Support SSE float environment in fenv.h functions.

	* cpu_features.c: New file.
	* cpu_features.h: New file.
	* crt1.c: Include "cpu_features.h".
	(__mingw_CRTStartup): Call cpu_features_init().
	* Makefile.in (MING_OBJS): Add cpu_features.c.
	(SRCDIST_FILES): Add cpu_features.c, cpu_features.h.
	* include/fenv,h ( fenv_t;): Append  __mxcsr field.
	(__MXCSR_EXCEPT_FLAG_SHIFT): New define.
	(__MXCSR_EXCEPT_MASK_SHIFT): New define.
	(__MXCSR_ROUND_FLAG_SHIFT): New define.
	* mingwex/feclearexcept.c: Include "cpu_features.h".
	Handle SSE environment.
	* mingwex/fegetenv.c: Likewise.
	* mingwex/feholdexcept.c: Likewise.
	* mingwex/fesetenv.c: Likewise.
	* mingwex/fesetexceptflag.c: Likewise.
	* mingwex/fesetround.c: Likewise.
	* mingwex/fetestexcept.c: Likewise.
	* mingwex/feupdateenv.c: Likewise.
	* mingwex/fegetround.c: Add comment.

9 files changed, 99 insertions, 11 deletions

diff --git a/winsup/mingw/mingwex/feclearexcept.c b/winsup/mingw/mingwex/feclearexcept.c
index 8c943893b..a68884f41 100644
--- a/winsup/mingw/mingwex/feclearexcept.c
+++ b/winsup/mingw/mingwex/feclearexcept.c
@@ -1,4 +1,5 @@
 #include <fenv.h>
+#include "cpu_features.h"
 
 /* 7.6.2.1
    The feclearexcept function clears the supported exceptions
@@ -7,9 +8,17 @@
 int feclearexcept (int excepts)
 {
   fenv_t _env;
+  excepts &= FE_ALL_EXCEPT;
   __asm__ volatile ("fnstenv %0;" : "=m" (_env)); /* get the env */
-  _env.__status_word &= ~(excepts & FE_ALL_EXCEPT); /* clear the except */
+  _env.__status_word &= ~excepts; /* clear the except */
   __asm__ volatile ("fldenv %0;" :: "m" (_env)); /*set the env */
 
+  if (__HAS_SSE)
+    {
+       unsigned _csr;
+       __asm__ volatile("stmxcsr %0" : "=m" (_csr)); /* get the register */
+       _csr &= ~excepts; /* clear the except */
+       __asm__ volatile("ldmxcsr %0" : : "m" (_csr)); /* set the register */
+    }
   return 0;
 }
diff --git a/winsup/mingw/mingwex/fegetenv.c b/winsup/mingw/mingwex/fegetenv.c
index 930105673..0c5d591e7 100644
--- a/winsup/mingw/mingwex/fegetenv.c
+++ b/winsup/mingw/mingwex/fegetenv.c
@@ -1,4 +1,5 @@
 #include <fenv.h>
+#include "cpu_features.h"
 
 /* 7.6.4.1
    The fegetenv function stores the current floating-point environment
@@ -10,5 +11,10 @@ int fegetenv (fenv_t * envp)
  /* fnstenv sets control word to non-stop for all exceptions, so we
     need to reload our env to restore the original mask.  */
   __asm__ ("fldenv %0" : : "m" (*envp));
+
+  /* And the SSE environment.  */
+  if (__HAS_SSE)
+    __asm__ ("stmxcsr %0" : "=m" (envp->__mxcsr));
+
   return 0;
 }
diff --git a/winsup/mingw/mingwex/fegetround.c b/winsup/mingw/mingwex/fegetround.c
index 076b1068c..01d42849f 100644
--- a/winsup/mingw/mingwex/fegetround.c
+++ b/winsup/mingw/mingwex/fegetround.c
@@ -1,4 +1,5 @@
 #include <fenv.h>
+#include "cpu_features.h"
 
