diff --git a/ChangeLog.hppa b/ChangeLog.hppa
index 401a2a2068..e937d9bbe2 100644
--- a/ChangeLog.hppa
+++ b/ChangeLog.hppa
@@ -1,3 +1,28 @@
+2006-04-20  Carlos O'Donell  <carlos@systemhalted.org>
+
+	* sysdeps/hppa/fpu/fclrexcpt.c (feclearexcept): Use union to
+	align parameters. Specify memory clobbers.
+	* sysdeps/hppa/fpu/fedisblxcpt.c (fedisableexcept): Likewise.
+	* sysdeps/hppa/fpu/feenablxcpt.c (feenableexcept): Likewise.
+	* sysdeps/hppa/fpu/fegetenv.c (fegetenv): Do not save exception
+	register. Use memcpy to align buffer.
+	* sysdeps/hppa/fpu/fegetexcept.c (fegetexcept): Store and reload
+	fr0. Use union to align parameters.
+	* sysdeps/hppa/fpu/fegetround.c (fegetround): Likewise.
+	* sysdeps/hppa/fpu/feholdexcpt.c (feholdexcept): Do not save
+	exception registers. Define libm_hidden_def.
+	* sysdeps/hppa/fpu/fesetenv.c (fesetenv): Do not save exception
+	registers.
+	* sysdeps/hppa/fpu/fesetround.c (fesetround): Use union to
+	align parameters, speficy memory clobbers. Define libm_hidde_def
+	* sysdeps/hppa/fpu/feupdateenv.c (feupdateenv): Use union to align
+	parameters. Use memcpy to align buffer.
+	* sysdeps/hppa/fpu/fgetexcptflg.c (fegetexceptflag): Likewise.
+	* sysdeps/hppa/fpu/fsetexcptflg.c (fesetexceptflag): Likewise.
+	* sysdeps/hppa/fpu/ftestexcept.c (fetestexcept): Likewise.
+	* sysdeps/hppa/fpu/libm-test-ulps: Update.
+	* sysdeps/hppa/fpu/bits/fenv.h: Add ABI comments.
+
 2006-04-19  Carlos O'Donell  <carlos@systemhalted.org>
 
 	* sysdeps/unix/sysv/linux/hppa/bits/mman.h [__USE_GNU]: 
diff --git a/sysdeps/hppa/fpu/bits/fenv.h b/sysdeps/hppa/fpu/bits/fenv.h
index c5f8c43459..6d83b141f8 100644
--- a/sysdeps/hppa/fpu/bits/fenv.h
+++ b/sysdeps/hppa/fpu/bits/fenv.h
@@ -62,7 +62,12 @@ typedef unsigned int fexcept_t;
 
 /* Type representing floating-point environment.  This structure
    corresponds to the layout of the status and exception words in the
-   register file. */
+   register file. The exception registers are never saved/stored by
+   userspace. This structure is also not correctly aligned ever, in
+   an ABI error we left out __aligned(8) and subsequently all of our
+   fenv functions must accept unaligned input, align the input, and
+   then use assembly to store fr0. This is a performance hit, but 
+   means the ABI is stable. */
 typedef struct
 {
   unsigned int __status_word;
diff --git a/sysdeps/hppa/fpu/fclrexcpt.c b/sysdeps/hppa/fpu/fclrexcpt.c
index a7c698206e..d74a449ebb 100644
--- a/sysdeps/hppa/fpu/fclrexcpt.c
+++ b/sysdeps/hppa/fpu/fclrexcpt.c
@@ -23,14 +23,13 @@
 int
 feclearexcept (int excepts)
 {
-  unsigned int sw[2];
+  union { unsigned long long l; unsigned int sw[2]; } s;
 
