[PATCH] ARM: 2837/2: Re: ARM: Make NWFPE preempt safe

Patch from Richard Purdie

NWFPE used global variables which meant it wasn't safe for use with
preemptive kernels. This patch removes them and communicates the
information between functions in a preempt safe manner. Generation
of some exceptions was broken and this has also been corrected.
Tests with glibc's maths test suite show no change in the results
before/after this patch.

Signed-off-by: Richard Purdie <rpurdie@rpsys.net>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

11 files changed, 304 insertions, 304 deletions

diff --git a/arch/arm/nwfpe/double_cpdo.c b/arch/arm/nwfpe/double_cpdo.c
index 7ffd8cb..c51d138 100644
--- a/arch/arm/nwfpe/double_cpdo.c
+++ b/arch/arm/nwfpe/double_cpdo.c
@@ -40,17 +40,17 @@ float64 float64_arccos(float64 rFm);
 float64 float64_pow(float64 rFn, float64 rFm);
 float64 float64_pol(float64 rFn, float64 rFm);
 
-static float64 float64_rsf(float64 rFn, float64 rFm)
+static float64 float64_rsf(struct roundingData *roundData, float64 rFn, float64 rFm)
 {
-	return float64_sub(rFm, rFn);
+	return float64_sub(roundData, rFm, rFn);
 }
 
-static float64 float64_rdv(float64 rFn, float64 rFm)
+static float64 float64_rdv(struct roundingData *roundData, float64 rFn, float64 rFm)
 {
-	return float64_div(rFm, rFn);
+	return float64_div(roundData, rFm, rFn);
 }
 
-static float64 (*const dyadic_double[16])(float64 rFn, float64 rFm) = {
+static float64 (*const dyadic_double[16])(struct roundingData*, float64 rFn, float64 rFm) = {
 	[ADF_CODE >> 20] = float64_add,
 	[MUF_CODE >> 20] = float64_mul,
 	[SUF_CODE >> 20] = float64_sub,
@@ -65,12 +65,12 @@ static float64 (*const dyadic_double[16])(float64 rFn, float64 rFm) = {
 	[FRD_CODE >> 20] = float64_rdv,
 };
 
-static float64 float64_mvf(float64 rFm)
+static float64 float64_mvf(struct roundingData *roundData,float64 rFm)
 {
 	return rFm;
 }
 
-static float64 float64_mnf(float64 rFm)
+static float64 float64_mnf(struct roundingData *roundData,float64 rFm)
 {
 	union float64_components u;
 
@@ -84,7 +84,7 @@ static float64 float64_mnf(float64 rFm)
 	return u.f64;
 }
 
-static float64 float64_abs(float64 rFm)
+static float64 float64_abs(struct roundingData *roundData,float64 rFm)
 {
 	union float64_components u;
 
@@ -98,7 +98,7 @@ static float64 float64_abs(float64 rFm)
 	return u.f64;
 }
 
-static float64 (*const monadic_double[16])(float64 rFm) = {
+static float64 (*const monadic_double[16])(struct roundingData *, float64 rFm) = {
 	[MVF_CODE >> 20] = float64_mvf,
 	[MNF_CODE >> 20] = float64_mnf,
 	[ABS_CODE >> 20] = float64_abs,
@@ -108,7 +108,7 @@ static float64 (*const monadic_double[16])(float64 rFm) = {
 	[NRM_CODE >> 20] = float64_mvf,
 };
 
-unsigned int DoubleCPDO(const unsigned int opcode, FPREG * rFd)
+unsigned int DoubleCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd)
 {
 	FPA11 *fpa11 = GET_FPA11();
 	float64 rFm;
@@ -151,13 +151,13 @@ unsigned int DoubleCPDO(const unsigned int opcode, FPREG * rFd)
 		}
 
 		if (dyadic_double[opc_mask_shift]) {
-			rFd->fDouble = dyadic_double[opc_mask_shift](rFn, rFm);
+			rFd->fDouble = dyadic_double[opc_mask_shift](roundData, rFn, rFm);
 		} else {
 			return 0;
 		}
 	} else {
 		if (monadic_double[opc_mask_shift]) {
-			rFd->fDouble = monadic_double[opc_mask_shift](rFm);
+			rFd->fDouble = monadic_double[opc_mask_shift](roundData, rFm);
 		} else {
 			return 0;
 		}
diff --git a/arch/arm/nwfpe/extended_cpdo.c b/arch/arm/nwfpe/extended_cpdo.c
index c39f68a..65a279b 100644
--- a/arch/arm/nwfpe/extended_cpdo.c
+++ b/arch/arm/nwfpe/extended_cpdo.c
@@ -35,17 +35,17 @@ floatx80 floatx80_arccos(floatx80 rFm);
 floatx80 floatx80_pow(floatx80 rFn, floatx80 rFm);
 floatx80 floatx80_pol(floatx80 rFn, floatx80 rFm);
 
-static floatx80 floatx80_rsf(floatx80 rFn, floatx80 rFm)
+static floatx80 floatx80_rsf(struct roundingData *roundData, floatx80 rFn, floatx80 rFm)
 {
-	return floatx80_sub(rFm, rFn);
+	return floatx80_sub(roundData, rFm, rFn);
 }
 
-static floatx80 floatx80_rdv(floatx80 rFn, floatx80 rFm)
+static floatx80 floatx80_rdv(struct roundingData *roundData, floatx80 rFn, floatx80 rFm)
 {
-	return floatx80_div(rFm, rFn);
+	return floatx80_div(roundData, rFm, rFn);
 }
 
-static floatx80 (*const dyadic_extended[16])(floatx80 rFn, floatx80 rFm) = {
+static floatx80 (*const dyadic_extended[16])(struct roundingData*, floatx80 rFn, floatx80 rFm) = {
 	[ADF_CODE >> 20] = floatx80_add,
 	[MUF_CODE >> 20] = floatx80_mul,
 	[SUF_CODE >> 20] = floatx80_sub,
@@ -60,24 +60,24 @@ static floatx80 (*const dyadic_extended[16])(floatx80 rFn, floatx80 rFm) = {
 	[FRD_CODE >> 20] = floatx80_rdv,
 };
 
-static floatx80 floatx80_mvf(floatx80 rFm)
+static floatx80 floatx80_mvf(struct roundingData *roundData, floatx80 rFm)
 {
 	return rFm;
 }
 
-static floatx80 floatx80_mnf(floatx80 rFm)
+static floatx80 floatx80_mnf(struct roundingData *roundData, floatx80 rFm)
 {
 	rFm.high ^= 0x8000;
 	return rFm;
 }
 
-static floatx80 floatx80_abs(floatx80 rFm)
+static floatx80 floatx80_abs(struct roundingData *roundData, floatx80 rFm)
 {
 	rFm.high &= 0x7fff;
 	return rFm;
 }
 
-static floatx80 (*const monadic_extended[16])(floatx80 rFm) = {
+static floatx80 (*const monadic_extended[16])(struct roundingData*, floatx80 rFm) = {
 	[MVF_CODE >> 20] = floatx80_mvf,
 	[MNF_CODE >> 20] = floatx80_mnf,
 	[ABS_CODE >> 20] = floatx80_abs,
@@ -87,7 +87,7 @@ static floatx80 (*const monadic_extended[16])(floatx80 rFm) = {
 	[NRM_CODE >> 20] = floatx80_mvf,
 };
 
-unsigned int ExtendedCPDO(const unsigned int opcode, FPREG * rFd)
+unsigned int ExtendedCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd)
 {
 	FPA11 *fpa11 = GET_FPA11();
 	floatx80 rFm;
@@ -138,13 +138,13 @@ unsigned int ExtendedCPDO(const unsigned int opcode, FPREG * rFd)
 		}
 
 		if (dyadic_extended[opc_mask_shift]) {
-			rFd->fExtended = dyadic_extended[opc_mask_shift](rFn, rFm);
+			rFd->fExtended = dyadic_extended[opc_mask_shift](roundData, rFn, rFm);
 		} else {
 			return 0;
 		}
 	} else {
 		if (monadic_extended[opc_mask_shift]) {
-			rFd->fExtended = monadic_extended[opc_mask_shift](rFm);
+			rFd->fExtended = monadic_extended[opc_mask_shift](roundData, rFm);
 		} else {
 			return 0;
 		}
diff --git a/arch/arm/nwfpe/fpa11.c b/arch/arm/nwfpe/fpa11.c
index bf61696..7690f73 100644
--- a/arch/arm/nwfpe/fpa11.c
+++ b/arch/arm/nwfpe/fpa11.c
@@ -51,48 +51,42 @@ static void resetFPA11(void)
 	fpa11->fpsr = FP_EMULATOR | BIT_AC;
 }
 
-void SetRoundingMode(const unsigned int opcode)
+int8 SetRoundingMode(const unsigned int opcode)
 {
 	switch (opcode & MASK_ROUNDING_MODE) {
 	default:
 	case ROUND_TO_NEAREST:
-		float_rounding_mode = float_round_nearest_even;
-		break;
+		return float_round_nearest_even;
 
 	case ROUND_TO_PLUS_INFINITY:
-		float_rounding_mode = float_round_up;
-		break;
+		return float_round_up;
 
 	case ROUND_TO_MINUS_INFINITY:
-		float_rounding_mode = float_round_down;
-		break;
+		return float_round_down;
 
 	case ROUND_TO_ZERO:
-		float_rounding_mode = float_round_to_zero;
-		break;
+		return float_round_to_zero;
 	}
 }
 
-void SetRoundingPrecision(const unsigned int opcode)
+int8 SetRoundingPrecision(const unsigned int opcode)
 {
 #ifdef CONFIG_FPE_NWFPE_XP
 	switch (opcode & MASK_ROUNDING_PRECISION) {
 	case ROUND_SINGLE:
-		floatx80_rounding_precision = 32;
-		break;
+		return 32;
 
 	case ROUND_DOUBLE:
-		floatx80_rounding_precision = 64;
-		break;
+		return 64;
 
 	case ROUND_EXTENDED:
-		floatx80_rounding_precision = 80;
-		break;
+		return 80;
 
 	default:
-		floatx80_rounding_precision = 80;
+		return 80;
 	}
 #endif
+	return 80;
 }
 
 void nwfpe_init_fpa(union fp_state *fp)
@@ -103,8 +97,6 @@ void nwfpe_init_fpa(union fp_state *fp)
 #endif
  	memset(fpa11, 0, sizeof(FPA11));
 	resetFPA11();
-	SetRoundingMode(ROUND_TO_NEAREST);
-	SetRoundingPrecision(ROUND_EXTENDED);
 	fpa11->initflag = 1;
 }
 
diff --git a/arch/arm/nwfpe/fpa11.h b/arch/arm/nwfpe/fpa11.h
index e4a61ae..93523ae 100644
--- a/arch/arm/nwfpe/fpa11.h
+++ b/arch/arm/nwfpe/fpa11.h
@@ -37,6 +37,13 @@
 /* includes */
 #include "fpsr.h"		/* FP control and status register definitions */
 #include "milieu.h"
+
+struct roundingData {
+    int8 mode;
+    int8 precision;
+    signed char exception;
+};
+
 #include "softfloat.h"
 
