Avoid "inexact" exceptions in i386/x86_64 ceil functions (bug 15479).

As discussed in
<https://sourceware.org/ml/libc-alpha/2016-05/msg00577.html>, TS
18661-1 disallows ceil, floor, round and trunc functions from raising
the "inexact" exception, in accordance with general IEEE 754 semantics
for when that exception is raised.  Fixing this for x87 floating point
is more complicated than for the other versions of these functions,
because they use the frndint instruction that raises "inexact" and
this can only be avoided by saving and restoring the whole
floating-point environment.

As I noted in
<https://sourceware.org/ml/libc-alpha/2016-06/msg00128.html>, I have
now implemented a GCC option -fno-fp-int-builtin-inexact for GCC 7,
such that GCC will inline these functions on x86, without caring about
"inexact", when the default -ffp-int-builtin-inexact is in effect.
This allows users to get optimized code depending on the options they
pass to the compiler, while making the out-of-line functions follow TS
18661-1 semantics and avoid "inexact".

This patch duly fixes the out-of-line ceil function implementations to
avoid "inexact", in the same way as the nearbyint implementations.

I do not know how the performance of implementations such as these
based on saving the environment and changing the rounding mode
temporarily compares to that of the C versions or SSE 4.1 versions (of
course, for 32-bit x86 SSE implementations still need to get the
return value in an x87 register); it's entirely possible other
implementations could be faster in some cases.

Tested for x86_64 and x86.

	[BZ #15479]
	* sysdeps/i386/fpu/s_ceil.S (__ceil): Save and restore
	floating-point environment rather than just control word.
	* sysdeps/i386/fpu/s_ceilf.S (__ceilf): Likewise.
	* sysdeps/i386/fpu/s_ceill.S (__ceill): Save and restore
	floating-point environment, with "invalid" exceptions merged in,
	rather than just control word.
	* sysdeps/x86_64/fpu/s_ceill.S (__ceill): Likewise.
	* math/libm-test.inc (ceil_test_data): Do not allow spurious
	"inexact" exceptions.
This commit is contained in:
Joseph Myers 2016-06-27 17:23:19 +00:00
parent b87c1ec3fa
commit 26b0bf9600
6 changed files with 115 additions and 93 deletions

View File

@ -1,3 +1,16 @@
2016-06-27 Joseph Myers <joseph@codesourcery.com>
[BZ #15479]
* sysdeps/i386/fpu/s_ceil.S (__ceil): Save and restore
floating-point environment rather than just control word.
* sysdeps/i386/fpu/s_ceilf.S (__ceilf): Likewise.
* sysdeps/i386/fpu/s_ceill.S (__ceill): Save and restore
floating-point environment, with "invalid" exceptions merged in,
rather than just control word.
* sysdeps/x86_64/fpu/s_ceill.S (__ceill): Likewise.
* math/libm-test.inc (ceil_test_data): Do not allow spurious
"inexact" exceptions.
2016-06-21 Aurelien Jarno <aurelien@aurel32.net>
* sysdeps/unix/sysv/linux/mips/vfork.S (__vfork): Rename into