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
-
+  __asm__ ("fstd %%fr0,0(%1)" : "=m" (s.l) : "r" (&s.l) : "%r0");
   /* Clear all the relevant bits. */
-  sw[0] &= ~((excepts & FE_ALL_EXCEPT) << 27);
-  __asm__ ("fldd 0(%0),%%fr0" : : "r" (sw));
+  s.sw[0] &= ~((excepts & FE_ALL_EXCEPT) << 27);
+  __asm__ ("fldd 0(%0),%%fr0" : : "r" (&s.l), "m" (s.l) : "%r0");
 
   /* Success.  */
   return 0;
diff --git a/sysdeps/hppa/fpu/fedisblxcpt.c b/sysdeps/hppa/fpu/fedisblxcpt.c
index aac6bbfa2a..8d2e664034 100644
--- a/sysdeps/hppa/fpu/fedisblxcpt.c
+++ b/sysdeps/hppa/fpu/fedisblxcpt.c
@@ -23,15 +23,16 @@
 int
 fedisableexcept (int excepts)
 {
-  unsigned int sw[2], old_exc;
+  union { unsigned long long l; unsigned int sw[2]; } s; 
+  unsigned int old_exc;
 
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
+  __asm__ ("fstd %%fr0,0(%1)" : "=m" (s.l) : "r" (&s.l) : "%r0");
 
-  old_exc = sw[0] & FE_ALL_EXCEPT;
+  old_exc = s.sw[0] & FE_ALL_EXCEPT;
 
-  sw[0] &= ~(excepts & FE_ALL_EXCEPT);
-  __asm__ ("fldd 0(%0),%%fr0" : : "r" (sw));
+  s.sw[0] &= ~(excepts & FE_ALL_EXCEPT);
+  __asm__ ("fldd 0(%0),%%fr0" : : "r" (&s.l), "m" (s.l) : "%r0");
 
   return old_exc;
 }
diff --git a/sysdeps/hppa/fpu/feenablxcpt.c b/sysdeps/hppa/fpu/feenablxcpt.c
index 9ce3ca82cc..4b17a60378 100644
--- a/sysdeps/hppa/fpu/feenablxcpt.c
+++ b/sysdeps/hppa/fpu/feenablxcpt.c
@@ -23,15 +23,16 @@
 int
 feenableexcept (int excepts)
 {
-  unsigned int sw[2], old_exc;
+  union { unsigned long long l; unsigned int sw[2]; } s;
+  unsigned int old_exc;
 
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
+  __asm__ ("fstd %%fr0,0(%1)" : "=m" (s.l) : "r" (&s.l) : "%r0");
 
-  old_exc = sw[0] & FE_ALL_EXCEPT;
+  old_exc = s.sw[0] & FE_ALL_EXCEPT;
 
-  sw[0] |= (excepts & FE_ALL_EXCEPT);
-  __asm__ ("fldd 0(%0),%%fr0" : : "r" (sw));
+  s.sw[0] |= (excepts & FE_ALL_EXCEPT);
+  __asm__ ("fldd 0(%0),%%fr0" : : "r" (&s.l), "m" (s.l) : "%r0");
 
   return old_exc;
 }
diff --git a/sysdeps/hppa/fpu/fegetenv.c b/sysdeps/hppa/fpu/fegetenv.c
index b87317b789..fcf5d2dcfe 100644
--- a/sysdeps/hppa/fpu/fegetenv.c
+++ b/sysdeps/hppa/fpu/fegetenv.c
@@ -19,15 +19,17 @@
    02111-1307 USA.  */
 
