From 02db27d9deaee71b244c91e720ec819c74dab150 Mon Sep 17 00:00:00 2001
From: Rich Felker <dalias@aerifal.cx>
Date: Mon, 19 Mar 2012 09:00:30 -0400
Subject: optimize exponential asm for i386

up to 30% faster exp2 by avoiding slow frndint and fscale functions.
expm1 also takes a much more direct path for small arguments (the
expected usage case).
---
 src/math/i386/expm1.s | 48 +-----------------------------------------------
 1 file changed, 1 insertion(+), 47 deletions(-)

(limited to 'src/math/i386/expm1.s')

diff --git a/src/math/i386/expm1.s b/src/math/i386/expm1.s
index bbb5d12e..f335a3e5 100644
--- a/src/math/i386/expm1.s
+++ b/src/math/i386/expm1.s
@@ -1,47 +1 @@
-.global expm1f
-.type expm1f,@function
-expm1f:
-	flds 4(%esp)
-	jmp 1f
-
-.global expm1l
-.type expm1l,@function
-expm1l:
-	fldt 4(%esp)
-	jmp 1f
-
-.global expm1
-.type expm1,@function
-expm1:
-	fldl 4(%esp)
-1:	fxam
-	fnstsw %ax
-	sahf
-	jnp 1f
-	jnc 1f
-	fstps 4(%esp)
-	mov $0xfe,%al
-	and %al,7(%esp)
-	flds 4(%esp)
-1:	fldl2e
-	fmulp
-	fld %st(0)
-	frndint
-	fldz
-	fcomp
-	fnstsw %ax
-	sahf
-	jnz 1f
-	fstp %st(0)
-	f2xm1
-	ret
-1:	fxch %st(1)
-	fsub %st(1)
-	f2xm1
-	fld1
-	faddp
-	fscale
-	fld1
-	fsubrp
-	fstp %st(1)
-	ret
+# see exp.s
-- 
cgit v1.2.1