Upgrade to 2.1.73.

42 files changed, 3628 insertions, 3047 deletions

diff --git a/arch/i386/defconfig b/arch/i386/defconfig
index 222b10939..f8793e75f 100644
--- a/arch/i386/defconfig
+++ b/arch/i386/defconfig
@@ -58,6 +58,7 @@ CONFIG_BLK_DEV_IDECD=y
 CONFIG_BLK_DEV_CMD640=y
 # CONFIG_BLK_DEV_CMD640_ENHANCED is not set
 CONFIG_BLK_DEV_RZ1000=y
+CONFIG_BLK_DEV_IDEPCI=y
 CONFIG_BLK_DEV_IDEDMA=y
 # CONFIG_IDE_CHIPSETS is not set
 
@@ -221,7 +222,6 @@ CONFIG_ISO9660_FS=y
 # CONFIG_VFAT_FS is not set
 CONFIG_PROC_FS=y
 CONFIG_NFS_FS=y
-# CONFIG_ROOT_NFS is not set
 CONFIG_NFSD=y
 CONFIG_SUNRPC=y
 CONFIG_LOCKD=y
diff --git a/arch/i386/kernel/process.c b/arch/i386/kernel/process.c
index 5a13e0179..352c5552a 100644
--- a/arch/i386/kernel/process.c
+++ b/arch/i386/kernel/process.c
@@ -40,6 +40,10 @@
 #include <asm/system.h>
 #include <asm/io.h>
 #include <asm/ldt.h>
+#include <asm/processor.h>
+#ifdef CONFIG_MATH_EMULATION
+#include <asm/math_emu.h>
+#endif
 
 #ifdef __SMP__
 asmlinkage void ret_from_smpfork(void) __asm__("ret_from_smpfork");
@@ -525,19 +529,16 @@ int dump_fpu (struct pt_regs * regs, struct user_i387_struct* fpu)
 {
 	int fpvalid;
 
-/* Flag indicating the math stuff is valid. We don't support this for the
-   soft-float routines yet */
-	if (hard_math) {
-		if ((fpvalid = current->used_math) != 0) {
-			if (last_task_used_math == current)
-				__asm__("clts ; fnsave %0": :"m" (*fpu));
+	if ((fpvalid = current->used_math) != 0) {
+		if (hard_math) {
+		  if (last_task_used_math == current) {
+			  __asm__("clts ; fsave %0; fwait": :"m" (*fpu));
+		  }
 			else
 				memcpy(fpu,&current->tss.i387.hard,sizeof(*fpu));
+		} else {
+			memcpy(fpu,&current->tss.i387.hard,sizeof(*fpu));
 		}
-	} else {
-		/* we should dump the emulator state here, but we need to
-		   convert it into standard 387 format first.. */
-		fpvalid = 0;
 	}
 
 	return fpvalid;
diff --git a/arch/i386/kernel/ptrace.c b/arch/i386/kernel/ptrace.c
index 246a6c29b..e08d75100 100644
--- a/arch/i386/kernel/ptrace.c
+++ b/arch/i386/kernel/ptrace.c
@@ -16,6 +16,7 @@
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
 #include <asm/system.h>
+#include <asm/processor.h>
 
 /*
  * does not yet catch signals sent when the child dies.
@@ -561,6 +562,102 @@ asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
 			goto out;
 		}
 
+		case PTRACE_GETREGS: { /* Get all gp regs from the child. */
+		  	if (!access_ok(VERIFY_WRITE, (unsigned *)data,
+				       17*sizeof(long)))
+			  {
+			    ret = -EIO;
+			    goto out;
+			  }
+			for ( i = 0; i < 17*sizeof(long); i += sizeof(long) )
+			  {
+			    __put_user(getreg(child, i),(unsigned long *) data);
+			    data += sizeof(long);
+			  }
+			ret = 0;
+			goto out;
+		  };
+
+		case PTRACE_SETREGS: { /* Set all gp regs in the child. */
+			unsigned long tmp;
+		  	if (!access_ok(VERIFY_READ, (unsigned *)data,
+				       17*sizeof(long)))
+			  {
+			    ret = -EIO;
+			    goto out;
+			  }
+			for ( i = 0; i < 17*sizeof(long); i += sizeof(long) )
+			  {
+			    __get_user(tmp, (unsigned long *) data);
+			    putreg(child, i, tmp);
+			    data += sizeof(long);
+			  }
+			ret = 0;
+			goto out;
+		  };
+
+		case PTRACE_GETFPREGS: { /* Get the child FPU state. */
+			if (!access_ok(VERIFY_WRITE, (unsigned *)data,
+				       sizeof(struct user_i387_struct)))
+			  {
+			    ret = -EIO;
+			    goto out;
+			  }
+			ret = 0;
+			if ( !child->used_math ) {
+			  /* Simulate an empty FPU. */
+			  child->tss.i387.hard.cwd = 0xffff037f;
+			  child->tss.i387.hard.swd = 0xffff0000;
+			  child->tss.i387.hard.twd = 0xffffffff;
+			}
+#ifdef CONFIG_MATH_EMULATION
+			if ( hard_math ) {
+#endif
+			  if (last_task_used_math == child) {
+			    clts();
+			    __asm__("fnsave %0; fwait":"=m" (child->tss.i387.hard));
+			    last_task_used_math = NULL;
+			    stts();
+			  }
+			  __copy_to_user((void *)data, &child->tss.i387.hard,
+					 sizeof(struct user_i387_struct));
+#ifdef CONFIG_MATH_EMULATION
+			} else {
+			  save_i387_soft(&child->tss.i387.soft,
+					 (struct _fpstate *)data);
+			}
+#endif
+			goto out;
+		  };
+
+		case PTRACE_SETFPREGS: { /* Set the child FPU state. */
+			if (!access_ok(VERIFY_READ, (unsigned *)data,
+				       sizeof(struct user_i387_struct)))
+			  {
+			    ret = -EIO;
+			    goto out;
+			  }
+			child->used_math = 1;
+#ifdef CONFIG_MATH_EMULATION
+			if ( hard_math ) {
+#endif
+			  if (last_task_used_math == child) {
+			    /* Discard the state of the FPU */
+			    last_task_used_math = NULL;
+			  }
+			  __copy_from_user(&child->tss.i387.hard, (void *)data,
+					   sizeof(struct user_i387_struct));
+			  child->flags &= ~PF_USEDFPU;
+#ifdef CONFIG_MATH_EMULATION
+			} else {
+			  restore_i387_soft(&child->tss.i387.soft,
+					    (struct _fpstate *)data);
+			}
+#endif
+			ret = 0;
+			goto out;
+		  };
+
 		default:
 			ret = -EIO;
 			goto out;
diff --git a/arch/i386/kernel/signal.c b/arch/i386/kernel/signal.c
index 7d5cf3a95..853b82100 100644
--- a/arch/i386/kernel/signal.c
+++ b/arch/i386/kernel/signal.c
@@ -156,7 +156,6 @@ static inline void restore_i387_hard(struct _fpstate *buf)
 		stts();
 	}
 #endif
-	current->used_math = 1;
 	current->flags &= ~PF_USEDFPU;
 	__copy_from_user(&current->tss.i387.hard, buf, sizeof(*buf));
 }
@@ -169,8 +168,9 @@ static inline void restore_i387(struct _fpstate *buf)
 	if (hard_math)
 		restore_i387_hard(buf);
 	else
-		restore_i387_soft(buf);
+		restore_i387_soft(&current->tss.i387.soft, buf);
 #endif
+	current->used_math = 1;
 }
 
 static int
@@ -309,7 +309,6 @@ static inline struct _fpstate * save_i387_hard(struct _fpstate * buf)
 #endif
 	current->tss.i387.hard.status = current->tss.i387.hard.swd;
 	copy_to_user(buf, &current->tss.i387.hard, sizeof(*buf));
-	current->used_math = 0;
 	return buf;
 }
 
@@ -318,10 +317,16 @@ static struct _fpstate * save_i387(struct _fpstate *buf)
 	if (!current->used_math)
 		return NULL;
 
+	/* This will cause a "finit" to be triggered by the next
+	   attempted FPU operation by the 'current' process.
+	   */
+	current->used_math = 0;
+
 #ifndef CONFIG_MATH_EMULATION
 	return save_i387_hard(buf);
 #else
-	return hard_math ? save_i387_hard(buf) : save_i387_soft(buf);
+	return hard_math ? save_i387_hard(buf)
+	  : save_i387_soft(&current->tss.i387.soft, buf);
 #endif
 }
 
diff --git a/arch/i386/math-emu/Makefile b/arch/i386/math-emu/Makefile
index 1bd2cb40d..588f7ada2 100644
--- a/arch/i386/math-emu/Makefile
+++ b/arch/i386/math-emu/Makefile
@@ -12,18 +12,23 @@ CFLAGS	:= $(CFLAGS) $(PARANOID) $(DEBUG) -fno-builtin $(MATH_EMULATION)
 .S.o:
 	$(CC) -D__ASSEMBLY__ $(PARANOID) -c $<
 
-L_OBJS =fpu_entry.o div_small.o errors.o \
-	fpu_arith.o fpu_aux.o fpu_etc.o fpu_trig.o \
+# From 'C' language sources:
+C_OBJS =fpu_entry.o errors.o \
+	fpu_arith.o fpu_aux.o fpu_etc.o fpu_tags.o fpu_trig.o \
 	load_store.o get_address.o \
 	poly_atan.o poly_l2.o poly_2xm1.o poly_sin.o poly_tan.o \
-	reg_add_sub.o reg_compare.o reg_constant.o reg_ld_str.o \
-	reg_div.o reg_mul.o reg_norm.o \
-	reg_u_add.o reg_u_div.o reg_u_mul.o reg_u_sub.o \
-	reg_round.o \
+	reg_add_sub.o reg_compare.o reg_constant.o reg_convert.o \
+	reg_ld_str.o reg_divide.o reg_mul.o
+
+# From 80x86 assembler sources:
+A_OBJS =reg_u_add.o reg_u_div.o reg_u_mul.o reg_u_sub.o \
+	div_small.o reg_norm.o reg_round.o \
 	wm_shrx.o wm_sqrt.o \
 	div_Xsig.o polynom_Xsig.o round_Xsig.o \
 	shr_Xsig.o mul_Xsig.o
 
+L_OBJS =$(C_OBJS) $(A_OBJS)
+
 include $(TOPDIR)/Rules.make
 
 proto:
diff --git a/arch/i386/math-emu/README b/arch/i386/math-emu/README
index 5158b4b81..c6c774092 100644
--- a/arch/i386/math-emu/README
+++ b/arch/i386/math-emu/README
@@ -1,7 +1,7 @@
  +---------------------------------------------------------------------------+
  |  wm-FPU-emu   an FPU emulator for 80386 and 80486SX microprocessors.      |
  |                                                                           |
- | Copyright (C) 1992,1993,1994,1995,1996                                    |
+ | Copyright (C) 1992,1993,1994,1995,1996,1997                               |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
  |                       Australia.  E-mail billm@suburbia.net               |
  |                                                                           |
@@ -44,9 +44,12 @@ some differences.
 Please report bugs, etc to me at:
        billm@suburbia.net
 
+For more information on the emulator and on floating point topics, see
+my web pages, currently at  http://www.suburbia.net/~billm/
+
 
 --Bill Metzenthen
-  October 1996
+  December 1997
 
 
 ----------------------- Internals of wm-FPU-emu -----------------------
@@ -95,8 +98,9 @@ form of re-entrancy which is required by the Linux kernel.
 ----------------------- Limitations of wm-FPU-emu -----------------------
 
 There are a number of differences between the current wm-FPU-emu
-(version 1.20) and the 80486 FPU (apart from bugs). Some of the more
-important differences are listed below:
+(version 2.00) and the 80486 FPU (apart from bugs).  The differences
+are fewer than those which applied to the 1.xx series of the emulator.
+Some of the more important differences are listed below:
 
 The Roundup flag does not have much meaning for the transcendental
 functions and its 80486 value with these functions is likely to differ
@@ -122,18 +126,6 @@ and Unnormals. None of these will be generated by an 80486 or by the
 emulator. Do not use them. The emulator treats them differently in
 detail from the way an 80486 does.
 
-The emulator treats PseudoDenormals differently from an 80486. These
-numbers are in fact properly normalised numbers with the exponent
-offset by 1, and the emulator treats them as such. Unlike the 80486,
-the emulator does not generate a Denormal Operand exception for these
-numbers. The arithmetical results produced when using such a number as
-an operand are the same for the emulator and a real 80486 (apart from
-any slight precision difference for the transcendental functions).
-Neither the emulator nor an 80486 produces one of these numbers as the
-result of any arithmetic operation. An 80486 can keep one of these
-numbers in an FPU register with its identity as a PseudoDenormal, but
-the emulator will not; they are always converted to a valid number.
-
 Self modifying code can cause the emulator to fail. An example of such
 code is:
           movl %esp,[%ebx]
diff --git a/arch/i386/math-emu/div_small.S b/arch/i386/math-emu/div_small.S
index 13ab2b7ae..47099628f 100644
--- a/arch/i386/math-emu/div_small.S
+++ b/arch/i386/math-emu/div_small.S
@@ -12,13 +12,13 @@
  +---------------------------------------------------------------------------*/
 
 /*---------------------------------------------------------------------------+
- |    unsigned long div_small(unsigned long long *x, unsigned long y)        |
+ |    unsigned long FPU_div_small(unsigned long long *x, unsigned long y)    |
  +---------------------------------------------------------------------------*/
 
 #include "fpu_emu.h"
 
 .text
-ENTRY(div_small)
+ENTRY(FPU_div_small)
 	pushl	%ebp
 	movl	%esp,%ebp
 
diff --git a/arch/i386/math-emu/errors.c b/arch/i386/math-emu/errors.c
index 38a72e572..aff145f8a 100644
--- a/arch/i386/math-emu/errors.c
+++ b/arch/i386/math-emu/errors.c
@@ -21,9 +21,9 @@
 
 #include <asm/uaccess.h>
 
+#include "fpu_emu.h"
 #include "fpu_system.h"
 #include "exception.h"
-#include "fpu_emu.h"
 #include "status_w.h"
 #include "control_w.h"
 #include "reg_constant.h"
@@ -36,7 +36,7 @@
 
 void Un_impl(void)
 {
-  unsigned char byte1, FPU_modrm;
+  u_char byte1, FPU_modrm;
   unsigned long address = FPU_ORIG_EIP;
 
   RE_ENTRANT_CHECK_OFF;
@@ -46,13 +46,13 @@ void Un_impl(void)
     {
       while ( 1 )
 	{
-	  get_user(byte1, (unsigned char *) address);
+	  FPU_get_user(byte1, (u_char *) address);
 	  if ( (byte1 & 0xf8) == 0xd8 ) break;
 	  printk("[%02x]", byte1);
 	  address++;
 	}
       printk("%02x ", byte1);
-      get_user(FPU_modrm, 1 + (unsigned char *) address);
+      FPU_get_user(FPU_modrm, 1 + (u_char *) address);
       
       if (FPU_modrm >= 0300)
 	printk("%02x (%02x+%d)\n", FPU_modrm, FPU_modrm & 0xf8, FPU_modrm & 7);
@@ -82,12 +82,12 @@ void FPU_illegal(void)
 
 
 
-void emu_printall(void)
+void FPU_printall(void)
 {
   int i;
-  static const char *tag_desc[] = { "Valid", "Zero", "ERROR", "ERROR",
-                              "DeNorm", "Inf", "NaN", "Empty" };
-  unsigned char byte1, FPU_modrm;
+  static const char *tag_desc[] = { "Valid", "Zero", "ERROR", "Empty",
+                              "DeNorm", "Inf", "NaN" };
+  u_char byte1, FPU_modrm;
   unsigned long address = FPU_ORIG_EIP;
 
   RE_ENTRANT_CHECK_OFF;
@@ -98,7 +98,7 @@ void emu_printall(void)
 #define MAX_PRINTED_BYTES 20
       for ( i = 0; i < MAX_PRINTED_BYTES; i++ )
 	{
-	  get_user(byte1, (unsigned char *) address);
+	  FPU_get_user(byte1, (u_char *) address);
 	  if ( (byte1 & 0xf8) == 0xd8 )
 	    {
 	      printk(" %02x", byte1);
@@ -111,7 +111,7 @@ void emu_printall(void)
 	printk(" [more..]\n");
       else
 	{
-	  get_user(FPU_modrm, 1 + (unsigned char *) address);
+	  FPU_get_user(FPU_modrm, 1 + (u_char *) address);
 	  
 	  if (FPU_modrm >= 0300)
 	    printk(" %02x (%02x+%d)\n", FPU_modrm, FPU_modrm & 0xf8, FPU_modrm & 7);
@@ -166,29 +166,23 @@ printk(" CW: ic=%d rc=%ld%ld pc=%ld%ld iem=%d     ef=%d%d%d%d%d%d\n",
   for ( i = 0; i < 8; i++ )
     {
       FPU_REG *r = &st(i);
-      char tagi = r->tag;
+      u_char tagi = FPU_gettagi(i);
       switch (tagi)
 	{
-	case TW_Empty:
+	case TAG_Empty:
 	  continue;
 	  break;
-	case TW_Zero:
-#if 0
-	  printk("st(%d)  %c .0000 0000 0000 0000         ",
-		 i, r->sign ? '-' : '+');
-	  break;
-#endif
-	case TW_Valid:
-	case TW_NaN:
-/*	case TW_Denormal: */
-	case TW_Infinity:
-	  printk("st(%d)  %c .%04lx %04lx %04lx %04lx e%+-6ld ", i,
-		 r->sign ? '-' : '+',
+	case TAG_Zero:
+	case TAG_Special:
+	  tagi = FPU_Special(r);
+	case TAG_Valid:
+	  printk("st(%d)  %c .%04lx %04lx %04lx %04lx e%+-6d ", i,
+		 getsign(r) ? '-' : '+',
 		 (long)(r->sigh >> 16),
 		 (long)(r->sigh & 0xFFFF),
 		 (long)(r->sigl >> 16),
 		 (long)(r->sigl & 0xFFFF),
-		 r->exp - EXP_BIAS + 1);
+		 exponent(r) - EXP_BIAS + 1);
 	  break;
 	default:
 	  printk("Whoops! Error in errors.c: tag%d is %d ", i, tagi);
@@ -262,6 +256,11 @@ static struct {
 	      0x161  in reg_ld_str.c
 	      0x162  in reg_ld_str.c
 	      0x163  in reg_ld_str.c
+	      0x164  in reg_ld_str.c
+	      0x170  in fpu_tags.c
+	      0x171  in fpu_tags.c
+	      0x172  in fpu_tags.c
+	      0x180  in reg_convert.c
        0x2nn  in an *.S file:
               0x201  in reg_u_add.S
               0x202  in reg_u_div.S
@@ -347,11 +346,11 @@ void FPU_exception(int n)
       if ( n == EX_INTERNAL )
 	{
 	  printk("FPU emulator: Internal error type 0x%04x\n", int_type);
-	  emu_printall();
+	  FPU_printall();
 	}
 #ifdef PRINT_MESSAGES
       else
-	emu_printall();
+	FPU_printall();
 #endif PRINT_MESSAGES
 
       /*
@@ -369,24 +368,97 @@ void FPU_exception(int n)
 }
 
 
-/* Real operation attempted on two operands, one a NaN. */
-/* Returns nz if the exception is unmasked */
-asmlinkage int real_2op_NaN(FPU_REG const *a, FPU_REG const *b, FPU_REG *dest)
+/* Real operation attempted on a NaN. */
+/* Returns < 0 if the exception is unmasked */
+int real_1op_NaN(FPU_REG *a)
 {
-  FPU_REG const *x;
-  int signalling;
+  int signalling, isNaN;
+
+  isNaN = (exponent(a) == EXP_OVER) && (a->sigh & 0x80000000);
 
   /* The default result for the case of two "equal" NaNs (signs may
      differ) is chosen to reproduce 80486 behaviour */
-  x = a;
-  if (a->tag == TW_NaN)
+  signalling = isNaN && !(a->sigh & 0x40000000);
+
+  if ( !signalling )
     {
-      if (b->tag == TW_NaN)
+      if ( !isNaN )  /* pseudo-NaN, or other unsupported? */
+	{
+	  if ( control_word & CW_Invalid )
+	    {
+	      /* Masked response */
+	      reg_copy(&CONST_QNaN, a);
+	    }
+	  EXCEPTION(EX_Invalid);
+	  return (!(control_word & CW_Invalid) ? FPU_Exception : 0) | TAG_Special;
+	}
+      return TAG_Special;
+    }
+
+  if ( control_word & CW_Invalid )
+    {
+      /* The masked response */
+      if ( !(a->sigh & 0x80000000) )  /* pseudo-NaN ? */
+	{
+	  reg_copy(&CONST_QNaN, a);
+	}
+      /* ensure a Quiet NaN */
+      a->sigh |= 0x40000000;
+    }
+
+  EXCEPTION(EX_Invalid);
+
+  return (!(control_word & CW_Invalid) ? FPU_Exception : 0) | TAG_Special;
+}
+
+
+/* Real operation attempted on two operands, one a NaN. */
+/* Returns < 0 if the exception is unmasked */
+int real_2op_NaN(FPU_REG const *b, u_char tagb,
+		 int deststnr,
+		 FPU_REG const *defaultNaN)
+{
+  FPU_REG *dest = &st(deststnr);
+  FPU_REG const *a = dest;
+  u_char taga = FPU_gettagi(deststnr);
+  FPU_REG const *x;
+  int signalling, unsupported;
+
+  if ( taga == TAG_Special )
+    taga = FPU_Special(a);
+  if ( tagb == TAG_Special )
+    tagb = FPU_Special(b);
+
+  /* TW_NaN is also used for unsupported data types. */
+  unsupported = ((taga == TW_NaN)
+		 && !((exponent(a) == EXP_OVER) && (a->sigh & 0x80000000)))
+    || ((tagb == TW_NaN)
+	&& !((exponent(b) == EXP_OVER) && (b->sigh & 0x80000000)));
+  if ( unsupported )
+    {
+      if ( control_word & CW_Invalid )
+	{
+	  /* Masked response */
+	  FPU_copy_to_regi(&CONST_QNaN, TAG_Special, deststnr);
+	}
+      EXCEPTION(EX_Invalid);
+      return (!(control_word & CW_Invalid) ? FPU_Exception : 0) | TAG_Special;
+    }
+
+  if (taga == TW_NaN)
+    {
+      x = a;
+      if (tagb == TW_NaN)
 	{
 	  signalling = !(a->sigh & b->sigh & 0x40000000);
-	  /* find the "larger" */
-	  if ( significand(a) < significand(b) )
+	  if ( significand(b) > significand(a) )
 	    x = b;
+	  else if ( significand(b) == significand(a) )
+	    {
+	      /* The default result for the case of two "equal" NaNs (signs may
+		 differ) is chosen to reproduce 80486 behaviour */
+	      x = defaultNaN;
+	    }
 	}
       else
 	{
@@ -396,7 +468,7 @@ asmlinkage int real_2op_NaN(FPU_REG const *a, FPU_REG const *b, FPU_REG *dest)
     }
   else
 #ifdef PARANOID
-    if (b->tag == TW_NaN)
+    if (tagb == TW_NaN)
 #endif PARANOID
     {
       signalling = !(b->sigh & 0x40000000);
@@ -411,33 +483,32 @@ asmlinkage int real_2op_NaN(FPU_REG const *a, FPU_REG const *b, FPU_REG *dest)
     }
 #endif PARANOID
 
-  if ( !signalling )
+  if ( (!signalling) || (control_word & CW_Invalid) )
     {
-      if ( !(x->sigh & 0x80000000) )  /* pseudo-NaN ? */
-	x = &CONST_QNaN;
-      reg_move(x, dest);
-      return 0;
-    }
+      if ( ! x )
+	x = b;
 
-  if ( control_word & CW_Invalid )
-    {
-      /* The masked response */
       if ( !(x->sigh & 0x80000000) )  /* pseudo-NaN ? */
 	x = &CONST_QNaN;
-      reg_move(x, dest);
+
+      FPU_copy_to_regi(x, TAG_Special, deststnr);
+
+      if ( !signalling )
+	return TAG_Special;
+
       /* ensure a Quiet NaN */
       dest->sigh |= 0x40000000;
     }
 
   EXCEPTION(EX_Invalid);
-  
-  return !(control_word & CW_Invalid);
+
+  return (!(control_word & CW_Invalid) ? FPU_Exception : 0) | TAG_Special;
 }
 
 
 /* Invalid arith operation on Valid registers */
-/* Returns nz if the exception is unmasked */
-asmlinkage int arith_invalid(FPU_REG *dest)
+/* Returns < 0 if the exception is unmasked */
+asmlinkage int arith_invalid(int deststnr)
 {
 
   EXCEPTION(EX_Invalid);
@@ -445,28 +516,31 @@ asmlinkage int arith_invalid(FPU_REG *dest)
   if ( control_word & CW_Invalid )
     {
       /* The masked response */
-      reg_move(&CONST_QNaN, dest);
+      FPU_copy_to_regi(&CONST_QNaN, TAG_Special, deststnr);
     }
   
-  return !(control_word & CW_Invalid);
+  return (!(control_word & CW_Invalid) ? FPU_Exception : 0) | TAG_Valid;
 
 }
 
 
 /* Divide a finite number by zero */
-asmlinkage int divide_by_zero(int sign, FPU_REG *dest)
+asmlinkage int FPU_divide_by_zero(int deststnr, u_char sign)
 {
+  FPU_REG *dest = &st(deststnr);
+  int tag = TAG_Valid;
 
   if ( control_word & CW_ZeroDiv )
     {
       /* The masked response */
-      reg_move(&CONST_INF, dest);
-      dest->sign = (unsigned char)sign;
+      FPU_copy_to_regi(&CONST_INF, TAG_Special, deststnr);
+      setsign(dest, sign);
+      tag = TAG_Special;
     }
  
   EXCEPTION(EX_ZeroDiv);
 
-  return !(control_word & CW_ZeroDiv);
+  return (!(control_word & CW_ZeroDiv) ? FPU_Exception : 0) | tag;
 
 }
 
@@ -495,7 +569,6 @@ asmlinkage void set_precision_flag_up(void)
     partial_status |= (SW_Precision | SW_C1);   /* The masked response */
   else
     EXCEPTION(EX_Precision | SW_C1);
-
 }
 
 
@@ -517,32 +590,31 @@ asmlinkage int denormal_operand(void)
   if ( control_word & CW_Denormal )
     {   /* The masked response */
       partial_status |= SW_Denorm_Op;
-      return 0;
+      return TAG_Special;
     }
   else
     {
       EXCEPTION(EX_Denormal);
-      return 1;
+      return TAG_Special | FPU_Exception;
     }
 }
 
 
 asmlinkage int arith_overflow(FPU_REG *dest)
 {
+  int tag = TAG_Valid;
 
   if ( control_word & CW_Overflow )
     {
-      char sign;
       /* The masked response */
 /* ###### The response here depends upon the rounding mode */
-      sign = dest->sign;
-      reg_move(&CONST_INF, dest);
-      dest->sign = sign;
+      reg_copy(&CONST_INF, dest);
+      tag = TAG_Special;
     }
   else
     {
       /* Subtract the magic number from the exponent */
-      dest->exp -= (3 * (1 << 13));
+      addexponent(dest, (-3 * (1 << 13)));
     }
 
   EXCEPTION(EX_Overflow);
@@ -553,30 +625,36 @@ asmlinkage int arith_overflow(FPU_REG *dest)
 	 The roundup bit (C1) is also set because we have
 	 "rounded" upwards to Infinity. */
       EXCEPTION(EX_Precision | SW_C1);
-      return !(control_word & CW_Precision);
+      return tag;
     }
 
-  return 0;
+  return tag;
 
 }
 
 
 asmlinkage int arith_underflow(FPU_REG *dest)
 {
+  int tag = TAG_Valid;
 
   if ( control_word & CW_Underflow )
     {
       /* The masked response */
-      if ( dest->exp <= EXP_UNDER - 63 )
+      if ( exponent16(dest) <= EXP_UNDER - 63 )
 	{
-	  reg_move(&CONST_Z, dest);
+	  reg_copy(&CONST_Z, dest);
 	  partial_status &= ~SW_C1;       /* Round down. */
+	  tag = TAG_Zero;
+	}
+      else
+	{
+	  stdexp(dest);
 	}
     }
   else
     {
       /* Add the magic number to the exponent. */
-      dest->exp += (3 * (1 << 13));
+      addexponent(dest, (3 * (1 << 13)) + EXTENDED_Ebias);
     }
 
   EXCEPTION(EX_Underflow);
@@ -584,22 +662,22 @@ asmlinkage int arith_underflow(FPU_REG *dest)
     {
       /* The underflow exception is masked. */
       EXCEPTION(EX_Precision);
-      return !(control_word & CW_Precision);
+      return tag;
     }
 
-  return 0;
+  return tag;
 
 }
 
 
-void stack_overflow(void)
+void FPU_stack_overflow(void)
 {
 
  if ( control_word & CW_Invalid )
     {
       /* The masked response */
       top--;
-      reg_move(&CONST_QNaN, &st(0));
+      FPU_copy_to_reg0(&CONST_QNaN, TAG_Special);
     }
 
   EXCEPTION(EX_StackOver);
@@ -609,13 +687,13 @@ void stack_overflow(void)
 }
 
 
-void stack_underflow(void)
+void FPU_stack_underflow(void)
 {
 
  if ( control_word & CW_Invalid )
     {
       /* The masked response */
-      reg_move(&CONST_QNaN, &st(0));
+      FPU_copy_to_reg0(&CONST_QNaN, TAG_Special);
     }
 
   EXCEPTION(EX_StackUnder);
@@ -625,13 +703,13 @@ void stack_underflow(void)
 }
 
 
-void stack_underflow_i(int i)
+void FPU_stack_underflow_i(int i)
 {
 
  if ( control_word & CW_Invalid )
     {
       /* The masked response */
-      reg_move(&CONST_QNaN, &(st(i)));
+      FPU_copy_to_regi(&CONST_QNaN, TAG_Special, i);
     }
 
   EXCEPTION(EX_StackUnder);
@@ -641,14 +719,14 @@ void stack_underflow_i(int i)
 }
 
 
-void stack_underflow_pop(int i)
+void FPU_stack_underflow_pop(int i)
 {
 
  if ( control_word & CW_Invalid )
     {
       /* The masked response */
-      reg_move(&CONST_QNaN, &(st(i)));
-      pop();
+      FPU_copy_to_regi(&CONST_QNaN, TAG_Special, i);
+      FPU_pop();
     }
 
   EXCEPTION(EX_StackUnder);
diff --git a/arch/i386/math-emu/fpu_arith.c b/arch/i386/math-emu/fpu_arith.c
index 96e6bd89b..fcad7ec08 100644
--- a/arch/i386/math-emu/fpu_arith.c
+++ b/arch/i386/math-emu/fpu_arith.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Code to implement the FPU register/register arithmetic instructions       |
  |                                                                           |
- | Copyright (C) 1992,1993                                                   |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1997                                              |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -19,16 +19,18 @@
 void fadd__()
 {
   /* fadd st,st(i) */
+  int i = FPU_rm;
   clear_C1();
-  reg_add(&st(0), &st(FPU_rm), &st(0), control_word);
+  FPU_add(&st(i), FPU_gettagi(i), 0, control_word);
 }
 
 
 void fmul__()
 {
   /* fmul st,st(i) */
+  int i = FPU_rm;
   clear_C1();
-  reg_mul(&st(0), &st(FPU_rm), &st(0), control_word);
+  FPU_mul(&st(i), FPU_gettagi(i), 0, control_word);
 }
 
 
@@ -37,7 +39,7 @@ void fsub__()
 {
   /* fsub st,st(i) */
   clear_C1();
-  reg_sub(&st(0), &st(FPU_rm), &st(0), control_word);
+  FPU_sub(0, FPU_rm, control_word);
 }
 
 
@@ -45,7 +47,7 @@ void fsubr_()
 {
   /* fsubr st,st(i) */
   clear_C1();
-  reg_sub(&st(FPU_rm), &st(0), &st(0), control_word);
+  FPU_sub(REV, FPU_rm, control_word);
 }
 
 
@@ -53,7 +55,7 @@ void fdiv__()
 {
   /* fdiv st,st(i) */
   clear_C1();
-  reg_div(&st(0), &st(FPU_rm), &st(0), control_word);
+  FPU_div(0, FPU_rm, control_word);
 }
 
 
@@ -61,7 +63,7 @@ void fdivr_()
 {
   /* fdivr st,st(i) */
   clear_C1();
-  reg_div(&st(FPU_rm), &st(0), &st(0), control_word);
+  FPU_div(REV, FPU_rm, control_word);
 }
 
 
@@ -69,8 +71,9 @@ void fdivr_()
 void fadd_i()
 {
   /* fadd st(i),st */
+  int i = FPU_rm;
   clear_C1();
-  reg_add(&st(0), &st(FPU_rm), &st(FPU_rm), control_word);
+  FPU_add(&st(i), FPU_gettagi(i), i, control_word);
 }
 
 
@@ -78,27 +81,23 @@ void fmul_i()
 {
   /* fmul st(i),st */
   clear_C1();
-  reg_mul(&st(0), &st(FPU_rm), &st(FPU_rm), control_word);
+  FPU_mul(&st(0), FPU_gettag0(), FPU_rm, control_word);
 }
 
 
 void fsubri()
 {
   /* fsubr st(i),st */
-  /* This is the sense of the 80486 manual
-     reg_sub(&st(FPU_rm), &st(0), &st(FPU_rm), control_word); */
   clear_C1();
-  reg_sub(&st(0), &st(FPU_rm), &st(FPU_rm), control_word);
+  FPU_sub(DEST_RM, FPU_rm, control_word);
 }
 
 
 void fsub_i()
 {
   /* fsub st(i),st */
-  /* This is the sense of the 80486 manual
-     reg_sub(&st(0), &st(FPU_rm), &st(FPU_rm), control_word); */
   clear_C1();
-  reg_sub(&st(FPU_rm), &st(0), &st(FPU_rm), control_word);
+  FPU_sub(REV|DEST_RM, FPU_rm, control_word);
 }
 
 
@@ -106,7 +105,7 @@ void fdivri()
 {
   /* fdivr st(i),st */
   clear_C1();
-  reg_div(&st(0), &st(FPU_rm), &st(FPU_rm), control_word);
+  FPU_div(DEST_RM, FPU_rm, control_word);
 }
 
 
@@ -114,7 +113,7 @@ void fdiv_i()
 {
   /* fdiv st(i),st */
   clear_C1();
-  reg_div(&st(FPU_rm), &st(0), &st(FPU_rm), control_word);
+  FPU_div(REV|DEST_RM, FPU_rm, control_word);
 }
 
 
@@ -122,9 +121,10 @@ void fdiv_i()
 void faddp_()
 {
   /* faddp st(i),st */
+  int i = FPU_rm;
   clear_C1();
-  if ( !reg_add(&st(0), &st(FPU_rm), &st(FPU_rm), control_word) )
-    pop();
+  if ( FPU_add(&st(i), FPU_gettagi(i), i, control_word) >= 0 )
+    FPU_pop();
 }
 
 
@@ -132,8 +132,8 @@ void fmulp_()
 {
   /* fmulp st(i),st */
   clear_C1();
-  if ( !reg_mul(&st(0), &st(FPU_rm), &st(FPU_rm), control_word) )
-    pop();
+  if ( FPU_mul(&st(0), FPU_gettag0(), FPU_rm, control_word) >= 0 )
+    FPU_pop();
 }
 
 
@@ -141,22 +141,18 @@ void fmulp_()
 void fsubrp()
 {
   /* fsubrp st(i),st */
-  /* This is the sense of the 80486 manual
-     reg_sub(&st(FPU_rm), &st(0), &st(FPU_rm), control_word); */
   clear_C1();
-  if ( !reg_sub(&st(0), &st(FPU_rm), &st(FPU_rm), control_word) )
-    pop();
+  if ( FPU_sub(DEST_RM, FPU_rm, control_word) >= 0 )
+    FPU_pop();
 }
 
 
 void fsubp_()
 {
   /* fsubp st(i),st */
-  /* This is the sense of the 80486 manual
-     reg_sub(&st(0), &st(FPU_rm), &st(FPU_rm), control_word); */
   clear_C1();
-  if ( !reg_sub(&st(FPU_rm), &st(0), &st(FPU_rm), control_word) )
-    pop();
+  if ( FPU_sub(REV|DEST_RM, FPU_rm, control_word) >= 0 )
+    FPU_pop();
 }
 
 
@@ -164,8 +160,8 @@ void fdivrp()
 {
   /* fdivrp st(i),st */
   clear_C1();
-  if ( !reg_div(&st(0), &st(FPU_rm), &st(FPU_rm), control_word) )
-    pop();
+  if ( FPU_div(DEST_RM, FPU_rm, control_word) >= 0 )
+    FPU_pop();
 }
 
 
@@ -173,7 +169,6 @@ void fdivp_()
 {
   /* fdivp st(i),st */
   clear_C1();
-  if ( !reg_div(&st(FPU_rm), &st(0), &st(FPU_rm), control_word) )
-    pop();
+  if ( FPU_div(REV|DEST_RM, FPU_rm, control_word) >= 0 )
+    FPU_pop();
 }
-
diff --git a/arch/i386/math-emu/fpu_asm.h b/arch/i386/math-emu/fpu_asm.h
index d08fbc874..d89d676c7 100644
--- a/arch/i386/math-emu/fpu_asm.h
+++ b/arch/i386/math-emu/fpu_asm.h
@@ -1,9 +1,9 @@
 /*---------------------------------------------------------------------------+
  |  fpu_asm.h                                                                |
  |                                                                           |
- | Copyright (C) 1992,1995                                                   |
+ | Copyright (C) 1992,1995,1997                                              |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
+ |                       Australia.  E-mail billm@suburbia.net               |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
@@ -19,13 +19,14 @@
 #define	PARAM2	12(%ebp)
 #define	PARAM3	16(%ebp)
 #define	PARAM4	20(%ebp)
+#define	PARAM5	24(%ebp)
+#define	PARAM6	28(%ebp)
+#define	PARAM7	32(%ebp)
 
-#define SIGL_OFFSET 8
-#define SIGN(x)	(x)
-#define	TAG(x)	1(x)
-#define	EXP(x)	4(x)
+#define SIGL_OFFSET 0
+#define	EXP(x)	8(x)
 #define SIG(x)	SIGL_OFFSET##(x)
 #define	SIGL(x)	SIGL_OFFSET##(x)
-#define	SIGH(x)	12(x)
+#define	SIGH(x)	4(x)
 
 #endif _FPU_ASM_H_
diff --git a/arch/i386/math-emu/fpu_aux.c b/arch/i386/math-emu/fpu_aux.c
index 0d35fe19b..112c173f2 100644
--- a/arch/i386/math-emu/fpu_aux.c
+++ b/arch/i386/math-emu/fpu_aux.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Code to implement some of the FPU auxiliary instructions.                 |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1994,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -32,15 +32,11 @@ void fclex(void)
 /* Needs to be externally visible */
 void finit()
 {
-  int r;
   control_word = 0x037f;
   partial_status = 0;
   top = 0;            /* We don't keep top in the status word internally. */
-  for (r = 0; r < 8; r++)
-    {
-      regs[r].tag = TW_Empty;
-    }
-  /* The behaviour is different to that detailed in
+  fpu_tag_word = 0xffff;
+  /* The behaviour is different from that detailed in
      Section 15.1.6 of the Intel manual */
   operand_address.offset = 0;
   operand_address.selector = 0;
@@ -99,19 +95,27 @@ void fp_nop()
 void fld_i_()
 {
   FPU_REG *st_new_ptr;
+  int i;
+  u_char tag;
 
   if ( STACK_OVERFLOW )
-    { stack_overflow(); return; }
+    { FPU_stack_overflow(); return; }
 
   /* fld st(i) */
-  if ( NOT_EMPTY(FPU_rm) )
-    { reg_move(&st(FPU_rm), st_new_ptr); push(); }
+  i = FPU_rm;
+  if ( NOT_EMPTY(i) )
+    {
+      reg_copy(&st(i), st_new_ptr);
+      tag = FPU_gettagi(i);
+      push();
+      FPU_settag0(tag);
+    }
   else
     {
       if ( control_word & CW_Invalid )
 	{
 	  /* The masked response */
-	  stack_underflow();
+	  FPU_stack_underflow();
 	}
       else
 	EXCEPTION(EX_StackUnder);
@@ -124,61 +128,77 @@ void fxch_i()
 {
   /* fxch st(i) */
   FPU_REG t;
-  register FPU_REG *sti_ptr = &st(FPU_rm), *st0_ptr = &st(0);
-
-  if ( st0_ptr->tag == TW_Empty )
+  int i = FPU_rm;
+  FPU_REG *st0_ptr = &st(0), *sti_ptr = &st(i);
+  long tag_word = fpu_tag_word;
+  int regnr = top & 7, regnri = ((regnr + i) & 7);
+  u_char st0_tag = (tag_word >> (regnr*2)) & 3;
+  u_char sti_tag = (tag_word >> (regnri*2)) & 3;
+
+  if ( st0_tag == TAG_Empty )
     {
-      if ( sti_ptr->tag == TW_Empty )
+      if ( sti_tag == TAG_Empty )
 	{
-	  stack_underflow();
-	  stack_underflow_i(FPU_rm);
+	  FPU_stack_underflow();
+	  FPU_stack_underflow_i(i);
 	  return;
 	}
       if ( control_word & CW_Invalid )
-	reg_move(sti_ptr, st0_ptr);   /* Masked response */
-      stack_underflow_i(FPU_rm);
+	{
+	  /* Masked response */
+	  FPU_copy_to_reg0(sti_ptr, sti_tag);
+	}
+      FPU_stack_underflow_i(i);
       return;
     }
-  if ( sti_ptr->tag == TW_Empty )
+  if ( sti_tag == TAG_Empty )
     {
       if ( control_word & CW_Invalid )
-	reg_move(st0_ptr, sti_ptr);   /* Masked response */
-      stack_underflow();
+	{
+	  /* Masked response */
+	  FPU_copy_to_regi(st0_ptr, st0_tag, i);
+	}
+      FPU_stack_underflow();
       return;
     }
   clear_C1();
-  reg_move(st0_ptr, &t);
-  reg_move(sti_ptr, st0_ptr);
-  reg_move(&t, sti_ptr);
+
+  reg_copy(st0_ptr, &t);
+  reg_copy(sti_ptr, st0_ptr);
+  reg_copy(&t, sti_ptr);
+
+  tag_word &= ~(3 << (regnr*2)) & ~(3 << (regnri*2));
+  tag_word |= (sti_tag << (regnr*2)) | (st0_tag << (regnri*2));
+  fpu_tag_word = tag_word;
 }
 
 
 void ffree_()
 {
   /* ffree st(i) */
-  st(FPU_rm).tag = TW_Empty;
+  FPU_settagi(FPU_rm, TAG_Empty);
 }
 
 
 void ffreep()
 {
   /* ffree st(i) + pop - unofficial code */
-  st(FPU_rm).tag = TW_Empty;
-  pop();
+  FPU_settagi(FPU_rm, TAG_Empty);
+  FPU_pop();
 }
 
 
 void fst_i_()
 {
   /* fst st(i) */
-  reg_move(&st(0), &st(FPU_rm));
+  FPU_copy_to_regi(&st(0), FPU_gettag0(), FPU_rm);
 }
 
 
 void fstp_i()
 {
   /* fstp st(i) */
-  reg_move(&st(0), &st(FPU_rm));
-  pop();
+  FPU_copy_to_regi(&st(0), FPU_gettag0(), FPU_rm);
+  FPU_pop();
 }
 
diff --git a/arch/i386/math-emu/fpu_emu.h b/arch/i386/math-emu/fpu_emu.h
index b8385db1f..e5734c82d 100644
--- a/arch/i386/math-emu/fpu_emu.h
+++ b/arch/i386/math-emu/fpu_emu.h
@@ -1,9 +1,9 @@
 /*---------------------------------------------------------------------------+
  |  fpu_emu.h                                                                |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
+ | Copyright (C) 1992,1993,1994,1997                                         |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ |                       Australia.  E-mail   billm@suburbia.net             |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
@@ -12,14 +12,6 @@
 #define _FPU_EMU_H_
 
 /*
- * Define DENORM_OPERAND to make the emulator detect denormals
- * and use the denormal flag of the status word. Note: this only
- * affects the flag and corresponding interrupt, the emulator
- * will always generate denormals and operate upon them as required.
- */
-#define DENORM_OPERAND
-
-/*
  * Define PECULIAR_486 to get a closer approximation to 80486 behaviour,
  * rather than behaviour which appears to be cleaner.
  * This is a matter of opinion: for all I know, the 80486 may simply
@@ -38,28 +30,51 @@
 #define EXP_BIAS	Const(0)
 #define EXP_OVER	Const(0x4000)    /* smallest invalid large exponent */
 #define	EXP_UNDER	Const(-0x3fff)   /* largest invalid small exponent */
+#define EXP_WAY_UNDER   Const(-0x6000)   /* Below the smallest denormal, but
+					    still a 16 bit nr. */
 #define EXP_Infinity    EXP_OVER
 #define EXP_NaN         EXP_OVER
 
+#define EXTENDED_Ebias Const(0x3fff)
+#define EXTENDED_Emin (-0x3ffe)  /* smallest valid exponent */
+
 #define SIGN_POS	Const(0)
-#define SIGN_NEG	Const(1)
+#define SIGN_NEG	Const(0x80)
+
+#define SIGN_Positive	Const(0)
+#define SIGN_Negative	Const(0x8000)
 
-/* Keep the order TW_Valid, TW_Zero, TW_Denormal */
-#define TW_Valid	Const(0)	/* valid */
-#define TW_Zero		Const(1)	/* zero */
+
+/* Keep the order TAG_Valid, TAG_Zero, TW_Denormal */
 /* The following fold to 2 (Special) in the Tag Word */
-/* #define TW_Denormal     Const(4) */       /* De-normal */
+#define TW_Denormal     Const(4)        /* De-normal */
 #define TW_Infinity	Const(5)	/* + or - infinity */
 #define	TW_NaN		Const(6)	/* Not a Number */
+#define	TW_Unsupported	Const(7)	/* Not supported by an 80486 */
+
+#define TAG_Valid	Const(0)	/* valid */
+#define TAG_Zero	Const(1)	/* zero */
+#define TAG_Special	Const(2)	/* De-normal, + or - infinity,
+					   or Not a Number */
+#define TAG_Empty	Const(3)	/* empty */
+
+#define LOADED_DATA	Const(10101)	/* Special st() number to identify
+					   loaded data (not on stack). */
+
+/* A few flags (must be >= 0x10). */
+#define REV             0x10
+#define DEST_RM         0x20
+#define LOADED          0x40
 
-#define TW_Empty	Const(7)	/* empty */
+#define FPU_Exception   Const(0x80000000)   /* Added to tag returns. */
 
 
 #ifndef __ASSEMBLY__
 
-#include <asm/sigcontext.h>	/* for struct _fpstate */
-#include <asm/math_emu.h>
+#include "fpu_system.h"
 
+#include <asm/sigcontext.h>   /* for struct _fpstate */
+#include <asm/math_emu.h>
 #include <linux/linkage.h>
 
 /*
@@ -67,7 +82,7 @@
  */
 
 #ifdef RE_ENTRANT_CHECKING
-extern char emulating;
+extern u_char emulating;
 #  define RE_ENTRANT_CHECK_OFF emulating = 0
 #  define RE_ENTRANT_CHECK_ON emulating = 1
 #else
@@ -97,18 +112,24 @@ extern char emulating;
 
 struct address {
   unsigned int offset;
-  unsigned short selector;
-  unsigned short opcode:11,
-		 empty:5;
+  unsigned int selector:16;
+  unsigned int opcode:11;
+  unsigned int empty:5;
+};
+struct fpu__reg {
+  unsigned sigl;
+  unsigned sigh;
+  short exp;
 };
+
 typedef void (*FUNC)(void);
-typedef struct fpu_reg FPU_REG;
-typedef void (*FUNC_ST0)(FPU_REG *st0_ptr);
-typedef struct { unsigned char address_size, operand_size, segment; }
+typedef struct fpu__reg FPU_REG;
+typedef void (*FUNC_ST0)(FPU_REG *st0_ptr, u_char st0_tag);
+typedef struct { u_char address_size, operand_size, segment; }
         overrides;
 /* This structure is 32 bits: */
 typedef struct { overrides override;
-		 unsigned char default_mode; } fpu_addr_modes;
+		 u_char default_mode; } fpu_addr_modes;
 /* PROTECTED has a restricted meaning in the emulator; it is used
    to signal that the emulator needs to do special things to ensure
    that protection is respected in a segmented model. */
@@ -117,27 +138,50 @@ typedef struct { overrides override;
 #define VM86      SIXTEEN
 #define PM16      (SIXTEEN | PROTECTED)
 #define SEG32     PROTECTED
-extern unsigned char const data_sizes_16[32];
+extern u_char const data_sizes_16[32];
+
+#define register_base ((u_char *) registers )
+#define fpu_register(x)  ( * ((FPU_REG *)( register_base + 10 * (x & 7) )) )
+#define	st(x)      ( * ((FPU_REG *)( register_base + 10 * ((top+x) & 7) )) )
 
-#define	st(x)	( regs[((top+x) &7 )] )
+#define	STACK_OVERFLOW	(FPU_stackoverflow(&st_new_ptr))
+#define	NOT_EMPTY(i)	(!FPU_empty_i(i))
 
-#define	STACK_OVERFLOW	(st_new_ptr = &st(-1), st_new_ptr->tag != TW_Empty)
-#define	NOT_EMPTY(i)	(st(i).tag != TW_Empty)
-#define	NOT_EMPTY_ST0	(st0_tag ^ TW_Empty)
+#define	NOT_EMPTY_ST0	(st0_tag ^ TAG_Empty)
 
-#define pop()	{ regs[(top++ & 7 )].tag = TW_Empty; }
-#define poppop() { regs[((top + 1) & 7 )].tag \
-		     = regs[(top & 7 )].tag = TW_Empty; \
-		   top += 2; }
+#define poppop() { FPU_pop(); FPU_pop(); }
 
 /* push() does not affect the tags */
 #define push()	{ top--; }
 
+#define signbyte(a) (((u_char *)(a))[9])
+#define getsign(a) (signbyte(a) & 0x80)
+#define setsign(a,b) { if (b) signbyte(a) |= 0x80; else signbyte(a) &= 0x7f; }
+#define copysign(a,b) { if (getsign(a)) signbyte(b) |= 0x80; \
+                        else signbyte(b) &= 0x7f; }
+#define changesign(a) { signbyte(a) ^= 0x80; }
+#define setpositive(a) { signbyte(a) &= 0x7f; }
+#define setnegative(a) { signbyte(a) |= 0x80; }
+#define signpositive(a) ( (signbyte(a) & 0x80) == 0 )
+#define signnegative(a) (signbyte(a) & 0x80)
+
+#include "fpu_proto.h"
+
+static inline void reg_copy(FPU_REG const *x, FPU_REG *y)
+{
+  *(short *)&(y->exp) = *(const short *)&(x->exp); 
+  *(long long *)&(y->sigl) = *(const long long *)&(x->sigl);
+}
+
+#define exponent(x)  (((*(short *)&((x)->exp)) & 0x7fff) - EXTENDED_Ebias)
+#define setexponentpos(x,y) { (*(short *)&((x)->exp)) = \
+  ((y) + EXTENDED_Ebias) & 0x7fff; }
+#define exponent16(x)         (*(short *)&((x)->exp))
+#define setexponent16(x,y)  { (*(short *)&((x)->exp)) = (y); }
+#define addexponent(x,y)    { (*(short *)&((x)->exp)) += (y); }
+#define stdexp(x)           { (*(short *)&((x)->exp)) += EXTENDED_Ebias; }
 