 /* 7.6.3.1
    The fegetround function returns the value of the rounding direction
@@ -9,6 +10,10 @@ fegetround (void)
 {
   unsigned short _cw;
   __asm__ ("fnstcw %0;"	: "=m" (_cw));
+
+  /* If the MXCSR flag is different, there is no way to indicate, so just
+     report the FPU flag. */
   return _cw
           & (FE_TONEAREST | FE_DOWNWARD | FE_UPWARD | FE_TOWARDZERO);
+
 }
diff --git a/winsup/mingw/mingwex/feholdexcept.c b/winsup/mingw/mingwex/feholdexcept.c
index 8d4a3f67b..66912d85d 100644
--- a/winsup/mingw/mingwex/feholdexcept.c
+++ b/winsup/mingw/mingwex/feholdexcept.c
@@ -1,4 +1,5 @@
 #include <fenv.h>
+#include "cpu_features.h"
 
 /* 7.6.4.2
    The feholdexcept function saves the current floating-point
@@ -12,5 +13,18 @@ int feholdexcept (fenv_t * envp)
  /* fnstenv sets control word to non-stop for all exceptions, so all we
     need to do is clear the exception flags.  */
   __asm__ ("fnclex");
+
+  if (__HAS_SSE)
+    {
+       unsigned int _csr;
+       /* Save the SSE MXCSR register.  */
+       __asm__ ("stmxcsr %0" :  "=m" (envp->__mxcsr));  
+       /* Clear the exception flags. */
+       _csr = envp->__mxcsr & ~FE_ALL_EXCEPT;
+       /* Set exception mask to non-stop  */
+       _csr |= (FE_ALL_EXCEPT << __MXCSR_EXCEPT_MASK_SHIFT) /*= 0x1f80 */;
+       __asm__ volatile ("ldmxcsr %0" : : "m" (_csr));
+    } 
+
   return 0;
 }
diff --git a/winsup/mingw/mingwex/fesetenv.c b/winsup/mingw/mingwex/fesetenv.c
index 3d7b5604a..105a0af10 100644
--- a/winsup/mingw/mingwex/fesetenv.c
+++ b/winsup/mingw/mingwex/fesetenv.c
@@ -1,5 +1,6 @@
 #include <fenv.h>
 #include <float.h>
+#include "cpu_features.h"
 
 /* 7.6.4.3
    The fesetenv function establishes the floating-point environment
@@ -15,6 +16,11 @@ extern void (*_imp___fpreset)( void ) ;
 
 int fesetenv (const fenv_t * envp)
 {
+  /* Default mxcsr status is to mask all exceptions.  All other bits
+     are zero.  */
+     
+  unsigned int _csr = FE_ALL_EXCEPT << __MXCSR_EXCEPT_MASK_SHIFT /*= 0x1f80 */; 
+  
   if (envp == FE_PC64_ENV)
    /*
     *  fninit initializes the control register to 0x37f,
@@ -37,7 +43,15 @@ int fesetenv (const fenv_t * envp)
     _fpreset();
 
   else
-    __asm__ ("fldenv %0;" : : "m" (*envp));
-
+    {
+      __asm__ ("fldenv %0;" : : "m" (*envp));
+       /* Setting the reserved high order bits of MXCSR causes a segfault */
+       _csr = envp ->__mxcsr & 0xffff;
+    }
+
+  /* Set MXCSR */   
+   if (__HAS_SSE)
+     __asm__ volatile ("ldmxcsr %0" : : "m" (_csr));
+ 
   return 0;
 }
diff --git a/winsup/mingw/mingwex/fesetexceptflag.c b/winsup/mingw/mingwex/fesetexceptflag.c
index 7f4b8e562..876174b69 100644
--- a/winsup/mingw/mingwex/fesetexceptflag.c
+++ b/winsup/mingw/mingwex/fesetexceptflag.c
@@ -1,4 +1,5 @@
 #include <fenv.h> 
+#include "cpu_features.h"
 