 #define		typeNone		0x00
@@ -84,8 +91,8 @@ typedef struct tagFPA11 {
 				   initialised. */
 } FPA11;
 
-extern void SetRoundingMode(const unsigned int);
-extern void SetRoundingPrecision(const unsigned int);
+extern int8 SetRoundingMode(const unsigned int);
+extern int8 SetRoundingPrecision(const unsigned int);
 extern void nwfpe_init_fpa(union fp_state *fp);
 
 #endif
diff --git a/arch/arm/nwfpe/fpa11_cpdo.c b/arch/arm/nwfpe/fpa11_cpdo.c
index 1bea674..4a31dfd 100644
--- a/arch/arm/nwfpe/fpa11_cpdo.c
+++ b/arch/arm/nwfpe/fpa11_cpdo.c
@@ -24,15 +24,16 @@
 #include "fpa11.h"
 #include "fpopcode.h"
 
-unsigned int SingleCPDO(const unsigned int opcode, FPREG * rFd);
-unsigned int DoubleCPDO(const unsigned int opcode, FPREG * rFd);
-unsigned int ExtendedCPDO(const unsigned int opcode, FPREG * rFd);
+unsigned int SingleCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd);
+unsigned int DoubleCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd);
+unsigned int ExtendedCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd);
 
 unsigned int EmulateCPDO(const unsigned int opcode)
 {
 	FPA11 *fpa11 = GET_FPA11();
 	FPREG *rFd;
 	unsigned int nType, nDest, nRc;
+	struct roundingData roundData;
 
 	/* Get the destination size.  If not valid let Linux perform
 	   an invalid instruction trap. */
@@ -40,7 +41,9 @@ unsigned int EmulateCPDO(const unsigned int opcode)
 	if (typeNone == nDest)
 		return 0;
 
-	SetRoundingMode(opcode);
+	roundData.mode = SetRoundingMode(opcode);
+	roundData.precision = SetRoundingPrecision(opcode);
+	roundData.exception = 0;
 
 	/* Compare the size of the operands in Fn and Fm.
 	   Choose the largest size and perform operations in that size,
@@ -63,14 +66,14 @@ unsigned int EmulateCPDO(const unsigned int opcode)
 
 	switch (nType) {
 	case typeSingle:
-		nRc = SingleCPDO(opcode, rFd);
+		nRc = SingleCPDO(&roundData, opcode, rFd);
 		break;
 	case typeDouble:
-		nRc = DoubleCPDO(opcode, rFd);
+		nRc = DoubleCPDO(&roundData, opcode, rFd);
 		break;
 #ifdef CONFIG_FPE_NWFPE_XP
 	case typeExtended:
-		nRc = ExtendedCPDO(opcode, rFd);
+		nRc = ExtendedCPDO(&roundData, opcode, rFd);
 		break;
 #endif
 	default:
@@ -93,9 +96,9 @@ unsigned int EmulateCPDO(const unsigned int opcode)
 			case typeSingle:
 				{
 					if (typeDouble == nType)
-						rFd->fSingle = float64_to_float32(rFd->fDouble);
+						rFd->fSingle = float64_to_float32(&roundData, rFd->fDouble);
 					else
-						rFd->fSingle = floatx80_to_float32(rFd->fExtended);
+						rFd->fSingle = floatx80_to_float32(&roundData, rFd->fExtended);
 				}
 				break;
 
@@ -104,7 +107,7 @@ unsigned int EmulateCPDO(const unsigned int opcode)
 					if (typeSingle == nType)
 						rFd->fDouble = float32_to_float64(rFd->fSingle);
 					else
-						rFd->fDouble = floatx80_to_float64(rFd->fExtended);
+						rFd->fDouble = floatx80_to_float64(&roundData, rFd->fExtended);
 				}
 				break;
 
@@ -121,12 +124,15 @@ unsigned int EmulateCPDO(const unsigned int opcode)
 #else
 		if (nDest != nType) {
 			if (nDest == typeSingle)
-				rFd->fSingle = float64_to_float32(rFd->fDouble);
+				rFd->fSingle = float64_to_float32(&roundData, rFd->fDouble);
 			else
 				rFd->fDouble = float32_to_float64(rFd->fSingle);
 		}
 #endif
 	}
 
+	if (roundData.exception)
+		float_raise(roundData.exception);
+
 	return nRc;
 }
diff --git a/arch/arm/nwfpe/fpa11_cpdt.c b/arch/arm/nwfpe/fpa11_cpdt.c
index 95fb63f..b0db5cb 100644
--- a/arch/arm/nwfpe/fpa11_cpdt.c
+++ b/arch/arm/nwfpe/fpa11_cpdt.c
@@ -96,7 +96,7 @@ static inline void loadMultiple(const unsigned int Fn, const unsigned int __user
 	}
 }
 
-static inline void storeSingle(const unsigned int Fn, unsigned int __user *pMem)
+static inline void storeSingle(struct roundingData *roundData, const unsigned int Fn, unsigned int __user *pMem)
 {
 	FPA11 *fpa11 = GET_FPA11();
 	union {
@@ -106,12 +106,12 @@ static inline void storeSingle(const unsigned int Fn, unsigned int __user *pMem)
 
 	switch (fpa11->fType[Fn]) {
 	case typeDouble:
-		val.f = float64_to_float32(fpa11->fpreg[Fn].fDouble);
+		val.f = float64_to_float32(roundData, fpa11->fpreg[Fn].fDouble);
 		break;
 
 #ifdef CONFIG_FPE_NWFPE_XP
 	case typeExtended:
-		val.f = floatx80_to_float32(fpa11->fpreg[Fn].fExtended);
+		val.f = floatx80_to_float32(roundData, fpa11->fpreg[Fn].fExtended);
 		break;
 #endif
 
@@ -122,7 +122,7 @@ static inline void storeSingle(const unsigned int Fn, unsigned int __user *pMem)
 	put_user(val.i[0], pMem);
 }
 
-static inline void storeDouble(const unsigned int Fn, unsigned int __user *pMem)
+static inline void storeDouble(struct roundingData *roundData, const unsigned int Fn, unsigned int __user *pMem)
 {
 	FPA11 *fpa11 = GET_FPA11();
 	union {
@@ -137,7 +137,7 @@ static inline void storeDouble(const unsigned int Fn, unsigned int __user *pMem)
 
 #ifdef CONFIG_FPE_NWFPE_XP
 	case typeExtended:
-		val.f = floatx80_to_float64(fpa11->fpreg[Fn].fExtended);
+		val.f = floatx80_to_float64(roundData, fpa11->fpreg[Fn].fExtended);
 		break;
 #endif
 
@@ -259,8 +259,11 @@ unsigned int PerformSTF(const unsigned int opcode)
 {
 	unsigned int __user *pBase, *pAddress, *pFinal;
 	unsigned int nRc = 1, write_back = WRITE_BACK(opcode);
+	struct roundingData roundData;
 
-	SetRoundingMode(ROUND_TO_NEAREST);
+	roundData.mode = SetRoundingMode(opcode);
+	roundData.precision = SetRoundingPrecision(opcode);
+	roundData.exception = 0;
 
 	pBase = (unsigned int __user *) readRegister(getRn(opcode));
 	if (REG_PC == getRn(opcode)) {
@@ -281,10 +284,10 @@ unsigned int PerformSTF(const unsigned int opcode)
 
 	switch (opcode & MASK_TRANSFER_LENGTH) {
 	case TRANSFER_SINGLE:
-		storeSingle(getFd(opcode), pAddress);
+		storeSingle(&roundData, getFd(opcode), pAddress);
 		break;
 	case TRANSFER_DOUBLE:
-		storeDouble(getFd(opcode), pAddress);
+		storeDouble(&roundData, getFd(opcode), pAddress);
 		break;
 #ifdef CONFIG_FPE_NWFPE_XP
 	case TRANSFER_EXTENDED:
@@ -295,6 +298,9 @@ unsigned int PerformSTF(const unsigned int opcode)
 		nRc = 0;
 	}
 
+	if (roundData.exception)
+		float_raise(roundData.exception);
+
 	if (write_back)
 		writeRegister(getRn(opcode), (unsigned long) pFinal);
 	return nRc;
diff --git a/arch/arm/nwfpe/fpa11_cprt.c b/arch/arm/nwfpe/fpa11_cprt.c
index db01fbc..adf8d30 100644
--- a/arch/arm/nwfpe/fpa11_cprt.c
+++ b/arch/arm/nwfpe/fpa11_cprt.c
@@ -33,8 +33,6 @@ extern flag floatx80_is_nan(floatx80);
 extern flag float64_is_nan(float64);
 extern flag float32_is_nan(float32);
 
-void SetRoundingMode(const unsigned int opcode);
-
 unsigned int PerformFLT(const unsigned int opcode);
 unsigned int PerformFIX(const unsigned int opcode);
 
@@ -77,14 +75,17 @@ unsigned int EmulateCPRT(const unsigned int opcode)
 unsigned int PerformFLT(const unsigned int opcode)
 {
 	FPA11 *fpa11 = GET_FPA11();
-	SetRoundingMode(opcode);
-	SetRoundingPrecision(opcode);
+	struct roundingData roundData;
+
+	roundData.mode = SetRoundingMode(opcode);
+	roundData.precision = SetRoundingPrecision(opcode);
+	roundData.exception = 0;
 
 	switch (opcode & MASK_ROUNDING_PRECISION) {
 	case ROUND_SINGLE:
 		{
 			fpa11->fType[getFn(opcode)] = typeSingle;
-			fpa11->fpreg[getFn(opcode)].fSingle = int32_to_float32(readRegister(getRd(opcode)));
+			fpa11->fpreg[getFn(opcode)].fSingle = int32_to_float32(&roundData, readRegister(getRd(opcode)));
 		}
 		break;
 
@@ -108,6 +109,9 @@ unsigned int PerformFLT(const unsigned int opcode)
 		return 0;
 	}
 
+	if (roundData.exception)
+		float_raise(roundData.exception);
+
 	return 1;
 }
 
@@ -115,26 +119,29 @@ unsigned int PerformFIX(const unsigned int opcode)
 {
 	FPA11 *fpa11 = GET_FPA11();
 	unsigned int Fn = getFm(opcode);
+	struct roundingData roundData;
 
-	SetRoundingMode(opcode);
+	roundData.mode = SetRoundingMode(opcode);
+	roundData.precision = SetRoundingPrecision(opcode);
+	roundData.exception = 0;
 
 	switch (fpa11->fType[Fn]) {
 	case typeSingle:
 		{
-			writeRegister(getRd(opcode), float32_to_int32(fpa11->fpreg[Fn].fSingle));
+			writeRegister(getRd(opcode), float32_to_int32(&roundData, fpa11->fpreg[Fn].fSingle));
 		}
 		break;
 
 	case typeDouble:
 		{
-			writeRegister(getRd(opcode), float64_to_int32(fpa11->fpreg[Fn].fDouble));
+			writeRegister(getRd(opcode), float64_to_int32(&roundData, fpa11->fpreg[Fn].fDouble));
 		}
 		break;
 
