about summary refs log tree commit diff
path: root/sysdeps/powerpc/fpu
diff options
context:
space:
mode:
authorAdhemerval Zanella <azanella@linux.vnet.ibm.com>2014-04-28 14:38:24 -0500
committerAdhemerval Zanella <azanella@linux.vnet.ibm.com>2014-04-29 07:05:39 -0500
commit18f2945ae9216cfcd53a162080a73e3d719de9e6 (patch)
tree8d529fe01c41f0d3c6dd290aa69dbf4e5d6e083f /sysdeps/powerpc/fpu
parent5abebba403181de898bbea4ee1bcce5f088c663b (diff)
downloadglibc-18f2945ae9216cfcd53a162080a73e3d719de9e6.tar.gz
glibc-18f2945ae9216cfcd53a162080a73e3d719de9e6.tar.xz
glibc-18f2945ae9216cfcd53a162080a73e3d719de9e6.zip
PowerPC: Suppress unnecessary FPSCR write
This patch optimizes the FPSCR update on exception and rounding change
functions by just updating its value if new value if different from
current one.  It also optimizes fedisableexcept and feenableexcept by
removing an unecessary FPSCR read.
Diffstat (limited to 'sysdeps/powerpc/fpu')
-rw-r--r--sysdeps/powerpc/fpu/fclrexcpt.c7
-rw-r--r--sysdeps/powerpc/fpu/fedisblxcpt.c14
-rw-r--r--sysdeps/powerpc/fpu/feenablxcpt.c14
-rw-r--r--sysdeps/powerpc/fpu/feholdexcpt.c3
-rw-r--r--sysdeps/powerpc/fpu/fenv_libc.h17
-rw-r--r--sysdeps/powerpc/fpu/fesetenv.c2
-rw-r--r--sysdeps/powerpc/fpu/fsetexcptflg.c7
7 files changed, 48 insertions, 16 deletions
diff --git a/sysdeps/powerpc/fpu/fclrexcpt.c b/sysdeps/powerpc/fpu/fclrexcpt.c
index cda2810120..4607f62bb1 100644
--- a/sysdeps/powerpc/fpu/fclrexcpt.c
+++ b/sysdeps/powerpc/fpu/fclrexcpt.c
@@ -22,17 +22,18 @@
 int
 __feclearexcept (int excepts)
 {
-  fenv_union_t u;
+  fenv_union_t u, n;
 
   /* Get the current state.  */
   u.fenv = fegetenv_register ();
 
   /* Clear the relevant bits.  */
-  u.l = u.l & ~((-(excepts >> (31 - FPSCR_VX) & 1) & FE_ALL_INVALID)
+  n.l = u.l & ~((-(excepts >> (31 - FPSCR_VX) & 1) & FE_ALL_INVALID)
 		| (excepts & FPSCR_STICKY_BITS));
 
   /* Put the new state in effect.  */
-  fesetenv_register (u.fenv);
+  if (u.l != n.l)
+    fesetenv_register (n.fenv);
 
   /* Success.  */
   return 0;
diff --git a/sysdeps/powerpc/fpu/fedisblxcpt.c b/sysdeps/powerpc/fpu/fedisblxcpt.c
index 5883e09263..94c01abb86 100644
--- a/sysdeps/powerpc/fpu/fedisblxcpt.c
+++ b/sysdeps/powerpc/fpu/fedisblxcpt.c
@@ -22,15 +22,17 @@
 int
 fedisableexcept (int excepts)
 {
-  fenv_union_t fe;
-  int result, new;
+  fenv_union_t fe, curr;
+  int result = 0, new;
 
-  result = __fegetexcept ();
+  /* Get current exception mask to return.  */
+  fe.fenv = curr.fenv = fegetenv_register ();
+  result = fenv_reg_to_exceptions (fe.l);
 
   if ((excepts & FE_ALL_INVALID) == FE_ALL_INVALID)
     excepts = (excepts | FE_INVALID) & ~ FE_ALL_INVALID;
 
-  fe.fenv = fegetenv_register ();
+  /* Sets the new exception mask.  */
   if (excepts & FE_INEXACT)
     fe.l &= ~(1 << (31 - FPSCR_XE));
   if (excepts & FE_DIVBYZERO)
@@ -41,7 +43,9 @@ fedisableexcept (int excepts)
     fe.l &= ~(1 << (31 - FPSCR_OE));
   if (excepts & FE_INVALID)
     fe.l &= ~(1 << (31 - FPSCR_VE));
-  fesetenv_register (fe.fenv);
+
+  if (fe.l != curr.l)
+    fesetenv_register (fe.fenv);
 
   new = __fegetexcept ();
   if (new == 0 && result != 0)
diff --git a/sysdeps/powerpc/fpu/feenablxcpt.c b/sysdeps/powerpc/fpu/feenablxcpt.c
index 8ad0f97ac7..01a68cf086 100644
--- a/sysdeps/powerpc/fpu/feenablxcpt.c
+++ b/sysdeps/powerpc/fpu/feenablxcpt.c
@@ -22,15 +22,17 @@
 int
 feenableexcept (int excepts)
 {
-  fenv_union_t fe;
-  int result, new;
+  fenv_union_t fe, curr;
+  int result = 0, new;
 