 #include <fenv.h>
+#include <string.h>
 
 int
 fegetenv (fenv_t *envp)
 {
+  unsigned long long buf[4], *bufptr = buf;
+  
   __asm__ (
-	   "fstd,ma %%fr0,8(%1)\n"
-	   "fstd,ma %%fr1,8(%1)\n"
-	   "fstd,ma %%fr2,8(%1)\n"
-	   "fstd %%fr3,0(%1)\n"
-	   : "=m" (*envp), "+r" (envp));
+	   "fstd,ma %%fr0,8(%1)	\n\t"
+	   "fldd -8(%1),%%fr0	\n\t"
+	   : "=m" (buf), "+r" (bufptr) : : "%r0");
+  memcpy(envp, buf, sizeof (*envp));
   return 0;
 }
diff --git a/sysdeps/hppa/fpu/fegetexcept.c b/sysdeps/hppa/fpu/fegetexcept.c
index efd1d7df05..d249dc6b9f 100644
--- a/sysdeps/hppa/fpu/fegetexcept.c
+++ b/sysdeps/hppa/fpu/fegetexcept.c
@@ -23,10 +23,12 @@
 int
 fegetexcept (void)
 {
-  unsigned int sw[2];
+  union { unsigned long long l; unsigned int sw[2] } s;
 
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
+  __asm__ ("fstd %%fr0,0(%1)	\n\t" 
+           "fldd 0(%1),%%fr0	\n\t"
+      	   : "=m" (s.l) : "r" (&s.l) : "%r0");
 
-  return sw[0] & FE_ALL_EXCEPT;
+  return (s.sw[0] & FE_ALL_EXCEPT);
 }
diff --git a/sysdeps/hppa/fpu/fegetround.c b/sysdeps/hppa/fpu/fegetround.c
index aefedbc071..1e606c9ecd 100644
--- a/sysdeps/hppa/fpu/fegetround.c
+++ b/sysdeps/hppa/fpu/fegetround.c
@@ -23,10 +23,12 @@
 int
 fegetround (void)
 {
-  unsigned int sw[2];
+  union { unsigned long long l; unsigned int sw[2] } s;
 
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
+  __asm__ ("fstd %%fr0,0(%1)	\n\t" 
+	   "fldd 0(%1),%%fr0	\n\t" 
+           : "=m" (s.l) : "r" (&s.l));
 
-  return sw[0] & FE_DOWNWARD;
+  return (s.sw[0] & FE_DOWNWARD);
 }
diff --git a/sysdeps/hppa/fpu/feholdexcpt.c b/sysdeps/hppa/fpu/feholdexcpt.c
index 5aec0151f8..ac6eb58055 100644
--- a/sysdeps/hppa/fpu/feholdexcpt.c
+++ b/sysdeps/hppa/fpu/feholdexcpt.c
@@ -24,33 +24,32 @@
 int
 feholdexcept (fenv_t *envp)
 {
-  fenv_t clear;
-  fenv_t * _regs = envp;
+  union { unsigned long long buf[4]; fenv_t env; } clear;
+  unsigned long long *bufptr;
 
   /* Store the environment.  */
+  bufptr = clear.buf;
   __asm__ (
 	   "fstd,ma %%fr0,8(%1)\n"
-	   "fstd,ma %%fr1,8(%1)\n"
-	   "fstd,ma %%fr2,8(%1)\n"
-	   "fstd %%fr3,0(%1)\n"
-	   : "=m" (*_regs), "+r" (_regs));
-  memcpy (&clear, envp, sizeof (clear));
-
-  /* Now clear all exceptions.  */
-  clear.__status_word &= ~(FE_ALL_EXCEPT << 27);
-  memset (clear.__exception, 0, sizeof (clear.__exception));
+	   : "=m" (clear), "+r" (bufptr) : : "%r0");
+  memcpy (envp, &clear.env, sizeof (fenv_t));
 
+  /* Clear exception queues */
+  memset (clear.env.__exception, 0, sizeof (clear.env.__exception));
   /* And set all exceptions to non-stop.  */
-  clear.__status_word &= ~FE_ALL_EXCEPT;
+  clear.env.__status_word &= ~FE_ALL_EXCEPT;
+  /* Now clear all flags  */
+  clear.env.__status_word &= ~(FE_ALL_EXCEPT << 27);
 