 #ifdef CONFIG_FPE_NWFPE_XP
 	case typeExtended:
 		{
-			writeRegister(getRd(opcode), floatx80_to_int32(fpa11->fpreg[Fn].fExtended));
+			writeRegister(getRd(opcode), floatx80_to_int32(&roundData, fpa11->fpreg[Fn].fExtended));
 		}
 		break;
 #endif
@@ -143,6 +150,9 @@ unsigned int PerformFIX(const unsigned int opcode)
 		return 0;
 	}
 
+	if (roundData.exception)
+		float_raise(roundData.exception);
+
 	return 1;
 }
 
diff --git a/arch/arm/nwfpe/fpmodule.c b/arch/arm/nwfpe/fpmodule.c
index 12885f3..2dfe1ac 100644
--- a/arch/arm/nwfpe/fpmodule.c
+++ b/arch/arm/nwfpe/fpmodule.c
@@ -116,8 +116,6 @@ fpmodule.c to integrate with the NetBSD kernel (I hope!).
 code to access data in user space in some other source files at the 
 moment (grep for get_user / put_user calls).  --philb]
 
-float_exception_flags is a global variable in SoftFloat.
-
 This function is called by the SoftFloat routines to raise a floating
 point exception.  We check the trap enable byte in the FPSR, and raise
 a SIGFPE exception if necessary.  If not the relevant bits in the 
@@ -129,15 +127,14 @@ void float_raise(signed char flags)
 	register unsigned int fpsr, cumulativeTraps;
 
 #ifdef CONFIG_DEBUG_USER
-	printk(KERN_DEBUG
-	       "NWFPE: %s[%d] takes exception %08x at %p from %08lx\n",
-	       current->comm, current->pid, flags,
-	       __builtin_return_address(0), GET_USERREG()->ARM_pc);
+ 	/* Ignore inexact errors as there are far too many of them to log */
+ 	if (flags & ~BIT_IXC)
+ 		printk(KERN_DEBUG
+		       "NWFPE: %s[%d] takes exception %08x at %p from %08lx\n",
+		       current->comm, current->pid, flags,
+		       __builtin_return_address(0), GET_USERREG()->ARM_pc);
 #endif
 
-	/* Keep SoftFloat exception flags up to date.  */
-	float_exception_flags |= flags;
-
 	/* Read fpsr and initialize the cumulativeTraps.  */
 	fpsr = readFPSR();
 	cumulativeTraps = 0;
diff --git a/arch/arm/nwfpe/single_cpdo.c b/arch/arm/nwfpe/single_cpdo.c
index 705808e..c66981d 100644
--- a/arch/arm/nwfpe/single_cpdo.c
+++ b/arch/arm/nwfpe/single_cpdo.c
@@ -36,17 +36,17 @@ float32 float32_arccos(float32 rFm);
 float32 float32_pow(float32 rFn, float32 rFm);
 float32 float32_pol(float32 rFn, float32 rFm);
 
-static float32 float32_rsf(float32 rFn, float32 rFm)
+static float32 float32_rsf(struct roundingData *roundData, float32 rFn, float32 rFm)
 {
-	return float32_sub(rFm, rFn);
+	return float32_sub(roundData, rFm, rFn);
 }
 
-static float32 float32_rdv(float32 rFn, float32 rFm)
+static float32 float32_rdv(struct roundingData *roundData, float32 rFn, float32 rFm)
 {
-	return float32_div(rFm, rFn);
+	return float32_div(roundData, rFm, rFn);
 }
 
-static float32 (*const dyadic_single[16])(float32 rFn, float32 rFm) = {
+static float32 (*const dyadic_single[16])(struct roundingData *, float32 rFn, float32 rFm) = {
 	[ADF_CODE >> 20] = float32_add,
 	[MUF_CODE >> 20] = float32_mul,
 	[SUF_CODE >> 20] = float32_sub,
@@ -60,22 +60,22 @@ static float32 (*const dyadic_single[16])(float32 rFn, float32 rFm) = {
 	[FRD_CODE >> 20] = float32_rdv,
 };
 
-static float32 float32_mvf(float32 rFm)
+static float32 float32_mvf(struct roundingData *roundData, float32 rFm)
 {
 	return rFm;
 }
 
-static float32 float32_mnf(float32 rFm)
+static float32 float32_mnf(struct roundingData *roundData, float32 rFm)
 {
 	return rFm ^ 0x80000000;
 }
 
-static float32 float32_abs(float32 rFm)
+static float32 float32_abs(struct roundingData *roundData, float32 rFm)
 {
 	return rFm & 0x7fffffff;
 }
 
-static float32 (*const monadic_single[16])(float32 rFm) = {
+static float32 (*const monadic_single[16])(struct roundingData*, float32 rFm) = {
 	[MVF_CODE >> 20] = float32_mvf,
 	[MNF_CODE >> 20] = float32_mnf,
 	[ABS_CODE >> 20] = float32_abs,
@@ -85,7 +85,7 @@ static float32 (*const monadic_single[16])(float32 rFm) = {
 	[NRM_CODE >> 20] = float32_mvf,
 };
 
-unsigned int SingleCPDO(const unsigned int opcode, FPREG * rFd)
+unsigned int SingleCPDO(struct roundingData *roundData, const unsigned int opcode, FPREG * rFd)
 {
 	FPA11 *fpa11 = GET_FPA11();
 	float32 rFm;
@@ -108,13 +108,13 @@ unsigned int SingleCPDO(const unsigned int opcode, FPREG * rFd)
 		if (fpa11->fType[Fn] == typeSingle &&
 		    dyadic_single[opc_mask_shift]) {
 			rFn = fpa11->fpreg[Fn].fSingle;
-			rFd->fSingle = dyadic_single[opc_mask_shift](rFn, rFm);
+			rFd->fSingle = dyadic_single[opc_mask_shift](roundData, rFn, rFm);
 		} else {
 			return 0;
 		}
 	} else {
 		if (monadic_single[opc_mask_shift]) {
-			rFd->fSingle = monadic_single[opc_mask_shift](rFm);
+			rFd->fSingle = monadic_single[opc_mask_shift](roundData, rFm);
 		} else {
 			return 0;
 		}
diff --git a/arch/arm/nwfpe/softfloat.c b/arch/arm/nwfpe/softfloat.c
index e038dd3..8b75a6e 100644
--- a/arch/arm/nwfpe/softfloat.c
+++ b/arch/arm/nwfpe/softfloat.c
@@ -36,16 +36,6 @@ this code that are retained.
 
 /*
 -------------------------------------------------------------------------------
-Floating-point rounding mode, extended double-precision rounding precision,
-and exception flags.
--------------------------------------------------------------------------------
-*/
-int8 float_rounding_mode = float_round_nearest_even;
-int8 floatx80_rounding_precision = 80;
-int8 float_exception_flags;
-
-/*
--------------------------------------------------------------------------------
 Primitive arithmetic functions, including multi-word arithmetic, and
 division and square root approximations.  (Can be specialized to target if
 desired.)
@@ -77,14 +67,14 @@ input is too large, however, the invalid exception is raised and the largest
 positive or negative integer is returned.
 -------------------------------------------------------------------------------
 */
-static int32 roundAndPackInt32( flag zSign, bits64 absZ )
+static int32 roundAndPackInt32( struct roundingData *roundData, flag zSign, bits64 absZ )
 {
     int8 roundingMode;
     flag roundNearestEven;
     int8 roundIncrement, roundBits;
     int32 z;
 
-    roundingMode = float_rounding_mode;
+    roundingMode = roundData->mode;
     roundNearestEven = ( roundingMode == float_round_nearest_even );
     roundIncrement = 0x40;
     if ( ! roundNearestEven ) {
@@ -107,10 +97,10 @@ static int32 roundAndPackInt32( flag zSign, bits64 absZ )
     z = absZ;
     if ( zSign ) z = - z;
     if ( ( absZ>>32 ) || ( z && ( ( z < 0 ) ^ zSign ) ) ) {
-        float_exception_flags |= float_flag_invalid;
+        roundData->exception |= float_flag_invalid;
         return zSign ? 0x80000000 : 0x7FFFFFFF;
     }
-    if ( roundBits ) float_exception_flags |= float_flag_inexact;
+    if ( roundBits ) roundData->exception |= float_flag_inexact;
     return z;
 
 }
@@ -224,14 +214,14 @@ The handling of underflow and overflow follows the IEC/IEEE Standard for
 Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-static float32 roundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig )
+static float32 roundAndPackFloat32( struct roundingData *roundData, flag zSign, int16 zExp, bits32 zSig )
 {
     int8 roundingMode;
     flag roundNearestEven;
     int8 roundIncrement, roundBits;
     flag isTiny;
 
-    roundingMode = float_rounding_mode;
+    roundingMode = roundData->mode;
     roundNearestEven = ( roundingMode == float_round_nearest_even );
     roundIncrement = 0x40;
     if ( ! roundNearestEven ) {
@@ -254,7 +244,7 @@ static float32 roundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig )
              || (    ( zExp == 0xFD )
                   && ( (sbits32) ( zSig + roundIncrement ) < 0 ) )
            ) {
-            float_raise( float_flag_overflow | float_flag_inexact );
+            roundData->exception |= float_flag_overflow | float_flag_inexact;
             return packFloat32( zSign, 0xFF, 0 ) - ( roundIncrement == 0 );
         }
         if ( zExp < 0 ) {
@@ -265,10 +255,10 @@ static float32 roundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig )
             shift32RightJamming( zSig, - zExp, &zSig );
             zExp = 0;
             roundBits = zSig & 0x7F;
-            if ( isTiny && roundBits ) float_raise( float_flag_underflow );
+            if ( isTiny && roundBits ) roundData->exception |= float_flag_underflow;
         }
     }
-    if ( roundBits ) float_exception_flags |= float_flag_inexact;
+    if ( roundBits ) roundData->exception |= float_flag_inexact;
     zSig = ( zSig + roundIncrement )>>7;
     zSig &= ~ ( ( ( roundBits ^ 0x40 ) == 0 ) & roundNearestEven );
     if ( zSig == 0 ) zExp = 0;
@@ -287,12 +277,12 @@ point exponent.
 -------------------------------------------------------------------------------
 */
 static float32
- normalizeRoundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig )
+ normalizeRoundAndPackFloat32( struct roundingData *roundData, flag zSign, int16 zExp, bits32 zSig )
 {
     int8 shiftCount;
 
     shiftCount = countLeadingZeros32( zSig ) - 1;
-    return roundAndPackFloat32( zSign, zExp - shiftCount, zSig<<shiftCount );
+    return roundAndPackFloat32( roundData, zSign, zExp - shiftCount, zSig<<shiftCount );
 