 /* 7.6.2.4
    The fesetexceptflag function sets the complete status for those
@@ -18,5 +19,15 @@ int fesetexceptflag (const fexcept_t * flagp, int excepts)
   _env.__status_word &= ~excepts;
   _env.__status_word |= (*flagp & excepts);
   __asm__ volatile ("fldenv %0;" : : "m" (_env));
+
+  if (__HAS_SSE)
+   {
+      unsigned int _csr;
+      __asm__ __volatile__("stmxcsr %0" : "=m" (_csr));
+      _csr &= ~excepts;
+      _csr |= *flagp & excepts;
+      __asm__ volatile ("ldmxcsr %0" : : "m" (_csr));
+   }
+
   return 0;
 }
diff --git a/winsup/mingw/mingwex/fesetround.c b/winsup/mingw/mingwex/fesetround.c
index a8cef86a4..e90dbb4e7 100644
--- a/winsup/mingw/mingwex/fesetround.c
+++ b/winsup/mingw/mingwex/fesetround.c
@@ -1,4 +1,6 @@
 #include <fenv.h>
+#include "cpu_features.h"
+
  /* 7.6.3.2
     The fesetround function establishes the rounding direction
     represented by its argument round. If the argument is not equal
@@ -15,5 +17,14 @@ int fesetround (int mode)
   _cw &= ~(FE_TONEAREST | FE_DOWNWARD | FE_UPWARD | FE_TOWARDZERO);
   _cw |= mode;
   __asm__ volatile ("fldcw %0;" : : "m" (_cw));
+
+ if (__HAS_SSE)
+    {
+      __asm__ volatile ("stmxcsr %0" : "=m" (_cw));
+      _cw &= ~ 0x6000;
+      _cw |= (mode <<  __MXCSR_ROUND_FLAG_SHIFT);
+      __asm__ volatile ("ldmxcsr %0" : : "m" (_cw));
+    }
+
   return 0;
 }
diff --git a/winsup/mingw/mingwex/fetestexcept.c b/winsup/mingw/mingwex/fetestexcept.c
index 6934ed925..062d585eb 100644
--- a/winsup/mingw/mingwex/fetestexcept.c
+++ b/winsup/mingw/mingwex/fetestexcept.c
@@ -1,4 +1,5 @@
-#include <fenv.h> 
+#include <fenv.h>
+#include "cpu_features.h"
 /* 7.6.2.5 
    The fetestexcept function determines which of a specified subset of
    the exception flags are currently set. The excepts argument
@@ -9,7 +10,18 @@
 
 int fetestexcept (int excepts)
 {
-  unsigned short _sw;
-  __asm__ ("fnstsw %%ax" : "=a" (_sw));
-  return _sw & excepts & FE_ALL_EXCEPT;
+  
+  unsigned int _res;
+  __asm__ ("fnstsw %%ax" : "=a" (_res));
+
+
+  /* If SSE supported, return the union of the FPU and SSE flags.  */
+  if (__HAS_SSE)
+    {     
+        unsigned int _csr; 	
+      __asm__ volatile("stmxcsr %0" : "=m" (_csr));
+      _res |= _csr;    
+    }
+
+  return (_res & excepts & FE_ALL_EXCEPT);
 }
diff --git a/winsup/mingw/mingwex/feupdateenv.c b/winsup/mingw/mingwex/feupdateenv.c
index f414837f5..e37566b02 100644
--- a/winsup/mingw/mingwex/feupdateenv.c
+++ b/winsup/mingw/mingwex/feupdateenv.c
@@ -1,4 +1,5 @@
 #include <fenv.h>
+#include "cpu_features.h"
 
 /* 7.6.4.4
    The feupdateenv function saves the currently raised exceptions in
@@ -8,13 +9,18 @@
    set by a call to feholdexcept or fegetenv, or equal the macro
    FE_DFL_ENV or an implementation-defined environment macro. */
 
-/* FIXME: this works but surely there must be a better way.  */
 
 int feupdateenv (const fenv_t * envp)
 {
-  unsigned int _fexcept = fetestexcept (FE_ALL_EXCEPT); /*save excepts */
+  unsigned int _fexcept;
+  __asm__ ("fnstsw %%ax" : "=a" (_fexcept)); /*save excepts */
+  if (__HAS_SSE)
+    {
+      unsigned int  _csr;
+      __asm__ ("stmxcsr %0" : "=m" (_csr));
+      _fexcept |= _csr;
+    }
   fesetenv (envp); /* install the env  */
-  feraiseexcept (_fexcept); /* raise the execept */
+  feraiseexcept (_fexcept & FE_ALL_EXCEPT); /* raise the execeptions */
   return 0;
 }
-