View File

@ -6078,14 +6078,13 @@ static const struct test_f_f_data ceil_test_data[] =
TEST_f_f (ceil, snan_value, qnan_value, NO_INEXACT_EXCEPTION|INVALID_EXCEPTION),
TEST_f_f (ceil, -snan_value, qnan_value, NO_INEXACT_EXCEPTION|INVALID_EXCEPTION),
/* Bug 15479: spurious "inexact" exception may occur. */
TEST_f_f (ceil, lit_pi, 4.0, ERRNO_UNCHANGED),
TEST_f_f (ceil, -lit_pi, -3.0, ERRNO_UNCHANGED),
TEST_f_f (ceil, min_subnorm_value, 1.0, ERRNO_UNCHANGED),
TEST_f_f (ceil, min_value, 1.0, ERRNO_UNCHANGED),
TEST_f_f (ceil, 0.1, 1.0, ERRNO_UNCHANGED),
TEST_f_f (ceil, 0.25, 1.0, ERRNO_UNCHANGED),
TEST_f_f (ceil, 0.625, 1.0, ERRNO_UNCHANGED),
TEST_f_f (ceil, lit_pi, 4.0, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -lit_pi, -3.0, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, min_subnorm_value, 1.0, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, min_value, 1.0, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 0.1, 1.0, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 0.25, 1.0, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 0.625, 1.0, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 1, 1, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 2, 2, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 0x1p23, 0x1p23, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
@ -6104,11 +6103,11 @@ static const struct test_f_f_data ceil_test_data[] =
TEST_f_f (ceil, 0x1p113, 0x1p113, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 0x1p114, 0x1p114, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, max_value, max_value, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -min_subnorm_value, minus_zero, ERRNO_UNCHANGED),
TEST_f_f (ceil, -min_value, minus_zero, ERRNO_UNCHANGED),
TEST_f_f (ceil, -0.1, minus_zero, ERRNO_UNCHANGED),
TEST_f_f (ceil, -0.25, minus_zero, ERRNO_UNCHANGED),
TEST_f_f (ceil, -0.625, minus_zero, ERRNO_UNCHANGED),
TEST_f_f (ceil, -min_subnorm_value, minus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -min_value, minus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -0.1, minus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -0.25, minus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -0.625, minus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -1, -1, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -2, -2, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -0x1p23, -0x1p23, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
@ -6130,81 +6129,81 @@ static const struct test_f_f_data ceil_test_data[] =
#if MANT_DIG >= 64
/* The result can only be represented in long double. */
TEST_f_f (ceil, 4503599627370495.5L, 4503599627370496.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370496.25L, 4503599627370497.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370496.5L, 4503599627370497.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370496.75L, 4503599627370497.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370497.5L, 4503599627370498.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370495.5L, 4503599627370496.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370496.25L, 4503599627370497.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370496.5L, 4503599627370497.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370496.75L, 4503599627370497.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370497.5L, 4503599627370498.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370495.5L, -4503599627370495.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370496.25L, -4503599627370496.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370496.5L, -4503599627370496.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370496.75L, -4503599627370496.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370497.5L, -4503599627370497.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370495.5L, -4503599627370495.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370496.25L, -4503599627370496.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370496.5L, -4503599627370496.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370496.75L, -4503599627370496.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370497.5L, -4503599627370497.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
# if MANT_DIG > 100
TEST_f_f (ceil, 4503599627370494.5000000000001L, 4503599627370495.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370495.5000000000001L, 4503599627370496.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370496.5000000000001L, 4503599627370497.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370494.5000000000001L, -4503599627370494.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370495.5000000000001L, -4503599627370495.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370496.5000000000001L, -4503599627370496.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370494.5000000000001L, 4503599627370495.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370495.5000000000001L, 4503599627370496.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 4503599627370496.5000000000001L, 4503599627370497.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370494.5000000000001L, -4503599627370494.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370495.5000000000001L, -4503599627370495.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -4503599627370496.5000000000001L, -4503599627370496.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
# endif
TEST_f_f (ceil, 9007199254740991.5L, 9007199254740992.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.25L, 9007199254740993.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.5L, 9007199254740993.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.75L, 9007199254740993.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740993.5L, 9007199254740994.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740991.5L, 9007199254740992.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.25L, 9007199254740993.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.5L, 9007199254740993.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.75L, 9007199254740993.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740993.5L, 9007199254740994.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740991.5L, -9007199254740991.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.25L, -9007199254740992.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.5L, -9007199254740992.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.75L, -9007199254740992.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740993.5L, -9007199254740993.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740991.5L, -9007199254740991.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.25L, -9007199254740992.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.5L, -9007199254740992.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.75L, -9007199254740992.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740993.5L, -9007199254740993.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
# if MANT_DIG > 100
TEST_f_f (ceil, 9007199254740991.0000000000001L, 9007199254740992.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.0000000000001L, 9007199254740993.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740993.0000000000001L, 9007199254740994.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740991.5000000000001L, 9007199254740992.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.5000000000001L, 9007199254740993.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740993.5000000000001L, 9007199254740994.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740991.0000000000001L, 9007199254740992.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.0000000000001L, 9007199254740993.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740993.0000000000001L, 9007199254740994.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740991.5000000000001L, 9007199254740992.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740992.5000000000001L, 9007199254740993.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 9007199254740993.5000000000001L, 9007199254740994.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740991.0000000000001L, -9007199254740991.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.0000000000001L, -9007199254740992.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740993.0000000000001L, -9007199254740993.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740991.5000000000001L, -9007199254740991.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.5000000000001L, -9007199254740992.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740993.5000000000001L, -9007199254740993.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740991.0000000000001L, -9007199254740991.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.0000000000001L, -9007199254740992.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740993.0000000000001L, -9007199254740993.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740991.5000000000001L, -9007199254740991.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740992.5000000000001L, -9007199254740992.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -9007199254740993.5000000000001L, -9007199254740993.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
# endif
TEST_f_f (ceil, 72057594037927935.5L, 72057594037927936.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927936.25L, 72057594037927937.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927936.5L, 72057594037927937.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927936.75L, 72057594037927937.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927937.5L, 72057594037927938.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927935.5L, 72057594037927936.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927936.25L, 72057594037927937.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927936.5L, 72057594037927937.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927936.75L, 72057594037927937.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 72057594037927937.5L, 72057594037927938.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927935.5L, -72057594037927935.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927936.25L, -72057594037927936.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927936.5L, -72057594037927936.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927936.75L, -72057594037927936.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927937.5L, -72057594037927937.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927935.5L, -72057594037927935.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927936.25L, -72057594037927936.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927936.5L, -72057594037927936.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927936.75L, -72057594037927936.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -72057594037927937.5L, -72057594037927937.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
/* Check cases where first double is a exact integer higher than 2^52 and
the precision is determined by second long double for IBM long double. */
TEST_f_f (ceil, 34503599627370498.515625L, 34503599627370499.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -34503599627370498.515625L, -34503599627370498.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 34503599627370498.515625L, 34503599627370499.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -34503599627370498.515625L, -34503599627370498.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
# if MANT_DIG >= 106
TEST_f_f (ceil, 1192568192774434123539907640624.484375L, 1192568192774434123539907640625.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, -1192568192774434123539907640624.484375L, -1192568192774434123539907640624.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 1192568192774434123539907640624.484375L, 1192568192774434123539907640625.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, -1192568192774434123539907640624.484375L, -1192568192774434123539907640624.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
# endif
TEST_f_f (ceil, 10141204801825835211973625643007.5L, 10141204801825835211973625643008.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643008.25L, 10141204801825835211973625643009.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643008.5L, 10141204801825835211973625643009.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643008.75L, 10141204801825835211973625643009.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643009.5L, 10141204801825835211973625643010.0L, ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643007.5L, 10141204801825835211973625643008.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643008.25L, 10141204801825835211973625643009.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643008.5L, 10141204801825835211973625643009.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643008.75L, 10141204801825835211973625643009.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
TEST_f_f (ceil, 10141204801825835211973625643009.5L, 10141204801825835211973625643010.0L, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
#endif
};