 }
 
@@ -395,14 +385,14 @@ The handling of underflow and overflow follows the IEC/IEEE Standard for
 Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-static float64 roundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig )
+static float64 roundAndPackFloat64( struct roundingData *roundData, flag zSign, int16 zExp, bits64 zSig )
 {
     int8 roundingMode;
     flag roundNearestEven;
     int16 roundIncrement, roundBits;
     flag isTiny;
 
-    roundingMode = float_rounding_mode;
+    roundingMode = roundData->mode;
     roundNearestEven = ( roundingMode == float_round_nearest_even );
     roundIncrement = 0x200;
     if ( ! roundNearestEven ) {
@@ -427,7 +417,7 @@ static float64 roundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig )
            ) {
             //register int lr = __builtin_return_address(0);
             //printk("roundAndPackFloat64 called from 0x%08x\n",lr);
-            float_raise( float_flag_overflow | float_flag_inexact );
+            roundData->exception |= float_flag_overflow | float_flag_inexact;
             return packFloat64( zSign, 0x7FF, 0 ) - ( roundIncrement == 0 );
         }
         if ( zExp < 0 ) {
@@ -438,10 +428,10 @@ static float64 roundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig )
             shift64RightJamming( zSig, - zExp, &zSig );
             zExp = 0;
             roundBits = zSig & 0x3FF;
-            if ( isTiny && roundBits ) float_raise( float_flag_underflow );
+            if ( isTiny && roundBits ) roundData->exception |= float_flag_underflow;
         }
     }
-    if ( roundBits ) float_exception_flags |= float_flag_inexact;
+    if ( roundBits ) roundData->exception |= float_flag_inexact;
     zSig = ( zSig + roundIncrement )>>10;
     zSig &= ~ ( ( ( roundBits ^ 0x200 ) == 0 ) & roundNearestEven );
     if ( zSig == 0 ) zExp = 0;
@@ -460,12 +450,12 @@ point exponent.
 -------------------------------------------------------------------------------
 */
 static float64
- normalizeRoundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig )
+ normalizeRoundAndPackFloat64( struct roundingData *roundData, flag zSign, int16 zExp, bits64 zSig )
 {
     int8 shiftCount;
 
     shiftCount = countLeadingZeros64( zSig ) - 1;
-    return roundAndPackFloat64( zSign, zExp - shiftCount, zSig<<shiftCount );
+    return roundAndPackFloat64( roundData, zSign, zExp - shiftCount, zSig<<shiftCount );
 
 }
 
@@ -572,14 +562,15 @@ Floating-point Arithmetic.
 */
 static floatx80
  roundAndPackFloatx80(
-     int8 roundingPrecision, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1
+     struct roundingData *roundData, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1
  )
 {
-    int8 roundingMode;
+    int8 roundingMode, roundingPrecision;
     flag roundNearestEven, increment, isTiny;
     int64 roundIncrement, roundMask, roundBits;
 
-    roundingMode = float_rounding_mode;
+    roundingMode = roundData->mode;
+    roundingPrecision = roundData->precision;
     roundNearestEven = ( roundingMode == float_round_nearest_even );
     if ( roundingPrecision == 80 ) goto precision80;
     if ( roundingPrecision == 64 ) {
@@ -623,8 +614,8 @@ static floatx80
             shift64RightJamming( zSig0, 1 - zExp, &zSig0 );
             zExp = 0;
             roundBits = zSig0 & roundMask;
-            if ( isTiny && roundBits ) float_raise( float_flag_underflow );
-            if ( roundBits ) float_exception_flags |= float_flag_inexact;
+            if ( isTiny && roundBits ) roundData->exception |= float_flag_underflow;
+            if ( roundBits ) roundData->exception |= float_flag_inexact;
             zSig0 += roundIncrement;
             if ( (sbits64) zSig0 < 0 ) zExp = 1;
             roundIncrement = roundMask + 1;
@@ -635,7 +626,7 @@ static floatx80
             return packFloatx80( zSign, zExp, zSig0 );
         }
     }
-    if ( roundBits ) float_exception_flags |= float_flag_inexact;
+    if ( roundBits ) roundData->exception |= float_flag_inexact;
     zSig0 += roundIncrement;
     if ( zSig0 < roundIncrement ) {
         ++zExp;
@@ -672,7 +663,7 @@ static floatx80
            ) {
             roundMask = 0;
  overflow:
-            float_raise( float_flag_overflow | float_flag_inexact );
+            roundData->exception |= float_flag_overflow | float_flag_inexact;
             if (    ( roundingMode == float_round_to_zero )
                  || ( zSign && ( roundingMode == float_round_up ) )
                  || ( ! zSign && ( roundingMode == float_round_down ) )
@@ -689,8 +680,8 @@ static floatx80
                 || ( zSig0 < LIT64( 0xFFFFFFFFFFFFFFFF ) );
             shift64ExtraRightJamming( zSig0, zSig1, 1 - zExp, &zSig0, &zSig1 );
             zExp = 0;
-            if ( isTiny && zSig1 ) float_raise( float_flag_underflow );
-            if ( zSig1 ) float_exception_flags |= float_flag_inexact;
+            if ( isTiny && zSig1 ) roundData->exception |= float_flag_underflow;
+            if ( zSig1 ) roundData->exception |= float_flag_inexact;
             if ( roundNearestEven ) {
                 increment = ( (sbits64) zSig1 < 0 );
             }
@@ -710,7 +701,7 @@ static floatx80
             return packFloatx80( zSign, zExp, zSig0 );
         }
     }
-    if ( zSig1 ) float_exception_flags |= float_flag_inexact;
+    if ( zSig1 ) roundData->exception |= float_flag_inexact;
     if ( increment ) {
         ++zSig0;
         if ( zSig0 == 0 ) {
@@ -740,7 +731,7 @@ normalized.
 */
 static floatx80
  normalizeRoundAndPackFloatx80(
-     int8 roundingPrecision, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1
+     struct roundingData *roundData, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1
  )
 {
     int8 shiftCount;
@@ -754,7 +745,7 @@ static floatx80
     shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 );
     zExp -= shiftCount;
     return
-        roundAndPackFloatx80( roundingPrecision, zSign, zExp, zSig0, zSig1 );
+        roundAndPackFloatx80( roundData, zSign, zExp, zSig0, zSig1 );
 
 }
 
@@ -767,14 +758,14 @@ the single-precision floating-point format.  The conversion is performed
 according to the IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 int32_to_float32( int32 a )
+float32 int32_to_float32(struct roundingData *roundData, int32 a)
 {
     flag zSign;
 
     if ( a == 0 ) return 0;
     if ( a == 0x80000000 ) return packFloat32( 1, 0x9E, 0 );
     zSign = ( a < 0 );
-    return normalizeRoundAndPackFloat32( zSign, 0x9C, zSign ? - a : a );
+    return normalizeRoundAndPackFloat32( roundData, zSign, 0x9C, zSign ? - a : a );
 
 }
 
@@ -840,7 +831,7 @@ positive integer is returned.  Otherwise, if the conversion overflows, the
 largest integer with the same sign as `a' is returned.
 -------------------------------------------------------------------------------
 */
-int32 float32_to_int32( float32 a )
+int32 float32_to_int32( struct roundingData *roundData, float32 a )
 {
     flag aSign;
     int16 aExp, shiftCount;
@@ -856,7 +847,7 @@ int32 float32_to_int32( float32 a )
     zSig = aSig;
     zSig <<= 32;
     if ( 0 < shiftCount ) shift64RightJamming( zSig, shiftCount, &zSig );
-    return roundAndPackInt32( aSign, zSig );
+    return roundAndPackInt32( roundData, aSign, zSig );
 
 }
 
@@ -889,13 +880,13 @@ int32 float32_to_int32_round_to_zero( float32 a )
         return 0x80000000;
     }
     else if ( aExp <= 0x7E ) {
-        if ( aExp | aSig ) float_exception_flags |= float_flag_inexact;
+        if ( aExp | aSig ) float_raise( float_flag_inexact );
         return 0;
     }
     aSig = ( aSig | 0x00800000 )<<8;
     z = aSig>>( - shiftCount );
     if ( (bits32) ( aSig<<( shiftCount & 31 ) ) ) {
-        float_exception_flags |= float_flag_inexact;
+        float_raise( float_flag_inexact );
     }
     return aSign ? - z : z;
 
@@ -973,7 +964,7 @@ operation is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 float32_round_to_int( float32 a )
+float32 float32_round_to_int( struct roundingData *roundData, float32 a )
 {
     flag aSign;
     int16 aExp;
@@ -988,11 +979,12 @@ float32 float32_round_to_int( float32 a )
         }
         return a;
     }
+    roundingMode = roundData->mode;
     if ( aExp <= 0x7E ) {
         if ( (bits32) ( a<<1 ) == 0 ) return a;
-        float_exception_flags |= float_flag_inexact;
+        roundData->exception |= float_flag_inexact;
         aSign = extractFloat32Sign( a );
-        switch ( float_rounding_mode ) {
+        switch ( roundingMode ) {
          case float_round_nearest_even:
             if ( ( aExp == 0x7E ) && extractFloat32Frac( a ) ) {
                 return packFloat32( aSign, 0x7F, 0 );
@@ -1009,7 +1001,6 @@ float32 float32_round_to_int( float32 a )
     lastBitMask <<= 0x96 - aExp;
     roundBitsMask = lastBitMask - 1;
     z = a;
-    roundingMode = float_rounding_mode;
     if ( roundingMode == float_round_nearest_even ) {
         z += lastBitMask>>1;
         if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask;
@@ -1020,7 +1011,7 @@ float32 float32_round_to_int( float32 a )
         }
     }
     z &= ~ roundBitsMask;
-    if ( z != a ) float_exception_flags |= float_flag_inexact;
+    if ( z != a ) roundData->exception |= float_flag_inexact;
     return z;
 
 }
@@ -1034,7 +1025,7 @@ addition is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-static float32 addFloat32Sigs( float32 a, float32 b, flag zSign )
+static float32 addFloat32Sigs( struct roundingData *roundData, float32 a, float32 b, flag zSign )
 {
     int16 aExp, bExp, zExp;
     bits32 aSig, bSig, zSig;
@@ -1093,7 +1084,7 @@ static float32 addFloat32Sigs( float32 a, float32 b, flag zSign )
         ++zExp;
     }
  roundAndPack:
-    return roundAndPackFloat32( zSign, zExp, zSig );
+    return roundAndPackFloat32( roundData, zSign, zExp, zSig );
 