-  /* Load the new environment. */
-  _regs = &clear;
+  /* Load the new environment. Note: fr0 must load last to enable T-bit 
+     Thus we start bufptr at the end and work backwards */
+  bufptr = (unsigned int)(clear.buf) + sizeof(unsigned int)*4;
   __asm__ (
-	   "fldd,ma 8(%0),%%fr0\n"
-	   "fldd,ma 8(%0),%%fr1\n"
-	   "fldd,ma 8(%0),%%fr2\n"
-	   "fldd 0(%0),%%fr3\n"
-	   : : "r" (_regs));
+	   "fldd,mb -8(%0),%%fr0\n"
+	   : : "r" (bufptr), "m" (clear) : "%r0");
 
   return 0;
 }
+
+libm_hidden_def (feholdexcept)
+
diff --git a/sysdeps/hppa/fpu/fesetenv.c b/sysdeps/hppa/fpu/fesetenv.c
index 526773214b..b5753efabd 100644
--- a/sysdeps/hppa/fpu/fesetenv.c
+++ b/sysdeps/hppa/fpu/fesetenv.c
@@ -25,40 +25,38 @@
 int
 fesetenv (const fenv_t *envp)
 {
-  fenv_t temp;
-  fenv_t * _regs = &temp;
+  union { unsigned long long buf[4]; fenv_t env; } temp;
+  unsigned long long *bufptr;
 
   /* Install the environment specified by ENVP.  But there are a few
      values which we do not want to come from the saved environment.
      Therefore, we get the current environment and replace the values
      we want to use from the environment specified by the parameter.  */
+  bufptr = temp.buf;
   __asm__ (
 	   "fstd,ma %%fr0,8(%1)\n"
-	   "fstd,ma %%fr1,8(%1)\n"
-	   "fstd,ma %%fr2,8(%1)\n"
-	   "fstd %%fr3,0(%1)\n"
-	   : "=m" (*_regs), "+r" (_regs));
+	   : "=m" (temp), "+r" (bufptr) : : "%r0");
 
-  temp.__status_word &= ~(FE_ALL_EXCEPT
-			  | (FE_ALL_EXCEPT << 27)
-			  | FE_DOWNWARD);
+  temp.env.__status_word &= ~(FE_ALL_EXCEPT
+			    | (FE_ALL_EXCEPT << 27)
+			    | FE_DOWNWARD);
   if (envp == FE_DFL_ENV)
     ;
   else if (envp == FE_NOMASK_ENV)
-    temp.__status_word |= FE_ALL_EXCEPT;
+    temp.env.__status_word |= FE_ALL_EXCEPT;
   else
-    temp.__status_word |= (envp->__status_word
-			   & (FE_ALL_EXCEPT
-			      | FE_DOWNWARD
-			      | (FE_ALL_EXCEPT << 27)));
+    temp.env.__status_word |= (envp->__status_word
+			       & (FE_ALL_EXCEPT
+				  | FE_DOWNWARD
+				  | (FE_ALL_EXCEPT << 27)));
 
-  /* Load the new environment. */
+  /* Load the new environment. We use bufptr again since the 
+     initial asm has modified the value of the register and here
+     we take advantage of that to load in reverse order so fr0
+     is loaded last and T-Bit is enabled. */
   __asm__ (
-	   "fldd,ma -8(%1),%%fr3\n"
-	   "fldd,ma -8(%1),%%fr2\n"
-	   "fldd,ma -8(%1),%%fr1\n"
-	   "fldd 0(%1),%%fr0\n"
-	   : "=m" (*_regs), "+r" (_regs));
+	   "fldd,mb -8(%1),%%fr0\n"
+	   : "=m" (temp), "+r" (bufptr) : : "%r0" );
 