-#define reg_move(x, y) { \
-		 *(short *)&((y)->sign) = *(const short *)&((x)->sign); \
-		 *(long *)&((y)->exp) = *(const long *)&((x)->exp); \
-		 *(long long *)&((y)->sigl) = *(const long long *)&((x)->sigl); }
+#define isdenormal(ptr)   (exponent(ptr) == EXP_BIAS+EXP_UNDER)
 
 #define significand(x) ( ((unsigned long long *)&((x)->sigl))[0] )
 
@@ -145,24 +189,26 @@ extern unsigned char const data_sizes_16[32];
 /*----- Prototypes for functions written in assembler -----*/
 /* extern void reg_move(FPU_REG *a, FPU_REG *b); */
 
-asmlinkage void normalize(FPU_REG *x);
-asmlinkage void normalize_nuo(FPU_REG *x);
-asmlinkage int reg_div(FPU_REG const *arg1, FPU_REG const *arg2,
-		       FPU_REG *answ, unsigned int control_w);
-asmlinkage int reg_u_sub(FPU_REG const *arg1, FPU_REG const *arg2,
-			 FPU_REG *answ, unsigned int control_w);
-asmlinkage int reg_u_mul(FPU_REG const *arg1, FPU_REG const *arg2,
-			 FPU_REG *answ, unsigned int control_w);
-asmlinkage int reg_u_div(FPU_REG const *arg1, FPU_REG const *arg2,
-			 FPU_REG *answ, unsigned int control_w);
-asmlinkage int reg_u_add(FPU_REG const *arg1, FPU_REG const *arg2,
-			 FPU_REG *answ, unsigned int control_w);
-asmlinkage int wm_sqrt(FPU_REG *n, unsigned int control_w);
-asmlinkage unsigned	shrx(void *l, unsigned x);
-asmlinkage unsigned	shrxs(void *v, unsigned x);
-asmlinkage unsigned long div_small(unsigned long long *x, unsigned long y);
-asmlinkage void round_reg(FPU_REG *arg, unsigned int extent,
-		      unsigned int control_w);
+asmlinkage int FPU_normalize(FPU_REG *x);
+asmlinkage int FPU_normalize_nuo(FPU_REG *x);
+asmlinkage int FPU_u_sub(FPU_REG const *arg1, FPU_REG const *arg2,
+			 FPU_REG *answ, unsigned int control_w, u_char sign,
+			 int expa, int expb);
+asmlinkage int FPU_u_mul(FPU_REG const *arg1, FPU_REG const *arg2,
+			 FPU_REG *answ, unsigned int control_w, u_char sign,
+			 int expon);
+asmlinkage int FPU_u_div(FPU_REG const *arg1, FPU_REG const *arg2,
+			 FPU_REG *answ, unsigned int control_w, u_char sign);
+asmlinkage int FPU_u_add(FPU_REG const *arg1, FPU_REG const *arg2,
+			 FPU_REG *answ, unsigned int control_w, u_char sign,
+			 int expa, int expb);
+asmlinkage int wm_sqrt(FPU_REG *n, int dummy1, int dummy2,
+		       unsigned int control_w, u_char sign);
+asmlinkage unsigned	FPU_shrx(void *l, unsigned x);
+asmlinkage unsigned	FPU_shrxs(void *v, unsigned x);
+asmlinkage unsigned long FPU_div_small(unsigned long long *x, unsigned long y);
+asmlinkage int FPU_round(FPU_REG *arg, unsigned int extent, int dummy,
+			 unsigned int control_w, u_char sign);
 
 #ifndef MAKING_PROTO
 #include "fpu_proto.h"
diff --git a/arch/i386/math-emu/fpu_entry.c b/arch/i386/math-emu/fpu_entry.c
index 943daf3b9..36ca90a35 100644
--- a/arch/i386/math-emu/fpu_entry.c
+++ b/arch/i386/math-emu/fpu_entry.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | The entry functions for wm-FPU-emu                                        |
  |                                                                           |
- | Copyright (C) 1992,1993,1994,1996                                         |
+ | Copyright (C) 1992,1993,1994,1996,1997                                    |
  |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
- |                  E-mail   billm@jacobi.maths.monash.edu.au                |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  | See the files "README" and "COPYING" for further copyright and warranty   |
  | information.                                                              |
@@ -54,27 +54,27 @@
 #define _df_d8_ fstp_i    /* unofficial code (1f) */
 
 static FUNC const st_instr_table[64] = {
-  fadd__,   fld_i_,  __BAD__, __BAD__, fadd_i,  ffree_,  faddp_,  _df_c0_,
-  fmul__,   fxch_i,  __BAD__, __BAD__, fmul_i,  _dd_c8_, fmulp_,  _df_c8_,
-  fcom_st,  fp_nop,  __BAD__, __BAD__, _dc_d0_, fst_i_,  _de_d0_, _df_d0_,
-  fcompst,  _d9_d8_, __BAD__, __BAD__, _dc_d8_, fstp_i,  fcompp,  _df_d8_,
-  fsub__,   fp_etc,  __BAD__, finit_,  fsubri,  fucom_,  fsubrp,  fstsw_,
-  fsubr_,   fconst,  fucompp, __BAD__, fsub_i,  fucomp,  fsubp_,  __BAD__,
-  fdiv__,   trig_a,  __BAD__, __BAD__, fdivri,  __BAD__, fdivrp,  __BAD__,
-  fdivr_,   trig_b,  __BAD__, __BAD__, fdiv_i,  __BAD__, fdivp_,  __BAD__,
+  fadd__,   fld_i_,     __BAD__, __BAD__, fadd_i,  ffree_,  faddp_,  _df_c0_,
+  fmul__,   fxch_i,     __BAD__, __BAD__, fmul_i,  _dd_c8_, fmulp_,  _df_c8_,
+  fcom_st,  fp_nop,     __BAD__, __BAD__, _dc_d0_, fst_i_,  _de_d0_, _df_d0_,
+  fcompst,  _d9_d8_,    __BAD__, __BAD__, _dc_d8_, fstp_i,  fcompp,  _df_d8_,
+  fsub__,   FPU_etc,    __BAD__, finit_,  fsubri,  fucom_,  fsubrp,  fstsw_,
+  fsubr_,   fconst,     fucompp, __BAD__, fsub_i,  fucomp,  fsubp_,  __BAD__,
+  fdiv__,   FPU_triga,  __BAD__, __BAD__, fdivri,  __BAD__, fdivrp,  __BAD__,
+  fdivr_,   FPU_trigb,  __BAD__, __BAD__, fdiv_i,  __BAD__, fdivp_,  __BAD__,
 };
 
 #else     /* Support only documented FPU op-codes */
 
 static FUNC const st_instr_table[64] = {
-  fadd__,   fld_i_,  __BAD__, __BAD__, fadd_i,  ffree_,  faddp_,  __BAD__,
-  fmul__,   fxch_i,  __BAD__, __BAD__, fmul_i,  __BAD__, fmulp_,  __BAD__,
-  fcom_st,  fp_nop,  __BAD__, __BAD__, __BAD__, fst_i_,  __BAD__, __BAD__,
-  fcompst,  __BAD__, __BAD__, __BAD__, __BAD__, fstp_i,  fcompp,  __BAD__,
-  fsub__,   fp_etc,  __BAD__, finit_,  fsubri,  fucom_,  fsubrp,  fstsw_,
-  fsubr_,   fconst,  fucompp, __BAD__, fsub_i,  fucomp,  fsubp_,  __BAD__,
-  fdiv__,   trig_a,  __BAD__, __BAD__, fdivri,  __BAD__, fdivrp,  __BAD__,
-  fdivr_,   trig_b,  __BAD__, __BAD__, fdiv_i,  __BAD__, fdivp_,  __BAD__,
+  fadd__,   fld_i_,     __BAD__, __BAD__, fadd_i,  ffree_,  faddp_,  __BAD__,
+  fmul__,   fxch_i,     __BAD__, __BAD__, fmul_i,  __BAD__, fmulp_,  __BAD__,
+  fcom_st,  fp_nop,     __BAD__, __BAD__, __BAD__, fst_i_,  __BAD__, __BAD__,
+  fcompst,  __BAD__,    __BAD__, __BAD__, __BAD__, fstp_i,  fcompp,  __BAD__,
+  fsub__,   FPU_etc,    __BAD__, finit_,  fsubri,  fucom_,  fsubrp,  fstsw_,
+  fsubr_,   fconst,     fucompp, __BAD__, fsub_i,  fucomp,  fsubp_,  __BAD__,
+  fdiv__,   FPU_triga,  __BAD__, __BAD__, fdivri,  __BAD__, fdivrp,  __BAD__,
+  fdivr_,   FPU_trigb,  __BAD__, __BAD__, fdiv_i,  __BAD__, fdivp_,  __BAD__,
 };
 
 #endif NO_UNDOC_CODE
@@ -95,7 +95,7 @@ static FUNC const st_instr_table[64] = {
 
 /* Un-documented FPU op-codes supported by default. (see above) */
 
-static unsigned char const type_table[64] = {
+static u_char const type_table[64] = {
   _REGI_, _NONE_, _null_, _null_, _REGIi, _REGi_, _REGIp, _REGi_,
   _REGI_, _REGIn, _null_, _null_, _REGIi, _REGI_, _REGIp, _REGI_,
   _REGIc, _NONE_, _null_, _null_, _REGIc, _REG0_, _REGIc, _REG0_,
@@ -108,7 +108,7 @@ static unsigned char const type_table[64] = {
 
 #else     /* Support only documented FPU op-codes */
 
-static unsigned char const type_table[64] = {
+static u_char const type_table[64] = {
   _REGI_, _NONE_, _null_, _null_, _REGIi, _REGi_, _REGIp, _null_,
   _REGI_, _REGIn, _null_, _null_, _REGIi, _null_, _REGIp, _null_,
   _REGIc, _NONE_, _null_, _null_, _null_, _REG0_, _null_, _null_,
@@ -123,26 +123,26 @@ static unsigned char const type_table[64] = {
 
 
 #ifdef RE_ENTRANT_CHECKING
-char emulating=0;
+u_char emulating=0;
 #endif RE_ENTRANT_CHECKING
 
-static int valid_prefix(unsigned char *Byte, unsigned char **fpu_eip,
+static int valid_prefix(u_char *Byte, u_char **fpu_eip,
 			overrides *override);
 
 asmlinkage void math_emulate(long arg)
 {
-  unsigned char  FPU_modrm, byte1;
+  u_char  FPU_modrm, byte1;
   unsigned short code;
   fpu_addr_modes addr_modes;
   int unmasked;
   FPU_REG loaded_data;
+  FPU_REG *st0_ptr;
+  u_char	  loaded_tag, st0_tag;
   void *data_address;
   struct address data_sel_off;
   struct address entry_sel_off;
   unsigned long code_base = 0;
   unsigned long code_limit = 0;  /* Initialized to stop compiler warnings */
-  char	       st0_tag;
-  FPU_REG      *st0_ptr;
   struct desc_struct code_descriptor;
 
 #ifdef RE_ENTRANT_CHECKING
@@ -155,15 +155,6 @@ asmlinkage void math_emulate(long arg)
 
   if (!current->used_math)
     {
-      int i;
-      for ( i = 0; i < 8; i++ )
-	{
-	  /* Make sure that the registers are compatible
-	     with the assumptions of the emulator. */
-	  if ( !((regs[i].exp == EXP_UNDER) && (regs[i].sigh == 0)
-		 && (regs[i].sigl == 0)) )
-	    regs[i].sigh |= 0x80000000;
-	}
       finit();
       current->used_math = 1;
     }
@@ -221,7 +212,7 @@ asmlinkage void math_emulate(long arg)
   if (current->flags & PF_PTRACED)
     FPU_lookahead = 0;
 
-  if ( !valid_prefix(&byte1, (unsigned char **)&FPU_EIP,
+  if ( !valid_prefix(&byte1, (u_char **)&FPU_EIP,
 		     &addr_modes.override) )
     {
       RE_ENTRANT_CHECK_OFF;
@@ -264,7 +255,7 @@ do_another_FPU_instruction:
 
   RE_ENTRANT_CHECK_OFF;
   FPU_code_verify_area(1);
-  get_user(FPU_modrm, (unsigned char *) FPU_EIP);
+  FPU_get_user(FPU_modrm, (u_char *) FPU_EIP);
   RE_ENTRANT_CHECK_ON;
   FPU_EIP++;
 
@@ -287,6 +278,7 @@ do_another_FPU_instruction:
 	   *  interrupts here.
 	   */
 	do_the_FPU_interrupt:
+
 	  FPU_EIP = FPU_ORIG_EIP;	/* Point to current FPU instruction. */
 
 	  RE_ENTRANT_CHECK_OFF;
@@ -309,11 +301,11 @@ do_another_FPU_instruction:
 
       if ( (addr_modes.default_mode & SIXTEEN)
 	  ^ (addr_modes.override.address_size == ADDR_SIZE_PREFIX) )
-	data_address = get_address_16(FPU_modrm, &FPU_EIP, &data_sel_off,
-				      addr_modes);
+	data_address = FPU_get_address_16(FPU_modrm, &FPU_EIP, &data_sel_off,
+					  addr_modes);
       else
-	data_address = get_address(FPU_modrm, &FPU_EIP, &data_sel_off,
-				   addr_modes);
+	data_address = FPU_get_address(FPU_modrm, &FPU_EIP, &data_sel_off,
+				       addr_modes);
 
       if ( addr_modes.default_mode )
 	{
@@ -326,7 +318,7 @@ do_another_FPU_instruction:
 	  unsigned short status1 = partial_status;
 
 	  st0_ptr = &st(0);
-	  st0_tag = st0_ptr->tag;
+	  st0_tag = FPU_gettag0();
 
 	  /* Stack underflow has priority */
 	  if ( NOT_EMPTY_ST0 )
@@ -342,29 +334,34 @@ do_another_FPU_instruction:
 	      switch ( (byte1 >> 1) & 3 )
 		{
 		case 0:
-		  unmasked = reg_load_single((float *)data_address,
+		  unmasked = FPU_load_single((float *)data_address,
 					     &loaded_data);
+		  loaded_tag = unmasked & 0xff;
+		  unmasked &= ~0xff;
 		  break;
 		case 1:
-		  reg_load_int32((long *)data_address, &loaded_data);
+		  loaded_tag = FPU_load_int32((long *)data_address, &loaded_data);
 		  break;
 		case 2:
-		  unmasked = reg_load_double((double *)data_address,
+		  unmasked = FPU_load_double((double *)data_address,
 					     &loaded_data);
+		  loaded_tag = unmasked & 0xff;
+		  unmasked &= ~0xff;
 		  break;
 		case 3:
-		  reg_load_int16((short *)data_address, &loaded_data);
+		default:  /* Used here to suppress gcc warnings. */
+		  loaded_tag = FPU_load_int16((short *)data_address, &loaded_data);
 		  break;
 		}
-	      
+
 	      /* No more access to user memory, it is safe
 		 to use static data now */
 
 	      /* NaN operands have the next priority. */
 	      /* We have to delay looking at st(0) until after
 		 loading the data, because that data might contain an SNaN */
-	      if ( (st0_tag == TW_NaN) ||
-		  (loaded_data.tag == TW_NaN) )
+	      if ( ((st0_tag == TAG_Special) && isNaN(st0_ptr)) ||
+		  ((loaded_tag == TAG_Special) && isNaN(&loaded_data)) )
 		{
 		  /* Restore the status word; we might have loaded a
 		     denormal. */
@@ -375,22 +372,22 @@ do_another_FPU_instruction:
 		      EXCEPTION(EX_Invalid);
 		      setcc(SW_C3 | SW_C2 | SW_C0);
 		      if ( (FPU_modrm & 0x08) && (control_word & CW_Invalid) )
-			pop();             /* fcomp, masked, so we pop. */
+			FPU_pop();             /* fcomp, masked, so we pop. */
 		    }
 		  else
 		    {
+		      if ( loaded_tag == TAG_Special )
+			loaded_tag = FPU_Special(&loaded_data);
 #ifdef PECULIAR_486
 		      /* This is not really needed, but gives behaviour
 			 identical to an 80486 */
 		      if ( (FPU_modrm & 0x28) == 0x20 )
 			/* fdiv or fsub */
-			real_2op_NaN(&loaded_data, st0_ptr,
-				     st0_ptr);
+			real_2op_NaN(&loaded_data, loaded_tag, 0, &loaded_data);
 		      else
 #endif PECULIAR_486
 			/* fadd, fdivr, fmul, or fsubr */
-			real_2op_NaN(st0_ptr, &loaded_data,
-				     st0_ptr);
+			real_2op_NaN(&loaded_data, loaded_tag, 0, st0_ptr);
 		    }
 		  goto reg_mem_instr_done;
 		}
@@ -401,11 +398,13 @@ do_another_FPU_instruction:
 		  if ( (FPU_modrm & 0x38) == 0x38 )
 		    {
 		      /* fdivr */
-		      if ( (st0_tag == TW_Zero) &&
-			  (loaded_data.tag == TW_Valid) )
+		      if ( (st0_tag == TAG_Zero) &&
+			   ((loaded_tag == TAG_Valid)
+			    || (loaded_tag == TAG_Special
+				&& isdenormal(&loaded_data))) )
 			{
-			  if ( divide_by_zero(loaded_data.sign,
-					      st0_ptr) )
+			  if ( FPU_divide_by_zero(0, getsign(&loaded_data))
+			       < 0 )
 			    {
 			      /* We use the fact here that the unmasked
 				 exception in the loaded data was for a
@@ -414,6 +413,8 @@ do_another_FPU_instruction:
 			      partial_status &= ~SW_Denorm_Op;
 			      partial_status |= status1 & SW_Denorm_Op;
 			    }
+			  else
+			    setsign(st0_ptr, getsign(&loaded_data));
 			}
 		    }
 		  goto reg_mem_instr_done;
@@ -423,43 +424,38 @@ do_another_FPU_instruction:
 		{
 		case 0:         /* fadd */
 		  clear_C1();
-		  reg_add(st0_ptr, &loaded_data, st0_ptr,
-			  control_word);
+		  FPU_add(&loaded_data, loaded_tag, 0, control_word);
 		  break;
 		case 1:         /* fmul */
 		  clear_C1();
-		  reg_mul(st0_ptr, &loaded_data, st0_ptr,
-			  control_word);
+		  FPU_mul(&loaded_data, loaded_tag, 0, control_word);
 		  break;
 		case 2:         /* fcom */
-		  compare_st_data(&loaded_data);
+		  FPU_compare_st_data(&loaded_data, loaded_tag);
 		  break;
 		case 3:         /* fcomp */
-		  if ( !compare_st_data(&loaded_data) && !unmasked )
-		    pop();
+		  if ( !FPU_compare_st_data(&loaded_data, loaded_tag)
+		       && !unmasked )
+		    FPU_pop();
 		  break;
 		case 4:         /* fsub */
 		  clear_C1();
-		  reg_sub(st0_ptr, &loaded_data, st0_ptr,
-			  control_word);
+		  FPU_sub(LOADED|loaded_tag, (int)&loaded_data, control_word);
 		  break;
 		case 5:         /* fsubr */
 		  clear_C1();
-		  reg_sub(&loaded_data, st0_ptr, st0_ptr,
-			  control_word);
+		  FPU_sub(REV|LOADED|loaded_tag, (int)&loaded_data, control_word);
 		  break;
 		case 6:         /* fdiv */
 		  clear_C1();
-		  reg_div(st0_ptr, &loaded_data, st0_ptr,
-			  control_word);
+		  FPU_div(LOADED|loaded_tag, (int)&loaded_data, control_word);
 		  break;
 		case 7:         /* fdivr */
 		  clear_C1();
-		  if ( st0_tag == TW_Zero )
+		  if ( st0_tag == TAG_Zero )
 		    partial_status = status1;  /* Undo any denorm tag,
-					       zero-divide has priority. */
-		  reg_div(&loaded_data, st0_ptr, st0_ptr,
-			  control_word);
+						  zero-divide has priority. */
+		  FPU_div(REV|LOADED|loaded_tag, (int)&loaded_data, control_word);
 		  break;
 		}
 	    }
@@ -471,10 +467,10 @@ do_another_FPU_instruction:
 		  EXCEPTION(EX_StackUnder);
 		  setcc(SW_C3 | SW_C2 | SW_C0);
 		  if ( (FPU_modrm & 0x08) && (control_word & CW_Invalid) )
-		    pop();             /* fcomp */
+		    FPU_pop();             /* fcomp */
 		}
 	      else
-		stack_underflow();
+		FPU_stack_underflow();
 	    }
 	reg_mem_instr_done:
 	  operand_address = data_sel_off;
@@ -482,8 +478,8 @@ do_another_FPU_instruction:
       else
 	{
 	  if ( !(no_ip_update =
-		 load_store_instr(((FPU_modrm & 0x38) | (byte1 & 6)) >> 1,
-				  addr_modes, data_address)) )
+		 FPU_load_store(((FPU_modrm & 0x38) | (byte1 & 6)) >> 1,
+				addr_modes, data_address)) )
 	    {
 	      operand_address = data_sel_off;
 	    }
@@ -493,7 +489,7 @@ do_another_FPU_instruction:
   else
     {
       /* None of these instructions access user memory */
-      unsigned char instr_index = (FPU_modrm & 0x38) | (byte1 & 7);
+      u_char instr_index = (FPU_modrm & 0x38) | (byte1 & 7);
 
 #ifdef PECULIAR_486
       /* This is supposed to be undefined, but a real 80486 seems
@@ -503,7 +499,7 @@ do_another_FPU_instruction:
 #endif PECULIAR_486
 
       st0_ptr = &st(0);
-      st0_tag = st0_ptr->tag;
+      st0_tag = FPU_gettag0();
       switch ( type_table[(int) instr_index] )
 	{
 	case _NONE_:   /* also _REGIc: _REGIn */
@@ -511,28 +507,28 @@ do_another_FPU_instruction:
 	case _REG0_:
 	  if ( !NOT_EMPTY_ST0 )
 	    {
-	      stack_underflow();
+	      FPU_stack_underflow();
 	      goto FPU_instruction_done;
 	    }
 	  break;
 	case _REGIi:
 	  if ( !NOT_EMPTY_ST0 || !NOT_EMPTY(FPU_rm) )
 	    {
-	      stack_underflow_i(FPU_rm);
+	      FPU_stack_underflow_i(FPU_rm);
 	      goto FPU_instruction_done;
 	    }
 	  break;
 	case _REGIp:
 	  if ( !NOT_EMPTY_ST0 || !NOT_EMPTY(FPU_rm) )
 	    {
-	      stack_underflow_pop(FPU_rm);
+	      FPU_stack_underflow_pop(FPU_rm);
 	      goto FPU_instruction_done;
 	    }
 	  break;
 	case _REGI_:
 	  if ( !NOT_EMPTY_ST0 || !NOT_EMPTY(FPU_rm) )
 	    {
-	      stack_underflow();
+	      FPU_stack_underflow();
 	      goto FPU_instruction_done;
 	    }
 	  break;
@@ -558,14 +554,14 @@ FPU_fwait_done:
 
 #ifdef DEBUG
   RE_ENTRANT_CHECK_OFF;
-  emu_printall();
+  FPU_printall();
   RE_ENTRANT_CHECK_ON;
 #endif DEBUG
 
   if (FPU_lookahead && !need_resched)
     {
       FPU_ORIG_EIP = FPU_EIP - code_base;
-      if ( valid_prefix(&byte1, (unsigned char **)&FPU_EIP,
+      if ( valid_prefix(&byte1, (u_char **)&FPU_EIP,
 			&addr_modes.override) )
 	goto do_another_FPU_instruction;
     }
@@ -581,17 +577,17 @@ FPU_fwait_done:
    all prefix bytes, further changes are needed in the emulator code
    which accesses user address space. Access to separate segments is
    important for msdos emulation. */
-static int valid_prefix(unsigned char *Byte, unsigned char **fpu_eip,
+static int valid_prefix(u_char *Byte, u_char **fpu_eip,
 			overrides *override)
 {
-  unsigned char byte;
-  unsigned char *ip = *fpu_eip;
+  u_char byte;
+  u_char *ip = *fpu_eip;
 
   *override = (overrides) { 0, 0, PREFIX_DEFAULT };       /* defaults */
 
   RE_ENTRANT_CHECK_OFF;
   FPU_code_verify_area(1);
-  get_user(byte, ip);
+  FPU_get_user(byte, ip);
   RE_ENTRANT_CHECK_ON;
 
   while ( 1 )
@@ -637,7 +633,7 @@ static int valid_prefix(unsigned char *Byte, unsigned char **fpu_eip,
 	  ip++;
 	  RE_ENTRANT_CHECK_OFF;
 	  FPU_code_verify_area(1);
-	  get_user(byte, ip);
+	  FPU_get_user(byte, ip);
 	  RE_ENTRANT_CHECK_ON;
 	  break;
 	case FWAIT_OPCODE:
@@ -677,19 +673,79 @@ void math_abort(struct info * info, unsigned int signal)
 
 
 
-void restore_i387_soft(struct _fpstate *buf)
+#define S387 ((struct i387_soft_struct *)s387)
+#define sstatus_word() \
+  ((S387->swd & ~SW_Top & 0xffff) | ((S387->ftop << SW_Top_Shift) & SW_Top))
+
+void restore_i387_soft(void *s387, struct _fpstate *buf)
 {
-  fpu_addr_modes addr_modes = {{ 0, 0, PREFIX_DEFAULT }, 0};
+  u_char *d = (u_char *)buf;
+  int offset, other, i, tags, regnr, tag, newtop;
+
+  RE_ENTRANT_CHECK_OFF;
+  FPU_verify_area(VERIFY_READ, d, 7*4 + 8*10);
+  __copy_from_user(&S387->cwd, d, 7*4);
+  RE_ENTRANT_CHECK_ON;
+
+  d += 7*4;
+
+  S387->ftop = (S387->swd >> SW_Top_Shift) & 7;
+  offset = (S387->ftop & 7) * 10;
+  other = 80 - offset;
+
+  RE_ENTRANT_CHECK_OFF;
+  /* Copy all registers in stack order. */
+  __copy_from_user(((u_char *)&S387->st_space)+offset, d, other);
+  if ( offset )
+    __copy_from_user((u_char *)&S387->st_space, d+other, offset);
+  RE_ENTRANT_CHECK_ON;
+
+  /* The tags may need to be corrected now. */
+  tags = S387->twd;
+  newtop = S387->ftop;
+  for ( i = 0; i < 8; i++ )
+    {
+      regnr = (i+newtop) & 7;
+      if ( ((tags >> ((regnr & 7)*2)) & 3) != TAG_Empty )
+	{
+	  /* The loaded data over-rides all other cases. */
+	  tag = FPU_tagof((FPU_REG *)((u_char *)S387->st_space + 10*regnr));
+	  tags &= ~(3 << (regnr*2));
+	  tags |= (tag & 3) << (regnr*2);
+	}
+    }
+  S387->twd = tags;
 
-  frstor(addr_modes, (char *)buf);
 }
 
 
-struct _fpstate * save_i387_soft(struct _fpstate * buf)
+struct _fpstate * save_i387_soft(void *s387, struct _fpstate * buf)
 {
-  fpu_addr_modes addr_modes = {{ 0, 0, PREFIX_DEFAULT }, 0};
+  u_char *d = (u_char *)buf;
+  int offset = (S387->ftop & 7) * 10, other = 80 - offset;
 
-  fsave(addr_modes, (char *)buf);
+  RE_ENTRANT_CHECK_OFF;
+  FPU_verify_area(VERIFY_WRITE, d, 7*4 + 8*10);
+#ifdef PECULIAR_486
+  S387->cwd &= ~0xe080;
+  /* An 80486 sets all the reserved bits to 1. */
+  S387->cwd |= 0xffff0000;
+  S387->swd = sstatus_word() | 0xffff0000;
+  S387->twd |= 0xffff0000;
+  S387->fcs |= 0xf8000000;
+  S387->fos |= 0xffff0000;
+#endif PECULIAR_486
+  __copy_to_user(d, &S387->cwd, 7*4);
+  RE_ENTRANT_CHECK_ON;
+
+  d += 7*4;
+
+  RE_ENTRANT_CHECK_OFF;
+  /* Copy all registers in stack order. */
+  __copy_to_user(d, ((u_char *)&S387->st_space)+offset, other);
+  if ( offset )
+    __copy_to_user(d+other, (u_char *)&S387->st_space, offset);
+  RE_ENTRANT_CHECK_ON;
 
   return buf;
 }
diff --git a/arch/i386/math-emu/fpu_etc.c b/arch/i386/math-emu/fpu_etc.c
index 20e3294ca..64c5a70cd 100644
--- a/arch/i386/math-emu/fpu_etc.c
+++ b/arch/i386/math-emu/fpu_etc.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Implement a few FPU instructions.                                         |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
+ | Copyright (C) 1992,1993,1994,1997                                         |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ |                       Australia.  E-mail   billm@suburbia.net             |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -17,102 +17,116 @@
 #include "reg_constant.h"
 
 
-static void fchs(FPU_REG *st0_ptr)
+static void fchs(FPU_REG *st0_ptr, u_char st0tag)
 {
-  if ( st0_ptr->tag ^ TW_Empty )
+  if ( st0tag ^ TAG_Empty )
     {
-      st0_ptr->sign ^= SIGN_POS^SIGN_NEG;
+      signbyte(st0_ptr) ^= SIGN_NEG;
       clear_C1();
     }
   else
-    stack_underflow();
+    FPU_stack_underflow();
 }
 
-static void fabs(FPU_REG *st0_ptr)
+
+static void fabs(FPU_REG *st0_ptr, u_char st0tag)
 {
-  if ( st0_ptr->tag ^ TW_Empty )
+  if ( st0tag ^ TAG_Empty )
     {
-      st0_ptr->sign = SIGN_POS;
+      setpositive(st0_ptr);
       clear_C1();
     }
   else
-    stack_underflow();
+    FPU_stack_underflow();
 }
 
 
-static void ftst_(FPU_REG *st0_ptr)
+static void ftst_(FPU_REG *st0_ptr, u_char st0tag)
 {
-  switch (st0_ptr->tag)
+  switch (st0tag)
     {
-    case TW_Zero:
+    case TAG_Zero:
       setcc(SW_C3);
       break;
-    case TW_Valid:
-      if (st0_ptr->sign == SIGN_POS)
+    case TAG_Valid:
+      if (getsign(st0_ptr) == SIGN_POS)
         setcc(0);
       else
         setcc(SW_C0);
-
-#ifdef DENORM_OPERAND
-      if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+      break;
+    case TAG_Special:
+      switch ( FPU_Special(st0_ptr) )
 	{
+	case TW_Denormal:
+	  if (getsign(st0_ptr) == SIGN_POS)
+	    setcc(0);
+	  else
+	    setcc(SW_C0);
+	  if ( denormal_operand() < 0 )
+	    {
 #ifdef PECULIAR_486
-	  /* This is weird! */
-	  if (st0_ptr->sign == SIGN_POS)
-	    setcc(SW_C3);
+	      /* This is weird! */
+	      if (getsign(st0_ptr) == SIGN_POS)
+		setcc(SW_C3);
 #endif PECULIAR_486
-	  return;
+	      return;
+	    }
+	  break;
+	case TW_NaN:
+	  setcc(SW_C0|SW_C2|SW_C3);   /* Operand is not comparable */ 
+	  EXCEPTION(EX_Invalid);
+	  break;
+	case TW_Infinity:
+	  if (getsign(st0_ptr) == SIGN_POS)
+	    setcc(0);
+	  else
+	    setcc(SW_C0);
+	  break;
+	default:
+	  setcc(SW_C0|SW_C2|SW_C3);   /* Operand is not comparable */ 
+	  EXCEPTION(EX_INTERNAL|0x14);
+	  break;
 	}
-#endif DENORM_OPERAND
-
-      break;
-    case TW_NaN:
-      setcc(SW_C0|SW_C2|SW_C3);   /* Operand is not comparable */ 
-      EXCEPTION(EX_Invalid);
-      break;
-    case TW_Infinity:
-      if (st0_ptr->sign == SIGN_POS)
-        setcc(0);
-      else
-        setcc(SW_C0);
       break;
-    case TW_Empty:
+    case TAG_Empty:
       setcc(SW_C0|SW_C2|SW_C3);
       EXCEPTION(EX_StackUnder);
       break;
-    default:
-      setcc(SW_C0|SW_C2|SW_C3);   /* Operand is not comparable */ 
-      EXCEPTION(EX_INTERNAL|0x14);
-      break;
     }
 }
 
-static void fxam(FPU_REG *st0_ptr)
+
+static void fxam(FPU_REG *st0_ptr, u_char st0tag)
 {
-  int c=0;
-  switch (st0_ptr->tag)
+  int c = 0;
+  switch (st0tag)
     {
-    case TW_Empty:
+    case TAG_Empty:
       c = SW_C3|SW_C0;
       break;
-    case TW_Zero:
+    case TAG_Zero:
       c = SW_C3;
       break;
-    case TW_Valid:
-      /* This will need to be changed if TW_Denormal is ever used. */
-      if ( st0_ptr->exp <= EXP_UNDER )
-        c = SW_C2|SW_C3;  /* Denormal */
-      else
-        c = SW_C2;
-      break;
-    case TW_NaN:
-      c = SW_C0;
-      break;
-    case TW_Infinity:
-      c = SW_C2|SW_C0;
+    case TAG_Valid:
+      c = SW_C2;
       break;
+    case TAG_Special:
+      switch ( FPU_Special(st0_ptr) )
+	{
+	case TW_Denormal:
+	  c = SW_C2|SW_C3;  /* Denormal */
+	  break;
+	case TW_NaN:
+	  /* We also use NaN for unsupported types. */
+	  if ( (st0_ptr->sigh & 0x80000000) && (exponent(st0_ptr) == EXP_OVER) )
+	    c = SW_C0;
+	  break;
+	case TW_Infinity:
+	  c = SW_C2|SW_C0;
+	  break;
+	}
     }
-  if (st0_ptr->sign == SIGN_NEG)
+  if ( getsign(st0_ptr) == SIGN_NEG )
     c |= SW_C1;
   setcc(c);
 }
@@ -123,7 +137,7 @@ static FUNC_ST0 const fp_etc_table[] = {
   ftst_, fxam, (FUNC_ST0)FPU_illegal, (FUNC_ST0)FPU_illegal
 };
 
-void fp_etc()
+void FPU_etc()
 {
-  (fp_etc_table[FPU_rm])(&st(0));
+  (fp_etc_table[FPU_rm])(&st(0), FPU_gettag0());
 }
diff --git a/arch/i386/math-emu/fpu_proto.h b/arch/i386/math-emu/fpu_proto.h
index d0e58ed0e..4fccf05fe 100644
--- a/arch/i386/math-emu/fpu_proto.h
+++ b/arch/i386/math-emu/fpu_proto.h
@@ -1,22 +1,26 @@
+#ifndef _FPU_PROTO_H
+#define _FPU_PROTO_H
+
 /* errors.c */
 extern void Un_impl(void);
 extern void FPU_illegal(void);
-extern void emu_printall(void);
-extern void stack_overflow(void);
-extern void stack_underflow(void);
-extern void stack_underflow_i(int i);
-extern void stack_underflow_pop(int i);
-extern int set_precision_flag(int flags);
+extern void FPU_printall(void);
 asmlinkage void FPU_exception(int n);
-asmlinkage int real_2op_NaN(FPU_REG const *a, FPU_REG const *b, FPU_REG *dest);
-asmlinkage int arith_invalid(FPU_REG *dest);
-asmlinkage int divide_by_zero(int sign, FPU_REG *dest);
-asmlinkage void set_precision_flag_up(void);
-asmlinkage void set_precision_flag_down(void);
-asmlinkage int denormal_operand(void);
-asmlinkage int arith_overflow(FPU_REG *dest);
-asmlinkage int arith_underflow(FPU_REG *dest);
-
+extern int real_1op_NaN(FPU_REG *a);
+extern int real_2op_NaN(FPU_REG const *b, u_char tagb, int deststnr,
+			FPU_REG const *defaultNaN);
+extern int arith_invalid(int deststnr);
+extern int FPU_divide_by_zero(int deststnr, u_char sign);
+extern int set_precision_flag(int flags);
+extern void set_precision_flag_up(void);
+extern void set_precision_flag_down(void);
+extern int denormal_operand(void);
+extern int arith_overflow(FPU_REG *dest);
+extern int arith_underflow(FPU_REG *dest);
+extern void FPU_stack_overflow(void);
+extern void FPU_stack_underflow(void);
+extern void FPU_stack_underflow_i(int i);
+extern void FPU_stack_underflow_pop(int i);
 /* fpu_arith.c */
 extern void fadd__(void);
 extern void fmul__(void);
@@ -36,7 +40,6 @@ extern void fsubrp(void);
 extern void fsubp_(void);
 extern void fdivrp(void);
 extern void fdivp_(void);
-
 /* fpu_aux.c */
 extern void fclex(void);
 extern void finit(void);
@@ -49,89 +52,92 @@ extern void ffree_(void);
 extern void ffreep(void);
 extern void fst_i_(void);
 extern void fstp_i(void);
-
 /* fpu_entry.c */
-asmlinkage void math_emulate(long arg);
+extern void math_emulate(long arg);
 extern void math_abort(struct info *info, unsigned int signal);
-
 /* fpu_etc.c */
-extern void fp_etc(void);
-
+extern void FPU_etc(void);
+/* fpu_tags.c */
+extern int FPU_gettag0(void);
+extern int FPU_gettagi(int stnr);
+extern int FPU_gettag(int regnr);
+extern void FPU_settag0(int tag);
+extern void FPU_settagi(int stnr, int tag);
+extern void FPU_settag(int regnr, int tag);
+extern int FPU_Special(FPU_REG const *ptr);
+extern int isNaN(FPU_REG const *ptr);
+extern void FPU_pop(void);
+extern int FPU_empty_i(int stnr);
+extern int FPU_stackoverflow(FPU_REG **st_new_ptr);
+extern void FPU_sync_tags(void);
+extern void FPU_copy_to_regi(FPU_REG const *r, u_char tag, int stnr);
+extern void FPU_copy_to_reg1(FPU_REG const *r, u_char tag);
+extern void FPU_copy_to_reg0(FPU_REG const *r, u_char tag);
 /* fpu_trig.c */
-extern void convert_l2reg(long const *arg, FPU_REG *dest);
-extern void trig_a(void);
-extern void trig_b(void);
-
+extern void FPU_triga(void);
+extern void FPU_trigb(void);
 /* get_address.c */
-extern void *get_address(unsigned char FPU_modrm, unsigned long *fpu_eip,
-			 struct address *addr,
-			 fpu_addr_modes);
-extern void *get_address_16(unsigned char FPU_modrm, unsigned long *fpu_eip,
-			    struct address *addr,
-			    fpu_addr_modes);
-
+extern void *FPU_get_address(u_char FPU_modrm, unsigned long *fpu_eip,
+			 struct address *addr, fpu_addr_modes addr_modes);
+extern void *FPU_get_address_16(u_char FPU_modrm, unsigned long *fpu_eip,
+			    struct address *addr, fpu_addr_modes addr_modes);
 /* load_store.c */
-extern int load_store_instr(unsigned char type, fpu_addr_modes addr_modes,
-			     void *address);
-
+extern int FPU_load_store(u_char type, fpu_addr_modes addr_modes,
+			    void *data_address);
 /* poly_2xm1.c */
-extern int poly_2xm1(FPU_REG const *arg, FPU_REG *result);
-
+extern int poly_2xm1(u_char sign, FPU_REG *arg, FPU_REG *result);
 /* poly_atan.c */
-extern void poly_atan(FPU_REG *arg1, FPU_REG *arg2, FPU_REG *result);
-
+extern void poly_atan(FPU_REG *st0_ptr, u_char st0_tag, FPU_REG *st1_ptr,
+		      u_char st1_tag);
 /* poly_l2.c */
-extern void poly_l2(FPU_REG const *arg, FPU_REG const *y, FPU_REG *result);
-extern int poly_l2p1(FPU_REG const *arg, FPU_REG const *y, FPU_REG *result);
-
+extern void poly_l2(FPU_REG *st0_ptr, FPU_REG *st1_ptr, u_char st1_sign);
+extern int poly_l2p1(u_char s0, u_char s1, FPU_REG *r0, FPU_REG *r1,
+		     FPU_REG *d);
 /* poly_sin.c */
-extern void poly_sine(FPU_REG const *arg, FPU_REG *result);
-extern void poly_cos(FPU_REG const *arg, FPU_REG *result);
-
+extern void poly_sine(FPU_REG *st0_ptr);
+extern void poly_cos(FPU_REG *st0_ptr);
 /* poly_tan.c */
-extern void poly_tan(FPU_REG const *arg, FPU_REG *result);
-
+extern void poly_tan(FPU_REG *st0_ptr);
 /* reg_add_sub.c */
-extern int reg_add(FPU_REG const *a, FPU_REG const *b,
-		   FPU_REG *dest, int control_w);
-extern int reg_sub(FPU_REG const *a, FPU_REG const *b,
-		   FPU_REG *dest, int control_w);
-
+extern int FPU_add(FPU_REG const *b, u_char tagb, int destrnr, int control_w);
+extern int FPU_sub(int flags, int rm, int control_w);
 /* reg_compare.c */
-extern int compare(FPU_REG const *b);
-extern int compare_st_data(FPU_REG const *b);
+extern int FPU_compare_st_data(FPU_REG const *loaded_data, u_char loaded_tag);
 extern void fcom_st(void);
 extern void fcompst(void);
 extern void fcompp(void);
 extern void fucom_(void);
 extern void fucomp(void);
 extern void fucompp(void);
-
 /* reg_constant.c */
 extern void fconst(void);
-
 /* reg_ld_str.c */
-extern int reg_load_extended(long double *addr, FPU_REG *loaded_data);
-extern int reg_load_double(double *dfloat, FPU_REG *loaded_data);
-extern int reg_load_single(float *single, FPU_REG *loaded_data);
-extern void reg_load_int64(long long *_s, FPU_REG *loaded_data);
-extern void reg_load_int32(long *_s, FPU_REG *loaded_data);
-extern void reg_load_int16(short *_s, FPU_REG *loaded_data);
-extern void reg_load_bcd(char *s, FPU_REG *loaded_data);
-extern int reg_store_extended(long double *d, FPU_REG *st0_ptr);
-extern int reg_store_double(double *dfloat, FPU_REG *st0_ptr);
-extern int reg_store_single(float *single, FPU_REG *st0_ptr);
-extern int reg_store_int64(long long *d, FPU_REG *st0_ptr);
-extern int reg_store_int32(long *d, FPU_REG *st0_ptr);
-extern int reg_store_int16(short *d, FPU_REG *st0_ptr);
-extern int reg_store_bcd(char *d, FPU_REG *st0_ptr);
-extern int round_to_int(FPU_REG *r);
-extern char *fldenv(fpu_addr_modes addr_modes, char *address);
-extern void frstor(fpu_addr_modes addr_modes, char *address);
-extern unsigned short tag_word(void);
-extern char *fstenv(fpu_addr_modes addr_modes, char *address);
-extern void fsave(fpu_addr_modes addr_modes, char *address);
-
+extern int FPU_load_extended(long double *s, int stnr);
+extern int FPU_load_double(double *dfloat, FPU_REG *loaded_data);
+extern int FPU_load_single(float *single, FPU_REG *loaded_data);
+extern int FPU_load_int64(long long *_s);
+extern int FPU_load_int32(long *_s, FPU_REG *loaded_data);
+extern int FPU_load_int16(short *_s, FPU_REG *loaded_data);
+extern int FPU_load_bcd(u_char *s);
+extern int FPU_store_extended(FPU_REG *st0_ptr, u_char st0_tag,
+			      long double *d);
+extern int FPU_store_double(FPU_REG *st0_ptr, u_char st0_tag, double *dfloat);
+extern int FPU_store_single(FPU_REG *st0_ptr, u_char st0_tag, float *single);
+extern int FPU_store_int64(FPU_REG *st0_ptr, u_char st0_tag, long long *d);
+extern int FPU_store_int32(FPU_REG *st0_ptr, u_char st0_tag, long *d);
+extern int FPU_store_int16(FPU_REG *st0_ptr, u_char st0_tag, short *d);
+extern int FPU_store_bcd(FPU_REG *st0_ptr, u_char st0_tag, u_char *d);
+extern int FPU_round_to_int(FPU_REG *r, u_char tag);
+extern u_char *fldenv(fpu_addr_modes addr_modes, u_char *s);
+extern void frstor(fpu_addr_modes addr_modes, u_char *data_address);
+extern u_char *fstenv(fpu_addr_modes addr_modes, u_char *d);
+extern void fsave(fpu_addr_modes addr_modes, u_char *data_address);
+extern int FPU_tagof(FPU_REG *ptr);
 /* reg_mul.c */
-extern int reg_mul(FPU_REG const *a, FPU_REG const *b,
-		   FPU_REG *dest, unsigned int control_w);
+extern int FPU_mul(FPU_REG const *b, u_char tagb, int deststnr, int control_w);
+
+extern int FPU_div(int flags, int regrm, int control_w);
+/* reg_convert.c */
+extern int FPU_to_exp16(FPU_REG const *a, FPU_REG *x);
+#endif /* _FPU_PROTO_H */
+
diff --git a/arch/i386/math-emu/fpu_system.h b/arch/i386/math-emu/fpu_system.h
index d2c3fa716..42303f679 100644
--- a/arch/i386/math-emu/fpu_system.h
+++ b/arch/i386/math-emu/fpu_system.h
@@ -1,9 +1,9 @@
 /*---------------------------------------------------------------------------+
  |  fpu_system.h                                                             |
  |                                                                           |
- | Copyright (C) 1992,1994                                                   |
+ | Copyright (C) 1992,1994,1997                                              |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ |                       Australia.  E-mail   billm@suburbia.net             |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
@@ -18,19 +18,19 @@
 
 /* This sets the pointer FPU_info to point to the argument part
    of the stack frame of math_emulate() */
-#define SETUP_DATA_AREA(arg)    FPU_info = (struct info *) &arg
-
-#define LDT_DESCRIPTOR(s)       (current->ldt[(s) >> 3])
-#define SEG_D_SIZE(x)           ((x).b & (3 << 21))
-#define SEG_G_BIT(x)            ((x).b & (1 << 23))
-#define SEG_GRANULARITY(x)      (((x).b & (1 << 23)) ? 4096 : 1)
-#define SEG_286_MODE(x)         ((x).b & ( 0xff000000 | 0xf0000 | (1 << 23)))
-#define SEG_BASE_ADDR(s)        (((s).b & 0xff000000) \
+#define SETUP_DATA_AREA(arg)	FPU_info = (struct info *) &arg
+
+#define LDT_DESCRIPTOR(s)	(current->ldt[(s) >> 3])
+#define SEG_D_SIZE(x)		((x).b & (3 << 21))
+#define SEG_G_BIT(x)		((x).b & (1 << 23))
+#define SEG_GRANULARITY(x)	(((x).b & (1 << 23)) ? 4096 : 1)
+#define SEG_286_MODE(x)		((x).b & ( 0xff000000 | 0xf0000 | (1 << 23)))
+#define SEG_BASE_ADDR(s)	(((s).b & 0xff000000) \
 				 | (((s).b & 0xff) << 16) | ((s).a >> 16))
-#define SEG_LIMIT(s)            (((s).b & 0xff0000) | ((s).a & 0xffff))
-#define SEG_EXECUTE_ONLY(s)     (((s).b & ((1 << 11) | (1 << 9))) == (1 << 11))
-#define SEG_WRITE_PERM(s)       (((s).b & ((1 << 11) | (1 << 9))) == (1 << 9))
-#define SEG_EXPAND_DOWN(s)      (((s).b & ((1 << 11) | (1 << 10))) \
+#define SEG_LIMIT(s)		(((s).b & 0xff0000) | ((s).a & 0xffff))
+#define SEG_EXECUTE_ONLY(s)	(((s).b & ((1 << 11) | (1 << 9))) == (1 << 11))
+#define SEG_WRITE_PERM(s)	(((s).b & ((1 << 11) | (1 << 9))) == (1 << 9))
+#define SEG_EXPAND_DOWN(s)	(((s).b & ((1 << 11) | (1 << 10))) \
 				 == (1 << 10))
 
 #define I387			(current->tss.i387)
@@ -48,23 +48,24 @@
 
 /* nz if ip_offset and cs_selector are not to be set for the current
    instruction. */
-#define no_ip_update            (((char *)&(I387.soft.twd))[0])
-#define FPU_rm                  (((unsigned char *)&(I387.soft.twd))[1])
+#define no_ip_update		(*(u_char *)&(I387.soft.no_update))
+#define FPU_rm			(*(u_char *)&(I387.soft.rm))
 
 /* Number of bytes of data which can be legally accessed by the current
    instruction. This only needs to hold a number <= 108, so a byte will do. */
-#define access_limit            (((unsigned char *)&(I387.soft.twd))[2])
+#define access_limit		(*(u_char *)&(I387.soft.alimit))
 
-#define partial_status       	(I387.soft.swd)
+#define partial_status		(I387.soft.swd)
 #define control_word		(I387.soft.cwd)
-#define regs			(I387.soft.regs)
-#define top			(I387.soft.top)
+#define fpu_tag_word		(I387.soft.twd)
+#define registers		(I387.soft.st_space)
+#define top			(I387.soft.ftop)
 
-#define instruction_address     (*(struct address *)&I387.soft.fip)
-#define operand_address         (*(struct address *)&I387.soft.foo)
+#define instruction_address	(*(struct address *)&I387.soft.fip)
+#define operand_address		(*(struct address *)&I387.soft.foo)
 
-#define FPU_verify_area(x,y,z)  if ( verify_area(x,y,z) ) \
-                                math_abort(FPU_info,SIGSEGV)
+#define FPU_verify_area(x,y,z)	if ( verify_area(x,y,z) ) \
+				math_abort(FPU_info,SIGSEGV)
 
 #undef FPU_IGNORE_CODE_SEGV
 #ifdef FPU_IGNORE_CODE_SEGV
@@ -80,4 +81,7 @@
 #define	FPU_code_verify_area(z) FPU_verify_area(VERIFY_READ,(void *)FPU_EIP,z)
 #endif
 