 }
 
@@ -1106,7 +1097,7 @@ result is a NaN.  The subtraction is performed according to the IEC/IEEE
 Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-static float32 subFloat32Sigs( float32 a, float32 b, flag zSign )
+static float32 subFloat32Sigs( struct roundingData *roundData, float32 a, float32 b, flag zSign )
 {
     int16 aExp, bExp, zExp;
     bits32 aSig, bSig, zSig;
@@ -1123,7 +1114,7 @@ static float32 subFloat32Sigs( float32 a, float32 b, flag zSign )
     if ( expDiff < 0 ) goto bExpBigger;
     if ( aExp == 0xFF ) {
         if ( aSig | bSig ) return propagateFloat32NaN( a, b );
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return float32_default_nan;
     }
     if ( aExp == 0 ) {
@@ -1132,7 +1123,7 @@ static float32 subFloat32Sigs( float32 a, float32 b, flag zSign )
     }
     if ( bSig < aSig ) goto aBigger;
     if ( aSig < bSig ) goto bBigger;
-    return packFloat32( float_rounding_mode == float_round_down, 0, 0 );
+    return packFloat32( roundData->mode == float_round_down, 0, 0 );
  bExpBigger:
     if ( bExp == 0xFF ) {
         if ( bSig ) return propagateFloat32NaN( a, b );
@@ -1169,7 +1160,7 @@ static float32 subFloat32Sigs( float32 a, float32 b, flag zSign )
     zExp = aExp;
  normalizeRoundAndPack:
     --zExp;
-    return normalizeRoundAndPackFloat32( zSign, zExp, zSig );
+    return normalizeRoundAndPackFloat32( roundData, zSign, zExp, zSig );
 
 }
 
@@ -1180,17 +1171,17 @@ and `b'.  The operation is performed according to the IEC/IEEE Standard for
 Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 float32_add( float32 a, float32 b )
+float32 float32_add( struct roundingData *roundData, float32 a, float32 b )
 {
     flag aSign, bSign;
 
     aSign = extractFloat32Sign( a );
     bSign = extractFloat32Sign( b );
     if ( aSign == bSign ) {
-        return addFloat32Sigs( a, b, aSign );
+        return addFloat32Sigs( roundData, a, b, aSign );
     }
     else {
-        return subFloat32Sigs( a, b, aSign );
+        return subFloat32Sigs( roundData, a, b, aSign );
     }
 
 }
@@ -1202,17 +1193,17 @@ Returns the result of subtracting the single-precision floating-point values
 for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 float32_sub( float32 a, float32 b )
+float32 float32_sub( struct roundingData *roundData, float32 a, float32 b )
 {
     flag aSign, bSign;
 
     aSign = extractFloat32Sign( a );
     bSign = extractFloat32Sign( b );
     if ( aSign == bSign ) {
-        return subFloat32Sigs( a, b, aSign );
+        return subFloat32Sigs( roundData, a, b, aSign );
     }
     else {
-        return addFloat32Sigs( a, b, aSign );
+        return addFloat32Sigs( roundData, a, b, aSign );
     }
 
 }
@@ -1224,7 +1215,7 @@ Returns the result of multiplying the single-precision floating-point values
 for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 float32_mul( float32 a, float32 b )
+float32 float32_mul( struct roundingData *roundData, float32 a, float32 b )
 {
     flag aSign, bSign, zSign;
     int16 aExp, bExp, zExp;
@@ -1244,7 +1235,7 @@ float32 float32_mul( float32 a, float32 b )
             return propagateFloat32NaN( a, b );
         }
         if ( ( bExp | bSig ) == 0 ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             return float32_default_nan;
         }
         return packFloat32( zSign, 0xFF, 0 );
@@ -1252,7 +1243,7 @@ float32 float32_mul( float32 a, float32 b )
     if ( bExp == 0xFF ) {
         if ( bSig ) return propagateFloat32NaN( a, b );
         if ( ( aExp | aSig ) == 0 ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             return float32_default_nan;
         }
         return packFloat32( zSign, 0xFF, 0 );
@@ -1274,7 +1265,7 @@ float32 float32_mul( float32 a, float32 b )
         zSig <<= 1;
         --zExp;
     }
-    return roundAndPackFloat32( zSign, zExp, zSig );
+    return roundAndPackFloat32( roundData, zSign, zExp, zSig );
 
 }
 
@@ -1285,7 +1276,7 @@ by the corresponding value `b'.  The operation is performed according to the
 IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 float32_div( float32 a, float32 b )
+float32 float32_div( struct roundingData *roundData, float32 a, float32 b )
 {
     flag aSign, bSign, zSign;
     int16 aExp, bExp, zExp;
@@ -1302,7 +1293,7 @@ float32 float32_div( float32 a, float32 b )
         if ( aSig ) return propagateFloat32NaN( a, b );
         if ( bExp == 0xFF ) {
             if ( bSig ) return propagateFloat32NaN( a, b );
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             return float32_default_nan;
         }
         return packFloat32( zSign, 0xFF, 0 );
@@ -1314,10 +1305,10 @@ float32 float32_div( float32 a, float32 b )
     if ( bExp == 0 ) {
         if ( bSig == 0 ) {
             if ( ( aExp | aSig ) == 0 ) {
-                float_raise( float_flag_invalid );
+                roundData->exception |= float_flag_invalid;
                 return float32_default_nan;
             }
-            float_raise( float_flag_divbyzero );
+            roundData->exception |= float_flag_divbyzero;
             return packFloat32( zSign, 0xFF, 0 );
         }
         normalizeFloat32Subnormal( bSig, &bExp, &bSig );
@@ -1341,7 +1332,7 @@ float32 float32_div( float32 a, float32 b )
     if ( ( zSig & 0x3F ) == 0 ) {
         zSig |= ( ( (bits64) bSig ) * zSig != ( (bits64) aSig )<<32 );
     }
-    return roundAndPackFloat32( zSign, zExp, zSig );
+    return roundAndPackFloat32( roundData, zSign, zExp, zSig );
 
 }
 
@@ -1352,7 +1343,7 @@ with respect to the corresponding value `b'.  The operation is performed
 according to the IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 float32_rem( float32 a, float32 b )
+float32 float32_rem( struct roundingData *roundData, float32 a, float32 b )
 {
     flag aSign, bSign, zSign;
     int16 aExp, bExp, expDiff;
@@ -1372,7 +1363,7 @@ float32 float32_rem( float32 a, float32 b )
         if ( aSig || ( ( bExp == 0xFF ) && bSig ) ) {
             return propagateFloat32NaN( a, b );
         }
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return float32_default_nan;
     }
     if ( bExp == 0xFF ) {
@@ -1381,7 +1372,7 @@ float32 float32_rem( float32 a, float32 b )
     }
     if ( bExp == 0 ) {
         if ( bSig == 0 ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             return float32_default_nan;
         }
         normalizeFloat32Subnormal( bSig, &bExp, &bSig );
@@ -1444,7 +1435,7 @@ float32 float32_rem( float32 a, float32 b )
     }
     zSign = ( (sbits32) aSig < 0 );
     if ( zSign ) aSig = - aSig;
-    return normalizeRoundAndPackFloat32( aSign ^ zSign, bExp, aSig );
+    return normalizeRoundAndPackFloat32( roundData, aSign ^ zSign, bExp, aSig );
 
 }
 
@@ -1455,7 +1446,7 @@ The operation is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 float32_sqrt( float32 a )
+float32 float32_sqrt( struct roundingData *roundData, float32 a )
 {
     flag aSign;
     int16 aExp, zExp;
@@ -1468,12 +1459,12 @@ float32 float32_sqrt( float32 a )
     if ( aExp == 0xFF ) {
         if ( aSig ) return propagateFloat32NaN( a, 0 );
         if ( ! aSign ) return a;
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return float32_default_nan;
     }
     if ( aSign ) {
         if ( ( aExp | aSig ) == 0 ) return a;
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return float32_default_nan;
     }
     if ( aExp == 0 ) {
@@ -1499,7 +1490,7 @@ float32 float32_sqrt( float32 a )
         }
     }
     shift32RightJamming( zSig, 1, &zSig );
-    return roundAndPackFloat32( 0, zExp, zSig );
+    return roundAndPackFloat32( roundData, 0, zExp, zSig );
 
 }
 
@@ -1661,7 +1652,7 @@ positive integer is returned.  Otherwise, if the conversion overflows, the
 largest integer with the same sign as `a' is returned.
 -------------------------------------------------------------------------------
 */
-int32 float64_to_int32( float64 a )
+int32 float64_to_int32( struct roundingData *roundData, float64 a )
 {
     flag aSign;
     int16 aExp, shiftCount;
@@ -1674,7 +1665,7 @@ int32 float64_to_int32( float64 a )
     if ( aExp ) aSig |= LIT64( 0x0010000000000000 );
     shiftCount = 0x42C - aExp;
     if ( 0 < shiftCount ) shift64RightJamming( aSig, shiftCount, &aSig );
-    return roundAndPackInt32( aSign, aSig );
+    return roundAndPackInt32( roundData, aSign, aSig );
 
 }
 
@@ -1705,7 +1696,7 @@ int32 float64_to_int32_round_to_zero( float64 a )
         goto invalid;
     }
     else if ( 52 < shiftCount ) {
-        if ( aExp || aSig ) float_exception_flags |= float_flag_inexact;
+        if ( aExp || aSig ) float_raise( float_flag_inexact );
         return 0;
     }
     aSig |= LIT64( 0x0010000000000000 );
@@ -1715,11 +1706,11 @@ int32 float64_to_int32_round_to_zero( float64 a )
     if ( aSign ) z = - z;
     if ( ( z < 0 ) ^ aSign ) {
  invalid:
-        float_exception_flags |= float_flag_invalid;
+        float_raise( float_flag_invalid );
         return aSign ? 0x80000000 : 0x7FFFFFFF;
     }
     if ( ( aSig<<shiftCount ) != savedASig ) {
-        float_exception_flags |= float_flag_inexact;
+        float_raise( float_flag_inexact );
     }
     return z;
 
@@ -1736,7 +1727,7 @@ positive integer is returned.  Otherwise, if the conversion overflows, the
 largest positive integer is returned.
 -------------------------------------------------------------------------------
 */
-int32 float64_to_uint32( float64 a )
+int32 float64_to_uint32( struct roundingData *roundData, float64 a )
 {
     flag aSign;
     int16 aExp, shiftCount;
@@ -1749,7 +1740,7 @@ int32 float64_to_uint32( float64 a )
     if ( aExp ) aSig |= LIT64( 0x0010000000000000 );
     shiftCount = 0x42C - aExp;
     if ( 0 < shiftCount ) shift64RightJamming( aSig, shiftCount, &aSig );
-    return roundAndPackInt32( aSign, aSig );
+    return roundAndPackInt32( roundData, aSign, aSig );
 }
 
