ARM: Merge latest VFP fixes from 3dmoo team.

4 files changed, 2388 insertions, 2096 deletions

diff --git a/src/core/arm/skyeye_common/vfp/vfp.cpp b/src/core/arm/skyeye_common/vfp/vfp.cpp
index e4fa3c20..454f6009 100644
--- a/src/core/arm/skyeye_common/vfp/vfp.cpp
+++ b/src/core/arm/skyeye_common/vfp/vfp.cpp
@@ -28,230 +28,270 @@
 #include "core/arm/skyeye_common/armdefs.h"
 #include "core/arm/skyeye_common/vfp/vfp.h"
 
+#define DEBUG DBG
+
 //ARMul_State* persistent_state; /* function calls from SoftFloat lib don't have an access to ARMul_state. */
 
 unsigned
 VFPInit (ARMul_State *state)
 {
-	state->VFP[VFP_OFFSET(VFP_FPSID)] = VFP_FPSID_IMPLMEN<<24 | VFP_FPSID_SW<<23 | VFP_FPSID_SUBARCH<<16 | 
-		VFP_FPSID_PARTNUM<<8 | VFP_FPSID_VARIANT<<4 | VFP_FPSID_REVISION;
-	state->VFP[VFP_OFFSET(VFP_FPEXC)] = 0;
-	state->VFP[VFP_OFFSET(VFP_FPSCR)] = 0;
-	
-	//persistent_state = state;
-	/* Reset only specify VFP_FPEXC_EN = '0' */
-
-	return No_exp;
+    state->VFP[VFP_OFFSET(VFP_FPSID)] = VFP_FPSID_IMPLMEN<<24 | VFP_FPSID_SW<<23 | VFP_FPSID_SUBARCH<<16 |
+                                        VFP_FPSID_PARTNUM<<8 | VFP_FPSID_VARIANT<<4 | VFP_FPSID_REVISION;
+    state->VFP[VFP_OFFSET(VFP_FPEXC)] = 0;
+    state->VFP[VFP_OFFSET(VFP_FPSCR)] = 0;
+
+    //persistent_state = state;
+    /* Reset only specify VFP_FPEXC_EN = '0' */
+
+    return 0;
 }
 
 unsigned
-VFPMRC (ARMul_State * state, unsigned type, ARMword instr, ARMword * value)
+VFPMRC (ARMul_State * state, unsigned type, u32 instr, u32 * value)
 {
-	/* MRC<c> <coproc>,<opc1>,<Rt>,<CRn>,<CRm>{,<opc2>} */
-	int CoProc = BITS (8, 11); /* 10 or 11 */
-	int OPC_1 = BITS (21, 23);
-	int Rt = BITS (12, 15);
-	int CRn = BITS (16, 19);
-	int CRm = BITS (0, 3);
-	int OPC_2 = BITS (5, 7);
-	
-	/* TODO check access permission */
-	
-	/* CRn/opc1 CRm/opc2 */
-	
-	if (CoProc == 10 || CoProc == 11)
-	{
-		#define VFP_MRC_TRANS
-		#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
-		#undef VFP_MRC_TRANS
-	}
-	DEBUG_LOG(ARM11, "Can't identify %x, CoProc %x, OPC_1 %x, Rt %x, CRn %x, CRm %x, OPC_2 %x\n", 
-	       instr, CoProc, OPC_1, Rt, CRn, CRm, OPC_2);
-
-	return ARMul_CANT;
+    /* MRC<c> <coproc>,<opc1>,<Rt>,<CRn>,<CRm>{,<opc2>} */
+    int CoProc = BITS (8, 11); /* 10 or 11 */
+    int OPC_1 = BITS (21, 23);
+    int Rt = BITS (12, 15);
+    int CRn = BITS (16, 19);
+    int CRm = BITS (0, 3);
+    int OPC_2 = BITS (5, 7);
+
+    /* TODO check access permission */
+
+    /* CRn/opc1 CRm/opc2 */
+
+    if (CoProc == 10 || CoProc == 11) {
+#define VFP_MRC_TRANS
+#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
+#undef VFP_MRC_TRANS
+    }
+    DEBUG("Can't identify %x, CoProc %x, OPC_1 %x, Rt %x, CRn %x, CRm %x, OPC_2 %x\n",
+          instr, CoProc, OPC_1, Rt, CRn, CRm, OPC_2);
+
+    return ARMul_CANT;
 }
 
 unsigned
-VFPMCR (ARMul_State * state, unsigned type, ARMword instr, ARMword value)
+VFPMCR (ARMul_State * state, unsigned type, u32 instr, u32 value)
 {
-	/* MCR<c> <coproc>,<opc1>,<Rt>,<CRn>,<CRm>{,<opc2>} */
-	int CoProc = BITS (8, 11); /* 10 or 11 */
-	int OPC_1 = BITS (21, 23);
-	int Rt = BITS (12, 15);
-	int CRn = BITS (16, 19);
-	int CRm = BITS (0, 3);
-	int OPC_2 = BITS (5, 7);
-	
-	/* TODO check access permission */
-	
-	/* CRn/opc1 CRm/opc2 */
-	if (CoProc == 10 || CoProc == 11)
-	{
-		#define VFP_MCR_TRANS
-		#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
-		#undef VFP_MCR_TRANS
-	}
-	DEBUG_LOG(ARM11, "Can't identify %x, CoProc %x, OPC_1 %x, Rt %x, CRn %x, CRm %x, OPC_2 %x\n", 
-	       instr, CoProc, OPC_1, Rt, CRn, CRm, OPC_2);
-
-	return ARMul_CANT;
+    /* MCR<c> <coproc>,<opc1>,<Rt>,<CRn>,<CRm>{,<opc2>} */
+    int CoProc = BITS (8, 11); /* 10 or 11 */
+    int OPC_1 = BITS (21, 23);
+    int Rt = BITS (12, 15);
+    int CRn = BITS (16, 19);
+    int CRm = BITS (0, 3);
+    int OPC_2 = BITS (5, 7);
+
+    /* TODO check access permission */
+
+    /* CRn/opc1 CRm/opc2 */
+    if (CoProc == 10 || CoProc == 11) {
+#define VFP_MCR_TRANS
+#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
+#undef VFP_MCR_TRANS
+    }
+    DEBUG("Can't identify %x, CoProc %x, OPC_1 %x, Rt %x, CRn %x, CRm %x, OPC_2 %x\n",
+          instr, CoProc, OPC_1, Rt, CRn, CRm, OPC_2);
+
+    return ARMul_CANT;
 }
 
 unsigned
-VFPMRRC (ARMul_State * state, unsigned type, ARMword instr, ARMword * value1, ARMword * value2)
+VFPMRRC (ARMul_State * state, unsigned type, u32 instr, u32 * value1, u32 * value2)
 {
-	/* MCRR<c> <coproc>,<opc1>,<Rt>,<Rt2>,<CRm> */
-	int CoProc = BITS (8, 11); /* 10 or 11 */
-	int OPC_1 = BITS (4, 7);
-	int Rt = BITS (12, 15);
-	int Rt2 = BITS (16, 19);
-	int CRm = BITS (0, 3);
-	
-	if (CoProc == 10 || CoProc == 11)
-	{
-		#define VFP_MRRC_TRANS
-		#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
-		#undef VFP_MRRC_TRANS
-	}
-	DEBUG_LOG(ARM11, "Can't identify %x, CoProc %x, OPC_1 %x, Rt %x, Rt2 %x, CRm %x\n", 
-	       instr, CoProc, OPC_1, Rt, Rt2, CRm);
-
-	return ARMul_CANT;
+    /* MCRR<c> <coproc>,<opc1>,<Rt>,<Rt2>,<CRm> */
+    int CoProc = BITS (8, 11); /* 10 or 11 */
+    int OPC_1 = BITS (4, 7);
+    int Rt = BITS (12, 15);
+    int Rt2 = BITS (16, 19);
+    int CRm = BITS (0, 3);
+
+    if (CoProc == 10 || CoProc == 11) {
+#define VFP_MRRC_TRANS
+#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
+#undef VFP_MRRC_TRANS
+    }
+    DEBUG("Can't identify %x, CoProc %x, OPC_1 %x, Rt %x, Rt2 %x, CRm %x\n",
+          instr, CoProc, OPC_1, Rt, Rt2, CRm);
+
+    return ARMul_CANT;
 }
 
 unsigned
-VFPMCRR (ARMul_State * state, unsigned type, ARMword instr, ARMword value1, ARMword value2)
+VFPMCRR (ARMul_State * state, unsigned type, u32 instr, u32 value1, u32 value2)
 {
-	/* MCRR<c> <coproc>,<opc1>,<Rt>,<Rt2>,<CRm> */
-	int CoProc = BITS (8, 11); /* 10 or 11 */
-	int OPC_1 = BITS (4, 7);
-	int Rt = BITS (12, 15);
-	int Rt2 = BITS (16, 19);
-	int CRm = BITS (0, 3);
-	
-	/* TODO check access permission */
-	
-	/* CRn/opc1 CRm/opc2 */
-	
-	if (CoProc == 11 || CoProc == 10)
-	{
-		#define VFP_MCRR_TRANS
-		#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
-		#undef VFP_MCRR_TRANS
-	}
-	DEBUG_LOG(ARM11, "Can't identify %x, CoProc %x, OPC_1 %x, Rt %x, Rt2 %x, CRm %x\n", 
-	       instr, CoProc, OPC_1, Rt, Rt2, CRm);
-
-	return ARMul_CANT;
+    /* MCRR<c> <coproc>,<opc1>,<Rt>,<Rt2>,<CRm> */
+    int CoProc = BITS (8, 11); /* 10 or 11 */
+    int OPC_1 = BITS (4, 7);
+    int Rt = BITS (12, 15);
+    int Rt2 = BITS (16, 19);
+    int CRm = BITS (0, 3);
+
+    /* TODO check access permission */
+
+    /* CRn/opc1 CRm/opc2 */
+
+    if (CoProc == 11 || CoProc == 10) {
+#define VFP_MCRR_TRANS
+#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
+#undef VFP_MCRR_TRANS
+    }
+    DEBUG("Can't identify %x, CoProc %x, OPC_1 %x, Rt %x, Rt2 %x, CRm %x\n",
+          instr, CoProc, OPC_1, Rt, Rt2, CRm);
+
+    return ARMul_CANT;
 }
 
 unsigned
-VFPSTC (ARMul_State * state, unsigned type, ARMword instr, ARMword * value)
+VFPSTC (ARMul_State * state, unsigned type, u32 instr, u32 * value)
 {
-	/* STC{L}<c> <coproc>,<CRd>,[<Rn>],<option> */
-	int CoProc = BITS (8, 11); /* 10 or 11 */
-	int CRd = BITS (12, 15);
-	int Rn = BITS (16, 19);
-	int imm8 = BITS (0, 7);
-	int P = BIT(24);
-	int U = BIT(23);
-	int D = BIT(22);
-	int W = BIT(21);
-	
-	/* TODO check access permission */
-	
-	/* VSTM */
-	if ( (P|U|D|W) == 0 )
-	{
-		DEBUG_LOG(ARM11, "In %s, UNDEFINED\n", __FUNCTION__); exit(-1);
-	}
-	if (CoProc == 10 || CoProc == 11)
-	{
-		#if 1
-		if (P == 0 && U == 0 && W == 0)
-		{
-			DEBUG_LOG(ARM11, "VSTM Related encodings\n"); exit(-1);
-		}
-		if (P == U && W == 1)
-		{
-			DEBUG_LOG(ARM11, "UNDEFINED\n"); exit(-1);
-		}
-		#endif
-
-		#define VFP_STC_TRANS
-		#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
-		#undef VFP_STC_TRANS
-	}
-	DEBUG_LOG(ARM11, "Can't identify %x, CoProc %x, CRd %x, Rn %x, imm8 %x, P %x, U %x, D %x, W %x\n", 
-	       instr, CoProc, CRd, Rn, imm8, P, U, D, W);
-
-	return ARMul_CANT;
+    /* STC{L}<c> <coproc>,<CRd>,[<Rn>],<option> */
+    int CoProc = BITS (8, 11); /* 10 or 11 */
+    int CRd = BITS (12, 15);
+    int Rn = BITS (16, 19);
+    int imm8 = BITS (0, 7);
+    int P = BIT(24);
+    int U = BIT(23);
+    int D = BIT(22);
+    int W = BIT(21);
+
+    /* TODO check access permission */
+
+    /* VSTM */
+    if ( (P|U|D|W) == 0 ) {
+        DEBUG("In %s, UNDEFINED\n", __FUNCTION__);
+        exit(-1);
+    }
+    if (CoProc == 10 || CoProc == 11) {
+#if 1
+        if (P == 0 && U == 0 && W == 0) {
+            DEBUG("VSTM Related encodings\n");
+            exit(-1);
+        }
+        if (P == U && W == 1) {
+            DEBUG("UNDEFINED\n");
+            exit(-1);
+        }
+#endif
+
+#define VFP_STC_TRANS
+#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
+#undef VFP_STC_TRANS
+    }
+    DEBUG("Can't identify %x, CoProc %x, CRd %x, Rn %x, imm8 %x, P %x, U %x, D %x, W %x\n",
+          instr, CoProc, CRd, Rn, imm8, P, U, D, W);
+
+    return ARMul_CANT;
 }
 
 unsigned
-VFPLDC (ARMul_State * state, unsigned type, ARMword instr, ARMword value)
+VFPLDC (ARMul_State * state, unsigned type, u32 instr, u32 value)
 {
-	/* LDC{L}<c> <coproc>,<CRd>,[<Rn>] */
-	int CoProc = BITS (8, 11); /* 10 or 11 */
-	int CRd = BITS (12, 15);
-	int Rn = BITS (16, 19);
-	int imm8 = BITS (0, 7);
-	int P = BIT(24);
-	int U = BIT(23);
-	int D = BIT(22);
-	int W = BIT(21);
-	
-	/* TODO check access permission */
-	
-	if ( (P|U|D|W) == 0 )
-	{
-		DEBUG_LOG(ARM11, "In %s, UNDEFINED\n", __FUNCTION__); exit(-1);
-	}
-	if (CoProc == 10 || CoProc == 11)
-	{
-		#define VFP_LDC_TRANS
-		#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
-		#undef VFP_LDC_TRANS
-	}
-	DEBUG_LOG(ARM11, "Can't identify %x, CoProc %x, CRd %x, Rn %x, imm8 %x, P %x, U %x, D %x, W %x\n", 
-	       instr, CoProc, CRd, Rn, imm8, P, U, D, W);
-
-	return ARMul_CANT;
+    /* LDC{L}<c> <coproc>,<CRd>,[<Rn>] */
+    int CoProc = BITS (8, 11); /* 10 or 11 */
+    int CRd = BITS (12, 15);
+    int Rn = BITS (16, 19);
+    int imm8 = BITS (0, 7);
+    int P = BIT(24);
+    int U = BIT(23);
+    int D = BIT(22);
+    int W = BIT(21);
+
+    /* TODO check access permission */
+
+    if ( (P|U|D|W) == 0 ) {
+        DEBUG("In %s, UNDEFINED\n", __FUNCTION__);
+        exit(-1);
+    }
+    if (CoProc == 10 || CoProc == 11) {
+#define VFP_LDC_TRANS
+#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
+#undef VFP_LDC_TRANS
+    }
+    DEBUG("Can't identify %x, CoProc %x, CRd %x, Rn %x, imm8 %x, P %x, U %x, D %x, W %x\n",
+          instr, CoProc, CRd, Rn, imm8, P, U, D, W);
+
+    return ARMul_CANT;
 }
 
 unsigned
-VFPCDP (ARMul_State * state, unsigned type, ARMword instr)
+VFPCDP (ARMul_State * state, unsigned type, u32 instr)
 {
-	/* CDP<c> <coproc>,<opc1>,<CRd>,<CRn>,<CRm>,<opc2> */
-	int CoProc = BITS (8, 11); /* 10 or 11 */
-	int OPC_1 = BITS (20, 23);
-	int CRd = BITS (12, 15);
-	int CRn = BITS (16, 19);
-	int CRm = BITS (0, 3);
-	int OPC_2 = BITS (5, 7);
-	
-	/* TODO check access permission */
-	
-	/* CRn/opc1 CRm/opc2 */
-
-	if (CoProc == 10 || CoProc == 11)
-	{
-		#define VFP_CDP_TRANS
-		#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
-		#undef VFP_CDP_TRANS
-		
-		int exceptions = 0;
-		if (CoProc == 10)
-			exceptions = vfp_single_cpdo(state, instr, state->VFP[VFP_OFFSET(VFP_FPSCR)]);
-		else 
-			exceptions = vfp_double_cpdo(state, instr, state->VFP[VFP_OFFSET(VFP_FPSCR)]);
-
-		vfp_raise_exceptions(state, exceptions, instr, state->VFP[VFP_OFFSET(VFP_FPSCR)]);
-
-		return ARMul_DONE;
-	}
-	DEBUG_LOG(ARM11, "Can't identify %x\n", instr);
-	return ARMul_CANT;
+    /* CDP<c> <coproc>,<opc1>,<CRd>,<CRn>,<CRm>,<opc2> */
+    int CoProc = BITS (8, 11); /* 10 or 11 */
+    int OPC_1 = BITS (20, 23);
+    int CRd = BITS (12, 15);
+    int CRn = BITS (16, 19);
+    int CRm = BITS (0, 3);
+    int OPC_2 = BITS (5, 7);
+
+    //ichfly
+    /*if ((instr & 0x0FBF0FD0) == 0x0EB70AC0) //vcvt.f64.f32	d8, s16 (s is bit 0-3 and LSB bit 22) (d is bit 12 - 15 MSB is Bit 6)
+    {
+        struct vfp_double vdd;
+        struct vfp_single vsd;
+        int dn = BITS(12, 15) + (BIT(22) << 4);
+        int sd = (BITS(0, 3) << 1) + BIT(5);
+        s32 n = vfp_get_float(state, sd);
+        vfp_single_unpack(&vsd, n);
+        if (vsd.exponent & 0x80)
+        {
+            vdd.exponent = (vsd.exponent&~0x80) | 0x400;
+        }
+        else
+        {
+            vdd.exponent = vsd.exponent | 0x380;
+        }
+        vdd.sign = vsd.sign;
+        vdd.significand = (u64)(vsd.significand & ~0xC0000000) << 32; // I have no idea why but the 2 uppern bits are not from the significand
+        vfp_put_double(state, vfp_double_pack(&vdd), dn);
+        return ARMul_DONE;
+    }
+    if ((instr & 0x0FBF0FD0) == 0x0EB70BC0) //vcvt.f32.f64	s15, d6
+    {
+        struct vfp_double vdd;
+        struct vfp_single vsd;
+        int sd = BITS(0, 3) + (BIT(5) << 4);
+        int dn = (BITS(12, 15) << 1) + BIT(22);
+        vfp_double_unpack(&vdd, vfp_get_double(state, sd));
+        if (vdd.exponent & 0x400) //todo if the exponent is to low or to high for this convert
+        {
+            vsd.exponent = (vdd.exponent) | 0x80;
+        }
+        else
+        {
+            vsd.exponent = vdd.exponent & ~0x80;
+        }
+        vsd.exponent &= 0xFF;
+       // vsd.exponent = vdd.exponent >> 3;
+        vsd.sign = vdd.sign;
+        vsd.significand = ((u64)(vdd.significand ) >> 32)& ~0xC0000000;
+        vfp_put_float(state, vfp_single_pack(&vsd), dn);
+        return ARMul_DONE;
+    }*/
+
+    /* TODO check access permission */
+
+    /* CRn/opc1 CRm/opc2 */
+
+    if (CoProc == 10 || CoProc == 11) {
+#define VFP_CDP_TRANS
+#include "core/arm/skyeye_common/vfp/vfpinstr.cpp"
+#undef VFP_CDP_TRANS
+
+        int exceptions = 0;
+        if (CoProc == 10)
+            exceptions = vfp_single_cpdo(state, instr, state->VFP[VFP_OFFSET(VFP_FPSCR)]);
+        else
+            exceptions = vfp_double_cpdo(state, instr, state->VFP[VFP_OFFSET(VFP_FPSCR)]);
+
+        vfp_raise_exceptions(state, exceptions, instr, state->VFP[VFP_OFFSET(VFP_FPSCR)]);
+
+        return ARMul_DONE;
+    }
+    DEBUG("Can't identify %x\n", instr);
+    return ARMul_CANT;
 }
 
 
@@ -301,29 +341,29 @@ VFPCDP (ARMul_State * state, unsigned type, ARMword instr)
 /* Miscellaneous functions */
 int32_t vfp_get_float(arm_core_t* state, unsigned int reg)
 {
-	DBG("VFP get float: s%d=[%08x]\n", reg, state->ExtReg[reg]);
-	return state->ExtReg[reg];
+    DEBUG("VFP get float: s%d=[%08x]\n", reg, state->ExtReg[reg]);
+    return state->ExtReg[reg];
 }
 
 void vfp_put_float(arm_core_t* state, int32_t val, unsigned int reg)
 {
-	DBG("VFP put float: s%d <= [%08x]\n", reg, val);
-	state->ExtReg[reg] = val;
+    DEBUG("VFP put float: s%d <= [%08x]\n", reg, val);
+    state->ExtReg[reg] = val;
 }
 
 uint64_t vfp_get_double(arm_core_t* state, unsigned int reg)
 {
-	uint64_t result;
-	result = ((uint64_t) state->ExtReg[reg*2+1])<<32 | state->ExtReg[reg*2];
-	DBG("VFP get double: s[%d-%d]=[%016llx]\n", reg*2+1, reg*2, result);
-	return result;
+    uint64_t result;
+    result = ((uint64_t) state->ExtReg[reg*2+1])<<32 | state->ExtReg[reg*2];
+    DEBUG("VFP get double: s[%d-%d]=[%016llx]\n", reg*2+1, reg*2, result);
+    return result;
 }
 
 void vfp_put_double(arm_core_t* state, uint64_t val, unsigned int reg)
 {
-	DBG("VFP put double: s[%d-%d] <= [%08x-%08x]\n", reg*2+1, reg*2, (uint32_t) (val>>32), (uint32_t) (val & 0xffffffff));
-	state->ExtReg[reg*2] = (uint32_t) (val & 0xffffffff);
-	state->ExtReg[reg*2+1] = (uint32_t) (val>>32);
+    DEBUG("VFP put double: s[%d-%d] <= [%08x-%08x]\n", reg*2+1, reg*2, (uint32_t) (val>>32), (uint32_t) (val & 0xffffffff));
+    state->ExtReg[reg*2] = (uint32_t) (val & 0xffffffff);
+    state->ExtReg[reg*2+1] = (uint32_t) (val>>32);
 }
 