+#define FPU_get_user(x,y)       get_user((x),(y))
+#define FPU_put_user(x,y)       put_user((x),(y))
+
 #endif
diff --git a/arch/i386/math-emu/fpu_tags.c b/arch/i386/math-emu/fpu_tags.c
new file mode 100644
index 000000000..cb436fe20
--- /dev/null
+++ b/arch/i386/math-emu/fpu_tags.c
@@ -0,0 +1,127 @@
+/*---------------------------------------------------------------------------+
+ |  fpu_tags.c                                                               |
+ |                                                                           |
+ |  Set FPU register tags.                                                   |
+ |                                                                           |
+ | Copyright (C) 1997                                                        |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@jacobi.maths.monash.edu.au                |
+ |                                                                           |
+ |                                                                           |
+ +---------------------------------------------------------------------------*/
+
+#include "fpu_emu.h"
+#include "fpu_system.h"
+#include "exception.h"
+
+
+void FPU_pop(void)
+{
+  fpu_tag_word |= 3 << ((top & 7)*2);
+  top++;
+}
+
+
+int FPU_gettag0(void)
+{
+  return (fpu_tag_word >> ((top & 7)*2)) & 3;
+}
+
+
+int FPU_gettagi(int stnr)
+{
+  return (fpu_tag_word >> (((top+stnr) & 7)*2)) & 3;
+}
+
+
+int FPU_gettag(int regnr)
+{
+  return (fpu_tag_word >> ((regnr & 7)*2)) & 3;
+}
+
+
+void FPU_settag0(int tag)
+{
+  int regnr = top;
+  regnr &= 7;
+  fpu_tag_word &= ~(3 << (regnr*2));
+  fpu_tag_word |= (tag & 3) << (regnr*2);
+}
+
+
+void FPU_settagi(int stnr, int tag)
+{
+  int regnr = stnr+top;
+  regnr &= 7;
+  fpu_tag_word &= ~(3 << (regnr*2));
+  fpu_tag_word |= (tag & 3) << (regnr*2);
+}
+
+
+void FPU_settag(int regnr, int tag)
+{
+  regnr &= 7;
+  fpu_tag_word &= ~(3 << (regnr*2));
+  fpu_tag_word |= (tag & 3) << (regnr*2);
+}
+
+
+int FPU_Special(FPU_REG const *ptr)
+{
+  int exp = exponent(ptr);
+
+  if ( exp == EXP_BIAS+EXP_UNDER )
+    return TW_Denormal;
+  else if ( exp != EXP_BIAS+EXP_OVER )
+    return TW_NaN;
+  else if ( (ptr->sigh == 0x80000000) && (ptr->sigl == 0) )
+    return TW_Infinity;
+  return TW_NaN;
+}
+
+
+int isNaN(FPU_REG const *ptr)
+{
+  return ( (exponent(ptr) == EXP_BIAS+EXP_OVER)
+	   && !((ptr->sigh == 0x80000000) && (ptr->sigl == 0)) );
+}
+
+
+int FPU_empty_i(int stnr)
+{
+  int regnr = (top+stnr) & 7;
+
+  return ((fpu_tag_word >> (regnr*2)) & 3) == TAG_Empty;
+}
+
+
+int FPU_stackoverflow(FPU_REG **st_new_ptr)
+{
+  *st_new_ptr = &st(-1);
+
+  return ((fpu_tag_word >> (((top - 1) & 7)*2)) & 3) != TAG_Empty;
+}
+
+
+void FPU_copy_to_regi(FPU_REG const *r, u_char tag, int stnr)
+{
+  reg_copy(r, &st(stnr));
+  FPU_settagi(stnr, tag);
+}
+
+void FPU_copy_to_reg1(FPU_REG const *r, u_char tag)
+{
+  reg_copy(r, &st(1));
+  FPU_settagi(1, tag);
+}
+
+void FPU_copy_to_reg0(FPU_REG const *r, u_char tag)
+{
+  int regnr = top;
+  regnr &= 7;
+
+  reg_copy(r, &st(0));
+
+  fpu_tag_word &= ~(3 << (regnr*2));
+  fpu_tag_word |= (tag & 3) << (regnr*2);
+}
diff --git a/arch/i386/math-emu/fpu_trig.c b/arch/i386/math-emu/fpu_trig.c
index 05241f700..57a902162 100644
--- a/arch/i386/math-emu/fpu_trig.c
+++ b/arch/i386/math-emu/fpu_trig.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Implementation of the FPU "transcendental" functions.                     |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
+ | Copyright (C) 1992,1993,1994,1997                                         |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ |                       Australia.  E-mail   billm@suburbia.net             |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -17,7 +17,6 @@
 #include "control_w.h"
 #include "reg_constant.h"	
 
-
 static void rem_kernel(unsigned long long st0, unsigned long long *y,
 		       unsigned long long st1,
 		       unsigned long long q, int n);
@@ -25,9 +24,6 @@ static void rem_kernel(unsigned long long st0, unsigned long long *y,
 #define BETTER_THAN_486
 
 #define FCOS  4
-/* Not needed now with new code
-#define FPTAN 1
- */
 
 /* Used only by fptan, fsin, fcos, and fsincos. */
 /* This routine produces very accurate results, similar to
@@ -35,13 +31,15 @@ static void rem_kernel(unsigned long long st0, unsigned long long *y,
 /* Limited measurements show no results worse than 64 bit precision
    except for the results for arguments close to 2^63, where the
    precision of the result sometimes degrades to about 63.9 bits */
-static int trig_arg(FPU_REG *X, int even)
+static int trig_arg(FPU_REG *st0_ptr, int even)
 {
   FPU_REG tmp;
+  u_char tmptag;
   unsigned long long q;
   int old_cw = control_word, saved_status = partial_status;
+  int tag, st0_tag = TAG_Valid;
 
-  if ( X->exp >= EXP_BIAS + 63 )
+  if ( exponent(st0_ptr) >= 63 )
     {
       partial_status |= SW_C2;     /* Reduction incomplete. */
       return -1;
@@ -50,58 +48,52 @@ static int trig_arg(FPU_REG *X, int even)
   control_word &= ~CW_RC;
   control_word |= RC_CHOP;
 
-  reg_div(X, &CONST_PI2, &tmp, PR_64_BITS | RC_CHOP | 0x3f);
-  round_to_int(&tmp);  /* Fortunately, this can't overflow
-			  to 2^64 */
+  setpositive(st0_ptr);
+  tag = FPU_u_div(st0_ptr, &CONST_PI2, &tmp, PR_64_BITS | RC_CHOP | 0x3f,
+		  SIGN_POS);
+
+  FPU_round_to_int(&tmp, tag);  /* Fortunately, this can't overflow
+				   to 2^64 */
   q = significand(&tmp);
   if ( q )
     {
-      rem_kernel(significand(X),
+      rem_kernel(significand(st0_ptr),
 		 &significand(&tmp),
 		 significand(&CONST_PI2),
-		 q, X->exp - CONST_PI2.exp);
-      tmp.exp = CONST_PI2.exp;
-      normalize(&tmp);
-      reg_move(&tmp, X);
-    }
-
-#ifdef FPTAN
-  if ( even == FPTAN )
-    {
-      if ( ((X->exp >= EXP_BIAS) ||
-	    ((X->exp == EXP_BIAS-1)
-	     && (X->sigh >= 0xc90fdaa2))) ^ (q & 1) )
-	even = FCOS;
-      else
-	even = 0;
+		 q, exponent(st0_ptr) - exponent(&CONST_PI2));
+      setexponent16(&tmp, exponent(&CONST_PI2));
+      st0_tag = FPU_normalize(&tmp);
+      FPU_copy_to_reg0(&tmp, st0_tag);
     }
-#endif FPTAN
 
   if ( (even && !(q & 1)) || (!even && (q & 1)) )
     {
-      reg_sub(&CONST_PI2, X, X, FULL_PRECISION);
+      st0_tag = FPU_sub(REV|LOADED|TAG_Valid, (int)&CONST_PI2, FULL_PRECISION);
+
 #ifdef BETTER_THAN_486
       /* So far, the results are exact but based upon a 64 bit
 	 precision approximation to pi/2. The technique used
 	 now is equivalent to using an approximation to pi/2 which
 	 is accurate to about 128 bits. */
-      if ( (X->exp <= CONST_PI2extra.exp + 64) || (q > 1) )
+      if ( (exponent(st0_ptr) <= exponent(&CONST_PI2extra) + 64) || (q > 1) )
 	{
-	  /* This code gives the effect of having p/2 to better than
+	  /* This code gives the effect of having pi/2 to better than
 	     128 bits precision. */
+
 	  significand(&tmp) = q + 1;
-	  tmp.exp = EXP_BIAS + 63;
-	  tmp.tag = TW_Valid;
-	  normalize(&tmp);
-	  reg_mul(&CONST_PI2extra, &tmp, &tmp, FULL_PRECISION);
-	  reg_add(X, &tmp,  X, FULL_PRECISION);
-	  if ( X->sign == SIGN_NEG )
+	  setexponent16(&tmp, 63);
+	  FPU_normalize(&tmp);
+	  tmptag =
+	    FPU_u_mul(&CONST_PI2extra, &tmp, &tmp, FULL_PRECISION, SIGN_POS,
+		      exponent16(&CONST_PI2extra) + exponent16(&tmp));
+	  st0_tag = FPU_add(&tmp, tmptag, 0, FULL_PRECISION);
+	  if ( signnegative(st0_ptr) )
 	    {
 	      /* CONST_PI2extra is negative, so the result of the addition
 		 can be negative. This means that the argument is actually
 		 in a different quadrant. The correction is always < pi/2,
 		 so it can't overflow into yet another quadrant. */
-	      X->sign = SIGN_POS;
+	      setpositive(st0_ptr);
 	      q++;
 	    }
 	}
@@ -114,33 +106,39 @@ static int trig_arg(FPU_REG *X, int even)
 	 precision approximation to pi/2. The technique used
 	 now is equivalent to using an approximation to pi/2 which
 	 is accurate to about 128 bits. */
-      if ( ((q > 0) && (X->exp <= CONST_PI2extra.exp + 64)) || (q > 1) )
+      if ( ((q > 0) && (exponent(st0_ptr) <= exponent(&CONST_PI2extra) + 64))
+	   || (q > 1) )
 	{
 	  /* This code gives the effect of having p/2 to better than
 	     128 bits precision. */
+
 	  significand(&tmp) = q;
-	  tmp.exp = EXP_BIAS + 63;
-	  tmp.tag = TW_Valid;
-	  normalize(&tmp);
-	  reg_mul(&CONST_PI2extra, &tmp, &tmp, FULL_PRECISION);
-	  reg_sub(X, &tmp, X, FULL_PRECISION);
-	  if ( (X->exp == CONST_PI2.exp) &&
-	      ((X->sigh > CONST_PI2.sigh)
-	       || ((X->sigh == CONST_PI2.sigh)
-		   && (X->sigl > CONST_PI2.sigl))) )
+	  setexponent16(&tmp, 63);
+	  FPU_normalize(&tmp);         /* This must return TAG_Valid */
+	  tmptag = FPU_u_mul(&CONST_PI2extra, &tmp, &tmp, FULL_PRECISION,
+			     SIGN_POS,
+			     exponent16(&CONST_PI2extra) + exponent16(&tmp));
+	  st0_tag = FPU_sub(LOADED|(tmptag & 0x0f), (int)&tmp,
+			    FULL_PRECISION);
+	  if ( (exponent(st0_ptr) == exponent(&CONST_PI2)) &&
+	      ((st0_ptr->sigh > CONST_PI2.sigh)
+	       || ((st0_ptr->sigh == CONST_PI2.sigh)
+		   && (st0_ptr->sigl > CONST_PI2.sigl))) )
 	    {
 	      /* CONST_PI2extra is negative, so the result of the
 		 subtraction can be larger than pi/2. This means
 		 that the argument is actually in a different quadrant.
 		 The correction is always < pi/2, so it can't overflow
 		 into yet another quadrant. */
-	      reg_sub(&CONST_PI, X, X, FULL_PRECISION);
+	      st0_tag = FPU_sub(REV|LOADED|TAG_Valid, (int)&CONST_PI2,
+				FULL_PRECISION);
 	      q++;
 	    }
 	}
     }
 #endif BETTER_THAN_486
 
+  FPU_settag0(st0_tag);
   control_word = old_cw;
   partial_status = saved_status & ~SW_C2;     /* Reduction complete. */
 
@@ -149,57 +147,56 @@ static int trig_arg(FPU_REG *X, int even)
 
 
 /* Convert a long to register */
-void convert_l2reg(long const *arg, FPU_REG *dest)
+static void convert_l2reg(long const *arg, int deststnr)
 {
+  int tag;
   long num = *arg;
+  u_char sign;
+  FPU_REG *dest = &st(deststnr);
 
   if (num == 0)
-    { reg_move(&CONST_Z, dest); return; }
+    {
+      FPU_copy_to_regi(&CONST_Z, TAG_Zero, deststnr);
+      return;
+    }
 
   if (num > 0)
-    dest->sign = SIGN_POS;
+    { sign = SIGN_POS; }
   else
-    { num = -num; dest->sign = SIGN_NEG; }
+    { num = -num; sign = SIGN_NEG; }
 
   dest->sigh = num;
   dest->sigl = 0;
-  dest->exp = EXP_BIAS + 31;
-  dest->tag = TW_Valid;
-  normalize(dest);
+  setexponent16(dest, 31);
+  tag = FPU_normalize(dest);
+  FPU_settagi(deststnr, tag);
+  setsign(dest, sign);
+  return;
 }
 
 
-static void single_arg_error(FPU_REG *st0_ptr)
+static void single_arg_error(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  switch ( st0_ptr->tag )
-    {
-    case TW_NaN:
-      if ( !(st0_ptr->sigh & 0x40000000) )   /* Signaling ? */
-	{
-	  EXCEPTION(EX_Invalid);
-	  if ( control_word & CW_Invalid )
-	    st0_ptr->sigh |= 0x40000000;	  /* Convert to a QNaN */
-	}
-      break;              /* return with a NaN in st(0) */
-    case TW_Empty:
-      stack_underflow();  /* Puts a QNaN in st(0) */
-      break;
+  if ( st0_tag == TAG_Empty )
+    FPU_stack_underflow();  /* Puts a QNaN in st(0) */
+  else if ( st0_tag == TW_NaN )
+    real_1op_NaN(st0_ptr);       /* return with a NaN in st(0) */
 #ifdef PARANOID
-    default:
-      EXCEPTION(EX_INTERNAL|0x0112);
+  else
+    EXCEPTION(EX_INTERNAL|0x0112);
 #endif PARANOID
-    }
 }
 
 
-static void single_arg_2_error(FPU_REG *st0_ptr)
+static void single_arg_2_error(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  FPU_REG *st_new_ptr;
+  int isNaN;
 
-  switch ( st0_ptr->tag )
+  switch ( st0_tag )
     {
     case TW_NaN:
-      if ( !(st0_ptr->sigh & 0x40000000) )   /* Signaling ? */
+      isNaN = (exponent(st0_ptr) == EXP_OVER) && (st0_ptr->sigh & 0x80000000);
+      if ( isNaN && !(st0_ptr->sigh & 0x40000000) )   /* Signaling ? */
 	{
 	  EXCEPTION(EX_Invalid);
 	  if ( control_word & CW_Invalid )
@@ -207,17 +204,27 @@ static void single_arg_2_error(FPU_REG *st0_ptr)
 	      /* The masked response */
 	      /* Convert to a QNaN */
 	      st0_ptr->sigh |= 0x40000000;
-	      st_new_ptr = &st(-1);
 	      push();
-	      reg_move(&st(1), st_new_ptr);
+	      FPU_copy_to_reg0(st0_ptr, TAG_Special);
 	    }
 	}
-      else
+      else if ( isNaN )
 	{
 	  /* A QNaN */
-	  st_new_ptr = &st(-1);
 	  push();
-	  reg_move(&st(1), st_new_ptr);
+	  FPU_copy_to_reg0(st0_ptr, TAG_Special);
+	}
+      else
+	{
+	  /* pseudoNaN or other unsupported */
+	  EXCEPTION(EX_Invalid);
+	  if ( control_word & CW_Invalid )
+	    {
+	      /* The masked response */
+	      FPU_copy_to_reg0(&CONST_QNaN, TAG_Special);
+	      push();
+	      FPU_copy_to_reg0(&CONST_QNaN, TAG_Special);
+	    }
 	}
       break;              /* return with a NaN in st(0) */
 #ifdef PARANOID
@@ -230,92 +237,88 @@ static void single_arg_2_error(FPU_REG *st0_ptr)
 
 /*---------------------------------------------------------------------------*/
 
-static void f2xm1(FPU_REG *st0_ptr)
+static void f2xm1(FPU_REG *st0_ptr, u_char tag)
 {
+  FPU_REG a;
+
   clear_C1();
-  switch ( st0_ptr->tag )
-    {
-    case TW_Valid:
-      {
-	if ( st0_ptr->exp >= 0 )
-	  {
-	    /* For an 80486 FPU, the result is undefined. */
-	  }
-#ifdef DENORM_OPERAND
-	else if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
-	  return;
-#endif DENORM_OPERAND
-	else
-	  {
-	    /* poly_2xm1(x) requires 0 < x < 1. */
-	    poly_2xm1(st0_ptr, st0_ptr);
-	  }
-	if ( st0_ptr->exp <= EXP_UNDER )
-	  {
-	    /* A denormal result has been produced.
-	       Precision must have been lost, this is always
-	       an underflow. */
-	    arith_underflow(st0_ptr);
-	  }
-	set_precision_flag_up();   /* 80486 appears to always do this */
-	return;
-      }
-    case TW_Zero:
+
+  if ( tag == TAG_Valid )
+    {
+      /* For an 80486 FPU, the result is undefined if the arg is >= 1.0 */
+      if ( exponent(st0_ptr) < 0 )
+	{
+	denormal_arg:
+
+	  FPU_to_exp16(st0_ptr, &a);
+
+	  /* poly_2xm1(x) requires 0 < st(0) < 1. */
+	  poly_2xm1(getsign(st0_ptr), &a, st0_ptr);
+	}
+      set_precision_flag_up();   /* 80486 appears to always do this */
       return;
+    }
+
+  if ( tag == TAG_Zero )
+    return;
+
+  if ( tag == TAG_Special )
+    tag = FPU_Special(st0_ptr);
+
+  switch ( tag )
+    {
+    case TW_Denormal:
+      if ( denormal_operand() < 0 )
+	return;
+      goto denormal_arg;
     case TW_Infinity:
-      if ( st0_ptr->sign == SIGN_NEG )
+      if ( signnegative(st0_ptr) )
 	{
 	  /* -infinity gives -1 (p16-10) */
-	  reg_move(&CONST_1, st0_ptr);
-	  st0_ptr->sign = SIGN_NEG;
+	  FPU_copy_to_reg0(&CONST_1, TAG_Valid);
+	  setnegative(st0_ptr);
 	}
       return;
     default:
-      single_arg_error(st0_ptr);
+      single_arg_error(st0_ptr, tag);
     }
 }
 
 
-static void fptan(FPU_REG *st0_ptr)
+static void fptan(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag;
   FPU_REG *st_new_ptr;
   int q;
-  char arg_sign = st0_ptr->sign;
+  u_char arg_sign = getsign(st0_ptr);
 
   /* Stack underflow has higher priority */
-  if ( st0_tag == TW_Empty )
+  if ( st0_tag == TAG_Empty )
     {
-      stack_underflow();  /* Puts a QNaN in st(0) */
+      FPU_stack_underflow();  /* Puts a QNaN in st(0) */
       if ( control_word & CW_Invalid )
 	{
 	  st_new_ptr = &st(-1);
 	  push();
-	  stack_underflow();  /* Puts a QNaN in the new st(0) */
+	  FPU_stack_underflow();  /* Puts a QNaN in the new st(0) */
 	}
       return;
     }
 
   if ( STACK_OVERFLOW )
-    { stack_overflow(); return; }
+    { FPU_stack_overflow(); return; }
 
-  switch ( st0_tag )
+  if ( st0_tag == TAG_Valid )
     {
-    case TW_Valid:
-      if ( st0_ptr->exp > EXP_BIAS - 40 )
+      if ( exponent(st0_ptr) > -40 )
 	{
-	  st0_ptr->sign = SIGN_POS;
-	  if ( (q = trig_arg(st0_ptr, 0)) != -1 )
-	    {
-	      poly_tan(st0_ptr, st0_ptr);
-	      st0_ptr->sign = (q & 1) ^ arg_sign;
-	    }
-	  else
+	  if ( (q = trig_arg(st0_ptr, 0)) == -1 )
 	    {
 	      /* Operand is out of range */
-	      st0_ptr->sign = arg_sign;         /* restore st(0) */
 	      return;
 	    }
+
+	  poly_tan(st0_ptr);
+	  setsign(st0_ptr, (q & 1) ^ (arg_sign != 0));
 	  set_precision_flag_up();  /* We do not really know if up or down */
 	}
       else
@@ -323,106 +326,134 @@ static void fptan(FPU_REG *st0_ptr)
 	  /* For a small arg, the result == the argument */
 	  /* Underflow may happen */
 
-	  if ( st0_ptr->exp <= EXP_UNDER )
-	    {
-#ifdef DENORM_OPERAND
-	      if ( denormal_operand() )
-		return;
-#endif DENORM_OPERAND
-	      /* A denormal result has been produced.
-		 Precision must have been lost, this is always
-		 an underflow. */
-	      if ( arith_underflow(st0_ptr) )
-		return;
-	    }
-	  set_precision_flag_down();  /* Must be down. */
+	denormal_arg:
+
+	  FPU_to_exp16(st0_ptr, st0_ptr);
+      
+	  st0_tag = FPU_round(st0_ptr, 1, 0, FULL_PRECISION, arg_sign);
+	  FPU_settag0(st0_tag);
 	}
       push();
-      reg_move(&CONST_1, st_new_ptr);
+      FPU_copy_to_reg0(&CONST_1, TAG_Valid);
       return;
-      break;
-    case TW_Infinity:
+    }
+
+  if ( st0_tag == TAG_Zero )
+    {
+      push();
+      FPU_copy_to_reg0(&CONST_1, TAG_Valid);
+      setcc(0);
+      return;
+    }
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+
+  if ( st0_tag == TW_Denormal )
+    {
+      if ( denormal_operand() < 0 )
+	return;
+
+      goto denormal_arg;
+    }
+
+  if ( st0_tag == TW_Infinity )
+    {
       /* The 80486 treats infinity as an invalid operand */
-      arith_invalid(st0_ptr);
-      if ( control_word & CW_Invalid )
+      if ( arith_invalid(0) >= 0 )
 	{
 	  st_new_ptr = &st(-1);
 	  push();
-	  arith_invalid(st_new_ptr);
+	  arith_invalid(0);
 	}
       return;
-    case TW_Zero:
-      push();
-      reg_move(&CONST_1, st_new_ptr);
-      setcc(0);
-      break;
-    default:
-      single_arg_2_error(st0_ptr);
-      break;
     }
+
+  single_arg_2_error(st0_ptr, st0_tag);
 }
 
 
-static void fxtract(FPU_REG *st0_ptr)
+static void fxtract(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag;
   FPU_REG *st_new_ptr;
+  u_char sign;
   register FPU_REG *st1_ptr = st0_ptr;  /* anticipate */
 
   if ( STACK_OVERFLOW )
-    {  stack_overflow(); return; }
+    {  FPU_stack_overflow(); return; }
+
   clear_C1();
-  if ( !(st0_tag ^ TW_Valid) )
+
+  if ( st0_tag == TAG_Valid )
     {
       long e;
 
-#ifdef DENORM_OPERAND
-      if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
-	return;
-#endif DENORM_OPERAND
-	  
       push();
-      reg_move(st1_ptr, st_new_ptr);
-      st_new_ptr->exp = EXP_BIAS;
-      e = st1_ptr->exp - EXP_BIAS;
-      convert_l2reg(&e, st1_ptr);
+      sign = getsign(st1_ptr);
+      reg_copy(st1_ptr, st_new_ptr);
+      setexponent16(st_new_ptr, exponent(st_new_ptr));
+
+    denormal_arg:
+
+      e = exponent16(st_new_ptr);
+      convert_l2reg(&e, 1);
+      setexponentpos(st_new_ptr, 0);
+      setsign(st_new_ptr, sign);
+      FPU_settag0(TAG_Valid);       /* Needed if arg was a denormal */
       return;
     }
-  else if ( st0_tag == TW_Zero )
+  else if ( st0_tag == TAG_Zero )
     {
-      char sign = st0_ptr->sign;
-      if ( divide_by_zero(SIGN_NEG, st0_ptr) )
+      sign = getsign(st0_ptr);
+
+      if ( FPU_divide_by_zero(0, SIGN_NEG) < 0 )
 	return;
+
       push();
-      reg_move(&CONST_Z, st_new_ptr);
-      st_new_ptr->sign = sign;
+      FPU_copy_to_reg0(&CONST_Z, TAG_Zero);
+      setsign(st_new_ptr, sign);
       return;
     }
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+
+  if ( st0_tag == TW_Denormal )
+    {
+      if (denormal_operand() < 0 )
+	return;
+
+      push();
+      sign = getsign(st1_ptr);
+      FPU_to_exp16(st1_ptr, st_new_ptr);
+      goto denormal_arg;
+    }
   else if ( st0_tag == TW_Infinity )
     {
-      char sign = st0_ptr->sign;
-      st0_ptr->sign = SIGN_POS;
+      sign = getsign(st0_ptr);
+      setpositive(st0_ptr);
       push();
-      reg_move(&CONST_INF, st_new_ptr);
-      st_new_ptr->sign = sign;
+      FPU_copy_to_reg0(&CONST_INF, TAG_Special);
+      setsign(st_new_ptr, sign);
       return;
     }
   else if ( st0_tag == TW_NaN )
     {
-      if ( real_2op_NaN(st0_ptr, st0_ptr, st0_ptr) )
+      if ( real_1op_NaN(st0_ptr) < 0 )
 	return;
+
       push();
-      reg_move(st1_ptr, st_new_ptr);
+      FPU_copy_to_reg0(st0_ptr, TAG_Special);
       return;
     }
-  else if ( st0_tag == TW_Empty )
+  else if ( st0_tag == TAG_Empty )
     {
       /* Is this the correct behaviour? */
       if ( control_word & EX_Invalid )
 	{
-	  stack_underflow();
+	  FPU_stack_underflow();
 	  push();
-	  stack_underflow();
+	  FPU_stack_underflow();
 	}
       else
 	EXCEPTION(EX_StackUnder);
@@ -434,193 +465,233 @@ static void fxtract(FPU_REG *st0_ptr)
 }
 
 
-static void fdecstp(FPU_REG *st0_ptr)
+static void fdecstp(void)
 {
   clear_C1();
-  top--;  /* st0_ptr will be fixed in math_emulate() before the next instr */
+  top--;
 }
 
-static void fincstp(FPU_REG *st0_ptr)
+static void fincstp(void)
 {
   clear_C1();
-  top++;  /* st0_ptr will be fixed in math_emulate() before the next instr */
+  top++;
 }
 
 
-static void fsqrt_(FPU_REG *st0_ptr)
+static void fsqrt_(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag;
+  int expon;
 
   clear_C1();
-  if ( !(st0_tag ^ TW_Valid) )
+
+  if ( st0_tag == TAG_Valid )
     {
-      int expon;
+      u_char tag;
       
-      if (st0_ptr->sign == SIGN_NEG)
+      if (signnegative(st0_ptr))
 	{
-	  arith_invalid(st0_ptr);  /* sqrt(negative) is invalid */
+	  arith_invalid(0);  /* sqrt(negative) is invalid */
 	  return;
 	}
 
-#ifdef DENORM_OPERAND
-      if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
-	return;
-#endif DENORM_OPERAND
+      /* make st(0) in  [1.0 .. 4.0) */
+      expon = exponent(st0_ptr);
 
-      expon = st0_ptr->exp - EXP_BIAS;
-      st0_ptr->exp = EXP_BIAS + (expon & 1);  /* make st(0) in  [1.0 .. 4.0) */
-      
-      wm_sqrt(st0_ptr, control_word);	/* Do the computation */
-      
-      st0_ptr->exp += expon >> 1;
-      st0_ptr->sign = SIGN_POS;
+    denormal_arg:
+
+      setexponent16(st0_ptr, (expon & 1));
+
+      /* Do the computation, the sign of the result will be positive. */
+      tag = wm_sqrt(st0_ptr, 0, 0, control_word, SIGN_POS);
+      addexponent(st0_ptr, expon >> 1);
+      FPU_settag0(tag);
+      return;
     }
-  else if ( st0_tag == TW_Zero )
+
+  if ( st0_tag == TAG_Zero )
     return;
-  else if ( st0_tag == TW_Infinity )
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+
+  if ( st0_tag == TW_Infinity )
     {
-      if ( st0_ptr->sign == SIGN_NEG )
-	arith_invalid(st0_ptr);  /* sqrt(-Infinity) is invalid */
+      if ( signnegative(st0_ptr) )
+	arith_invalid(0);  /* sqrt(-Infinity) is invalid */
       return;
     }
-  else
-    { single_arg_error(st0_ptr); return; }
+  else if ( st0_tag == TW_Denormal )
+    {
+      if (signnegative(st0_ptr))
+	{
+	  arith_invalid(0);  /* sqrt(negative) is invalid */
+	  return;
+	}
+
+      if ( denormal_operand() < 0 )
+	return;
+
+      FPU_to_exp16(st0_ptr, st0_ptr);
+
+      expon = exponent16(st0_ptr);
+
+      goto denormal_arg;
+    }
+
+  single_arg_error(st0_ptr, st0_tag);
 
 }
 
 
-static void frndint_(FPU_REG *st0_ptr)
+static void frndint_(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag;
-  int flags;
+  int flags, tag;
 
-  if ( !(st0_tag ^ TW_Valid) )
+  if ( st0_tag == TAG_Valid )
     {
-      if (st0_ptr->exp > EXP_BIAS+63)
-	return;
+      u_char sign;
+
+    denormal_arg:
+
+      sign = getsign(st0_ptr);
 
-#ifdef DENORM_OPERAND
-      if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+      if (exponent(st0_ptr) > 63)
 	return;
-#endif DENORM_OPERAND
+
+      if ( st0_tag == TW_Denormal )
+	{
+	  if (denormal_operand() < 0 )
+	    return;
+	}
 
       /* Fortunately, this can't overflow to 2^64 */
-      if ( (flags = round_to_int(st0_ptr)) )
+      if ( (flags = FPU_round_to_int(st0_ptr, st0_tag)) )
 	set_precision_flag(flags);
 
-      st0_ptr->exp = EXP_BIAS + 63;
-      normalize(st0_ptr);
+      setexponent16(st0_ptr, 63);
+      tag = FPU_normalize(st0_ptr);
+      setsign(st0_ptr, sign);
+      FPU_settag0(tag);
       return;
     }
-  else if ( (st0_tag == TW_Zero) || (st0_tag == TW_Infinity) )
+
+  if ( st0_tag == TAG_Zero )
+    return;
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+
+  if ( st0_tag == TW_Denormal )
+    goto denormal_arg;
+  else if ( st0_tag == TW_Infinity )
     return;
   else
-    single_arg_error(st0_ptr);
+    single_arg_error(st0_ptr, st0_tag);
 }
 
 
-static void fsin(FPU_REG *st0_ptr)
+static int fsin(FPU_REG *st0_ptr, u_char tag)
 {
-  char st0_tag = st0_ptr->tag;
-  char arg_sign = st0_ptr->sign;
+  u_char arg_sign = getsign(st0_ptr);
 
-  if ( st0_tag == TW_Valid )
+  if ( tag == TAG_Valid )
     {
-      FPU_REG rv;
       int q;
 
-      if ( st0_ptr->exp > EXP_BIAS - 40 )
+      if ( exponent(st0_ptr) > -40 )
 	{
-	  st0_ptr->sign = SIGN_POS;
-	  if ( (q = trig_arg(st0_ptr, 0)) != -1 )
+	  if ( (q = trig_arg(st0_ptr, 0)) == -1 )
 	    {
+	      /* Operand is out of range */
+	      return 1;
+	    }
 
-	      poly_sine(st0_ptr, &rv);
+	  poly_sine(st0_ptr);
+	  
+	  if (q & 2)
+	    changesign(st0_ptr);
 
-	      if (q & 2)
-		rv.sign ^= SIGN_POS ^ SIGN_NEG;
-	      rv.sign ^= arg_sign;
-	      reg_move(&rv, st0_ptr);
+	  setsign(st0_ptr, getsign(st0_ptr) ^ arg_sign);
 
-	      /* We do not really know if up or down */
-	      set_precision_flag_up();
-	      return;
-	    }
-	  else
-	    {
-	      /* Operand is out of range */
-	      st0_ptr->sign = arg_sign;         /* restore st(0) */
-	      return;
-	    }
+	  /* We do not really know if up or down */
+	  set_precision_flag_up();
+	  return 0;
 	}
       else
 	{
 	  /* For a small arg, the result == the argument */
-	  /* Underflow may happen */
-
-	  if ( st0_ptr->exp <= EXP_UNDER )
-	    {
-#ifdef DENORM_OPERAND
-	      if ( denormal_operand() )
-		return;
-#endif DENORM_OPERAND
-	      /* A denormal result has been produced.
-		 Precision must have been lost, this is always
-		 an underflow. */
-	      arith_underflow(st0_ptr);
-	      return;
-	    }
-
 	  set_precision_flag_up();  /* Must be up. */
+	  return 0;
 	}
     }
-  else if ( st0_tag == TW_Zero )
+
+  if ( tag == TAG_Zero )
     {
       setcc(0);
-      return;
+      return 0;
     }
-  else if ( st0_tag == TW_Infinity )
+
+  if ( tag == TAG_Special )
+    tag = FPU_Special(st0_ptr);
+
+  if ( tag == TW_Denormal )
+    {
+      if ( denormal_operand() < 0 )
+	return 1;
+
+      /* For a small arg, the result == the argument */
+      /* Underflow may happen */
+      FPU_to_exp16(st0_ptr, st0_ptr);
+      
+      tag = FPU_round(st0_ptr, 1, 0, FULL_PRECISION, arg_sign);
+
+      FPU_settag0(tag);
+
+      return 0;
+    }
+  else if ( tag == TW_Infinity )
     {
       /* The 80486 treats infinity as an invalid operand */
-      arith_invalid(st0_ptr);
-      return;
+      arith_invalid(0);
+      return 1;
     }
   else
-    single_arg_error(st0_ptr);
+    {
+      single_arg_error(st0_ptr, tag);
+      return 1;
+    }
 }
 
 
-static int f_cos(FPU_REG *arg)
+static int f_cos(FPU_REG *st0_ptr, u_char tag)
 {
-  char arg_sign = arg->sign;
+  u_char st0_sign;
+
+  st0_sign = getsign(st0_ptr);
 
-  if ( arg->tag == TW_Valid )
+  if ( tag == TAG_Valid )
     {
-      FPU_REG rv;
       int q;
 
-      if ( arg->exp > EXP_BIAS - 40 )
+      if ( exponent(st0_ptr) > -40 )
 	{
-	  arg->sign = SIGN_POS;
-	  if ( (arg->exp < EXP_BIAS)
-	      || ((arg->exp == EXP_BIAS)
-		  && (significand(arg) <= 0xc90fdaa22168c234LL)) )
+	  if ( (exponent(st0_ptr) < 0)
+	      || ((exponent(st0_ptr) == 0)
+		  && (significand(st0_ptr) <= 0xc90fdaa22168c234LL)) )
 	    {
-	      poly_cos(arg, &rv);
-	      reg_move(&rv, arg);
+	      poly_cos(st0_ptr);
 
 	      /* We do not really know if up or down */
 	      set_precision_flag_down();
 	  
 	      return 0;
 	    }
-	  else if ( (q = trig_arg(arg, FCOS)) != -1 )
+	  else if ( (q = trig_arg(st0_ptr, FCOS)) != -1 )
 	    {
-	      poly_sine(arg, &rv);
+	      poly_sine(st0_ptr);
 
 	      if ((q+1) & 2)
-		rv.sign ^= SIGN_POS ^ SIGN_NEG;
-	      reg_move(&rv, arg);
+		changesign(st0_ptr);
 
 	      /* We do not really know if up or down */
 	      set_precision_flag_down();
@@ -630,19 +701,15 @@ static int f_cos(FPU_REG *arg)
 	  else
 	    {
 	      /* Operand is out of range */
-	      arg->sign = arg_sign;         /* restore st(0) */
 	      return 1;
 	    }
 	}
       else
 	{
-#ifdef DENORM_OPERAND
-	  if ( (arg->exp <= EXP_UNDER) && (denormal_operand()) )
-	    return 1;
-#endif DENORM_OPERAND
+	denormal_arg:
 
 	  setcc(0);
-	  reg_move(&CONST_1, arg);
+	  FPU_copy_to_reg0(&CONST_1, TAG_Valid);
 #ifdef PECULIAR_486
 	  set_precision_flag_down();  /* 80486 appears to do this. */
 #else
@@ -651,79 +718,99 @@ static int f_cos(FPU_REG *arg)
 	  return 0;
 	}
     }
-  else if ( arg->tag == TW_Zero )
+  else if ( tag == TAG_Zero )
     {
-      reg_move(&CONST_1, arg);
+      FPU_copy_to_reg0(&CONST_1, TAG_Valid);
       setcc(0);
       return 0;
     }
-  else if ( arg->tag == TW_Infinity )
+
+  if ( tag == TAG_Special )
+    tag = FPU_Special(st0_ptr);
+
+  if ( tag == TW_Denormal )
+    {
+      if ( denormal_operand() < 0 )
+	return 1;
+
+      goto denormal_arg;
+    }
+  else if ( tag == TW_Infinity )
     {
       /* The 80486 treats infinity as an invalid operand */
-      arith_invalid(arg);
+      arith_invalid(0);
       return 1;
     }
   else
     {
-      single_arg_error(arg);  /* requires arg == &st(0) */
+      single_arg_error(st0_ptr, tag);  /* requires st0_ptr == &st(0) */
       return 1;
     }
 }
 
 
-static void fcos(FPU_REG *st0_ptr)
+static void fcos(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  f_cos(st0_ptr);
+  f_cos(st0_ptr, st0_tag);
 }
 
 
-static void fsincos(FPU_REG *st0_ptr)
+static void fsincos(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag;
   FPU_REG *st_new_ptr;
   FPU_REG arg;
+  u_char tag;
 
   /* Stack underflow has higher priority */
-  if ( st0_tag == TW_Empty )
+  if ( st0_tag == TAG_Empty )
     {
-      stack_underflow();  /* Puts a QNaN in st(0) */
+      FPU_stack_underflow();  /* Puts a QNaN in st(0) */
       if ( control_word & CW_Invalid )
 	{
 	  st_new_ptr = &st(-1);
 	  push();
-	  stack_underflow();  /* Puts a QNaN in the new st(0) */
+	  FPU_stack_underflow();  /* Puts a QNaN in the new st(0) */
 	}
       return;
     }
 
   if ( STACK_OVERFLOW )
-    { stack_overflow(); return; }
+    { FPU_stack_overflow(); return; }
+
+  if ( st0_tag == TAG_Special )
+    tag = FPU_Special(st0_ptr);
+  else
+    tag = st0_tag;
 
-  if ( st0_tag == TW_NaN )
+  if ( tag == TW_NaN )
     {
-      single_arg_2_error(st0_ptr);
+      single_arg_2_error(st0_ptr, TW_NaN);
       return;
     }
-  else if ( st0_tag == TW_Infinity )
+  else if ( tag == TW_Infinity )
     {
       /* The 80486 treats infinity as an invalid operand */
-      if ( !arith_invalid(st0_ptr) )
+      if ( arith_invalid(0) >= 0 )
 	{
-	  /* unmasked response */
+	  /* Masked response */
 	  push();
-	  arith_invalid(st_new_ptr);
+	  arith_invalid(0);
 	}
       return;
     }
 
-  reg_move(st0_ptr,&arg);
-  if ( !f_cos(&arg) )
+  reg_copy(st0_ptr, &arg);
+  if ( !fsin(st0_ptr, st0_tag) )
     {
-      fsin(st0_ptr);
       push();
-      reg_move(&arg,st_new_ptr);
+      FPU_copy_to_reg0(&arg, st0_tag);
+      f_cos(&st(0), st0_tag);
+    }
+  else
+    {
+      /* An error, so restore st(0) */
+      FPU_copy_to_reg0(&arg, st0_tag);
     }
-
 }
 
 
@@ -760,79 +847,86 @@ static void rem_kernel(unsigned long long st0, unsigned long long *y,
 /* Remainder of st(0) / st(1) */
 /* This routine produces exact results, i.e. there is never any
    rounding or truncation, etc of the result. */
-static void do_fprem(FPU_REG *st0_ptr, int round)
+static void do_fprem(FPU_REG *st0_ptr, u_char st0_tag, int round)
 {
   FPU_REG *st1_ptr = &st(1);
-  char st1_tag = st1_ptr->tag;
-  char st0_tag = st0_ptr->tag;
-  char sign = st0_ptr->sign;
+  u_char st1_tag = FPU_gettagi(1);
 
-  if ( !((st0_tag ^ TW_Valid) | (st1_tag ^ TW_Valid)) )
+  if ( !((st0_tag ^ TAG_Valid) | (st1_tag ^ TAG_Valid)) )
     {
-      FPU_REG tmp;
-      int old_cw = control_word;
-      int expdif = st0_ptr->exp - st1_ptr->exp;
+      FPU_REG tmp, st0, st1;
+      u_char st0_sign, st1_sign;
+      u_char tmptag;
+      int tag;
+      int old_cw;
+      int expdif;
       long long q;
       unsigned short saved_status;
-      int cc = 0;
+      int cc;
+
+    fprem_valid:
+      /* Convert registers for internal use. */
+      st0_sign = FPU_to_exp16(st0_ptr, &st0);
+      st1_sign = FPU_to_exp16(st1_ptr, &st1);
+      expdif = exponent16(&st0) - exponent16(&st1);
+
+      old_cw = control_word;
+      cc = 0;
 
-#ifdef DENORM_OPERAND
-      if ( ((st0_ptr->exp <= EXP_UNDER) ||
-	    (st1_ptr->exp <= EXP_UNDER)) && (denormal_operand()) )
-	return;
-#endif DENORM_OPERAND
-      
       /* We want the status following the denorm tests, but don't want
 	 the status changed by the arithmetic operations. */
       saved_status = partial_status;
       control_word &= ~CW_RC;
       control_word |= RC_CHOP;
 
-      if (expdif < 64)
+      if ( expdif < 64 )
 	{
 	  /* This should be the most common case */
 
 	  if ( expdif > -2 )
 	    {
-	      reg_div(st0_ptr, st1_ptr, &tmp, PR_64_BITS | RC_CHOP | 0x3f);
+	      u_char sign = st0_sign ^ st1_sign;
+	      tag = FPU_u_div(&st0, &st1, &tmp,
+			      PR_64_BITS | RC_CHOP | 0x3f,
+			      sign);
+	      setsign(&tmp, sign);
 
-	      if ( tmp.exp >= EXP_BIAS )
+	      if ( exponent(&tmp) >= 0 )
 		{
-		  round_to_int(&tmp);  /* Fortunately, this can't overflow
-					  to 2^64 */
+		  FPU_round_to_int(&tmp, tag);  /* Fortunately, this can't
+						   overflow to 2^64 */
 		  q = significand(&tmp);
 
-		  rem_kernel(significand(st0_ptr),
+		  rem_kernel(significand(&st0),
 			     &significand(&tmp),
-			     significand(st1_ptr),
+			     significand(&st1),
 			     q, expdif);
 
-		  tmp.exp = st1_ptr->exp;
+		  setexponent16(&tmp, exponent16(&st1));
 		}
 	      else
 		{
-		  reg_move(st0_ptr, &tmp);
+		  reg_copy(&st0, &tmp);
 		  q = 0;
 		}
-	      tmp.sign = sign;
 
 	      if ( (round == RC_RND) && (tmp.sigh & 0xc0000000) )
 		{
 		  /* We may need to subtract st(1) once more,
 		     to get a result <= 1/2 of st(1). */
 		  unsigned long long x;
-		  expdif = st1_ptr->exp - tmp.exp;
+		  expdif = exponent16(&st1) - exponent16(&tmp);
 		  if ( expdif <= 1 )
 		    {
 		      if ( expdif == 0 )
-			x = significand(st1_ptr) - significand(&tmp);
+			x = significand(&st1) - significand(&tmp);
 		      else /* expdif is 1 */
-			x = (significand(st1_ptr) << 1) - significand(&tmp);
+			x = (significand(&st1) << 1) - significand(&tmp);
 		      if ( (x < significand(&tmp)) ||
 			  /* or equi-distant (from 0 & st(1)) and q is odd */
 			  ((x == significand(&tmp)) && (q & 1) ) )
 			{
-			  tmp.sign ^= (SIGN_POS^SIGN_NEG);
+			  st0_sign = ! st0_sign;
 			  significand(&tmp) = x;
 			  q++;
 			}
@@ -855,28 +949,35 @@ static void do_fprem(FPU_REG *st0_ptr, int round)
 	  /* There is a large exponent difference ( >= 64 ) */
 	  /* To make much sense, the code in this section should
 	     be done at high precision. */
-	  int exp_1;
+	  int exp_1, N;
+	  u_char sign;
 
 	  /* prevent overflow here */
 	  /* N is 'a number between 32 and 63' (p26-113) */
-	  reg_move(st0_ptr, &tmp);
-	  tmp.exp = EXP_BIAS + 56;
-	  exp_1 = st1_ptr->exp;      st1_ptr->exp = EXP_BIAS;
-	  expdif -= 56;
-
-	  reg_div(&tmp, st1_ptr, &tmp, PR_64_BITS | RC_CHOP | 0x3f);
-	  st1_ptr->exp = exp_1;
-
-	  round_to_int(&tmp);  /* Fortunately, this can't overflow to 2^64 */
-
-	  rem_kernel(significand(st0_ptr),
+	  reg_copy(&st0, &tmp);
+	  tmptag = st0_tag;
+	  N = (expdif & 0x0000001f) + 32;  /* This choice gives results
+					      identical to an AMD 486 */
+	  setexponent16(&tmp, N);
+	  exp_1 = exponent16(&st1);
+	  setexponent16(&st1, 0);
+	  expdif -= N;
+
+	  sign = getsign(&tmp) ^ st1_sign;
+	  tag = FPU_u_div(&tmp, &st1, &tmp, PR_64_BITS | RC_CHOP | 0x3f,
+			  sign);
+	  setsign(&tmp, sign);
+
+	  FPU_round_to_int(&tmp, tag);  /* Fortunately, this can't
+					   overflow to 2^64 */
+
+	  rem_kernel(significand(&st0),
 		     &significand(&tmp),
-		     significand(st1_ptr),
+		     significand(&st1),
 		     significand(&tmp),
-		     tmp.exp - EXP_BIAS
+		     exponent(&tmp)
 		     ); 
-	  tmp.exp = exp_1 + expdif;
-	  tmp.sign = sign;
+	  setexponent16(&tmp, exp_1 + expdif);
 
 	  /* It is possible for the operation to be complete here.
 	     What does the IEEE standard say? The Intel 80486 manual
@@ -888,8 +989,8 @@ static void do_fprem(FPU_REG *st0_ptr, int round)
 	      /* The result is zero */
 	      control_word = old_cw;
 	      partial_status = saved_status;
-	      reg_move(&CONST_Z, st0_ptr);
-	      st0_ptr->sign = sign;
+	      FPU_copy_to_reg0(&CONST_Z, TAG_Zero);
+	      setsign(&st0, st0_sign);
 #ifdef PECULIAR_486
 	      setcc(SW_C2);
 #else
@@ -902,52 +1003,82 @@ static void do_fprem(FPU_REG *st0_ptr, int round)
 
       control_word = old_cw;
       partial_status = saved_status;
-      normalize_nuo(&tmp);
-      reg_move(&tmp, st0_ptr);
-      setcc(cc);
+      tag = FPU_normalize_nuo(&tmp);
+      reg_copy(&tmp, st0_ptr);
 
       /* The only condition to be looked for is underflow,
 	 and it can occur here only if underflow is unmasked. */
-      if ( (st0_ptr->exp <= EXP_UNDER) && (st0_ptr->tag != TW_Zero)
+      if ( (exponent16(&tmp) <= EXP_UNDER) && (tag != TAG_Zero)
 	  && !(control_word & CW_Underflow) )
-	arith_underflow(st0_ptr);
+	{
+	  setcc(cc);
+	  tag = arith_underflow(st0_ptr);
+	  setsign(st0_ptr, st0_sign);
+	  FPU_settag0(tag);
+	  return;
+	}
+      else if ( (exponent16(&tmp) > EXP_UNDER) || (tag == TAG_Zero) )
+	{
+	  stdexp(st0_ptr);
+	  setsign(st0_ptr, st0_sign);
+	}
+      else
+	{
+	  tag = FPU_round(st0_ptr, 0, 0, FULL_PRECISION, st0_sign);
+	}
+      FPU_settag0(tag);
+      setcc(cc);
 
       return;
     }
-  else if ( (st0_tag == TW_Empty) | (st1_tag == TW_Empty) )
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+  if ( st1_tag == TAG_Special )
+    st1_tag = FPU_Special(st1_ptr);
+
+  if ( ((st0_tag == TAG_Valid) && (st1_tag == TW_Denormal))
+	    || ((st0_tag == TW_Denormal) && (st1_tag == TAG_Valid))
+	    || ((st0_tag == TW_Denormal) && (st1_tag == TW_Denormal)) )
+    {
+      if ( denormal_operand() < 0 )
+	return;
+      goto fprem_valid;
+    }
+  else if ( (st0_tag == TAG_Empty) | (st1_tag == TAG_Empty) )
     {
-      stack_underflow();
+      FPU_stack_underflow();
       return;
     }
-  else if ( st0_tag == TW_Zero )
+  else if ( st0_tag == TAG_Zero )
     {
-      if ( st1_tag == TW_Valid )
+      if ( st1_tag == TAG_Valid )
 	{
-#ifdef DENORM_OPERAND
-	  if ( (st1_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	  setcc(0); return;
+	}
+      else if ( st1_tag == TW_Denormal )
+	{
+	  if ( denormal_operand() < 0 )
 	    return;
-#endif DENORM_OPERAND
-
 	  setcc(0); return;
 	}
-      else if ( st1_tag == TW_Zero )
-	{ arith_invalid(st0_ptr); return; } /* fprem(?,0) always invalid */
+      else if ( st1_tag == TAG_Zero )
+	{ arith_invalid(0); return; } /* fprem(?,0) always invalid */
       else if ( st1_tag == TW_Infinity )
 	{ setcc(0); return; }
     }
-  else if ( st0_tag == TW_Valid )
+  else if ( (st0_tag == TAG_Valid) || (st0_tag == TW_Denormal) )
     {
-      if ( st1_tag == TW_Zero )
+      if ( st1_tag == TAG_Zero )
 	{
-	  arith_invalid(st0_ptr); /* fprem(Valid,Zero) is invalid */
+	  arith_invalid(0); /* fprem(Valid,Zero) is invalid */
 	  return;
 	}
       else if ( st1_tag != TW_NaN )
 	{
-#ifdef DENORM_OPERAND
-	  if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	  if ( ((st0_tag == TW_Denormal) || (st1_tag == TW_Denormal))
+	       && (denormal_operand() < 0) )
 	    return;
-#endif DENORM_OPERAND
 
 	  if ( st1_tag == TW_Infinity )
 	    {
@@ -960,729 +1091,710 @@ static void do_fprem(FPU_REG *st0_ptr, int round)
     {
       if ( st1_tag != TW_NaN )
 	{
-	  arith_invalid(st0_ptr); /* fprem(Infinity,?) is invalid */
+	  arith_invalid(0); /* fprem(Infinity,?) is invalid */
 	  return;
 	}
     }
 
-  /* One of the registers must contain a NaN is we got here. */
+  /* One of the registers must contain a NaN if we got here. */
 
 #ifdef PARANOID
   if ( (st0_tag != TW_NaN) && (st1_tag != TW_NaN) )
       EXCEPTION(EX_INTERNAL | 0x118);
 #endif PARANOID
 
-  real_2op_NaN(st1_ptr, st0_ptr, st0_ptr);
+  real_2op_NaN(st1_ptr, st1_tag, 0, st1_ptr);
 
