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.

1 files changed, 27 insertions, 12 deletions

diff --git a/src/fenv/i386/fenv.s b/src/fenv/i386/fenv.s
index 9bba40a5..a8540add 100644
--- a/src/fenv/i386/fenv.s
+++ b/src/fenv/i386/fenv.s
@@ -4,26 +4,41 @@
 .type feclearexcept,@function
 feclearexcept:	
 	mov 4(%esp),%ecx
-	not %ecx
+	fnstsw %ax
 		# consider sse fenv as well if the cpu has XMM capability
 	call 1f
 1:	addl $__hwcap-1b,(%esp)
 	pop %edx
 	testl $0x02000000,(%edx)
+	jz 2f
+		# maintain exceptions in the sse mxcsr, clear x87 exceptions
+	test %eax,%ecx
 	jz 1f
-	stmxcsr 4(%esp)
-	and %ecx,4(%esp)
-	ldmxcsr 4(%esp)
-1:	test $0x3f,%ecx
-	jnz 2f
-1:	fnclex
-	xor %eax,%eax
+	fnclex
+1:	push %edx
+	stmxcsr (%esp)
+	pop %edx
+	and $0x3f,%eax
+	or %eax,%edx
+	test %edx,%ecx
+	jz 1f
+	not %ecx
+	and %ecx,%edx
+	push %edx
+	ldmxcsr (%esp)
+	pop %edx
+1:	xor %eax,%eax
 	ret
-2:	fnstsw %ax
-		# TODO: only load/store fenv if exceptions arent clear yet
-	and %ecx,%eax
+		# only do the expensive x87 fenv load/store when needed
+2:	test %eax,%ecx
 	jz 1b
-	sub $32,%esp
+	not %ecx
+	and %ecx,%eax
+	test $0x3f,%eax
+	jz 1f
+	fnclex
+	jmp 1b
+1:	sub $32,%esp
 	fnstenv (%esp)
 	mov %al,4(%esp)
 	fldenv (%esp)