   /* Success.  */
   return 0;
diff --git a/sysdeps/hppa/fpu/fesetround.c b/sysdeps/hppa/fpu/fesetround.c
index 3687624c2b..9f30c2495f 100644
--- a/sysdeps/hppa/fpu/fesetround.c
+++ b/sysdeps/hppa/fpu/fesetround.c
@@ -23,17 +23,19 @@
 int
 fesetround (int round)
 {
-  unsigned int sw[2];
+  union { unsigned long long l; unsigned int sw[2]; } s;
 
   if (round & ~FE_DOWNWARD)
-    /* ROUND is not a valid rounding mode.  */
+    /* round is not a valid rounding mode. */
     return 1;
-
+  
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
-  sw[0] &= ~FE_DOWNWARD;
-  sw[0] |= round;
-  __asm__ ("fldd 0(%0),%%fr0" : : "r" (sw));
+  __asm__ ("fstd %%fr0,0(%1)" : "=m" (s.l) : "r" (&s.l) : "%r0");
+  s.sw[0] &= ~FE_DOWNWARD;
+  s.sw[0] |= round & FE_DOWNWARD;
+  __asm__ ("fldd 0(%0),%%fr0" : : "r" (&s.l), "m" (s.l) : "%r0");
 
   return 0;
 }
+
+libm_hidden_def (fesetround)
diff --git a/sysdeps/hppa/fpu/feupdateenv.c b/sysdeps/hppa/fpu/feupdateenv.c
index 7d50282e05..17140060ad 100644
--- a/sysdeps/hppa/fpu/feupdateenv.c
+++ b/sysdeps/hppa/fpu/feupdateenv.c
@@ -19,19 +19,22 @@
    02111-1307 USA.  */
 
 #include <fenv.h>
+#include <string.h>
 
 int
 feupdateenv (const fenv_t *envp)
 {
-  unsigned int sw[2];
-
-  /* Get the current exception status. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
+  union { unsigned long long l; unsigned int sw[2]; } s;
+  fenv_t temp;
+  /* Get the current exception status */
+  __asm__ ("fstd %%fr0,0(%1)	\n\t" 
+           "fldd 0(%1),%%fr0	\n\t" 
+	   : "=m" (s.l) : "r" (&s.l));
+  memcpy(&temp, envp, sizeof(fenv_t));
+  /* Currently raised exceptions not cleared */
+  temp.__status_word |= s.sw[0] & (FE_ALL_EXCEPT << 27);
   /* Install new environment.  */
-  fesetenv (envp);
-  /* Raise the saved exceptions */
-  feraiseexcept(sw[0] & FE_ALL_EXCEPT);
-
+  fesetenv (&temp);
   /* Success.  */
   return 0;
 }
diff --git a/sysdeps/hppa/fpu/fgetexcptflg.c b/sysdeps/hppa/fpu/fgetexcptflg.c
index 27766ecf58..d5bcfe31d6 100644
--- a/sysdeps/hppa/fpu/fgetexcptflg.c
+++ b/sysdeps/hppa/fpu/fgetexcptflg.c
@@ -23,12 +23,14 @@
 int
 fegetexceptflag (fexcept_t *flagp, int excepts)
 {
-  unsigned int sw[2];
+  union { unsigned long long l; unsigned int sw[2]; } s;
 
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
+  __asm__ ("fstd %%fr0,0(%1)	\n\t" 
+           "fldd 0(%1),%%fr0	\n\t" 
+      	   : "=m" (s.l) : "r" (&s.l) : "%r0");
 
-  *flagp = (sw[0] >> 27) & excepts & FE_ALL_EXCEPT;
+  *flagp = (s.sw[0] >> 27) & excepts & FE_ALL_EXCEPT;
 
   /* Success.  */
   return 0;
diff --git a/sysdeps/hppa/fpu/fsetexcptflg.c b/sysdeps/hppa/fpu/fsetexcptflg.c
index af35f5ae35..4ec3a92dcc 100644
--- a/sysdeps/hppa/fpu/fsetexcptflg.c
+++ b/sysdeps/hppa/fpu/fsetexcptflg.c
@@ -24,16 +24,14 @@
 int
 fesetexceptflag (const fexcept_t *flagp, int excepts)
 {
-  unsigned int sw[2];
+  union { unsigned long long l; unsigned int sw[2]; } s;
 