 /*
@@ -1778,7 +1769,7 @@ int32 float64_to_uint32_round_to_zero( float64 a )
         goto invalid;
     }
     else if ( 52 < shiftCount ) {
-        if ( aExp || aSig ) float_exception_flags |= float_flag_inexact;
+        if ( aExp || aSig ) float_raise( float_flag_inexact );
         return 0;
     }
     aSig |= LIT64( 0x0010000000000000 );
@@ -1788,11 +1779,11 @@ int32 float64_to_uint32_round_to_zero( float64 a )
     if ( aSign ) z = - z;
     if ( ( z < 0 ) ^ aSign ) {
  invalid:
-        float_exception_flags |= float_flag_invalid;
+        float_raise( float_flag_invalid );
         return aSign ? 0x80000000 : 0x7FFFFFFF;
     }
     if ( ( aSig<<shiftCount ) != savedASig ) {
-        float_exception_flags |= float_flag_inexact;
+        float_raise( float_flag_inexact );
     }
     return z;
 }
@@ -1805,7 +1796,7 @@ performed according to the IEC/IEEE Standard for Binary Floating-point
 Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 float64_to_float32( float64 a )
+float32 float64_to_float32( struct roundingData *roundData, float64 a )
 {
     flag aSign;
     int16 aExp;
@@ -1825,7 +1816,7 @@ float32 float64_to_float32( float64 a )
         zSig |= 0x40000000;
         aExp -= 0x381;
     }
-    return roundAndPackFloat32( aSign, aExp, zSig );
+    return roundAndPackFloat32( roundData, aSign, aExp, zSig );
 
 }
 
@@ -1872,7 +1863,7 @@ operation is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float64 float64_round_to_int( float64 a )
+float64 float64_round_to_int( struct roundingData *roundData, float64 a )
 {
     flag aSign;
     int16 aExp;
@@ -1889,9 +1880,9 @@ float64 float64_round_to_int( float64 a )
     }
     if ( aExp <= 0x3FE ) {
         if ( (bits64) ( a<<1 ) == 0 ) return a;
-        float_exception_flags |= float_flag_inexact;
+        roundData->exception |= float_flag_inexact;
         aSign = extractFloat64Sign( a );
-        switch ( float_rounding_mode ) {
+        switch ( roundData->mode ) {
          case float_round_nearest_even:
             if ( ( aExp == 0x3FE ) && extractFloat64Frac( a ) ) {
                 return packFloat64( aSign, 0x3FF, 0 );
@@ -1909,7 +1900,7 @@ float64 float64_round_to_int( float64 a )
     lastBitMask <<= 0x433 - aExp;
     roundBitsMask = lastBitMask - 1;
     z = a;
-    roundingMode = float_rounding_mode;
+    roundingMode = roundData->mode;
     if ( roundingMode == float_round_nearest_even ) {
         z += lastBitMask>>1;
         if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask;
@@ -1920,7 +1911,7 @@ float64 float64_round_to_int( float64 a )
         }
     }
     z &= ~ roundBitsMask;
-    if ( z != a ) float_exception_flags |= float_flag_inexact;
+    if ( z != a ) roundData->exception |= float_flag_inexact;
     return z;
 
 }
@@ -1934,7 +1925,7 @@ addition is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-static float64 addFloat64Sigs( float64 a, float64 b, flag zSign )
+static float64 addFloat64Sigs( struct roundingData *roundData, float64 a, float64 b, flag zSign )
 {
     int16 aExp, bExp, zExp;
     bits64 aSig, bSig, zSig;
@@ -1993,7 +1984,7 @@ static float64 addFloat64Sigs( float64 a, float64 b, flag zSign )
         ++zExp;
     }
  roundAndPack:
-    return roundAndPackFloat64( zSign, zExp, zSig );
+    return roundAndPackFloat64( roundData, zSign, zExp, zSig );
 
 }
 
@@ -2006,7 +1997,7 @@ result is a NaN.  The subtraction is performed according to the IEC/IEEE
 Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-static float64 subFloat64Sigs( float64 a, float64 b, flag zSign )
+static float64 subFloat64Sigs( struct roundingData *roundData, float64 a, float64 b, flag zSign )
 {
     int16 aExp, bExp, zExp;
     bits64 aSig, bSig, zSig;
@@ -2023,7 +2014,7 @@ static float64 subFloat64Sigs( float64 a, float64 b, flag zSign )
     if ( expDiff < 0 ) goto bExpBigger;
     if ( aExp == 0x7FF ) {
         if ( aSig | bSig ) return propagateFloat64NaN( a, b );
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return float64_default_nan;
     }
     if ( aExp == 0 ) {
@@ -2032,7 +2023,7 @@ static float64 subFloat64Sigs( float64 a, float64 b, flag zSign )
     }
     if ( bSig < aSig ) goto aBigger;
     if ( aSig < bSig ) goto bBigger;
-    return packFloat64( float_rounding_mode == float_round_down, 0, 0 );
+    return packFloat64( roundData->mode == float_round_down, 0, 0 );
  bExpBigger:
     if ( bExp == 0x7FF ) {
         if ( bSig ) return propagateFloat64NaN( a, b );
@@ -2069,7 +2060,7 @@ static float64 subFloat64Sigs( float64 a, float64 b, flag zSign )
     zExp = aExp;
  normalizeRoundAndPack:
     --zExp;
-    return normalizeRoundAndPackFloat64( zSign, zExp, zSig );
+    return normalizeRoundAndPackFloat64( roundData, zSign, zExp, zSig );
 
 }
 
@@ -2080,17 +2071,17 @@ and `b'.  The operation is performed according to the IEC/IEEE Standard for
 Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float64 float64_add( float64 a, float64 b )
+float64 float64_add( struct roundingData *roundData, float64 a, float64 b )
 {
     flag aSign, bSign;
 
     aSign = extractFloat64Sign( a );
     bSign = extractFloat64Sign( b );
     if ( aSign == bSign ) {
-        return addFloat64Sigs( a, b, aSign );
+        return addFloat64Sigs( roundData, a, b, aSign );
     }
     else {
-        return subFloat64Sigs( a, b, aSign );
+        return subFloat64Sigs( roundData, a, b, aSign );
     }
 
 }
@@ -2102,17 +2093,17 @@ Returns the result of subtracting the double-precision floating-point values
 for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float64 float64_sub( float64 a, float64 b )
+float64 float64_sub( struct roundingData *roundData, float64 a, float64 b )
 {
     flag aSign, bSign;
 
     aSign = extractFloat64Sign( a );
     bSign = extractFloat64Sign( b );
     if ( aSign == bSign ) {
-        return subFloat64Sigs( a, b, aSign );
+        return subFloat64Sigs( roundData, a, b, aSign );
     }
     else {
-        return addFloat64Sigs( a, b, aSign );
+        return addFloat64Sigs( roundData, a, b, aSign );
     }
 
 }
@@ -2124,7 +2115,7 @@ Returns the result of multiplying the double-precision floating-point values
 for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float64 float64_mul( float64 a, float64 b )
+float64 float64_mul( struct roundingData *roundData, float64 a, float64 b )
 {
     flag aSign, bSign, zSign;
     int16 aExp, bExp, zExp;
@@ -2142,7 +2133,7 @@ float64 float64_mul( float64 a, float64 b )
             return propagateFloat64NaN( a, b );
         }
         if ( ( bExp | bSig ) == 0 ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             return float64_default_nan;
         }
         return packFloat64( zSign, 0x7FF, 0 );
@@ -2150,7 +2141,7 @@ float64 float64_mul( float64 a, float64 b )
     if ( bExp == 0x7FF ) {
         if ( bSig ) return propagateFloat64NaN( a, b );
         if ( ( aExp | aSig ) == 0 ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             return float64_default_nan;
         }
         return packFloat64( zSign, 0x7FF, 0 );
@@ -2172,7 +2163,7 @@ float64 float64_mul( float64 a, float64 b )
         zSig0 <<= 1;
         --zExp;
     }
-    return roundAndPackFloat64( zSign, zExp, zSig0 );
+    return roundAndPackFloat64( roundData, zSign, zExp, zSig0 );
 
 }
 
@@ -2183,7 +2174,7 @@ by the corresponding value `b'.  The operation is performed according to
 the IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float64 float64_div( float64 a, float64 b )
+float64 float64_div( struct roundingData *roundData, float64 a, float64 b )
 {
     flag aSign, bSign, zSign;
     int16 aExp, bExp, zExp;
@@ -2202,7 +2193,7 @@ float64 float64_div( float64 a, float64 b )
         if ( aSig ) return propagateFloat64NaN( a, b );
         if ( bExp == 0x7FF ) {
             if ( bSig ) return propagateFloat64NaN( a, b );
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             return float64_default_nan;
         }
         return packFloat64( zSign, 0x7FF, 0 );
@@ -2214,10 +2205,10 @@ float64 float64_div( float64 a, float64 b )
     if ( bExp == 0 ) {
         if ( bSig == 0 ) {
             if ( ( aExp | aSig ) == 0 ) {
-                float_raise( float_flag_invalid );
+                roundData->exception |= float_flag_invalid;
                 return float64_default_nan;
             }
-            float_raise( float_flag_divbyzero );
+            roundData->exception |= float_flag_divbyzero;
             return packFloat64( zSign, 0x7FF, 0 );
         }
         normalizeFloat64Subnormal( bSig, &bExp, &bSig );
@@ -2243,7 +2234,7 @@ float64 float64_div( float64 a, float64 b )
         }
         zSig |= ( rem1 != 0 );
     }
-    return roundAndPackFloat64( zSign, zExp, zSig );
+    return roundAndPackFloat64( roundData, zSign, zExp, zSig );
 
 }
 
@@ -2254,7 +2245,7 @@ with respect to the corresponding value `b'.  The operation is performed
 according to the IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float64 float64_rem( float64 a, float64 b )
+float64 float64_rem( struct roundingData *roundData, float64 a, float64 b )
 {
     flag aSign, bSign, zSign;
     int16 aExp, bExp, expDiff;
@@ -2272,7 +2263,7 @@ float64 float64_rem( float64 a, float64 b )
         if ( aSig || ( ( bExp == 0x7FF ) && bSig ) ) {
             return propagateFloat64NaN( a, b );
         }
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return float64_default_nan;
     }
     if ( bExp == 0x7FF ) {
@@ -2281,7 +2272,7 @@ float64 float64_rem( float64 a, float64 b )
     }
     if ( bExp == 0 ) {
         if ( bSig == 0 ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             return float64_default_nan;
         }
         normalizeFloat64Subnormal( bSig, &bExp, &bSig );
@@ -2329,7 +2320,7 @@ float64 float64_rem( float64 a, float64 b )
     }
     zSign = ( (sbits64) aSig < 0 );
     if ( zSign ) aSig = - aSig;
-    return normalizeRoundAndPackFloat64( aSign ^ zSign, bExp, aSig );
+    return normalizeRoundAndPackFloat64( roundData, aSign ^ zSign, bExp, aSig );
 
 }
 
@@ -2340,7 +2331,7 @@ The operation is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float64 float64_sqrt( float64 a )
+float64 float64_sqrt( struct roundingData *roundData, float64 a )
 {
     flag aSign;
     int16 aExp, zExp;
@@ -2354,12 +2345,12 @@ float64 float64_sqrt( float64 a )
     if ( aExp == 0x7FF ) {
         if ( aSig ) return propagateFloat64NaN( a, a );
         if ( ! aSign ) return a;
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return float64_default_nan;
     }
     if ( aSign ) {
         if ( ( aExp | aSig ) == 0 ) return a;
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return float64_default_nan;
     }
     if ( aExp == 0 ) {
@@ -2390,7 +2381,7 @@ float64 float64_sqrt( float64 a )
         }
     }
     shift64RightJamming( zSig, 1, &zSig );
-    return roundAndPackFloat64( 0, zExp, zSig );
+    return roundAndPackFloat64( roundData, 0, zExp, zSig );
 