 
@@ -333,25 +373,25 @@ void vfp_put_double(arm_core_t* state, uint64_t val, unsigned int reg)
  */
 void vfp_raise_exceptions(ARMul_State* state, u32 exceptions, u32 inst, u32 fpscr)
 {
-	int si_code = 0;
+    int si_code = 0;
 
-	vfpdebug("VFP: raising exceptions %08x\n", exceptions);
+    vfpdebug("VFP: raising exceptions %08x\n", exceptions);
 
-	if (exceptions == VFP_EXCEPTION_ERROR) {
-		DEBUG_LOG(ARM11, "unhandled bounce %x\n", inst);
-		exit(-1);
-		return;
-	}
+    if (exceptions == VFP_EXCEPTION_ERROR) {
+        DEBUG("unhandled bounce %x\n", inst);
+        exit(-1);
+        return;
+    }
 
-	/*
-	 * If any of the status flags are set, update the FPSCR.
-	 * Comparison instructions always return at least one of
-	 * these flags set.
-	 */
-	if (exceptions & (FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V))
-		fpscr &= ~(FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V);
+    /*
+     * If any of the status flags are set, update the FPSCR.
+     * Comparison instructions always return at least one of
+     * these flags set.
+     */
+    if (exceptions & (FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V))
+        fpscr &= ~(FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V);
 
-	fpscr |= exceptions;
+    fpscr |= exceptions;
 
-	state->VFP[VFP_OFFSET(VFP_FPSCR)] = fpscr;
+    state->VFP[VFP_OFFSET(VFP_FPSCR)] = fpscr;
 }
diff --git a/src/core/arm/skyeye_common/vfp/vfp_helper.h b/src/core/arm/skyeye_common/vfp/vfp_helper.h
index 5076e59f..f0896fc8 100644
--- a/src/core/arm/skyeye_common/vfp/vfp_helper.h
+++ b/src/core/arm/skyeye_common/vfp/vfp_helper.h
@@ -44,7 +44,7 @@
 #define pr_info //printf
 #define pr_debug //printf
 
-static u32 vfp_fls(int x);
+static u32 fls(int x);
 #define do_div(n, base) {n/=base;}
 
 /* From vfpinstr.h */
@@ -502,7 +502,7 @@ struct op {
 	u32 flags;
 };
 
-static u32 vfp_fls(int x)
+static u32 fls(int x)
 {
 	int r = 32;
 
@@ -532,4 +532,9 @@ static u32 vfp_fls(int x)
 
 }
 
+u32 vfp_double_normaliseroundintern(ARMul_State* state, struct vfp_double *vd, u32 fpscr, u32 exceptions, const char *func);
+u32 vfp_double_multiply(struct vfp_double *vdd, struct vfp_double *vdn, struct vfp_double *vdm, u32 fpscr);
+u32 vfp_double_add(struct vfp_double *vdd, struct vfp_double *vdn, struct vfp_double *vdm, u32 fpscr);
+u32 vfp_double_fcvtsinterncutting(ARMul_State* state, int sd, struct vfp_double* dm, u32 fpscr);
+
 #endif
diff --git a/src/core/arm/skyeye_common/vfp/vfpdouble.cpp b/src/core/arm/skyeye_common/vfp/vfpdouble.cpp
index 13411ad8..765c1f6b 100644
--- a/src/core/arm/skyeye_common/vfp/vfpdouble.cpp
+++ b/src/core/arm/skyeye_common/vfp/vfpdouble.cpp
@@ -56,163 +56,291 @@
 #include "core/arm/skyeye_common/vfp/asm_vfp.h"
 
 static struct vfp_double vfp_double_default_qnan = {
-	//.exponent	= 2047,
-	//.sign		= 0,
-	//.significand	= VFP_DOUBLE_SIGNIFICAND_QNAN,
+    2047,
+    0,
+    VFP_DOUBLE_SIGNIFICAND_QNAN,
 };
 
 static void vfp_double_dump(const char *str, struct vfp_double *d)
 {
-	pr_debug("VFP: %s: sign=%d exponent=%d significand=%016llx\n",
-		 str, d->sign != 0, d->exponent, d->significand);
+    pr_debug("VFP: %s: sign=%d exponent=%d significand=%016llx\n",
+             str, d->sign != 0, d->exponent, d->significand);
 }
 
 static void vfp_double_normalise_denormal(struct vfp_double *vd)
 {
-	int bits = 31 - vfp_fls(vd->significand >> 32);
-	if (bits == 31)
-		bits = 63 - vfp_fls(vd->significand);
+    int bits = 31 - fls((ARMword)(vd->significand >> 32));
+    if (bits == 31)
+        bits = 63 - fls((ARMword)vd->significand);
 
-	vfp_double_dump("normalise_denormal: in", vd);
+    vfp_double_dump("normalise_denormal: in", vd);
 
-	if (bits) {
-		vd->exponent -= bits - 1;
-		vd->significand <<= bits;
-	}
+    if (bits) {
+        vd->exponent -= bits - 1;
+        vd->significand <<= bits;
+    }
 
-	vfp_double_dump("normalise_denormal: out", vd);
+    vfp_double_dump("normalise_denormal: out", vd);
 }
 
-u32 vfp_double_normaliseround(ARMul_State* state, int dd, struct vfp_double *vd, u32 fpscr, u32 exceptions, const char *func)
+u32 vfp_double_normaliseroundintern(ARMul_State* state, struct vfp_double *vd, u32 fpscr, u32 exceptions, const char *func)
 {
-	u64 significand, incr;
-	int exponent, shift, underflow;
-	u32 rmode;
-
-	vfp_double_dump("pack: in", vd);
-
-	/*
-	 * Infinities and NaNs are a special case.
-	 */
-	if (vd->exponent == 2047 && (vd->significand == 0 || exceptions))
-		goto pack;
-
-	/*
-	 * Special-case zero.
-	 */
-	if (vd->significand == 0) {
-		vd->exponent = 0;
-		goto pack;
-	}
-
-	exponent = vd->exponent;
-	significand = vd->significand;
-
-	shift = 32 - vfp_fls(significand >> 32);
-	if (shift == 32)
-		shift = 64 - vfp_fls(significand);
-	if (shift) {
-		exponent -= shift;
-		significand <<= shift;
-	}
+    u64 significand, incr;
+    int exponent, shift, underflow;
+    u32 rmode;
+
+    vfp_double_dump("pack: in", vd);
+
+    /*
+    * Infinities and NaNs are a special case.
+    */
+    if (vd->exponent == 2047 && (vd->significand == 0 || exceptions))
+        goto pack;
+
+    /*
+    * Special-case zero.
+    */
+    if (vd->significand == 0) {
+        vd->exponent = 0;
+        goto pack;
+    }
+
+    exponent = vd->exponent;
+    significand = vd->significand;
+
+    shift = 32 - fls((ARMword)(significand >> 32));
+    if (shift == 32)
+        shift = 64 - fls((ARMword)significand);
+    if (shift) {
+        exponent -= shift;
+        significand <<= shift;
+    }
 
 #if 1
-	vd->exponent = exponent;
-	vd->significand = significand;
-	vfp_double_dump("pack: normalised", vd);
+    vd->exponent = exponent;
+    vd->significand = significand;
+    vfp_double_dump("pack: normalised", vd);
 #endif
 
-	/*
-	 * Tiny number?
-	 */
-	underflow = exponent < 0;
-	if (underflow) {
-		significand = vfp_shiftright64jamming(significand, -exponent);
-		exponent = 0;
+    /*
+    * Tiny number?
+    */
+    underflow = exponent < 0;
+    if (underflow) {
+        significand = vfp_shiftright64jamming(significand, -exponent);
+        exponent = 0;
 #if 1
-		vd->exponent = exponent;
-		vd->significand = significand;
-		vfp_double_dump("pack: tiny number", vd);
+        vd->exponent = exponent;
+        vd->significand = significand;
+        vfp_double_dump("pack: tiny number", vd);
 #endif
-		if (!(significand & ((1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1)))
-			underflow = 0;
-	}
-
-	/*
-	 * Select rounding increment.
-	 */
-	incr = 0;
-	rmode = fpscr & FPSCR_RMODE_MASK;
-
-	if (rmode == FPSCR_ROUND_NEAREST) {
-		incr = 1ULL << VFP_DOUBLE_LOW_BITS;
-		if ((significand & (1ULL << (VFP_DOUBLE_LOW_BITS + 1))) == 0)
-			incr -= 1;
-	} else if (rmode == FPSCR_ROUND_TOZERO) {
-		incr = 0;
-	} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vd->sign != 0))
-		incr = (1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1;
-
-	pr_debug("VFP: rounding increment = 0x%08llx\n", incr);
-
-	/*
-	 * Is our rounding going to overflow?
-	 */
-	if ((significand + incr) < significand) {
-		exponent += 1;
-		significand = (significand >> 1) | (significand & 1);
-		incr >>= 1;
+        if (!(significand & ((1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1)))
+            underflow = 0;
+    }
+
+    /*
+    * Select rounding increment.
+    */
+    incr = 0;
+    rmode = fpscr & FPSCR_RMODE_MASK;
+
+    if (rmode == FPSCR_ROUND_NEAREST) {
+        incr = 1ULL << VFP_DOUBLE_LOW_BITS;
+        if ((significand & (1ULL << (VFP_DOUBLE_LOW_BITS + 1))) == 0)
+            incr -= 1;
+    }
+    else if (rmode == FPSCR_ROUND_TOZERO) {
+        incr = 0;
+    }
+    else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vd->sign != 0))
+        incr = (1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1;
+
+    pr_debug("VFP: rounding increment = 0x%08llx\n", incr);
+
+    /*
+    * Is our rounding going to overflow?
+    */
+    if ((significand + incr) < significand) {
+        exponent += 1;
+        significand = (significand >> 1) | (significand & 1);
+        incr >>= 1;
 #if 1
-		vd->exponent = exponent;
-		vd->significand = significand;
-		vfp_double_dump("pack: overflow", vd);
+        vd->exponent = exponent;
+        vd->significand = significand;
+        vfp_double_dump("pack: overflow", vd);
 #endif
-	}
-
-	/*
-	 * If any of the low bits (which will be shifted out of the
-	 * number) are non-zero, the result is inexact.
-	 */
-	if (significand & ((1 << (VFP_DOUBLE_LOW_BITS + 1)) - 1))
-		exceptions |= FPSCR_IXC;
-
-	/*
-	 * Do our rounding.
-	 */
-	significand += incr;
-
-	/*
-	 * Infinity?
-	 */
-	if (exponent >= 2046) {
-		exceptions |= FPSCR_OFC | FPSCR_IXC;
-		if (incr == 0) {
-			vd->exponent = 2045;
-			vd->significand = 0x7fffffffffffffffULL;
-		} else {
-			vd->exponent = 2047;		/* infinity */
-			vd->significand = 0;
-		}
-	} else {
-		if (significand >> (VFP_DOUBLE_LOW_BITS + 1) == 0)
-			exponent = 0;
-		if (exponent || significand > 0x8000000000000000ULL)
-			underflow = 0;
-		if (underflow)
-			exceptions |= FPSCR_UFC;
-		vd->exponent = exponent;
-		vd->significand = significand >> 1;
-	}
-
+    }
+
+    /*
+    * If any of the low bits (which will be shifted out of the
+    * number) are non-zero, the result is inexact.
+    */
+    if (significand & ((1 << (VFP_DOUBLE_LOW_BITS + 1)) - 1))
+        exceptions |= FPSCR_IXC;
+
+    /*
+    * Do our rounding.
+    */
+    significand += incr;
+
+    /*
+    * Infinity?
+    */
+    if (exponent >= 2046) {
+        exceptions |= FPSCR_OFC | FPSCR_IXC;
+        if (incr == 0) {
+            vd->exponent = 2045;
+            vd->significand = 0x7fffffffffffffffULL;
+        }
+        else {
+            vd->exponent = 2047;		/* infinity */
+            vd->significand = 0;
+        }
+    }
+    else {
+        if (significand >> (VFP_DOUBLE_LOW_BITS + 1) == 0)
+            exponent = 0;
+        if (exponent || significand > 0x8000000000000000ULL)
+            underflow = 0;
+        if (underflow)
+            exceptions |= FPSCR_UFC;
+        vd->exponent = exponent;
+        vd->significand = significand >> 1;
+    }
  pack:
-	vfp_double_dump("pack: final", vd);
-	{
-		s64 d = vfp_double_pack(vd);
-		pr_debug("VFP: %s: d(d%d)=%016llx exceptions=%08x\n", func,
-			 dd, d, exceptions);
-		vfp_put_double(state, d, dd);
-	}
-	return exceptions;
+    return 0;
+}
+
+u32 vfp_double_normaliseround(ARMul_State* state, int dd, struct vfp_double *vd, u32 fpscr, u32 exceptions, const char *func)
+{
+    u64 significand, incr;
+    int exponent, shift, underflow;
+    u32 rmode;
+
+    vfp_double_dump("pack: in", vd);
+
+    /*
+     * Infinities and NaNs are a special case.
+     */
+    if (vd->exponent == 2047 && (vd->significand == 0 || exceptions))
+        goto pack;
+
+    /*
+     * Special-case zero.
+     */
+    if (vd->significand == 0) {
+        vd->exponent = 0;
+        goto pack;
+    }
+
+    exponent = vd->exponent;
+    significand = vd->significand;
+
+    shift = 32 - fls((ARMword)(significand >> 32));
+    if (shift == 32)
+        shift = 64 - fls((ARMword)significand);
+    if (shift) {
+        exponent -= shift;
+        significand <<= shift;
+    }
+
+#if 1
+    vd->exponent = exponent;
+    vd->significand = significand;
+    vfp_double_dump("pack: normalised", vd);
+#endif
+
+    /*
+     * Tiny number?
+     */
+    underflow = exponent < 0;
+    if (underflow) {
+        significand = vfp_shiftright64jamming(significand, -exponent);
+        exponent = 0;
+#if 1
+        vd->exponent = exponent;
+        vd->significand = significand;
+        vfp_double_dump("pack: tiny number", vd);
+#endif
+        if (!(significand & ((1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1)))
+            underflow = 0;
+    }
+
+    /*
+     * Select rounding increment.
+     */
+    incr = 0;
+    rmode = fpscr & FPSCR_RMODE_MASK;
+
+    if (rmode == FPSCR_ROUND_NEAREST) {
+        incr = 1ULL << VFP_DOUBLE_LOW_BITS;
+        if ((significand & (1ULL << (VFP_DOUBLE_LOW_BITS + 1))) == 0)
+            incr -= 1;
+    } else if (rmode == FPSCR_ROUND_TOZERO) {
+        incr = 0;
+    } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vd->sign != 0))
+        incr = (1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1;
+
+    pr_debug("VFP: rounding increment = 0x%08llx\n", incr);
+
+    /*
+     * Is our rounding going to overflow?
+     */
+    if ((significand + incr) < significand) {
+        exponent += 1;
+        significand = (significand >> 1) | (significand & 1);
+        incr >>= 1;
+#if 1
+        vd->exponent = exponent;
+        vd->significand = significand;
+        vfp_double_dump("pack: overflow", vd);
+#endif
+    }
+
+    /*
+     * If any of the low bits (which will be shifted out of the
+     * number) are non-zero, the result is inexact.
+     */
+    if (significand & ((1 << (VFP_DOUBLE_LOW_BITS + 1)) - 1))
+        exceptions |= FPSCR_IXC;
+
+    /*
+     * Do our rounding.
+     */
+    significand += incr;
+
+    /*
+     * Infinity?
+     */
+    if (exponent >= 2046) {
+        exceptions |= FPSCR_OFC | FPSCR_IXC;
+        if (incr == 0) {
+            vd->exponent = 2045;
+            vd->significand = 0x7fffffffffffffffULL;
+        } else {
+            vd->exponent = 2047;		/* infinity */
+            vd->significand = 0;
+        }
+    } else {
+        if (significand >> (VFP_DOUBLE_LOW_BITS + 1) == 0)
+            exponent = 0;
+        if (exponent || significand > 0x8000000000000000ULL)
+            underflow = 0;
+        if (underflow)
+            exceptions |= FPSCR_UFC;
+        vd->exponent = exponent;
+        vd->significand = significand >> 1;
+    }
+
+pack:
+    vfp_double_dump("pack: final", vd);
+    {
+        s64 d = vfp_double_pack(vd);
+        pr_debug("VFP: %s: d(d%d)=%016llx exceptions=%08x\n", func,
+                 dd, d, exceptions);
+        vfp_put_double(state, d, dd);
+    }
+    return exceptions;
 }
 
 /*
@@ -221,43 +349,43 @@ u32 vfp_double_normaliseround(ARMul_State* state, int dd, struct vfp_double *vd,
  */
 static u32
 vfp_propagate_nan(struct vfp_double *vdd, struct vfp_double *vdn,
-		  struct vfp_double *vdm, u32 fpscr)
+                  struct vfp_double *vdm, u32 fpscr)
 {
-	struct vfp_double *nan;
-	int tn, tm = 0;
-
-	tn = vfp_double_type(vdn);
-
-	if (vdm)
-		tm = vfp_double_type(vdm);
-
-	if (fpscr & FPSCR_DEFAULT_NAN)
-		/*
-		 * Default NaN mode - always returns a quiet NaN
-		 */
-		nan = &vfp_double_default_qnan;
-	else {
-		/*
-		 * Contemporary mode - select the first signalling
-		 * NAN, or if neither are signalling, the first
-		 * quiet NAN.
-		 */
-		if (tn == VFP_SNAN || (tm != VFP_SNAN && tn == VFP_QNAN))
-			nan = vdn;
-		else
-			nan = vdm;
-		/*
-		 * Make the NaN quiet.
-		 */
-		nan->significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
-	}
-
-	*vdd = *nan;
-
-	/*
-	 * If one was a signalling NAN, raise invalid operation.
-	 */
-	return tn == VFP_SNAN || tm == VFP_SNAN ? FPSCR_IOC : VFP_NAN_FLAG;
+    struct vfp_double *nan;
+    int tn, tm = 0;
+
+    tn = vfp_double_type(vdn);
+
+    if (vdm)
+        tm = vfp_double_type(vdm);
+
+    if (fpscr & FPSCR_DEFAULT_NAN)
+        /*
+         * Default NaN mode - always returns a quiet NaN
+         */
+        nan = &vfp_double_default_qnan;
+    else {
+        /*
+         * Contemporary mode - select the first signalling
+         * NAN, or if neither are signalling, the first
+         * quiet NAN.
+         */
+        if (tn == VFP_SNAN || (tm != VFP_SNAN && tn == VFP_QNAN))
+            nan = vdn;
+        else
+            nan = vdm;
+        /*
+         * Make the NaN quiet.
+         */
+        nan->significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
+    }
+
+    *vdd = *nan;
+
+    /*
+     * If one was a signalling NAN, raise invalid operation.
+     */
+    return tn == VFP_SNAN || tm == VFP_SNAN ? FPSCR_IOC : VFP_NAN_FLAG;
 }
 
