From d8764bf84022397ff9d22310f78fcd78d801e2bf Mon Sep 17 00:00:00 2001
From: Szabolcs Nagy <nsz@port70.net>
Date: Sun, 18 Aug 2013 15:34:07 +0000
Subject: optimize x86 feclearexcept: only use save/restore x87 fenv if needed

the x87 exception summary (ES) and stack fault (SF) flags may be
spuriously cleared by feclearexcept using the fnclex instruction,
but these flags are not observable through libc hence maintaining
their state is not critical.
---
 src/fenv/x86_64/fenv.s | 26 +++++++++++---------------
 1 file changed, 11 insertions(+), 15 deletions(-)

(limited to 'src/fenv/x86_64/fenv.s')

diff --git a/src/fenv/x86_64/fenv.s b/src/fenv/x86_64/fenv.s
index c48dade3..dda6b61a 100644
--- a/src/fenv/x86_64/fenv.s
+++ b/src/fenv/x86_64/fenv.s
@@ -1,25 +1,21 @@
 .global feclearexcept
 .type feclearexcept,@function
 feclearexcept:
+		# maintain exceptions in the sse mxcsr, clear x87 exceptions
 	mov %edi,%ecx
+	fnstsw %ax
+	test %eax,%ecx
+	jz 1f
+	fnclex
+1:	stmxcsr -8(%rsp)
+	and $0x3f,%eax
+	or %eax,-8(%rsp)
+	test %ecx,-8(%rsp)
+	jz 1f
 	not %ecx
-	stmxcsr -8(%rsp)
 	and %ecx,-8(%rsp)
 	ldmxcsr -8(%rsp)
-	test $0x3f,%ecx
-	jnz 2f
-1:	fnclex
-	xor %eax,%eax
-	ret
-2:	fnstsw %ax
-	and %ecx,%eax
-	jz 1b
-	sub $32,%rsp
-	fnstenv (%rsp)
-	mov %al,4(%rsp)
-	fldenv (%rsp)
-	add $32,%rsp
-	xor %eax,%eax
+1:	xor %eax,%eax
 	ret
 
 .global feraiseexcept
-- 
cgit v1.2.1