 }
 
@@ -2554,7 +2545,7 @@ largest positive integer is returned.  Otherwise, if the conversion
 overflows, the largest integer with the same sign as `a' is returned.
 -------------------------------------------------------------------------------
 */
-int32 floatx80_to_int32( floatx80 a )
+int32 floatx80_to_int32( struct roundingData *roundData, floatx80 a )
 {
     flag aSign;
     int32 aExp, shiftCount;
@@ -2567,7 +2558,7 @@ int32 floatx80_to_int32( floatx80 a )
     shiftCount = 0x4037 - aExp;
     if ( shiftCount <= 0 ) shiftCount = 1;
     shift64RightJamming( aSig, shiftCount, &aSig );
-    return roundAndPackInt32( aSign, aSig );
+    return roundAndPackInt32( roundData, aSign, aSig );
 
 }
 
@@ -2598,7 +2589,7 @@ int32 floatx80_to_int32_round_to_zero( floatx80 a )
         goto invalid;
     }
     else if ( 63 < shiftCount ) {
-        if ( aExp || aSig ) float_exception_flags |= float_flag_inexact;
+        if ( aExp || aSig ) float_raise( float_flag_inexact );
         return 0;
     }
     savedASig = aSig;
@@ -2607,11 +2598,11 @@ int32 floatx80_to_int32_round_to_zero( floatx80 a )
     if ( aSign ) z = - z;
     if ( ( z < 0 ) ^ aSign ) {
  invalid:
-        float_exception_flags |= float_flag_invalid;
+        float_raise( float_flag_invalid );
         return aSign ? 0x80000000 : 0x7FFFFFFF;
     }
     if ( ( aSig<<shiftCount ) != savedASig ) {
-        float_exception_flags |= float_flag_inexact;
+        float_raise( float_flag_inexact );
     }
     return z;
 
@@ -2625,7 +2616,7 @@ conversion is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float32 floatx80_to_float32( floatx80 a )
+float32 floatx80_to_float32( struct roundingData *roundData, floatx80 a )
 {
     flag aSign;
     int32 aExp;
@@ -2642,7 +2633,7 @@ float32 floatx80_to_float32( floatx80 a )
     }
     shift64RightJamming( aSig, 33, &aSig );
     if ( aExp || aSig ) aExp -= 0x3F81;
-    return roundAndPackFloat32( aSign, aExp, aSig );
+    return roundAndPackFloat32( roundData, aSign, aExp, aSig );
 
 }
 
@@ -2654,7 +2645,7 @@ conversion is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-float64 floatx80_to_float64( floatx80 a )
+float64 floatx80_to_float64( struct roundingData *roundData, floatx80 a )
 {
     flag aSign;
     int32 aExp;
@@ -2671,7 +2662,7 @@ float64 floatx80_to_float64( floatx80 a )
     }
     shift64RightJamming( aSig, 1, &zSig );
     if ( aExp || aSig ) aExp -= 0x3C01;
-    return roundAndPackFloat64( aSign, aExp, zSig );
+    return roundAndPackFloat64( roundData, aSign, aExp, zSig );
 
 }
 
@@ -2683,7 +2674,7 @@ value.  The operation is performed according to the IEC/IEEE Standard for
 Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-floatx80 floatx80_round_to_int( floatx80 a )
+floatx80 floatx80_round_to_int( struct roundingData *roundData, floatx80 a )
 {
     flag aSign;
     int32 aExp;
@@ -2703,9 +2694,9 @@ floatx80 floatx80_round_to_int( floatx80 a )
              && ( (bits64) ( extractFloatx80Frac( a )<<1 ) == 0 ) ) {
             return a;
         }
-        float_exception_flags |= float_flag_inexact;
+        roundData->exception |= float_flag_inexact;
         aSign = extractFloatx80Sign( a );
-        switch ( float_rounding_mode ) {
+        switch ( roundData->mode ) {
          case float_round_nearest_even:
             if ( ( aExp == 0x3FFE ) && (bits64) ( extractFloatx80Frac( a )<<1 )
                ) {
@@ -2729,7 +2720,7 @@ floatx80 floatx80_round_to_int( floatx80 a )
     lastBitMask <<= 0x403E - aExp;
     roundBitsMask = lastBitMask - 1;
     z = a;
-    roundingMode = float_rounding_mode;
+    roundingMode = roundData->mode;
     if ( roundingMode == float_round_nearest_even ) {
         z.low += lastBitMask>>1;
         if ( ( z.low & roundBitsMask ) == 0 ) z.low &= ~ lastBitMask;
@@ -2744,7 +2735,7 @@ floatx80 floatx80_round_to_int( floatx80 a )
         ++z.high;
         z.low = LIT64( 0x8000000000000000 );
     }
-    if ( z.low != a.low ) float_exception_flags |= float_flag_inexact;
+    if ( z.low != a.low ) roundData->exception |= float_flag_inexact;
     return z;
 
 }
@@ -2758,7 +2749,7 @@ The addition is performed according to the IEC/IEEE Standard for Binary
 Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-static floatx80 addFloatx80Sigs( floatx80 a, floatx80 b, flag zSign )
+static floatx80 addFloatx80Sigs( struct roundingData *roundData, floatx80 a, floatx80 b, flag zSign )
 {
     int32 aExp, bExp, zExp;
     bits64 aSig, bSig, zSig0, zSig1;
@@ -2814,7 +2805,7 @@ static floatx80 addFloatx80Sigs( floatx80 a, floatx80 b, flag zSign )
  roundAndPack:
     return
         roundAndPackFloatx80(
-            floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 );
+            roundData, zSign, zExp, zSig0, zSig1 );
 
 }
 
@@ -2827,7 +2818,7 @@ result is a NaN.  The subtraction is performed according to the IEC/IEEE
 Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign )
+static floatx80 subFloatx80Sigs( struct roundingData *roundData, floatx80 a, floatx80 b, flag zSign )
 {
     int32 aExp, bExp, zExp;
     bits64 aSig, bSig, zSig0, zSig1;
@@ -2845,7 +2836,7 @@ static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign )
         if ( (bits64) ( ( aSig | bSig )<<1 ) ) {
             return propagateFloatx80NaN( a, b );
         }
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         z.low = floatx80_default_nan_low;
         z.high = floatx80_default_nan_high;
         return z;
@@ -2857,7 +2848,7 @@ static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign )
     zSig1 = 0;
     if ( bSig < aSig ) goto aBigger;
     if ( aSig < bSig ) goto bBigger;
-    return packFloatx80( float_rounding_mode == float_round_down, 0, 0 );
+    return packFloatx80( roundData->mode == float_round_down, 0, 0 );
  bExpBigger:
     if ( bExp == 0x7FFF ) {
         if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b );
@@ -2883,7 +2874,7 @@ static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign )
  normalizeRoundAndPack:
     return
         normalizeRoundAndPackFloatx80(
-            floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 );
+            roundData, zSign, zExp, zSig0, zSig1 );
 
 }
 
@@ -2894,17 +2885,17 @@ values `a' and `b'.  The operation is performed according to the IEC/IEEE
 Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-floatx80 floatx80_add( floatx80 a, floatx80 b )
+floatx80 floatx80_add( struct roundingData *roundData, floatx80 a, floatx80 b )
 {
     flag aSign, bSign;
     
     aSign = extractFloatx80Sign( a );
     bSign = extractFloatx80Sign( b );
     if ( aSign == bSign ) {
-        return addFloatx80Sigs( a, b, aSign );
+        return addFloatx80Sigs( roundData, a, b, aSign );
     }
     else {
-        return subFloatx80Sigs( a, b, aSign );
+        return subFloatx80Sigs( roundData, a, b, aSign );
     }
     
 }
@@ -2916,17 +2907,17 @@ point values `a' and `b'.  The operation is performed according to the
 IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-floatx80 floatx80_sub( floatx80 a, floatx80 b )
+floatx80 floatx80_sub( struct roundingData *roundData, floatx80 a, floatx80 b )
 {
     flag aSign, bSign;
 
     aSign = extractFloatx80Sign( a );
     bSign = extractFloatx80Sign( b );
     if ( aSign == bSign ) {
-        return subFloatx80Sigs( a, b, aSign );
+        return subFloatx80Sigs( roundData, a, b, aSign );
     }
     else {
-        return addFloatx80Sigs( a, b, aSign );
+        return addFloatx80Sigs( roundData, a, b, aSign );
     }
 
 }
@@ -2938,7 +2929,7 @@ point values `a' and `b'.  The operation is performed according to the
 IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-floatx80 floatx80_mul( floatx80 a, floatx80 b )
+floatx80 floatx80_mul( struct roundingData *roundData, floatx80 a, floatx80 b )
 {
     flag aSign, bSign, zSign;
     int32 aExp, bExp, zExp;
@@ -2964,7 +2955,7 @@ floatx80 floatx80_mul( floatx80 a, floatx80 b )
         if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b );
         if ( ( aExp | aSig ) == 0 ) {
  invalid:
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             z.low = floatx80_default_nan_low;
             z.high = floatx80_default_nan_high;
             return z;
@@ -2987,7 +2978,7 @@ floatx80 floatx80_mul( floatx80 a, floatx80 b )
     }
     return
         roundAndPackFloatx80(
-            floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 );
+            roundData, zSign, zExp, zSig0, zSig1 );
 
 }
 
@@ -2998,7 +2989,7 @@ value `a' by the corresponding value `b'.  The operation is performed
 according to the IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-floatx80 floatx80_div( floatx80 a, floatx80 b )
+floatx80 floatx80_div( struct roundingData *roundData, floatx80 a, floatx80 b )
 {
     flag aSign, bSign, zSign;
     int32 aExp, bExp, zExp;
@@ -3029,12 +3020,12 @@ floatx80 floatx80_div( floatx80 a, floatx80 b )
         if ( bSig == 0 ) {
             if ( ( aExp | aSig ) == 0 ) {
  invalid:
-                float_raise( float_flag_invalid );
+                roundData->exception |= float_flag_invalid;
                 z.low = floatx80_default_nan_low;
                 z.high = floatx80_default_nan_high;
                 return z;
             }
-            float_raise( float_flag_divbyzero );
+            roundData->exception |= float_flag_divbyzero;
             return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) );
         }
         normalizeFloatx80Subnormal( bSig, &bExp, &bSig );
@@ -3068,7 +3059,7 @@ floatx80 floatx80_div( floatx80 a, floatx80 b )
     }
     return
         roundAndPackFloatx80(
-            floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 );
+            roundData, zSign, zExp, zSig0, zSig1 );
 