-  result = __fegetexcept ();
+  /* Get current exception mask to return.  */
+  fe.fenv = curr.fenv = fegetenv_register ();
+  result = fenv_reg_to_exceptions (fe.l);
 
   if ((excepts & FE_ALL_INVALID) == FE_ALL_INVALID)
     excepts = (excepts | FE_INVALID) & ~ FE_ALL_INVALID;
 
-  fe.fenv = fegetenv_register ();
+  /* Sets the new exception mask.  */
   if (excepts & FE_INEXACT)
     fe.l |= (1 << (31 - FPSCR_XE));
   if (excepts & FE_DIVBYZERO)
@@ -41,7 +43,9 @@ feenableexcept (int excepts)
     fe.l |= (1 << (31 - FPSCR_OE));
   if (excepts & FE_INVALID)
     fe.l |= (1 << (31 - FPSCR_VE));
-  fesetenv_register (fe.fenv);
+
+  if (fe.l != curr.l)
+    fesetenv_register (fe.fenv);
 
   new = __fegetexcept ();
   if (new != 0 && result == 0)
diff --git a/sysdeps/powerpc/fpu/feholdexcpt.c b/sysdeps/powerpc/fpu/feholdexcpt.c
index 1375a2f5ad..764dd38f3d 100644
--- a/sysdeps/powerpc/fpu/feholdexcpt.c
+++ b/sysdeps/powerpc/fpu/feholdexcpt.c
@@ -32,6 +32,9 @@ feholdexcept (fenv_t *envp)
      flag.  */
   new.l = old.l & 0xffffffff00000007LL;
 
+  if (new.l == old.l)
+    return 0;
+
   /* If the old env had any enabled exceptions, then mask SIGFPE in the
      MSR FE0/FE1 bits.  This may allow the FPU to run faster because it
      always takes the default action and can not generate SIGFPE. */
diff --git a/sysdeps/powerpc/fpu/fenv_libc.h b/sysdeps/powerpc/fpu/fenv_libc.h
index 28e4d1f7c9..c1df5ce091 100644
--- a/sysdeps/powerpc/fpu/fenv_libc.h
+++ b/sysdeps/powerpc/fpu/fenv_libc.h
@@ -146,6 +146,23 @@ enum {
   /* the remaining two least-significant bits keep the rounding mode */
 };
 
+static inline int
+fenv_reg_to_exceptions (unsigned long long l)
+{
+  int result = 0;
+  if (l & (1 << (31 - FPSCR_XE)))
+    result |= FE_INEXACT;
+  if (l & (1 << (31 - FPSCR_ZE)))
+    result |= FE_DIVBYZERO;
+  if (l & (1 << (31 - FPSCR_UE)))
+    result |= FE_UNDERFLOW;
+  if (l & (1 << (31 - FPSCR_OE)))
+    result |= FE_OVERFLOW;
+  if (l & (1 << (31 - FPSCR_VE)))
+    result |= FE_INVALID;
+  return result;
+}
+
 #ifdef _ARCH_PWR6
   /* Not supported in ISA 2.05.  Provided for source compat only.  */
 # define FPSCR_NI 29
diff --git a/sysdeps/powerpc/fpu/fesetenv.c b/sysdeps/powerpc/fpu/fesetenv.c
index fa99ddbd40..138bde0685 100644
--- a/sysdeps/powerpc/fpu/fesetenv.c
+++ b/sysdeps/powerpc/fpu/fesetenv.c
@@ -29,6 +29,8 @@ __fesetenv (const fenv_t *envp)
   /* get the currently set exceptions.  */
   new.fenv = *envp;
   old.fenv = fegetenv_register ();
+  if (old.l == new.l)
+    return 0;
 
   /* If the old env has no enabled exceptions and the new env has any enabled
      exceptions, then unmask SIGFPE in the MSR FE0/FE1 bits.  This will put the
diff --git a/sysdeps/powerpc/fpu/fsetexcptflg.c b/sysdeps/powerpc/fpu/fsetexcptflg.c
index 95193c1010..cee6682b64 100644
--- a/sysdeps/powerpc/fpu/fsetexcptflg.c
+++ b/sysdeps/powerpc/fpu/fsetexcptflg.c
@@ -21,7 +21,7 @@
 int
 __fesetexceptflag (const fexcept_t *flagp, int excepts)
 {
-  fenv_union_t u;
+  fenv_union_t u, n;
   fexcept_t flag;
 
   /* Get the current state.  */
@@ -31,7 +31,7 @@ __fesetexceptflag (const fexcept_t *flagp, int excepts)
   flag = *flagp & excepts;
 
   /* Replace the exception status */
-  u.l = ((u.l & ~(FPSCR_STICKY_BITS & excepts))
+  n.l = ((u.l & ~(FPSCR_STICKY_BITS & excepts))
 	 | (flag & FPSCR_STICKY_BITS)
 	 | (flag >> ((31 - FPSCR_VX) - (31 - FPSCR_VXSOFT))
 	    & FE_INVALID_SOFTWARE));
@@ -39,7 +39,8 @@ __fesetexceptflag (const fexcept_t *flagp, int excepts)
   /* Store the new status word (along with the rest of the environment).
      This may cause floating-point exceptions if the restored state
      requests it.  */
-  fesetenv_register (u.fenv);
+  if (n.l != u.l)
+    fesetenv_register (u.fenv);
 
   /* Deal with FE_INVALID_SOFTWARE not being implemented on some chips.  */
   if (flag & FE_INVALID)