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
-
-  /* Install new enable trap bits  */
-  sw[0] |= (*flagp & excepts & FE_ALL_EXCEPT) << 27;
-
+  __asm__ ("fstd %%fr0,0(%1)" : "=m" (s.l) : "r" (&s.l) : "%r0");
+  /* Install new raised trap bits */
+  s.sw[0] |= (*flagp & excepts & FE_ALL_EXCEPT) << 27;
   /* Store the new status word.  */
-  __asm__ ("fldd 0(%0),%%fr0" : : "r" (sw));
+  __asm__ ("fldd 0(%0),%%fr0" : : "r" (&s.l), "m" (s.l) : "%r0");
 
   /* Success.  */
   return 0;
diff --git a/sysdeps/hppa/fpu/ftestexcept.c b/sysdeps/hppa/fpu/ftestexcept.c
index d08d4d6eb9..ac6d4b2e35 100644
--- a/sysdeps/hppa/fpu/ftestexcept.c
+++ b/sysdeps/hppa/fpu/ftestexcept.c
@@ -23,10 +23,12 @@
 int
 fetestexcept (int excepts)
 {
-  unsigned int sw[2];
+  union { unsigned long long l; unsigned int sw[2] } s;
 
   /* Get the current status word. */
-  __asm__ ("fstd %%fr0,0(%1)" : "=m" (*sw) : "r" (sw));
+  __asm__ ("fstd %%fr0,0(%1)	\n\t" 
+           "fldd 0(%1),%%fr0	\n\t" 
+      	   : "=m" (s.l) : "r" (&s.l));
 
-  return (sw[0] >> 27) & excepts & FE_ALL_EXCEPT;
+  return (s.sw[0] >> 27) & excepts & FE_ALL_EXCEPT;
 }
diff --git a/sysdeps/hppa/fpu/libm-test-ulps b/sysdeps/hppa/fpu/libm-test-ulps
index b514496676..c4ffefaa86 100644
--- a/sysdeps/hppa/fpu/libm-test-ulps
+++ b/sysdeps/hppa/fpu/libm-test-ulps
@@ -1,6 +1,9 @@
 # Begin of automatic generation
 
 # atan2
+Test "atan2 (-0.00756827042671106339, -.001792735857538728036) == -1.80338464113663849327153994380":
+float: 6
+ifloat: 6
 Test "atan2 (-0.75, -1.0) == -2.49809154479650885165983415456218025":
 float: 3
 ifloat: 3
@@ -258,9 +261,6 @@ float: 1
 ifloat: 1
 
 # ctan
-Test "Real part of: ctan (-2 - 3 i) == 0.376402564150424829275122113032269084e-2 - 1.00323862735360980144635859782192726 i":
-double: 1
-idouble: 1
 Test "Imaginary part of: ctan (0.75 + 1.25 i) == 0.160807785916206426725166058173438663 + 0.975363285031235646193581759755216379 i":
 double: 1
 idouble: 1
@@ -479,6 +479,11 @@ Test "log1p (-0.25) == -0.287682072451780927439219005993827432":
 float: 1
 ifloat: 1
 
+# lround
+Test "lround (1071930.0008) == 1071930":
+double: -214511494
+idouble: -214511494
+
 # sincos
 Test "sincos (M_PI_6l*2.0, &sin_res, &cos_res) puts 0.5 in cos_res":
 double: 1
@@ -640,8 +645,8 @@ idouble: 1
 
 # Maximal error of functions:
 Function: "atan2":
-float: 3
-ifloat: 3
+float: 6
+ifloat: 6
 
 Function: "atanh":
 float: 1
@@ -777,10 +782,6 @@ Function: Real part of "csqrt":
 float: 1
 ifloat: 1
 
-Function: Real part of "ctan":
-double: 1
-idouble: 1
-
 Function: Imaginary part of "ctan":
 double: 1
 idouble: 1