 }
 
@@ -3079,7 +3070,7 @@ Returns the remainder of the extended double-precision floating-point value
 according to the IEC/IEEE Standard for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-floatx80 floatx80_rem( floatx80 a, floatx80 b )
+floatx80 floatx80_rem( struct roundingData *roundData, floatx80 a, floatx80 b )
 {
     flag aSign, bSign, zSign;
     int32 aExp, bExp, expDiff;
@@ -3107,7 +3098,7 @@ floatx80 floatx80_rem( floatx80 a, floatx80 b )
     if ( bExp == 0 ) {
         if ( bSig == 0 ) {
  invalid:
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
             z.low = floatx80_default_nan_low;
             z.high = floatx80_default_nan_high;
             return z;
@@ -3164,9 +3155,10 @@ floatx80 floatx80_rem( floatx80 a, floatx80 b )
         aSig1 = alternateASig1;
         zSign = ! zSign;
     }
+
     return
         normalizeRoundAndPackFloatx80(
-            80, zSign, bExp + expDiff, aSig0, aSig1 );
+            roundData, zSign, bExp + expDiff, aSig0, aSig1 );
 
 }
 
@@ -3177,7 +3169,7 @@ value `a'.  The operation is performed according to the IEC/IEEE Standard
 for Binary Floating-point Arithmetic.
 -------------------------------------------------------------------------------
 */
-floatx80 floatx80_sqrt( floatx80 a )
+floatx80 floatx80_sqrt( struct roundingData *roundData, floatx80 a )
 {
     flag aSign;
     int32 aExp, zExp;
@@ -3197,7 +3189,7 @@ floatx80 floatx80_sqrt( floatx80 a )
     if ( aSign ) {
         if ( ( aExp | aSig0 ) == 0 ) return a;
  invalid:
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         z.low = floatx80_default_nan_low;
         z.high = floatx80_default_nan_high;
         return z;
@@ -3242,7 +3234,7 @@ floatx80 floatx80_sqrt( floatx80 a )
     }
     return
         roundAndPackFloatx80(
-            floatx80_rounding_precision, 0, zExp, zSig0, zSig1 );
+            roundData, 0, zExp, zSig0, zSig1 );
 
 }
 
@@ -3264,7 +3256,7 @@ flag floatx80_eq( floatx80 a, floatx80 b )
        ) {
         if (    floatx80_is_signaling_nan( a )
              || floatx80_is_signaling_nan( b ) ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
         }
         return 0;
     }
@@ -3294,7 +3286,7 @@ flag floatx80_le( floatx80 a, floatx80 b )
          || (    ( extractFloatx80Exp( b ) == 0x7FFF )
               && (bits64) ( extractFloatx80Frac( b )<<1 ) )
        ) {
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return 0;
     }
     aSign = extractFloatx80Sign( a );
@@ -3328,7 +3320,7 @@ flag floatx80_lt( floatx80 a, floatx80 b )
          || (    ( extractFloatx80Exp( b ) == 0x7FFF )
               && (bits64) ( extractFloatx80Frac( b )<<1 ) )
        ) {
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return 0;
     }
     aSign = extractFloatx80Sign( a );
@@ -3361,7 +3353,7 @@ flag floatx80_eq_signaling( floatx80 a, floatx80 b )
          || (    ( extractFloatx80Exp( b ) == 0x7FFF )
               && (bits64) ( extractFloatx80Frac( b )<<1 ) )
        ) {
-        float_raise( float_flag_invalid );
+        roundData->exception |= float_flag_invalid;
         return 0;
     }
     return
@@ -3392,7 +3384,7 @@ flag floatx80_le_quiet( floatx80 a, floatx80 b )
        ) {
         if (    floatx80_is_signaling_nan( a )
              || floatx80_is_signaling_nan( b ) ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
         }
         return 0;
     }
@@ -3429,7 +3421,7 @@ flag floatx80_lt_quiet( floatx80 a, floatx80 b )
        ) {
         if (    floatx80_is_signaling_nan( a )
              || floatx80_is_signaling_nan( b ) ) {
-            float_raise( float_flag_invalid );
+            roundData->exception |= float_flag_invalid;
         }
         return 0;
     }
diff --git a/arch/arm/nwfpe/softfloat.h b/arch/arm/nwfpe/softfloat.h
index 1e17431..1c8799b 100644
--- a/arch/arm/nwfpe/softfloat.h
+++ b/arch/arm/nwfpe/softfloat.h
@@ -74,7 +74,7 @@ enum {
 Software IEC/IEEE floating-point rounding mode.
 -------------------------------------------------------------------------------
 */
-extern signed char float_rounding_mode;
+//extern int8 float_rounding_mode;
 enum {
     float_round_nearest_even = 0,
     float_round_to_zero      = 1,
@@ -86,7 +86,6 @@ enum {
 -------------------------------------------------------------------------------
 Software IEC/IEEE floating-point exception flags.
 -------------------------------------------------------------------------------
-extern signed char float_exception_flags;
 enum {
     float_flag_inexact   =  1,
     float_flag_underflow =  2,
@@ -99,7 +98,6 @@ ScottB: November 4, 1998
 Changed the enumeration to match the bit order in the FPA11.
 */
 
-extern signed char float_exception_flags;
 enum {
     float_flag_invalid   =  1,
     float_flag_divbyzero =  2,
@@ -121,7 +119,7 @@ void float_raise( signed char );
 Software IEC/IEEE integer-to-floating-point conversion routines.
 -------------------------------------------------------------------------------
 */
-float32 int32_to_float32( signed int );
+float32 int32_to_float32( struct roundingData *, signed int );
 float64 int32_to_float64( signed int );
 #ifdef FLOATX80
 floatx80 int32_to_floatx80( signed int );
@@ -132,7 +130,7 @@ floatx80 int32_to_floatx80( signed int );
 Software IEC/IEEE single-precision conversion routines.
 -------------------------------------------------------------------------------
 */
-signed int float32_to_int32( float32 );
+signed int float32_to_int32( struct roundingData *, float32 );
 signed int float32_to_int32_round_to_zero( float32 );
 float64 float32_to_float64( float32 );
 #ifdef FLOATX80
@@ -144,13 +142,13 @@ floatx80 float32_to_floatx80( float32 );
 Software IEC/IEEE single-precision operations.
 -------------------------------------------------------------------------------
 */
-float32 float32_round_to_int( float32 );
-float32 float32_add( float32, float32 );
-float32 float32_sub( float32, float32 );
-float32 float32_mul( float32, float32 );
-float32 float32_div( float32, float32 );
-float32 float32_rem( float32, float32 );
-float32 float32_sqrt( float32 );
+float32 float32_round_to_int( struct roundingData*, float32 );
+float32 float32_add( struct roundingData *, float32, float32 );
+float32 float32_sub( struct roundingData *, float32, float32 );
+float32 float32_mul( struct roundingData *, float32, float32 );
+float32 float32_div( struct roundingData *, float32, float32 );
+float32 float32_rem( struct roundingData *, float32, float32 );
+float32 float32_sqrt( struct roundingData*, float32 );
 char float32_eq( float32, float32 );
 char float32_le( float32, float32 );
 char float32_lt( float32, float32 );
@@ -164,9 +162,9 @@ char float32_is_signaling_nan( float32 );
 Software IEC/IEEE double-precision conversion routines.
 -------------------------------------------------------------------------------
 */
-signed int float64_to_int32( float64 );
+signed int float64_to_int32( struct roundingData *, float64 );
 signed int float64_to_int32_round_to_zero( float64 );
-float32 float64_to_float32( float64 );
+float32 float64_to_float32( struct roundingData *, float64 );
 #ifdef FLOATX80
 floatx80 float64_to_floatx80( float64 );
 #endif
@@ -176,13 +174,13 @@ floatx80 float64_to_floatx80( float64 );
 Software IEC/IEEE double-precision operations.
 -------------------------------------------------------------------------------
 */
-float64 float64_round_to_int( float64 );
-float64 float64_add( float64, float64 );
-float64 float64_sub( float64, float64 );
-float64 float64_mul( float64, float64 );
-float64 float64_div( float64, float64 );
-float64 float64_rem( float64, float64 );
-float64 float64_sqrt( float64 );
+float64 float64_round_to_int( struct roundingData *, float64 );
+float64 float64_add( struct roundingData *, float64, float64 );
+float64 float64_sub( struct roundingData *, float64, float64 );
+float64 float64_mul( struct roundingData *, float64, float64 );
+float64 float64_div( struct roundingData *, float64, float64 );
+float64 float64_rem( struct roundingData *, float64, float64 );
+float64 float64_sqrt( struct roundingData *, float64 );
 char float64_eq( float64, float64 );
 char float64_le( float64, float64 );
 char float64_lt( float64, float64 );
@@ -198,31 +196,23 @@ char float64_is_signaling_nan( float64 );
 Software IEC/IEEE extended double-precision conversion routines.
 -------------------------------------------------------------------------------
 */
-signed int floatx80_to_int32( floatx80 );
+signed int floatx80_to_int32( struct roundingData *, floatx80 );
 signed int floatx80_to_int32_round_to_zero( floatx80 );
-float32 floatx80_to_float32( floatx80 );
-float64 floatx80_to_float64( floatx80 );
-
-/*
--------------------------------------------------------------------------------
-Software IEC/IEEE extended double-precision rounding precision.  Valid
-values are 32, 64, and 80.
--------------------------------------------------------------------------------
-*/
-extern signed char floatx80_rounding_precision;
+float32 floatx80_to_float32( struct roundingData *, floatx80 );
+float64 floatx80_to_float64( struct roundingData *, floatx80 );
 
 /*
 -------------------------------------------------------------------------------
 Software IEC/IEEE extended double-precision operations.
 -------------------------------------------------------------------------------
 */
-floatx80 floatx80_round_to_int( floatx80 );
-floatx80 floatx80_add( floatx80, floatx80 );
-floatx80 floatx80_sub( floatx80, floatx80 );
-floatx80 floatx80_mul( floatx80, floatx80 );
-floatx80 floatx80_div( floatx80, floatx80 );
-floatx80 floatx80_rem( floatx80, floatx80 );
-floatx80 floatx80_sqrt( floatx80 );
+floatx80 floatx80_round_to_int( struct roundingData *, floatx80 );
+floatx80 floatx80_add( struct roundingData *, floatx80, floatx80 );
+floatx80 floatx80_sub( struct roundingData *, floatx80, floatx80 );
+floatx80 floatx80_mul( struct roundingData *, floatx80, floatx80 );
+floatx80 floatx80_div( struct roundingData *, floatx80, floatx80 );
+floatx80 floatx80_rem( struct roundingData *, floatx80, floatx80 );
+floatx80 floatx80_sqrt( struct roundingData *, floatx80 );
 char floatx80_eq( floatx80, floatx80 );
 char floatx80_le( floatx80, floatx80 );
 char floatx80_lt( floatx80, floatx80 );