 /*
@@ -265,108 +393,108 @@ vfp_propagate_nan(struct vfp_double *vdd, struct vfp_double *vdn,
  */
 static u32 vfp_double_fabs(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_put_double(state, vfp_double_packed_abs(vfp_get_double(state, dm)), dd);
-	return 0;
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_put_double(state, vfp_double_packed_abs(vfp_get_double(state, dm)), dd);
+    return 0;
 }
 
 static u32 vfp_double_fcpy(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_put_double(state, vfp_get_double(state, dm), dd);
-	return 0;
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_put_double(state, vfp_get_double(state, dm), dd);
+    return 0;
 }
 
 static u32 vfp_double_fneg(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_put_double(state, vfp_double_packed_negate(vfp_get_double(state, dm)), dd);
-	return 0;
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_put_double(state, vfp_double_packed_negate(vfp_get_double(state, dm)), dd);
+    return 0;
 }
 
 static u32 vfp_double_fsqrt(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	struct vfp_double vdm, vdd, *vdp;
-	int ret, tm;
-
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-	tm = vfp_double_type(&vdm);
-	if (tm & (VFP_NAN|VFP_INFINITY)) {
-		vdp = &vdd;
-
-		if (tm & VFP_NAN)
-			ret = vfp_propagate_nan(vdp, &vdm, NULL, fpscr);
-		else if (vdm.sign == 0) {
- sqrt_copy:
-			vdp = &vdm;
-			ret = 0;
-		} else {
- sqrt_invalid:
-			vdp = &vfp_double_default_qnan;
-			ret = FPSCR_IOC;
-		}
-		vfp_put_double(state, vfp_double_pack(vdp), dd);
-		return ret;
-	}
-
-	/*
-	 * sqrt(+/- 0) == +/- 0
-	 */
-	if (tm & VFP_ZERO)
-		goto sqrt_copy;
-
-	/*
-	 * Normalise a denormalised number
-	 */
-	if (tm & VFP_DENORMAL)
-		vfp_double_normalise_denormal(&vdm);
-
-	/*
-	 * sqrt(<0) = invalid
-	 */
-	if (vdm.sign)
-		goto sqrt_invalid;
-
-	vfp_double_dump("sqrt", &vdm);
-
-	/*
-	 * Estimate the square root.
-	 */
-	vdd.sign = 0;
-	vdd.exponent = ((vdm.exponent - 1023) >> 1) + 1023;
-	vdd.significand = (u64)vfp_estimate_sqrt_significand(vdm.exponent, vdm.significand >> 32) << 31;
-
-	vfp_double_dump("sqrt estimate1", &vdd);
-
-	vdm.significand >>= 1 + (vdm.exponent & 1);
-	vdd.significand += 2 + vfp_estimate_div128to64(vdm.significand, 0, vdd.significand);
-
-	vfp_double_dump("sqrt estimate2", &vdd);
-
-	/*
-	 * And now adjust.
-	 */
-	if ((vdd.significand & VFP_DOUBLE_LOW_BITS_MASK) <= 5) {
-		if (vdd.significand < 2) {
-			vdd.significand = ~0ULL;
-		} else {
-			u64 termh, terml, remh, reml;
-			vdm.significand <<= 2;
-			mul64to128(&termh, &terml, vdd.significand, vdd.significand);
-			sub128(&remh, &reml, vdm.significand, 0, termh, terml);
-			while ((s64)remh < 0) {
-				vdd.significand -= 1;
-				shift64left(&termh, &terml, vdd.significand);
-				terml |= 1;
-				add128(&remh, &reml, remh, reml, termh, terml);
-			}
-			vdd.significand |= (remh | reml) != 0;
-		}
-	}
-	vdd.significand = vfp_shiftright64jamming(vdd.significand, 1);
-
-	return vfp_double_normaliseround(state, dd, &vdd, fpscr, 0, "fsqrt");
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double vdm, vdd, *vdp;
+    int ret, tm;
+
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+    tm = vfp_double_type(&vdm);
+    if (tm & (VFP_NAN|VFP_INFINITY)) {
+        vdp = &vdd;
+
+        if (tm & VFP_NAN)
+            ret = vfp_propagate_nan(vdp, &vdm, NULL, fpscr);
+        else if (vdm.sign == 0) {
+sqrt_copy:
+            vdp = &vdm;
+            ret = 0;
+        } else {
+sqrt_invalid:
+            vdp = &vfp_double_default_qnan;
+            ret = FPSCR_IOC;
+        }
+        vfp_put_double(state, vfp_double_pack(vdp), dd);
+        return ret;
+    }
+
+    /*
+     * sqrt(+/- 0) == +/- 0
+     */
+    if (tm & VFP_ZERO)
+        goto sqrt_copy;
+
+    /*
+     * Normalise a denormalised number
+     */
+    if (tm & VFP_DENORMAL)
+        vfp_double_normalise_denormal(&vdm);
+
+    /*
+     * sqrt(<0) = invalid
+     */
+    if (vdm.sign)
+        goto sqrt_invalid;
+
+    vfp_double_dump("sqrt", &vdm);
+
+    /*
+     * Estimate the square root.
+     */
+    vdd.sign = 0;
+    vdd.exponent = ((vdm.exponent - 1023) >> 1) + 1023;
+    vdd.significand = (u64)vfp_estimate_sqrt_significand(vdm.exponent, vdm.significand >> 32) << 31;
+
+    vfp_double_dump("sqrt estimate1", &vdd);
+
+    vdm.significand >>= 1 + (vdm.exponent & 1);
+    vdd.significand += 2 + vfp_estimate_div128to64(vdm.significand, 0, vdd.significand);
+
+    vfp_double_dump("sqrt estimate2", &vdd);
+
+    /*
+     * And now adjust.
+     */
+    if ((vdd.significand & VFP_DOUBLE_LOW_BITS_MASK) <= 5) {
+        if (vdd.significand < 2) {
+            vdd.significand = ~0ULL;
+        } else {
+            u64 termh, terml, remh, reml;
+            vdm.significand <<= 2;
+            mul64to128(&termh, &terml, vdd.significand, vdd.significand);
+            sub128(&remh, &reml, vdm.significand, 0, termh, terml);
+            while ((s64)remh < 0) {
+                vdd.significand -= 1;
+                shift64left(&termh, &terml, vdd.significand);
+                terml |= 1;
+                add128(&remh, &reml, remh, reml, termh, terml);
+            }
+            vdd.significand |= (remh | reml) != 0;
+        }
+    }
+    vdd.significand = vfp_shiftright64jamming(vdd.significand, 1);
+
+    return vfp_double_normaliseround(state, dd, &vdd, fpscr, 0, "fsqrt");
 }
 
 /*
@@ -377,319 +505,362 @@ static u32 vfp_double_fsqrt(ARMul_State* state, int dd, int unused, int dm, u32
  */
 static u32 vfp_compare(ARMul_State* state, int dd, int signal_on_qnan, int dm, u32 fpscr)
 {
-	s64 d, m;
-	u32 ret = 0;
-
-	pr_debug("In %s, state=0x%x, fpscr=0x%x\n", __FUNCTION__, state, fpscr);
-	m = vfp_get_double(state, dm);
-	if (vfp_double_packed_exponent(m) == 2047 && vfp_double_packed_mantissa(m)) {
-		ret |= FPSCR_C | FPSCR_V;
-		if (signal_on_qnan || !(vfp_double_packed_mantissa(m) & (1ULL << (VFP_DOUBLE_MANTISSA_BITS - 1))))
-			/*
-			 * Signalling NaN, or signalling on quiet NaN
-			 */
-			ret |= FPSCR_IOC;
-	}
-
-	d = vfp_get_double(state, dd);
-	if (vfp_double_packed_exponent(d) == 2047 && vfp_double_packed_mantissa(d)) {
-		ret |= FPSCR_C | FPSCR_V;
-		if (signal_on_qnan || !(vfp_double_packed_mantissa(d) & (1ULL << (VFP_DOUBLE_MANTISSA_BITS - 1))))
-			/*
-			 * Signalling NaN, or signalling on quiet NaN
-			 */
-			ret |= FPSCR_IOC;
-	}
-
-	if (ret == 0) {
-		//printf("In %s, d=%lld, m =%lld\n ", __FUNCTION__, d, m);
-		if (d == m || vfp_double_packed_abs(d | m) == 0) {
-			/*
-			 * equal
-			 */
-			ret |= FPSCR_Z | FPSCR_C;
-			//printf("In %s,1 ret=0x%x\n", __FUNCTION__, ret);
-		} else if (vfp_double_packed_sign(d ^ m)) {
-			/*
-			 * different signs
-			 */
-			if (vfp_double_packed_sign(d))
-				/*
-				 * d is negative, so d < m
-				 */
-				ret |= FPSCR_N;
-			else
-				/*
-				 * d is positive, so d > m
-				 */
-				ret |= FPSCR_C;
-		} else if ((vfp_double_packed_sign(d) != 0) ^ (d < m)) {
-			/*
-			 * d < m
-			 */
-			ret |= FPSCR_N;
-		} else if ((vfp_double_packed_sign(d) != 0) ^ (d > m)) {
-			/*
-			 * d > m
-			 */
-			ret |= FPSCR_C;
-		}
-	}
-	pr_debug("In %s, state=0x%x, ret=0x%x\n", __FUNCTION__, state, ret);
-
-	return ret;
+    s64 d, m;
+    u32 ret = 0;
+
+    pr_debug("In %s, state=0x%x, fpscr=0x%x\n", __FUNCTION__, state, fpscr);
+    m = vfp_get_double(state, dm);
+    if (vfp_double_packed_exponent(m) == 2047 && vfp_double_packed_mantissa(m)) {
+        ret |= FPSCR_C | FPSCR_V;
+        if (signal_on_qnan || !(vfp_double_packed_mantissa(m) & (1ULL << (VFP_DOUBLE_MANTISSA_BITS - 1))))
+            /*
+             * Signalling NaN, or signalling on quiet NaN
+             */
+            ret |= FPSCR_IOC;
+    }
+
+    d = vfp_get_double(state, dd);
+    if (vfp_double_packed_exponent(d) == 2047 && vfp_double_packed_mantissa(d)) {
+        ret |= FPSCR_C | FPSCR_V;
+        if (signal_on_qnan || !(vfp_double_packed_mantissa(d) & (1ULL << (VFP_DOUBLE_MANTISSA_BITS - 1))))
+            /*
+             * Signalling NaN, or signalling on quiet NaN
+             */
+            ret |= FPSCR_IOC;
+    }
+
+    if (ret == 0) {
+        //printf("In %s, d=%lld, m =%lld\n ", __FUNCTION__, d, m);
+        if (d == m || vfp_double_packed_abs(d | m) == 0) {
+            /*
+             * equal
+             */
+            ret |= FPSCR_Z | FPSCR_C;
+            //printf("In %s,1 ret=0x%x\n", __FUNCTION__, ret);
+        } else if (vfp_double_packed_sign(d ^ m)) {
+            /*
+             * different signs
+             */
+            if (vfp_double_packed_sign(d))
+                /*
+                 * d is negative, so d < m
+                 */
+                ret |= FPSCR_N;
+            else
+                /*
+                 * d is positive, so d > m
+                 */
+                ret |= FPSCR_C;
+        } else if ((vfp_double_packed_sign(d) != 0) ^ (d < m)) {
+            /*
+             * d < m
+             */
+            ret |= FPSCR_N;
+        } else if ((vfp_double_packed_sign(d) != 0) ^ (d > m)) {
+            /*
+             * d > m
+             */
+            ret |= FPSCR_C;
+        }
+    }
+    pr_debug("In %s, state=0x%x, ret=0x%x\n", __FUNCTION__, state, ret);
+
+    return ret;
 }
 
 static u32 vfp_double_fcmp(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_compare(state, dd, 0, dm, fpscr);
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_compare(state, dd, 0, dm, fpscr);
 }
 
 static u32 vfp_double_fcmpe(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_compare(state, dd, 1, dm, fpscr);
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_compare(state, dd, 1, dm, fpscr);
 }
 
 static u32 vfp_double_fcmpz(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_compare(state, dd, 0, VFP_REG_ZERO, fpscr);
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_compare(state, dd, 0, VFP_REG_ZERO, fpscr);
 }
 
 static u32 vfp_double_fcmpez(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_compare(state, dd, 1, VFP_REG_ZERO, fpscr);
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_compare(state, dd, 1, VFP_REG_ZERO, fpscr);
+}
+
+u32 vfp_double_fcvtsinterncutting(ARMul_State* state, int sd, struct vfp_double* dm, u32 fpscr) //ichfly for internal use only
+{
+    struct vfp_single vsd;
+    int tm;
+    u32 exceptions = 0;
+
+    pr_debug("In %s\n", __FUNCTION__);
+
+    tm = vfp_double_type(dm);
+
+    /*
+    * If we have a signalling NaN, signal invalid operation.
+    */
+    if (tm == VFP_SNAN)
+        exceptions = FPSCR_IOC;
+
+    if (tm & VFP_DENORMAL)
+        vfp_double_normalise_denormal(dm);
+
+    vsd.sign = dm->sign;
+    vsd.significand = vfp_hi64to32jamming(dm->significand);
+
+    /*
+    * If we have an infinity or a NaN, the exponent must be 255
+    */
+    if (tm & (VFP_INFINITY | VFP_NAN)) {
+        vsd.exponent = 255;
+        if (tm == VFP_QNAN)
+            vsd.significand |= VFP_SINGLE_SIGNIFICAND_QNAN;
+        goto pack_nan;
+    }
+    else if (tm & VFP_ZERO)
+        vsd.exponent = 0;
+    else
+        vsd.exponent = dm->exponent - (1023 - 127);
+
+    return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fcvts");
+
+pack_nan:
+    vfp_put_float(state, vfp_single_pack(&vsd), sd);
+    return exceptions;
 }
 
 static u32 vfp_double_fcvts(ARMul_State* state, int sd, int unused, int dm, u32 fpscr)
 {
-	struct vfp_double vdm;
-	struct vfp_single vsd;
-	int tm;
-	u32 exceptions = 0;
-
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-
-	tm = vfp_double_type(&vdm);
-
-	/*
-	 * If we have a signalling NaN, signal invalid operation.
-	 */
-	if (tm == VFP_SNAN)
-		exceptions = FPSCR_IOC;
-
-	if (tm & VFP_DENORMAL)
-		vfp_double_normalise_denormal(&vdm);
-
-	vsd.sign = vdm.sign;
-	vsd.significand = vfp_hi64to32jamming(vdm.significand);
-
-	/*
-	 * If we have an infinity or a NaN, the exponent must be 255
-	 */
-	if (tm & (VFP_INFINITY|VFP_NAN)) {
-		vsd.exponent = 255;
-		if (tm == VFP_QNAN)
-			vsd.significand |= VFP_SINGLE_SIGNIFICAND_QNAN;
-		goto pack_nan;
-	} else if (tm & VFP_ZERO)
-		vsd.exponent = 0;
-	else
-		vsd.exponent = vdm.exponent - (1023 - 127);
-
-	return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fcvts");
-
- pack_nan:
-	vfp_put_float(state, vfp_single_pack(&vsd), sd);
-	return exceptions;
+    struct vfp_double vdm;
+    struct vfp_single vsd;
+    int tm;
+    u32 exceptions = 0;
+
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+
+    tm = vfp_double_type(&vdm);
+
+    /*
+     * If we have a signalling NaN, signal invalid operation.
+     */
+    if (tm == VFP_SNAN)
+        exceptions = FPSCR_IOC;
+
+    if (tm & VFP_DENORMAL)
+        vfp_double_normalise_denormal(&vdm);
+
+    vsd.sign = vdm.sign;
+    vsd.significand = vfp_hi64to32jamming(vdm.significand);
+
+    /*
+     * If we have an infinity or a NaN, the exponent must be 255
+     */
+    if (tm & (VFP_INFINITY|VFP_NAN)) {
+        vsd.exponent = 255;
+        if (tm == VFP_QNAN)
+            vsd.significand |= VFP_SINGLE_SIGNIFICAND_QNAN;
+        goto pack_nan;
+    } else if (tm & VFP_ZERO)
+        vsd.exponent = 0;
+    else
+        vsd.exponent = vdm.exponent - (1023 - 127);
+
+    return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fcvts");
+
+pack_nan:
+    vfp_put_float(state, vfp_single_pack(&vsd), sd);
+    return exceptions;
 }
 
 static u32 vfp_double_fuito(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	struct vfp_double vdm;
-	u32 m = vfp_get_float(state, dm);
+    struct vfp_double vdm;
+    u32 m = vfp_get_float(state, dm);
 
-	pr_debug("In %s\n", __FUNCTION__);
-	vdm.sign = 0;
-	vdm.exponent = 1023 + 63 - 1;
-	vdm.significand = (u64)m;
+    pr_debug("In %s\n", __FUNCTION__);
+    vdm.sign = 0;
+    vdm.exponent = 1023 + 63 - 1;
+    vdm.significand = (u64)m;
 
-	return vfp_double_normaliseround(state, dd, &vdm, fpscr, 0, "fuito");
+    return vfp_double_normaliseround(state, dd, &vdm, fpscr, 0, "fuito");
 }
 
 static u32 vfp_double_fsito(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	struct vfp_double vdm;
-	u32 m = vfp_get_float(state, dm);
+    struct vfp_double vdm;
+    u32 m = vfp_get_float(state, dm);
 
-	pr_debug("In %s\n", __FUNCTION__);
-	vdm.sign = (m & 0x80000000) >> 16;
-	vdm.exponent = 1023 + 63 - 1;
-	vdm.significand = vdm.sign ? -m : m;
+    pr_debug("In %s\n", __FUNCTION__);
+    vdm.sign = (m & 0x80000000) >> 16;
+    vdm.exponent = 1023 + 63 - 1;
+    vdm.significand = vdm.sign ? -m : m;
 
-	return vfp_double_normaliseround(state, dd, &vdm, fpscr, 0, "fsito");
+    return vfp_double_normaliseround(state, dd, &vdm, fpscr, 0, "fsito");
 }
 
 static u32 vfp_double_ftoui(ARMul_State* state, int sd, int unused, int dm, u32 fpscr)
 {
-	struct vfp_double vdm;
-	u32 d, exceptions = 0;
-	int rmode = fpscr & FPSCR_RMODE_MASK;
-	int tm;
-
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-
-	/*
-	 * Do we have a denormalised number?
-	 */
-	tm = vfp_double_type(&vdm);
-	if (tm & VFP_DENORMAL)
-		exceptions |= FPSCR_IDC;
-
-	if (tm & VFP_NAN)
-		vdm.sign = 0;
-
-	if (vdm.exponent >= 1023 + 32) {
-		d = vdm.sign ? 0 : 0xffffffff;
-		exceptions = FPSCR_IOC;
-	} else if (vdm.exponent >= 1023 - 1) {
-		int shift = 1023 + 63 - vdm.exponent;
-		u64 rem, incr = 0;
-
-		/*
-		 * 2^0 <= m < 2^32-2^8
-		 */
-		d = (vdm.significand << 1) >> shift;
-		rem = vdm.significand << (65 - shift);
-
-		if (rmode == FPSCR_ROUND_NEAREST) {
-			incr = 0x8000000000000000ULL;
-			if ((d & 1) == 0)
-				incr -= 1;
-		} else if (rmode == FPSCR_ROUND_TOZERO) {
-			incr = 0;
-		} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vdm.sign != 0)) {
-			incr = ~0ULL;
-		}
-
-		if ((rem + incr) < rem) {
-			if (d < 0xffffffff)
-				d += 1;
-			else
-				exceptions |= FPSCR_IOC;
-		}
-
-		if (d && vdm.sign) {
-			d = 0;
-			exceptions |= FPSCR_IOC;
-		} else if (rem)
-			exceptions |= FPSCR_IXC;
-	} else {
-		d = 0;
-		if (vdm.exponent | vdm.significand) {
-			exceptions |= FPSCR_IXC;
-			if (rmode == FPSCR_ROUND_PLUSINF && vdm.sign == 0)
-				d = 1;
-			else if (rmode == FPSCR_ROUND_MINUSINF && vdm.sign) {
-				d = 0;
-				exceptions |= FPSCR_IOC;
-			}
-		}
-	}
-
-	pr_debug("VFP: ftoui: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
-
-	vfp_put_float(state, d, sd);
-
-	return exceptions;
+    struct vfp_double vdm;
+    u32 d, exceptions = 0;
+    int rmode = fpscr & FPSCR_RMODE_MASK;
+    int tm;
+
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+
+    /*
+     * Do we have a denormalised number?
+     */
+    tm = vfp_double_type(&vdm);
+    if (tm & VFP_DENORMAL)
+        exceptions |= FPSCR_IDC;
+
+    if (tm & VFP_NAN)
+        vdm.sign = 0;
+
+    if (vdm.exponent >= 1023 + 32) {
+        d = vdm.sign ? 0 : 0xffffffff;
+        exceptions = FPSCR_IOC;
+    } else if (vdm.exponent >= 1023 - 1) {
+        int shift = 1023 + 63 - vdm.exponent;
+        u64 rem, incr = 0;
+
+        /*
+         * 2^0 <= m < 2^32-2^8
+         */
+        d = (ARMword)((vdm.significand << 1) >> shift);
+        rem = vdm.significand << (65 - shift);
+
+        if (rmode == FPSCR_ROUND_NEAREST) {
+            incr = 0x8000000000000000ULL;
+            if ((d & 1) == 0)
+                incr -= 1;
+        } else if (rmode == FPSCR_ROUND_TOZERO) {
+            incr = 0;
+        } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vdm.sign != 0)) {
+            incr = ~0ULL;
+        }
+
+        if ((rem + incr) < rem) {
+            if (d < 0xffffffff)
+                d += 1;
+            else
+                exceptions |= FPSCR_IOC;
+        }
+
+        if (d && vdm.sign) {
+            d = 0;
+            exceptions |= FPSCR_IOC;
+        } else if (rem)
+            exceptions |= FPSCR_IXC;
+    } else {
+        d = 0;
+        if (vdm.exponent | vdm.significand) {
+            exceptions |= FPSCR_IXC;
+            if (rmode == FPSCR_ROUND_PLUSINF && vdm.sign == 0)
+                d = 1;
+            else if (rmode == FPSCR_ROUND_MINUSINF && vdm.sign) {
+                d = 0;
+                exceptions |= FPSCR_IOC;
+            }
+        }
+    }
+
+    pr_debug("VFP: ftoui: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
+
+    vfp_put_float(state, d, sd);
+
+    return exceptions;
 }
 
 static u32 vfp_double_ftouiz(ARMul_State* state, int sd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_double_ftoui(state, sd, unused, dm, FPSCR_ROUND_TOZERO);
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_double_ftoui(state, sd, unused, dm, FPSCR_ROUND_TOZERO);
 }
 