 }
 
 
 /* ST(1) <- ST(1) * log ST;  pop ST */
-static void fyl2x(FPU_REG *st0_ptr)
+static void fyl2x(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag;
   FPU_REG *st1_ptr = &st(1), exponent;
-  char st1_tag = st1_ptr->tag;
-  int e;
+  u_char st1_tag = FPU_gettagi(1);
+  u_char sign;
+  int e, tag;
 
   clear_C1();
-  if ( !((st0_tag ^ TW_Valid) | (st1_tag ^ TW_Valid)) )
+
+  if ( (st0_tag == TAG_Valid) && (st1_tag == TAG_Valid) )
     {
-      if ( st0_ptr->sign == SIGN_POS )
+    both_valid:
+      /* Both regs are Valid or Denormal */
+      if ( signpositive(st0_ptr) )
 	{
-#ifdef DENORM_OPERAND
-	  if ( ((st0_ptr->exp <= EXP_UNDER) ||
-		(st1_ptr->exp <= EXP_UNDER)) && (denormal_operand()) )
-	    return;
-#endif DENORM_OPERAND
+	  if ( st0_tag == TW_Denormal )
+	    FPU_to_exp16(st0_ptr, st0_ptr);
+	  else
+	    /* Convert st(0) for internal use. */
+	    setexponent16(st0_ptr, exponent(st0_ptr));
 
 	  if ( (st0_ptr->sigh == 0x80000000) && (st0_ptr->sigl == 0) )
 	    {
 	      /* Special case. The result can be precise. */
-	      e = st0_ptr->exp - EXP_BIAS;
-	      if ( e > 0 )
+	      u_char esign;
+	      e = exponent16(st0_ptr);
+	      if ( e >= 0 )
 		{
 		  exponent.sigh = e;
-		  exponent.sign = SIGN_POS;
+		  esign = SIGN_POS;
 		}
 	      else
 		{
 		  exponent.sigh = -e;
-		  exponent.sign = SIGN_NEG;
+		  esign = SIGN_NEG;
 		}
 	      exponent.sigl = 0;
-	      exponent.exp = EXP_BIAS + 31;
-	      exponent.tag = TW_Valid;
-	      normalize_nuo(&exponent);
-	      reg_mul(&exponent, st1_ptr, st1_ptr, FULL_PRECISION);
+	      setexponent16(&exponent, 31);
+	      tag = FPU_normalize_nuo(&exponent);
+	      stdexp(&exponent);
+	      setsign(&exponent, esign);
+	      tag = FPU_mul(&exponent, tag, 1, FULL_PRECISION);
+	      if ( tag >= 0 )
+		FPU_settagi(1, tag);
 	    }
 	  else
 	    {
 	      /* The usual case */
-	      poly_l2(st0_ptr, st1_ptr, st1_ptr);
-	      if ( st1_ptr->exp <= EXP_UNDER )
-		{
-		  /* A denormal result has been produced.
-		     Precision must have been lost, this is always
-		     an underflow. */
-		  arith_underflow(st1_ptr);
-		}
+	      sign = getsign(st1_ptr);
+	      if ( st1_tag == TW_Denormal )
+		FPU_to_exp16(st1_ptr, st1_ptr);
 	      else
-		set_precision_flag_up();  /* 80486 appears to always do this */
+		/* Convert st(1) for internal use. */
+		setexponent16(st1_ptr, exponent(st1_ptr));
+	      poly_l2(st0_ptr, st1_ptr, sign);
 	    }
-	  pop();
-	  return;
 	}
       else
 	{
 	  /* negative */
-	  if ( !arith_invalid(st1_ptr) )
-	    pop();
-	  return;
+	  if ( arith_invalid(1) < 0 )
+	    return;
 	}
-    }
-  else if ( (st0_tag == TW_Empty) || (st1_tag == TW_Empty) )
-    {
-      stack_underflow_pop(1);
+
+      FPU_pop();
+
       return;
     }
-  else if ( (st0_tag == TW_NaN) || (st1_tag == TW_NaN) )
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+  if ( st1_tag == TAG_Special )
+    st1_tag = FPU_Special(st1_ptr);
+
+  if ( (st0_tag == TAG_Empty) || (st1_tag == TAG_Empty) )
     {
-      if ( !real_2op_NaN(st0_ptr, st1_ptr, st1_ptr) )
-	pop();
+      FPU_stack_underflow_pop(1);
       return;
     }
-  else if ( (st0_tag <= TW_Zero) && (st1_tag <= TW_Zero) )
+  else if ( (st0_tag <= TW_Denormal) && (st1_tag <= TW_Denormal) )
     {
-      /* one of the args is zero, the other valid, or both zero */
-      if ( st0_tag == TW_Zero )
+      if ( st0_tag == TAG_Zero )
 	{
-	  if ( st1_tag == TW_Zero )
+	  if ( st1_tag == TAG_Zero )
 	    {
 	      /* Both args zero is invalid */
-	      if ( !arith_invalid(st1_ptr) )
-		pop();
-	    }
-#ifdef PECULIAR_486
-	  /* This case is not specifically covered in the manual,
-	     but divide-by-zero would seem to be the best response.
-	     However, a real 80486 does it this way... */
-	  else if ( st0_ptr->tag == TW_Infinity )
-	    {
-	      reg_move(&CONST_INF, st1_ptr);
-	      pop();
+	      if ( arith_invalid(1) < 0 )
+		return;
 	    }
-#endif PECULIAR_486
 	  else
 	    {
-	      if ( !divide_by_zero(st1_ptr->sign^SIGN_NEG^SIGN_POS, st1_ptr) )
-		pop();
+	      u_char sign;
+	      sign = getsign(st1_ptr)^SIGN_NEG;
+	      if ( FPU_divide_by_zero(1, sign) < 0 )
+		return;
+
+	      setsign(st1_ptr, sign);
 	    }
-	  return;
 	}
-      else
+      else if ( st1_tag == TAG_Zero )
 	{
 	  /* st(1) contains zero, st(0) valid <> 0 */
 	  /* Zero is the valid answer */
-	  char sign = st1_ptr->sign;
-
-	  if ( st0_ptr->sign == SIGN_NEG )
+	  sign = getsign(st1_ptr);
+	  
+	  if ( signnegative(st0_ptr) )
 	    {
 	      /* log(negative) */
-	      if ( !arith_invalid(st1_ptr) )
-		pop();
-	      return;
+	      if ( arith_invalid(1) < 0 )
+		return;
 	    }
-
-#ifdef DENORM_OPERAND
-	  if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	  else if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
 	    return;
-#endif DENORM_OPERAND
+	  else
+	    {
+	      if ( exponent(st0_ptr) < 0 )
+		sign ^= SIGN_NEG;
 
-	  if ( st0_ptr->exp < EXP_BIAS ) sign ^= SIGN_NEG^SIGN_POS;
-	  pop(); st0_ptr = &st(0);
-	  reg_move(&CONST_Z, st0_ptr);
-	  st0_ptr->sign = sign;
-	  return;
+	      FPU_copy_to_reg1(&CONST_Z, TAG_Zero);
+	      setsign(st1_ptr, sign);
+	    }
 	}
+      else
+	{
+	  /* One or both operands are denormals. */
+	  if ( denormal_operand() < 0 )
+	    return;
+	  goto both_valid;
+	}
+    }
+  else if ( (st0_tag == TW_NaN) || (st1_tag == TW_NaN) )
+    {
+      if ( real_2op_NaN(st0_ptr, st0_tag, 1, st0_ptr) < 0 )
+	return;
     }
   /* One or both arg must be an infinity */
   else if ( st0_tag == TW_Infinity )
     {
-      if ( (st0_ptr->sign == SIGN_NEG) || (st1_tag == TW_Zero) )
+      if ( (signnegative(st0_ptr)) || (st1_tag == TAG_Zero) )
 	{
 	  /* log(-infinity) or 0*log(infinity) */
-	  if ( !arith_invalid(st1_ptr) )
-	    pop();
-	  return;
+	  if ( arith_invalid(1) < 0 )
+	    return;
 	}
       else
 	{
-	  char sign = st1_ptr->sign;
+	  u_char sign = getsign(st1_ptr);
 
-#ifdef DENORM_OPERAND
-	  if ( (st1_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	  if ( (st1_tag == TW_Denormal) && (denormal_operand() < 0) )
 	    return;
-#endif DENORM_OPERAND
 
-	  pop(); st0_ptr = &st(0);
-	  reg_move(&CONST_INF, st0_ptr);
-	  st0_ptr->sign = sign;
-	  return;
+	  FPU_copy_to_reg1(&CONST_INF, TAG_Special);
+	  setsign(st1_ptr, sign);
 	}
     }
   /* st(1) must be infinity here */
-  else if ( (st0_tag == TW_Valid) && (st0_ptr->sign == SIGN_POS) )
+  else if ( ((st0_tag == TAG_Valid) || (st0_tag == TW_Denormal))
+	    && ( signpositive(st0_ptr) ) )
     {
-      if ( st0_ptr->exp >= EXP_BIAS )
+      if ( exponent(st0_ptr) >= 0 )
 	{
-	  if ( (st0_ptr->exp == EXP_BIAS) &&
+	  if ( (exponent(st0_ptr) == 0) &&
 	      (st0_ptr->sigh == 0x80000000) &&
 	      (st0_ptr->sigl == 0) )
 	    {
 	      /* st(0) holds 1.0 */
 	      /* infinity*log(1) */
-	      if ( !arith_invalid(st1_ptr) )
-		pop();
-	      return;
+	      if ( arith_invalid(1) < 0 )
+		return;
 	    }
-	  /* st(0) is positive and > 1.0 */
-	  pop();
+	  /* else st(0) is positive and > 1.0 */
 	}
       else
 	{
 	  /* st(0) is positive and < 1.0 */
 
-#ifdef DENORM_OPERAND
-	  if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	  if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
 	    return;
-#endif DENORM_OPERAND
 
-	  st1_ptr->sign ^= SIGN_NEG;
-	  pop();
+	  changesign(st1_ptr);
 	}
-      return;
     }
   else
     {
       /* st(0) must be zero or negative */
-      if ( st0_ptr->tag == TW_Zero )
+      if ( st0_tag == TAG_Zero )
 	{
 	  /* This should be invalid, but a real 80486 is happy with it. */
+
 #ifndef PECULIAR_486
-	  if ( !divide_by_zero(st1_ptr->sign, st1_ptr) )
+	  sign = getsign(st1_ptr);
+	  if ( FPU_divide_by_zero(1, sign) < 0 )
+	    return;
 #endif PECULIAR_486
-	    {
-	      st1_ptr->sign ^= SIGN_NEG^SIGN_POS;
-	      pop();
-	    }
-	}
-      else
-	{
-	  /* log(negative) */
-	  if ( !arith_invalid(st1_ptr) )
-	    pop();
+
+	  changesign(st1_ptr);
 	}
-      return;
+      else if ( arith_invalid(1) < 0 )	  /* log(negative) */
+	return;
     }
+
+  FPU_pop();
 }
 
 
-static void fpatan(FPU_REG *st0_ptr)
+static void fpatan(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag;
   FPU_REG *st1_ptr = &st(1);
-  char st1_tag = st1_ptr->tag;
+  u_char st1_tag = FPU_gettagi(1);
+  int tag;
 
   clear_C1();
-  if ( !((st0_tag ^ TW_Valid) | (st1_tag ^ TW_Valid)) )
+  if ( !((st0_tag ^ TAG_Valid) | (st1_tag ^ TAG_Valid)) )
     {
-#ifdef DENORM_OPERAND
-      if ( ((st0_ptr->exp <= EXP_UNDER) ||
-	    (st1_ptr->exp <= EXP_UNDER)) && (denormal_operand()) )
-	return;
-#endif DENORM_OPERAND
+    valid_atan:
 
-      poly_atan(st0_ptr, st1_ptr, st1_ptr);
+      poly_atan(st0_ptr, st0_tag, st1_ptr, st1_tag);
 
-      if ( st1_ptr->exp <= EXP_UNDER )
-	{
-	  /* A denormal result has been produced.
-	     Precision must have been lost.
-	     This is by definition an underflow. */
-	  arith_underflow(st1_ptr);
-	  pop();
-	  return;
-	}
+      FPU_pop();
+
+      return;
     }
-  else if ( (st0_tag == TW_Empty) || (st1_tag == TW_Empty) )
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+  if ( st1_tag == TAG_Special )
+    st1_tag = FPU_Special(st1_ptr);
+
+  if ( ((st0_tag == TAG_Valid) && (st1_tag == TW_Denormal))
+	    || ((st0_tag == TW_Denormal) && (st1_tag == TAG_Valid))
+	    || ((st0_tag == TW_Denormal) && (st1_tag == TW_Denormal)) )
     {
-      stack_underflow_pop(1);
+      if ( denormal_operand() < 0 )
+	return;
+
+      goto valid_atan;
+    }
+  else if ( (st0_tag == TAG_Empty) || (st1_tag == TAG_Empty) )
+    {
+      FPU_stack_underflow_pop(1);
       return;
     }
   else if ( (st0_tag == TW_NaN) || (st1_tag == TW_NaN) )
     {
-      if ( !real_2op_NaN(st0_ptr, st1_ptr, st1_ptr) )
-	  pop();
+      if ( real_2op_NaN(st0_ptr, st0_tag, 1, st0_ptr) >= 0 )
+	  FPU_pop();
       return;
     }
   else if ( (st0_tag == TW_Infinity) || (st1_tag == TW_Infinity) )
     {
-      char sign = st1_ptr->sign;
+      u_char sign = getsign(st1_ptr);
       if ( st0_tag == TW_Infinity )
 	{
 	  if ( st1_tag == TW_Infinity )
 	    {
-	      if ( st0_ptr->sign == SIGN_POS )
-		{ reg_move(&CONST_PI4, st1_ptr); }
+	      if ( signpositive(st0_ptr) )
+		{
+		  FPU_copy_to_reg1(&CONST_PI4, TAG_Valid);
+		}
 	      else
-		reg_add(&CONST_PI4, &CONST_PI2, st1_ptr, FULL_PRECISION);
+		{
+		  setpositive(st1_ptr);
+		  tag = FPU_u_add(&CONST_PI4, &CONST_PI2, st1_ptr,
+				  FULL_PRECISION, SIGN_POS,
+				  exponent(&CONST_PI4), exponent(&CONST_PI2));
+		  if ( tag >= 0 )
+		    FPU_settagi(1, tag);
+		}
 	    }
 	  else
 	    {
-#ifdef DENORM_OPERAND
-	      if ( st1_tag != TW_Zero )
-		{
-		  if ( (st1_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
-		    return;
-		}
-#endif DENORM_OPERAND
+	      if ( (st1_tag == TW_Denormal) && (denormal_operand() < 0) )
+		return;
 
-	      if ( st0_ptr->sign == SIGN_POS )
+	      if ( signpositive(st0_ptr) )
 		{
-		  reg_move(&CONST_Z, st1_ptr);
-		  st1_ptr->sign = sign;   /* An 80486 preserves the sign */
-		  pop();
+		  FPU_copy_to_reg1(&CONST_Z, TAG_Zero);
+		  setsign(st1_ptr, sign);   /* An 80486 preserves the sign */
+		  FPU_pop();
 		  return;
 		}
 	      else
-		reg_move(&CONST_PI, st1_ptr);
+		{
+		  FPU_copy_to_reg1(&CONST_PI, TAG_Valid);
+		}
 	    }
 	}
       else
 	{
 	  /* st(1) is infinity, st(0) not infinity */
-#ifdef DENORM_OPERAND
-	  if ( st0_tag != TW_Zero )
-	    {
-	      if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
-		return;
-	    }
-#endif DENORM_OPERAND
+	  if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
+	    return;
 
-	  reg_move(&CONST_PI2, st1_ptr);
+	  FPU_copy_to_reg1(&CONST_PI2, TAG_Valid);
 	}
-      st1_ptr->sign = sign;
+      setsign(st1_ptr, sign);
     }
-  else if ( st1_tag == TW_Zero )
+  else if ( st1_tag == TAG_Zero )
     {
       /* st(0) must be valid or zero */
-      char sign = st1_ptr->sign;
+      u_char sign = getsign(st1_ptr);
+
+      if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
+	return;
 
-#ifdef DENORM_OPERAND
-      if ( st0_tag != TW_Zero )
+      if ( signpositive(st0_ptr) )
 	{
-	  if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
-	    return;
+	  /* An 80486 preserves the sign */
+	  FPU_pop();
+	  return;
 	}
-#endif DENORM_OPERAND
 
-      if ( st0_ptr->sign == SIGN_POS )
-	{ /* An 80486 preserves the sign */ pop(); return; }
-      else
-	reg_move(&CONST_PI, st1_ptr);
-      st1_ptr->sign = sign;
+      FPU_copy_to_reg1(&CONST_PI, TAG_Valid);
+      setsign(st1_ptr, sign);
     }
-  else if ( st0_tag == TW_Zero )
+  else if ( st0_tag == TAG_Zero )
     {
-      /* st(1) must be TW_Valid here */
-      char sign = st1_ptr->sign;
+      /* st(1) must be TAG_Valid here */
+      u_char sign = getsign(st1_ptr);
 
-#ifdef DENORM_OPERAND
-      if ( (st1_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+      if ( (st1_tag == TW_Denormal) && (denormal_operand() < 0) )
 	return;
-#endif DENORM_OPERAND
 
-      reg_move(&CONST_PI2, st1_ptr);
-      st1_ptr->sign = sign;
+      FPU_copy_to_reg1(&CONST_PI2, TAG_Valid);
+      setsign(st1_ptr, sign);
     }
 #ifdef PARANOID
   else
     EXCEPTION(EX_INTERNAL | 0x125);
 #endif PARANOID
 
-  pop();
+  FPU_pop();
   set_precision_flag_up();  /* We do not really know if up or down */
 }
 
 
-static void fprem(FPU_REG *st0_ptr)
+static void fprem(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  do_fprem(st0_ptr, RC_CHOP);
+  do_fprem(st0_ptr, st0_tag, RC_CHOP);
 }
 
 
-static void fprem1(FPU_REG *st0_ptr)
+static void fprem1(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  do_fprem(st0_ptr, RC_RND);
+  do_fprem(st0_ptr, st0_tag, RC_RND);
 }
 
 
-static void fyl2xp1(FPU_REG *st0_ptr)
+static void fyl2xp1(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag, sign;
-  FPU_REG *st1_ptr = &st(1);
-  char st1_tag = st1_ptr->tag;
+  u_char sign, sign1;
+  FPU_REG *st1_ptr = &st(1), a, b;
+  u_char st1_tag = FPU_gettagi(1);
 
   clear_C1();
-  if ( !((st0_tag ^ TW_Valid) | (st1_tag ^ TW_Valid)) )
+  if ( !((st0_tag ^ TAG_Valid) | (st1_tag ^ TAG_Valid)) )
     {
-#ifdef DENORM_OPERAND
-      if ( ((st0_ptr->exp <= EXP_UNDER) ||
-	    (st1_ptr->exp <= EXP_UNDER)) && denormal_operand() )
+    valid_yl2xp1:
+
+      sign = getsign(st0_ptr);
+      sign1 = getsign(st1_ptr);
+
+      FPU_to_exp16(st0_ptr, &a);
+      FPU_to_exp16(st1_ptr, &b);
+
+      if ( poly_l2p1(sign, sign1, &a, &b, st1_ptr) )
 	return;
-#endif DENORM_OPERAND
 
-      if ( poly_l2p1(st0_ptr, st1_ptr, st1_ptr) )
-	{
-#ifdef PECULIAR_486   /* Stupid 80486 doesn't worry about log(negative). */
-	  st1_ptr->sign ^= SIGN_POS^SIGN_NEG;
-#else
-	  if ( arith_invalid(st1_ptr) )  /* poly_l2p1() returned invalid */
-	    return;
-#endif PECULIAR_486
-	}
-      if ( st1_ptr->exp <= EXP_UNDER )
-	{
-	  /* A denormal result has been produced.
-	     Precision must have been lost, this is always
-	     an underflow. */
-	  sign = st1_ptr->sign;
-	  arith_underflow(st1_ptr);
-	  st1_ptr->sign = sign;
-	}
-      else
-	set_precision_flag_up();   /* 80486 appears to always do this */
-      pop();
+      FPU_pop();
       return;
     }
-  else if ( (st0_tag == TW_Empty) | (st1_tag == TW_Empty) )
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+  if ( st1_tag == TAG_Special )
+    st1_tag = FPU_Special(st1_ptr);
+
+  if ( ((st0_tag == TAG_Valid) && (st1_tag == TW_Denormal))
+	    || ((st0_tag == TW_Denormal) && (st1_tag == TAG_Valid))
+	    || ((st0_tag == TW_Denormal) && (st1_tag == TW_Denormal)) )
+    {
+      if ( denormal_operand() < 0 )
+	return;
+
+      goto valid_yl2xp1;
+    }
+  else if ( (st0_tag == TAG_Empty) | (st1_tag == TAG_Empty) )
     {
-      stack_underflow_pop(1);
+      FPU_stack_underflow_pop(1);
       return;
     }
-  else if ( st0_tag == TW_Zero )
+  else if ( st0_tag == TAG_Zero )
     {
-      if ( st1_tag <= TW_Zero )
+      switch ( st1_tag )
 	{
-#ifdef DENORM_OPERAND
-	  if ( (st1_tag == TW_Valid) && (st1_ptr->exp <= EXP_UNDER) &&
-	      (denormal_operand()) )
+	case TW_Denormal:
+	  if ( denormal_operand() < 0 )
 	    return;
-#endif DENORM_OPERAND
-	  
-	  st0_ptr->sign ^= st1_ptr->sign;
-	  reg_move(st0_ptr, st1_ptr);
-	}
-      else if ( st1_tag == TW_Infinity )
-	{
+
+	case TAG_Zero:
+	case TAG_Valid:
+	  setsign(st0_ptr, getsign(st0_ptr) ^ getsign(st1_ptr));
+	  FPU_copy_to_reg1(st0_ptr, st0_tag);
+	  break;
+
+	case TW_Infinity:
 	  /* Infinity*log(1) */
-	  if ( !arith_invalid(st1_ptr) )
-	    pop();
-	  return;
-	}
-      else if ( st1_tag == TW_NaN )
-	{
-	  if ( !real_2op_NaN(st0_ptr, st1_ptr, st1_ptr) )
-	    pop();
-	  return;
-	}
+	  if ( arith_invalid(1) < 0 )
+	    return;
+	  break;
+
+	case TW_NaN:
+	  if ( real_2op_NaN(st0_ptr, st0_tag, 1, st0_ptr) < 0 )
+	    return;
+	  break;
+
+	default:
 #ifdef PARANOID
-      else
-	{
 	  EXCEPTION(EX_INTERNAL | 0x116);
 	  return;
-	}
 #endif PARANOID
-      pop(); return;
+	}
     }
-  else if ( st0_tag == TW_Valid )
+  else if ( (st0_tag == TAG_Valid) || (st0_tag == TW_Denormal) )
     {
-      if ( st1_tag == TW_Zero )
+      switch ( st1_tag )
 	{
-	  if ( st0_ptr->sign == SIGN_NEG )
+	case TAG_Zero:
+	  if ( signnegative(st0_ptr) )
 	    {
-	      if ( st0_ptr->exp >= EXP_BIAS )
+	      if ( exponent(st0_ptr) >= 0 )
 		{
 		  /* st(0) holds <= -1.0 */
 #ifdef PECULIAR_486   /* Stupid 80486 doesn't worry about log(negative). */
-		  st1_ptr->sign ^= SIGN_POS^SIGN_NEG;
+		  changesign(st1_ptr);
 #else
-		  if ( arith_invalid(st1_ptr) ) return;
+		  if ( arith_invalid(1) < 0 )
+		    return;
 #endif PECULIAR_486
-		  pop(); return;
 		}
-#ifdef DENORM_OPERAND
-	      if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	      else if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
 		return;
-#endif DENORM_OPERAND
-	      st1_ptr->sign ^= SIGN_POS^SIGN_NEG;
-	      pop(); return;
+	      else
+		changesign(st1_ptr);
 	    }
-#ifdef DENORM_OPERAND
-	  if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	  else if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
 	    return;
-#endif DENORM_OPERAND
-	  pop(); return;
-	}
-      if ( st1_tag == TW_Infinity )
-	{
-	  if ( st0_ptr->sign == SIGN_NEG )
+	  break;
+
+	case TW_Infinity:
+	  if ( signnegative(st0_ptr) )
 	    {
-	      if ( (st0_ptr->exp >= EXP_BIAS) &&
+	      if ( (exponent(st0_ptr) >= 0) &&
 		  !((st0_ptr->sigh == 0x80000000) &&
 		    (st0_ptr->sigl == 0)) )
 		{
 		  /* st(0) holds < -1.0 */
 #ifdef PECULIAR_486   /* Stupid 80486 doesn't worry about log(negative). */
-		  st1_ptr->sign ^= SIGN_POS^SIGN_NEG;
+		  changesign(st1_ptr);
 #else
-		  if ( arith_invalid(st1_ptr) ) return;
+		  if ( arith_invalid(1) < 0 ) return;
 #endif PECULIAR_486
-		  pop(); return;
 		}
-#ifdef DENORM_OPERAND
-	      if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	      else if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
 		return;
-#endif DENORM_OPERAND
-	      st1_ptr->sign ^= SIGN_POS^SIGN_NEG;
-	      pop(); return;
+	      else
+		changesign(st1_ptr);
 	    }
-#ifdef DENORM_OPERAND
-	  if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	  else if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
+	    return;
+	  break;
+
+	case TW_NaN:
+	  if ( real_2op_NaN(st0_ptr, st0_tag, 1, st0_ptr) < 0 )
 	    return;
-#endif DENORM_OPERAND
-	  pop(); return;
-	}
-      if ( st1_tag == TW_NaN )
-	{
-	  if ( !real_2op_NaN(st0_ptr, st1_ptr, st1_ptr) )
-	    pop();
-	  return;
 	}
+
     }
   else if ( st0_tag == TW_NaN )
     {
-      if ( !real_2op_NaN(st0_ptr, st1_ptr, st1_ptr) )
-	pop();
-      return;
+      if ( real_2op_NaN(st0_ptr, st0_tag, 1, st0_ptr) < 0 )
+	return;
     }
   else if ( st0_tag == TW_Infinity )
     {
       if ( st1_tag == TW_NaN )
 	{
-	  if ( !real_2op_NaN(st0_ptr, st1_ptr, st1_ptr) )
-	    pop();
-	  return;
+	  if ( real_2op_NaN(st0_ptr, st0_tag, 1, st0_ptr) < 0 )
+	    return;
 	}
-      else if ( st0_ptr->sign == SIGN_NEG )
+      else if ( signnegative(st0_ptr) )
 	{
-	  int exponent = st1_ptr->exp;
 #ifndef PECULIAR_486
 	  /* This should have higher priority than denormals, but... */
-	  if ( arith_invalid(st1_ptr) )  /* log(-infinity) */
+	  if ( arith_invalid(1) < 0 )  /* log(-infinity) */
 	    return;
 #endif PECULIAR_486
-#ifdef DENORM_OPERAND
-	  if ( st1_tag != TW_Zero )
-	    {
-	      if ( (exponent <= EXP_UNDER) && (denormal_operand()) )
-		return;
-	    }
-#endif DENORM_OPERAND
+	  if ( (st1_tag == TW_Denormal) && (denormal_operand() < 0) )
+	    return;
 #ifdef PECULIAR_486
 	  /* Denormal operands actually get higher priority */
-	  if ( arith_invalid(st1_ptr) )  /* log(-infinity) */
+	  if ( arith_invalid(1) < 0 )  /* log(-infinity) */
 	    return;
 #endif PECULIAR_486
-	  pop();
-	  return;
 	}
-      else if ( st1_tag == TW_Zero )
+      else if ( st1_tag == TAG_Zero )
 	{
 	  /* log(infinity) */
-	  if ( !arith_invalid(st1_ptr) )
-	    pop();
-	  return;
+	  if ( arith_invalid(1) < 0 )
+	    return;
 	}
 	
       /* st(1) must be valid here. */
 
-#ifdef DENORM_OPERAND
-      if ( (st1_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+      else if ( (st1_tag == TW_Denormal) && (denormal_operand() < 0) )
 	return;
-#endif DENORM_OPERAND
 
       /* The Manual says that log(Infinity) is invalid, but a real
 	 80486 sensibly says that it is o.k. */
-      { char sign = st1_ptr->sign;
-	reg_move(&CONST_INF, st1_ptr);
-	st1_ptr->sign = sign;
-      }
-      pop();
-      return;
+      else
+	{
+	  u_char sign = getsign(st1_ptr);
+	  FPU_copy_to_reg1(&CONST_INF, TAG_Special);
+	  setsign(st1_ptr, sign);
+	}
     }
 #ifdef PARANOID
   else
     {
       EXCEPTION(EX_INTERNAL | 0x117);
+      return;
     }
 #endif PARANOID
+
+  FPU_pop();
+  return;
+
 }
 
 
-static void fscale(FPU_REG *st0_ptr)
+static void fscale(FPU_REG *st0_ptr, u_char st0_tag)
 {
-  char st0_tag = st0_ptr->tag;
   FPU_REG *st1_ptr = &st(1);
-  char st1_tag = st1_ptr->tag;
+  u_char st1_tag = FPU_gettagi(1);
   int old_cw = control_word;
-  char sign = st0_ptr->sign;
+  u_char sign = getsign(st0_ptr);
 
   clear_C1();
-  if ( !((st0_tag ^ TW_Valid) | (st1_tag ^ TW_Valid)) )
+  if ( !((st0_tag ^ TAG_Valid) | (st1_tag ^ TAG_Valid)) )
     {
       long scale;
       FPU_REG tmp;
 
-#ifdef DENORM_OPERAND
-      if ( ((st0_ptr->exp <= EXP_UNDER) ||
-	    (st1_ptr->exp <= EXP_UNDER)) && (denormal_operand()) )
-	return;
-#endif DENORM_OPERAND
+      /* Convert register for internal use. */
+      setexponent16(st0_ptr, exponent(st0_ptr));
+
+    valid_scale:
 
-      if ( st1_ptr->exp > EXP_BIAS + 30 )
+      if ( exponent(st1_ptr) > 30 )
 	{
 	  /* 2^31 is far too large, would require 2^(2^30) or 2^(-2^30) */
-	  char sign;
 
-	  if ( st1_ptr->sign == SIGN_POS )
+	  if ( signpositive(st1_ptr) )
 	    {
 	      EXCEPTION(EX_Overflow);
-	      sign = st0_ptr->sign;
-	      reg_move(&CONST_INF, st0_ptr);
-	      st0_ptr->sign = sign;
+	      FPU_copy_to_reg0(&CONST_INF, TAG_Special);
 	    }
 	  else
 	    {
 	      EXCEPTION(EX_Underflow);
-	      sign = st0_ptr->sign;
-	      reg_move(&CONST_Z, st0_ptr);
-	      st0_ptr->sign = sign;
+	      FPU_copy_to_reg0(&CONST_Z, TAG_Zero);
 	    }
+	  setsign(st0_ptr, sign);
 	  return;
 	}
 
       control_word &= ~CW_RC;
       control_word |= RC_CHOP;
-      reg_move(st1_ptr, &tmp);
-      round_to_int(&tmp);               /* This can never overflow here */
+      reg_copy(st1_ptr, &tmp);
+      FPU_round_to_int(&tmp, st1_tag);      /* This can never overflow here */
       control_word = old_cw;
-      scale = st1_ptr->sign ? -tmp.sigl : tmp.sigl;
-      scale += st0_ptr->exp;
-      st0_ptr->exp = scale;
+      scale = signnegative(st1_ptr) ? -tmp.sigl : tmp.sigl;
+      scale += exponent16(st0_ptr);
 
-      /* Use round_reg() to properly detect under/overflow etc */
-      round_reg(st0_ptr, 0, control_word);
+      setexponent16(st0_ptr, scale);
+
+      /* Use FPU_round() to properly detect under/overflow etc */
+      FPU_round(st0_ptr, 0, 0, control_word, sign);
 
       return;
     }
-  else if ( st0_tag == TW_Valid )
+
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+  if ( st1_tag == TAG_Special )
+    st1_tag = FPU_Special(st1_ptr);
+
+  if ( (st0_tag == TAG_Valid) || (st0_tag == TW_Denormal) )
     {
-      if ( st1_tag == TW_Zero )
+      switch ( st1_tag )
 	{
-
-#ifdef DENORM_OPERAND
-	  if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+	case TAG_Valid:
+	  /* st(0) must be a denormal */
+	  if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
 	    return;
-#endif DENORM_OPERAND
 
+	  FPU_to_exp16(st0_ptr, st0_ptr);  /* Will not be left on stack */
+	  goto valid_scale;
+
+	case TAG_Zero:
+	  if ( st0_tag == TW_Denormal )
+	    denormal_operand();
 	  return;
-	}
-      if ( st1_tag == TW_Infinity )
-	{
-#ifdef DENORM_OPERAND
-	  if ( (st0_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
+
+	case TW_Denormal:
+	  denormal_operand();
+	  return;
+
+	case TW_Infinity:
+	  if ( (st0_tag == TW_Denormal) && (denormal_operand() < 0) )
 	    return;
-#endif DENORM_OPERAND
 
-	  if ( st1_ptr->sign == SIGN_POS )
-	    { reg_move(&CONST_INF, st0_ptr); }
+	  if ( signpositive(st1_ptr) )
+	    FPU_copy_to_reg0(&CONST_INF, TAG_Special);
 	  else
-	      reg_move(&CONST_Z, st0_ptr);
-	  st0_ptr->sign = sign;
+	    FPU_copy_to_reg0(&CONST_Z, TAG_Zero);
+	  setsign(st0_ptr, sign);
+	  return;
+
+	case TW_NaN:
+	  real_2op_NaN(st1_ptr, st1_tag, 0, st0_ptr);
 	  return;
 	}
-      if ( st1_tag == TW_NaN )
-	{ real_2op_NaN(st0_ptr, st1_ptr, st0_ptr); return; }
     }
-  else if ( st0_tag == TW_Zero )
+  else if ( st0_tag == TAG_Zero )
     {
-      if ( st1_tag == TW_Valid )
+      switch ( st1_tag )
 	{
+	case TAG_Valid:
+	case TAG_Zero:
+	  return;
 
-#ifdef DENORM_OPERAND
-	  if ( (st1_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
-	    return;
-#endif DENORM_OPERAND
+	case TW_Denormal:
+	  denormal_operand();
+	  return;
 
+	case TW_Infinity:
+	  if ( signpositive(st1_ptr) )
+	    arith_invalid(0); /* Zero scaled by +Infinity */
+	  return;
+
+	case TW_NaN:
+	  real_2op_NaN(st1_ptr, st1_tag, 0, st0_ptr);
 	  return;
 	}
-      else if ( st1_tag == TW_Zero ) { return; }
-      else if ( st1_tag == TW_Infinity )
-	{
-	  if ( st1_ptr->sign == SIGN_NEG )
-	    return;
-	  else
-	    {
-	      arith_invalid(st0_ptr); /* Zero scaled by +Infinity */
-	      return;
-	    }
-	}
-      else if ( st1_tag == TW_NaN )
-	{ real_2op_NaN(st0_ptr, st1_ptr, st0_ptr); return; }
     }
   else if ( st0_tag == TW_Infinity )
     {
-      if ( st1_tag == TW_Valid )
+      switch ( st1_tag )
 	{
+	case TAG_Valid:
+	case TAG_Zero:
+	  return;
 
-#ifdef DENORM_OPERAND
-	  if ( (st1_ptr->exp <= EXP_UNDER) && (denormal_operand()) )
-	    return;
-#endif DENORM_OPERAND
+	case TW_Denormal:
+	  denormal_operand();
+	  return;
 
+	case TW_Infinity:
+	  if ( signnegative(st1_ptr) )
+	    arith_invalid(0); /* Infinity scaled by -Infinity */
 	  return;
-	}
-      if ( ((st1_tag == TW_Infinity) && (st1_ptr->sign == SIGN_POS))
-	  || (st1_tag == TW_Zero) )
-	return;
-      else if ( st1_tag == TW_Infinity )
-	{
-	  arith_invalid(st0_ptr); /* Infinity scaled by -Infinity */
+
+	case TW_NaN:
+	  real_2op_NaN(st1_ptr, st1_tag, 0, st0_ptr);
 	  return;
 	}
-      else if ( st1_tag == TW_NaN )
-	{ real_2op_NaN(st0_ptr, st1_ptr, st0_ptr); return; }
     }
   else if ( st0_tag == TW_NaN )
     {
-      if ( st1_tag != TW_Empty )
-	{ real_2op_NaN(st0_ptr, st1_ptr, st0_ptr); return; }
+      if ( st1_tag != TAG_Empty )
+	{ real_2op_NaN(st1_ptr, st1_tag, 0, st0_ptr); return; }
     }
 
 #ifdef PARANOID
-  if ( !((st0_tag == TW_Empty) || (st1_tag == TW_Empty)) )
+  if ( !((st0_tag == TAG_Empty) || (st1_tag == TAG_Empty)) )
     {
       EXCEPTION(EX_INTERNAL | 0x115);
       return;
@@ -1690,7 +1802,7 @@ static void fscale(FPU_REG *st0_ptr)
 #endif
 
   /* At least one of st(0), st(1) must be empty */
-  stack_underflow();
+  FPU_stack_underflow();
 
 }
 
@@ -1698,21 +1810,22 @@ static void fscale(FPU_REG *st0_ptr)
 /*---------------------------------------------------------------------------*/
 
 static FUNC_ST0 const trig_table_a[] = {
-  f2xm1, fyl2x, fptan, fpatan, fxtract, fprem1, fdecstp, fincstp
+  f2xm1, fyl2x, fptan, fpatan,
+  fxtract, fprem1, (FUNC_ST0)fdecstp, (FUNC_ST0)fincstp
 };
 
-void trig_a(void)
+void FPU_triga(void)
 {
-  (trig_table_a[FPU_rm])(&st(0));
+  (trig_table_a[FPU_rm])(&st(0), FPU_gettag0());
 }
 
 
 static FUNC_ST0 const trig_table_b[] =
   {
-    fprem, fyl2xp1, fsqrt_, fsincos, frndint_, fscale, fsin, fcos
+    fprem, fyl2xp1, fsqrt_, fsincos, frndint_, fscale, (FUNC_ST0)fsin, fcos
   };
 
-void trig_b(void)
+void FPU_trigb(void)
 {
-  (trig_table_b[FPU_rm])(&st(0));
+  (trig_table_b[FPU_rm])(&st(0), FPU_gettag0());
 }
diff --git a/arch/i386/math-emu/get_address.c b/arch/i386/math-emu/get_address.c
index 0749cdc70..799bc1c41 100644
--- a/arch/i386/math-emu/get_address.c
+++ b/arch/i386/math-emu/get_address.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Get the effective address from an FPU instruction.                        |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
+ | Copyright (C) 1992,1993,1994,1997                                         |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ |                       Australia.  E-mail   billm@suburbia.net             |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -41,7 +41,7 @@ static int reg_offset[] = {
 	offsetof(struct info,___edi)
 };
 
-#define REG_(x) (*(long *)(reg_offset[(x)]+(char *) FPU_info))
+#define REG_(x) (*(long *)(reg_offset[(x)]+(u_char *) FPU_info))
 
 static int reg_offset_vm86[] = {
 	offsetof(struct info,___cs),
@@ -54,7 +54,7 @@ static int reg_offset_vm86[] = {
       };
 
 #define VM86_REG_(x) (*(unsigned short *) \
-		      (reg_offset_vm86[((unsigned)x)]+(char *) FPU_info))
+		      (reg_offset_vm86[((unsigned)x)]+(u_char *) FPU_info))
 
 /* These are dummy, fs and gs are not saved on the stack. */
 #define ___FS ___ds
@@ -71,18 +71,18 @@ static int reg_offset_pm[] = {
       };
 
 #define PM_REG_(x) (*(unsigned short *) \
-		      (reg_offset_pm[((unsigned)x)]+(char *) FPU_info))
+		      (reg_offset_pm[((unsigned)x)]+(u_char *) FPU_info))
 
 
 /* Decode the SIB byte. This function assumes mod != 0 */
 static int sib(int mod, unsigned long *fpu_eip)
 {
-  unsigned char ss,index,base;
+  u_char ss,index,base;
   long offset;
 
   RE_ENTRANT_CHECK_OFF;
   FPU_code_verify_area(1);
-  get_user(base, (unsigned char *) (*fpu_eip));   /* The SIB byte */
+  FPU_get_user(base, (u_char *) (*fpu_eip));   /* The SIB byte */
   RE_ENTRANT_CHECK_ON;
   (*fpu_eip)++;
   ss = base >> 6;
@@ -112,7 +112,7 @@ static int sib(int mod, unsigned long *fpu_eip)
       long displacement;
       RE_ENTRANT_CHECK_OFF;
       FPU_code_verify_area(1);
-      get_user(displacement, (signed char *) (*fpu_eip));
+      FPU_get_user(displacement, (signed char *) (*fpu_eip));
       offset += displacement;
       RE_ENTRANT_CHECK_ON;
       (*fpu_eip)++;
@@ -123,7 +123,7 @@ static int sib(int mod, unsigned long *fpu_eip)
       long displacement;
       RE_ENTRANT_CHECK_OFF;
       FPU_code_verify_area(4);
-      get_user(displacement, (signed long *) (*fpu_eip));
+      FPU_get_user(displacement, (long *) (*fpu_eip));
       offset += displacement;
       RE_ENTRANT_CHECK_ON;
       (*fpu_eip) += 4;
@@ -133,7 +133,7 @@ static int sib(int mod, unsigned long *fpu_eip)
 }
 
 
-static unsigned long vm86_segment(unsigned char segment,
+static unsigned long vm86_segment(u_char segment,
 				  unsigned short *selector)
 { 
   segment--;
@@ -150,7 +150,7 @@ static unsigned long vm86_segment(unsigned char segment,
 
 
 /* This should work for 16 and 32 bit protected mode. */
-static long pm_address(unsigned char FPU_modrm, unsigned char segment,
+static long pm_address(u_char FPU_modrm, u_char segment,
 		       unsigned short *selector, long offset)
 { 
   struct desc_struct descriptor;
@@ -233,12 +233,11 @@ static long pm_address(unsigned char FPU_modrm, unsigned char segment,
 
 */
 
-void *get_address(unsigned char FPU_modrm, unsigned long *fpu_eip,
+void *FPU_get_address(u_char FPU_modrm, unsigned long *fpu_eip,
 		  struct address *addr,
-/*		  unsigned short *selector, unsigned long *offset, */
 		  fpu_addr_modes addr_modes)
 {
-  unsigned char mod;
+  u_char mod;
   unsigned rm = FPU_modrm & 7;
   long *cpu_reg_ptr;
   int address = 0;     /* Initialized just to stop compiler warnings. */
@@ -270,7 +269,7 @@ void *get_address(unsigned char FPU_modrm, unsigned long *fpu_eip,
 	      /* Special case: disp32 */
 	      RE_ENTRANT_CHECK_OFF;
 	      FPU_code_verify_area(4);
-	      get_user(address, (unsigned long *) (*fpu_eip));
+	      FPU_get_user(address, (unsigned long *) (*fpu_eip));
 	      (*fpu_eip) += 4;
 	      RE_ENTRANT_CHECK_ON;
 	      addr->offset = address;
@@ -287,7 +286,7 @@ void *get_address(unsigned char FPU_modrm, unsigned long *fpu_eip,
 	  /* 8 bit signed displacement */
 	  RE_ENTRANT_CHECK_OFF;
 	  FPU_code_verify_area(1);
-	  get_user(address, (signed char *) (*fpu_eip));
+	  FPU_get_user(address, (signed char *) (*fpu_eip));
 	  RE_ENTRANT_CHECK_ON;
 	  (*fpu_eip)++;
 	  break;
@@ -295,7 +294,7 @@ void *get_address(unsigned char FPU_modrm, unsigned long *fpu_eip,
 	  /* 32 bit displacement */
 	  RE_ENTRANT_CHECK_OFF;
 	  FPU_code_verify_area(4);
-	  get_user(address, (long *) (*fpu_eip));
+	  FPU_get_user(address, (long *) (*fpu_eip));
 	  (*fpu_eip) += 4;
 	  RE_ENTRANT_CHECK_ON;
 	  break;
@@ -329,12 +328,11 @@ void *get_address(unsigned char FPU_modrm, unsigned long *fpu_eip,
 }
 
 
-void *get_address_16(unsigned char FPU_modrm, unsigned long *fpu_eip,
+void *FPU_get_address_16(u_char FPU_modrm, unsigned long *fpu_eip,
 		     struct address *addr,
-/*		     unsigned short *selector, unsigned long *offset, */
 		     fpu_addr_modes addr_modes)
 {
-  unsigned char mod;
+  u_char mod;
   unsigned rm = FPU_modrm & 7;
   int address = 0;     /* Default used for mod == 0 */
 
@@ -358,7 +356,7 @@ void *get_address_16(unsigned char FPU_modrm, unsigned long *fpu_eip,
 	  /* Special case: disp16 */
 	  RE_ENTRANT_CHECK_OFF;
 	  FPU_code_verify_area(2);
-	  get_user(address, (unsigned short *) (*fpu_eip));
+	  FPU_get_user(address, (unsigned short *) (*fpu_eip));
 	  (*fpu_eip) += 2;
 	  RE_ENTRANT_CHECK_ON;
 	  goto add_segment;
@@ -368,7 +366,7 @@ void *get_address_16(unsigned char FPU_modrm, unsigned long *fpu_eip,
       /* 8 bit signed displacement */
       RE_ENTRANT_CHECK_OFF;
       FPU_code_verify_area(1);
-      get_user(address, (signed char *) (*fpu_eip));
+      FPU_get_user(address, (signed char *) (*fpu_eip));
       RE_ENTRANT_CHECK_ON;
       (*fpu_eip)++;
       break;
@@ -376,7 +374,7 @@ void *get_address_16(unsigned char FPU_modrm, unsigned long *fpu_eip,
       /* 16 bit displacement */
       RE_ENTRANT_CHECK_OFF;
       FPU_code_verify_area(2);
-      get_user(address, (unsigned short *) (*fpu_eip));
+      FPU_get_user(address, (unsigned short *) (*fpu_eip));
       (*fpu_eip) += 2;
       RE_ENTRANT_CHECK_ON;
       break;
diff --git a/arch/i386/math-emu/load_store.c b/arch/i386/math-emu/load_store.c
index cc288a9ab..4ed4bb7a6 100644
--- a/arch/i386/math-emu/load_store.c
+++ b/arch/i386/math-emu/load_store.c
@@ -4,9 +4,9 @@
  | This file contains most of the code to interpret the FPU instructions     |
  | which load and store from user memory.                                    |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
+ | Copyright (C) 1992,1993,1994,1997                                         |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ |                       Australia.  E-mail   billm@suburbia.net             |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -32,10 +32,10 @@
 #define _PUSH_ 3   /* Need to check for space to push onto stack */
 #define _null_ 4   /* Function illegal or not implemented */
 
-#define pop_0()	{ st0_ptr->tag = TW_Empty; top++; }
+#define pop_0()	{ FPU_settag0(TAG_Empty); top++; }
 
 
-static unsigned char const type_table[32] = {
+static u_char const type_table[32] = {
   _PUSH_, _PUSH_, _PUSH_, _PUSH_,
   _null_, _null_, _null_, _null_,
   _REG0_, _REG0_, _REG0_, _REG0_,
@@ -46,25 +46,27 @@ static unsigned char const type_table[32] = {
   _NONE_, _REG0_, _NONE_, _REG0_
   };
 
-unsigned char const data_sizes_16[32] = {
+u_char const data_sizes_16[32] = {
   4,  4,  8,  2,  0,  0,  0,  0,
   4,  4,  8,  2,  4,  4,  8,  2,
   14, 0, 94, 10,  2, 10,  0,  8,  
   14, 0, 94, 10,  2, 10,  2,  8
 };
 
-unsigned char const data_sizes_32[32] = {
+u_char const data_sizes_32[32] = {
   4,  4,  8,  2,  0,  0,  0,  0,
   4,  4,  8,  2,  4,  4,  8,  2,
   28, 0,108, 10,  2, 10,  0,  8,  
   28, 0,108, 10,  2, 10,  2,  8
 };
 