View File

@ -9,10 +9,10 @@ RCSID("$NetBSD: s_ceil.S,v 1.4 1995/05/08 23:52:13 jtc Exp $")
ENTRY(__ceil)
fldl 4(%esp)
subl $8,%esp
cfi_adjust_cfa_offset (8)
subl $32,%esp
cfi_adjust_cfa_offset (32)
fstcw 4(%esp) /* store fpu control word */
fnstenv 4(%esp) /* store fpu environment */
/* We use here %edx although only the low 1 bits are defined.
But none of the operations should care and they are faster
@ -25,10 +25,10 @@ ENTRY(__ceil)
frndint /* round */
fldcw 4(%esp) /* restore original control word */
fldenv 4(%esp) /* restore original environment */
addl $8,%esp
cfi_adjust_cfa_offset (-8)
addl $32,%esp
cfi_adjust_cfa_offset (-32)
ret
END (__ceil)
weak_alias (__ceil, ceil)

View File

@ -9,10 +9,10 @@ RCSID("$NetBSD: s_ceilf.S,v 1.3 1995/05/08 23:52:44 jtc Exp $")
ENTRY(__ceilf)
flds 4(%esp)
subl $8,%esp
cfi_adjust_cfa_offset (8)
subl $32,%esp
cfi_adjust_cfa_offset (32)
fstcw 4(%esp) /* store fpu control word */
fnstenv 4(%esp) /* store fpu environment */
/* We use here %edx although only the low 1 bits are defined.
But none of the operations should care and they are faster
@ -25,10 +25,10 @@ ENTRY(__ceilf)
frndint /* round */
fldcw 4(%esp) /* restore original control word */
fldenv 4(%esp) /* restore original environment */
addl $8,%esp
cfi_adjust_cfa_offset (-8)
addl $32,%esp
cfi_adjust_cfa_offset (-32)
ret
END (__ceilf)
weak_alias (__ceilf, ceilf)

View File

@ -10,10 +10,10 @@ RCSID("$NetBSD: $")
ENTRY(__ceill)
fldt 4(%esp)
subl $8,%esp
cfi_adjust_cfa_offset (8)
subl $32,%esp
cfi_adjust_cfa_offset (32)
fstcw 4(%esp) /* store fpu control word */
fnstenv 4(%esp) /* store fpu environment */
/* We use here %edx although only the low 1 bits are defined.
But none of the operations should care and they are faster
@ -26,10 +26,15 @@ ENTRY(__ceill)
frndint /* round */
fldcw 4(%esp) /* restore original control word */
/* Preserve "invalid" exceptions from sNaN input. */
fnstsw
andl $0x1, %eax
orl %eax, 8(%esp)
addl $8,%esp
cfi_adjust_cfa_offset (-8)
fldenv 4(%esp) /* restore original environment */
addl $32,%esp
cfi_adjust_cfa_offset (-32)
ret
END (__ceill)
weak_alias (__ceill, ceill)

View File

@ -11,20 +11,25 @@
ENTRY(__ceill)
fldt 8(%rsp)
fstcw -4(%rsp) /* store fpu control word */
fnstenv -28(%rsp) /* store fpu environment */
/* We use here %edx although only the low 1 bits are defined.
But none of the operations should care and they are faster
than the 16 bit operations. */
movl $0x0800,%edx /* round towards +oo */
orl -4(%rsp),%edx
orl -28(%rsp),%edx
andl $0xfbff,%edx
movl %edx,-8(%rsp)
fldcw -8(%rsp) /* load modified control word */
movl %edx,-32(%rsp)
fldcw -32(%rsp) /* load modified control word */
frndint /* round */
fldcw -4(%rsp) /* restore original control word */
/* Preserve "invalid" exceptions from sNaN input. */
fnstsw
andl $0x1, %eax
orl %eax, -24(%rsp)
fldenv -28(%rsp) /* restore original environment */
ret
END (__ceill)