 static u32 vfp_double_ftosi(ARMul_State* state, int sd, int unused, int dm, u32 fpscr)
 {
-	struct vfp_double vdm;
-	u32 d, exceptions = 0;
-	int rmode = fpscr & FPSCR_RMODE_MASK;
-	int tm;
-
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-	vfp_double_dump("VDM", &vdm);
-
-	/*
-	 * Do we have denormalised number?
-	 */
-	tm = vfp_double_type(&vdm);
-	if (tm & VFP_DENORMAL)
-		exceptions |= FPSCR_IDC;
-
-	if (tm & VFP_NAN) {
-		d = 0;
-		exceptions |= FPSCR_IOC;
-	} else if (vdm.exponent >= 1023 + 32) {
-		d = 0x7fffffff;
-		if (vdm.sign)
-			d = ~d;
-		exceptions |= FPSCR_IOC;
-	} else if (vdm.exponent >= 1023 - 1) {
-		int shift = 1023 + 63 - vdm.exponent;	/* 58 */
-		u64 rem, incr = 0;
-
-		d = (vdm.significand << 1) >> shift;
-		rem = vdm.significand << (65 - shift);
-
-		if (rmode == FPSCR_ROUND_NEAREST) {
-			incr = 0x8000000000000000ULL;
-			if ((d & 1) == 0)
-				incr -= 1;
-		} else if (rmode == FPSCR_ROUND_TOZERO) {
-			incr = 0;
-		} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vdm.sign != 0)) {
-			incr = ~0ULL;
-		}
-
-		if ((rem + incr) < rem && d < 0xffffffff)
-			d += 1;
-		if (d > 0x7fffffff + (vdm.sign != 0)) {
-			d = 0x7fffffff + (vdm.sign != 0);
-			exceptions |= FPSCR_IOC;
-		} else if (rem)
-			exceptions |= FPSCR_IXC;
-
-		if (vdm.sign)
-			d = -d;
-	} else {
-		d = 0;
-		if (vdm.exponent | vdm.significand) {
-			exceptions |= FPSCR_IXC;
-			if (rmode == FPSCR_ROUND_PLUSINF && vdm.sign == 0)
-				d = 1;
-			else if (rmode == FPSCR_ROUND_MINUSINF && vdm.sign)
-				d = -1;
-		}
-	}
-
-	pr_debug("VFP: ftosi: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
-
-	vfp_put_float(state, (s32)d, sd);
-
-	return exceptions;
+    struct vfp_double vdm;
+    u32 d, exceptions = 0;
+    int rmode = fpscr & FPSCR_RMODE_MASK;
+    int tm;
+
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+    vfp_double_dump("VDM", &vdm);
+
+    /*
+     * Do we have denormalised number?
+     */
+    tm = vfp_double_type(&vdm);
+    if (tm & VFP_DENORMAL)
+        exceptions |= FPSCR_IDC;
+
+    if (tm & VFP_NAN) {
+        d = 0;
+        exceptions |= FPSCR_IOC;
+    } else if (vdm.exponent >= 1023 + 32) {
+        d = 0x7fffffff;
+        if (vdm.sign)
+            d = ~d;
+        exceptions |= FPSCR_IOC;
+    } else if (vdm.exponent >= 1023 - 1) {
+        int shift = 1023 + 63 - vdm.exponent;	/* 58 */
+        u64 rem, incr = 0;
+
+        d = (ARMword)((vdm.significand << 1) >> shift);
+        rem = vdm.significand << (65 - shift);
+
+        if (rmode == FPSCR_ROUND_NEAREST) {
+            incr = 0x8000000000000000ULL;
+            if ((d & 1) == 0)
+                incr -= 1;
+        } else if (rmode == FPSCR_ROUND_TOZERO) {
+            incr = 0;
+        } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vdm.sign != 0)) {
+            incr = ~0ULL;
+        }
+
+        if ((rem + incr) < rem && d < 0xffffffff)
+            d += 1;
+        if (d > (0x7fffffff + (vdm.sign != 0))) {
+            d = (0x7fffffff + (vdm.sign != 0));
+            exceptions |= FPSCR_IOC;
+        } else if (rem)
+            exceptions |= FPSCR_IXC;
+
+        if (vdm.sign)
+            d = -d;
+    } else {
+        d = 0;
+        if (vdm.exponent | vdm.significand) {
+            exceptions |= FPSCR_IXC;
+            if (rmode == FPSCR_ROUND_PLUSINF && vdm.sign == 0)
+                d = 1;
+            else if (rmode == FPSCR_ROUND_MINUSINF && vdm.sign)
+                d = -1;
+        }
+    }
+
+    pr_debug("VFP: ftosi: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
+
+    vfp_put_float(state, (s32)d, sd);
+
+    return exceptions;
 }
 
 static u32 vfp_double_ftosiz(ARMul_State* state, int dd, int unused, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_double_ftosi(state, dd, unused, dm, FPSCR_ROUND_TOZERO);
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_double_ftosi(state, dd, unused, dm, FPSCR_ROUND_TOZERO);
 }
 
 static struct op fops_ext[] = {
@@ -728,197 +899,195 @@ static struct op fops_ext[] = {
 
 static u32
 vfp_double_fadd_nonnumber(struct vfp_double *vdd, struct vfp_double *vdn,
-			  struct vfp_double *vdm, u32 fpscr)
+                          struct vfp_double *vdm, u32 fpscr)
 {
-	struct vfp_double *vdp;
-	u32 exceptions = 0;
-	int tn, tm;
-
-	tn = vfp_double_type(vdn);
-	tm = vfp_double_type(vdm);
-
-	if (tn & tm & VFP_INFINITY) {
-		/*
-		 * Two infinities.  Are they different signs?
-		 */
-		if (vdn->sign ^ vdm->sign) {
-			/*
-			 * different signs -> invalid
-			 */
-			exceptions = FPSCR_IOC;
-			vdp = &vfp_double_default_qnan;
-		} else {
-			/*
-			 * same signs -> valid
-			 */
-			vdp = vdn;
-		}
-	} else if (tn & VFP_INFINITY && tm & VFP_NUMBER) {
-		/*
-		 * One infinity and one number -> infinity
-		 */
-		vdp = vdn;
-	} else {
-		/*
-		 * 'n' is a NaN of some type
-		 */
-		return vfp_propagate_nan(vdd, vdn, vdm, fpscr);
-	}
-	*vdd = *vdp;
-	return exceptions;
+    struct vfp_double *vdp;
+    u32 exceptions = 0;
+    int tn, tm;
+
+    tn = vfp_double_type(vdn);
+    tm = vfp_double_type(vdm);
+
+    if (tn & tm & VFP_INFINITY) {
+        /*
+         * Two infinities.  Are they different signs?
+         */
+        if (vdn->sign ^ vdm->sign) {
+            /*
+             * different signs -> invalid
+             */
+            exceptions = FPSCR_IOC;
+            vdp = &vfp_double_default_qnan;
+        } else {
+            /*
+             * same signs -> valid
+             */
+            vdp = vdn;
+        }
+    } else if (tn & VFP_INFINITY && tm & VFP_NUMBER) {
+        /*
+         * One infinity and one number -> infinity
+         */
+        vdp = vdn;
+    } else {
+        /*
+         * 'n' is a NaN of some type
+         */
+        return vfp_propagate_nan(vdd, vdn, vdm, fpscr);
+    }
+    *vdd = *vdp;
+    return exceptions;
 }
 
-static u32
-vfp_double_add(struct vfp_double *vdd, struct vfp_double *vdn,
-	       struct vfp_double *vdm, u32 fpscr)
+u32 vfp_double_add(struct vfp_double *vdd, struct vfp_double *vdn,struct vfp_double *vdm, u32 fpscr)
 {
-	u32 exp_diff;
-	u64 m_sig;
-
-	if (vdn->significand & (1ULL << 63) ||
-	    vdm->significand & (1ULL << 63)) {
-		pr_info("VFP: bad FP values\n");
-		vfp_double_dump("VDN", vdn);
-		vfp_double_dump("VDM", vdm);
-	}
-
-	/*
-	 * Ensure that 'n' is the largest magnitude number.  Note that
-	 * if 'n' and 'm' have equal exponents, we do not swap them.
-	 * This ensures that NaN propagation works correctly.
-	 */
-	if (vdn->exponent < vdm->exponent) {
-		struct vfp_double *t = vdn;
-		vdn = vdm;
-		vdm = t;
-	}
-
-	/*
-	 * Is 'n' an infinity or a NaN?  Note that 'm' may be a number,
-	 * infinity or a NaN here.
-	 */
-	if (vdn->exponent == 2047)
-		return vfp_double_fadd_nonnumber(vdd, vdn, vdm, fpscr);
-
-	/*
-	 * We have two proper numbers, where 'vdn' is the larger magnitude.
-	 *
-	 * Copy 'n' to 'd' before doing the arithmetic.
-	 */
-	*vdd = *vdn;
-
-	/*
-	 * Align 'm' with the result.
-	 */
-	exp_diff = vdn->exponent - vdm->exponent;
-	m_sig = vfp_shiftright64jamming(vdm->significand, exp_diff);
-
-	/*
-	 * If the signs are different, we are really subtracting.
-	 */
-	if (vdn->sign ^ vdm->sign) {
-		m_sig = vdn->significand - m_sig;
-		if ((s64)m_sig < 0) {
-			vdd->sign = vfp_sign_negate(vdd->sign);
-			m_sig = -m_sig;
-		} else if (m_sig == 0) {
-			vdd->sign = (fpscr & FPSCR_RMODE_MASK) ==
-				      FPSCR_ROUND_MINUSINF ? 0x8000 : 0;
-		}
-	} else {
-		m_sig += vdn->significand;
-	}
-	vdd->significand = m_sig;
-
-	return 0;
+    u32 exp_diff;
+    u64 m_sig;
+
+    if (vdn->significand & (1ULL << 63) ||
+            vdm->significand & (1ULL << 63)) {
+        pr_info("VFP: bad FP values in %s\n", __func__);
+        vfp_double_dump("VDN", vdn);
+        vfp_double_dump("VDM", vdm);
+    }
+
+    /*
+     * Ensure that 'n' is the largest magnitude number.  Note that
+     * if 'n' and 'm' have equal exponents, we do not swap them.
+     * This ensures that NaN propagation works correctly.
+     */
+    if (vdn->exponent < vdm->exponent) {
+        struct vfp_double *t = vdn;
+        vdn = vdm;
+        vdm = t;
+    }
+
+    /*
+     * Is 'n' an infinity or a NaN?  Note that 'm' may be a number,
+     * infinity or a NaN here.
+     */
+    if (vdn->exponent == 2047)
+        return vfp_double_fadd_nonnumber(vdd, vdn, vdm, fpscr);
+
+    /*
+     * We have two proper numbers, where 'vdn' is the larger magnitude.
+     *
+     * Copy 'n' to 'd' before doing the arithmetic.
+     */
+    *vdd = *vdn;
+
+    /*
+     * Align 'm' with the result.
+     */
+    exp_diff = vdn->exponent - vdm->exponent;
+    m_sig = vfp_shiftright64jamming(vdm->significand, exp_diff);
+
+    /*
+     * If the signs are different, we are really subtracting.
+     */
+    if (vdn->sign ^ vdm->sign) {
+        m_sig = vdn->significand - m_sig;
+        if ((s64)m_sig < 0) {
+            vdd->sign = vfp_sign_negate(vdd->sign);
+            m_sig = -m_sig;
+        } else if (m_sig == 0) {
+            vdd->sign = (fpscr & FPSCR_RMODE_MASK) ==
+                        FPSCR_ROUND_MINUSINF ? 0x8000 : 0;
+        }
+    } else {
+        m_sig += vdn->significand;
+    }
+    vdd->significand = m_sig;
+
+    return 0;
 }
 
-static u32
+u32
 vfp_double_multiply(struct vfp_double *vdd, struct vfp_double *vdn,
-		    struct vfp_double *vdm, u32 fpscr)
+                    struct vfp_double *vdm, u32 fpscr)
 {
-	vfp_double_dump("VDN", vdn);
-	vfp_double_dump("VDM", vdm);
-
-	/*
-	 * Ensure that 'n' is the largest magnitude number.  Note that
-	 * if 'n' and 'm' have equal exponents, we do not swap them.
-	 * This ensures that NaN propagation works correctly.
-	 */
-	if (vdn->exponent < vdm->exponent) {
-		struct vfp_double *t = vdn;
-		vdn = vdm;
-		vdm = t;
-		pr_debug("VFP: swapping M <-> N\n");
-	}
-
-	vdd->sign = vdn->sign ^ vdm->sign;
-
-	/*
-	 * If 'n' is an infinity or NaN, handle it.  'm' may be anything.
-	 */
-	if (vdn->exponent == 2047) {
-		if (vdn->significand || (vdm->exponent == 2047 && vdm->significand))
-			return vfp_propagate_nan(vdd, vdn, vdm, fpscr);
-		if ((vdm->exponent | vdm->significand) == 0) {
-			*vdd = vfp_double_default_qnan;
-			return FPSCR_IOC;
-		}
-		vdd->exponent = vdn->exponent;
-		vdd->significand = 0;
-		return 0;
-	}
-
-	/*
-	 * If 'm' is zero, the result is always zero.  In this case,
-	 * 'n' may be zero or a number, but it doesn't matter which.
-	 */
-	if ((vdm->exponent | vdm->significand) == 0) {
-		vdd->exponent = 0;
-		vdd->significand = 0;
-		return 0;
-	}
-
-	/*
-	 * We add 2 to the destination exponent for the same reason
-	 * as the addition case - though this time we have +1 from
-	 * each input operand.
-	 */
-	vdd->exponent = vdn->exponent + vdm->exponent - 1023 + 2;
-	vdd->significand = vfp_hi64multiply64(vdn->significand, vdm->significand);
-
-	vfp_double_dump("VDD", vdd);
-	return 0;
+    vfp_double_dump("VDN", vdn);
+    vfp_double_dump("VDM", vdm);
+
+    /*
+     * Ensure that 'n' is the largest magnitude number.  Note that
+     * if 'n' and 'm' have equal exponents, we do not swap them.
+     * This ensures that NaN propagation works correctly.
+     */
+    if (vdn->exponent < vdm->exponent) {
+        struct vfp_double *t = vdn;
+        vdn = vdm;
+        vdm = t;
+        pr_debug("VFP: swapping M <-> N\n");
+    }
+
+    vdd->sign = vdn->sign ^ vdm->sign;
+
+    /*
+     * If 'n' is an infinity or NaN, handle it.  'm' may be anything.
+     */
+    if (vdn->exponent == 2047) {
+        if (vdn->significand || (vdm->exponent == 2047 && vdm->significand))
+            return vfp_propagate_nan(vdd, vdn, vdm, fpscr);
+        if ((vdm->exponent | vdm->significand) == 0) {
+            *vdd = vfp_double_default_qnan;
+            return FPSCR_IOC;
+        }
+        vdd->exponent = vdn->exponent;
+        vdd->significand = 0;
+        return 0;
+    }
+
+    /*
+     * If 'm' is zero, the result is always zero.  In this case,
+     * 'n' may be zero or a number, but it doesn't matter which.
+     */
+    if ((vdm->exponent | vdm->significand) == 0) {
+        vdd->exponent = 0;
+        vdd->significand = 0;
+        return 0;
+    }
+
+    /*
+     * We add 2 to the destination exponent for the same reason
+     * as the addition case - though this time we have +1 from
+     * each input operand.
+     */
+    vdd->exponent = vdn->exponent + vdm->exponent - 1023 + 2;
+    vdd->significand = vfp_hi64multiply64(vdn->significand, vdm->significand);
+
+    vfp_double_dump("VDD", vdd);
+    return 0;
 }
 
 #define NEG_MULTIPLY	(1 << 0)
 #define NEG_SUBTRACT	(1 << 1)
 
 static u32
-vfp_double_multiply_accumulate(ARMul_State* state, int dd, int dn, int dm, u32 fpscr, u32 negate, const char *func)
+vfp_double_multiply_accumulate(ARMul_State* state, int dd, int dn, int dm, u32 fpscr, u32 negate, char *func)
 {
-	struct vfp_double vdd, vdp, vdn, vdm;
-	u32 exceptions;
+    struct vfp_double vdd, vdp, vdn, vdm;
+    u32 exceptions;
 
-	vfp_double_unpack(&vdn, vfp_get_double(state, dn));
-	if (vdn.exponent == 0 && vdn.significand)
-		vfp_double_normalise_denormal(&vdn);
+    vfp_double_unpack(&vdn, vfp_get_double(state, dn));
+    if (vdn.exponent == 0 && vdn.significand)
+        vfp_double_normalise_denormal(&vdn);
 
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-	if (vdm.exponent == 0 && vdm.significand)
-		vfp_double_normalise_denormal(&vdm);
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+    if (vdm.exponent == 0 && vdm.significand)
+        vfp_double_normalise_denormal(&vdm);
 
-	exceptions = vfp_double_multiply(&vdp, &vdn, &vdm, fpscr);
-	if (negate & NEG_MULTIPLY)
-		vdp.sign = vfp_sign_negate(vdp.sign);
+    exceptions = vfp_double_multiply(&vdp, &vdn, &vdm, fpscr);
+    if (negate & NEG_MULTIPLY)
+        vdp.sign = vfp_sign_negate(vdp.sign);
 
-	vfp_double_unpack(&vdn, vfp_get_double(state, dd));
-	if (negate & NEG_SUBTRACT)
-		vdn.sign = vfp_sign_negate(vdn.sign);
+    vfp_double_unpack(&vdn, vfp_get_double(state, dd));
+    if (negate & NEG_SUBTRACT)
+        vdn.sign = vfp_sign_negate(vdn.sign);
 
-	exceptions |= vfp_double_add(&vdd, &vdn, &vdp, fpscr);
+    exceptions |= vfp_double_add(&vdd, &vdn, &vdp, fpscr);
 
-	return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, func);
+    return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, func);
 }
 
 /*
@@ -930,8 +1099,8 @@ vfp_double_multiply_accumulate(ARMul_State* state, int dd, int dn, int dm, u32 f
  */
 static u32 vfp_double_fmac(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_double_multiply_accumulate(state, dd, dn, dm, fpscr, 0, "fmac");
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_double_multiply_accumulate(state, dd, dn, dm, fpscr, 0, "fmac");
 }
 
 /*
@@ -939,8 +1108,8 @@ static u32 vfp_double_fmac(ARMul_State* state, int dd, int dn, int dm, u32 fpscr
  */
 static u32 vfp_double_fnmac(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_double_multiply_accumulate(state, dd, dn, dm, fpscr, NEG_MULTIPLY, "fnmac");
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_double_multiply_accumulate(state, dd, dn, dm, fpscr, NEG_MULTIPLY, "fnmac");
 }
 
 /*
@@ -948,8 +1117,8 @@ static u32 vfp_double_fnmac(ARMul_State* state, int dd, int dn, int dm, u32 fpsc
  */
 static u32 vfp_double_fmsc(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_double_multiply_accumulate(state, dd, dn, dm, fpscr, NEG_SUBTRACT, "fmsc");
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_double_multiply_accumulate(state, dd, dn, dm, fpscr, NEG_SUBTRACT, "fmsc");
 }
 
 /*
@@ -957,8 +1126,8 @@ static u32 vfp_double_fmsc(ARMul_State* state, int dd, int dn, int dm, u32 fpscr
  */
 static u32 vfp_double_fnmsc(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	pr_debug("In %s\n", __FUNCTION__);
-	return vfp_double_multiply_accumulate(state, dd, dn, dm, fpscr, NEG_SUBTRACT | NEG_MULTIPLY, "fnmsc");
+    pr_debug("In %s\n", __FUNCTION__);
+    return vfp_double_multiply_accumulate(state, dd, dn, dm, fpscr, NEG_SUBTRACT | NEG_MULTIPLY, "fnmsc");
 }
 
 /*
@@ -966,20 +1135,20 @@ static u32 vfp_double_fnmsc(ARMul_State* state, int dd, int dn, int dm, u32 fpsc
  */
 static u32 vfp_double_fmul(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	struct vfp_double vdd, vdn, vdm;
-	u32 exceptions;
+    struct vfp_double vdd, vdn, vdm;
+    u32 exceptions;
 
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_double_unpack(&vdn, vfp_get_double(state, dn));
-	if (vdn.exponent == 0 && vdn.significand)
-		vfp_double_normalise_denormal(&vdn);
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double_unpack(&vdn, vfp_get_double(state, dn));
+    if (vdn.exponent == 0 && vdn.significand)
+        vfp_double_normalise_denormal(&vdn);
 
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-	if (vdm.exponent == 0 && vdm.significand)
-		vfp_double_normalise_denormal(&vdm);
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+    if (vdm.exponent == 0 && vdm.significand)
+        vfp_double_normalise_denormal(&vdm);
 
-	exceptions = vfp_double_multiply(&vdd, &vdn, &vdm, fpscr);
-	return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fmul");
+    exceptions = vfp_double_multiply(&vdd, &vdn, &vdm, fpscr);
+    return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fmul");
 }
 
 /*
@@ -987,22 +1156,22 @@ static u32 vfp_double_fmul(ARMul_State* state, int dd, int dn, int dm, u32 fpscr
  */
 static u32 vfp_double_fnmul(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	struct vfp_double vdd, vdn, vdm;
-	u32 exceptions;
+    struct vfp_double vdd, vdn, vdm;
+    u32 exceptions;
 
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_double_unpack(&vdn, vfp_get_double(state, dn));
-	if (vdn.exponent == 0 && vdn.significand)
-		vfp_double_normalise_denormal(&vdn);
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double_unpack(&vdn, vfp_get_double(state, dn));
+    if (vdn.exponent == 0 && vdn.significand)
+        vfp_double_normalise_denormal(&vdn);
 
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-	if (vdm.exponent == 0 && vdm.significand)
-		vfp_double_normalise_denormal(&vdm);
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+    if (vdm.exponent == 0 && vdm.significand)
+        vfp_double_normalise_denormal(&vdm);
 
-	exceptions = vfp_double_multiply(&vdd, &vdn, &vdm, fpscr);
-	vdd.sign = vfp_sign_negate(vdd.sign);
+    exceptions = vfp_double_multiply(&vdd, &vdn, &vdm, fpscr);
+    vdd.sign = vfp_sign_negate(vdd.sign);
 
-	return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fnmul");
+    return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fnmul");
 }
 
 /*
@@ -1010,21 +1179,21 @@ static u32 vfp_double_fnmul(ARMul_State* state, int dd, int dn, int dm, u32 fpsc
  */
 static u32 vfp_double_fadd(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	struct vfp_double vdd, vdn, vdm;
-	u32 exceptions;
+    struct vfp_double vdd, vdn, vdm;
+    u32 exceptions;
 
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_double_unpack(&vdn, vfp_get_double(state, dn));
-	if (vdn.exponent == 0 && vdn.significand)
-		vfp_double_normalise_denormal(&vdn);
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double_unpack(&vdn, vfp_get_double(state, dn));
+    if (vdn.exponent == 0 && vdn.significand)
+        vfp_double_normalise_denormal(&vdn);
 
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-	if (vdm.exponent == 0 && vdm.significand)
-		vfp_double_normalise_denormal(&vdm);
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+    if (vdm.exponent == 0 && vdm.significand)
+        vfp_double_normalise_denormal(&vdm);
 
-	exceptions = vfp_double_add(&vdd, &vdn, &vdm, fpscr);
+    exceptions = vfp_double_add(&vdd, &vdn, &vdm, fpscr);
 
-	return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fadd");
+    return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fadd");
 }
 
 /*
@@ -1032,26 +1201,26 @@ static u32 vfp_double_fadd(ARMul_State* state, int dd, int dn, int dm, u32 fpscr
  */
 static u32 vfp_double_fsub(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	struct vfp_double vdd, vdn, vdm;
-	u32 exceptions;
+    struct vfp_double vdd, vdn, vdm;
+    u32 exceptions;
 
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_double_unpack(&vdn, vfp_get_double(state, dn));
-	if (vdn.exponent == 0 && vdn.significand)
-		vfp_double_normalise_denormal(&vdn);
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double_unpack(&vdn, vfp_get_double(state, dn));
+    if (vdn.exponent == 0 && vdn.significand)
+        vfp_double_normalise_denormal(&vdn);
 
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-	if (vdm.exponent == 0 && vdm.significand)
-		vfp_double_normalise_denormal(&vdm);
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+    if (vdm.exponent == 0 && vdm.significand)
+        vfp_double_normalise_denormal(&vdm);
 