-int load_store_instr(unsigned char type, fpu_addr_modes addr_modes,
+int FPU_load_store(u_char type, fpu_addr_modes addr_modes,
 		     void *data_address)
 {
   FPU_REG loaded_data;
   FPU_REG *st0_ptr;
+  u_char st0_tag = TAG_Empty;  /* This is just to stop a gcc warning. */
+  u_char loaded_tag;
 
   st0_ptr = NULL;    /* Initialized just to stop compiler warnings. */
 
@@ -93,13 +95,14 @@ int load_store_instr(unsigned char type, fpu_addr_modes addr_modes,
     case _REG0_:
       st0_ptr = &st(0);       /* Some of these instructions pop after
 				 storing */
+      st0_tag = FPU_gettag0();
       break;
     case _PUSH_:
       {
-	st0_ptr = &st(-1);
-	if ( st0_ptr->tag != TW_Empty )
-	  { stack_overflow(); return 0; }
+	if ( FPU_gettagi(-1) != TAG_Empty )
+	  { FPU_stack_overflow(); return 0; }
 	top--;
+	st0_ptr = &st(0);
       }
       break;
     case _null_:
@@ -116,92 +119,97 @@ int load_store_instr(unsigned char type, fpu_addr_modes addr_modes,
     {
     case 000:       /* fld m32real */
       clear_C1();
-      reg_load_single((float *)data_address, &loaded_data);
-      if ( (loaded_data.tag == TW_NaN) &&
-	  real_2op_NaN(&loaded_data, &loaded_data, &loaded_data) )
+      loaded_tag = FPU_load_single((float *)data_address, &loaded_data);
+      if ( (loaded_tag == TAG_Special)
+	   && isNaN(&loaded_data)
+	   && (real_1op_NaN(&loaded_data) < 0) )
 	{
 	  top++;
 	  break;
 	}
-      reg_move(&loaded_data, st0_ptr);
+      FPU_copy_to_reg0(&loaded_data, loaded_tag);
       break;
     case 001:      /* fild m32int */
       clear_C1();
-      reg_load_int32((long *)data_address, st0_ptr);
+      loaded_tag = FPU_load_int32((long *)data_address, &loaded_data);
+      FPU_copy_to_reg0(&loaded_data, loaded_tag);
       break;
     case 002:      /* fld m64real */
       clear_C1();
-      reg_load_double((double *)data_address, &loaded_data);
-      if ( (loaded_data.tag == TW_NaN) &&
-	  real_2op_NaN(&loaded_data, &loaded_data, &loaded_data) )
+      loaded_tag = FPU_load_double((double *)data_address, &loaded_data);
+      if ( (loaded_tag == TAG_Special)
+	   && isNaN(&loaded_data)
+	   && (real_1op_NaN(&loaded_data) < 0) )
 	{
 	  top++;
 	  break;
 	}
-      reg_move(&loaded_data, st0_ptr);
+      FPU_copy_to_reg0(&loaded_data, loaded_tag);
       break;
     case 003:      /* fild m16int */
       clear_C1();
-      reg_load_int16((short *)data_address, st0_ptr);
+      loaded_tag = FPU_load_int16((short *)data_address, &loaded_data);
+      FPU_copy_to_reg0(&loaded_data, loaded_tag);
       break;
     case 010:      /* fst m32real */
       clear_C1();
-      reg_store_single((float *)data_address, st0_ptr);
+      FPU_store_single(st0_ptr, st0_tag, (float *)data_address);
       break;
     case 011:      /* fist m32int */
       clear_C1();
-      reg_store_int32((long *)data_address, st0_ptr);
+      FPU_store_int32(st0_ptr, st0_tag, (long *)data_address);
       break;
     case 012:     /* fst m64real */
       clear_C1();
-      reg_store_double((double *)data_address, st0_ptr);
+      FPU_store_double(st0_ptr, st0_tag, (double *)data_address);
       break;
     case 013:     /* fist m16int */
       clear_C1();
-      reg_store_int16((short *)data_address, st0_ptr);
+      FPU_store_int16(st0_ptr, st0_tag, (short *)data_address);
       break;
     case 014:     /* fstp m32real */
       clear_C1();
-      if ( reg_store_single((float *)data_address, st0_ptr) )
+      if ( FPU_store_single(st0_ptr, st0_tag, (float *)data_address) )
 	pop_0();  /* pop only if the number was actually stored
 		     (see the 80486 manual p16-28) */
       break;
     case 015:     /* fistp m32int */
       clear_C1();
-      if ( reg_store_int32((long *)data_address, st0_ptr) )
+      if ( FPU_store_int32(st0_ptr, st0_tag, (long *)data_address) )
 	pop_0();  /* pop only if the number was actually stored
 		     (see the 80486 manual p16-28) */
       break;
     case 016:     /* fstp m64real */
       clear_C1();
-      if ( reg_store_double((double *)data_address, st0_ptr) )
+      if ( FPU_store_double(st0_ptr, st0_tag, (double *)data_address) )
 	pop_0();  /* pop only if the number was actually stored
 		     (see the 80486 manual p16-28) */
       break;
     case 017:     /* fistp m16int */
       clear_C1();
-      if ( reg_store_int16((short *)data_address, st0_ptr) )
+      if ( FPU_store_int16(st0_ptr, st0_tag, (short *)data_address) )
 	pop_0();  /* pop only if the number was actually stored
 		     (see the 80486 manual p16-28) */
       break;
     case 020:     /* fldenv  m14/28byte */
-      fldenv(addr_modes, (char *)data_address);
+      fldenv(addr_modes, (u_char *)data_address);
       /* Ensure that the values just loaded are not changed by
 	 fix-up operations. */
       return 1;
     case 022:     /* frstor m94/108byte */
-      frstor(addr_modes, (char *)data_address);
+      frstor(addr_modes, (u_char *)data_address);
       /* Ensure that the values just loaded are not changed by
 	 fix-up operations. */
       return 1;
     case 023:     /* fbld m80dec */
       clear_C1();
-      reg_load_bcd((char *)data_address, st0_ptr);
+      loaded_tag = FPU_load_bcd((u_char *)data_address);
+      FPU_settag0(loaded_tag);
       break;
     case 024:     /* fldcw */
       RE_ENTRANT_CHECK_OFF;
       FPU_verify_area(VERIFY_READ, data_address, 2);
-      get_user(control_word, (unsigned short *) data_address);
+      FPU_get_user(control_word, (unsigned short *) data_address);
       RE_ENTRANT_CHECK_ON;
       if ( partial_status & ~control_word & CW_Exceptions )
 	partial_status |= (SW_Summary | SW_Backward);
@@ -213,45 +221,47 @@ int load_store_instr(unsigned char type, fpu_addr_modes addr_modes,
       return 1;
     case 025:      /* fld m80real */
       clear_C1();
-      reg_load_extended((long double *)data_address, st0_ptr);
+      loaded_tag = FPU_load_extended((long double *)data_address, 0);
+      FPU_settag0(loaded_tag);
       break;
     case 027:      /* fild m64int */
       clear_C1();
-      reg_load_int64((long long *)data_address, st0_ptr);
+      loaded_tag = FPU_load_int64((long long *)data_address);
+      FPU_settag0(loaded_tag);
       break;
     case 030:     /* fstenv  m14/28byte */
-      fstenv(addr_modes, (char *)data_address);
+      fstenv(addr_modes, (u_char *)data_address);
       return 1;
     case 032:      /* fsave */
-      fsave(addr_modes, (char *)data_address);
+      fsave(addr_modes, (u_char *)data_address);
       return 1;
     case 033:      /* fbstp m80dec */
       clear_C1();
-      if ( reg_store_bcd((char *)data_address, st0_ptr) )
+      if ( FPU_store_bcd(st0_ptr, st0_tag, (u_char *)data_address) )
 	pop_0();  /* pop only if the number was actually stored
 		     (see the 80486 manual p16-28) */
       break;
     case 034:      /* fstcw m16int */
       RE_ENTRANT_CHECK_OFF;
       FPU_verify_area(VERIFY_WRITE,data_address,2);
-      put_user(control_word, (unsigned short *) data_address);
+      FPU_put_user(control_word, (unsigned short *) data_address);
       RE_ENTRANT_CHECK_ON;
       return 1;
     case 035:      /* fstp m80real */
       clear_C1();
-      if ( reg_store_extended((long double *)data_address, st0_ptr) )
+      if ( FPU_store_extended(st0_ptr, st0_tag, (long double *)data_address) )
 	pop_0();  /* pop only if the number was actually stored
 		     (see the 80486 manual p16-28) */
       break;
     case 036:      /* fstsw m2byte */
       RE_ENTRANT_CHECK_OFF;
       FPU_verify_area(VERIFY_WRITE,data_address,2);
-      put_user(status_word(),(unsigned short *) data_address);
+      FPU_put_user(status_word(),(unsigned short *) data_address);
       RE_ENTRANT_CHECK_ON;
       return 1;
     case 037:      /* fistp m64int */
       clear_C1();
-      if ( reg_store_int64((long long *)data_address, st0_ptr) )
+      if ( FPU_store_int64(st0_ptr, st0_tag, (long long *)data_address) )
 	pop_0();  /* pop only if the number was actually stored
 		     (see the 80486 manual p16-28) */
       break;
diff --git a/arch/i386/math-emu/poly_2xm1.c b/arch/i386/math-emu/poly_2xm1.c
index f7c585d60..51fb4481c 100644
--- a/arch/i386/math-emu/poly_2xm1.c
+++ b/arch/i386/math-emu/poly_2xm1.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Function to compute 2^x-1 by a polynomial approximation.                  |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1994,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -13,6 +13,7 @@
 #include "exception.h"
 #include "reg_constant.h"
 #include "fpu_emu.h"
+#include "fpu_system.h"
 #include "control_w.h"
 #include "poly.h"
 
@@ -48,20 +49,19 @@ static const Xsig *shiftterm[] = { &shiftterm0, &shiftterm1,
 
 
 /*--- poly_2xm1() -----------------------------------------------------------+
- | Requires an argument which is TW_Valid and < 1.                           |
+ | Requires st(0) which is TAG_Valid and < 1.                                |
  +---------------------------------------------------------------------------*/
-int	poly_2xm1(FPU_REG const *arg, FPU_REG *result)
+int	poly_2xm1(u_char sign, FPU_REG *arg, FPU_REG *result)
 {
-  long int               exponent, shift;
-  unsigned long long     Xll;
-  Xsig                   accumulator, Denom, argSignif;
+  long int              exponent, shift;
+  unsigned long long    Xll;
+  Xsig                  accumulator, Denom, argSignif;
+  u_char                tag;
 
-
-  exponent = arg->exp - EXP_BIAS;
+  exponent = exponent16(arg);
 
 #ifdef PARANOID
-  if (   (exponent >= 0)    	/* Don't want a |number| >= 1.0 */
-      || (arg->tag != TW_Valid) )
+  if ( exponent >= 0 )    	/* Don't want a |number| >= 1.0 */
     {
       /* Number negative, too large, or not Valid. */
       EXCEPTION(EX_INTERNAL|0x127);
@@ -94,7 +94,7 @@ int	poly_2xm1(FPU_REG const *arg, FPU_REG *result)
   if ( exponent < -2 )
     {
       /* Shift the argument right by the required places. */
-      if ( shrx(&Xll, -2-exponent) >= 0x80000000U )
+      if ( FPU_shrx(&Xll, -2-exponent) >= 0x80000000U )
 	Xll++;	/* round up */
     }
 
@@ -118,7 +118,7 @@ int	poly_2xm1(FPU_REG const *arg, FPU_REG *result)
       exponent = 1;
     }
 
-  if ( arg->sign != SIGN_POS )
+  if ( sign != SIGN_POS )
     {
       /* The argument is negative, use the identity:
 	     f(-x) = -f(x) / (1 + f(x))
@@ -142,10 +142,14 @@ int	poly_2xm1(FPU_REG const *arg, FPU_REG *result)
   /* Convert to 64 bit signed-compatible */
   exponent += round_Xsig(&accumulator);
 
+  result = &st(0);
   significand(result) = XSIG_LL(accumulator);
-  result->tag = TW_Valid;
-  result->exp = exponent + EXP_BIAS;
-  result->sign = arg->sign;
+  setexponent16(result, exponent);
+
+  tag = FPU_round(result, 1, 0, FULL_PRECISION, sign);
+
+  setsign(result, sign);
+  FPU_settag0(tag);
 
   return 0;
 
diff --git a/arch/i386/math-emu/poly_atan.c b/arch/i386/math-emu/poly_atan.c
index 6edca625f..a5d5af882 100644
--- a/arch/i386/math-emu/poly_atan.c
+++ b/arch/i386/math-emu/poly_atan.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Compute the arctan of a FPU_REG, using a polynomial approximation.        |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1994,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -13,6 +13,7 @@
 #include "exception.h"
 #include "reg_constant.h"
 #include "fpu_emu.h"
+#include "fpu_system.h"
 #include "status_w.h"
 #include "control_w.h"
 #include "poly.h"
@@ -51,31 +52,57 @@ static const Xsig pi_signif = MK_XSIG(0xc90fdaa2, 0x2168c234, 0xc4c6628b);
 /*--- poly_atan() -----------------------------------------------------------+
  |                                                                           |
  +---------------------------------------------------------------------------*/
-void	poly_atan(FPU_REG *arg1, FPU_REG *arg2, FPU_REG *result)
+void	poly_atan(FPU_REG *st0_ptr, u_char st0_tag,
+		  FPU_REG *st1_ptr, u_char st1_tag)
 {
-  char		        transformed, inverted,
-                        sign1 = arg1->sign, sign2 = arg2->sign;
-  long int   		exponent, dummy_exp;
-  Xsig                  accumulator, Numer, Denom, accumulatore, argSignif,
-                        argSq, argSqSq;
+  u_char	transformed, inverted,
+                sign1, sign2;
+  int           exponent;
+  long int   	dummy_exp;
+  Xsig          accumulator, Numer, Denom, accumulatore, argSignif,
+                argSq, argSqSq;
+  u_char        tag;
   
+  sign1 = getsign(st0_ptr);
+  sign2 = getsign(st1_ptr);
+  if ( st0_tag == TAG_Valid )
+    {
+      exponent = exponent(st0_ptr);
+    }
+  else
+    {
+      /* This gives non-compatible stack contents... */
+      FPU_to_exp16(st0_ptr, st0_ptr);
+      exponent = exponent16(st0_ptr);
+    }
+  if ( st1_tag == TAG_Valid )
+    {
+      exponent -= exponent(st1_ptr);
+    }
+  else
+    {
+      /* This gives non-compatible stack contents... */
+      FPU_to_exp16(st1_ptr, st1_ptr);
+      exponent -= exponent16(st1_ptr);
+    }
 
-  arg1->sign = arg2->sign = SIGN_POS;
-  if ( (compare(arg2) & ~COMP_Denormal) == COMP_A_lt_B )
+  if ( (exponent < 0) || ((exponent == 0) &&
+			  ((st0_ptr->sigh < st1_ptr->sigh) ||
+			   ((st0_ptr->sigh == st1_ptr->sigh) &&
+			    (st0_ptr->sigl < st1_ptr->sigl))) ) )
     {
       inverted = 1;
-      exponent = arg1->exp - arg2->exp;
       Numer.lsw = Denom.lsw = 0;
-      XSIG_LL(Numer) = significand(arg1);
-      XSIG_LL(Denom) = significand(arg2);
+      XSIG_LL(Numer) = significand(st0_ptr);
+      XSIG_LL(Denom) = significand(st1_ptr);
     }
   else
     {
       inverted = 0;
-      exponent = arg2->exp - arg1->exp;
+      exponent = -exponent;
       Numer.lsw = Denom.lsw = 0;
-      XSIG_LL(Numer) = significand(arg2);
-      XSIG_LL(Denom) = significand(arg1);
+      XSIG_LL(Numer) = significand(st1_ptr);
+      XSIG_LL(Denom) = significand(st0_ptr);
      }
   div_Xsig(&Numer, &Denom, &argSignif);
   exponent += norm_Xsig(&argSignif);
@@ -189,9 +216,14 @@ void	poly_atan(FPU_REG *arg1, FPU_REG *arg2, FPU_REG *result)
     }
 
   exponent += round_Xsig(&accumulator);
-  significand(result) = XSIG_LL(accumulator);
-  result->exp = exponent + EXP_BIAS;
-  result->tag = TW_Valid;
-  result->sign = sign2;
+
+  significand(st1_ptr) = XSIG_LL(accumulator);
+  setexponent16(st1_ptr, exponent);
+
+  tag = FPU_round(st1_ptr, 1, 0, FULL_PRECISION, sign2);
+  FPU_settagi(1, tag);
+
+  set_precision_flag_up();  /* We do not really know if up or down,
+			       use this as the default. */
 
 }
diff --git a/arch/i386/math-emu/poly_l2.c b/arch/i386/math-emu/poly_l2.c
index 1677f4aff..07b2da4b9 100644
--- a/arch/i386/math-emu/poly_l2.c
+++ b/arch/i386/math-emu/poly_l2.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Compute the base 2 log of a FPU_REG, using a polynomial approximation.    |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1994,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -14,96 +14,101 @@
 #include "exception.h"
 #include "reg_constant.h"
 #include "fpu_emu.h"
+#include "fpu_system.h"
 #include "control_w.h"
 #include "poly.h"
 
 
-
-static void log2_kernel(FPU_REG const *arg,
+static void log2_kernel(FPU_REG const *arg, u_char argsign,
 			Xsig *accum_result, long int *expon);
 
 
 /*--- poly_l2() -------------------------------------------------------------+
  |   Base 2 logarithm by a polynomial approximation.                         |
  +---------------------------------------------------------------------------*/
-void	poly_l2(FPU_REG const *arg, FPU_REG const *y, FPU_REG *result)
+void	poly_l2(FPU_REG *st0_ptr, FPU_REG *st1_ptr, u_char st1_sign)
 {
   long int	       exponent, expon, expon_expon;
   Xsig                 accumulator, expon_accum, yaccum;
-  char		       sign;
+  u_char		       sign, argsign;
   FPU_REG              x;
+  int                  tag;
 
+  exponent = exponent16(st0_ptr);
 
-  exponent = arg->exp - EXP_BIAS;
-
-  /* From arg, make a number > sqrt(2)/2 and < sqrt(2) */
-  if ( arg->sigh > (unsigned)0xb504f334 )
+  /* From st0_ptr, make a number > sqrt(2)/2 and < sqrt(2) */
+  if ( st0_ptr->sigh > (unsigned)0xb504f334 )
     {
-      /* Treat as  sqrt(2)/2 < arg < 1 */
-      significand(&x) = - significand(arg);
-      x.sign = SIGN_NEG;
-      x.tag = TW_Valid;
-      x.exp = EXP_BIAS-1;
+      /* Treat as  sqrt(2)/2 < st0_ptr < 1 */
+      significand(&x) = - significand(st0_ptr);
+      setexponent16(&x, -1);
       exponent++;
-      normalize(&x);
+      argsign = SIGN_NEG;
     }
   else
     {
-      /* Treat as  1 <= arg < sqrt(2) */
-      x.sigh = arg->sigh - 0x80000000;
-      x.sigl = arg->sigl;
-      x.sign = SIGN_POS;
-      x.tag = TW_Valid;
-      x.exp = EXP_BIAS;
-      normalize(&x);
+      /* Treat as  1 <= st0_ptr < sqrt(2) */
+      x.sigh = st0_ptr->sigh - 0x80000000;
+      x.sigl = st0_ptr->sigl;
+      setexponent16(&x, 0);
+      argsign = SIGN_POS;
     }
+  tag = FPU_normalize_nuo(&x);
 
-  if ( x.tag == TW_Zero )
+  if ( tag == TAG_Zero )
     {
       expon = 0;
       accumulator.msw = accumulator.midw = accumulator.lsw = 0;
     }
   else
     {
-      log2_kernel(&x, &accumulator, &expon);
+      log2_kernel(&x, argsign, &accumulator, &expon);
     }
 
-  sign = exponent < 0;
-  if ( sign ) exponent = -exponent;
+  if ( exponent < 0 )
+    {
+      sign = SIGN_NEG;
+      exponent = -exponent;
+    }
+  else
+    sign = SIGN_POS;
   expon_accum.msw = exponent; expon_accum.midw = expon_accum.lsw = 0;
   if ( exponent )
     {
       expon_expon = 31 + norm_Xsig(&expon_accum);
       shr_Xsig(&accumulator, expon_expon - expon);
 
-      if ( sign ^ (x.sign == SIGN_NEG) )
+      if ( sign ^ argsign )
 	negate_Xsig(&accumulator);
       add_Xsig_Xsig(&accumulator, &expon_accum);
     }
   else
     {
       expon_expon = expon;
-      sign = x.sign;
+      sign = argsign;
     }
 
-  yaccum.lsw = 0; XSIG_LL(yaccum) = significand(y);
+  yaccum.lsw = 0; XSIG_LL(yaccum) = significand(st1_ptr);
   mul_Xsig_Xsig(&accumulator, &yaccum);
 
   expon_expon += round_Xsig(&accumulator);
 
   if ( accumulator.msw == 0 )
     {
-      reg_move(&CONST_Z, y);
-    }
-  else
-    {
-      result->exp = expon_expon + y->exp + 1;
-      significand(result) = XSIG_LL(accumulator);
-      result->tag = TW_Valid; /* set the tags to Valid */
-      result->sign = sign ^ y->sign;
+      FPU_copy_to_reg1(&CONST_Z, TAG_Zero);
+      return;
     }
 
+  significand(st1_ptr) = XSIG_LL(accumulator);
+  setexponent16(st1_ptr, expon_expon + exponent16(st1_ptr) + 1);
+
+  tag = FPU_round(st1_ptr, 1, 0, FULL_PRECISION, sign ^ st1_sign);
+  FPU_settagi(1, tag);
+
+  set_precision_flag_up();  /* 80486 appears to always do this */
+
   return;
+
 }
 
 
@@ -111,47 +116,62 @@ void	poly_l2(FPU_REG const *arg, FPU_REG const *y, FPU_REG *result)
  |   Base 2 logarithm by a polynomial approximation.                         |
  |   log2(x+1)                                                               |
  +---------------------------------------------------------------------------*/
-int	poly_l2p1(FPU_REG const *arg, FPU_REG const *y, FPU_REG *result)
+int	poly_l2p1(u_char sign0, u_char sign1,
+		  FPU_REG *st0_ptr, FPU_REG *st1_ptr, FPU_REG *dest)
 {
-  char                 sign;
-  long int             exponent;
-  Xsig                 accumulator, yaccum;
+  u_char             	tag;
+  long int        	exponent;
+  Xsig              	accumulator, yaccum;
 
-
-  sign = arg->sign;
-
-  if ( arg->exp < EXP_BIAS )
+  if ( exponent16(st0_ptr) < 0 )
     {
-      log2_kernel(arg, &accumulator, &exponent);
+      log2_kernel(st0_ptr, sign0, &accumulator, &exponent);
 
       yaccum.lsw = 0;
-      XSIG_LL(yaccum) = significand(y);
+      XSIG_LL(yaccum) = significand(st1_ptr);
       mul_Xsig_Xsig(&accumulator, &yaccum);
 
       exponent += round_Xsig(&accumulator);
 
-      result->exp = exponent + y->exp + 1;
-      significand(result) = XSIG_LL(accumulator);
-      result->tag = TW_Valid; /* set the tags to Valid */
-      result->sign = sign ^ y->sign;
+      exponent += exponent16(st1_ptr) + 1;
+      if ( exponent < EXP_WAY_UNDER ) exponent = EXP_WAY_UNDER;
+
+      significand(dest) = XSIG_LL(accumulator);
+      setexponent16(dest, exponent);
 
-      return 0;
+      tag = FPU_round(dest, 1, 0, FULL_PRECISION, sign0 ^ sign1);
+      FPU_settagi(1, tag);
+
+      if ( tag == TAG_Valid )
+	set_precision_flag_up();   /* 80486 appears to always do this */
     }
   else
     {
-      /* The magnitude of arg is far too large. */
-      reg_move(y, result);
-      if ( sign != SIGN_POS )
+      /* The magnitude of st0_ptr is far too large. */
+
+      if ( sign0 != SIGN_POS )
 	{
 	  /* Trying to get the log of a negative number. */
-	  return 1;
+#ifdef PECULIAR_486   /* Stupid 80486 doesn't worry about log(negative). */
+	  changesign(st1_ptr);
+#else
+	  if ( arith_invalid(1) < 0 )
+	    return 1;
+#endif PECULIAR_486
 	}
+
+      /* 80486 appears to do this */
+      if ( sign0 == SIGN_NEG )
+	set_precision_flag_down();
       else
-	{
-	  return 0;
-	}
+	set_precision_flag_up();
     }
 
+  if ( exponent(dest) <= EXP_UNDER )
+    EXCEPTION(EX_Underflow);
+
+  return 0;
+
 }
 
 
@@ -180,20 +200,17 @@ static const unsigned long leadterm = 0xb8000000;
  |   Base 2 logarithm by a polynomial approximation.                         |
  |   log2(x+1)                                                               |
  +---------------------------------------------------------------------------*/
-static void log2_kernel(FPU_REG const *arg, Xsig *accum_result,
+static void log2_kernel(FPU_REG const *arg, u_char argsign, Xsig *accum_result,
 			long int *expon)
 {
-  char                 sign;
   long int             exponent, adj;
   unsigned long long   Xsq;
   Xsig                 accumulator, Numer, Denom, argSignif, arg_signif;
 
-  sign = arg->sign;
-
-  exponent = arg->exp - EXP_BIAS;
+  exponent = exponent16(arg);
   Numer.lsw = Denom.lsw = 0;
   XSIG_LL(Numer) = XSIG_LL(Denom) = significand(arg);
-  if ( sign == SIGN_POS )
+  if ( argsign == SIGN_POS )
     {
       shr_Xsig(&Denom, 2 - (1 + exponent));
       Denom.msw |= 0x80000000;
diff --git a/arch/i386/math-emu/poly_sin.c b/arch/i386/math-emu/poly_sin.c
index 03db5b6aa..f03df4c00 100644
--- a/arch/i386/math-emu/poly_sin.c
+++ b/arch/i386/math-emu/poly_sin.c
@@ -4,9 +4,9 @@
  |  Computation of an approximation of the sin function and the cosine       |
  |  function by a polynomial.                                                |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1994,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -15,6 +15,7 @@
 #include "exception.h"
 #include "reg_constant.h"
 #include "fpu_emu.h"
+#include "fpu_system.h"
 #include "control_w.h"
 #include "poly.h"
 
@@ -62,35 +63,26 @@ static const unsigned long long neg_terms_h[N_COEFF_NH] =
 /*--- poly_sine() -----------------------------------------------------------+
  |                                                                           |
  +---------------------------------------------------------------------------*/
-void	poly_sine(FPU_REG const *arg, FPU_REG *result)
+void	poly_sine(FPU_REG *st0_ptr)
 {
   int                 exponent, echange;
   Xsig                accumulator, argSqrd, argTo4;
   unsigned long       fix_up, adj;
   unsigned long long  fixed_arg;
+  FPU_REG	      result;
 
-
-#ifdef PARANOID
-  if ( arg->tag == TW_Zero )
-    {
-      /* Return 0.0 */
-      reg_move(&CONST_Z, result);
-      return;
-    }
-#endif PARANOID
-
-  exponent = arg->exp - EXP_BIAS;
+  exponent = exponent(st0_ptr);
 
   accumulator.lsw = accumulator.midw = accumulator.msw = 0;
 
   /* Split into two ranges, for arguments below and above 1.0 */
   /* The boundary between upper and lower is approx 0.88309101259 */
-  if ( (exponent < -1) || ((exponent == -1) && (arg->sigh <= 0xe21240aa)) )
+  if ( (exponent < -1) || ((exponent == -1) && (st0_ptr->sigh <= 0xe21240aa)) )
     {
       /* The argument is <= 0.88309101259 */
 
-      argSqrd.msw = arg->sigh; argSqrd.midw = arg->sigl; argSqrd.lsw = 0;
-      mul64_Xsig(&argSqrd, &significand(arg));
+      argSqrd.msw = st0_ptr->sigh; argSqrd.midw = st0_ptr->sigl; argSqrd.lsw = 0;
+      mul64_Xsig(&argSqrd, &significand(st0_ptr));
       shr_Xsig(&argSqrd, 2*(-1-exponent));
       argTo4.msw = argSqrd.msw; argTo4.midw = argSqrd.midw;
       argTo4.lsw = argSqrd.lsw;
@@ -107,29 +99,29 @@ void	poly_sine(FPU_REG const *arg, FPU_REG *result)
       shr_Xsig(&accumulator, 2);    /* Divide by four */
       accumulator.msw |= 0x80000000;  /* Add 1.0 */
 
-      mul64_Xsig(&accumulator, &significand(arg));
-      mul64_Xsig(&accumulator, &significand(arg));
-      mul64_Xsig(&accumulator, &significand(arg));
+      mul64_Xsig(&accumulator, &significand(st0_ptr));
+      mul64_Xsig(&accumulator, &significand(st0_ptr));
+      mul64_Xsig(&accumulator, &significand(st0_ptr));
 
       /* Divide by four, FPU_REG compatible, etc */
-      exponent = 3*exponent + EXP_BIAS;
+      exponent = 3*exponent;
 
       /* The minimum exponent difference is 3 */
-      shr_Xsig(&accumulator, arg->exp - exponent);
+      shr_Xsig(&accumulator, exponent(st0_ptr) - exponent);
 
       negate_Xsig(&accumulator);
-      XSIG_LL(accumulator) += significand(arg);
+      XSIG_LL(accumulator) += significand(st0_ptr);
 
       echange = round_Xsig(&accumulator);
 
-      result->exp = arg->exp + echange;
+      setexponentpos(&result, exponent(st0_ptr) + echange);
     }
   else
     {
       /* The argument is > 0.88309101259 */
-      /* We use sin(arg) = cos(pi/2-arg) */
+      /* We use sin(st(0)) = cos(pi/2-st(0)) */
 
-      fixed_arg = significand(arg);
+      fixed_arg = significand(st0_ptr);
 
       if ( exponent == 0 )
 	{
@@ -192,16 +184,16 @@ void	poly_sine(FPU_REG const *arg, FPU_REG *result)
 
       echange = round_Xsig(&accumulator);
 
-      result->exp = EXP_BIAS - 1 + echange;
+      setexponentpos(&result, echange - 1);
     }
 
-  significand(result) = XSIG_LL(accumulator);
-  result->tag = TW_Valid;
-  result->sign = arg->sign;
+  significand(&result) = XSIG_LL(accumulator);
+  setsign(&result, getsign(st0_ptr));
+  FPU_copy_to_reg0(&result, TAG_Valid);
 
 #ifdef PARANOID
-  if ( (result->exp >= EXP_BIAS)
-      && (significand(result) > 0x8000000000000000LL) )
+  if ( (exponent(&result) >= 0)
+      && (significand(&result) > 0x8000000000000000LL) )
     {
       EXCEPTION(EX_INTERNAL|0x150);
     }
@@ -214,42 +206,36 @@ void	poly_sine(FPU_REG const *arg, FPU_REG *result)
 /*--- poly_cos() ------------------------------------------------------------+
  |                                                                           |
  +---------------------------------------------------------------------------*/
-void	poly_cos(FPU_REG const *arg, FPU_REG *result)
+void	poly_cos(FPU_REG *st0_ptr)
 {
+  FPU_REG	      result;
   long int            exponent, exp2, echange;
   Xsig                accumulator, argSqrd, fix_up, argTo4;
   unsigned long       adj;
   unsigned long long  fixed_arg;
 
-
 #ifdef PARANOID
-  if ( arg->tag == TW_Zero )
-    {
-      /* Return 1.0 */
-      reg_move(&CONST_1, result);
-      return;
-    }
-
-  if ( (arg->exp > EXP_BIAS)
-      || ((arg->exp == EXP_BIAS)
-	  && (significand(arg) > 0xc90fdaa22168c234LL)) )
+  if ( (exponent(st0_ptr) > 0)
+      || ((exponent(st0_ptr) == 0)
+	  && (significand(st0_ptr) > 0xc90fdaa22168c234LL)) )
     {
       EXCEPTION(EX_Invalid);
-      reg_move(&CONST_QNaN, result);
+      FPU_copy_to_reg0(&CONST_QNaN, TAG_Special);
       return;
     }
 #endif PARANOID
 
-  exponent = arg->exp - EXP_BIAS;
+  exponent = exponent(st0_ptr);
 
   accumulator.lsw = accumulator.midw = accumulator.msw = 0;
 
-  if ( (exponent < -1) || ((exponent == -1) && (arg->sigh <= 0xb00d6f54)) )
+  if ( (exponent < -1) || ((exponent == -1) && (st0_ptr->sigh <= 0xb00d6f54)) )
     {
       /* arg is < 0.687705 */
 
-      argSqrd.msw = arg->sigh; argSqrd.midw = arg->sigl; argSqrd.lsw = 0;
-      mul64_Xsig(&argSqrd, &significand(arg));
+      argSqrd.msw = st0_ptr->sigh; argSqrd.midw = st0_ptr->sigl;
+      argSqrd.lsw = 0;
+      mul64_Xsig(&argSqrd, &significand(st0_ptr));
 
       if ( exponent < -1 )
 	{
@@ -270,8 +256,8 @@ void	poly_cos(FPU_REG const *arg, FPU_REG *result)
 		      N_COEFF_PH-1);
       negate_Xsig(&accumulator);
 
-      mul64_Xsig(&accumulator, &significand(arg));
-      mul64_Xsig(&accumulator, &significand(arg));
+      mul64_Xsig(&accumulator, &significand(st0_ptr));
+      mul64_Xsig(&accumulator, &significand(st0_ptr));
       shr_Xsig(&accumulator, -2*(1+exponent));
 
       shr_Xsig(&accumulator, 3);
@@ -290,20 +276,20 @@ void	poly_cos(FPU_REG const *arg, FPU_REG *result)
       if ( accumulator.msw == 0 )
 	{
 	  /* The result is 1.0 */
-	  reg_move(&CONST_1, result);
+	  FPU_copy_to_reg0(&CONST_1, TAG_Valid);
+	  return;
 	}
       else
 	{
-	  significand(result) = XSIG_LL(accumulator);
+	  significand(&result) = XSIG_LL(accumulator);
       
 	  /* will be a valid positive nr with expon = -1 */
-	  *(short *)&(result->sign) = 0;
-	  result->exp = EXP_BIAS - 1;
+	  setexponentpos(&result, -1);
 	}
     }
   else
     {
-      fixed_arg = significand(arg);
+      fixed_arg = significand(st0_ptr);
 
       if ( exponent == 0 )
 	{
@@ -392,14 +378,15 @@ void	poly_cos(FPU_REG const *arg, FPU_REG *result)
 
       echange = round_Xsig(&accumulator);
 
-      result->exp = exp2 + EXP_BIAS + echange;
-      *(short *)&(result->sign) = 0;      /* Is a valid positive nr */
-      significand(result) = XSIG_LL(accumulator);
+      setexponentpos(&result, exp2 + echange);
+      significand(&result) = XSIG_LL(accumulator);
     }
 
+  FPU_copy_to_reg0(&result, TAG_Valid);
+
 #ifdef PARANOID
-  if ( (result->exp >= EXP_BIAS)
-      && (significand(result) > 0x8000000000000000LL) )
+  if ( (exponent(&result) >= 0)
+      && (significand(&result) > 0x8000000000000000LL) )
     {
       EXCEPTION(EX_INTERNAL|0x151);
     }
diff --git a/arch/i386/math-emu/poly_tan.c b/arch/i386/math-emu/poly_tan.c
index d9b09e438..1743d6f0f 100644
--- a/arch/i386/math-emu/poly_tan.c
+++ b/arch/i386/math-emu/poly_tan.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Compute the tan of a FPU_REG, using a polynomial approximation.           |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
+ | Copyright (C) 1992,1993,1994,1997                                         |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ |                       Australia.  E-mail   billm@suburbia.net             |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -13,6 +13,7 @@
 #include "exception.h"
 #include "reg_constant.h"
 #include "fpu_emu.h"
+#include "fpu_system.h"
 #include "control_w.h"
 #include "poly.h"
 
@@ -52,7 +53,7 @@ static const unsigned long long twothirds = 0xaaaaaaaaaaaaaaabLL;
 /*--- poly_tan() ------------------------------------------------------------+
  |                                                                           |
  +---------------------------------------------------------------------------*/
-void	poly_tan(FPU_REG const *arg, FPU_REG *result)
+void	poly_tan(FPU_REG *st0_ptr)
 {
   long int    		exponent;
   int                   invert;
@@ -60,20 +61,20 @@ void	poly_tan(FPU_REG const *arg, FPU_REG *result)
                         argSignif, fix_up;
   unsigned long         adj;
 
-  exponent = arg->exp - EXP_BIAS;
+  exponent = exponent(st0_ptr);
 
 #ifdef PARANOID
-  if ( arg->sign != 0 )	/* Can't hack a number < 0.0 */
-    { arith_invalid(result); return; }  /* Need a positive number */
+  if ( signnegative(st0_ptr) )	/* Can't hack a number < 0.0 */
+    { arith_invalid(0); return; }  /* Need a positive number */
 #endif PARANOID
 
   /* Split the problem into two domains, smaller and larger than pi/4 */
-  if ( (exponent == 0) || ((exponent == -1) && (arg->sigh > 0xc90fdaa2)) )
+  if ( (exponent == 0) || ((exponent == -1) && (st0_ptr->sigh > 0xc90fdaa2)) )
     {
       /* The argument is greater than (approx) pi/4 */
       invert = 1;
       accum.lsw = 0;
-      XSIG_LL(accum) = significand(arg);
+      XSIG_LL(accum) = significand(st0_ptr);
  
       if ( exponent == 0 )
 	{
@@ -92,12 +93,12 @@ void	poly_tan(FPU_REG const *arg, FPU_REG *result)
     {
       invert = 0;
       argSignif.lsw = 0;
-      XSIG_LL(accum) = XSIG_LL(argSignif) = significand(arg);
+      XSIG_LL(accum) = XSIG_LL(argSignif) = significand(st0_ptr);
  
       if ( exponent < -1 )
 	{
 	  /* shift the argument right by the required places */
-	  if ( shrx(&XSIG_LL(accum), -1-exponent) >= 0x80000000U )
+	  if ( FPU_shrx(&XSIG_LL(accum), -1-exponent) >= 0x80000000U )
 	    XSIG_LL(accum) ++;	/* round up */
 	}
     }
@@ -206,8 +207,8 @@ void	poly_tan(FPU_REG const *arg, FPU_REG *result)
 
   /* Transfer the result */
   round_Xsig(&accum);
-  *(short *)&(result->sign) = 0;
-  significand(result) = XSIG_LL(accum);
-  result->exp = EXP_BIAS + exponent;
+  FPU_settag0(TAG_Valid);
+  significand(st0_ptr) = XSIG_LL(accum);
+  setexponent16(st0_ptr, exponent + EXTENDED_Ebias);  /* Result is positive. */
 
 }
diff --git a/arch/i386/math-emu/reg_add_sub.c b/arch/i386/math-emu/reg_add_sub.c
index d70889b40..05e86d624 100644
--- a/arch/i386/math-emu/reg_add_sub.c
+++ b/arch/i386/math-emu/reg_add_sub.c
@@ -3,16 +3,19 @@
  |                                                                           |
  | Functions to add or subtract two registers and put the result in a third. |
  |                                                                           |
- | Copyright (C) 1992,1993                                                   |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1997                                              |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
 /*---------------------------------------------------------------------------+
- | For each function, the destination may be any FPU_REG, including one of   |
+ |  For each function, the destination may be any FPU_REG, including one of  |
  | the source FPU_REGs.                                                      |
+ |  Each function returns 0 if the answer is o.k., otherwise a non-zero      |
+ | value is returned, indicating either an exception condition or an         |
+ | internal error.                                                           |
  +---------------------------------------------------------------------------*/
 
 #include "exception.h"
@@ -21,156 +24,164 @@
 #include "control_w.h"
 #include "fpu_system.h"
 
+static
+int add_sub_specials(FPU_REG const *a, u_char taga, u_char signa,
+		     FPU_REG const *b, u_char tagb, u_char signb,
+		     FPU_REG *dest, int deststnr, int control_w);
 
-int reg_add(FPU_REG const *a, FPU_REG const *b, FPU_REG *dest, int control_w)
+/*
+  Operates on st(0) and st(n), or on st(0) and temporary data.
+  The destination must be one of the source st(x).
+  */
+int FPU_add(FPU_REG const *b, u_char tagb, int deststnr, int control_w)
 {
-  char saved_sign = dest->sign;
-  int diff;
+  FPU_REG *a = &st(0);
+  FPU_REG *dest = &st(deststnr);
+  u_char signb = getsign(b);
+  u_char taga = FPU_gettag0();
+  u_char signa = getsign(a);
+  u_char saved_sign = getsign(dest);
+  int diff, tag, expa, expb;
   
-  if ( !(a->tag | b->tag) )
+  if ( !(taga | tagb) )
     {
+      expa = exponent(a);
+      expb = exponent(b);
+
+    valid_add:
       /* Both registers are valid */
-      if (!(a->sign ^ b->sign))
+      if (!(signa ^ signb))
 	{
 	  /* signs are the same */
-	  dest->sign = a->sign;
-	  if ( reg_u_add(a, b, dest, control_w) )
-	    {
-	      dest->sign = saved_sign;
-	      return 1;
-	    }
-	  return 0;
+	  tag = FPU_u_add(a, b, dest, control_w, signa, expa, expb);
 	}
-      
-      /* The signs are different, so do a subtraction */
-      diff = a->exp - b->exp;
-      if (!diff)
+      else
 	{
-	  diff = a->sigh - b->sigh;  /* Works only if ms bits are identical */
+	  /* The signs are different, so do a subtraction */
+	  diff = expa - expb;
 	  if (!diff)
 	    {
-	      diff = a->sigl > b->sigl;
+	      diff = a->sigh - b->sigh;  /* This works only if the ms bits
+					    are identical. */
 	      if (!diff)
-		diff = -(a->sigl < b->sigl);
-	    }
-	}
-      
-      if (diff > 0)
-	{
-	  dest->sign = a->sign;
-	  if ( reg_u_sub(a, b, dest, control_w) )
-	    {
-	      dest->sign = saved_sign;
-	      return 1;
-	    }
-	}
-      else if ( diff == 0 )
-	{
-#ifdef DENORM_OPERAND
-	  if ( (b->tag == TW_Valid) && (b->exp <= EXP_UNDER) &&
-	      denormal_operand() )
-	    return 1;
-#endif DENORM_OPERAND
-	  reg_move(&CONST_Z, dest);
-	  /* sign depends upon rounding mode */
-	  dest->sign = ((control_w & CW_RC) != RC_DOWN)
-	    ? SIGN_POS : SIGN_NEG;
-	}
-      else
-	{
-	  dest->sign = b->sign;
-	  if ( reg_u_sub(b, a, dest, control_w) )
-	    {
-	      dest->sign = saved_sign;
-	      return 1;
-	    }
-	}
-      return 0;
-    }
-  else
-    {
-      if ( (a->tag == TW_NaN) || (b->tag == TW_NaN) )
-	{ return real_2op_NaN(a, b, dest); }
-      else if (a->tag == TW_Zero)
-	{
-	  if (b->tag == TW_Zero)
-	    {
-	      char different_signs = a->sign ^ b->sign;
-	      /* Both are zero, result will be zero. */
-	      reg_move(a, dest);
-	      if (different_signs)
 		{
-		  /* Signs are different. */
-		  /* Sign of answer depends upon rounding mode. */
-		  dest->sign = ((control_w & CW_RC) != RC_DOWN)
-		    ? SIGN_POS : SIGN_NEG;
+		  diff = a->sigl > b->sigl;
+		  if (!diff)
+		    diff = -(a->sigl < b->sigl);
 		}
 	    }
-	  else
+      
+	  if (diff > 0)
 	    {
-#ifdef DENORM_OPERAND
-	      if ( (b->tag == TW_Valid) && (b->exp <= EXP_UNDER) &&
-		  denormal_operand() )
-		return 1;
-#endif DENORM_OPERAND
-	      reg_move(b, dest);
+	      tag = FPU_u_sub(a, b, dest, control_w, signa, expa, expb);
 	    }
-	  return 0;
-	}
-      else if (b->tag == TW_Zero)
-	{
-#ifdef DENORM_OPERAND
-	  if ( (a->tag == TW_Valid) && (a->exp <= EXP_UNDER) &&
-	      denormal_operand() )
-	    return 1;
-#endif DENORM_OPERAND
-	  reg_move(a, dest); return 0;
-	}
-      else if (a->tag == TW_Infinity)
-	{
-	  if (b->tag != TW_Infinity)
+	  else if ( diff < 0 )
 	    {
-#ifdef DENORM_OPERAND
-	      if ( (b->tag == TW_Valid) && (b->exp <= EXP_UNDER) &&
-		  denormal_operand() )
-		return 1;
-#endif DENORM_OPERAND
-	      reg_move(a, dest); return 0;
+	      tag = FPU_u_sub(b, a, dest, control_w, signb, expb, expa);
 	    }
-	  if (a->sign == b->sign)
+	  else
 	    {
-	      /* They are both + or - infinity */
-	      reg_move(a, dest); return 0;
+	      FPU_copy_to_regi(&CONST_Z, TAG_Zero, deststnr);
+	      /* sign depends upon rounding mode */
+	      setsign(dest, ((control_w & CW_RC) != RC_DOWN)
+		      ? SIGN_POS : SIGN_NEG);
+	      return TAG_Zero;
 	    }
-	  return arith_invalid(dest);	/* Infinity-Infinity is undefined. */
 	}
-      else if (b->tag == TW_Infinity)
+
+      if ( tag < 0 )
 	{
-#ifdef DENORM_OPERAND
-	  if ( (a->tag == TW_Valid) && (a->exp <= EXP_UNDER) &&
-	      denormal_operand() )
-	    return 1;
-#endif DENORM_OPERAND
-	  reg_move(b, dest); return 0;
+	  setsign(dest, saved_sign);
+	  return tag;
 	}
+      FPU_settagi(deststnr, tag);
+      return tag;
     }
-#ifdef PARANOID
-  EXCEPTION(EX_INTERNAL|0x101);
-#endif
-  return 1;
+
+  if ( taga == TAG_Special )
+    taga = FPU_Special(a);
+  if ( tagb == TAG_Special )
+    tagb = FPU_Special(b);
+
+  if ( ((taga == TAG_Valid) && (tagb == TW_Denormal))
+	    || ((taga == TW_Denormal) && (tagb == TAG_Valid))
+	    || ((taga == TW_Denormal) && (tagb == TW_Denormal)) )
+    {
+      FPU_REG x, y;
+
+      if ( denormal_operand() < 0 )
+	return FPU_Exception;
+
+      FPU_to_exp16(a, &x);
+      FPU_to_exp16(b, &y);
+      a = &x;
+      b = &y;
+      expa = exponent16(a);
+      expb = exponent16(b);
+      goto valid_add;
+    }
+
+  if ( (taga == TW_NaN) || (tagb == TW_NaN) )
+    {
+      if ( deststnr == 0 )
+	return real_2op_NaN(b, tagb, deststnr, a);
+      else
+	return real_2op_NaN(a, taga, deststnr, a);
+    }
+
+  return add_sub_specials(a, taga, signa, b, tagb, signb,
+			  dest, deststnr, control_w);
 }
 
 
 /* Subtract b from a.  (a-b) -> dest */
-int reg_sub(FPU_REG const *a, FPU_REG const *b, FPU_REG *dest, int control_w)
+int FPU_sub(int flags, int rm, int control_w)
 {
-  char saved_sign = dest->sign;
-  int diff;
+  FPU_REG const *a, *b;
+  FPU_REG *dest;
+  u_char taga, tagb, signa, signb, saved_sign, sign;
+  int diff, tag, expa, expb, deststnr;
+
+  a = &st(0);
+  taga = FPU_gettag0();
+
+  deststnr = 0;
+  if ( flags & LOADED )
+    {
+      b = (FPU_REG *)rm;
+      tagb = flags & 0x0f;
+    }
+  else
+    {
+      b = &st(rm);
+      tagb = FPU_gettagi(rm);
+
+      if ( flags & DEST_RM )
+	deststnr = rm;
+    }
+
+  signa = getsign(a);
+  signb = getsign(b);
+
+  if ( flags & REV )
+    {
+      signa ^= SIGN_NEG;
+      signb ^= SIGN_NEG;
+    }
 
-  if ( !(a->tag | b->tag) )
+  dest = &st(deststnr);
+  saved_sign = getsign(dest);
+
+  if ( !(taga | tagb) )
     {
+      expa = exponent(a);
+      expb = exponent(b);
+
+    valid_subtract:
       /* Both registers are valid */
-      diff = a->exp - b->exp;
+
+      diff = expa - expb;
+
       if (!diff)
 	{
 	  diff = a->sigh - b->sigh;  /* Works only if ms bits are identical */
@@ -182,137 +193,182 @@ int reg_sub(FPU_REG const *a, FPU_REG const *b, FPU_REG *dest, int control_w)
 	    }
 	}
 
-      switch (a->sign*2 + b->sign)
+      switch ( (((int)signa)*2 + signb) / SIGN_NEG )
 	{
 	case 0: /* P - P */
 	case 3: /* N - N */
 	  if (diff > 0)
 	    {
 	      /* |a| > |b| */
-	      dest->sign = a->sign;
-	      if ( reg_u_sub(a, b, dest, control_w) )
-		{
-		  dest->sign = saved_sign;
-		  return 1;
-		}
-	      return 0;
+	      tag = FPU_u_sub(a, b, dest, control_w, signa, expa, expb);
 	    }
 	  else if ( diff == 0 )
 	    {
-#ifdef DENORM_OPERAND
-	      if ( (b->tag == TW_Valid) && (b->exp <= EXP_UNDER) &&
-		  denormal_operand() )
-		return 1;
-#endif DENORM_OPERAND
-	      reg_move(&CONST_Z, dest);
+	      FPU_copy_to_regi(&CONST_Z, TAG_Zero, deststnr);
+
 	      /* sign depends upon rounding mode */
-	      dest->sign = ((control_w & CW_RC) != RC_DOWN)
-		? SIGN_POS : SIGN_NEG;
+	      setsign(dest, ((control_w & CW_RC) != RC_DOWN)
+		? SIGN_POS : SIGN_NEG);
+	      return TAG_Zero;
 	    }
 	  else
 	    {
-	      dest->sign = a->sign ^ SIGN_POS^SIGN_NEG;
-	      if ( reg_u_sub(b, a, dest, control_w) )
-		{
-		  dest->sign = saved_sign;
-		  return 1;
-		}
+	      sign = signa ^ SIGN_NEG;
+	      tag = FPU_u_sub(b, a, dest, control_w, sign, expb, expa);
 	    }
 	  break;
 	case 1: /* P - N */
-	  dest->sign = SIGN_POS;
-	  if ( reg_u_add(a, b, dest, control_w) )
-	    {
-	      dest->sign = saved_sign;
-	      return 1;
-	    }
+	  tag = FPU_u_add(a, b, dest, control_w, SIGN_POS, expa, expb);
 	  break;
 	case 2: /* N - P */
-	  dest->sign = SIGN_NEG;
-	  if ( reg_u_add(a, b, dest, control_w) )
-	    {
-	      dest->sign = saved_sign;
-	      return 1;
-	    }
+	  tag = FPU_u_add(a, b, dest, control_w, SIGN_NEG, expa, expb);
 	  break;
+#ifdef PARANOID
+	default:
+	  EXCEPTION(EX_INTERNAL|0x111);
+	  return -1;
+#endif
+	}
+      if ( tag < 0 )
+	{
+	  setsign(dest, saved_sign);
+	  return tag;
 	}
-      return 0;
+      FPU_settagi(deststnr, tag);
+      return tag;
     }
-  else
+
+  if ( taga == TAG_Special )
+    taga = FPU_Special(a);
+  if ( tagb == TAG_Special )
+    tagb = FPU_Special(b);
+
+  if ( ((taga == TAG_Valid) && (tagb == TW_Denormal))
+	    || ((taga == TW_Denormal) && (tagb == TAG_Valid))
+	    || ((taga == TW_Denormal) && (tagb == TW_Denormal)) )
     {
-      if ( (a->tag == TW_NaN) || (b->tag == TW_NaN) )
-	{ return real_2op_NaN(b, a, dest); }
-      else if (b->tag == TW_Zero)
-	{ 
-	  if (a->tag == TW_Zero)
-	    {
-	      char same_signs = !(a->sign ^ b->sign);
-	      /* Both are zero, result will be zero. */
-	      reg_move(a, dest); /* Answer for different signs. */
-	      if (same_signs)
-		{
-		  /* Sign depends upon rounding mode */
-		  dest->sign = ((control_w & CW_RC) != RC_DOWN)
-		    ? SIGN_POS : SIGN_NEG;
-		}
-	    }
-	  else
-	    {
-#ifdef DENORM_OPERAND
-	      if ( (a->tag == TW_Valid) && (a->exp <= EXP_UNDER) &&
-		  denormal_operand() )
-		return 1;
-#endif DENORM_OPERAND
-	      reg_move(a, dest);
-	    }
-	  return 0;
+      FPU_REG x, y;
+
+      if ( denormal_operand() < 0 )
+	return FPU_Exception;
+
+      FPU_to_exp16(a, &x);
+      FPU_to_exp16(b, &y);
+      a = &x;
+      b = &y;
+      expa = exponent16(a);
+      expb = exponent16(b);
+
+      goto valid_subtract;
+    }
+
+  if ( (taga == TW_NaN) || (tagb == TW_NaN) )
+    {
+      FPU_REG const *d1, *d2;
+      if ( flags & REV )
+	{
+	  d1 = b;
+	  d2 = a;
 	}
-      else if (a->tag == TW_Zero)
+      else
 	{
-#ifdef DENORM_OPERAND
-	  if ( (b->tag == TW_Valid) && (b->exp <= EXP_UNDER) &&
-	      denormal_operand() )
-	    return 1;
-#endif DENORM_OPERAND
-	  reg_move(b, dest);
-	  dest->sign ^= SIGN_POS^SIGN_NEG;
-	  return 0;
+	  d1 = a;
+	  d2 = b;
 	}
-      else if (a->tag == TW_Infinity)
+      if ( flags & LOADED )
+	return real_2op_NaN(b, tagb, deststnr, d1);
+      if ( flags & DEST_RM )
+	return real_2op_NaN(a, taga, deststnr, d2);
+      else
+	return real_2op_NaN(b, tagb, deststnr, d2);
+    }
+
+    return add_sub_specials(a, taga, signa, b, tagb, signb ^ SIGN_NEG,
+			    dest, deststnr, control_w);
+}
+
+
+static
+int add_sub_specials(FPU_REG const *a, u_char taga, u_char signa,
+		     FPU_REG const *b, u_char tagb, u_char signb,
+		     FPU_REG *dest, int deststnr, int control_w)
+{
+  if ( ((taga == TW_Denormal) || (tagb == TW_Denormal))
+       && (denormal_operand() < 0) )
+    return FPU_Exception;
+
+  if (taga == TAG_Zero)
+    {
+      if (tagb == TAG_Zero)
 	{
-	  if (b->tag != TW_Infinity)
+	  /* Both are zero, result will be zero. */
+	  u_char different_signs = signa ^ signb;
+
+	  FPU_copy_to_regi(a, TAG_Zero, deststnr);
+	  if ( different_signs )
 	    {
-#ifdef DENORM_OPERAND
-	      if ( (b->tag == TW_Valid) && (b->exp <= EXP_UNDER) &&
-		  denormal_operand() )
-		return 1;
-#endif DENORM_OPERAND
-	      reg_move(a, dest); return 0;
+	      /* Signs are different. */
+	      /* Sign of answer depends upon rounding mode. */
+	      setsign(dest, ((control_w & CW_RC) != RC_DOWN)
+		      ? SIGN_POS : SIGN_NEG);
 	    }
-	  /* Both args are Infinity */
-	  if (a->sign == b->sign)
+	  else
+	    setsign(dest, signa);  /* signa may differ from the sign of a. */
+	  return TAG_Zero;
+	}
+      else
+	{
+	  reg_copy(b, dest);
+	  if ( (tagb == TW_Denormal) && (b->sigh & 0x80000000) )
 	    {
-	      /* Infinity-Infinity is undefined. */
-	      return arith_invalid(dest);
+	      /* A pseudoDenormal, convert it. */
+	      addexponent(dest, 1);
+	      tagb = TAG_Valid;
 	    }
-	  reg_move(a, dest);
-	  return 0;
+	  else if ( tagb > TAG_Empty )
+	    tagb = TAG_Special;
+	  setsign(dest, signb);  /* signb may differ from the sign of b. */
+	  FPU_settagi(deststnr, tagb);
+	  return tagb;
 	}
-      else if (b->tag == TW_Infinity)
+    }
+  else if (tagb == TAG_Zero)
+    {
+      reg_copy(a, dest);
+      if ( (taga == TW_Denormal) && (a->sigh & 0x80000000) )
 	{
-#ifdef DENORM_OPERAND
-	  if ( (a->tag == TW_Valid) && (a->exp <= EXP_UNDER) &&
-	      denormal_operand() )
-	    return 1;
-#endif DENORM_OPERAND
-	  reg_move(b, dest);
-	  dest->sign ^= SIGN_POS^SIGN_NEG;
-	  return 0;
+	  /* A pseudoDenormal */
+	  addexponent(dest, 1);
+	  taga = TAG_Valid;
 	}
+      else if ( taga > TAG_Empty )
+	taga = TAG_Special;
+      setsign(dest, signa);  /* signa may differ from the sign of a. */
+      FPU_settagi(deststnr, taga);
+      return taga;
     }
+  else if (taga == TW_Infinity)
+    {
+      if ( (tagb != TW_Infinity) || (signa == signb) )
+	{
+	  FPU_copy_to_regi(a, TAG_Special, deststnr);
+	  setsign(dest, signa);  /* signa may differ from the sign of a. */
+	  return taga;
+	}
+      /* Infinity-Infinity is undefined. */
+      return arith_invalid(deststnr);
+    }
+  else if (tagb == TW_Infinity)
+    {
+      FPU_copy_to_regi(b, TAG_Special, deststnr);
+      setsign(dest, signb);  /* signb may differ from the sign of b. */
+      return tagb;
+    }
+
 #ifdef PARANOID
-  EXCEPTION(EX_INTERNAL|0x110);
+  EXCEPTION(EX_INTERNAL|0x101);
 #endif
-  return 1;
+
+  return FPU_Exception;
 }
 
diff --git a/arch/i386/math-emu/reg_compare.c b/arch/i386/math-emu/reg_compare.c
index eb4a1fa99..20023977f 100644
--- a/arch/i386/math-emu/reg_compare.c
+++ b/arch/i386/math-emu/reg_compare.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | Compare two floating point registers                                      |
  |                                                                           |
- | Copyright (C) 1992,1993,1994                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1994,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -21,86 +21,87 @@
 #include "status_w.h"
 
