From 91c28f61f43ba029166772e8ac25808ea3c3dc98 Mon Sep 17 00:00:00 2001
From: nsz <nsz@port70.net>
Date: Tue, 20 Mar 2012 22:49:19 +0100
Subject: nearbyint optimization (only clear inexact when necessary)

old code saved/restored the fenv (the new code is only as slow
as that when inexact is not set before the call, but some other
flag is set and the rounding is inexact, which is rare)

before:
bench_nearbyint_exact              5000000 N        261 ns/op
bench_nearbyint_inexact_set        5000000 N        262 ns/op
bench_nearbyint_inexact_unset      5000000 N        261 ns/op

after:
bench_nearbyint_exact             10000000 N         94.99 ns/op
bench_nearbyint_inexact_set       25000000 N         65.81 ns/op
bench_nearbyint_inexact_unset     10000000 N         94.97 ns/op
---
 src/math/nearbyint.c  | 23 +++++++++++------------
 src/math/nearbyintf.c | 14 ++++++++++----
 src/math/nearbyintl.c | 11 ++++++++---
 3 files changed, 29 insertions(+), 19 deletions(-)

(limited to 'src')

diff --git a/src/math/nearbyint.c b/src/math/nearbyint.c
index 714c55ca..7a4c58cf 100644
--- a/src/math/nearbyint.c
+++ b/src/math/nearbyint.c
@@ -1,20 +1,19 @@
 #include <fenv.h>
 #include <math.h>
 
-/*
-rint may raise inexact (and it should not alter the fenv otherwise)
-nearbyint must not raise inexact
+/* nearbyint is the same as rint, but it must not raise the inexact exception */
 
-(according to ieee754r section 7.9 both functions should raise invalid
-when the input is signaling nan, but c99 does not define snan so saving
-and restoring the entire fenv should be fine)
-*/
+double nearbyint(double x)
+{
+#ifdef FE_INEXACT
+	int e;
 
-double nearbyint(double x) {
-	fenv_t e;
-
-	fegetenv(&e);
+	e = fetestexcept(FE_INEXACT);
+#endif
 	x = rint(x);
-	fesetenv(&e);
+#ifdef FE_INEXACT
+	if (!e)
+		feclearexcept(FE_INEXACT);
+#endif
 	return x;
 }
diff --git a/src/math/nearbyintf.c b/src/math/nearbyintf.c
index 07df8f54..39c3d73b 100644
--- a/src/math/nearbyintf.c
+++ b/src/math/nearbyintf.c
@@ -1,11 +1,17 @@
 #include <fenv.h>
 #include <math.h>
 
-float nearbyintf(float x) {
-	fenv_t e;
+float nearbyintf(float x)
+{
+#ifdef FE_INEXACT
+	int e;
 
-	fegetenv(&e);
+	e = fetestexcept(FE_INEXACT);
+#endif
 	x = rintf(x);
-	fesetenv(&e);
+#ifdef FE_INEXACT
+	if (!e)
+		feclearexcept(FE_INEXACT);
+#endif
 	return x;
 }
diff --git a/src/math/nearbyintl.c b/src/math/nearbyintl.c
index 2906f383..0ff4b1f9 100644
--- a/src/math/nearbyintl.c
+++ b/src/math/nearbyintl.c
@@ -10,11 +10,16 @@ long double nearbyintl(long double x)
 #include <fenv.h>
 long double nearbyintl(long double x)
 {
-	fenv_t e;
+#ifdef FE_INEXACT
+	int e;
 
-	fegetenv(&e);
+	e = fetestexcept(FE_INEXACT);
+#endif
 	x = rintl(x);
-	fesetenv(&e);
+#ifdef FE_INEXACT
+	if (!e)
+		feclearexcept(FE_INEXACT);
+#endif
 	return x;
 }
 #endif
-- 
cgit v1.2.1