)
      return unop(Iop_32to16, e);
   if (src_ty == Ity_I32 && dst_ty == Ity_I8)
      return unop(Iop_32to8, e);
   if (src_ty == Ity_I64 && dst_ty == Ity_I32)
      return unop(Iop_64to32, e);
   if (src_ty == Ity_I64 && dst_ty == Ity_I16)
      return unop(Iop_64to16, e);
   if (src_ty == Ity_I64 && dst_ty == Ity_I8)
      return unop(Iop_64to8, e);

   vex_printf("\nsrc, dst tys are: ");
   ppIRType(src_ty);
   vex_printf(", ");
   ppIRType(dst_ty);
   vex_printf("\n");
   vpanic("narrowTo(amd64)");
}


/* Set the flags thunk OP, DEP1 and DEP2 fields.  The supplied op is
   auto-sized up to the real op. */

static 
void setFlags_DEP1_DEP2 ( IROp op8, IRTemp dep1, IRTemp dep2, IRType ty )
{
   Int ccOp = 0;
   switch (ty) {
      case Ity_I8:  ccOp = 0; break;
      case Ity_I16: ccOp = 1; break;
      case Ity_I32: ccOp = 2; break;
      case Ity_I64: ccOp = 3; break;
      default: vassert(0);
   }
   switch (op8) {
      case Iop_Add8: ccOp += AMD64G_CC_OP_ADDB;   break;
      case Iop_Sub8: ccOp += AMD64G_CC_OP_SUBB;   break;
      default:       ppIROp(op8);
                     vpanic("setFlags_DEP1_DEP2(amd64)");
   }
   stmt( IRStmt_Put( OFFB_CC_OP,   mkU64(ccOp)) );
   stmt( IRStmt_Put( OFFB_CC_DEP1, widenUto64(mkexpr(dep1))) );
   stmt( IRStmt_Put( OFFB_CC_DEP2, widenUto64(mkexpr(dep2))) );
}


/* Set the OP and DEP1 fields only, and write zero to DEP2. */

static 
void setFlags_DEP1 ( IROp op8, IRTemp dep1, IRType ty )
{
   Int ccOp = 0;
   switch (ty) {
      case Ity_I8:  ccOp = 0; break;
      case Ity_I16: ccOp = 1; break;
      case Ity_I32: ccOp = 2; break;
      case Ity_I64: ccOp = 3; break;
      default: vassert(0);
   }
   switch (op8) {
      case Iop_Or8:
      case Iop_And8:
      case Iop_Xor8: ccOp += AMD64G_CC_OP_LOGICB; break;
      default:       ppIROp(op8);
                     vpanic("setFlags_DEP1(amd64)");
   }
   stmt( IRStmt_Put( OFFB_CC_OP,   mkU64(ccOp)) );
   stmt( IRStmt_Put( OFFB_CC_DEP1, widenUto64(mkexpr(dep1))) );
   stmt( IRStmt_Put( OFFB_CC_DEP2, mkU64(0)) );
}


/* For shift operations, we put in the result and the undershifted
   result.  Except if the shift amount is zero, the thunk is left
   unchanged. */

static void setFlags_DEP1_DEP2_shift ( IROp    op64,
                                       IRTemp  res,
                                       IRTemp  resUS,
                                       IRType  ty,
                                       IRTemp  guard )
{
   Int ccOp = 0;
   switch (ty) {
      case Ity_I8:  ccOp = 0; break;
      case Ity_I16: ccOp = 1; break;
      case Ity_I32: ccOp = 2; break;
      case Ity_I64: ccOp = 3; break;
      default: vassert(0);
   }

   vassert(guard);

   /* Both kinds of right shifts are handled by the same thunk
      operation. */
   switch (op64) {
      case Iop_Shr64:
      case Iop_Sar64: ccOp += AMD64G_CC_OP_SHRB; break;
      case Iop_Shl64: ccOp += AMD64G_CC_OP_SHLB; break;
      default:        ppIROp(op64);
                      vpanic("setFlags_DEP1_DEP2_shift(amd64)");
   }

   /* DEP1 contains the result, DEP2 contains the undershifted value. */
   stmt( IRStmt_Put( OFFB_CC_OP,
                     IRExpr_Mux0X( mkexpr(guard),
                                   IRExpr_Get(OFFB_CC_OP,Ity_I64),
                                   mkU64(ccOp))) );
   stmt( IRStmt_Put( OFFB_CC_DEP1,
                     IRExpr_Mux0X( mkexpr(guard),
                                   IRExpr_Get(OFFB_CC_DEP1,Ity_I64),
                                   widenUto64(mkexpr(res)))) );
   stmt( IRStmt_Put( OFFB_CC_DEP2, 
                     IRExpr_Mux0X( mkexpr(guard),
                                   IRExpr_Get(OFFB_CC_DEP2,Ity_I64),
                                   widenUto64(mkexpr(resUS)))) );
}


/* For the inc/dec case, we store in DEP1 the result value and in NDEP
   the former value of the carry flag, which unfortunately we have to
   compute. */

static void setFlags_INC_DEC ( Bool inc, IRTemp res, IRType ty )
{
   Int ccOp = inc ? AMD64G_CC_OP_INCB : AMD64G_CC_OP_DECB;

   switch (ty) {
      case Ity_I8:  ccOp += 0; break;
      case Ity_I16: ccOp += 1; break;
      case Ity_I32: ccOp += 2; break;
      case Ity_I64: ccOp += 3; break;
      default: vassert(0);
   }
   
   /* This has to come first, because calculating the C flag 
      may require reading all four thunk fields. */
   stmt( IRStmt_Put( OFFB_CC_NDEP, mk_amd64g_calculate_rflags_c()) );
   stmt( IRStmt_Put( OFFB_CC_OP,   mkU64(ccOp)) );
   stmt( IRStmt_Put( OFFB_CC_DEP1, mkexpr(res)) );
   stmt( IRStmt_Put( OFFB_CC_DEP2, mkU64(0)) );
}


/* Multiplies are pretty much like add and sub: DEP1 and DEP2 hold the
   two arguments. */

static
void setFlags_MUL ( IRType ty, IRTemp arg1, IRTemp arg2, ULong base_op )
{
   switch (ty) {
      case Ity_I8:
         stmt( IRStmt_Put( OFFB_CC_OP, mkU64(base_op+0) ) );
         break;
      case Ity_I16:
         stmt( IRStmt_Put( OFFB_CC_OP, mkU64(base_op+1) ) );
         break;
      case Ity_I32:
         stmt( IRStmt_Put( OFFB_CC_OP, mkU64(base_op+2) ) );
         break;
      case Ity_I64:
         stmt( IRStmt_Put( OFFB_CC_OP, mkU64(base_op+3) ) );
         break;
      default:
         vpanic("setFlags_MUL(amd64)");