 
-int compare(FPU_REG const *b)
+static int compare(FPU_REG const *b, int tagb)
 {
-  int diff;
-  char	       st0_tag;
-  FPU_REG      *st0_ptr;
+  int diff, exp0, expb;
+  u_char	  	st0_tag;
+  FPU_REG  	*st0_ptr;
+  FPU_REG	x, y;
+  u_char		st0_sign, signb = getsign(b);
 
   st0_ptr = &st(0);
-  st0_tag = st0_ptr->tag;
+  st0_tag = FPU_gettag0();
+  st0_sign = getsign(st0_ptr);
 
-  if ( st0_tag | b->tag )
+  if ( tagb == TAG_Special )
+    tagb = FPU_Special(b);
+  if ( st0_tag == TAG_Special )
+    st0_tag = FPU_Special(st0_ptr);
+
+  if ( ((st0_tag != TAG_Valid) && (st0_tag != TW_Denormal))
+       || ((tagb != TAG_Valid) && (tagb != TW_Denormal)) )
     {
-      if ( st0_tag == TW_Zero )
+      if ( st0_tag == TAG_Zero )
 	{
-	  if ( b->tag == TW_Zero ) return COMP_A_eq_B;
-	  if ( b->tag == TW_Valid )
-	    {
-	      return ((b->sign == SIGN_POS) ? COMP_A_lt_B : COMP_A_gt_B)
-#ifdef DENORM_OPERAND
-		| ((b->exp <= EXP_UNDER) ?
-		   COMP_Denormal : 0)
-#endif DENORM_OPERAND
-		  ;
-	    }
+	  if ( tagb == TAG_Zero ) return COMP_A_eq_B;
+	  if ( tagb == TAG_Valid )
+	    return ((signb == SIGN_POS) ? COMP_A_lt_B : COMP_A_gt_B);
+	  if ( tagb == TW_Denormal )
+	    return ((signb == SIGN_POS) ? COMP_A_lt_B : COMP_A_gt_B)
+	    | COMP_Denormal;
 	}
-      else if ( b->tag == TW_Zero )
+      else if ( tagb == TAG_Zero )
 	{
-	  if ( st0_tag == TW_Valid )
-	    {
-	      return ((st0_ptr->sign == SIGN_POS) ? COMP_A_gt_B
-		      : COMP_A_lt_B)
-#ifdef DENORM_OPERAND
-		| ((st0_ptr->exp <= EXP_UNDER )
-		   ? COMP_Denormal : 0 )
-#endif DENORM_OPERAND
-		  ;
-	    }
+	  if ( st0_tag == TAG_Valid )
+	    return ((st0_sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B);
+	  if ( st0_tag == TW_Denormal )
+	    return ((st0_sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B)
+	    | COMP_Denormal;
 	}
 
       if ( st0_tag == TW_Infinity )
 	{
-	  if ( (b->tag == TW_Valid) || (b->tag == TW_Zero) )
-	    {
-	      return ((st0_ptr->sign == SIGN_POS) ? COMP_A_gt_B
-		      : COMP_A_lt_B)
-#ifdef DENORM_OPERAND
-	      | (((b->tag == TW_Valid) && (b->exp <= EXP_UNDER)) ?
-		COMP_Denormal : 0 )
-#endif DENORM_OPERAND
-;
-	    }
-	  else if ( b->tag == TW_Infinity )
+	  if ( (tagb == TAG_Valid) || (tagb == TAG_Zero) )
+	    return ((st0_sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B);
+	  else if ( tagb == TW_Denormal )
+	    return ((st0_sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B)
+	      | COMP_Denormal;
+	  else if ( tagb == TW_Infinity )
 	    {
 	      /* The 80486 book says that infinities can be equal! */
-	      return (st0_ptr->sign == b->sign) ? COMP_A_eq_B :
-		((st0_ptr->sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B);
+	      return (st0_sign == signb) ? COMP_A_eq_B :
+		((st0_sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B);
 	    }
 	  /* Fall through to the NaN code */
 	}
-      else if ( b->tag == TW_Infinity )
+      else if ( tagb == TW_Infinity )
 	{
-	  if ( (st0_tag == TW_Valid) || (st0_tag == TW_Zero) )
-	    {
-	      return ((b->sign == SIGN_POS) ? COMP_A_lt_B : COMP_A_gt_B)
-#ifdef DENORM_OPERAND
-		| (((st0_tag == TW_Valid)
-		    && (st0_ptr->exp <= EXP_UNDER)) ?
-		   COMP_Denormal : 0)
-#endif DENORM_OPERAND
-		  ;
-	    }
+	  if ( (st0_tag == TAG_Valid) || (st0_tag == TAG_Zero) )
+	    return ((signb == SIGN_POS) ? COMP_A_lt_B : COMP_A_gt_B);
+	  if ( st0_tag == TW_Denormal )
+	    return ((signb == SIGN_POS) ? COMP_A_lt_B : COMP_A_gt_B)
+		| COMP_Denormal;
 	  /* Fall through to the NaN code */
 	}
 
       /* The only possibility now should be that one of the arguments
 	 is a NaN */
-      if ( (st0_tag == TW_NaN) || (b->tag == TW_NaN) )
+      if ( (st0_tag == TW_NaN) || (tagb == TW_NaN) )
 	{
-	  if ( ((st0_tag == TW_NaN) && !(st0_ptr->sigh & 0x40000000))
-	      || ((b->tag == TW_NaN) && !(b->sigh & 0x40000000)) )
-	    /* At least one arg is a signaling NaN */
+	  int signalling = 0, unsupported = 0;
+	  if ( st0_tag == TW_NaN )
+	    {
+	      signalling = (st0_ptr->sigh & 0xc0000000) == 0x80000000;
+	      unsupported = !((exponent(st0_ptr) == EXP_OVER)
+			      && (st0_ptr->sigh & 0x80000000));
+	    }
+	  if ( tagb == TW_NaN )
+	    {
+	      signalling |= (b->sigh & 0xc0000000) == 0x80000000;
+	      unsupported |= !((exponent(b) == EXP_OVER)
+			       && (b->sigh & 0x80000000));
+	    }
+	  if ( signalling || unsupported )
 	    return COMP_No_Comp | COMP_SNaN | COMP_NaN;
 	  else
 	    /* Neither is a signaling NaN */
@@ -110,24 +111,34 @@ int compare(FPU_REG const *b)
       EXCEPTION(EX_Invalid);
     }
   
+  if (st0_sign != signb)
+    {
+      return ((st0_sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B)
+	| ( ((st0_tag == TW_Denormal) || (tagb == TW_Denormal)) ?
+	    COMP_Denormal : 0);
+    }
+
+  if ( (st0_tag == TW_Denormal) || (tagb == TW_Denormal) )
+    {
+      FPU_to_exp16(st0_ptr, &x);
+      FPU_to_exp16(b, &y);
+      st0_ptr = &x;
+      b = &y;
+      exp0 = exponent16(st0_ptr);
+      expb = exponent16(b);
+    }
+  else
+    {
+      exp0 = exponent(st0_ptr);
+      expb = exponent(b);
+    }
+
 #ifdef PARANOID
   if (!(st0_ptr->sigh & 0x80000000)) EXCEPTION(EX_Invalid);
   if (!(b->sigh & 0x80000000)) EXCEPTION(EX_Invalid);
 #endif PARANOID
 
-  
-  if (st0_ptr->sign != b->sign)
-    {
-      return ((st0_ptr->sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B)
-#ifdef DENORM_OPERAND
-	|
-	  ( ((st0_ptr->exp <= EXP_UNDER) || (b->exp <= EXP_UNDER)) ?
-	   COMP_Denormal : 0)
-#endif DENORM_OPERAND
-	    ;
-    }
-
-  diff = st0_ptr->exp - b->exp;
+  diff = exp0 - expb;
   if ( diff == 0 )
     {
       diff = st0_ptr->sigh - b->sigh;  /* Works only if ms bits are
@@ -142,42 +153,30 @@ int compare(FPU_REG const *b)
 
   if ( diff > 0 )
     {
-      return ((st0_ptr->sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B)
-#ifdef DENORM_OPERAND
-	|
-	  ( ((st0_ptr->exp <= EXP_UNDER) || (b->exp <= EXP_UNDER)) ?
-	   COMP_Denormal : 0)
-#endif DENORM_OPERAND
-	    ;
+      return ((st0_sign == SIGN_POS) ? COMP_A_gt_B : COMP_A_lt_B)
+	| ( ((st0_tag == TW_Denormal) || (tagb == TW_Denormal)) ?
+	    COMP_Denormal : 0);
     }
   if ( diff < 0 )
     {
-      return ((st0_ptr->sign == SIGN_POS) ? COMP_A_lt_B : COMP_A_gt_B)
-#ifdef DENORM_OPERAND
-	|
-	  ( ((st0_ptr->exp <= EXP_UNDER) || (b->exp <= EXP_UNDER)) ?
-	   COMP_Denormal : 0)
-#endif DENORM_OPERAND
-	    ;
+      return ((st0_sign == SIGN_POS) ? COMP_A_lt_B : COMP_A_gt_B)
+	| ( ((st0_tag == TW_Denormal) || (tagb == TW_Denormal)) ?
+	    COMP_Denormal : 0);
     }
 
   return COMP_A_eq_B
-#ifdef DENORM_OPERAND
-    |
-      ( ((st0_ptr->exp <= EXP_UNDER) || (b->exp <= EXP_UNDER)) ?
-       COMP_Denormal : 0)
-#endif DENORM_OPERAND
-	;
+    | ( ((st0_tag == TW_Denormal) || (tagb == TW_Denormal)) ?
+	COMP_Denormal : 0);
 
 }
 
 
 /* This function requires that st(0) is not empty */
-int compare_st_data(FPU_REG const *loaded_data)
+int FPU_compare_st_data(FPU_REG const *loaded_data, u_char loaded_tag)
 {
   int f, c;
 
-  c = compare(loaded_data);
+  c = compare(loaded_data, loaded_tag);
 
   if (c & COMP_NaN)
     {
@@ -209,7 +208,7 @@ int compare_st_data(FPU_REG const *loaded_data)
   setcc(f);
   if (c & COMP_Denormal)
     {
-      return denormal_operand();
+      return denormal_operand() < 0;
     }
   return 0;
 }
@@ -218,6 +217,7 @@ int compare_st_data(FPU_REG const *loaded_data)
 static int compare_st_st(int nr)
 {
   int f, c;
+  FPU_REG *st_ptr;
 
   if ( !NOT_EMPTY(0) || !NOT_EMPTY(nr) )
     {
@@ -227,7 +227,8 @@ static int compare_st_st(int nr)
       return !(control_word & CW_Invalid);
     }
 
-  c = compare(&st(nr));
+  st_ptr = &st(nr);
+  c = compare(st_ptr, FPU_gettagi(nr));
   if (c & COMP_NaN)
     {
       setcc(SW_C3 | SW_C2 | SW_C0);
@@ -259,7 +260,7 @@ static int compare_st_st(int nr)
   setcc(f);
   if (c & COMP_Denormal)
     {
-      return denormal_operand();
+      return denormal_operand() < 0;
     }
   return 0;
 }
@@ -268,6 +269,7 @@ static int compare_st_st(int nr)
 static int compare_u_st_st(int nr)
 {
   int f, c;
+  FPU_REG *st_ptr;
 
   if ( !NOT_EMPTY(0) || !NOT_EMPTY(nr) )
     {
@@ -277,7 +279,8 @@ static int compare_u_st_st(int nr)
       return !(control_word & CW_Invalid);
     }
 
-  c = compare(&st(nr));
+  st_ptr = &st(nr);
+  c = compare(st_ptr, FPU_gettagi(nr));
   if (c & COMP_NaN)
     {
       setcc(SW_C3 | SW_C2 | SW_C0);
@@ -314,7 +317,7 @@ static int compare_u_st_st(int nr)
   setcc(f);
   if (c & COMP_Denormal)
     {
-      return denormal_operand();
+      return denormal_operand() < 0;
     }
   return 0;
 }
@@ -332,7 +335,7 @@ void fcompst()
 {
   /* fcomp st(i) */
   if ( !compare_st_st(FPU_rm) )
-    pop();
+    FPU_pop();
 }
 
 
@@ -361,7 +364,7 @@ void fucomp()
 {
   /* fucomp st(i) */
   if ( !compare_u_st_st(FPU_rm) )
-    pop();
+    FPU_pop();
 }
 
 
diff --git a/arch/i386/math-emu/reg_constant.c b/arch/i386/math-emu/reg_constant.c
index 1b2458eea..ffbfebd51 100644
--- a/arch/i386/math-emu/reg_constant.c
+++ b/arch/i386/math-emu/reg_constant.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | All of the constant FPU_REGs                                              |
  |                                                                           |
- | Copyright (C) 1992,1993,1994,1996                                         |
+ | Copyright (C) 1992,1993,1994,1997                                         |
  |                     W. Metzenthen, 22 Parker St, Ormond, Vic 3163,        |
- |                     Australia.  E-mail   billm@jacobi.maths.monash.edu.au |
+ |                     Australia.  E-mail   billm@suburbia.net               |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -17,59 +17,52 @@
 #include "control_w.h"
 
 
-FPU_REG const CONST_1    = { SIGN_POS, TW_Valid, EXP_BIAS,
-			    0x00000000, 0x80000000 };
-FPU_REG const CONST_2    = { SIGN_POS, TW_Valid, EXP_BIAS+1,
-			    0x00000000, 0x80000000 };
-FPU_REG const CONST_HALF = { SIGN_POS, TW_Valid, EXP_BIAS-1,
-			    0x00000000, 0x80000000 };
-FPU_REG const CONST_L2T  = { SIGN_POS, TW_Valid, EXP_BIAS+1,
-			    0xcd1b8afe, 0xd49a784b };
-FPU_REG const CONST_L2E  = { SIGN_POS, TW_Valid, EXP_BIAS,
-			    0x5c17f0bc, 0xb8aa3b29 };
-FPU_REG const CONST_PI   = { SIGN_POS, TW_Valid, EXP_BIAS+1,
-			    0x2168c235, 0xc90fdaa2 };
-FPU_REG const CONST_PI2  = { SIGN_POS, TW_Valid, EXP_BIAS,
-			    0x2168c235, 0xc90fdaa2 };
-FPU_REG const CONST_PI4  = { SIGN_POS, TW_Valid, EXP_BIAS-1,
-			    0x2168c235, 0xc90fdaa2 };
-FPU_REG const CONST_LG2  = { SIGN_POS, TW_Valid, EXP_BIAS-2,
-			    0xfbcff799, 0x9a209a84 };
-FPU_REG const CONST_LN2  = { SIGN_POS, TW_Valid, EXP_BIAS-1,
-			    0xd1cf79ac, 0xb17217f7 };
+#define MAKE_REG(s,e,l,h) { l, h, \
+                            ((EXTENDED_Ebias+(e)) | ((SIGN_##s != 0)*0x8000)) }
+
+FPU_REG const CONST_1    = MAKE_REG(POS, 0, 0x00000000, 0x80000000);
+FPU_REG const CONST_2    = MAKE_REG(POS, 1, 0x00000000, 0x80000000);
+FPU_REG const CONST_HALF = MAKE_REG(POS, -1, 0x00000000, 0x80000000);
+FPU_REG const CONST_L2T  = MAKE_REG(POS, 1, 0xcd1b8afe, 0xd49a784b);
+FPU_REG const CONST_L2E  = MAKE_REG(POS, 0, 0x5c17f0bc, 0xb8aa3b29);
+FPU_REG const CONST_PI   = MAKE_REG(POS, 1, 0x2168c235, 0xc90fdaa2);
+FPU_REG const CONST_PI2  = MAKE_REG(POS, 0, 0x2168c235, 0xc90fdaa2);
+FPU_REG const CONST_PI4  = MAKE_REG(POS, -1, 0x2168c235, 0xc90fdaa2);
+FPU_REG const CONST_LG2  = MAKE_REG(POS, -2, 0xfbcff799, 0x9a209a84);
+FPU_REG const CONST_LN2  = MAKE_REG(POS, -1, 0xd1cf79ac, 0xb17217f7);
 
 /* Extra bits to take pi/2 to more than 128 bits precision. */
-FPU_REG const CONST_PI2extra = { SIGN_NEG, TW_Valid, EXP_BIAS-66,
-			    0xfc8f8cbb, 0xece675d1 };
+FPU_REG const CONST_PI2extra = MAKE_REG(NEG, -66,
+					 0xfc8f8cbb, 0xece675d1);
 
 /* Only the sign (and tag) is used in internal zeroes */
-FPU_REG const CONST_Z    = { SIGN_POS, TW_Zero, EXP_UNDER, 0x0, 0x0 };
+FPU_REG const CONST_Z    = MAKE_REG(POS, EXP_UNDER, 0x0, 0x0);
 
 /* Only the sign and significand (and tag) are used in internal NaNs */
 /* The 80486 never generates one of these 
-FPU_REG const CONST_SNAN = { SIGN_POS, TW_NaN, EXP_OVER, 0x00000001, 0x80000000 };
+FPU_REG const CONST_SNAN = MAKE_REG(POS, EXP_OVER, 0x00000001, 0x80000000);
  */
 /* This is the real indefinite QNaN */
-FPU_REG const CONST_QNaN = { SIGN_NEG, TW_NaN, EXP_OVER, 0x00000000, 0xC0000000 };
+FPU_REG const CONST_QNaN = MAKE_REG(NEG, EXP_OVER, 0x00000000, 0xC0000000);
 
 /* Only the sign (and tag) is used in internal infinities */
-FPU_REG const CONST_INF  = { SIGN_POS, TW_Infinity, EXP_OVER, 0x00000000, 0x80000000 };
-
+FPU_REG const CONST_INF  = MAKE_REG(POS, EXP_OVER, 0x00000000, 0x80000000);
 
 
-static void fld_const(FPU_REG const *c, int adj)
+static void fld_const(FPU_REG const *c, int adj, u_char tag)
 {
   FPU_REG *st_new_ptr;
 
   if ( STACK_OVERFLOW )
     {
-      stack_overflow();
+      FPU_stack_overflow();
       return;
     }
   push();
-  reg_move(c, st_new_ptr);
+  reg_copy(c, st_new_ptr);
   st_new_ptr->sigl += adj;  /* For all our fldxxx constants, we don't need to
 			       borrow or carry. */
+  FPU_settag0(tag);
   clear_C1();
 }
 
@@ -80,37 +73,37 @@ static void fld_const(FPU_REG const *c, int adj)
 
 static void fld1(int rc)
 {
-  fld_const(&CONST_1, 0);
+  fld_const(&CONST_1, 0, TAG_Valid);
 }
 
 static void fldl2t(int rc)
 {
-  fld_const(&CONST_L2T, (rc == RC_UP) ? 1 : 0);
+  fld_const(&CONST_L2T, (rc == RC_UP) ? 1 : 0, TAG_Valid);
 }
 
 static void fldl2e(int rc)
 {
-  fld_const(&CONST_L2E, DOWN_OR_CHOP(rc) ? -1 : 0);
+  fld_const(&CONST_L2E, DOWN_OR_CHOP(rc) ? -1 : 0, TAG_Valid);
 }
 
 static void fldpi(int rc)
 {
-  fld_const(&CONST_PI, DOWN_OR_CHOP(rc) ? -1 : 0);
+  fld_const(&CONST_PI, DOWN_OR_CHOP(rc) ? -1 : 0, TAG_Valid);
 }
 
 static void fldlg2(int rc)
 {
-  fld_const(&CONST_LG2, DOWN_OR_CHOP(rc) ? -1 : 0);
+  fld_const(&CONST_LG2, DOWN_OR_CHOP(rc) ? -1 : 0, TAG_Valid);
 }
 
 static void fldln2(int rc)
 {
-  fld_const(&CONST_LN2, DOWN_OR_CHOP(rc) ? -1 : 0);
+  fld_const(&CONST_LN2, DOWN_OR_CHOP(rc) ? -1 : 0, TAG_Valid);
 }
 
 static void fldz(int rc)
 {
-  fld_const(&CONST_Z, 0);
+  fld_const(&CONST_Z, 0, TAG_Zero);
 }
 
 typedef void (*FUNC_RC)(int);
diff --git a/arch/i386/math-emu/reg_convert.c b/arch/i386/math-emu/reg_convert.c
new file mode 100644
index 000000000..45a258752
--- /dev/null
+++ b/arch/i386/math-emu/reg_convert.c
@@ -0,0 +1,53 @@
+/*---------------------------------------------------------------------------+
+ |  reg_convert.c                                                            |
+ |                                                                           |
+ |  Convert register representation.                                         |
+ |                                                                           |
+ | Copyright (C) 1992,1993,1994,1996,1997                                    |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
+ |                                                                           |
+ |                                                                           |
+ +---------------------------------------------------------------------------*/
+
+#include "exception.h"
+#include "fpu_emu.h"
+
+
+int FPU_to_exp16(FPU_REG const *a, FPU_REG *x)
+{
+  int sign = getsign(a);
+
+  *(long long *)&(x->sigl) = *(const long long *)&(a->sigl);
+
+  /* Set up the exponent as a 16 bit quantity. */
+  setexponent16(x, exponent(a));
+
+  if ( exponent16(x) == EXP_UNDER )
+    {
+      /* The number is a de-normal or pseudodenormal. */
+      /* We only deal with the significand and exponent. */
+
+      if (x->sigh & 0x80000000)
+	{
+	  /* Is a pseudodenormal. */
+	  /* This is non-80486 behaviour because the number
+	     loses its 'denormal' identity. */
+	  addexponent(x, 1);
+	}
+      else
+	{
+	  /* Is a denormal. */
+	  addexponent(x, 1);
+	  FPU_normalize_nuo(x);
+	}
+    }
+
+  if ( !(x->sigh & 0x80000000) )
+    {
+      EXCEPTION(EX_INTERNAL | 0x180);
+    }
+
+  return sign;
+}
+
diff --git a/arch/i386/math-emu/reg_div.S b/arch/i386/math-emu/reg_div.S
deleted file mode 100644
index 24d44ac6c..000000000
--- a/arch/i386/math-emu/reg_div.S
+++ /dev/null
@@ -1,248 +0,0 @@
-	.file	"reg_div.S"
-/*---------------------------------------------------------------------------+
- |  reg_div.S                                                                |
- |                                                                           |
- | Divide one FPU_REG by another and put the result in a destination FPU_REG.|
- |                                                                           |
- | Copyright (C) 1992,1993,1994,1995                                         |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
- |                                                                           |
- | Call from C as:                                                           |
- |   void reg_div(FPU_REG *a, FPU_REG *b, FPU_REG *dest,                     |
- |                                    unsigned int control_word)             |
- |                                                                           |
- +---------------------------------------------------------------------------*/
-
-#include "exception.h"
-#include "fpu_emu.h"
-
-
-.text
-ENTRY(reg_div)
-	pushl	%ebp
-	movl	%esp,%ebp
-#ifndef NON_REENTRANT_FPU
-	subl	$28,%esp	/* Needed by divide_kernel */
-#endif NON_REENTRANT_FPU
-
-	pushl	%esi
-	pushl	%edi
-	pushl	%ebx
-
-	movl	PARAM1,%esi
-	movl	PARAM2,%ebx
-	movl	PARAM3,%edi
-
-	movb	TAG(%esi),%al
-	orb	TAG(%ebx),%al
-
-	jne	L_div_special		/* Not (both numbers TW_Valid) */
-
-#ifdef DENORM_OPERAND
-/* Check for denormals */
-	cmpl	EXP_UNDER,EXP(%esi)
-	jg	xL_arg1_not_denormal
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-
-xL_arg1_not_denormal:
-	cmpl	EXP_UNDER,EXP(%ebx)
-	jg	xL_arg2_not_denormal
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-
-xL_arg2_not_denormal:
-#endif DENORM_OPERAND
-
-/* Both arguments are TW_Valid */
-	movb	TW_Valid,TAG(%edi)
-
-	movb	SIGN(%esi),%cl
-	cmpb	%cl,SIGN(%ebx)
-	setne	(%edi)	      /* Set the sign, requires SIGN_NEG=1, SIGN_POS=0 */
-
-	movl	EXP(%esi),%edx
-	movl	EXP(%ebx),%eax
-	subl	%eax,%edx
-	addl	EXP_BIAS,%edx
-	movl	%edx,EXP(%edi)
-
-	jmp	SYMBOL_NAME(divide_kernel)
-
-
-/*-----------------------------------------------------------------------*/
-L_div_special:
-	cmpb	TW_NaN,TAG(%esi)	/* A NaN with anything to give NaN */
-	je	L_arg1_NaN
-
-	cmpb	TW_NaN,TAG(%ebx)	/* A NaN with anything to give NaN */
-	jne	L_no_NaN_arg
-
-/* Operations on NaNs */
-L_arg1_NaN:
-L_arg2_NaN:
-	pushl	%edi			/* Destination */
-	pushl	%esi
-	pushl	%ebx			/* Ordering is important here */
-	call	SYMBOL_NAME(real_2op_NaN)
-	jmp	LDiv_exit
-
-/* Invalid operations */
-L_zero_zero:
-L_inf_inf:
-	pushl	%edi			/* Destination */
-	call	SYMBOL_NAME(arith_invalid) /* 0/0 or Infinity/Infinity */
-	jmp	LDiv_exit
-
-L_no_NaN_arg:
-	cmpb	TW_Infinity,TAG(%esi)
-	jne	L_arg1_not_inf
-
-	cmpb	TW_Infinity,TAG(%ebx)
-	je	L_inf_inf		/* invalid operation */
-
-	cmpb	TW_Valid,TAG(%ebx)
-	je	L_inf_valid
-
-#ifdef PARANOID
-	/* arg2 must be zero or valid */
-	cmpb	TW_Zero,TAG(%ebx)
-	ja	L_unknown_tags
-#endif PARANOID
-
-	/* Note that p16-9 says that infinity/0 returns infinity */
-	jmp	L_copy_arg1		/* Answer is Inf */
-
-L_inf_valid:
-#ifdef DENORM_OPERAND
-	cmpl	EXP_UNDER,EXP(%ebx)
-	jg	L_copy_arg1		/* Answer is Inf */
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-#endif DENORM_OPERAND
-
-	jmp	L_copy_arg1		/* Answer is Inf */
-
-L_arg1_not_inf:
-	cmpb	TW_Zero,TAG(%ebx)	/* Priority to div-by-zero error */
-	jne	L_arg2_not_zero
-
-	cmpb	TW_Zero,TAG(%esi)
-	je	L_zero_zero		/* invalid operation */
-
-#ifdef PARANOID
-	/* arg1 must be valid */
-	cmpb	TW_Valid,TAG(%esi)
-	ja	L_unknown_tags
-#endif PARANOID
-
-/* Division by zero error */
-	pushl	%edi			/* destination */
-	movb	SIGN(%esi),%al
-	xorb	SIGN(%ebx),%al
-	pushl	%eax			/* lower 8 bits have the sign */
-	call	SYMBOL_NAME(divide_by_zero)
-	jmp	LDiv_exit
-
-L_arg2_not_zero:
-	cmpb	TW_Infinity,TAG(%ebx)
-	jne	L_arg2_not_inf
-
-#ifdef DENORM_OPERAND
-	cmpb	TW_Valid,TAG(%esi)
-	jne	L_return_zero
-
-	cmpl	EXP_UNDER,EXP(%esi)
-	jg	L_return_zero		/* Answer is zero */
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-#endif DENORM_OPERAND
-
-	jmp	L_return_zero		/* Answer is zero */
-
-L_arg2_not_inf:
-
-#ifdef PARANOID
-	cmpb	TW_Zero,TAG(%esi)
-	jne	L_unknown_tags
-#endif PARANOID
-
-	/* arg1 is zero, arg2 is not Infinity or a NaN */
-
-#ifdef DENORM_OPERAND
-	cmpl	EXP_UNDER,EXP(%ebx)
-	jg	L_copy_arg1		/* Answer is zero */
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-#endif DENORM_OPERAND
-
-L_copy_arg1:
-	movb	TAG(%esi),%ax
-	movb	%ax,TAG(%edi)
-	movl	EXP(%esi),%eax
-	movl	%eax,EXP(%edi)
-	movl	SIGL(%esi),%eax
-	movl	%eax,SIGL(%edi)
-	movl	SIGH(%esi),%eax
-	movl	%eax,SIGH(%edi)
-
-LDiv_set_result_sign:
-	movb	SIGN(%esi),%cl
-	cmpb	%cl,SIGN(%ebx)
-	jne	LDiv_negative_result
-
-	movb	SIGN_POS,SIGN(%edi)
-	xorl	%eax,%eax		/* Valid result */
-	jmp	LDiv_exit
-
-LDiv_negative_result:
-	movb	SIGN_NEG,SIGN(%edi)
-	xorl	%eax,%eax		/* Valid result */
-
-LDiv_exit:
-#ifndef NON_REENTRANT_FPU
-	leal	-40(%ebp),%esp
-#else
-	leal	-12(%ebp),%esp
-#endif NON_REENTRANT_FPU
-
-	popl	%ebx
-	popl	%edi
-	popl	%esi
-	leave
-	ret
-
-
-L_return_zero:
-	xorl	%eax,%eax
-	movl	%eax,SIGH(%edi)
-	movl	%eax,SIGL(%edi)
-	movl	EXP_UNDER,EXP(%edi)
-	movb	TW_Zero,TAG(%edi)
-	jmp	LDiv_set_result_sign
-
-#ifdef PARANOID
-L_unknown_tags:
-	pushl	EX_INTERNAL | 0x208
-	call	EXCEPTION
-
-	/* Generate a NaN for unknown tags */
-	movl	SYMBOL_NAME(CONST_QNaN),%eax
-	movl	%eax,(%edi)
-	movl	SYMBOL_NAME(CONST_QNaN)+4,%eax
-	movl	%eax,SIGL(%edi)
-	movl	SYMBOL_NAME(CONST_QNaN)+8,%eax
-	movl	%eax,SIGH(%edi)
-	jmp	LDiv_exit		/* %eax is nz */
-#endif PARANOID
diff --git a/arch/i386/math-emu/reg_divide.c b/arch/i386/math-emu/reg_divide.c
new file mode 100644
index 000000000..266775aca
--- /dev/null
+++ b/arch/i386/math-emu/reg_divide.c
@@ -0,0 +1,206 @@
+/*---------------------------------------------------------------------------+
+ |  reg_divide.c                                                             |
+ |                                                                           |
+ | Divide one FPU_REG by another and put the result in a destination FPU_REG.|
+ |                                                                           |
+ | Copyright (C) 1996                                                        |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@jacobi.maths.monash.edu.au                |
+ |                                                                           |
+ |    Return value is the tag of the answer, or-ed with FPU_Exception if     |
+ |    one was raised, or -1 on internal error.                               |
+ |                                                                           |
+ +---------------------------------------------------------------------------*/
+
+/*---------------------------------------------------------------------------+
+ | The destination may be any FPU_REG, including one of the source FPU_REGs. |
+ +---------------------------------------------------------------------------*/
+
+#include "exception.h"
+#include "reg_constant.h"
+#include "fpu_emu.h"
+#include "fpu_system.h"
+
+/*
+  Divide one register by another and put the result into a third register.
+  */
+int FPU_div(int flags, int rm, int control_w)
+{
+  FPU_REG x, y;
+  FPU_REG const *a, *b, *st0_ptr, *st_ptr;
+  FPU_REG *dest;
+  u_char taga, tagb, signa, signb, sign, saved_sign;
+  int tag, deststnr;
+
+  if ( flags & DEST_RM )
+    deststnr = rm;
+  else
+    deststnr = 0;
+
+  if ( flags & REV )
+    {
+      b = &st(0);
+      st0_ptr = b;
+      tagb = FPU_gettag0();
+      if ( flags & LOADED )
+	{
+	  a = (FPU_REG *)rm;
+	  taga = flags & 0x0f;
+	}
+      else
+	{
+	  a = &st(rm);
+	  st_ptr = a;
+	  taga = FPU_gettagi(rm);
+	}
+    }
+  else
+    {
+      a = &st(0);
+      st0_ptr = a;
+      taga = FPU_gettag0();
+      if ( flags & LOADED )
+	{
+	  b = (FPU_REG *)rm;
+	  tagb = flags & 0x0f;
+	}
+      else
+	{
+	  b = &st(rm);
+	  st_ptr = b;
+	  tagb = FPU_gettagi(rm);
+	}
+    }
+
+  signa = getsign(a);
+  signb = getsign(b);
+
+  sign = signa ^ signb;
+
+  dest = &st(deststnr);
+  saved_sign = getsign(dest);
+
+  if ( !(taga | tagb) )
+    {
+      /* Both regs Valid, this should be the most common case. */
+      reg_copy(a, &x);
+      reg_copy(b, &y);
+      setpositive(&x);
+      setpositive(&y);
+      tag = FPU_u_div(&x, &y, dest, control_w, sign);
+
+      if ( tag < 0 )
+	return tag;
+
+      FPU_settagi(deststnr, tag);
+      return tag;
+    }
+
+  if ( taga == TAG_Special )
+    taga = FPU_Special(a);
+  if ( tagb == TAG_Special )
+    tagb = FPU_Special(b);
+
+  if ( ((taga == TAG_Valid) && (tagb == TW_Denormal))
+	    || ((taga == TW_Denormal) && (tagb == TAG_Valid))
+	    || ((taga == TW_Denormal) && (tagb == TW_Denormal)) )
+    {
+      if ( denormal_operand() < 0 )
+	return FPU_Exception;
+
+      FPU_to_exp16(a, &x);
+      FPU_to_exp16(b, &y);
+      tag = FPU_u_div(&x, &y, dest, control_w, sign);
+      if ( tag < 0 )
+	return tag;
+
+      FPU_settagi(deststnr, tag);
+      return tag;
+    }
+  else if ( (taga <= TW_Denormal) && (tagb <= TW_Denormal) )
+    {
+      if ( tagb != TAG_Zero )
+	{
+	  /* Want to find Zero/Valid */
+	  if ( tagb == TW_Denormal )
+	    {
+	      if ( denormal_operand() < 0 )
+		return FPU_Exception;
+	    }
+
+	  /* The result is zero. */
+	  FPU_copy_to_regi(&CONST_Z, TAG_Zero, deststnr);
+	  setsign(dest, sign);
+	  return TAG_Zero;
+	}
+      /* We have an exception condition, either 0/0 or Valid/Zero. */
+      if ( taga == TAG_Zero )
+	{
+	  /* 0/0 */
+	  return arith_invalid(deststnr);
+	}
+      /* Valid/Zero */
+      return FPU_divide_by_zero(deststnr, sign);
+    }
+  /* Must have infinities, NaNs, etc */
+  else if ( (taga == TW_NaN) || (tagb == TW_NaN) )
+    {
+      if ( flags & LOADED )
+	return real_2op_NaN((FPU_REG *)rm, flags & 0x0f, 0, st0_ptr);
+
+      if ( flags & DEST_RM )
+	{
+	  int tag;
+	  tag = FPU_gettag0();
+	  if ( tag == TAG_Special )
+	    tag = FPU_Special(st0_ptr);
+	  return real_2op_NaN(st0_ptr, tag, rm, (flags & REV) ? st0_ptr : &st(rm));
+	}
+      else
+	{
+	  int tag;
+	  tag = FPU_gettagi(rm);
+	  if ( tag == TAG_Special )
+	    tag = FPU_Special(&st(rm));
+	  return real_2op_NaN(&st(rm), tag, 0, (flags & REV) ? st0_ptr : &st(rm));
+	}
+    }
+  else if (taga == TW_Infinity)
+    {
+      if (tagb == TW_Infinity)
+	{
+	  /* infinity/infinity */
+	  return arith_invalid(deststnr);
+	}
+      else
+	{
+	  /* tagb must be Valid or Zero */
+	  if ( (tagb == TW_Denormal) && (denormal_operand() < 0) )
+	    return FPU_Exception;
+	  
+	  /* Infinity divided by Zero or Valid does
+	     not raise and exception, but returns Infinity */
+	  FPU_copy_to_regi(a, TAG_Special, deststnr);
+	  setsign(dest, sign);
+	  return taga;
+	}
+    }
+  else if (tagb == TW_Infinity)
+    {
+      if ( (taga == TW_Denormal) && (denormal_operand() < 0) )
+	return FPU_Exception;
+
+      /* The result is zero. */
+      FPU_copy_to_regi(&CONST_Z, TAG_Zero, deststnr);
+      setsign(dest, sign);
+      return TAG_Zero;
+    }
+#ifdef PARANOID
+  else
+    {
+      EXCEPTION(EX_INTERNAL|0x102);
+      return FPU_Exception;
+    }
+#endif PARANOID
+
+}
diff --git a/arch/i386/math-emu/reg_ld_str.c b/arch/i386/math-emu/reg_ld_str.c
index 3e258a0c6..468e51cc8 100644
--- a/arch/i386/math-emu/reg_ld_str.c
+++ b/arch/i386/math-emu/reg_ld_str.c
@@ -3,9 +3,9 @@
  |                                                                           |
  | All of the functions which transfer data between user memory and FPU_REGs.|
  |                                                                           |
- | Copyright (C) 1992,1993,1994,1996                                         |
+ | Copyright (C) 1992,1993,1994,1996,1997                                    |
  |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
- |                  E-mail   billm@jacobi.maths.monash.edu.au                |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -17,19 +17,17 @@
  |    other processes using the emulator while swapping is in progress.      |
  +---------------------------------------------------------------------------*/
 
+#include "fpu_emu.h"
+
 #include <asm/uaccess.h>
 
 #include "fpu_system.h"
 #include "exception.h"
 #include "reg_constant.h"
-#include "fpu_emu.h"
 #include "control_w.h"
 #include "status_w.h"
 
 
-#define EXTENDED_Ebias 0x3fff
-#define EXTENDED_Emin (-0x3ffe)  /* smallest valid exponent */
-
 #define DOUBLE_Emax 1023         /* largest valid exponent */
 #define DOUBLE_Ebias 1023
 #define DOUBLE_Emin (-1022)      /* smallest valid exponent */
@@ -38,123 +36,85 @@
 #define SINGLE_Ebias 127
 #define SINGLE_Emin (-126)       /* smallest valid exponent */
 
-static void write_to_extended(FPU_REG *rp, char *d);
 
-
-/* Get a long double from user memory */
-int reg_load_extended(long double *s, FPU_REG *loaded_data)
+static u_char normalize_no_excep(FPU_REG *r, int exp, int sign)
 {
-  unsigned long sigl, sigh, exp;
+  u_char tag;
 
-  RE_ENTRANT_CHECK_OFF;
-  FPU_verify_area(VERIFY_READ, s, 10);
-  get_user(sigl, (unsigned long *) s);
-  get_user(sigh, 1 + (unsigned long *) s);
-  get_user(exp, 4 + (unsigned short *) s);
-  RE_ENTRANT_CHECK_ON;
+  setexponent16(r, exp);
 
-  loaded_data->tag = TW_Valid;   /* Default */
-  loaded_data->sigl = sigl;
-  loaded_data->sigh = sigh;
-  if (exp & 0x8000)
-    loaded_data->sign = SIGN_NEG;
-  else
-    loaded_data->sign = SIGN_POS;
-  exp &= 0x7fff;
-  loaded_data->exp = exp - EXTENDED_Ebias + EXP_BIAS;
+  tag = FPU_normalize_nuo(r);
+  stdexp(r);
+  if ( sign )
+    setnegative(r);
+
+  return tag;
+}
+
+
+int FPU_tagof(FPU_REG *ptr)
+{
+  int exp;
 
+  exp = exponent16(ptr) & 0x7fff;
   if ( exp == 0 )
     {
-      if ( !(sigh | sigl) )
+      if ( !(ptr->sigh | ptr->sigl) )
 	{
-	  loaded_data->tag = TW_Zero;
-	  return 0;
+	  return TAG_Zero;
 	}
       /* The number is a de-normal or pseudodenormal. */
-      if (sigh & 0x80000000)
-	{
-	  /* Is a pseudodenormal. */
-	  /* Convert it for internal use. */
-	  /* This is non-80486 behaviour because the number
-	     loses its 'denormal' identity. */
-	  loaded_data->exp++;
-	  return 1;
-	}
-      else
-	{
-	  /* Is a denormal. */
-	  /* Convert it for internal use. */
-	  loaded_data->exp++;
-	  normalize_nuo(loaded_data);
-	  return 0;
-	}
+      return TAG_Special;
     }
-  else if ( exp == 0x7fff )
-    {
-      if ( !((sigh ^ 0x80000000) | sigl) )
-	{
-	  /* Matches the bit pattern for Infinity. */
-	  loaded_data->exp = EXP_Infinity;
-	  loaded_data->tag = TW_Infinity;
-	  return 0;
-	}
 
-      loaded_data->exp = EXP_NaN;
-      loaded_data->tag = TW_NaN;
-      if ( !(sigh & 0x80000000) )
-	{
-	  /* NaNs have the ms bit set to 1. */
-	  /* This is therefore an Unsupported NaN data type. */
-	  /* This is non 80486 behaviour */
-	  /* This should generate an Invalid Operand exception
-	     later, so we convert it to a SNaN */
-	  loaded_data->sigh = 0x80000000;
-	  loaded_data->sigl = 0x00000001;
-	  loaded_data->sign = SIGN_NEG;
-	  return 1;
-	}
-      return 0;
+  if ( exp == 0x7fff )
+    {
+      /* Is an Infinity, a NaN, or an unsupported data type. */
+      return TAG_Special;
     }
 
-  if ( !(sigh & 0x80000000) )
+  if ( !(ptr->sigh & 0x80000000) )
     {
       /* Unsupported data type. */
       /* Valid numbers have the ms bit set to 1. */
       /* Unnormal. */
-      /* Convert it for internal use. */
-      /* This is non-80486 behaviour */
-      /* This should generate an Invalid Operand exception
-	 later, so we convert it to a SNaN */
-      loaded_data->sigh = 0x80000000;
-      loaded_data->sigl = 0x00000001;
-      loaded_data->sign = SIGN_NEG;
-      loaded_data->exp = EXP_NaN;
-      loaded_data->tag = TW_NaN;
-      return 1;
+      return TAG_Special;
     }
-  return 0;
+
+  return TAG_Valid;
+}
+
+
+/* Get a long double from user memory */
+int FPU_load_extended(long double *s, int stnr)
+{
+  FPU_REG *sti_ptr = &st(stnr);
+
+  RE_ENTRANT_CHECK_OFF;
+  FPU_verify_area(VERIFY_READ, s, 10);
+  __copy_from_user(sti_ptr, s, 10);
+  RE_ENTRANT_CHECK_ON;
+
+  return FPU_tagof(sti_ptr);
 }
 
 
 /* Get a double from user memory */
-int reg_load_double(double *dfloat, FPU_REG *loaded_data)
+int FPU_load_double(double *dfloat, FPU_REG *loaded_data)
 {
-  int exp;
+  int exp, tag, negative;
   unsigned m64, l64;
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_READ, dfloat, 8);
-  get_user(m64, 1 + (unsigned long *) dfloat);
-  get_user(l64, (unsigned long *) dfloat);
+  FPU_get_user(m64, 1 + (unsigned long *) dfloat);
+  FPU_get_user(l64, (unsigned long *) dfloat);
   RE_ENTRANT_CHECK_ON;
 
-  if (m64 & 0x80000000)
-    loaded_data->sign = SIGN_NEG;
-  else
-    loaded_data->sign = SIGN_POS;
-  exp = ((m64 & 0x7ff00000) >> 20) - DOUBLE_Ebias;
+  negative = (m64 & 0x80000000) ? SIGN_Negative : SIGN_Positive;
+  exp = ((m64 & 0x7ff00000) >> 20) - DOUBLE_Ebias + EXTENDED_Ebias;
   m64 &= 0xfffff;
-  if (exp > DOUBLE_Emax)
+  if ( exp > DOUBLE_Emax + EXTENDED_Ebias )
     {
       /* Infinity or NaN */
       if ((m64 == 0) && (l64 == 0))
@@ -162,93 +122,87 @@ int reg_load_double(double *dfloat, FPU_REG *loaded_data)
 	  /* +- infinity */
 	  loaded_data->sigh = 0x80000000;
 	  loaded_data->sigl = 0x00000000;
-	  loaded_data->exp = EXP_Infinity;
-	  loaded_data->tag = TW_Infinity;
-	  return 0;
+	  exp = EXP_Infinity + EXTENDED_Ebias;
+	  tag = TAG_Special;
 	}
       else
 	{
 	  /* Must be a signaling or quiet NaN */
-	  loaded_data->exp = EXP_NaN;
-	  loaded_data->tag = TW_NaN;
+	  exp = EXP_NaN + EXTENDED_Ebias;
 	  loaded_data->sigh = (m64 << 11) | 0x80000000;
 	  loaded_data->sigh |= l64 >> 21;
 	  loaded_data->sigl = l64 << 11;
-	  return 0; /* The calling function must look for NaNs */
+	  tag = TAG_Special;    /* The calling function must look for NaNs */
 	}
     }
-  else if ( exp < DOUBLE_Emin )
+  else if ( exp < DOUBLE_Emin + EXTENDED_Ebias )
     {
       /* Zero or de-normal */
       if ((m64 == 0) && (l64 == 0))
 	{
 	  /* Zero */
-	  int c = loaded_data->sign;
-	  reg_move(&CONST_Z, loaded_data);
-	  loaded_data->sign = c;
-	  return 0;
+	  reg_copy(&CONST_Z, loaded_data);
+	  exp = 0;
+	  tag = TAG_Zero;
 	}
       else
 	{
 	  /* De-normal */
-	  loaded_data->exp = DOUBLE_Emin + EXP_BIAS;
-	  loaded_data->tag = TW_Valid;
 	  loaded_data->sigh = m64 << 11;
 	  loaded_data->sigh |= l64 >> 21;
 	  loaded_data->sigl = l64 << 11;
-	  normalize_nuo(loaded_data);
-	  return denormal_operand();
+
+	  return normalize_no_excep(loaded_data, DOUBLE_Emin, negative)
+	    | (denormal_operand() < 0 ? FPU_Exception : 0);
 	}
     }
   else
     {
-      loaded_data->exp = exp + EXP_BIAS;
-      loaded_data->tag = TW_Valid;
       loaded_data->sigh = (m64 << 11) | 0x80000000;
       loaded_data->sigh |= l64 >> 21;
       loaded_data->sigl = l64 << 11;
 
-      return 0;
+      tag = TAG_Valid;
     }
+
+  setexponent16(loaded_data, exp | negative);
+
+  return tag;
 }
 
 
 /* Get a float from user memory */
-int reg_load_single(float *single, FPU_REG *loaded_data)
+int FPU_load_single(float *single, FPU_REG *loaded_data)
 {
   unsigned m32;
-  int exp;
+  int exp, tag, negative;
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_READ, single, 4);
-  get_user(m32, (unsigned long *) single);
+  FPU_get_user(m32, (unsigned long *) single);
   RE_ENTRANT_CHECK_ON;
 
-  if (m32 & 0x80000000)
-    loaded_data->sign = SIGN_NEG;
-  else
-    loaded_data->sign = SIGN_POS;
+  negative = (m32 & 0x80000000) ? SIGN_Negative : SIGN_Positive;
+
   if (!(m32 & 0x7fffffff))
     {
       /* Zero */
-      int c = loaded_data->sign;
-      reg_move(&CONST_Z, loaded_data);
-      loaded_data->sign = c;
-      return 0;
+      reg_copy(&CONST_Z, loaded_data);
+      addexponent(loaded_data, negative);
+      return TAG_Zero;
     }
-  exp = ((m32 & 0x7f800000) >> 23) - SINGLE_Ebias;
+  exp = ((m32 & 0x7f800000) >> 23) - SINGLE_Ebias + EXTENDED_Ebias;
   m32 = (m32 & 0x7fffff) << 8;
-  if ( exp < SINGLE_Emin )
+  if ( exp < SINGLE_Emin + EXTENDED_Ebias )
     {
       /* De-normals */
-      loaded_data->exp = SINGLE_Emin + EXP_BIAS;
-      loaded_data->tag = TW_Valid;
       loaded_data->sigh = m32;
       loaded_data->sigl = 0;
-      normalize_nuo(loaded_data);
-      return denormal_operand();
+
+      return normalize_no_excep(loaded_data, SINGLE_Emin, negative)
+	| (denormal_operand() < 0 ? FPU_Exception : 0);
     }
-  else if ( exp > SINGLE_Emax )
+  else if ( exp > SINGLE_Emax + EXTENDED_Ebias )
     {
     /* Infinity or NaN */
       if ( m32 == 0 )
@@ -256,36 +210,37 @@ int reg_load_single(float *single, FPU_REG *loaded_data)
 	  /* +- infinity */
 	  loaded_data->sigh = 0x80000000;
 	  loaded_data->sigl = 0x00000000;
-	  loaded_data->exp = EXP_Infinity;
-	  loaded_data->tag = TW_Infinity;
-	  return 0;
+	  exp = EXP_Infinity + EXTENDED_Ebias;
+	  tag = TAG_Special;
 	}
       else
 	{
 	  /* Must be a signaling or quiet NaN */
-	  loaded_data->exp = EXP_NaN;
-	  loaded_data->tag = TW_NaN;
+	  exp = EXP_NaN + EXTENDED_Ebias;
 	  loaded_data->sigh = m32 | 0x80000000;
 	  loaded_data->sigl = 0;
-	  return 0; /* The calling function must look for NaNs */
+	  tag = TAG_Special;  /* The calling function must look for NaNs */
 	}
     }
   else
     {
-      loaded_data->exp = exp + EXP_BIAS;
       loaded_data->sigh = m32 | 0x80000000;
       loaded_data->sigl = 0;
-      loaded_data->tag = TW_Valid;
-      return 0;
+      tag = TAG_Valid;
     }
+
+  setexponent16(loaded_data, exp | negative);  /* Set the sign. */
+
+  return tag;
 }
 