-	/*
-	 * Subtraction is like addition, but with a negated operand.
-	 */
-	vdm.sign = vfp_sign_negate(vdm.sign);
+    /*
+     * Subtraction is like addition, but with a negated operand.
+     */
+    vdm.sign = vfp_sign_negate(vdm.sign);
 
-	exceptions = vfp_double_add(&vdd, &vdn, &vdm, fpscr);
+    exceptions = vfp_double_add(&vdd, &vdn, &vdm, fpscr);
 
-	return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fsub");
+    return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fsub");
 }
 
 /*
@@ -1059,120 +1228,120 @@ static u32 vfp_double_fsub(ARMul_State* state, int dd, int dn, int dm, u32 fpscr
  */
 static u32 vfp_double_fdiv(ARMul_State* state, int dd, int dn, int dm, u32 fpscr)
 {
-	struct vfp_double vdd, vdn, vdm;
-	u32 exceptions = 0;
-	int tm, tn;
-
-	pr_debug("In %s\n", __FUNCTION__);
-	vfp_double_unpack(&vdn, vfp_get_double(state, dn));
-	vfp_double_unpack(&vdm, vfp_get_double(state, dm));
-
-	vdd.sign = vdn.sign ^ vdm.sign;
-
-	tn = vfp_double_type(&vdn);
-	tm = vfp_double_type(&vdm);
-
-	/*
-	 * Is n a NAN?
-	 */
-	if (tn & VFP_NAN)
-		goto vdn_nan;
-
-	/*
-	 * Is m a NAN?
-	 */
-	if (tm & VFP_NAN)
-		goto vdm_nan;
-
-	/*
-	 * If n and m are infinity, the result is invalid
-	 * If n and m are zero, the result is invalid
-	 */
-	if (tm & tn & (VFP_INFINITY|VFP_ZERO))
-		goto invalid;
-
-	/*
-	 * If n is infinity, the result is infinity
-	 */
-	if (tn & VFP_INFINITY)
-		goto infinity;
-
-	/*
-	 * If m is zero, raise div0 exceptions
-	 */
-	if (tm & VFP_ZERO)
-		goto divzero;
-
-	/*
-	 * If m is infinity, or n is zero, the result is zero
-	 */
-	if (tm & VFP_INFINITY || tn & VFP_ZERO)
-		goto zero;
-
-	if (tn & VFP_DENORMAL)
-		vfp_double_normalise_denormal(&vdn);
-	if (tm & VFP_DENORMAL)
-		vfp_double_normalise_denormal(&vdm);
-
-	/*
-	 * Ok, we have two numbers, we can perform division.
-	 */	
-	vdd.exponent = vdn.exponent - vdm.exponent + 1023 - 1;
-	vdm.significand <<= 1;
-	if (vdm.significand <= (2 * vdn.significand)) {
-		vdn.significand >>= 1;
-		vdd.exponent++;
-	}
-	vdd.significand = vfp_estimate_div128to64(vdn.significand, 0, vdm.significand);
-	if ((vdd.significand & 0x1ff) <= 2) {
-		u64 termh, terml, remh, reml;
-		mul64to128(&termh, &terml, vdm.significand, vdd.significand);
-		sub128(&remh, &reml, vdn.significand, 0, termh, terml);
-		while ((s64)remh < 0) {
-			vdd.significand -= 1;
-			add128(&remh, &reml, remh, reml, 0, vdm.significand);
-		}
-		vdd.significand |= (reml != 0);
-	}
-	return vfp_double_normaliseround(state, dd, &vdd, fpscr, 0, "fdiv");
-
- vdn_nan:
-	exceptions = vfp_propagate_nan(&vdd, &vdn, &vdm, fpscr);
- pack:
-	vfp_put_double(state, vfp_double_pack(&vdd), dd);
-	return exceptions;
-
- vdm_nan:
-	exceptions = vfp_propagate_nan(&vdd, &vdm, &vdn, fpscr);
-	goto pack;
-
- zero:
-	vdd.exponent = 0;
-	vdd.significand = 0;
-	goto pack;
-
- divzero:
-	exceptions = FPSCR_DZC;
- infinity:
-	vdd.exponent = 2047;
-	vdd.significand = 0;
-	goto pack;
-
- invalid:
-	vfp_put_double(state, vfp_double_pack(&vfp_double_default_qnan), dd);
-	return FPSCR_IOC;
+    struct vfp_double vdd, vdn, vdm;
+    u32 exceptions = 0;
+    int tm, tn;
+
+    pr_debug("In %s\n", __FUNCTION__);
+    vfp_double_unpack(&vdn, vfp_get_double(state, dn));
+    vfp_double_unpack(&vdm, vfp_get_double(state, dm));
+
+    vdd.sign = vdn.sign ^ vdm.sign;
+
+    tn = vfp_double_type(&vdn);
+    tm = vfp_double_type(&vdm);
+
+    /*
+     * Is n a NAN?
+     */
+    if (tn & VFP_NAN)
+        goto vdn_nan;
+
+    /*
+     * Is m a NAN?
+     */
+    if (tm & VFP_NAN)
+        goto vdm_nan;
+
+    /*
+     * If n and m are infinity, the result is invalid
+     * If n and m are zero, the result is invalid
+     */
+    if (tm & tn & (VFP_INFINITY|VFP_ZERO))
+        goto invalid;
+
+    /*
+     * If n is infinity, the result is infinity
+     */
+    if (tn & VFP_INFINITY)
+        goto infinity;
+
+    /*
+     * If m is zero, raise div0 exceptions
+     */
+    if (tm & VFP_ZERO)
+        goto divzero;
+
+    /*
+     * If m is infinity, or n is zero, the result is zero
+     */
+    if (tm & VFP_INFINITY || tn & VFP_ZERO)
+        goto zero;
+
+    if (tn & VFP_DENORMAL)
+        vfp_double_normalise_denormal(&vdn);
+    if (tm & VFP_DENORMAL)
+        vfp_double_normalise_denormal(&vdm);
+
+    /*
+     * Ok, we have two numbers, we can perform division.
+     */
+    vdd.exponent = vdn.exponent - vdm.exponent + 1023 - 1;
+    vdm.significand <<= 1;
+    if (vdm.significand <= (2 * vdn.significand)) {
+        vdn.significand >>= 1;
+        vdd.exponent++;
+    }
+    vdd.significand = vfp_estimate_div128to64(vdn.significand, 0, vdm.significand);
+    if ((vdd.significand & 0x1ff) <= 2) {
+        u64 termh, terml, remh, reml;
+        mul64to128(&termh, &terml, vdm.significand, vdd.significand);
+        sub128(&remh, &reml, vdn.significand, 0, termh, terml);
+        while ((s64)remh < 0) {
+            vdd.significand -= 1;
+            add128(&remh, &reml, remh, reml, 0, vdm.significand);
+        }
+        vdd.significand |= (reml != 0);
+    }
+    return vfp_double_normaliseround(state, dd, &vdd, fpscr, 0, "fdiv");
+
+vdn_nan:
+    exceptions = vfp_propagate_nan(&vdd, &vdn, &vdm, fpscr);
+pack:
+    vfp_put_double(state, vfp_double_pack(&vdd), dd);
+    return exceptions;
+
+vdm_nan:
+    exceptions = vfp_propagate_nan(&vdd, &vdm, &vdn, fpscr);
+    goto pack;
+
+zero:
+    vdd.exponent = 0;
+    vdd.significand = 0;
+    goto pack;
+
+divzero:
+    exceptions = FPSCR_DZC;
+infinity:
+    vdd.exponent = 2047;
+    vdd.significand = 0;
+    goto pack;
+
+invalid:
+    vfp_put_double(state, vfp_double_pack(&vfp_double_default_qnan), dd);
+    return FPSCR_IOC;
 }
 
 static struct op fops[] = {
-	{ vfp_double_fmac,  0 },
-	{ vfp_double_fmsc,  0 },
-	{ vfp_double_fmul,  0 },
-	{ vfp_double_fadd,  0 },
-	{ vfp_double_fnmac, 0 },
-	{ vfp_double_fnmsc, 0 },
-	{ vfp_double_fnmul, 0 },
-	{ vfp_double_fsub,  0 },
-	{ vfp_double_fdiv,  0 },
+    { vfp_double_fmac,  0 },
+    { vfp_double_fmsc,  0 },
+    { vfp_double_fmul,  0 },
+    { vfp_double_fadd,  0 },
+    { vfp_double_fnmac, 0 },
+    { vfp_double_fnmsc, 0 },
+    { vfp_double_fnmul, 0 },
+    { vfp_double_fsub,  0 },
+    { vfp_double_fdiv,  0 },
 };
 
 #define FREG_BANK(x)	((x) & 0x0c)
@@ -1180,84 +1349,84 @@ static struct op fops[] = {
 
 u32 vfp_double_cpdo(ARMul_State* state, u32 inst, u32 fpscr)
 {
-	u32 op = inst & FOP_MASK;
-	u32 exceptions = 0;
-	unsigned int dest;
-	unsigned int dn = vfp_get_dn(inst);
-	unsigned int dm;
-	unsigned int vecitr, veclen, vecstride;
-	struct op *fop;
-
-	pr_debug("In %s\n", __FUNCTION__);
-	vecstride = (1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK));
-
-	fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)];
-
-	/*
-	 * fcvtds takes an sN register number as destination, not dN.
-	 * It also always operates on scalars.
-	 */
-	if (fop->flags & OP_SD)
-		dest = vfp_get_sd(inst);
-	else
-		dest = vfp_get_dd(inst);
-
-	/*
-	 * f[us]ito takes a sN operand, not a dN operand.
-	 */
-	if (fop->flags & OP_SM)
-		dm = vfp_get_sm(inst);
-	else
-		dm = vfp_get_dm(inst);
-
-	/*
-	 * If destination bank is zero, vector length is always '1'.
-	 * ARM DDI0100F C5.1.3, C5.3.2.
-	 */
-	if ((fop->flags & OP_SCALAR) || (FREG_BANK(dest) == 0))
-		veclen = 0;
-	else
-		veclen = fpscr & FPSCR_LENGTH_MASK;
-
-	pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
-		 (veclen >> FPSCR_LENGTH_BIT) + 1);
-
-	if (!fop->fn) {
-		printf("VFP: could not find double op %d\n", FEXT_TO_IDX(inst));
-		goto invalid;
-	}
-
-	for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
-		u32 except;
-		char type;
-
-		type = fop->flags & OP_SD ? 's' : 'd';
-		if (op == FOP_EXT)
-			pr_debug("VFP: itr%d (%c%u) = op[%u] (d%u)\n",
-				 vecitr >> FPSCR_LENGTH_BIT,
-				 type, dest, dn, dm);
-		else
-			pr_debug("VFP: itr%d (%c%u) = (d%u) op[%u] (d%u)\n",
-				 vecitr >> FPSCR_LENGTH_BIT,
-				 type, dest, dn, FOP_TO_IDX(op), dm);
-
-		except = fop->fn(state, dest, dn, dm, fpscr);
-		pr_debug("VFP: itr%d: exceptions=%08x\n",
-			 vecitr >> FPSCR_LENGTH_BIT, except);
-
-		exceptions |= except;
-
-		/*
-		 * CHECK: It appears to be undefined whether we stop when
-		 * we encounter an exception.  We continue.
-		 */
-		dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 3);
-		dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 3);
-		if (FREG_BANK(dm) != 0)
-			dm = FREG_BANK(dm) + ((FREG_IDX(dm) + vecstride) & 3);
-	}
-	return exceptions;
-
- invalid:
-	return ~0;
+    u32 op = inst & FOP_MASK;
+    u32 exceptions = 0;
+    unsigned int dest;
+    unsigned int dn = vfp_get_dn(inst);
+    unsigned int dm;
+    unsigned int vecitr, veclen, vecstride;
+    struct op *fop;
+
+    pr_debug("In %s\n", __FUNCTION__);
+    vecstride = (1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK));
+
+    fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)];
+
+    /*
+     * fcvtds takes an sN register number as destination, not dN.
+     * It also always operates on scalars.
+     */
+    if (fop->flags & OP_SD)
+        dest = vfp_get_sd(inst);
+    else
+        dest = vfp_get_dd(inst);
+
+    /*
+     * f[us]ito takes a sN operand, not a dN operand.
+     */
+    if (fop->flags & OP_SM)
+        dm = vfp_get_sm(inst);
+    else
+        dm = vfp_get_dm(inst);
+
+    /*
+     * If destination bank is zero, vector length is always '1'.
+     * ARM DDI0100F C5.1.3, C5.3.2.
+     */
+    if ((fop->flags & OP_SCALAR) || (FREG_BANK(dest) == 0))
+        veclen = 0;
+    else
+        veclen = fpscr & FPSCR_LENGTH_MASK;
+
+    pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
+             (veclen >> FPSCR_LENGTH_BIT) + 1);
+
+    if (!fop->fn) {
+        printf("VFP: could not find double op %d\n", FEXT_TO_IDX(inst));
+        goto invalid;
+    }
+
+    for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
+        u32 except;
+        char type;
+
+        type = fop->flags & OP_SD ? 's' : 'd';
+        if (op == FOP_EXT)
+            pr_debug("VFP: itr%d (%c%u) = op[%u] (d%u)\n",
+                     vecitr >> FPSCR_LENGTH_BIT,
+                     type, dest, dn, dm);
+        else
+            pr_debug("VFP: itr%d (%c%u) = (d%u) op[%u] (d%u)\n",
+                     vecitr >> FPSCR_LENGTH_BIT,
+                     type, dest, dn, FOP_TO_IDX(op), dm);
+
+        except = fop->fn(state, dest, dn, dm, fpscr);
+        pr_debug("VFP: itr%d: exceptions=%08x\n",
+                 vecitr >> FPSCR_LENGTH_BIT, except);
+
+        exceptions |= except;
+
+        /*
+         * CHECK: It appears to be undefined whether we stop when
+         * we encounter an exception.  We continue.
+         */
+        dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 3);
+        dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 3);
+        if (FREG_BANK(dm) != 0)
+            dm = FREG_BANK(dm) + ((FREG_IDX(dm) + vecstride) & 3);
+    }
+    return exceptions;
+
+invalid:
+    return ~0;
 }
diff --git a/src/core/arm/skyeye_common/vfp/vfpsingle.cpp b/src/core/arm/skyeye_common/vfp/vfpsingle.cpp
index 8bcbd4fe..07d0c1f4 100644
--- a/src/core/arm/skyeye_common/vfp/vfpsingle.cpp
+++ b/src/core/arm/skyeye_common/vfp/vfpsingle.cpp
@@ -56,167 +56,167 @@
 #include "core/arm/skyeye_common/vfp/vfp.h"
 
 static struct vfp_single vfp_single_default_qnan = {
-	//.exponent	= 255,
-	//.sign		= 0,
-	//.significand	= VFP_SINGLE_SIGNIFICAND_QNAN,
+    255,
+    0,
+    VFP_SINGLE_SIGNIFICAND_QNAN,
 };
 
 static void vfp_single_dump(const char *str, struct vfp_single *s)
 {
-	pr_debug("VFP: %s: sign=%d exponent=%d significand=%08x\n",
-		 str, s->sign != 0, s->exponent, s->significand);
+    pr_debug("VFP: %s: sign=%d exponent=%d significand=%08x\n",
+             str, s->sign != 0, s->exponent, s->significand);
 }
 
 static void vfp_single_normalise_denormal(struct vfp_single *vs)
 {
-	int bits = 31 - vfp_fls(vs->significand);
+    int bits = 31 - fls(vs->significand);
 
-	vfp_single_dump("normalise_denormal: in", vs);
+    vfp_single_dump("normalise_denormal: in", vs);
 
-	if (bits) {
-		vs->exponent -= bits - 1;
-		vs->significand <<= bits;
-	}
+    if (bits) {
+        vs->exponent -= bits - 1;
+        vs->significand <<= bits;
+    }
 
-	vfp_single_dump("normalise_denormal: out", vs);
+    vfp_single_dump("normalise_denormal: out", vs);
 }
 
 
 u32 vfp_single_normaliseround(ARMul_State* state, int sd, struct vfp_single *vs, u32 fpscr, u32 exceptions, const char *func)
 {
-	u32 significand, incr, rmode;
-	int exponent, shift, underflow;
-
-	vfp_single_dump("pack: in", vs);
-
-	/*
-	 * Infinities and NaNs are a special case.
-	 */
-	if (vs->exponent == 255 && (vs->significand == 0 || exceptions))
-		goto pack;
-
-	/*
-	 * Special-case zero.
-	 */
-	if (vs->significand == 0) {
-		vs->exponent = 0;
-		goto pack;
-	}
-
-	exponent = vs->exponent;
-	significand = vs->significand;
-
-	/*
-	 * Normalise first.  Note that we shift the significand up to
-	 * bit 31, so we have VFP_SINGLE_LOW_BITS + 1 below the least
-	 * significant bit.
-	 */
-	shift = 32 - vfp_fls(significand);
-	if (shift < 32 && shift) {
-		exponent -= shift;
-		significand <<= shift;
-	}
+    u32 significand, incr, rmode;
+    int exponent, shift, underflow;
+
+    vfp_single_dump("pack: in", vs);
+
+    /*
+     * Infinities and NaNs are a special case.
+     */
+    if (vs->exponent == 255 && (vs->significand == 0 || exceptions))
+        goto pack;
+
+    /*
+     * Special-case zero.
+     */
+    if (vs->significand == 0) {
+        vs->exponent = 0;
+        goto pack;
+    }
+
+    exponent = vs->exponent;
+    significand = vs->significand;
+
+    /*
+     * Normalise first.  Note that we shift the significand up to
+     * bit 31, so we have VFP_SINGLE_LOW_BITS + 1 below the least
+     * significant bit.
+     */
+    shift = 32 - fls(significand);
+    if (shift < 32 && shift) {
+        exponent -= shift;
+        significand <<= shift;
+    }
 
 #if 1
-	vs->exponent = exponent;
-	vs->significand = significand;
-	vfp_single_dump("pack: normalised", vs);
+    vs->exponent = exponent;
+    vs->significand = significand;
+    vfp_single_dump("pack: normalised", vs);
 #endif
 
-	/*
-	 * Tiny number?
-	 */
-	underflow = exponent < 0;
-	if (underflow) {
-		significand = vfp_shiftright32jamming(significand, -exponent);
-		exponent = 0;
+    /*
+     * Tiny number?
+     */
+    underflow = exponent < 0;
+    if (underflow) {
+        significand = vfp_shiftright32jamming(significand, -exponent);
+        exponent = 0;
 #if 1
-		vs->exponent = exponent;
-		vs->significand = significand;
-		vfp_single_dump("pack: tiny number", vs);
+        vs->exponent = exponent;
+        vs->significand = significand;
+        vfp_single_dump("pack: tiny number", vs);
 #endif
-		if (!(significand & ((1 << (VFP_SINGLE_LOW_BITS + 1)) - 1)))
-			underflow = 0;
-	}
-
-	/*
-	 * Select rounding increment.
-	 */
-	incr = 0;
-	rmode = fpscr & FPSCR_RMODE_MASK;
-
-	if (rmode == FPSCR_ROUND_NEAREST) {
-		incr = 1 << VFP_SINGLE_LOW_BITS;
-		if ((significand & (1 << (VFP_SINGLE_LOW_BITS + 1))) == 0)
-			incr -= 1;
-	} else if (rmode == FPSCR_ROUND_TOZERO) {
-		incr = 0;
-	} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vs->sign != 0))
-		incr = (1 << (VFP_SINGLE_LOW_BITS + 1)) - 1;
-
-	pr_debug("VFP: rounding increment = 0x%08x\n", incr);
-
-	/*
-	 * Is our rounding going to overflow?
-	 */
-	if ((significand + incr) < significand) {
-		exponent += 1;
-		significand = (significand >> 1) | (significand & 1);
-		incr >>= 1;
+        if (!(significand & ((1 << (VFP_SINGLE_LOW_BITS + 1)) - 1)))
+            underflow = 0;
+    }
+
+    /*
+     * Select rounding increment.
+     */
+    incr = 0;
+    rmode = fpscr & FPSCR_RMODE_MASK;
+
+    if (rmode == FPSCR_ROUND_NEAREST) {
+        incr = 1 << VFP_SINGLE_LOW_BITS;
+        if ((significand & (1 << (VFP_SINGLE_LOW_BITS + 1))) == 0)
+            incr -= 1;
+    } else if (rmode == FPSCR_ROUND_TOZERO) {
+        incr = 0;
+    } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vs->sign != 0))
+        incr = (1 << (VFP_SINGLE_LOW_BITS + 1)) - 1;
+
+    pr_debug("VFP: rounding increment = 0x%08x\n", incr);
+
+    /*
+     * Is our rounding going to overflow?
+     */
+    if ((significand + incr) < significand) {
+        exponent += 1;
+        significand = (significand >> 1) | (significand & 1);
+        incr >>= 1;
 #if 1
-		vs->exponent = exponent;
-		vs->significand = significand;
-		vfp_single_dump("pack: overflow", vs);
+        vs->exponent = exponent;
+        vs->significand = significand;
+        vfp_single_dump("pack: overflow", vs);
 #endif
-	}
-
-	/*
-	 * If any of the low bits (which will be shifted out of the
-	 * number) are non-zero, the result is inexact.
-	 */
-	if (significand & ((1 << (VFP_SINGLE_LOW_BITS + 1)) - 1))
-		exceptions |= FPSCR_IXC;
-
-	/*
-	 * Do our rounding.
-	 */
-	significand += incr;
-
-	/*
-	 * Infinity?
-	 */
-	if (exponent >= 254) {
-		exceptions |= FPSCR_OFC | FPSCR_IXC;
-		if (incr == 0) {
-			vs->exponent = 253;
-			vs->significand = 0x7fffffff;
-		} else {
-			vs->exponent = 255;		/* infinity */
-			vs->significand = 0;
-		}
-	} else {
-		if (significand >> (VFP_SINGLE_LOW_BITS + 1) == 0)
-			exponent = 0;
-		if (exponent || significand > 0x80000000)
-			underflow = 0;
-		if (underflow)
-			exceptions |= FPSCR_UFC;
-		vs->exponent = exponent;
-		vs->significand = significand >> 1;
-	}
-
- pack:
-	vfp_single_dump("pack: final", vs);
-	{
-		s32 d = vfp_single_pack(vs);
+    }
+
+    /*
+     * If any of the low bits (which will be shifted out of the
+     * number) are non-zero, the result is inexact.
+     */
+    if (significand & ((1 << (VFP_SINGLE_LOW_BITS + 1)) - 1))
+        exceptions |= FPSCR_IXC;
+
+    /*
+     * Do our rounding.
+     */
+    significand += incr;
+
+    /*
+     * Infinity?
+     */
+    if (exponent >= 254) {
+        exceptions |= FPSCR_OFC | FPSCR_IXC;
+        if (incr == 0) {
+            vs->exponent = 253;
+            vs->significand = 0x7fffffff;
+        } else {
+            vs->exponent = 255;		/* infinity */
+            vs->significand = 0;
+        }
+    } else {
+        if (significand >> (VFP_SINGLE_LOW_BITS + 1) == 0)
+            exponent = 0;
+        if (exponent || significand > 0x80000000)
+            underflow = 0;
+        if (underflow)
+            exceptions |= FPSCR_UFC;
+        vs->exponent = exponent;
+        vs->significand = significand >> 1;
+    }
+
+pack:
+    vfp_single_dump("pack: final", vs);
+    {
+        s32 d = vfp_single_pack(vs);
 #if 1
-		pr_debug("VFP: %s: d(s%d)=%08x exceptions=%08x\n", func,
-			 sd, d, exceptions);
+        pr_debug("VFP: %s: d(s%d)=%08x exceptions=%08x\n", func,
+                 sd, d, exceptions);
 #endif
-		vfp_put_float(state, d, sd);
-	}
+        vfp_put_float(state, d, sd);
+    }
 
-	return exceptions;
+    return exceptions;
 }
 
 /*
@@ -225,43 +225,43 @@ u32 vfp_single_normaliseround(ARMul_State* state, int sd, struct vfp_single *vs,
  */
 static u32
 vfp_propagate_nan(struct vfp_single *vsd, struct vfp_single *vsn,
-		  struct vfp_single *vsm, u32 fpscr)
+                  struct vfp_single *vsm, u32 fpscr)
 {
-	struct vfp_single *nan;
-	int tn, tm = 0;
-
-	tn = vfp_single_type(vsn);
-
-	if (vsm)
-		tm = vfp_single_type(vsm);
-
-	if (fpscr & FPSCR_DEFAULT_NAN)
-		/*
-		 * Default NaN mode - always returns a quiet NaN
-		 */
-		nan = &vfp_single_default_qnan;
-	else {
-		/*
-		 * Contemporary mode - select the first signalling
-		 * NAN, or if neither are signalling, the first
-		 * quiet NAN.
-		 */
-		if (tn == VFP_SNAN || (tm != VFP_SNAN && tn == VFP_QNAN))
-			nan = vsn;
-		else
-			nan = vsm;
-		/*
-		 * Make the NaN quiet.
-		 */
-		nan->significand |= VFP_SINGLE_SIGNIFICAND_QNAN;
-	}
-
-	*vsd = *nan;
-
-	/*
-	 * If one was a signalling NAN, raise invalid operation.
-	 */
-	return tn == VFP_SNAN || tm == VFP_SNAN ? FPSCR_IOC : VFP_NAN_FLAG;
+    struct vfp_single *nan;
+    int tn, tm = 0;
+
+    tn = vfp_single_type(vsn);
+
+    if (vsm)
+        tm = vfp_single_type(vsm);
+
+    if (fpscr & FPSCR_DEFAULT_NAN)
+        /*
+         * Default NaN mode - always returns a quiet NaN
+         */
+        nan = &vfp_single_default_qnan;
+    else {
+        /*
+         * Contemporary mode - select the first signalling
+         * NAN, or if neither are signalling, the first
+         * quiet NAN.
+         */
+        if (tn == VFP_SNAN || (tm != VFP_SNAN && tn == VFP_QNAN))
+            nan = vsn;
+        else
+            nan = vsm;
+        /*
+         * Make the NaN quiet.
+         */
+        nan->significand |= VFP_SINGLE_SIGNIFICAND_QNAN;
+    }
+
+    *vsd = *nan;
+
+    /*
+     * If one was a signalling NAN, raise invalid operation.
+     */
+    return tn == VFP_SNAN || tm == VFP_SNAN ? FPSCR_IOC : VFP_NAN_FLAG;
 }
 
 
@@ -270,140 +270,140 @@ vfp_propagate_nan(struct vfp_single *vsd, struct vfp_single *vsn,
  */
 static u32 vfp_single_fabs(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	vfp_put_float(state, vfp_single_packed_abs(m), sd);
-	return 0;
+    vfp_put_float(state, vfp_single_packed_abs(m), sd);
+    return 0;
 }
 
 static u32 vfp_single_fcpy(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	vfp_put_float(state, m, sd);
-	return 0;
+    vfp_put_float(state, m, sd);
+    return 0;
 }
 
 static u32 vfp_single_fneg(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	vfp_put_float(state, vfp_single_packed_negate(m), sd);
-	return 0;
+    vfp_put_float(state, vfp_single_packed_negate(m), sd);
+    return 0;
 }
 
 static const u16 sqrt_oddadjust[] = {
-	0x0004, 0x0022, 0x005d, 0x00b1, 0x011d, 0x019f, 0x0236, 0x02e0,
-	0x039c, 0x0468, 0x0545, 0x0631, 0x072b, 0x0832, 0x0946, 0x0a67
+    0x0004, 0x0022, 0x005d, 0x00b1, 0x011d, 0x019f, 0x0236, 0x02e0,
+    0x039c, 0x0468, 0x0545, 0x0631, 0x072b, 0x0832, 0x0946, 0x0a67
 };
 
 static const u16 sqrt_evenadjust[] = {
-	0x0a2d, 0x08af, 0x075a, 0x0629, 0x051a, 0x0429, 0x0356, 0x029e,
-	0x0200, 0x0179, 0x0109, 0x00af, 0x0068, 0x0034, 0x0012, 0x0002
+    0x0a2d, 0x08af, 0x075a, 0x0629, 0x051a, 0x0429, 0x0356, 0x029e,
+    0x0200, 0x0179, 0x0109, 0x00af, 0x0068, 0x0034, 0x0012, 0x0002
 };
 
 u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand)
 {
-	int index;
-	u32 z, a;
-
-	if ((significand & 0xc0000000) != 0x40000000) {
-		pr_debug("VFP: estimate_sqrt: invalid significand\n");
-	}
-
-	a = significand << 1;
-	index = (a >> 27) & 15;
-	if (exponent & 1) {
-		z = 0x4000 + (a >> 17) - sqrt_oddadjust[index];
-		z = ((a / z) << 14) + (z << 15);
-		a >>= 1;
-	} else {
-		z = 0x8000 + (a >> 17) - sqrt_evenadjust[index];
-		z = a / z + z;
-		z = (z >= 0x20000) ? 0xffff8000 : (z << 15);
-		if (z <= a)
-			return (s32)a >> 1;
-	}
-	{
-		u64 v = (u64)a << 31;
-		do_div(v, z);
-		return v + (z >> 1);
-	}
+    int index;
+    u32 z, a;
+
+    if ((significand & 0xc0000000) != 0x40000000) {
+        pr_debug("VFP: estimate_sqrt: invalid significand\n");
+    }
+
+    a = significand << 1;
+    index = (a >> 27) & 15;
+    if (exponent & 1) {
+        z = 0x4000 + (a >> 17) - sqrt_oddadjust[index];
+        z = ((a / z) << 14) + (z << 15);
+        a >>= 1;
+    } else {
+        z = 0x8000 + (a >> 17) - sqrt_evenadjust[index];
+        z = a / z + z;
+        z = (z >= 0x20000) ? 0xffff8000 : (z << 15);
+        if (z <= a)
+            return (s32)a >> 1;
+    }
+    {
+        u64 v = (u64)a << 31;
+        do_div(v, z);
+        return (u32)(v + (z >> 1));
+    }
 }
 
 static u32 vfp_single_fsqrt(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	struct vfp_single vsm, vsd, *vsp;
-	int ret, tm;
-
-	vfp_single_unpack(&vsm, m);
-	tm = vfp_single_type(&vsm);
-	if (tm & (VFP_NAN|VFP_INFINITY)) {
-		vsp = &vsd;
-
-		if (tm & VFP_NAN)
-			ret = vfp_propagate_nan(vsp, &vsm, NULL, fpscr);
-		else if (vsm.sign == 0) {
- sqrt_copy:
-			vsp = &vsm;
-			ret = 0;
-		} else {
- sqrt_invalid:
-			vsp = &vfp_single_default_qnan;
-			ret = FPSCR_IOC;
-		}
-		vfp_put_float(state, vfp_single_pack(vsp), sd);
-		return ret;
-	}
-
-	/*
-	 * sqrt(+/- 0) == +/- 0
-	 */
-	if (tm & VFP_ZERO)
-		goto sqrt_copy;
-
-	/*
-	 * Normalise a denormalised number
-	 */
-	if (tm & VFP_DENORMAL)
-		vfp_single_normalise_denormal(&vsm);
-
-	/*
-	 * sqrt(<0) = invalid
-	 */
-	if (vsm.sign)
-		goto sqrt_invalid;
-
-	vfp_single_dump("sqrt", &vsm);
-
-	/*
-	 * Estimate the square root.
-	 */
-	vsd.sign = 0;
-	vsd.exponent = ((vsm.exponent - 127) >> 1) + 127;
-	vsd.significand = vfp_estimate_sqrt_significand(vsm.exponent, vsm.significand) + 2;
-
-	vfp_single_dump("sqrt estimate", &vsd);
-
-	/*
-	 * And now adjust.
-	 */
-	if ((vsd.significand & VFP_SINGLE_LOW_BITS_MASK) <= 5) {
-		if (vsd.significand < 2) {
-			vsd.significand = 0xffffffff;
-		} else {
-			u64 term;
-			s64 rem;
-			vsm.significand <<= !(vsm.exponent & 1);
-			term = (u64)vsd.significand * vsd.significand;
-			rem = ((u64)vsm.significand << 32) - term;
-
-			pr_debug("VFP: term=%016llx rem=%016llx\n", term, rem);
-
-			while (rem < 0) {
-				vsd.significand -= 1;
-				rem += ((u64)vsd.significand << 1) | 1;
-			}
-			vsd.significand |= rem != 0;
-		}
-	}
-	vsd.significand = vfp_shiftright32jamming(vsd.significand, 1);
-
-	return vfp_single_normaliseround(state, sd, &vsd, fpscr, 0, "fsqrt");
+    struct vfp_single vsm, vsd, *vsp;
+    int ret, tm;
+
+    vfp_single_unpack(&vsm, m);
+    tm = vfp_single_type(&vsm);
+    if (tm & (VFP_NAN|VFP_INFINITY)) {
+        vsp = &vsd;
+
+        if (tm & VFP_NAN)
+            ret = vfp_propagate_nan(vsp, &vsm, NULL, fpscr);
+        else if (vsm.sign == 0) {
+sqrt_copy:
+            vsp = &vsm;
+            ret = 0;
+        } else {
+sqrt_invalid:
+            vsp = &vfp_single_default_qnan;
+            ret = FPSCR_IOC;
+        }
+        vfp_put_float(state, vfp_single_pack(vsp), sd);
+        return ret;
+    }
+
+    /*
+     * sqrt(+/- 0) == +/- 0
+     */
+    if (tm & VFP_ZERO)
+        goto sqrt_copy;
+
+    /*
+     * Normalise a denormalised number
+     */
+    if (tm & VFP_DENORMAL)
+        vfp_single_normalise_denormal(&vsm);
+
+    /*
+     * sqrt(<0) = invalid
+     */
+    if (vsm.sign)
+        goto sqrt_invalid;
+
+    vfp_single_dump("sqrt", &vsm);
+
+    /*
+     * Estimate the square root.
+     */
+    vsd.sign = 0;
+    vsd.exponent = ((vsm.exponent - 127) >> 1) + 127;
+    vsd.significand = vfp_estimate_sqrt_significand(vsm.exponent, vsm.significand) + 2;
+
+    vfp_single_dump("sqrt estimate", &vsd);
+
+    /*
+     * And now adjust.
+     */
+    if ((vsd.significand & VFP_SINGLE_LOW_BITS_MASK) <= 5) {
+        if (vsd.significand < 2) {
+            vsd.significand = 0xffffffff;
+        } else {
+            u64 term;
+            s64 rem;
+            vsm.significand <<= !(vsm.exponent & 1);
+            term = (u64)vsd.significand * vsd.significand;
+            rem = ((u64)vsm.significand << 32) - term;
+
+            pr_debug("VFP: term=%016llx rem=%016llx\n", term, rem);
+
+            while (rem < 0) {
+                vsd.significand -= 1;
+                rem += ((u64)vsd.significand << 1) | 1;
+            }
+            vsd.significand |= rem != 0;
+        }
+    }
+    vsd.significand = vfp_shiftright32jamming(vsd.significand, 1);
+
+    return vfp_single_normaliseround(state, sd, &vsd, fpscr, 0, "fsqrt");
 }
 
 /*
@@ -414,305 +414,346 @@ static u32 vfp_single_fsqrt(ARMul_State* state, int sd, int unused, s32 m, u32 f
  */
 static u32 vfp_compare(ARMul_State* state, int sd, int signal_on_qnan, s32 m, u32 fpscr)
 {
-	s32 d;
-	u32 ret = 0;
-
-	d = vfp_get_float(state, sd);
-	if (vfp_single_packed_exponent(m) == 255 && vfp_single_packed_mantissa(m)) {
-		ret |= FPSCR_C | FPSCR_V;
-		if (signal_on_qnan || !(vfp_single_packed_mantissa(m) & (1 << (VFP_SINGLE_MANTISSA_BITS - 1))))
-			/*
-			 * Signalling NaN, or signalling on quiet NaN
-			 */
-			ret |= FPSCR_IOC;
-	}
-
-	if (vfp_single_packed_exponent(d) == 255 && vfp_single_packed_mantissa(d)) {
-		ret |= FPSCR_C | FPSCR_V;
-		if (signal_on_qnan || !(vfp_single_packed_mantissa(d) & (1 << (VFP_SINGLE_MANTISSA_BITS - 1))))
-			/*
-			 * Signalling NaN, or signalling on quiet NaN
-			 */
-			ret |= FPSCR_IOC;
-	}
-
-	if (ret == 0) {
-		if (d == m || vfp_single_packed_abs(d | m) == 0) {
-			/*
-			 * equal
-			 */
-			ret |= FPSCR_Z | FPSCR_C;
-		} else if (vfp_single_packed_sign(d ^ m)) {
-			/*
-			 * different signs
-			 */
-			if (vfp_single_packed_sign(d))
-				/*
-				 * d is negative, so d < m
-				 */
-				ret |= FPSCR_N;
-			else
-				/*
-				 * d is positive, so d > m
-				 */
-				ret |= FPSCR_C;
-		} else if ((vfp_single_packed_sign(d) != 0) ^ (d < m)) {
-			/*
-			 * d < m
-			 */
-			ret |= FPSCR_N;
-		} else if ((vfp_single_packed_sign(d) != 0) ^ (d > m)) {
-			/*
-			 * d > m
-			 */
-			ret |= FPSCR_C;
-		}
-	}
-	return ret;
+    s32 d;
+    u32 ret = 0;
+
+    d = vfp_get_float(state, sd);
+    if (vfp_single_packed_exponent(m) == 255 && vfp_single_packed_mantissa(m)) {
+        ret |= FPSCR_C | FPSCR_V;
+        if (signal_on_qnan || !(vfp_single_packed_mantissa(m) & (1 << (VFP_SINGLE_MANTISSA_BITS - 1))))
+            /*
+             * Signalling NaN, or signalling on quiet NaN
+             */
+            ret |= FPSCR_IOC;
+    }
+
+    if (vfp_single_packed_exponent(d) == 255 && vfp_single_packed_mantissa(d)) {
+        ret |= FPSCR_C | FPSCR_V;
+        if (signal_on_qnan || !(vfp_single_packed_mantissa(d) & (1 << (VFP_SINGLE_MANTISSA_BITS - 1))))
+            /*
+             * Signalling NaN, or signalling on quiet NaN
+             */
+            ret |= FPSCR_IOC;
+    }
+
+    if (ret == 0) {
+        if (d == m || vfp_single_packed_abs(d | m) == 0) {
+            /*
+             * equal
+             */
+            ret |= FPSCR_Z | FPSCR_C;
+        } else if (vfp_single_packed_sign(d ^ m)) {
+            /*
+             * different signs
+             */
+            if (vfp_single_packed_sign(d))
+                /*
+                 * d is negative, so d < m
+                 */
+                ret |= FPSCR_N;
+            else
+                /*
+                 * d is positive, so d > m
+                 */
+                ret |= FPSCR_C;
+        } else if ((vfp_single_packed_sign(d) != 0) ^ (d < m)) {
+            /*
+             * d < m
+             */
+            ret |= FPSCR_N;
+        } else if ((vfp_single_packed_sign(d) != 0) ^ (d > m)) {
+            /*
+             * d > m
+             */
+            ret |= FPSCR_C;
+        }
+    }
+    return ret;
 }
 
 static u32 vfp_single_fcmp(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	return vfp_compare(state, sd, 0, m, fpscr);
+    return vfp_compare(state, sd, 0, m, fpscr);
 }
 
 static u32 vfp_single_fcmpe(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	return vfp_compare(state, sd, 1, m, fpscr);
+    return vfp_compare(state, sd, 1, m, fpscr);
 }
 
 static u32 vfp_single_fcmpz(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	return vfp_compare(state, sd, 0, 0, fpscr);
+    return vfp_compare(state, sd, 0, 0, fpscr);
 }
 
 static u32 vfp_single_fcmpez(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	return vfp_compare(state, sd, 1, 0, fpscr);
+    return vfp_compare(state, sd, 1, 0, fpscr);
+}
+
+static s64 vfp_single_to_doubleintern(ARMul_State* state, s32 m, u32 fpscr) //ichfly for internal use only
+{
+    struct vfp_single vsm;
+    struct vfp_double vdd;
+    int tm;
+    u32 exceptions = 0;
+
+    vfp_single_unpack(&vsm, m);
+
+    tm = vfp_single_type(&vsm);
+
+    /*
+    * If we have a signalling NaN, signal invalid operation.
+    */
+    if (tm == VFP_SNAN)
+        exceptions = FPSCR_IOC;
+
+    if (tm & VFP_DENORMAL)
+        vfp_single_normalise_denormal(&vsm);
+
+    vdd.sign = vsm.sign;
+    vdd.significand = (u64)vsm.significand << 32;
+
+    /*
+    * If we have an infinity or NaN, the exponent must be 2047.
+    */
+    if (tm & (VFP_INFINITY | VFP_NAN)) {
+        vdd.exponent = 2047;
+        if (tm == VFP_QNAN)
+            vdd.significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
+        goto pack_nan;
+    }
+    else if (tm & VFP_ZERO)
+        vdd.exponent = 0;
+    else
+        vdd.exponent = vsm.exponent + (1023 - 127);
+pack_nan:
+    vfp_double_normaliseroundintern(state, &vdd, fpscr, exceptions, "fcvtd");
+    return vfp_double_pack(&vdd);
 }
 
 static u32 vfp_single_fcvtd(ARMul_State* state, int dd, int unused, s32 m, u32 fpscr)
 {
-	struct vfp_single vsm;
-	struct vfp_double vdd;
-	int tm;
-	u32 exceptions = 0;
-
-	vfp_single_unpack(&vsm, m);
-
-	tm = vfp_single_type(&vsm);
-
-	/*
-	 * If we have a signalling NaN, signal invalid operation.
-	 */
-	if (tm == VFP_SNAN)
-		exceptions = FPSCR_IOC;
-
-	if (tm & VFP_DENORMAL)
-		vfp_single_normalise_denormal(&vsm);
-
-	vdd.sign = vsm.sign;
-	vdd.significand = (u64)vsm.significand << 32;
-
-	/*
-	 * If we have an infinity or NaN, the exponent must be 2047.
-	 */
-	if (tm & (VFP_INFINITY|VFP_NAN)) {
-		vdd.exponent = 2047;
-		if (tm == VFP_QNAN)
-			vdd.significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
-		goto pack_nan;
-	} else if (tm & VFP_ZERO)
-		vdd.exponent = 0;
-	else
-		vdd.exponent = vsm.exponent + (1023 - 127);
-
-	return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fcvtd");
-
- pack_nan:
-	vfp_put_double(state, vfp_double_pack(&vdd), dd);
-	return exceptions;
+    struct vfp_single vsm;
+    struct vfp_double vdd;
+    int tm;
+    u32 exceptions = 0;
+
+    vfp_single_unpack(&vsm, m);
+
+    tm = vfp_single_type(&vsm);
+
+    /*
+     * If we have a signalling NaN, signal invalid operation.
+     */
+    if (tm == VFP_SNAN)
+        exceptions = FPSCR_IOC;
+
+    if (tm & VFP_DENORMAL)
+        vfp_single_normalise_denormal(&vsm);
+
+    vdd.sign = vsm.sign;
+    vdd.significand = (u64)vsm.significand << 32;
+
+    /*
+     * If we have an infinity or NaN, the exponent must be 2047.
+     */
+    if (tm & (VFP_INFINITY|VFP_NAN)) {
+        vdd.exponent = 2047;
+        if (tm == VFP_QNAN)
+            vdd.significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
+        goto pack_nan;
+    } else if (tm & VFP_ZERO)
+        vdd.exponent = 0;
+    else
+        vdd.exponent = vsm.exponent + (1023 - 127);
+
+    return vfp_double_normaliseround(state, dd, &vdd, fpscr, exceptions, "fcvtd");
+
+pack_nan:
+    vfp_put_double(state, vfp_double_pack(&vdd), dd);
+    return exceptions;
 }
 
 static u32 vfp_single_fuito(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	struct vfp_single vs;
+    struct vfp_single vs;
 
-	vs.sign = 0;
-	vs.exponent = 127 + 31 - 1;
-	vs.significand = (u32)m;
+    vs.sign = 0;
+    vs.exponent = 127 + 31 - 1;
+    vs.significand = (u32)m;
 
-	return vfp_single_normaliseround(state, sd, &vs, fpscr, 0, "fuito");
+    return vfp_single_normaliseround(state, sd, &vs, fpscr, 0, "fuito");
 }
 
 static u32 vfp_single_fsito(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	struct vfp_single vs;
+    struct vfp_single vs;
 
-	vs.sign = (m & 0x80000000) >> 16;
-	vs.exponent = 127 + 31 - 1;
-	vs.significand = vs.sign ? -m : m;
+    vs.sign = (m & 0x80000000) >> 16;
+    vs.exponent = 127 + 31 - 1;
+    vs.significand = vs.sign ? -m : m;
 
-	return vfp_single_normaliseround(state, sd, &vs, fpscr, 0, "fsito");
+    return vfp_single_normaliseround(state, sd, &vs, fpscr, 0, "fsito");
 }
 
 static u32 vfp_single_ftoui(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	struct vfp_single vsm;
-	u32 d, exceptions = 0;
-	int rmode = fpscr & FPSCR_RMODE_MASK;
-	int tm;
-
-	vfp_single_unpack(&vsm, m);
-	vfp_single_dump("VSM", &vsm);
-
-	/*
-	 * Do we have a denormalised number?
-	 */
-	tm = vfp_single_type(&vsm);
-	if (tm & VFP_DENORMAL)
-		exceptions |= FPSCR_IDC;
-
-	if (tm & VFP_NAN)
-		vsm.sign = 0;
-
-	if (vsm.exponent >= 127 + 32) {
-		d = vsm.sign ? 0 : 0xffffffff;
-		exceptions = FPSCR_IOC;
-	} else if (vsm.exponent >= 127 - 1) {
-		int shift = 127 + 31 - vsm.exponent;
-		u32 rem, incr = 0;
-
-		/*
-		 * 2^0 <= m < 2^32-2^8
-		 */
-		d = (vsm.significand << 1) >> shift;
-		rem = vsm.significand << (33 - shift);
-
-		if (rmode == FPSCR_ROUND_NEAREST) {
-			incr = 0x80000000;
-			if ((d & 1) == 0)
-				incr -= 1;
-		} else if (rmode == FPSCR_ROUND_TOZERO) {
-			incr = 0;
-		} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vsm.sign != 0)) {
-			incr = ~0;
-		}
-
-		if ((rem + incr) < rem) {
-			if (d < 0xffffffff)
-				d += 1;
-			else
-				exceptions |= FPSCR_IOC;
-		}
-
-		if (d && vsm.sign) {
-			d = 0;
-			exceptions |= FPSCR_IOC;
-		} else if (rem)
-			exceptions |= FPSCR_IXC;
-	} else {
-		d = 0;
-		if (vsm.exponent | vsm.significand) {
-			exceptions |= FPSCR_IXC;
-			if (rmode == FPSCR_ROUND_PLUSINF && vsm.sign == 0)
-				d = 1;
-			else if (rmode == FPSCR_ROUND_MINUSINF && vsm.sign) {
-				d = 0;
-				exceptions |= FPSCR_IOC;
-			}
-		}
-	}
-
-	pr_debug("VFP: ftoui: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
-
-	vfp_put_float(state, d, sd);
-
-	return exceptions;
+    struct vfp_single vsm;
+    u32 d, exceptions = 0;
+    int rmode = fpscr & FPSCR_RMODE_MASK;
+    int tm;
+
+    vfp_single_unpack(&vsm, m);
+    vfp_single_dump("VSM", &vsm);
+
+    /*
+     * Do we have a denormalised number?
+     */
+    tm = vfp_single_type(&vsm);
+    if (tm & VFP_DENORMAL)
+        exceptions |= FPSCR_IDC;
+
+    if (tm & VFP_NAN)
+        vsm.sign = 0;
+
+    if (vsm.exponent >= 127 + 32) {
+        d = vsm.sign ? 0 : 0xffffffff;
+        exceptions = FPSCR_IOC;
+    } else if (vsm.exponent >= 127 - 1) {
+        int shift = 127 + 31 - vsm.exponent;
+        u32 rem, incr = 0;
+
+        /*
+         * 2^0 <= m < 2^32-2^8
+         */
+        d = (vsm.significand << 1) >> shift;
+        rem = vsm.significand << (33 - shift);
+
+        if (rmode == FPSCR_ROUND_NEAREST) {
+            incr = 0x80000000;
+            if ((d & 1) == 0)
+                incr -= 1;
+        } else if (rmode == FPSCR_ROUND_TOZERO) {
+            incr = 0;
+        } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vsm.sign != 0)) {
+            incr = ~0;
+        }
+
+        if ((rem + incr) < rem) {
+            if (d < 0xffffffff)
+                d += 1;
+            else
+                exceptions |= FPSCR_IOC;
+        }
+
+        if (d && vsm.sign) {
+            d = 0;
+            exceptions |= FPSCR_IOC;
+        } else if (rem)
+            exceptions |= FPSCR_IXC;
+    } else {
+        d = 0;
+        if (vsm.exponent | vsm.significand) {
+            exceptions |= FPSCR_IXC;
+            if (rmode == FPSCR_ROUND_PLUSINF && vsm.sign == 0)
+                d = 1;
+            else if (rmode == FPSCR_ROUND_MINUSINF && vsm.sign) {
+                d = 0;
+                exceptions |= FPSCR_IOC;
+            }
+        }
+    }
+
+    pr_debug("VFP: ftoui: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
+
+    vfp_put_float(state, d, sd);
+
+    return exceptions;
 }
 
 static u32 vfp_single_ftouiz(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	return vfp_single_ftoui(state, sd, unused, m, FPSCR_ROUND_TOZERO);
+    return vfp_single_ftoui(state, sd, unused, m, FPSCR_ROUND_TOZERO);
 }
 
 static u32 vfp_single_ftosi(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	struct vfp_single vsm;
-	u32 d, exceptions = 0;
-	int rmode = fpscr & FPSCR_RMODE_MASK;
-	int tm;
-
-	vfp_single_unpack(&vsm, m);
-	vfp_single_dump("VSM", &vsm);
-
-	/*
-	 * Do we have a denormalised number?
-	 */
-	tm = vfp_single_type(&vsm);
-	if (vfp_single_type(&vsm) & VFP_DENORMAL)
-		exceptions |= FPSCR_IDC;
-
-	if (tm & VFP_NAN) {
-		d = 0;
-		exceptions |= FPSCR_IOC;
-	} else if (vsm.exponent >= 127 + 32) {
-		/*
-		 * m >= 2^31-2^7: invalid
-		 */
-		d = 0x7fffffff;
-		if (vsm.sign)
-			d = ~d;
-		exceptions |= FPSCR_IOC;
-	} else if (vsm.exponent >= 127 - 1) {
-		int shift = 127 + 31 - vsm.exponent;
-		u32 rem, incr = 0;
-
-		/* 2^0 <= m <= 2^31-2^7 */
-		d = (vsm.significand << 1) >> shift;
-		rem = vsm.significand << (33 - shift);
-
-		if (rmode == FPSCR_ROUND_NEAREST) {
-			incr = 0x80000000;
-			if ((d & 1) == 0)
-				incr -= 1;
-		} else if (rmode == FPSCR_ROUND_TOZERO) {
-			incr = 0;
-		} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vsm.sign != 0)) {
-			incr = ~0;
-		}
-
-		if ((rem + incr) < rem && d < 0xffffffff)
-			d += 1;
-		if (d > 0x7fffffff + (vsm.sign != 0)) {
-			d = 0x7fffffff + (vsm.sign != 0);
-			exceptions |= FPSCR_IOC;
-		} else if (rem)
-			exceptions |= FPSCR_IXC;
-
-		if (vsm.sign)
-			d = -d;
-	} else {
-		d = 0;
-		if (vsm.exponent | vsm.significand) {
-			exceptions |= FPSCR_IXC;
-			if (rmode == FPSCR_ROUND_PLUSINF && vsm.sign == 0)
-				d = 1;
-			else if (rmode == FPSCR_ROUND_MINUSINF && vsm.sign)
-				d = -1;
-		}
-	}
-
-	pr_debug("VFP: ftosi: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
-
-	vfp_put_float(state, (s32)d, sd);
-
-	return exceptions;
+    struct vfp_single vsm;
+    u32 d, exceptions = 0;
+    int rmode = fpscr & FPSCR_RMODE_MASK;
+    int tm;
+
+    vfp_single_unpack(&vsm, m);
+    vfp_single_dump("VSM", &vsm);
+
+    /*
+     * Do we have a denormalised number?
+     */
+    tm = vfp_single_type(&vsm);
+    if (vfp_single_type(&vsm) & VFP_DENORMAL)
+        exceptions |= FPSCR_IDC;
+
+    if (tm & VFP_NAN) {
+        d = 0;
+        exceptions |= FPSCR_IOC;
+    } else if (vsm.exponent >= 127 + 32) {
+        /*
+         * m >= 2^31-2^7: invalid
+         */
+        d = 0x7fffffff;
+        if (vsm.sign)
+            d = ~d;
+        exceptions |= FPSCR_IOC;
+    } else if (vsm.exponent >= 127 - 1) {
+        int shift = 127 + 31 - vsm.exponent;
+        u32 rem, incr = 0;
+
+        /* 2^0 <= m <= 2^31-2^7 */
+        d = (vsm.significand << 1) >> shift;
+        rem = vsm.significand << (33 - shift);
+
+        if (rmode == FPSCR_ROUND_NEAREST) {
+            incr = 0x80000000;
+            if ((d & 1) == 0)
+                incr -= 1;
+        } else if (rmode == FPSCR_ROUND_TOZERO) {
+            incr = 0;
+        } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vsm.sign != 0)) {
+            incr = ~0;
+        }
+
+        if ((rem + incr) < rem && d < 0xffffffff)
+            d += 1;
+        if (d > (0x7fffffffu + (vsm.sign != 0))) {
+            d = (0x7fffffffu + (vsm.sign != 0));
+            exceptions |= FPSCR_IOC;
+        } else if (rem)
+            exceptions |= FPSCR_IXC;
+
+        if (vsm.sign)
+            d = 0-d;
+    } else {
+        d = 0;
+        if (vsm.exponent | vsm.significand) {
+            exceptions |= FPSCR_IXC;
+            if (rmode == FPSCR_ROUND_PLUSINF && vsm.sign == 0)
+                d = 1;
+            else if (rmode == FPSCR_ROUND_MINUSINF && vsm.sign)
+                d = -1;
+        }
+    }
+
+    pr_debug("VFP: ftosi: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
+
+    vfp_put_float(state, (s32)d, sd);
+
+    return exceptions;
 }
 
 static u32 vfp_single_ftosiz(ARMul_State* state, int sd, int unused, s32 m, u32 fpscr)
 {
-	return vfp_single_ftosi(state, sd, unused, m, FPSCR_ROUND_TOZERO);
+    return vfp_single_ftosi(state, sd, unused, m, FPSCR_ROUND_TOZERO);
 }
 
 static struct op fops_ext[] = {
@@ -752,200 +793,237 @@ static struct op fops_ext[] = {
 
 static u32
 vfp_single_fadd_nonnumber(struct vfp_single *vsd, struct vfp_single *vsn,
-			  struct vfp_single *vsm, u32 fpscr)
+                          struct vfp_single *vsm, u32 fpscr)
 {
-	struct vfp_single *vsp;
-	u32 exceptions = 0;
-	int tn, tm;
-
-	tn = vfp_single_type(vsn);
-	tm = vfp_single_type(vsm);
-
-	if (tn & tm & VFP_INFINITY) {
-		/*
-		 * Two infinities.  Are they different signs?
-		 */
-		if (vsn->sign ^ vsm->sign) {
-			/*
-			 * different signs -> invalid
-			 */
-			exceptions = FPSCR_IOC;
-			vsp = &vfp_single_default_qnan;
-		} else {
-			/*
-			 * same signs -> valid
-			 */
-			vsp = vsn;
-		}
-	} else if (tn & VFP_INFINITY && tm & VFP_NUMBER) {
-		/*
-		 * One infinity and one number -> infinity
-		 */
-		vsp = vsn;
-	} else {
-		/*
-		 * 'n' is a NaN of some type
-		 */
-		return vfp_propagate_nan(vsd, vsn, vsm, fpscr);
-	}
-	*vsd = *vsp;
-	return exceptions;
+    struct vfp_single *vsp;
+    u32 exceptions = 0;
+    int tn, tm;
+
+    tn = vfp_single_type(vsn);
+    tm = vfp_single_type(vsm);
+
+    if (tn & tm & VFP_INFINITY) {
+        /*
+         * Two infinities.  Are they different signs?
+         */
+        if (vsn->sign ^ vsm->sign) {
+            /*
+             * different signs -> invalid
+             */
+            exceptions = FPSCR_IOC;
+            vsp = &vfp_single_default_qnan;
+        } else {
+            /*
+             * same signs -> valid
+             */
+            vsp = vsn;
+        }
+    } else if (tn & VFP_INFINITY && tm & VFP_NUMBER) {
+        /*
+         * One infinity and one number -> infinity
+         */
+        vsp = vsn;
+    } else {
+        /*
+         * 'n' is a NaN of some type
+         */
+        return vfp_propagate_nan(vsd, vsn, vsm, fpscr);
+    }
+    *vsd = *vsp;
+    return exceptions;
 }
 
 static u32
 vfp_single_add(struct vfp_single *vsd, struct vfp_single *vsn,
-	       struct vfp_single *vsm, u32 fpscr)
+               struct vfp_single *vsm, u32 fpscr)
 {
-	u32 exp_diff, m_sig;
-
-	if (vsn->significand & 0x80000000 ||
-	    vsm->significand & 0x80000000) {
-		pr_info("VFP: bad FP values\n");
-		vfp_single_dump("VSN", vsn);
-		vfp_single_dump("VSM", vsm);
-	}
-
-	/*
-	 * Ensure that 'n' is the largest magnitude number.  Note that
-	 * if 'n' and 'm' have equal exponents, we do not swap them.
-	 * This ensures that NaN propagation works correctly.
-	 */
-	if (vsn->exponent < vsm->exponent) {
-		struct vfp_single *t = vsn;
-		vsn = vsm;
-		vsm = t;
-	}
-
-	/*
-	 * Is 'n' an infinity or a NaN?  Note that 'm' may be a number,
-	 * infinity or a NaN here.
-	 */
-	if (vsn->exponent == 255)
-		return vfp_single_fadd_nonnumber(vsd, vsn, vsm, fpscr);
-
-	/*
-	 * We have two proper numbers, where 'vsn' is the larger magnitude.
-	 *
-	 * Copy 'n' to 'd' before doing the arithmetic.
-	 */
-	*vsd = *vsn;
-
-	/*
-	 * Align both numbers.
-	 */
-	exp_diff = vsn->exponent - vsm->exponent;
-	m_sig = vfp_shiftright32jamming(vsm->significand, exp_diff);
-
-	/*
-	 * If the signs are different, we are really subtracting.
-	 */
-	if (vsn->sign ^ vsm->sign) {
-		m_sig = vsn->significand - m_sig;
-		if ((s32)m_sig < 0) {
-			vsd->sign = vfp_sign_negate(vsd->sign);
-			m_sig = -m_sig;
-		} else if (m_sig == 0) {
-			vsd->sign = (fpscr & FPSCR_RMODE_MASK) ==
-				      FPSCR_ROUND_MINUSINF ? 0x8000 : 0;
-		}
-	} else {
-		m_sig = vsn->significand + m_sig;
-	}
-	vsd->significand = m_sig;
-
-	return 0;
+    u32 exp_diff, m_sig;
+
+    if (vsn->significand & 0x80000000 ||
+            vsm->significand & 0x80000000) {
+        pr_info("VFP: bad FP values in %s\n", __func__);
+        vfp_single_dump("VSN", vsn);
+        vfp_single_dump("VSM", vsm);
+    }
+
+    /*
+     * Ensure that 'n' is the largest magnitude number.  Note that
+     * if 'n' and 'm' have equal exponents, we do not swap them.
+     * This ensures that NaN propagation works correctly.
+     */
+    if (vsn->exponent < vsm->exponent) {
+        struct vfp_single *t = vsn;
+        vsn = vsm;
+        vsm = t;
+    }
+
+    /*
+     * Is 'n' an infinity or a NaN?  Note that 'm' may be a number,
+     * infinity or a NaN here.
+     */
+    if (vsn->exponent == 255)
+        return vfp_single_fadd_nonnumber(vsd, vsn, vsm, fpscr);
+
+    /*
+     * We have two proper numbers, where 'vsn' is the larger magnitude.
+     *
+     * Copy 'n' to 'd' before doing the arithmetic.
+     */
+    *vsd = *vsn;
+
+    /*
+     * Align both numbers.
+     */
+    exp_diff = vsn->exponent - vsm->exponent;
+    m_sig = vfp_shiftright32jamming(vsm->significand, exp_diff);
+
+    /*
+     * If the signs are different, we are really subtracting.
+     */
+    if (vsn->sign ^ vsm->sign) {
+        m_sig = vsn->significand - m_sig;
+        if ((s32)m_sig < 0) {
+            vsd->sign = vfp_sign_negate(vsd->sign);
+            m_sig = 0-m_sig;
+        } else if (m_sig == 0) {
+            vsd->sign = (fpscr & FPSCR_RMODE_MASK) ==
+                        FPSCR_ROUND_MINUSINF ? 0x8000 : 0;
+        }
+    } else {
+        m_sig = vsn->significand + m_sig;
+    }
+    vsd->significand = m_sig;
+
+    return 0;
 }
 
 static u32
 vfp_single_multiply(struct vfp_single *vsd, struct vfp_single *vsn, struct vfp_single *vsm, u32 fpscr)
 {
-	vfp_single_dump("VSN", vsn);
-	vfp_single_dump("VSM", vsm);
-
-	/*
-	 * Ensure that 'n' is the largest magnitude number.  Note that
-	 * if 'n' and 'm' have equal exponents, we do not swap them.
-	 * This ensures that NaN propagation works correctly.
-	 */
-	if (vsn->exponent < vsm->exponent) {
-		struct vfp_single *t = vsn;
-		vsn = vsm;
-		vsm = t;
-		pr_debug("VFP: swapping M <-> N\n");
-	}
-
-	vsd->sign = vsn->sign ^ vsm->sign;
-
-	/*
-	 * If 'n' is an infinity or NaN, handle it.  'm' may be anything.
-	 */
-	if (vsn->exponent == 255) {
-		if (vsn->significand || (vsm->exponent == 255 && vsm->significand))
-			return vfp_propagate_nan(vsd, vsn, vsm, fpscr);
-		if ((vsm->exponent | vsm->significand) == 0) {
-			*vsd = vfp_single_default_qnan;
-			return FPSCR_IOC;
-		}
-		vsd->exponent = vsn->exponent;
-		vsd->significand = 0;
-		return 0;
-	}
-
-	/*
-	 * If 'm' is zero, the result is always zero.  In this case,
-	 * 'n' may be zero or a number, but it doesn't matter which.
-	 */
-	if ((vsm->exponent | vsm->significand) == 0) {
-		vsd->exponent = 0;
-		vsd->significand = 0;
-		return 0;
-	}
-
-	/*
-	 * We add 2 to the destination exponent for the same reason as
-	 * the addition case - though this time we have +1 from each
-	 * input operand.
-	 */
-	vsd->exponent = vsn->exponent + vsm->exponent - 127 + 2;
-	vsd->significand = vfp_hi64to32jamming((u64)vsn->significand * vsm->significand);
-
-	vfp_single_dump("VSD", vsd);
-	return 0;
+    vfp_single_dump("VSN", vsn);
+    vfp_single_dump("VSM", vsm);
+
+    /*
+     * Ensure that 'n' is the largest magnitude number.  Note that
+     * if 'n' and 'm' have equal exponents, we do not swap them.
+     * This ensures that NaN propagation works correctly.
+     */
+    if (vsn->exponent < vsm->exponent) {
+        struct vfp_single *t = vsn;
+        vsn = vsm;
+        vsm = t;
+        pr_debug("VFP: swapping M <-> N\n");
+    }
+
+    vsd->sign = vsn->sign ^ vsm->sign;
+
+    /*
+     * If 'n' is an infinity or NaN, handle it.  'm' may be anything.
+     */
+    if (vsn->exponent == 255) {
+        if (vsn->significand || (vsm->exponent == 255 && vsm->significand))
+            return vfp_propagate_nan(vsd, vsn, vsm, fpscr);
+        if ((vsm->exponent | vsm->significand) == 0) {
+            *vsd = vfp_single_default_qnan;
+            return FPSCR_IOC;
+        }
+        vsd->exponent = vsn->exponent;
+        vsd->significand = 0;
+        return 0;
+    }
+
+    /*
+     * If 'm' is zero, the result is always zero.  In this case,
+     * 'n' may be zero or a number, but it doesn't matter which.
+     */
+    if ((vsm->exponent | vsm->significand) == 0) {
+        vsd->exponent = 0;
+        vsd->significand = 0;
+        return 0;
+    }
+
+    /*
+     * We add 2 to the destination exponent for the same reason as
+     * the addition case - though this time we have +1 from each
+     * input operand.
+     */
+    vsd->exponent = vsn->exponent + vsm->exponent - 127 + 2;
+    vsd->significand = vfp_hi64to32jamming((u64)vsn->significand * vsm->significand);
+
+    vfp_single_dump("VSD", vsd);
+    return 0;
 }
 
 #define NEG_MULTIPLY	(1 << 0)
 #define NEG_SUBTRACT	(1 << 1)
 
 static u32
-vfp_single_multiply_accumulate(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr, u32 negate, const char *func)
+vfp_single_multiply_accumulate(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr, u32 negate, char *func)
 {
-	struct vfp_single vsd, vsp, vsn, vsm;
-	u32 exceptions;
-	s32 v;
+    
+    {
+        struct vfp_single vsd, vsp, vsn, vsm;
+        u32 exceptions;
+        s32 v;
+
+
+
+        v = vfp_get_float(state, sn);
+        pr_debug("VFP: s%u = %08x\n", sn, v);
+        vfp_single_unpack(&vsn, v);
+        if (vsn.exponent == 0 && vsn.significand)
+            vfp_single_normalise_denormal(&vsn);
+
+        vfp_single_unpack(&vsm, m);
+        if (vsm.exponent == 0 && vsm.significand)
+            vfp_single_normalise_denormal(&vsm);
+
+        exceptions = vfp_single_multiply(&vsp, &vsn, &vsm, fpscr);
+
+        if (negate & NEG_MULTIPLY)
+            vsp.sign = vfp_sign_negate(vsp.sign);
+
+        v = vfp_get_float(state, sd);
+        pr_debug("VFP: s%u = %08x\n", sd, v);
+        vfp_single_unpack(&vsn, v);
+        if (negate & NEG_SUBTRACT)
+            vsn.sign = vfp_sign_negate(vsn.sign);
+
+        exceptions |= vfp_single_add(&vsd, &vsn, &vsp, fpscr);
+
+        return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, func);
+    }
+
+    struct vfp_double vsd, vsp, vsn, vsm;
+    u32 exceptions;
+    s32 v;
+    s64 vd;
+    s64 md;
+
+    v = vfp_get_float(state, sn);
+    vd = vfp_single_to_doubleintern(state, v, fpscr);
+    vfp_double_unpack(&vsn, vd);
+
+    md = vfp_single_to_doubleintern(state, m, fpscr);
+    vfp_double_unpack(&vsm, md);
+
+    exceptions = vfp_double_multiply(&vsp, &vsn, &vsm, fpscr);
+    if (negate & NEG_MULTIPLY)
+        vsp.sign = vfp_sign_negate(vsp.sign);
 
-	v = vfp_get_float(state, sn);
-	pr_debug("VFP: s%u = %08x\n", sn, v);
-	vfp_single_unpack(&vsn, v);
-	if (vsn.exponent == 0 && vsn.significand)
-		vfp_single_normalise_denormal(&vsn);
+    v = vfp_get_float(state, sd);
+    vd = vfp_single_to_doubleintern(state, v, fpscr);
+    vfp_double_unpack(&vsn, vd);
 
-	vfp_single_unpack(&vsm, m);
-	if (vsm.exponent == 0 && vsm.significand)
-		vfp_single_normalise_denormal(&vsm);
+    if (negate & NEG_SUBTRACT)
+        vsn.sign = vfp_sign_negate(vsn.sign);
 
-	exceptions = vfp_single_multiply(&vsp, &vsn, &vsm, fpscr);
-	if (negate & NEG_MULTIPLY)
-		vsp.sign = vfp_sign_negate(vsp.sign);
+    exceptions |= vfp_double_add(&vsd, &vsn, &vsp, fpscr);
 
-	v = vfp_get_float(state, sd);
-	pr_debug("VFP: s%u = %08x\n", sd, v);
-	vfp_single_unpack(&vsn, v);
-	if (negate & NEG_SUBTRACT)
-		vsn.sign = vfp_sign_negate(vsn.sign);
+    s64 debug = vfp_double_pack(&vsd);
 
-	exceptions |= vfp_single_add(&vsd, &vsn, &vsp, fpscr);
+    return vfp_double_fcvtsinterncutting(state, sd, &vsd, fpscr);
 
-	return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, func);
 }
 
 /*
@@ -957,8 +1035,8 @@ vfp_single_multiply_accumulate(ARMul_State* state, int sd, int sn, s32 m, u32 fp
  */
 static u32 vfp_single_fmac(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, sd);
-	return vfp_single_multiply_accumulate(state, sd, sn, m, fpscr, 0, "fmac");
+    pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, sd);
+    return vfp_single_multiply_accumulate(state, sd, sn, m, fpscr, 0, "fmac");
 }
 
 /*
@@ -966,8 +1044,8 @@ static u32 vfp_single_fmac(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
  */
 static u32 vfp_single_fnmac(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sd, sn);
-	return vfp_single_multiply_accumulate(state, sd, sn, m, fpscr, NEG_MULTIPLY, "fnmac");
+    pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sd, sn);
+    return vfp_single_multiply_accumulate(state, sd, sn, m, fpscr, NEG_MULTIPLY, "fnmac");
 }
 
 /*
@@ -975,8 +1053,8 @@ static u32 vfp_single_fnmac(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr
  */
 static u32 vfp_single_fmsc(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, sd);
-	return vfp_single_multiply_accumulate(state, sd, sn, m, fpscr, NEG_SUBTRACT, "fmsc");
+    pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, sd);
+    return vfp_single_multiply_accumulate(state, sd, sn, m, fpscr, NEG_SUBTRACT, "fmsc");
 }
 
 /*
@@ -984,8 +1062,8 @@ static u32 vfp_single_fmsc(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
  */
 static u32 vfp_single_fnmsc(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, sd);
-	return vfp_single_multiply_accumulate(state, sd, sn, m, fpscr, NEG_SUBTRACT | NEG_MULTIPLY, "fnmsc");
+    pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, sd);
+    return vfp_single_multiply_accumulate(state, sd, sn, m, fpscr, NEG_SUBTRACT | NEG_MULTIPLY, "fnmsc");
 }
 
 /*
@@ -993,22 +1071,22 @@ static u32 vfp_single_fnmsc(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr
  */
 static u32 vfp_single_fmul(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	struct vfp_single vsd, vsn, vsm;
-	u32 exceptions;
-	s32 n = vfp_get_float(state, sn);
+    struct vfp_single vsd, vsn, vsm;
+    u32 exceptions;
+    s32 n = vfp_get_float(state, sn);
 
-	pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, n);
+    pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, n);
 
-	vfp_single_unpack(&vsn, n);
-	if (vsn.exponent == 0 && vsn.significand)
-		vfp_single_normalise_denormal(&vsn);
+    vfp_single_unpack(&vsn, n);
+    if (vsn.exponent == 0 && vsn.significand)
+        vfp_single_normalise_denormal(&vsn);
 
-	vfp_single_unpack(&vsm, m);
-	if (vsm.exponent == 0 && vsm.significand)
-		vfp_single_normalise_denormal(&vsm);
+    vfp_single_unpack(&vsm, m);
+    if (vsm.exponent == 0 && vsm.significand)
+        vfp_single_normalise_denormal(&vsm);
 
-	exceptions = vfp_single_multiply(&vsd, &vsn, &vsm, fpscr);
-	return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fmul");
+    exceptions = vfp_single_multiply(&vsd, &vsn, &vsm, fpscr);
+    return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fmul");
 }
 
 /*
@@ -1016,23 +1094,23 @@ static u32 vfp_single_fmul(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
  */
 static u32 vfp_single_fnmul(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	struct vfp_single vsd, vsn, vsm;
-	u32 exceptions;
-	s32 n = vfp_get_float(state, sn);
+    struct vfp_single vsd, vsn, vsm;
+    u32 exceptions;
+    s32 n = vfp_get_float(state, sn);
 
-	pr_debug("VFP: s%u = %08x\n", sn, n);
+    pr_debug("VFP: s%u = %08x\n", sn, n);
 
-	vfp_single_unpack(&vsn, n);
-	if (vsn.exponent == 0 && vsn.significand)
-		vfp_single_normalise_denormal(&vsn);
+    vfp_single_unpack(&vsn, n);
+    if (vsn.exponent == 0 && vsn.significand)
+        vfp_single_normalise_denormal(&vsn);
 
-	vfp_single_unpack(&vsm, m);
-	if (vsm.exponent == 0 && vsm.significand)
-		vfp_single_normalise_denormal(&vsm);
+    vfp_single_unpack(&vsm, m);
+    if (vsm.exponent == 0 && vsm.significand)
+        vfp_single_normalise_denormal(&vsm);
 
-	exceptions = vfp_single_multiply(&vsd, &vsn, &vsm, fpscr);
-	vsd.sign = vfp_sign_negate(vsd.sign);
-	return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fnmul");
+    exceptions = vfp_single_multiply(&vsd, &vsn, &vsm, fpscr);
+    vsd.sign = vfp_sign_negate(vsd.sign);
+    return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fnmul");
 }
 
 /*
@@ -1040,26 +1118,26 @@ static u32 vfp_single_fnmul(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr
  */
 static u32 vfp_single_fadd(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	struct vfp_single vsd, vsn, vsm;
-	u32 exceptions;
-	s32 n = vfp_get_float(state, sn);
+    struct vfp_single vsd, vsn, vsm;
+    u32 exceptions;
+    s32 n = vfp_get_float(state, sn);
 
-	pr_debug("VFP: s%u = %08x\n", sn, n);
+    pr_debug("VFP: s%u = %08x\n", sn, n);
 
-	/*
-	 * Unpack and normalise denormals.
-	 */
-	vfp_single_unpack(&vsn, n);
-	if (vsn.exponent == 0 && vsn.significand)
-		vfp_single_normalise_denormal(&vsn);
+    /*
+     * Unpack and normalise denormals.
+     */
+    vfp_single_unpack(&vsn, n);
+    if (vsn.exponent == 0 && vsn.significand)
+        vfp_single_normalise_denormal(&vsn);
 
-	vfp_single_unpack(&vsm, m);
-	if (vsm.exponent == 0 && vsm.significand)
-		vfp_single_normalise_denormal(&vsm);
+    vfp_single_unpack(&vsm, m);
+    if (vsm.exponent == 0 && vsm.significand)
+        vfp_single_normalise_denormal(&vsm);
 
-	exceptions = vfp_single_add(&vsd, &vsn, &vsm, fpscr);
+    exceptions = vfp_single_add(&vsd, &vsn, &vsm, fpscr);
 
-	return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fadd");
+    return vfp_single_normaliseround(state, sd, &vsd, fpscr, exceptions, "fadd");
 }
 
 /*
@@ -1067,11 +1145,11 @@ static u32 vfp_single_fadd(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
  */
 static u32 vfp_single_fsub(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, sd);
-	/*
-	 * Subtraction is addition with one sign inverted.
-	 */
-	return vfp_single_fadd(state, sd, sn, vfp_single_packed_negate(m), fpscr);
+    pr_debug("In %sVFP: s%u = %08x\n", __FUNCTION__, sn, sd);
+    /*
+     * Subtraction is addition with one sign inverted.
+     */
+    return vfp_single_fadd(state, sd, sn, vfp_single_packed_negate(m), fpscr);
 }
 
 /*
@@ -1079,107 +1157,107 @@ static u32 vfp_single_fsub(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
  */
 static u32 vfp_single_fdiv(ARMul_State* state, int sd, int sn, s32 m, u32 fpscr)
 {
-	struct vfp_single vsd, vsn, vsm;
-	u32 exceptions = 0;
-	s32 n = vfp_get_float(state, sn);
-	int tm, tn;
-
-	pr_debug("VFP: s%u = %08x\n", sn, n);
-
-	vfp_single_unpack(&vsn, n);
-	vfp_single_unpack(&vsm, m);
-
-	vsd.sign = vsn.sign ^ vsm.sign;
-
-	tn = vfp_single_type(&vsn);
-	tm = vfp_single_type(&vsm);
-
-	/*
-	 * Is n a NAN?
-	 */
-	if (tn & VFP_NAN)
-		goto vsn_nan;
-
-	/*
-	 * Is m a NAN?
-	 */
-	if (tm & VFP_NAN)
-		goto vsm_nan;
-
-	/*
-	 * If n and m are infinity, the result is invalid
-	 * If n and m are zero, the result is invalid
-	 */
-	if (tm & tn & (VFP_INFINITY|VFP_ZERO))
-		goto invalid;
-
-	/*
-	 * If n is infinity, the result is infinity
-	 */
-	if (tn & VFP_INFINITY)
-		goto infinity;
-
-	/*
-	 * If m is zero, raise div0 exception
-	 */
-	if (tm & VFP_ZERO)
-		goto divzero;
-
-	/*
-	 * If m is infinity, or n is zero, the result is zero
-	 */
-	if (tm & VFP_INFINITY || tn & VFP_ZERO)
-		goto zero;
-
-	if (tn & VFP_DENORMAL)
-		vfp_single_normalise_denormal(&vsn);
-	if (tm & VFP_DENORMAL)
-		vfp_single_normalise_denormal(&vsm);
-
-	/*
-	 * Ok, we have two numbers, we can perform division.
-	 */
-	vsd.exponent = vsn.exponent - vsm.exponent + 127 - 1;
-	vsm.significand <<= 1;
-	if (vsm.significand <= (2 * vsn.significand)) {
-		vsn.significand >>= 1;
-		vsd.exponent++;
-	}
-	{
-		u64 significand = (u64)vsn.significand << 32;
-		do_div(significand, vsm.significand);
-		vsd.significand = significand;
-	}
-	if ((vsd.significand & 0x3f) == 0)
-		vsd.significand |= ((u64)vsm.significand * vsd.significand != (u64)vsn.significand << 32);
-
-	return vfp_single_normaliseround(state, sd, &vsd, fpscr, 0, "fdiv");
-
- vsn_nan:
-	exceptions = vfp_propagate_nan(&vsd, &vsn, &vsm, fpscr);
- pack:
-	vfp_put_float(state, vfp_single_pack(&vsd), sd);
-	return exceptions;
-
- vsm_nan:
-	exceptions = vfp_propagate_nan(&vsd, &vsm, &vsn, fpscr);
-	goto pack;
-
- zero:
-	vsd.exponent = 0;
-	vsd.significand = 0;
-	goto pack;
-
- divzero:
-	exceptions = FPSCR_DZC;
- infinity:
-	vsd.exponent = 255;
-	vsd.significand = 0;
-	goto pack;
-
- invalid:
-	vfp_put_float(state, vfp_single_pack(&vfp_single_default_qnan), sd);
-	return FPSCR_IOC;
+    struct vfp_single vsd, vsn, vsm;
+    u32 exceptions = 0;
+    s32 n = vfp_get_float(state, sn);
+    int tm, tn;
+
+    pr_debug("VFP: s%u = %08x\n", sn, n);
+
+    vfp_single_unpack(&vsn, n);
+    vfp_single_unpack(&vsm, m);
+
+    vsd.sign = vsn.sign ^ vsm.sign;
+
+    tn = vfp_single_type(&vsn);
+    tm = vfp_single_type(&vsm);
+
+    /*
+     * Is n a NAN?
+     */
+    if (tn & VFP_NAN)
+        goto vsn_nan;
+
+    /*
+     * Is m a NAN?
+     */
+    if (tm & VFP_NAN)
+        goto vsm_nan;
+
+    /*
+     * If n and m are infinity, the result is invalid
+     * If n and m are zero, the result is invalid
+     */
+    if (tm & tn & (VFP_INFINITY|VFP_ZERO))
+        goto invalid;
+
+    /*
+     * If n is infinity, the result is infinity
+     */
+    if (tn & VFP_INFINITY)
+        goto infinity;
+
+    /*
+     * If m is zero, raise div0 exception
+     */
+    if (tm & VFP_ZERO)
+        goto divzero;
+
+    /*
+     * If m is infinity, or n is zero, the result is zero
+     */
+    if (tm & VFP_INFINITY || tn & VFP_ZERO)
+        goto zero;
+
+    if (tn & VFP_DENORMAL)
+        vfp_single_normalise_denormal(&vsn);
+    if (tm & VFP_DENORMAL)
+        vfp_single_normalise_denormal(&vsm);
+
+    /*
+     * Ok, we have two numbers, we can perform division.
+     */
+    vsd.exponent = vsn.exponent - vsm.exponent + 127 - 1;
+    vsm.significand <<= 1;
+    if (vsm.significand <= (2 * vsn.significand)) {
+        vsn.significand >>= 1;
+        vsd.exponent++;
+    }
+    {
+        u64 significand = (u64)vsn.significand << 32;
+        do_div(significand, vsm.significand);
+        vsd.significand = (u32)significand;
+    }
+    if ((vsd.significand & 0x3f) == 0)
+        vsd.significand |= ((u64)vsm.significand * vsd.significand != (u64)vsn.significand << 32);
+
+    return vfp_single_normaliseround(state, sd, &vsd, fpscr, 0, "fdiv");
+
+vsn_nan:
+    exceptions = vfp_propagate_nan(&vsd, &vsn, &vsm, fpscr);
+pack:
+    vfp_put_float(state, vfp_single_pack(&vsd), sd);
+    return exceptions;
+
+vsm_nan:
+    exceptions = vfp_propagate_nan(&vsd, &vsm, &vsn, fpscr);
+    goto pack;
+
+zero:
+    vsd.exponent = 0;
+    vsd.significand = 0;
+    goto pack;
+
+divzero:
+    exceptions = FPSCR_DZC;
+infinity:
+    vsd.exponent = 255;
+    vsd.significand = 0;
+    goto pack;
+
+invalid:
+    vfp_put_float(state, vfp_single_pack(&vfp_single_default_qnan), sd);
+    return FPSCR_IOC;
 }
 
 static struct op fops[] = {
@@ -1199,80 +1277,80 @@ static struct op fops[] = {
 
 u32 vfp_single_cpdo(ARMul_State* state, u32 inst, u32 fpscr)
 {
-	u32 op = inst & FOP_MASK;
-	u32 exceptions = 0;
-	unsigned int dest;
-	unsigned int sn = vfp_get_sn(inst);
-	unsigned int sm = vfp_get_sm(inst);
-	unsigned int vecitr, veclen, vecstride;
-	struct op *fop;
-	pr_debug("In %s\n", __FUNCTION__);
-
-	vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK);
-
-	fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)];
-
-	/*
-	 * fcvtsd takes a dN register number as destination, not sN.
-	 * Technically, if bit 0 of dd is set, this is an invalid
-	 * instruction.  However, we ignore this for efficiency.
-	 * It also only operates on scalars.
-	 */
-	if (fop->flags & OP_DD)
-		dest = vfp_get_dd(inst);
-	else
-		dest = vfp_get_sd(inst);
-
-	/*
-	 * If destination bank is zero, vector length is always '1'.
-	 * ARM DDI0100F C5.1.3, C5.3.2.
-	 */
-	if ((fop->flags & OP_SCALAR) || FREG_BANK(dest) == 0)
-		veclen = 0;
-	else
-		veclen = fpscr & FPSCR_LENGTH_MASK;
-
-	pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
-		 (veclen >> FPSCR_LENGTH_BIT) + 1);
-
-	if (!fop->fn) {
-		printf("VFP: could not find single op %d, inst=0x%x@0x%x\n", FEXT_TO_IDX(inst), inst, state->Reg[15]);
-		exit(-1);
-		goto invalid;
-	}
-
-	for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
-		s32 m = vfp_get_float(state, sm);
-		u32 except;
-		char type;
-
-		type = fop->flags & OP_DD ? 'd' : 's';
-		if (op == FOP_EXT)
-			pr_debug("VFP: itr%d (%c%u) = op[%u] (s%u=%08x)\n",
-				 vecitr >> FPSCR_LENGTH_BIT, type, dest, sn,
-				 sm, m);
-		else
-			pr_debug("VFP: itr%d (%c%u) = (s%u) op[%u] (s%u=%08x)\n",
-				 vecitr >> FPSCR_LENGTH_BIT, type, dest, sn,
-				 FOP_TO_IDX(op), sm, m);
-
-		except = fop->fn(state, dest, sn, m, fpscr);
-		pr_debug("VFP: itr%d: exceptions=%08x\n",
-			 vecitr >> FPSCR_LENGTH_BIT, except);
-
-		exceptions |= except;
-
-		/*
-		 * CHECK: It appears to be undefined whether we stop when
-		 * we encounter an exception.  We continue.
-		 */
-		dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7);
-		sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7);
-		if (FREG_BANK(sm) != 0)
-			sm = FREG_BANK(sm) + ((FREG_IDX(sm) + vecstride) & 7);
-	}
-	return exceptions;
-
- invalid:
-	return (u32)-1;
+    u32 op = inst & FOP_MASK;
+    u32 exceptions = 0;
+    unsigned int dest;
+    unsigned int sn = vfp_get_sn(inst);
+    unsigned int sm = vfp_get_sm(inst);
+    unsigned int vecitr, veclen, vecstride;
+    struct op *fop;
+    pr_debug("In %s\n", __FUNCTION__);
+
+    vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK);
+
+    fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)];
+
+    /*
+     * fcvtsd takes a dN register number as destination, not sN.
+     * Technically, if bit 0 of dd is set, this is an invalid
+     * instruction.  However, we ignore this for efficiency.
+     * It also only operates on scalars.
+     */
+    if (fop->flags & OP_DD)
+        dest = vfp_get_dd(inst);
+    else
+        dest = vfp_get_sd(inst);
+
+    /*
+     * If destination bank is zero, vector length is always '1'.
+     * ARM DDI0100F C5.1.3, C5.3.2.
+     */
+    if ((fop->flags & OP_SCALAR) || FREG_BANK(dest) == 0)
+        veclen = 0;
+    else
+        veclen = fpscr & FPSCR_LENGTH_MASK;
+
+    pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
+             (veclen >> FPSCR_LENGTH_BIT) + 1);
+
+    if (!fop->fn) {
+        printf("VFP: could not find single op %d, inst=0x%x@0x%x\n", FEXT_TO_IDX(inst), inst, state->Reg[15]);
+        exit(-1);
+        goto invalid;
+    }
+
+    for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
+        s32 m = vfp_get_float(state, sm);
+        u32 except;
+        char type;
+
+        type = fop->flags & OP_DD ? 'd' : 's';
+        if (op == FOP_EXT)
+            pr_debug("VFP: itr%d (%c%u) = op[%u] (s%u=%08x)\n",
+                     vecitr >> FPSCR_LENGTH_BIT, type, dest, sn,
+                     sm, m);
+        else
+            pr_debug("VFP: itr%d (%c%u) = (s%u) op[%u] (s%u=%08x)\n",
+                     vecitr >> FPSCR_LENGTH_BIT, type, dest, sn,
+                     FOP_TO_IDX(op), sm, m);
+
+        except = fop->fn(state, dest, sn, m, fpscr);
+        pr_debug("VFP: itr%d: exceptions=%08x\n",
+                 vecitr >> FPSCR_LENGTH_BIT, except);
+
+        exceptions |= except;
+
+        /*
+         * CHECK: It appears to be undefined whether we stop when
+         * we encounter an exception.  We continue.
+         */
+        dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7);
+        sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7);
+        if (FREG_BANK(sm) != 0)
+            sm = FREG_BANK(sm) + ((FREG_IDX(sm) + vecstride) & 7);
+    }
+    return exceptions;
+
+invalid:
+    return (u32)-1;
 }