 
 /* Get a long long from user memory */
-void reg_load_int64(long long *_s, FPU_REG *loaded_data)
+int FPU_load_int64(long long *_s)
 {
-  int e;
   long long s;
+  int sign;
+  FPU_REG *st0_ptr = &st(0);
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_READ, _s, 8);
@@ -293,93 +248,91 @@ void reg_load_int64(long long *_s, FPU_REG *loaded_data)
   RE_ENTRANT_CHECK_ON;
 
   if (s == 0)
-    { reg_move(&CONST_Z, loaded_data); return; }
+    {
+      reg_copy(&CONST_Z, st0_ptr);
+      return TAG_Zero;
+    }
 
   if (s > 0)
-    loaded_data->sign = SIGN_POS;
+    sign = SIGN_Positive;
   else
   {
     s = -s;
-    loaded_data->sign = SIGN_NEG;
+    sign = SIGN_Negative;
   }
 
-  e = EXP_BIAS + 63;
-  significand(loaded_data) = s;
-  loaded_data->exp = e;
-  loaded_data->tag = TW_Valid;
-  normalize_nuo(loaded_data);
+  significand(st0_ptr) = s;
+
+  return normalize_no_excep(st0_ptr, 63, sign);
 }
 
 
 /* Get a long from user memory */
-void reg_load_int32(long *_s, FPU_REG *loaded_data)
+int FPU_load_int32(long *_s, FPU_REG *loaded_data)
 {
   long s;
-  int e;
+  int negative;
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_READ, _s, 4);
-  get_user(s, _s);
+  FPU_get_user(s, _s);
   RE_ENTRANT_CHECK_ON;
 
   if (s == 0)
-    { reg_move(&CONST_Z, loaded_data); return; }
+    { reg_copy(&CONST_Z, loaded_data); return TAG_Zero; }
 
   if (s > 0)
-    loaded_data->sign = SIGN_POS;
+    negative = SIGN_Positive;
   else
-  {
-    s = -s;
-    loaded_data->sign = SIGN_NEG;
-  }
+    {
+      s = -s;
+      negative = SIGN_Negative;
+    }
 
-  e = EXP_BIAS + 31;
   loaded_data->sigh = s;
   loaded_data->sigl = 0;
-  loaded_data->exp = e;
-  loaded_data->tag = TW_Valid;
-  normalize_nuo(loaded_data);
+
+  return normalize_no_excep(loaded_data, 31, negative);
 }
 
 
 /* Get a short from user memory */
-void reg_load_int16(short *_s, FPU_REG *loaded_data)
+int FPU_load_int16(short *_s, FPU_REG *loaded_data)
 {
-  int s, e;
+  int s, negative;
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_READ, _s, 2);
   /* Cast as short to get the sign extended. */
-  get_user(s, _s);
+  FPU_get_user(s, _s);
   RE_ENTRANT_CHECK_ON;
 
   if (s == 0)
-    { reg_move(&CONST_Z, loaded_data); return; }
+    { reg_copy(&CONST_Z, loaded_data); return TAG_Zero; }
 
   if (s > 0)
-    loaded_data->sign = SIGN_POS;
+    negative = SIGN_Positive;
   else
-  {
-    s = -s;
-    loaded_data->sign = SIGN_NEG;
-  }
+    {
+      s = -s;
+      negative = SIGN_Negative;
+    }
 
-  e = EXP_BIAS + 15;
   loaded_data->sigh = s << 16;
-
   loaded_data->sigl = 0;
-  loaded_data->exp = e;
-  loaded_data->tag = TW_Valid;
-  normalize_nuo(loaded_data);
+
+  return normalize_no_excep(loaded_data, 15, negative);
 }
 
 
 /* Get a packed bcd array from user memory */
-void reg_load_bcd(char *s, FPU_REG *loaded_data)
+int FPU_load_bcd(u_char *s)
 {
+  FPU_REG *st0_ptr = &st(0);
   int pos;
-  unsigned char bcd;
+  u_char bcd;
   long long l=0;
+  int sign;
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_READ, s, 10);
@@ -388,7 +341,7 @@ void reg_load_bcd(char *s, FPU_REG *loaded_data)
     {
       l *= 10;
       RE_ENTRANT_CHECK_OFF;
-      get_user(bcd, (unsigned char *) s+pos);
+      FPU_get_user(bcd, (u_char *) s+pos);
       RE_ENTRANT_CHECK_ON;
       l += bcd >> 4;
       l *= 10;
@@ -396,32 +349,27 @@ void reg_load_bcd(char *s, FPU_REG *loaded_data)
     }
  
   RE_ENTRANT_CHECK_OFF;
-  {
-    unsigned char sign;
-    get_user(sign, (unsigned char *) s+9);
-    loaded_data->sign = (sign & 0x80) ? SIGN_NEG : SIGN_POS;
-  }
+  FPU_get_user(sign, (u_char *) s+9);
+  sign = sign & 0x80 ? SIGN_Negative : SIGN_Positive;
   RE_ENTRANT_CHECK_ON;
 
-  if (l == 0)
+  if ( l == 0 )
     {
-      char sign = loaded_data->sign;
-      reg_move(&CONST_Z, loaded_data);
-      loaded_data->sign = sign;
+      reg_copy(&CONST_Z, st0_ptr);
+      addexponent(st0_ptr, sign);   /* Set the sign. */
+      return TAG_Zero;
     }
   else
     {
-      significand(loaded_data) = l;
-      loaded_data->exp = EXP_BIAS + 63;
-      loaded_data->tag = TW_Valid;
-      normalize_nuo(loaded_data);
+      significand(st0_ptr) = l;
+      return normalize_no_excep(st0_ptr, 63, sign);
     }
 }
 
 /*===========================================================================*/
 
 /* Put a long double into user memory */
-int reg_store_extended(long double *d, FPU_REG *st0_ptr)
+int FPU_store_extended(FPU_REG *st0_ptr, u_char st0_tag, long double *d)
 {
   /*
     The only exception raised by an attempt to store to an
@@ -429,12 +377,16 @@ int reg_store_extended(long double *d, FPU_REG *st0_ptr)
     attempting to store from an empty register.
    */
 
-  if ( st0_ptr->tag != TW_Empty )
+  if ( st0_tag != TAG_Empty )
     {
       RE_ENTRANT_CHECK_OFF;
       FPU_verify_area(VERIFY_WRITE, d, 10);
+
+      FPU_put_user(st0_ptr->sigl, (unsigned long *) d);
+      FPU_put_user(st0_ptr->sigh, (unsigned long *) ((u_char *)d + 4));
+      FPU_put_user(exponent16(st0_ptr), (unsigned short *) ((u_char *)d + 8));
       RE_ENTRANT_CHECK_ON;
-      write_to_extended(st0_ptr, (char *) d);
+
       return 1;
     }
 
@@ -446,9 +398,9 @@ int reg_store_extended(long double *d, FPU_REG *st0_ptr)
       /* Put out the QNaN indefinite */
       RE_ENTRANT_CHECK_OFF;
       FPU_verify_area(VERIFY_WRITE,d,10);
-      put_user(0, (unsigned long *) d);
-      put_user(0xc0000000, 1 + (unsigned long *) d);
-      put_user(0xffff, 4 + (short *) d);
+      FPU_put_user(0, (unsigned long *) d);
+      FPU_put_user(0xc0000000, 1 + (unsigned long *) d);
+      FPU_put_user(0xffff, 4 + (short *) d);
       RE_ENTRANT_CHECK_ON;
       return 1;
     }
@@ -459,38 +411,26 @@ int reg_store_extended(long double *d, FPU_REG *st0_ptr)
 
 
 /* Put a double into user memory */
-int reg_store_double(double *dfloat, FPU_REG *st0_ptr)
+int FPU_store_double(FPU_REG *st0_ptr, u_char st0_tag, double *dfloat)
 {
   unsigned long l[2];
   unsigned long increment = 0;	/* avoid gcc warnings */
-  char st0_tag = st0_ptr->tag;
+  int precision_loss;
+  int exp;
+  FPU_REG tmp;
 
-  if (st0_tag == TW_Valid)
+  if ( st0_tag == TAG_Valid )
     {
-      int precision_loss;
-      int exp;
-      FPU_REG tmp;
-
-      reg_move(st0_ptr, &tmp);
-      exp = tmp.exp - EXP_BIAS;
+      reg_copy(st0_ptr, &tmp);
+      exp = exponent(&tmp);
 
       if ( exp < DOUBLE_Emin )     /* It may be a denormal */
 	{
-	  /* A denormal will always underflow. */
-#ifndef PECULIAR_486
-	  /* An 80486 is supposed to be able to generate
-	     a denormal exception here, but... */
-	  if ( st0_ptr->exp <= EXP_UNDER )
-	    {
-	      /* Underflow has priority. */
-	      if ( control_word & CW_Underflow )
-		denormal_operand();
-	    }
-#endif PECULIAR_486
+	  addexponent(&tmp, -DOUBLE_Emin + 52);  /* largest exp to be 51 */
 
-	  tmp.exp += -DOUBLE_Emin + 52;  /* largest exp to be 51 */
+	denormal_arg:
 
-	  if ( (precision_loss = round_to_int(&tmp)) )
+	  if ( (precision_loss = FPU_round_to_int(&tmp, st0_tag)) )
 	    {
 #ifdef PECULIAR_486
 	      /* Did it round to a non-denormal ? */
@@ -527,10 +467,10 @@ int reg_store_double(double *dfloat, FPU_REG *st0_ptr)
 		    ((tmp.sigl & 0xc00) == 0xc00);            /* odd -> even */
 		  break;
 		case RC_DOWN:   /* towards -infinity */
-		  increment = (tmp.sign == SIGN_POS) ? 0 : tmp.sigl & 0x7ff;
+		  increment = signpositive(&tmp) ? 0 : tmp.sigl & 0x7ff;
 		  break;
 		case RC_UP:     /* towards +infinity */
-		  increment = (tmp.sign == SIGN_POS) ? tmp.sigl & 0x7ff : 0;
+		  increment = signpositive(&tmp) ? tmp.sigl & 0x7ff : 0;
 		  break;
 		case RC_CHOP:
 		  increment = 0;
@@ -601,33 +541,64 @@ int reg_store_double(double *dfloat, FPU_REG *st0_ptr)
 	    }
 	}
     }
-  else if (st0_tag == TW_Zero)
+  else if (st0_tag == TAG_Zero)
     {
       /* Number is zero */
       l[0] = 0;
       l[1] = 0;
     }
-  else if (st0_tag == TW_Infinity)
-    {
-      l[0] = 0;
-      l[1] = 0x7ff00000;
-    }
-  else if (st0_tag == TW_NaN)
+  else if ( st0_tag == TAG_Special )
     {
-      /* See if we can get a valid NaN from the FPU_REG */
-      l[0] = (st0_ptr->sigl >> 11) | (st0_ptr->sigh << 21);
-      l[1] = ((st0_ptr->sigh >> 11) & 0xfffff);
-      if ( !(st0_ptr->sigh & 0x40000000) )
+      st0_tag = FPU_Special(st0_ptr);
+      if ( st0_tag == TW_Denormal )
 	{
-	  /* It is a signalling NaN */
-	  EXCEPTION(EX_Invalid);
-	  if ( !(control_word & CW_Invalid) )
-	    return 0;
-	  l[1] |= (0x40000000 >> 11);
+	  /* A denormal will always underflow. */
+#ifndef PECULIAR_486
+	  /* An 80486 is supposed to be able to generate
+	     a denormal exception here, but... */
+	  /* Underflow has priority. */
+	  if ( control_word & CW_Underflow )
+	    denormal_operand();
+#endif PECULIAR_486
+	  reg_copy(st0_ptr, &tmp);
+	  goto denormal_arg;
+	}
+      else if (st0_tag == TW_Infinity)
+	{
+	  l[0] = 0;
+	  l[1] = 0x7ff00000;
+	}
+      else if (st0_tag == TW_NaN)
+	{
+	  /* Is it really a NaN ? */
+	  if ( (exponent(st0_ptr) == EXP_OVER)
+	       && (st0_ptr->sigh & 0x80000000) )
+	    {
+	      /* See if we can get a valid NaN from the FPU_REG */
+	      l[0] = (st0_ptr->sigl >> 11) | (st0_ptr->sigh << 21);
+	      l[1] = ((st0_ptr->sigh >> 11) & 0xfffff);
+	      if ( !(st0_ptr->sigh & 0x40000000) )
+		{
+		  /* It is a signalling NaN */
+		  EXCEPTION(EX_Invalid);
+		  if ( !(control_word & CW_Invalid) )
+		    return 0;
+		  l[1] |= (0x40000000 >> 11);
+		}
+	      l[1] |= 0x7ff00000;
+	    }
+	  else
+	    {
+	      /* It is an unsupported data type */
+	      EXCEPTION(EX_Invalid);
+	      if ( !(control_word & CW_Invalid) )
+		return 0;
+	      l[0] = 0;
+	      l[1] = 0xfff80000;
+	    }
 	}
-      l[1] |= 0x7ff00000;
     }
-  else if ( st0_tag == TW_Empty )
+  else if ( st0_tag == TAG_Empty )
     {
       /* Empty register (stack underflow) */
       EXCEPTION(EX_StackUnder);
@@ -637,21 +608,21 @@ int reg_store_double(double *dfloat, FPU_REG *st0_ptr)
 	  /* Put out the QNaN indefinite */
 	  RE_ENTRANT_CHECK_OFF;
 	  FPU_verify_area(VERIFY_WRITE,(void *)dfloat,8);
-	  put_user(0, (unsigned long *) dfloat);
-	  put_user(0xfff80000, 1 + (unsigned long *) dfloat);
+	  FPU_put_user(0, (unsigned long *) dfloat);
+	  FPU_put_user(0xfff80000, 1 + (unsigned long *) dfloat);
 	  RE_ENTRANT_CHECK_ON;
 	  return 1;
 	}
       else
 	return 0;
     }
-  if ( st0_ptr->sign )
+  if ( getsign(st0_ptr) )
     l[1] |= 0x80000000;
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_WRITE,(void *)dfloat,8);
-  put_user(l[0], (unsigned long *)dfloat);
-  put_user(l[1], 1 + (unsigned long *)dfloat);
+  FPU_put_user(l[0], (unsigned long *)dfloat);
+  FPU_put_user(l[1], 1 + (unsigned long *)dfloat);
   RE_ENTRANT_CHECK_ON;
 
   return 1;
@@ -659,38 +630,27 @@ int reg_store_double(double *dfloat, FPU_REG *st0_ptr)
 
 
 /* Put a float into user memory */
-int reg_store_single(float *single, FPU_REG *st0_ptr)
+int FPU_store_single(FPU_REG *st0_ptr, u_char st0_tag, float *single)
 {
   long templ;
   unsigned long increment = 0;     	/* avoid gcc warnings */
-  char st0_tag = st0_ptr->tag;
+  int precision_loss;
+  int exp;
+  FPU_REG tmp;
 
-  if (st0_tag == TW_Valid)
+  if ( st0_tag == TAG_Valid )
     {
-      int precision_loss;
-      int exp;
-      FPU_REG tmp;
 
-      reg_move(st0_ptr, &tmp);
-      exp = tmp.exp - EXP_BIAS;
+      reg_copy(st0_ptr, &tmp);
+      exp = exponent(&tmp);
 
       if ( exp < SINGLE_Emin )
 	{
-	  /* A denormal will always underflow. */
-#ifndef PECULIAR_486
-	  /* An 80486 is supposed to be able to generate
-	     a denormal exception here, but... */
-	  if ( st0_ptr->exp <= EXP_UNDER )
-	    {
-	      /* Underflow has priority. */
-	      if ( control_word & CW_Underflow )
-		denormal_operand();
-	    }
-#endif PECULIAR_486
+	  addexponent(&tmp, -SINGLE_Emin + 23);  /* largest exp to be 22 */
 
-	  tmp.exp += -SINGLE_Emin + 23;  /* largest exp to be 22 */
+	denormal_arg:
 
-	  if ( (precision_loss = round_to_int(&tmp)) )
+	  if ( (precision_loss = FPU_round_to_int(&tmp, st0_tag)) )
 	    {
 #ifdef PECULIAR_486
 	      /* Did it round to a non-denormal ? */
@@ -704,15 +664,15 @@ int reg_store_single(float *single, FPU_REG *st0_ptr)
 		  EXCEPTION(EX_Underflow);
 		  /* This is a special case: see sec 16.2.5.1 of
 		     the 80486 book */
-		  if ( !(control_word & EX_Underflow) )
+		  if ( !(control_word & CW_Underflow) )
 		    return 0;
 		}
 	      EXCEPTION(precision_loss);
-	      if ( !(control_word & EX_Precision) )
+	      if ( !(control_word & CW_Precision) )
 		return 0;
 	    }
 	  templ = tmp.sigl;
-	}
+      }
       else
 	{
 	  if ( tmp.sigl | (tmp.sigh & 0x000000ff) )
@@ -726,15 +686,15 @@ int reg_store_single(float *single, FPU_REG *st0_ptr)
 		case RC_RND:
 		  increment = ((sigh & 0xff) > 0x80)       /* more than half */
 		    || (((sigh & 0xff) == 0x80) && sigl)   /* more than half */
-		      || ((sigh & 0x180) == 0x180);        /* round to even */
+		    || ((sigh & 0x180) == 0x180);        /* round to even */
 		  break;
 		case RC_DOWN:   /* towards -infinity */
-		  increment = (tmp.sign == SIGN_POS)
-		              ? 0 : (sigl | (sigh & 0xff));
+		  increment = signpositive(&tmp)
+		    ? 0 : (sigl | (sigh & 0xff));
 		  break;
 		case RC_UP:     /* towards +infinity */
-		  increment = (tmp.sign == SIGN_POS)
-		              ? (sigl | (sigh & 0xff)) : 0;
+		  increment = signpositive(&tmp)
+		    ? (sigl | (sigh & 0xff)) : 0;
 		  break;
 		case RC_CHOP:
 		  increment = 0;
@@ -767,7 +727,7 @@ int reg_store_single(float *single, FPU_REG *st0_ptr)
 	    }
 	  else
 	    precision_loss = 0;
-
+      
 	  templ = (tmp.sigh >> 8) & 0x007fffff;
 
 	  if ( exp > SINGLE_Emax )
@@ -798,29 +758,66 @@ int reg_store_single(float *single, FPU_REG *st0_ptr)
 	    }
 	}
     }
-  else if (st0_tag == TW_Zero)
+  else if (st0_tag == TAG_Zero)
     {
       templ = 0;
     }
-  else if (st0_tag == TW_Infinity)
+  else if ( st0_tag == TAG_Special )
     {
-      templ = 0x7f800000;
-    }
-  else if (st0_tag == TW_NaN)
-    {
-      /* See if we can get a valid NaN from the FPU_REG */
-      templ = st0_ptr->sigh >> 8;
-      if ( !(st0_ptr->sigh & 0x40000000) )
+      st0_tag = FPU_Special(st0_ptr);
+      if (st0_tag == TW_Denormal)
 	{
-	  /* It is a signalling NaN */
-	  EXCEPTION(EX_Invalid);
-	  if ( !(control_word & CW_Invalid) )
-	    return 0;
-	  templ |= (0x40000000 >> 8);
+	  reg_copy(st0_ptr, &tmp);
+
+	  /* A denormal will always underflow. */
+#ifndef PECULIAR_486
+	  /* An 80486 is supposed to be able to generate
+	     a denormal exception here, but... */
+	  /* Underflow has priority. */
+	  if ( control_word & CW_Underflow )
+	    denormal_operand();
+#endif PECULIAR_486
+	  goto denormal_arg;
+	}
+      else if (st0_tag == TW_Infinity)
+	{
+	  templ = 0x7f800000;
 	}
-      templ |= 0x7f800000;
+      else if (st0_tag == TW_NaN)
+	{
+	  /* Is it really a NaN ? */
+	  if ( (exponent(st0_ptr) == EXP_OVER) && (st0_ptr->sigh & 0x80000000) )
+	    {
+	      /* See if we can get a valid NaN from the FPU_REG */
+	      templ = st0_ptr->sigh >> 8;
+	      if ( !(st0_ptr->sigh & 0x40000000) )
+		{
+		  /* It is a signalling NaN */
+		  EXCEPTION(EX_Invalid);
+		  if ( !(control_word & CW_Invalid) )
+		    return 0;
+		  templ |= (0x40000000 >> 8);
+		}
+	      templ |= 0x7f800000;
+	    }
+	  else
+	    {
+	      /* It is an unsupported data type */
+	      EXCEPTION(EX_Invalid);
+	      if ( !(control_word & CW_Invalid) )
+		return 0;
+	      templ = 0xffc00000;
+	    }
+	}
+#ifdef PARANOID
+      else
+	{
+	  EXCEPTION(EX_INTERNAL|0x164);
+	  return 0;
+	}
+#endif
     }
-  else if ( st0_tag == TW_Empty )
+  else if ( st0_tag == TAG_Empty )
     {
       /* Empty register (stack underflow) */
       EXCEPTION(EX_StackUnder);
@@ -830,7 +827,7 @@ int reg_store_single(float *single, FPU_REG *st0_ptr)
 	  /* Put out the QNaN indefinite */
 	  RE_ENTRANT_CHECK_OFF;
 	  FPU_verify_area(VERIFY_WRITE,(void *)single,4);
-	  put_user(0xffc00000, (unsigned long *) single);
+	  FPU_put_user(0xffc00000, (unsigned long *) single);
 	  RE_ENTRANT_CHECK_ON;
 	  return 1;
 	}
@@ -844,12 +841,12 @@ int reg_store_single(float *single, FPU_REG *st0_ptr)
       return 0;
     }
 #endif
-  if (st0_ptr->sign)
+  if ( getsign(st0_ptr) )
     templ |= 0x80000000;
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_WRITE,(void *)single,4);
-  put_user(templ,(unsigned long *) single);
+  FPU_put_user(templ,(unsigned long *) single);
   RE_ENTRANT_CHECK_ON;
 
   return 1;
@@ -857,34 +854,37 @@ int reg_store_single(float *single, FPU_REG *st0_ptr)
 
 
 /* Put a long long into user memory */
-int reg_store_int64(long long *d, FPU_REG *st0_ptr)
+int FPU_store_int64(FPU_REG *st0_ptr, u_char st0_tag, long long *d)
 {
   FPU_REG t;
   long long tll;
   int precision_loss;
-  char st0_tag = st0_ptr->tag;
 
-  if ( st0_tag == TW_Empty )
+  if ( st0_tag == TAG_Empty )
     {
       /* Empty register (stack underflow) */
       EXCEPTION(EX_StackUnder);
       goto invalid_operand;
     }
-  else if ( (st0_tag == TW_Infinity) ||
-	   (st0_tag == TW_NaN) )
+  else if ( st0_tag == TAG_Special )
     {
-      EXCEPTION(EX_Invalid);
-      goto invalid_operand;
+      st0_tag = FPU_Special(st0_ptr);
+      if ( (st0_tag == TW_Infinity) ||
+	   (st0_tag == TW_NaN) )
+	{
+	  EXCEPTION(EX_Invalid);
+	  goto invalid_operand;
+	}
     }
 
-  reg_move(st0_ptr, &t);
-  precision_loss = round_to_int(&t);
+  reg_copy(st0_ptr, &t);
+  precision_loss = FPU_round_to_int(&t, st0_tag);
   ((long *)&tll)[0] = t.sigl;
   ((long *)&tll)[1] = t.sigh;
   if ( (precision_loss == 1) ||
       ((t.sigh & 0x80000000) &&
        !((t.sigh == 0x80000000) && (t.sigl == 0) &&
-	 (t.sign == SIGN_NEG))) )
+	 signnegative(&t))) )
     {
       EXCEPTION(EX_Invalid);
       /* This is a special case: see sec 16.2.5.1 of the 80486 book */
@@ -901,7 +901,7 @@ int reg_store_int64(long long *d, FPU_REG *st0_ptr)
     {
       if ( precision_loss )
 	set_precision_flag(precision_loss);
-      if ( t.sign )
+      if ( signnegative(&t) )
 	tll = - tll;
     }
 
@@ -915,30 +915,33 @@ int reg_store_int64(long long *d, FPU_REG *st0_ptr)
 
 
 /* Put a long into user memory */
-int reg_store_int32(long *d, FPU_REG *st0_ptr)
+int FPU_store_int32(FPU_REG *st0_ptr, u_char st0_tag, long *d)
 {
   FPU_REG t;
   int precision_loss;
-  char st0_tag = st0_ptr->tag;
 
-  if ( st0_tag == TW_Empty )
+  if ( st0_tag == TAG_Empty )
     {
       /* Empty register (stack underflow) */
       EXCEPTION(EX_StackUnder);
       goto invalid_operand;
     }
-  else if ( (st0_tag == TW_Infinity) ||
-	   (st0_tag == TW_NaN) )
+  else if ( st0_tag == TAG_Special )
     {
-      EXCEPTION(EX_Invalid);
-      goto invalid_operand;
+      st0_tag = FPU_Special(st0_ptr);
+      if ( (st0_tag == TW_Infinity) ||
+	   (st0_tag == TW_NaN) )
+	{
+	  EXCEPTION(EX_Invalid);
+	  goto invalid_operand;
+	}
     }
 
-  reg_move(st0_ptr, &t);
-  precision_loss = round_to_int(&t);
+  reg_copy(st0_ptr, &t);
+  precision_loss = FPU_round_to_int(&t, st0_tag);
   if (t.sigh ||
       ((t.sigl & 0x80000000) &&
-       !((t.sigl == 0x80000000) && (t.sign == SIGN_NEG))) )
+       !((t.sigl == 0x80000000) && signnegative(&t))) )
     {
       EXCEPTION(EX_Invalid);
       /* This is a special case: see sec 16.2.5.1 of the 80486 book */
@@ -955,13 +958,13 @@ int reg_store_int32(long *d, FPU_REG *st0_ptr)
     {
       if ( precision_loss )
 	set_precision_flag(precision_loss);
-      if ( t.sign )
+      if ( signnegative(&t) )
 	t.sigl = -(long)t.sigl;
     }
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_WRITE,d,4);
-  put_user(t.sigl, (unsigned long *) d);
+  FPU_put_user(t.sigl, (unsigned long *) d);
   RE_ENTRANT_CHECK_ON;
 
   return 1;
@@ -969,30 +972,33 @@ int reg_store_int32(long *d, FPU_REG *st0_ptr)
 
 
 /* Put a short into user memory */
-int reg_store_int16(short *d, FPU_REG *st0_ptr)
+int FPU_store_int16(FPU_REG *st0_ptr, u_char st0_tag, short *d)
 {
   FPU_REG t;
   int precision_loss;
-  char st0_tag = st0_ptr->tag;
 
-  if ( st0_tag == TW_Empty )
+  if ( st0_tag == TAG_Empty )
     {
       /* Empty register (stack underflow) */
       EXCEPTION(EX_StackUnder);
       goto invalid_operand;
     }
-  else if ( (st0_tag == TW_Infinity) ||
-	   (st0_tag == TW_NaN) )
+  else if ( st0_tag == TAG_Special )
     {
-      EXCEPTION(EX_Invalid);
-      goto invalid_operand;
+      st0_tag = FPU_Special(st0_ptr);
+      if ( (st0_tag == TW_Infinity) ||
+	   (st0_tag == TW_NaN) )
+	{
+	  EXCEPTION(EX_Invalid);
+	  goto invalid_operand;
+	}
     }
 
-  reg_move(st0_ptr, &t);
-  precision_loss = round_to_int(&t);
+  reg_copy(st0_ptr, &t);
+  precision_loss = FPU_round_to_int(&t, st0_tag);
   if (t.sigh ||
       ((t.sigl & 0xffff8000) &&
-       !((t.sigl == 0x8000) && (t.sign == SIGN_NEG))) )
+       !((t.sigl == 0x8000) && signnegative(&t))) )
     {
       EXCEPTION(EX_Invalid);
       /* This is a special case: see sec 16.2.5.1 of the 80486 book */
@@ -1009,13 +1015,13 @@ int reg_store_int16(short *d, FPU_REG *st0_ptr)
     {
       if ( precision_loss )
 	set_precision_flag(precision_loss);
-      if ( t.sign )
+      if ( signnegative(&t) )
 	t.sigl = -t.sigl;
     }
 
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_WRITE,d,2);
-  put_user((short)t.sigl,(short *) d);
+  FPU_put_user((short)t.sigl,(short *) d);
   RE_ENTRANT_CHECK_ON;
 
   return 1;
@@ -1023,24 +1029,33 @@ int reg_store_int16(short *d, FPU_REG *st0_ptr)
 
 
 /* Put a packed bcd array into user memory */
-int reg_store_bcd(char *d, FPU_REG *st0_ptr)
+int FPU_store_bcd(FPU_REG *st0_ptr, u_char st0_tag, u_char *d)
 {
   FPU_REG t;
   unsigned long long ll;
-  unsigned char b;
+  u_char b;
   int i, precision_loss;
-  unsigned char sign = (st0_ptr->sign == SIGN_NEG) ? 0x80 : 0;
-  char st0_tag = st0_ptr->tag;
+  u_char sign = (getsign(st0_ptr) == SIGN_NEG) ? 0x80 : 0;
 
-  if ( st0_tag == TW_Empty )
+  if ( st0_tag == TAG_Empty )
     {
       /* Empty register (stack underflow) */
       EXCEPTION(EX_StackUnder);
       goto invalid_operand;
     }
+  else if ( st0_tag == TAG_Special )
+    {
+      st0_tag = FPU_Special(st0_ptr);
+      if ( (st0_tag == TW_Infinity) ||
+	   (st0_tag == TW_NaN) )
+	{
+	  EXCEPTION(EX_Invalid);
+	  goto invalid_operand;
+	}
+    }
 
-  reg_move(st0_ptr, &t);
-  precision_loss = round_to_int(&t);
+  reg_copy(st0_ptr, &t);
+  precision_loss = FPU_round_to_int(&t, st0_tag);
   ll = significand(&t);
 
   /* Check for overflow, by comparing with 999999999999999999 decimal. */
@@ -1056,10 +1071,10 @@ int reg_store_bcd(char *d, FPU_REG *st0_ptr)
 	  RE_ENTRANT_CHECK_OFF;
 	  FPU_verify_area(VERIFY_WRITE,d,10);
 	  for ( i = 0; i < 7; i++)
-	    put_user(0, (unsigned char *) d+i); /* These bytes "undefined" */
-	  put_user(0xc0, (unsigned char *) d+7); /* This byte "undefined" */
-	  put_user(0xff, (unsigned char *) d+8);
-	  put_user(0xff, (unsigned char *) d+9);
+	    FPU_put_user(0, (u_char *) d+i); /* These bytes "undefined" */
+	  FPU_put_user(0xc0, (u_char *) d+7); /* This byte "undefined" */
+	  FPU_put_user(0xff, (u_char *) d+8);
+	  FPU_put_user(0xff, (u_char *) d+9);
 	  RE_ENTRANT_CHECK_ON;
 	  return 1;
 	}
@@ -1077,14 +1092,14 @@ int reg_store_bcd(char *d, FPU_REG *st0_ptr)
   RE_ENTRANT_CHECK_ON;
   for ( i = 0; i < 9; i++)
     {
-      b = div_small(&ll, 10);
-      b |= (div_small(&ll, 10)) << 4;
+      b = FPU_div_small(&ll, 10);
+      b |= (FPU_div_small(&ll, 10)) << 4;
       RE_ENTRANT_CHECK_OFF;
-      put_user(b,(unsigned char *) d+i);
+      FPU_put_user(b,(u_char *) d+i);
       RE_ENTRANT_CHECK_ON;
     }
   RE_ENTRANT_CHECK_OFF;
-  put_user(sign,(unsigned char *) d+9);
+  FPU_put_user(sign,(u_char *) d+9);
   RE_ENTRANT_CHECK_ON;
 
   return 1;
@@ -1100,25 +1115,25 @@ int reg_store_bcd(char *d, FPU_REG *st0_ptr)
 /* Overflow is signalled by a non-zero return value (in eax).
    In the case of overflow, the returned significand always has the
    largest possible value */
-int round_to_int(FPU_REG *r)
+int FPU_round_to_int(FPU_REG *r, u_char tag)
 {
-  char     very_big;
+  u_char     very_big;
   unsigned eax;
 
-  if (r->tag == TW_Zero)
+  if (tag == TAG_Zero)
     {
       /* Make sure that zero is returned */
       significand(r) = 0;
       return 0;        /* o.k. */
     }
-  
-  if (r->exp > EXP_BIAS + 63)
+
+  if (exponent(r) > 63)
     {
       r->sigl = r->sigh = ~0;      /* The largest representable number */
       return 1;        /* overflow */
     }
 
-  eax = shrxs(&r->sigl, EXP_BIAS + 63 - r->exp);
+  eax = FPU_shrxs(&r->sigl, 63 - exponent(r));
   very_big = !(~(r->sigh) | ~(r->sigl));  /* test for 0xfff...fff */
 #define	half_or_more	(eax & 0x80000000)
 #define	frac_part	(eax)
@@ -1135,7 +1150,7 @@ int round_to_int(FPU_REG *r)
 	}
       break;
     case RC_DOWN:
-      if (frac_part && r->sign)
+      if (frac_part && getsign(r))
 	{
 	  if ( very_big ) return 1;        /* overflow */
 	  significand(r) ++;
@@ -1143,7 +1158,7 @@ int round_to_int(FPU_REG *r)
 	}
       break;
     case RC_UP:
-      if (frac_part && !r->sign)
+      if (frac_part && !getsign(r))
 	{
 	  if ( very_big ) return 1;        /* overflow */
 	  significand(r) ++;
@@ -1160,10 +1175,10 @@ int round_to_int(FPU_REG *r)
 
 /*===========================================================================*/
 
-char *fldenv(fpu_addr_modes addr_modes, char *s)
+u_char *fldenv(fpu_addr_modes addr_modes, u_char *s)
 {
   unsigned short tag_word = 0;
-  unsigned char tag;
+  u_char tag;
   int i;
 
   if ( (addr_modes.default_mode == VM86) ||
@@ -1172,13 +1187,13 @@ char *fldenv(fpu_addr_modes addr_modes, char *s)
     {
       RE_ENTRANT_CHECK_OFF;
       FPU_verify_area(VERIFY_READ, s, 0x0e);
-      get_user(control_word, (unsigned short *) s);
-      get_user(partial_status, (unsigned short *) (s+2));
-      get_user(tag_word, (unsigned short *) (s+4));
-      get_user(instruction_address.offset, (unsigned short *) (s+6));
-      get_user(instruction_address.selector, (unsigned short *) (s+8));
-      get_user(operand_address.offset, (unsigned short *) (s+0x0a));
-      get_user(operand_address.selector, (unsigned short *) (s+0x0c));
+      FPU_get_user(control_word, (unsigned short *) s);
+      FPU_get_user(partial_status, (unsigned short *) (s+2));
+      FPU_get_user(tag_word, (unsigned short *) (s+4));
+      FPU_get_user(instruction_address.offset, (unsigned short *) (s+6));
+      FPU_get_user(instruction_address.selector, (unsigned short *) (s+8));
+      FPU_get_user(operand_address.offset, (unsigned short *) (s+0x0a));
+      FPU_get_user(operand_address.selector, (unsigned short *) (s+0x0c));
       RE_ENTRANT_CHECK_ON;
       s += 0x0e;
       if ( addr_modes.default_mode == VM86 )
@@ -1192,14 +1207,14 @@ char *fldenv(fpu_addr_modes addr_modes, char *s)
     {
       RE_ENTRANT_CHECK_OFF;
       FPU_verify_area(VERIFY_READ, s, 0x1c);
-      get_user(control_word, (unsigned short *) s);
-      get_user(partial_status, (unsigned short *) (s+4));
-      get_user(tag_word, (unsigned short *) (s+8));
-      get_user(instruction_address.offset, (unsigned long *) (s+0x0c));
-      get_user(instruction_address.selector, (unsigned short *) (s+0x10));
-      get_user(instruction_address.opcode, (unsigned short *) (s+0x12));
-      get_user(operand_address.offset, (unsigned long *) (s+0x14));
-      get_user(operand_address.selector, (unsigned long *) (s+0x18));
+      FPU_get_user(control_word, (unsigned short *) s);
+      FPU_get_user(partial_status, (unsigned short *) (s+4));
+      FPU_get_user(tag_word, (unsigned short *) (s+8));
+      FPU_get_user(instruction_address.offset, (unsigned long *) (s+0x0c));
+      FPU_get_user(instruction_address.selector, (unsigned short *) (s+0x10));
+      FPU_get_user(instruction_address.opcode, (unsigned short *) (s+0x12));
+      FPU_get_user(operand_address.offset, (unsigned long *) (s+0x14));
+      FPU_get_user(operand_address.selector, (unsigned long *) (s+0x18));
       RE_ENTRANT_CHECK_ON;
       s += 0x1c;
     }
@@ -1220,29 +1235,28 @@ char *fldenv(fpu_addr_modes addr_modes, char *s)
       tag = tag_word & 3;
       tag_word >>= 2;
 
-      if ( tag == 3 )
+      if ( tag == TAG_Empty )
 	/* New tag is empty.  Accept it */
-	regs[i].tag = TW_Empty;
-      else if ( regs[i].tag == TW_Empty )
+	FPU_settag(i, TAG_Empty);
+      else if ( FPU_gettag(i) == TAG_Empty )
 	{
 	  /* Old tag is empty and new tag is not empty.  New tag is determined
 	     by old reg contents */
-	  if ( regs[i].exp == EXP_BIAS - EXTENDED_Ebias )
+	  if ( exponent(&fpu_register(i)) == - EXTENDED_Ebias )
 	    {
-	      if ( !(regs[i].sigl | regs[i].sigh) )
-		regs[i].tag = TW_Zero;
+	      if ( !(fpu_register(i).sigl | fpu_register(i).sigh) )
+		FPU_settag(i, TAG_Zero);
 	      else
-		regs[i].tag = TW_Valid;
+		FPU_settag(i, TAG_Special);
 	    }
-	  else if ( regs[i].exp == 0x7fff + EXP_BIAS - EXTENDED_Ebias )
+	  else if ( exponent(&fpu_register(i)) == 0x7fff - EXTENDED_Ebias )
 	    {
-	      if ( !((regs[i].sigh & ~0x80000000) | regs[i].sigl) )
-		regs[i].tag = TW_Infinity;
-	      else
-		regs[i].tag = TW_NaN;
+	      FPU_settag(i, TAG_Special);
 	    }
+	  else if ( fpu_register(i).sigh & 0x80000000 )
+	    FPU_settag(i, TAG_Valid);
 	  else
-	    regs[i].tag = TW_Valid;
+	    FPU_settag(i, TAG_Special);   /* An Un-normal */
   	}
       /* Else old tag is not empty and new tag is not empty.  Old tag
 	 remains correct */
@@ -1252,56 +1266,32 @@ char *fldenv(fpu_addr_modes addr_modes, char *s)
 }
 
 
-void frstor(fpu_addr_modes addr_modes, char *data_address)
+void frstor(fpu_addr_modes addr_modes, u_char *data_address)
 {
-  int i, stnr;
-  unsigned char tag;
-  char *s = fldenv(addr_modes, data_address);
+  int i, regnr;
+  u_char *s = fldenv(addr_modes, data_address);
+  int offset = (top & 7) * 10, other = 80 - offset;
+
+  /* Copy all registers in stack order. */
+  RE_ENTRANT_CHECK_OFF;
+  FPU_verify_area(VERIFY_READ,s,80);
+  __copy_from_user(register_base+offset, s, other);
+  if ( offset )
+    __copy_from_user(register_base, s+other, offset);
+  RE_ENTRANT_CHECK_ON;
 
   for ( i = 0; i < 8; i++ )
     {
-      /* Load each register. */
-      stnr = (i+top) & 7;
-      tag = regs[stnr].tag;   /* Derived from the fldenv() loaded tag word. */
-      reg_load_extended((long double *)(s+i*10), &regs[stnr]);
-      if ( tag == TW_Empty )  /* The loaded data over-rides all other cases. */
-	regs[stnr].tag = tag;
+      regnr = (i+top) & 7;
+      if ( FPU_gettag(regnr) != TAG_Empty )
+	/* The loaded data over-rides all other cases. */
+	FPU_settag(regnr, FPU_tagof(&st(i)));
     }
 
 }
 
 
-unsigned short tag_word(void)
-{
-  unsigned short word = 0;
-  unsigned char tag;
-  int i;
-
-  for ( i = 7; i >= 0; i-- )
-    {
-      switch ( tag = regs[i].tag )
-	{
-	case TW_Valid:
-	  if ( regs[i].exp <= (EXP_BIAS - EXTENDED_Ebias) )
-	    tag = 2;
-	  break;
-	case TW_Infinity:
-	case TW_NaN:
-	  tag = 2;
-	  break;
-	case TW_Empty:
-	  tag = 3;
-	  break;
-	  /* TW_Zero already has the correct value */
-	}
-      word <<= 2;
-      word |= tag;
-    }
-  return word;
-}
-
-
-char *fstenv(fpu_addr_modes addr_modes, char *d)
+u_char *fstenv(fpu_addr_modes addr_modes, u_char *d)
 {
   if ( (addr_modes.default_mode == VM86) ||
       ((addr_modes.default_mode == PM16)
@@ -1310,25 +1300,25 @@ char *fstenv(fpu_addr_modes addr_modes, char *d)
       RE_ENTRANT_CHECK_OFF;
       FPU_verify_area(VERIFY_WRITE,d,14);
 #ifdef PECULIAR_486
-      put_user(control_word & ~0xe080, (unsigned long *) d);
+      FPU_put_user(control_word & ~0xe080, (unsigned long *) d);
 #else
-      put_user(control_word, (unsigned short *) d);
+      FPU_put_user(control_word, (unsigned short *) d);
 #endif PECULIAR_486
-      put_user(status_word(), (unsigned short *) (d+2));
-      put_user(tag_word(), (unsigned short *) (d+4));
-      put_user(instruction_address.offset, (unsigned short *) (d+6));
-      put_user(operand_address.offset, (unsigned short *) (d+0x0a));
+      FPU_put_user(status_word(), (unsigned short *) (d+2));
+      FPU_put_user(fpu_tag_word, (unsigned short *) (d+4));
+      FPU_put_user(instruction_address.offset, (unsigned short *) (d+6));
+      FPU_put_user(operand_address.offset, (unsigned short *) (d+0x0a));
       if ( addr_modes.default_mode == VM86 )
 	{
-	  put_user((instruction_address.offset & 0xf0000) >> 4,
+	  FPU_put_user((instruction_address.offset & 0xf0000) >> 4,
 		      (unsigned short *) (d+8));
-	  put_user((operand_address.offset & 0xf0000) >> 4,
+	  FPU_put_user((operand_address.offset & 0xf0000) >> 4,
 		      (unsigned short *) (d+0x0c));
 	}
       else
 	{
-	  put_user(instruction_address.selector, (unsigned short *) (d+8));
-	  put_user(operand_address.selector, (unsigned short *) (d+0x0c));
+	  FPU_put_user(instruction_address.selector, (unsigned short *) (d+8));
+	  FPU_put_user(operand_address.selector, (unsigned short *) (d+0x0c));
 	}
       RE_ENTRANT_CHECK_ON;
       d += 0x0e;
@@ -1336,28 +1326,17 @@ char *fstenv(fpu_addr_modes addr_modes, char *d)
   else
     {
       RE_ENTRANT_CHECK_OFF;
-      FPU_verify_area(VERIFY_WRITE,d,28);
+      FPU_verify_area(VERIFY_WRITE, d, 7*4);
 #ifdef PECULIAR_486
+      control_word &= ~0xe080;
       /* An 80486 sets all the reserved bits to 1. */
-      put_user(0xffff0040 | (control_word & ~0xe080), (unsigned long *) d);
-      put_user(0xffff0000 | status_word(), (unsigned long *) (d+4));
-      put_user(0xffff0000 | tag_word(), (unsigned long *) (d+8));
-#else
-      put_user(control_word, (unsigned short *) d);
-      put_user(status_word(), (unsigned short *) (d+4));
-      put_user(tag_word(), (unsigned short *) (d+8));
-#endif PECULIAR_486
-      put_user(instruction_address.offset, (unsigned long *) (d+0x0c));
-      put_user(instruction_address.selector, (unsigned short *) (d+0x10));
-      put_user(instruction_address.opcode, (unsigned short *) (d+0x12));
-      put_user(operand_address.offset, (unsigned long *) (d+0x14));
-#ifdef PECULIAR_486
-      /* An 80486 sets all the reserved bits to 1. */
-      put_user(operand_address.selector, (unsigned short *) (d+0x18));
-      put_user(0xffff, (unsigned short *) (d+0x1a));
-#else
-      put_user(operand_address.selector, (unsigned long *) (d+0x18));
+      control_word |= 0xffff0000;
+      partial_status = status_word() | 0xffff0000;
+      fpu_tag_word |= 0xffff0000;
+      I387.soft.fcs |= 0xf8000000;
+      I387.soft.fos |= 0xffff0000;
 #endif PECULIAR_486
+      __copy_to_user(d, &control_word, 7*4);
       RE_ENTRANT_CHECK_ON;
       d += 0x1c;
     }
@@ -1369,84 +1348,23 @@ char *fstenv(fpu_addr_modes addr_modes, char *d)
 }
 
 
-void fsave(fpu_addr_modes addr_modes, char *data_address)
+void fsave(fpu_addr_modes addr_modes, u_char *data_address)
 {
-  char *d;
-  int i;
+  u_char *d;
+  int offset = (top & 7) * 10, other = 80 - offset;
 
   d = fstenv(addr_modes, data_address);
+
   RE_ENTRANT_CHECK_OFF;
   FPU_verify_area(VERIFY_WRITE,d,80);
+
+  /* Copy all registers in stack order. */
+  __copy_to_user(d, register_base+offset, other);
+  if ( offset )
+    __copy_to_user(d+other, register_base, offset);
   RE_ENTRANT_CHECK_ON;
-  for ( i = 0; i < 8; i++ )
-    write_to_extended(&regs[(top + i) & 7], d + 10 * i);
 
   finit();
-
 }
 
 /*===========================================================================*/
-
-/*
-  A call to this function must be preceded by a call to
-  FPU_verify_area() to verify access to the 10 bytes at d
-  */
-static void write_to_extended(FPU_REG *rp, char *d)
-{
-  long e;
-  FPU_REG tmp;
-  
-  e = rp->exp - EXP_BIAS + EXTENDED_Ebias;
-
-#ifdef PARANOID
-  switch ( rp->tag )
-    {
-    case TW_Zero:
-      if ( rp->sigh | rp->sigl | e )
-	EXCEPTION(EX_INTERNAL | 0x160);
-      break;
-    case TW_Infinity:
-    case TW_NaN:
-      if ( (e ^ 0x7fff) | !(rp->sigh & 0x80000000) )
-	EXCEPTION(EX_INTERNAL | 0x161);
-      break;
-    default:
-      if (e > 0x7fff || e < -63)
-	EXCEPTION(EX_INTERNAL | 0x162);
-    }
-#endif PARANOID
-
-  /*
-    All numbers except denormals are stored internally in a
-    format which is compatible with the extended real number
-    format.
-   */
-  if ( e > 0 )
-    {
-      /* just copy the reg */
-      RE_ENTRANT_CHECK_OFF;
-      put_user(rp->sigl, (unsigned long *) d);
-      put_user(rp->sigh, (unsigned long *) (d + 4));
-      RE_ENTRANT_CHECK_ON;
-    }
-  else
-    {
-      /*
-	The number is a de-normal stored as a normal using our
-	extra exponent range, or is Zero.
-	Convert it back to a de-normal, or leave it as Zero.
-       */
-      reg_move(rp, &tmp);
-      tmp.exp += -EXTENDED_Emin + 63;  /* largest exp to be 63 */
-      round_to_int(&tmp);
-      e = 0;
-      RE_ENTRANT_CHECK_OFF;
-      put_user(tmp.sigl, (unsigned long *) d);
-      put_user(tmp.sigh, (unsigned long *) (d + 4));
-      RE_ENTRANT_CHECK_ON;
-    }
-  e |= rp->sign == SIGN_POS ? 0 : 0x8000;
-  RE_ENTRANT_CHECK_OFF;
-  put_user(e, (unsigned short *) (d + 8));
-  RE_ENTRANT_CHECK_ON;
-}
diff --git a/arch/i386/math-emu/reg_mul.c b/arch/i386/math-emu/reg_mul.c
index 75246187b..1ea92d48e 100644
--- a/arch/i386/math-emu/reg_mul.c
+++ b/arch/i386/math-emu/reg_mul.c
@@ -3,10 +3,11 @@
  |                                                                           |
  | Multiply one FPU_REG by another, put the result in a destination FPU_REG. |
  |                                                                           |
- | Copyright (C) 1992,1993                                                   |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail   billm@vaxc.cc.monash.edu.au    |
+ | Copyright (C) 1992,1993,1997                                              |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
+ | Returns the tag of the result if no exceptions or errors occured.         |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
@@ -14,92 +15,117 @@
  | The destination may be any FPU_REG, including one of the source FPU_REGs. |
  +---------------------------------------------------------------------------*/
 
+#include "fpu_emu.h"
 #include "exception.h"
 #include "reg_constant.h"
-#include "fpu_emu.h"
 #include "fpu_system.h"
 
 
+/*
+  Multiply two registers to give a register result.
+  The sources are st(deststnr) and (b,tagb,signb).
+  The destination is st(deststnr).
+  */
 /* This routine must be called with non-empty source registers */
-int reg_mul(FPU_REG const *a, FPU_REG const *b,
-	    FPU_REG *dest, unsigned int control_w)
+int FPU_mul(FPU_REG const *b, u_char tagb, int deststnr, int control_w)
 {
-  char saved_sign = dest->sign;
-  char sign = (a->sign ^ b->sign);
+  FPU_REG *a = &st(deststnr);
+  FPU_REG *dest = a;
+  u_char taga = FPU_gettagi(deststnr);
+  u_char saved_sign = getsign(dest);
+  u_char sign = (getsign(a) ^ getsign(b));
+  int tag;
+
 
-  if (!(a->tag | b->tag))
+  if ( !(taga | tagb) )
     {
       /* Both regs Valid, this should be the most common case. */
-      dest->sign = sign;
-      if ( reg_u_mul(a, b, dest, control_w) )
+
+      tag = FPU_u_mul(a, b, dest, control_w, sign, exponent(a) + exponent(b));
+      if ( tag < 0 )
 	{
-	  dest->sign = saved_sign;
-	  return 1;
+	  setsign(dest, saved_sign);
+	  return tag;
 	}
-      return 0;
+      FPU_settagi(deststnr, tag);
+      return tag;
     }
-  else if ((a->tag <= TW_Zero) && (b->tag <= TW_Zero))
+
+  if ( taga == TAG_Special )
+    taga = FPU_Special(a);
+  if ( tagb == TAG_Special )
+    tagb = FPU_Special(b);
+
+  if ( ((taga == TAG_Valid) && (tagb == TW_Denormal))
+	    || ((taga == TW_Denormal) && (tagb == TAG_Valid))
+	    || ((taga == TW_Denormal) && (tagb == TW_Denormal)) )
     {
-#ifdef DENORM_OPERAND
-      if ( ((b->tag == TW_Valid) && (b->exp <= EXP_UNDER)) ||
-	  ((a->tag == TW_Valid) && (a->exp <= EXP_UNDER)) )
+      FPU_REG x, y;
+      if ( denormal_operand() < 0 )
+	return FPU_Exception;
+
+      FPU_to_exp16(a, &x);
+      FPU_to_exp16(b, &y);
+      tag = FPU_u_mul(&x, &y, dest, control_w, sign,
+		      exponent16(&x) + exponent16(&y));
+      if ( tag < 0 )
 	{
-	  if ( denormal_operand() ) return 1;
+	  setsign(dest, saved_sign);
+	  return tag;
 	}
-#endif DENORM_OPERAND
+      FPU_settagi(deststnr, tag);
+      return tag;
+    }
+  else if ( (taga <= TW_Denormal) && (tagb <= TW_Denormal) )
+    {
+      if ( ((tagb == TW_Denormal) || (taga == TW_Denormal))
+	   && (denormal_operand() < 0) )
+	return FPU_Exception;
+
       /* Must have either both arguments == zero, or
 	 one valid and the other zero.
 	 The result is therefore zero. */
-      reg_move(&CONST_Z, dest);
+      FPU_copy_to_regi(&CONST_Z, TAG_Zero, deststnr);
       /* The 80486 book says that the answer is +0, but a real
 	 80486 behaves this way.
 	 IEEE-754 apparently says it should be this way. */
-      dest->sign = sign;
-      return 0;
+      setsign(dest, sign);
+      return TAG_Zero;
     }
-  else
-    {
       /* Must have infinities, NaNs, etc */
-      if ( (a->tag == TW_NaN) || (b->tag == TW_NaN) )
-	{ return real_2op_NaN(a, b, dest); }
-      else if (a->tag == TW_Infinity)
-	{
-	  if (b->tag == TW_Zero)
-	    { return arith_invalid(dest); }  /* Zero*Infinity is invalid */
-	  else
-	    {
-#ifdef DENORM_OPERAND
-	      if ( (b->tag == TW_Valid) && (b->exp <= EXP_UNDER) &&
-		  denormal_operand() )
-		return 1;
-#endif DENORM_OPERAND
-	      reg_move(a, dest);
-	      dest->sign = sign;
-	    }
-	  return 0;
-	}
-      else if (b->tag == TW_Infinity)
-	{
-	  if (a->tag == TW_Zero)
-	    { return arith_invalid(dest); }  /* Zero*Infinity is invalid */
-	  else
-	    {
-#ifdef DENORM_OPERAND
-	      if ( (a->tag == TW_Valid) && (a->exp <= EXP_UNDER) &&
-		  denormal_operand() )
-		return 1;
-#endif DENORM_OPERAND
-	      reg_move(b, dest);
-	      dest->sign = sign;
-	    }
-	  return 0;
-	}
+  else if ( (taga == TW_NaN) || (tagb == TW_NaN) )
+    {
+      return real_2op_NaN(b, tagb, deststnr, &st(0));
+    }
+  else if ( ((taga == TW_Infinity) && (tagb == TAG_Zero))
+	    || ((tagb == TW_Infinity) && (taga == TAG_Zero)) )
+    {
+      return arith_invalid(deststnr);  /* Zero*Infinity is invalid */
+    }
+  else if ( ((taga == TW_Denormal) || (tagb == TW_Denormal))
+	    && (denormal_operand() < 0) )
+    {
+      return FPU_Exception;
+    }
+  else if (taga == TW_Infinity)
+    {
+      FPU_copy_to_regi(a, TAG_Special, deststnr);
+      setsign(dest, sign);
+      return TAG_Special;
+    }
+  else if (tagb == TW_Infinity)
+    {
+      FPU_copy_to_regi(b, TAG_Special, deststnr);
+      setsign(dest, sign);
+      return TAG_Special;
+    }
+
 #ifdef PARANOID
-      else
-	{
-	  EXCEPTION(EX_INTERNAL|0x102);
-	  return 1;
-	}
-#endif PARANOID
+  else
+    {
+      EXCEPTION(EX_INTERNAL|0x102);
+      return FPU_Exception;
     }
+#endif PARANOID
+
 }
diff --git a/arch/i386/math-emu/reg_norm.S b/arch/i386/math-emu/reg_norm.S
index 781a2d466..f63a87755 100644
--- a/arch/i386/math-emu/reg_norm.S
+++ b/arch/i386/math-emu/reg_norm.S
@@ -1,16 +1,19 @@
 /*---------------------------------------------------------------------------+
  |  reg_norm.S                                                               |
  |                                                                           |
- | Copyright (C) 1992,1993,1994,1995                                         |
+ | Copyright (C) 1992,1993,1994,1995,1997                                    |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
+ |                       Australia.  E-mail billm@suburbia.net               |
  |                                                                           |
  | Normalize the value in a FPU_REG.                                         |
  |                                                                           |
  | Call from C as:                                                           |
- |   void normalize(FPU_REG *n)                                              |
+ |    int FPU_normalize(FPU_REG *n)                                          |
  |                                                                           |
- |   void normalize_nuo(FPU_REG *n)                                          |
+ |    int FPU_normalize_nuo(FPU_REG *n)                                      |
+ |                                                                           |
+ |    Return value is the tag of the answer, or-ed with FPU_Exception if     |
+ |    one was raised, or -1 on internal error.                               |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
@@ -18,24 +21,13 @@
 
 
 .text
-ENTRY(normalize)
+ENTRY(FPU_normalize)
 	pushl	%ebp
 	movl	%esp,%ebp
 	pushl	%ebx
 
 	movl	PARAM1,%ebx
 
-#ifdef PARANOID
-	cmpb	TW_Valid,TAG(%ebx)
-	je	L_ok
-
-	pushl	$0x220
-	call	SYMBOL_NAME(FPU_exception)
-	addl	$4,%esp
-
-L_ok:
-#endif PARANOID
-
 	movl	SIGH(%ebx),%edx
 	movl	SIGL(%ebx),%eax
 
@@ -48,7 +40,7 @@ L_ok:
 
 	movl	%eax,%edx
 	xorl	%eax,%eax
-	subl	$32,EXP(%ebx)	/* This can cause an underflow */
+	subw	$32,EXP(%ebx)	/* This can cause an underflow */
 
 /* We need to shift left by 1 - 31 bits */
 L_shift_1:
@@ -57,18 +49,25 @@ L_shift_1:
 	negl	%ecx
 	shld	%cl,%eax,%edx
 	shl	%cl,%eax
-	subl	%ecx,EXP(%ebx)	/* This can cause an underflow */
+	subw	%cx,EXP(%ebx)	/* This can cause an underflow */
 
 	movl	%edx,SIGH(%ebx)
 	movl	%eax,SIGL(%ebx)
 
 L_done:
-	cmpl	EXP_OVER,EXP(%ebx)
+	cmpw	EXP_OVER,EXP(%ebx)
 	jge	L_overflow
 
-	cmpl	EXP_UNDER,EXP(%ebx)
+	cmpw	EXP_UNDER,EXP(%ebx)
 	jle	L_underflow
 
+L_exit_valid:
+	movl	TAG_Valid,%eax
+
+	/* Convert the exponent to 80x87 form. */
+	addw	EXTENDED_Ebias,EXP(%ebx)
+	andw	$0x7fff,EXP(%ebx)
+
 L_exit:
 	popl	%ebx
 	leave
@@ -76,17 +75,21 @@ L_exit:
 
 
 L_zero:
-	movl	EXP_UNDER,EXP(%ebx)
-	movb	TW_Zero,TAG(%ebx)
+	movw	$0,EXP(%ebx)
+	movl	TAG_Zero,%eax
 	jmp	L_exit
 
 L_underflow:
+	/* Convert the exponent to 80x87 form. */
+	addw	EXTENDED_Ebias,EXP(%ebx)
 	push	%ebx
 	call	SYMBOL_NAME(arith_underflow)
 	pop	%ebx
 	jmp	L_exit
 
 L_overflow:
+	/* Convert the exponent to 80x87 form. */
+	addw	EXTENDED_Ebias,EXP(%ebx)
 	push	%ebx
 	call	SYMBOL_NAME(arith_overflow)
 	pop	%ebx
@@ -95,37 +98,26 @@ L_overflow:
 
 
 /* Normalise without reporting underflow or overflow */
-ENTRY(normalize_nuo)
+ENTRY(FPU_normalize_nuo)
 	pushl	%ebp
 	movl	%esp,%ebp
 	pushl	%ebx
 
 	movl	PARAM1,%ebx
 
-#ifdef PARANOID
-	cmpb	TW_Valid,TAG(%ebx)
-	je	L_ok_nuo
-
-	pushl	$0x221
-	call	SYMBOL_NAME(FPU_exception)
-	addl	$4,%esp
-
-L_ok_nuo:
-#endif PARANOID
-
 	movl	SIGH(%ebx),%edx
 	movl	SIGL(%ebx),%eax
 
 	orl	%edx,%edx	/* ms bits */
-	js	L_exit		/* Already normalized */
+	js	L_exit_nuo_valid	/* Already normalized */
 	jnz	L_nuo_shift_1	/* Shift left 1 - 31 bits */
 
 	orl	%eax,%eax
-	jz	L_zero		/* The contents are zero */
+	jz	L_exit_nuo_zero		/* The contents are zero */
 
 	movl	%eax,%edx
 	xorl	%eax,%eax
-	subl	$32,EXP(%ebx)	/* This can cause an underflow */
+	subw	$32,EXP(%ebx)	/* This can cause an underflow */
 
 /* We need to shift left by 1 - 31 bits */
 L_nuo_shift_1:
@@ -134,10 +126,22 @@ L_nuo_shift_1:
 	negl	%ecx
 	shld	%cl,%eax,%edx
 	shl	%cl,%eax
-	subl	%ecx,EXP(%ebx)	/* This can cause an underflow */
+	subw	%cx,EXP(%ebx)	/* This can cause an underflow */
 
 	movl	%edx,SIGH(%ebx)
 	movl	%eax,SIGL(%ebx)
-	jmp	L_exit
 
+L_exit_nuo_valid:
+	movl	TAG_Valid,%eax
+
+	popl	%ebx
+	leave
+	ret
 
+L_exit_nuo_zero:
+	movl	TAG_Zero,%eax
+	movw	EXP_UNDER,EXP(%ebx)
+
+	popl	%ebx
+	leave
+	ret
diff --git a/arch/i386/math-emu/reg_round.S b/arch/i386/math-emu/reg_round.S
index 4aac507a1..9ab32e999 100644
--- a/arch/i386/math-emu/reg_round.S
+++ b/arch/i386/math-emu/reg_round.S
@@ -4,17 +4,20 @@
  |                                                                           |
  | Rounding/truncation/etc for FPU basic arithmetic functions.               |
  |                                                                           |
- | Copyright (C) 1993,1995                                                   |
+ | Copyright (C) 1993,1995,1997                                              |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
+ |                       Australia.  E-mail billm@suburbia.net               |
  |                                                                           |
  | This code has four possible entry points.                                 |
  | The following must be entered by a jmp instruction:                       |
  |   fpu_reg_round, fpu_reg_round_sqrt, and fpu_Arith_exit.                  |
  |                                                                           |
- | The _round_reg entry point is intended to be used by C code.              |
+ | The FPU_round entry point is intended to be used by C code.               |
  | From C, call as:                                                          |
- | void round_reg(FPU_REG *arg, unsigned int extent, unsigned int control_w) |
+ |  int FPU_round(FPU_REG *arg, unsigned int extent, unsigned int control_w) |
+ |                                                                           |
+ |    Return value is the tag of the answer, or-ed with FPU_Exception if     |
+ |    one was raised, or -1 on internal error.                               |
  |                                                                           |
  | For correct "up" and "down" rounding, the argument must have the correct  |
  | sign.                                                                     |
@@ -106,7 +109,7 @@ FPU_denormal:
 .globl fpu_Arith_exit
 
 /* Entry point when called from C */
-ENTRY(round_reg)
+ENTRY(FPU_round)
 	pushl	%ebp
 	movl	%esp,%ebp
 	pushl	%esi
@@ -117,14 +120,10 @@ ENTRY(round_reg)
 	movl	SIGH(%edi),%eax
 	movl	SIGL(%edi),%ebx
 	movl	PARAM2,%edx
-	movl	PARAM3,%ecx
-	jmp	fpu_reg_round_sqrt
 
 fpu_reg_round:			/* Normal entry point */
 	movl	PARAM4,%ecx
 
-fpu_reg_round_sqrt:		/* Entry point from wm_sqrt.S */
-
 #ifndef NON_REENTRANT_FPU
 	pushl	%ebx		/* adjust the stack pointer */
 #endif NON_REENTRANT_FPU
@@ -135,12 +134,12 @@ fpu_reg_round_sqrt:		/* Entry point from wm_sqrt.S */
 /*	jns	L_entry_bugged */
 #endif PARANOID
 
-	cmpl	EXP_UNDER,EXP(%edi)
-	jle	xMake_denorm			/* The number is a de-normal */
+	cmpw	EXP_UNDER,EXP(%edi)
+	jle	L_Make_denorm			/* The number is a de-normal */
 
 	movb	$0,FPU_denormal			/* 0 -> not a de-normal */
 
-xDenorm_done:
+Denorm_done:
 	movb	$0,FPU_bits_lost		/* No bits yet lost in rounding */
 
 	movl	%ecx,%esi
@@ -190,13 +189,13 @@ LRound_To_24:
 #endif PARANOID
 
 LUp_24:
-	cmpb	SIGN_POS,SIGN(%edi)
+	cmpb	SIGN_POS,PARAM5
 	jne	LCheck_truncate_24	/* If negative then  up==truncate */
 
 	jmp	LCheck_24_round_up
 
 LDown_24:
-	cmpb	SIGN_POS,SIGN(%edi)
+	cmpb	SIGN_POS,PARAM5
 	je	LCheck_truncate_24	/* If positive then  down==truncate */
 
 LCheck_24_round_up:
@@ -205,7 +204,7 @@ LCheck_24_round_up:
 	orl	%ebx,%ecx
 	orl	%edx,%ecx
 	jnz	LDo_24_round_up
-	jmp	LRe_normalise
+	jmp	L_Re_normalise
 
 LRound_nearest_24:
 	/* Do rounding of the 24th bit if needed (nearest or even) */
@@ -240,13 +239,13 @@ LCheck_truncate_24:
 	andl	$0x000000ff,%ecx
 	orl	%ebx,%ecx
 	orl	%edx,%ecx
-	jz	LRe_normalise		/* No truncation needed */
+	jz	L_Re_normalise		/* No truncation needed */
 
 LDo_truncate_24:
 	andl	$0xffffff00,%eax	/* Truncate to 24 bits */
 	xorl	%ebx,%ebx
 	movb	LOST_DOWN,FPU_bits_lost
-	jmp	LRe_normalise
+	jmp	L_Re_normalise
 
 
 /* Round etc to 53 bit precision */
@@ -270,13 +269,13 @@ LRound_To_53:
 #endif PARANOID
 
 LUp_53:
-	cmpb	SIGN_POS,SIGN(%edi)
+	cmpb	SIGN_POS,PARAM5
 	jne	LCheck_truncate_53	/* If negative then  up==truncate */
 
 	jmp	LCheck_53_round_up
 
 LDown_53:
-	cmpb	SIGN_POS,SIGN(%edi)
+	cmpb	SIGN_POS,PARAM5
 	je	LCheck_truncate_53	/* If positive then  down==truncate */
 
 LCheck_53_round_up:
@@ -284,7 +283,7 @@ LCheck_53_round_up:
 	andl	$0x000007ff,%ecx
 	orl	%edx,%ecx
 	jnz	LDo_53_round_up
-	jmp	LRe_normalise
+	jmp	L_Re_normalise
 
 LRound_nearest_53:
 	/* Do rounding of the 53rd bit if needed (nearest or even) */
@@ -315,12 +314,12 @@ LCheck_truncate_53:
 	movl	%ebx,%ecx
 	andl	$0x000007ff,%ecx
 	orl	%edx,%ecx
-	jz	LRe_normalise
+	jz	L_Re_normalise
 
 LTruncate_53:
 	movb	LOST_DOWN,FPU_bits_lost
 	andl	$0xfffff800,%ebx	/* Truncate to 53 bits */
-	jmp	LRe_normalise
+	jmp	L_Re_normalise
 
 
 /* Round etc to 64 bit precision */
@@ -344,20 +343,20 @@ LRound_To_64:
 #endif PARANOID
 
 LUp_64:
-	cmpb	SIGN_POS,SIGN(%edi)
+	cmpb	SIGN_POS,PARAM5
 	jne	LCheck_truncate_64	/* If negative then  up==truncate */
 
 	orl	%edx,%edx
 	jnz	LDo_64_round_up
-	jmp	LRe_normalise
+	jmp	L_Re_normalise
 
 LDown_64:
-	cmpb	SIGN_POS,SIGN(%edi)
+	cmpb	SIGN_POS,PARAM5
 	je	LCheck_truncate_64	/* If positive then  down==truncate */
 
 	orl	%edx,%edx
 	jnz	LDo_64_round_up
-	jmp	LRe_normalise
+	jmp	L_Re_normalise
 
 LRound_nearest_64:
 	cmpl	$0x80000000,%edx
@@ -375,46 +374,60 @@ LDo_64_round_up:
 	adcl	$0,%eax
 
 LCheck_Round_Overflow:
-	jnc	LRe_normalise
+	jnc	L_Re_normalise
 
 	/* Overflow, adjust the result (significand to 1.0) */
 	rcrl	$1,%eax
 	rcrl	$1,%ebx
-	incl	EXP(%edi)
-	jmp	LRe_normalise
+	incw	EXP(%edi)
+	jmp	L_Re_normalise
 
 LCheck_truncate_64:
 	orl	%edx,%edx
-	jz	LRe_normalise
+	jz	L_Re_normalise
 
 LTruncate_64:
 	movb	LOST_DOWN,FPU_bits_lost
 
-LRe_normalise:
+L_Re_normalise:
 	testb	$0xff,FPU_denormal
-	jnz	xNormalise_result
+	jnz	Normalise_result
+
+L_Normalised:
+	movl	TAG_Valid,%edx
 
-xL_Normalised:
+L_deNormalised:
 	cmpb	LOST_UP,FPU_bits_lost
-	je	xL_precision_lost_up
+	je	L_precision_lost_up
 
 	cmpb	LOST_DOWN,FPU_bits_lost
-	je	xL_precision_lost_down
+	je	L_precision_lost_down
 
-xL_no_precision_loss:
+L_no_precision_loss:
 	/* store the result */
-	movb	TW_Valid,TAG(%edi)
 
-xL_Store_significand:
+L_Store_significand:
 	movl	%eax,SIGH(%edi)
 	movl	%ebx,SIGL(%edi)
 
-	xorl	%eax,%eax	/* No errors detected. */
-
-	cmpl	EXP_OVER,EXP(%edi)
+	cmpw	EXP_OVER,EXP(%edi)
 	jge	L_overflow
 
-fpu_reg_round_exit:
+	movl	%edx,%eax
+
+	/* Convert the exponent to 80x87 form. */
+	addw	EXTENDED_Ebias,EXP(%edi)
+	andw	$0x7fff,EXP(%edi)
+
+fpu_reg_round_signed_special_exit:
+
+	cmpb	SIGN_POS,PARAM5
+	je	fpu_reg_round_special_exit
+
+	orw	$0x8000,EXP(%edi)	/* Negative sign for the result. */
+
+fpu_reg_round_special_exit:
+
 #ifndef NON_REENTRANT_FPU
 	popl	%ebx		/* adjust the stack pointer */
 #endif NON_REENTRANT_FPU
@@ -431,21 +444,25 @@ fpu_Arith_exit:
  * Set the FPU status flags to represent precision loss due to
  * round-up.
  */
-xL_precision_lost_up:
+L_precision_lost_up:
+	push	%edx
 	push	%eax
 	call	SYMBOL_NAME(set_precision_flag_up)
 	popl	%eax
-	jmp	xL_no_precision_loss
+	popl	%edx
+	jmp	L_no_precision_loss
 
 /*
  * Set the FPU status flags to represent precision loss due to
  * truncation.
  */
-xL_precision_lost_down:
+L_precision_lost_down:
+	push	%edx
 	push	%eax
 	call	SYMBOL_NAME(set_precision_flag_down)
 	popl	%eax
-	jmp	xL_no_precision_loss
+	popl	%edx
+	jmp	L_no_precision_loss
 
 
 /*
@@ -453,30 +470,30 @@ xL_precision_lost_down:
  * Shift the number right the required number of bits, which will
  * have to be undone later...
  */
-xMake_denorm:
+L_Make_denorm:
 	/* The action to be taken depends upon whether the underflow
 	   exception is masked */
 	testb	CW_Underflow,%cl		/* Underflow mask. */
-	jz	xUnmasked_underflow		/* Do not make a denormal. */
+	jz	Unmasked_underflow		/* Do not make a denormal. */
 
 	movb	DENORMAL,FPU_denormal
 
 	pushl	%ecx		/* Save */
-	movl	EXP_UNDER+1,%ecx
-	subl	EXP(%edi),%ecx
+	movw	EXP_UNDER+1,%cx
+	subw	EXP(%edi),%cx
 
-	cmpl	$64,%ecx	/* shrd only works for 0..31 bits */
-	jnc	xDenorm_shift_more_than_63
+	cmpw	$64,%cx	/* shrd only works for 0..31 bits */
+	jnc	Denorm_shift_more_than_63
 
-	cmpl	$32,%ecx	/* shrd only works for 0..31 bits */
-	jnc	xDenorm_shift_more_than_32
+	cmpw	$32,%cx	/* shrd only works for 0..31 bits */
+	jnc	Denorm_shift_more_than_32
 
 /*
  * We got here without jumps by assuming that the most common requirement
  *   is for a small de-normalising shift.
  * Shift by [1..31] bits
  */
-	addl	%ecx,EXP(%edi)
+	addw	%cx,EXP(%edi)
 	orl	%edx,%edx	/* extension */
 	setne	%ch		/* Save whether %edx is non-zero */
 	xorl	%edx,%edx
@@ -485,11 +502,11 @@ xMake_denorm:
 	shr	%cl,%eax
 	orb	%ch,%dl
 	popl	%ecx
-	jmp	xDenorm_done
+	jmp	Denorm_done
 
 /* Shift by [32..63] bits */
-xDenorm_shift_more_than_32:
-	addl	%ecx,EXP(%edi)
+Denorm_shift_more_than_32:
+	addw	%cx,EXP(%edi)
 	subb	$32,%cl
 	orl	%edx,%edx
 	setne	%ch
@@ -506,15 +523,15 @@ xDenorm_shift_more_than_32:
 	movl	%eax,%ebx
 	xorl	%eax,%eax
 	popl	%ecx
-	jmp	xDenorm_done
+	jmp	Denorm_done
 
 /* Shift by [64..) bits */
-xDenorm_shift_more_than_63:
-	cmpl	$64,%ecx
-	jne	xDenorm_shift_more_than_64
+Denorm_shift_more_than_63:
+	cmpw	$64,%cx
+	jne	Denorm_shift_more_than_64
 
 /* Exactly 64 bit shift */
-	addl	%ecx,EXP(%edi)
+	addw	%cx,EXP(%edi)
 	xorl	%ecx,%ecx
 	orl	%edx,%edx
 	setne	%cl
@@ -526,32 +543,32 @@ xDenorm_shift_more_than_63:
 	xorl	%eax,%eax
 	xorl	%ebx,%ebx
 	popl	%ecx
-	jmp	xDenorm_done
+	jmp	Denorm_done
 
-xDenorm_shift_more_than_64:
-	movl	EXP_UNDER+1,EXP(%edi)
+Denorm_shift_more_than_64:
+	movw	EXP_UNDER+1,EXP(%edi)
 /* This is easy, %eax must be non-zero, so.. */
 	movl	$1,%edx
 	xorl	%eax,%eax
 	xorl	%ebx,%ebx
 	popl	%ecx
-	jmp	xDenorm_done
+	jmp	Denorm_done
 
 
-xUnmasked_underflow:
+Unmasked_underflow:
 	movb	UNMASKED_UNDERFLOW,FPU_denormal
-	jmp	xDenorm_done
+	jmp	Denorm_done
 
 
 /* Undo the de-normalisation. */
-xNormalise_result:
+Normalise_result:
 	cmpb	UNMASKED_UNDERFLOW,FPU_denormal
-	je	xSignal_underflow
+	je	Signal_underflow
 
 /* The number must be a denormal if we got here. */
 #ifdef PARANOID
 	/* But check it... just in case. */
-	cmpl	EXP_UNDER+1,EXP(%edi)
+	cmpw	EXP_UNDER+1,EXP(%edi)
 	jne	L_norm_bugged
 #endif PARANOID
 
@@ -565,41 +582,33 @@ xNormalise_result:
 	 * Actual 80486 behaviour differs from this in some circumstances.
 	 */
 	orl	%eax,%eax		/* ms bits */
-	js	LNormalise_shift_done	/* Will be masked underflow */
-#endif PECULIAR_486
-
+	js	LPseudoDenormal		/* Will be masked underflow */
+#else
 	orl	%eax,%eax		/* ms bits */
-	js	xL_Normalised		/* No longer a denormal */
+	js	L_Normalised		/* No longer a denormal */
+#endif PECULIAR_486
 
-	jnz	LNormalise_shift_up_to_31	/* Shift left 0 - 31 bits */
+	jnz	LDenormal_adj_exponent
 
 	orl	%ebx,%ebx
 	jz	L_underflow_to_zero	/* The contents are zero */
 
-/* Shift left 32 - 63 bits */
-	movl	%ebx,%eax
-	xorl	%ebx,%ebx
-	subl	$32,EXP(%edi)
-
-LNormalise_shift_up_to_31:
-	bsrl	%eax,%ecx	/* get the required shift in %ecx */
-	subl	$31,%ecx
-	negl	%ecx
-	shld	%cl,%ebx,%eax
-	shl	%cl,%ebx
-	subl	%ecx,EXP(%edi)
+LDenormal_adj_exponent:
+	decw	EXP(%edi)
 
-LNormalise_shift_done:
+LPseudoDenormal:
 	testb	$0xff,FPU_bits_lost	/* bits lost == underflow */
-	jz	xL_Normalised
+	movl	TAG_Special,%edx
+	jz	L_deNormalised
 
 	/* There must be a masked underflow */
 	push	%eax
 	pushl	EX_Underflow
-	call	SYMBOL_NAME(FPU_exception)
+	call	EXCEPTION
 	popl	%eax
 	popl	%eax
-	jmp	xL_Normalised
+	movl	TAG_Special,%edx
+	jmp	L_deNormalised
 
 
 /*
@@ -613,41 +622,42 @@ L_underflow_to_zero:
 
 	push	%eax
 	pushl	EX_Underflow
-	call	SYMBOL_NAME(FPU_exception)
+	call	EXCEPTION
 	popl	%eax
 	popl	%eax
 
 /* Reduce the exponent to EXP_UNDER */
-	movl	EXP_UNDER,EXP(%edi)
-	movb	TW_Zero,TAG(%edi)
-	jmp	xL_Store_significand
+	movw	EXP_UNDER,EXP(%edi)
+	movl	TAG_Zero,%edx
+	jmp	L_Store_significand
 
 
 /* The operations resulted in a number too large to represent. */
 L_overflow:
+	addw	EXTENDED_Ebias,EXP(%edi)	/* Set for unmasked response. */
 	push	%edi
 	call	SYMBOL_NAME(arith_overflow)
 	pop	%edi
-	jmp	fpu_reg_round_exit
+	jmp	fpu_reg_round_signed_special_exit
 
 
-xSignal_underflow:
+Signal_underflow:
 	/* The number may have been changed to a non-denormal */
 	/* by the rounding operations. */
-	cmpl	EXP_UNDER,EXP(%edi)
-	jle	xDo_unmasked_underflow
+	cmpw	EXP_UNDER,EXP(%edi)
+	jle	Do_unmasked_underflow
 
-	jmp	xL_Normalised
+	jmp	L_Normalised
 
-xDo_unmasked_underflow:
+Do_unmasked_underflow:
 	/* Increase the exponent by the magic number */
-	addl	$(3*(1<<13)),EXP(%edi)
+	addw	$(3*(1<<13)),EXP(%edi)
 	push	%eax
 	pushl	EX_Underflow
 	call	EXCEPTION
 	popl	%eax
 	popl	%eax
-	jmp	xL_Normalised
+	jmp	L_Normalised
 
 
 #ifdef PARANOID
@@ -694,6 +704,6 @@ L_entry_bugged:
 	call	EXCEPTION
 	popl	%ebx
 L_exception_exit:
-	mov	$1,%eax
-	jmp	fpu_reg_round_exit
+	mov	$-1,%eax
+	jmp	fpu_reg_round_special_exit
 #endif PARANOID
diff --git a/arch/i386/math-emu/reg_u_add.S b/arch/i386/math-emu/reg_u_add.S
index 1dc0d41df..c6664a49e 100644
--- a/arch/i386/math-emu/reg_u_add.S
+++ b/arch/i386/math-emu/reg_u_add.S
@@ -2,24 +2,26 @@
 /*---------------------------------------------------------------------------+
  |  reg_u_add.S                                                              |
  |                                                                           |
- | Add two valid (TW_Valid) FPU_REG numbers, of the same sign, and put the   |
+ | Add two valid (TAG_Valid) FPU_REG numbers, of the same sign, and put the  |
  |   result in a destination FPU_REG.                                        |
  |                                                                           |
- | Copyright (C) 1992,1993,1995                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
+ | Copyright (C) 1992,1993,1995,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  | Call from C as:                                                           |
- |   void reg_u_add(FPU_REG *arg1, FPU_REG *arg2, FPU_REG *answ,             |
+ |   int  FPU_u_add(FPU_REG *arg1, FPU_REG *arg2, FPU_REG *answ,             |
  |                                                int control_w)             |
+ |    Return value is the tag of the answer, or-ed with FPU_Exception if     |
+ |    one was raised, or -1 on internal error.                               |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
 /*
- |    Kernel addition routine reg_u_add(reg *arg1, reg *arg2, reg *answ).
- |    Takes two valid reg f.p. numbers (TW_Valid), which are
+ |    Kernel addition routine FPU_u_add(reg *arg1, reg *arg2, reg *answ).
+ |    Takes two valid reg f.p. numbers (TAG_Valid), which are
  |    treated as unsigned numbers,
- |    and returns their sum as a TW_Valid or TW_S f.p. number.
+ |    and returns their sum as a TAG_Valid or TAG_Special f.p. number.
  |    The returned number is normalized.
  |    Basic checks are performed if PARANOID is defined.
  */
@@ -29,7 +31,7 @@
 #include "control_w.h"
 
 .text
-ENTRY(reg_u_add)
+ENTRY(FPU_u_add)
 	pushl	%ebp
 	movl	%esp,%ebp
 	pushl	%esi
@@ -39,27 +41,9 @@ ENTRY(reg_u_add)
 	movl	PARAM1,%esi		/* source 1 */
 	movl	PARAM2,%edi		/* source 2 */
 
-#ifdef DENORM_OPERAND
-	cmpl	EXP_UNDER,EXP(%esi)
-	jg	xOp1_not_denorm
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-
-xOp1_not_denorm:
-	cmpl	EXP_UNDER,EXP(%edi)
-	jg	xOp2_not_denorm
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-
-xOp2_not_denorm:
-#endif DENORM_OPERAND
-
-	movl	EXP(%esi),%ecx
-	subl	EXP(%edi),%ecx		/* exp1 - exp2 */
+	movl	PARAM6,%ecx
+	movl	%ecx,%edx
+	subl	PARAM7,%ecx			/* exp1 - exp2 */
 	jge	L_arg1_larger
 
 	/* num1 is smaller */
@@ -67,6 +51,7 @@ xOp2_not_denorm:
 	movl	SIGH(%esi),%eax
 
 	movl	%edi,%esi
+	movl	PARAM7,%edx
 	negw	%cx
 	jmp	L_accum_loaded
 
@@ -77,12 +62,7 @@ L_arg1_larger:
 
 L_accum_loaded:
 	movl	PARAM3,%edi		/* destination */
-/*	movb	SIGN(%esi),%dl
-	movb	%dl,SIGN(%edi) */	/* Copy the sign from the first arg */
-
-
-	movl	EXP(%esi),%edx
-	movl	%edx,EXP(%edi)		/* Copy exponent to destination */
+	movw	%dx,EXP(%edi)		/* Copy exponent to destination */
 
 	xorl	%edx,%edx		/* clear the extension */
 
@@ -162,7 +142,7 @@ L_shift_done:
 	orl	$1,%edx
 
 L_no_bit_lost:
-	incl	EXP(%edi)
+	incw	EXP(%edi)
 
 L_round_the_result:
 	jmp	fpu_reg_round	/* Round the result */
@@ -175,9 +155,8 @@ L_bugged:
 	pushl	EX_INTERNAL|0x201
 	call	EXCEPTION
 	pop	%ebx
+	movl	$-1,%eax
 	jmp	L_exit
-#endif PARANOID
-
 
 L_exit:
 	popl	%ebx
@@ -185,3 +164,4 @@ L_exit:
 	popl	%esi
 	leave
 	ret
+#endif PARANOID
diff --git a/arch/i386/math-emu/reg_u_div.S b/arch/i386/math-emu/reg_u_div.S
index 5bba98dd8..36630de7d 100644
--- a/arch/i386/math-emu/reg_u_div.S
+++ b/arch/i386/math-emu/reg_u_div.S
@@ -2,22 +2,24 @@
 /*---------------------------------------------------------------------------+
  |  reg_u_div.S                                                              |
  |                                                                           |
- | Core division routines                                                    |
+ | Divide one FPU_REG by another and put the result in a destination FPU_REG.|
  |                                                                           |
- | Copyright (C) 1992,1993,1995                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
+ | Copyright (C) 1992,1993,1995,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
 /*---------------------------------------------------------------------------+
- |  Kernel for the division routines.                                        |
- |                                                                           |
- |  void reg_u_div(FPU_REG *a, FPU_REG *a,                                   |
- |                 FPU_REG *dest, unsigned int control_word)                 |
+ | Call from C as:                                                           |
+ |    int FPU_u_div(FPU_REG *a, FPU_REG *b, FPU_REG *dest,                   |
+ |                unsigned int control_word, char *sign)                     |
  |                                                                           |
  |  Does not compute the destination exponent, but does adjust it.           |
+ |                                                                           |
+ |    Return value is the tag of the answer, or-ed with FPU_Exception if     |
+ |    one was raised, or -1 on internal error.                               |
  +---------------------------------------------------------------------------*/
 
 #include "exception.h"
@@ -67,9 +69,12 @@ FPU_ovfl_flag:
 	.byte	0
 #endif NON_REENTRANT_FPU
 
+#define REGA	PARAM1
+#define REGB	PARAM2
+#define DEST	PARAM3
 
 .text
-ENTRY(reg_u_div)
+ENTRY(FPU_u_div)
 	pushl	%ebp
 	movl	%esp,%ebp
 #ifndef NON_REENTRANT_FPU
@@ -80,32 +85,28 @@ ENTRY(reg_u_div)
 	pushl	%edi
 	pushl	%ebx
 
-	movl	PARAM1,%esi	/* pointer to num */
-	movl	PARAM2,%ebx	/* pointer to denom */
-	movl	PARAM3,%edi	/* pointer to answer */
+	movl	REGA,%esi
+	movl	REGB,%ebx
+	movl	DEST,%edi
 
-#ifdef DENORM_OPERAND
-	movl	EXP(%esi),%eax
-	cmpl	EXP_UNDER,%eax
-	jg	xOp1_not_denorm
+	movw	EXP(%esi),%dx
+	movw	EXP(%ebx),%ax
+	.byte	0x0f,0xbf,0xc0	/* movsx	%ax,%eax */
+	.byte	0x0f,0xbf,0xd2	/* movsx	%dx,%edx */
+	subl	%eax,%edx
+	addl	EXP_BIAS,%edx
 
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
+	/* A denormal and a large number can cause an exponent underflow */
+	cmpl	EXP_WAY_UNDER,%edx
+	jg	xExp_not_underflow
 
-xOp1_not_denorm:
-	movl	EXP(%ebx),%eax
-	cmpl	EXP_UNDER,%eax
-	jg	xOp2_not_denorm
+	/* Set to a really low value allow correct handling */
+	movl	EXP_WAY_UNDER,%edx
 
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
+xExp_not_underflow:
 
-xOp2_not_denorm:
-#endif DENORM_OPERAND
+	movw    %dx,EXP(%edi)
 
-ENTRY(divide_kernel)
 #ifdef PARANOID
 /*	testl	$0x80000000, SIGH(%esi)	// Dividend */
 /*	je	L_bugged */
@@ -147,7 +148,7 @@ L_no_adjust:
 
 	/* Do the shifting here */
 	/* increase the exponent */
-	incl	EXP(%edi)
+	incw	EXP(%edi)
 
 	/* shift the mantissa right one bit */
 	stc			/* To set the ms bit */
@@ -423,7 +424,7 @@ LRound_ovfl:
 	testb	$255,FPU_ovfl_flag	/* was the num > denom ? */
 	je	LRound_precision
 
-	incl	EXP(%edi)
+	incw	EXP(%edi)
 
 	/* shift the mantissa right one bit */
 	stc			/* Will set the ms bit */
@@ -433,7 +434,7 @@ LRound_ovfl:
 
 /* Round the result as required */
 LRound_precision:
-	decl	EXP(%edi)	/* binary point between 1st & 2nd bits */
+	decw	EXP(%edi)	/* binary point between 1st & 2nd bits */
 
 	movl	%eax,%edx
 	movl	FPU_result_1,%ebx
@@ -462,6 +463,7 @@ L_bugged_2:
 	jmp	L_exit
 
 L_exit:
+	movl	$-1,%eax
 	popl	%ebx
 	popl	%edi
 	popl	%esi
diff --git a/arch/i386/math-emu/reg_u_mul.S b/arch/i386/math-emu/reg_u_mul.S
index 682fbec15..b9396598d 100644
--- a/arch/i386/math-emu/reg_u_mul.S
+++ b/arch/i386/math-emu/reg_u_mul.S
@@ -4,9 +4,9 @@
  |                                                                           |
  | Core multiplication routine                                               |
  |                                                                           |
- | Copyright (C) 1992,1993,1995                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
+ | Copyright (C) 1992,1993,1995,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
@@ -15,7 +15,7 @@
  |   Basic multiplication routine.                                           |
  |   Does not check the resulting exponent for overflow/underflow            |
  |                                                                           |
- |   reg_u_mul(FPU_REG *a, FPU_REG *b, FPU_REG *c, unsigned int cw);         |
+ |   FPU_u_mul(FPU_REG *a, FPU_REG *b, FPU_REG *c, unsigned int cw);         |
  |                                                                           |
  |   Internal working is at approx 128 bits.                                 |
  |   Result is rounded to nearest 53 or 64 bits, using "nearest or even".    |
@@ -44,7 +44,7 @@ FPU_accum_1:
 
 
 .text
-ENTRY(reg_u_mul)
+ENTRY(FPU_u_mul)
 	pushl	%ebp
 	movl	%esp,%ebp
 #ifndef NON_REENTRANT_FPU
@@ -65,27 +65,6 @@ ENTRY(reg_u_mul)
 	jz	L_bugged
 #endif PARANOID
 
-#ifdef DENORM_OPERAND
-	movl	EXP(%esi),%eax
-	cmpl	EXP_UNDER,%eax
-	jg	xOp1_not_denorm
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-
-xOp1_not_denorm:
-	movl	EXP(%edi),%eax
-	cmpl	EXP_UNDER,%eax
-	jg	xOp2_not_denorm
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-
-xOp2_not_denorm:
-#endif DENORM_OPERAND
-
 	xorl	%ecx,%ecx
 	xorl	%ebx,%ebx
 
@@ -111,13 +90,22 @@ xOp2_not_denorm:
 	addl	%eax,%ebx
 	adcl	%edx,%ecx
 
-	movl	EXP(%esi),%eax	/* Compute the exponent */
-	addl	EXP(%edi),%eax
+	/* Get the sum of the exponents. */
+	movl	PARAM6,%eax
 	subl	EXP_BIAS-1,%eax
 
+	/* Two denormals can cause an exponent underflow */
+	cmpl	EXP_WAY_UNDER,%eax
+	jg	Exp_not_underflow
+
+	/* Set to a really low value allow correct handling */
+	movl	EXP_WAY_UNDER,%eax
+
+Exp_not_underflow:
+
 /*  Have now finished with the sources */
 	movl	PARAM3,%edi	/* Point to the destination */
-	movl	%eax,EXP(%edi)
+	movw	%ax,EXP(%edi)
 
 /*  Now make sure that the result is normalized */
 	testl	$0x80000000,%ecx
@@ -128,7 +116,7 @@ xOp2_not_denorm:
 	rcll	$1,FPU_accum_1
 	rcll	$1,%ebx
 	rcll	$1,%ecx
-	decl	EXP(%edi)
+	decw	EXP(%edi)
 
 LResult_Normalised:
 	movl	FPU_accum_0,%eax
diff --git a/arch/i386/math-emu/reg_u_sub.S b/arch/i386/math-emu/reg_u_sub.S
index 891670260..092f956c0 100644
--- a/arch/i386/math-emu/reg_u_sub.S
+++ b/arch/i386/math-emu/reg_u_sub.S
@@ -4,21 +4,23 @@
  |                                                                           |
  | Core floating point subtraction routine.                                  |
  |                                                                           |
- | Copyright (C) 1992,1993,1995                                              |
- |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
+ | Copyright (C) 1992,1993,1995,1997                                         |
+ |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  | Call from C as:                                                           |
- |   void reg_u_sub(FPU_REG *arg1, FPU_REG *arg2, FPU_REG *answ,             |
+ |    int FPU_u_sub(FPU_REG *arg1, FPU_REG *arg2, FPU_REG *answ,             |
  |                                                int control_w)             |
+ |    Return value is the tag of the answer, or-ed with FPU_Exception if     |
+ |    one was raised, or -1 on internal error.                               |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
 /*
- |    Kernel subtraction routine reg_u_sub(reg *arg1, reg *arg2, reg *answ).
- |    Takes two valid reg f.p. numbers (TW_Valid), which are
+ |    Kernel subtraction routine FPU_u_sub(reg *arg1, reg *arg2, reg *answ).
+ |    Takes two valid reg f.p. numbers (TAG_Valid), which are
  |    treated as unsigned numbers,
- |    and returns their difference as a TW_Valid or TW_Zero f.p.
+ |    and returns their difference as a TAG_Valid or TAG_Zero f.p.
  |    number.
  |    The first number (arg1) must be the larger.
  |    The returned number is normalized.
@@ -30,7 +32,7 @@
 #include "control_w.h"
 
 .text
-ENTRY(reg_u_sub)
+ENTRY(FPU_u_sub)
 	pushl	%ebp
 	movl	%esp,%ebp
 	pushl	%esi
@@ -39,28 +41,9 @@ ENTRY(reg_u_sub)
 
 	movl	PARAM1,%esi	/* source 1 */
 	movl	PARAM2,%edi	/* source 2 */
-
-#ifdef DENORM_OPERAND
-	cmpl	EXP_UNDER,EXP(%esi)
-	jg	xOp1_not_denorm
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-
-xOp1_not_denorm:
-	cmpl	EXP_UNDER,EXP(%edi)
-	jg	xOp2_not_denorm
-
-	call	SYMBOL_NAME(denormal_operand)
-	orl	%eax,%eax
-	jnz	fpu_Arith_exit
-
-xOp2_not_denorm:
-#endif DENORM_OPERAND
-
-	movl	EXP(%esi),%ecx
-	subl	EXP(%edi),%ecx	/* exp1 - exp2 */
+	
+	movl	PARAM6,%ecx
+	subl	PARAM7,%ecx	/* exp1 - exp2 */
 
 #ifdef PARANOID
 	/* source 2 is always smaller than source 1 */
@@ -81,10 +64,8 @@ xOp2_not_denorm:
 	movl	SIGL(%edi),%ebx	/* register ls word */
 
 	movl	PARAM3,%edi	/* destination */
-	movl	EXP(%esi),%edx
-	movl	%edx,EXP(%edi)	/* Copy exponent to destination */
-/*	movb	SIGN(%esi),%dl
-	movb	%dl,SIGN(%edi) */	/* Copy the sign from the first arg */
+	movl	PARAM6,%edx
+	movw	%dx,EXP(%edi)	/* Copy exponent to destination */
 
 	xorl	%edx,%edx	/* register extension */
 
@@ -93,8 +74,8 @@ xOp2_not_denorm:
  |      right the required number of	|
  |	places.				|
  +--------------------------------------*/
-L_shift_r:
-	cmpl	$32,%ecx		/* shrd only works for 0..31 bits */
+
+	cmpw	$32,%cx		/* shrd only works for 0..31 bits */
 	jnc	L_more_than_31
 
 /* less than 32 bits */
@@ -104,7 +85,7 @@ L_shift_r:
 	jmp	L_shift_done
 
 L_more_than_31:
-	cmpl	$64,%ecx
+	cmpw	$64,%cx
 	jnc	L_more_than_63
 
 	subb	$32,%cl
@@ -210,7 +191,7 @@ L_subtr:
 	jnz	L_must_be_zero
 
 	/* Shift left 64 bits */
-	subl	$64,EXP(%edi)
+	subw	$64,EXP(%edi)
 	xchg	%edx,%eax
 	jmp	fpu_reg_round
 
@@ -221,17 +202,17 @@ L_must_be_zero:
 #endif PARANOID
 
 	/* The result is zero */
-	movb	TW_Zero,TAG(%edi)
-	movl	$0,EXP(%edi)		/* exponent */
+	movw	$0,EXP(%edi)		/* exponent */
 	movl	$0,SIGL(%edi)
 	movl	$0,SIGH(%edi)
-	jmp	L_exit		/* %eax contains zero */
+	movl	TAG_Zero,%eax
+	jmp	L_exit
 
 L_shift_32:
 	movl	%ebx,%eax
 	movl	%edx,%ebx
 	movl	$0,%edx
-	subl	$32,EXP(%edi)	/* Can get underflow here */
+	subw	$32,EXP(%edi)	/* Can get underflow here */
 
 /* We need to shift left by 1 - 31 bits */
 L_shift_1:
@@ -241,7 +222,7 @@ L_shift_1:
 	shld	%cl,%ebx,%eax
 	shld	%cl,%edx,%ebx
 	shl	%cl,%edx
-	subl	%ecx,EXP(%edi)	/* Can get underflow here */
+	subw	%cx,EXP(%edi)	/* Can get underflow here */
 
 L_round:
 	jmp	fpu_reg_round	/* Round the result */
@@ -277,11 +258,12 @@ L_bugged:
 	call	EXCEPTION
 	pop	%ebx
 	jmp	L_error_exit
-#endif PARANOID
-
 
 L_error_exit:
-	movl	$1,%eax
+	movl	$-1,%eax
+
+#endif PARANOID
+
 L_exit:
 	popl	%ebx
 	popl	%edi
diff --git a/arch/i386/math-emu/version.h b/arch/i386/math-emu/version.h
index d966ab193..2f819087d 100644
--- a/arch/i386/math-emu/version.h
+++ b/arch/i386/math-emu/version.h
@@ -2,11 +2,11 @@
  |  version.h                                                                |
  |                                                                           |
  |                                                                           |
- | Copyright (C) 1992,1993,1994,1996                                         |
+ | Copyright (C) 1992,1993,1994,1996,1997                                    |
  |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
- |                  E-mail   billm@jacobi.maths.monash.edu.au                |
+ |                  E-mail   billm@suburbia.net                              |
  |                                                                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
-#define FPU_VERSION "wm-FPU-emu version 1.22"
+#define FPU_VERSION "wm-FPU-emu version 2.00"
diff --git a/arch/i386/math-emu/wm_shrx.S b/arch/i386/math-emu/wm_shrx.S
index 1ea7ff7da..518428317 100644
--- a/arch/i386/math-emu/wm_shrx.S
+++ b/arch/i386/math-emu/wm_shrx.S
@@ -9,9 +9,9 @@
  |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
  |                                                                           |
  | Call from C as:                                                           |
- |   unsigned shrx(void *arg1, unsigned arg2)                                |
+ |   unsigned FPU_shrx(void *arg1, unsigned arg2)                            |
  | and                                                                       |
- |   unsigned shrxs(void *arg1, unsigned arg2)                               |
+ |   unsigned FPU_shrxs(void *arg1, unsigned arg2)                           |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
@@ -19,7 +19,7 @@
 
 .text
 /*---------------------------------------------------------------------------+
- |   unsigned shrx(void *arg1, unsigned arg2)                                |
+ |   unsigned FPU_shrx(void *arg1, unsigned arg2)                            |
  |                                                                           |
  |   Extended shift right function.                                          |
  |   Fastest for small shifts.                                               |
@@ -32,7 +32,7 @@
  |   Results returned in the 64 bit arg and eax.                             |
  +---------------------------------------------------------------------------*/
 
-ENTRY(shrx)
+ENTRY(FPU_shrx)
 	push	%ebp
 	movl	%esp,%ebp
 	pushl	%esi
@@ -95,7 +95,7 @@ L_more_than_95:
 
 
 /*---------------------------------------------------------------------------+
- |   unsigned shrxs(void *arg1, unsigned arg2)                               |
+ |   unsigned FPU_shrxs(void *arg1, unsigned arg2)                           |
  |                                                                           |
  |   Extended shift right function (optimized for small floating point       |
  |   integers).                                                              |
@@ -110,7 +110,7 @@ L_more_than_95:
  |   part which has been shifted out of the arg.                             |
  |   Results returned in the 64 bit arg and eax.                             |
  +---------------------------------------------------------------------------*/
-ENTRY(shrxs)
+ENTRY(FPU_shrxs)
 	push	%ebp
 	movl	%esp,%ebp
 	pushl	%esi
diff --git a/arch/i386/math-emu/wm_sqrt.S b/arch/i386/math-emu/wm_sqrt.S
index 848796188..acc6dbbb2 100644
--- a/arch/i386/math-emu/wm_sqrt.S
+++ b/arch/i386/math-emu/wm_sqrt.S
@@ -4,12 +4,12 @@
  |                                                                           |
  | Fixed point arithmetic square root evaluation.                            |
  |                                                                           |
- | Copyright (C) 1992,1993,1995                                              |
+ | Copyright (C) 1992,1993,1995,1997                                         |
  |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
- |                       Australia.  E-mail billm@jacobi.maths.monash.edu.au |
+ |                       Australia.  E-mail billm@suburbia.net               |
  |                                                                           |
  | Call from C as:                                                           |
- |   void wm_sqrt(FPU_REG *n, unsigned int control_word)                     |
+ |    int wm_sqrt(FPU_REG *n, unsigned int control_word)                     |
  |                                                                           |
  +---------------------------------------------------------------------------*/
 
@@ -92,7 +92,7 @@ ENTRY(wm_sqrt)
 
 /* We use a rough linear estimate for the first guess.. */
 
-	cmpl	EXP_BIAS,EXP(%esi)
+	cmpw	EXP_BIAS,EXP(%esi)
 	jnz	sqrt_arg_ge_2
 
 	shrl	$1,%eax			/* arg is in the range  [1.0 .. 2.0) */
@@ -347,9 +347,8 @@ sqrt_round_result:
 	movl	%esi,%eax
 	movl	%edi,%ebx
 	movl	PARAM1,%edi
-	movl	EXP_BIAS,EXP(%edi)	/* Result is in  [1.0 .. 2.0) */
-	movl	PARAM2,%ecx
-	jmp	fpu_reg_round_sqrt
+	movw	EXP_BIAS,EXP(%edi)	/* Result is in  [1.0 .. 2.0) */
+	jmp	fpu_reg_round
 
 
 sqrt_near_exact_x: