AuroraMiddleware/glibc
1
0
Fork 0
mirror of https://sourceware.org/git/glibc.git synced 2024-12-22 02:40:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix x86 / x86_64 expl / expl10l wild results in directed rounding modes (bug 16356).

Browse Source 
This patch fixes bug 16356, bad results from x86 / x86_64 expl /
exp10l in directed rounding modes, the most serious of the bugs shown
up by my patch expanding libm test coverage.  When I fixed bug 16293,
I thought it was only necessary to set round-to-nearest when using
frndint in expm1 functions, because in other cases the cancellation
error from having the resulting fractional part close to 1 or -1 would
not be significant.  However, in expl and exp10l, the way the final
fractional part gets computed (something more complicated than a
simple subtraction, because more precision is needed than you'd get
that way) can result in a value outside the range [-1, 1] when the
argument to frndint was very close to an integer and was rounded the
"wrong" way because of the rounding mode - and the f2xm1 instruction
has undefined results if its argument is outside [-1, 1], so resulting
in the large errors seen.  So this patch removes the USE_AS_EXPM1L
conditionals on the round-to-nearest settings, so all of expl, expm1l
and exp10l now get round-to-nearest used for frndint (meaning the
final fractional part can at most be slightly above 0.5 in
magnitude).  Associated tests of exp and exp10 are added and testing
of exp10 in directed rounding modes enabled.

Tested x86_64 and x86 and ulps updated accordingly.

	* sysdeps/i386/fpu/e_expl.S (IEEE754_EXPL): Also set
	round-to-nearest for [!USE_AS_EXPM1L].
	* sysdeps/x86_64/fpu/e_expl.S (IEEE754_EXPL): Likewise.
	* math/auto-libm-test-in: Do not expect cosh tests to fail.  Add
	more tests of exp and exp10.  Expect some exp10 tests to miss
	exceptions or fail in directed rounding modes.
	* math/auto-libm-test-out: Regenerated.
	* math/libm-test.inc (exp10_tonearest_test_data): New array.
	(exp10_test_tonearest): New function.
	(exp10_towardzero_test_data): New array.
	(exp10_test_towardzero): New function.
	(exp10_downward_test_data): New array.
	(exp10_test_downward): New function.
	(exp10_upward_test_data): New array.
	(exp10_test_upward): New function.
	(main): Call the new functions.
	* sysdeps/i386/fpu/libm-test-ulps: Update.
	* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
This commit is contained in:
Joseph Myers 2013-12-21 13:07:16 +00:00
parent 31e3a40588
commit 5b0626b9c5
9 changed files with 1059 additions and 550 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
ChangeLog
View File

@ -1,3 +1,25 @@
2013-12-21 Joseph Myers <joseph@codesourcery.com>
[BZ #16356]
* sysdeps/i386/fpu/e_expl.S (IEEE754_EXPL): Also set
round-to-nearest for [!USE_AS_EXPM1L].
* sysdeps/x86_64/fpu/e_expl.S (IEEE754_EXPL): Likewise.
* math/auto-libm-test-in: Do not expect cosh tests to fail. Add
more tests of exp and exp10. Expect some exp10 tests to miss
exceptions or fail in directed rounding modes.
* math/auto-libm-test-out: Regenerated.
* math/libm-test.inc (exp10_tonearest_test_data): New array.
(exp10_test_tonearest): New function.
(exp10_towardzero_test_data): New array.
(exp10_test_towardzero): New function.
(exp10_downward_test_data): New array.
(exp10_test_downward): New function.
(exp10_upward_test_data): New array.
(exp10_test_upward): New function.
(main): Call the new functions.
* sysdeps/i386/fpu/libm-test-ulps: Update.
* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
2013-12-20 Joseph Myers <joseph@codesourcery.com>
* math/auto-libm-test-in: Add more tests of acos, acosh, asin,

2
NEWS
View File

@ -22,7 +22,7 @@ Version 2.19
15966, 15985, 15988, 15997, 16032, 16034, 16036, 16037, 16038, 16041,
16055, 16071, 16072, 16074, 16077, 16078, 16103, 16112, 16143, 16144,
16146, 16150, 16151, 16153, 16167, 16172, 16195, 16214, 16245, 16271,
16274, 16283, 16289, 16293, 16314, 16316, 16330, 16338.
16274, 16283, 16289, 16293, 16314, 16316, 16330, 16338, 16356.
* The public headers no longer use __unused nor __block. This change is to
support compiling programs that are derived from BSD sources and use

27
math/auto-libm-test-in
View File

@ -519,11 +519,10 @@ cosh 0x2.c679d1f73f0fap+8
cosh 0x2.c679d1f73f0fcp+8
cosh -0x2.c679d1f73f0fap+8
cosh -0x2.c679d1f73f0fcp+8
# Bug 16356: bad results from expl (and so coshl) in round-upward mode.
cosh 0x2.c679d1f73f0fb624d358b213a7p+8 xfail-rounding:ldbl-96-intel:x86 xfail-rounding:ldbl-96-intel:x86_64
cosh 0x2.c679d1f73f0fb624d358b213a8p+8 xfail-rounding:ldbl-96-intel:x86 xfail-rounding:ldbl-96-intel:x86_64
cosh -0x2.c679d1f73f0fb624d358b213a7p+8 xfail-rounding:ldbl-96-intel:x86 xfail-rounding:ldbl-96-intel:x86_64
cosh -0x2.c679d1f73f0fb624d358b213a8p+8 xfail-rounding:ldbl-96-intel:x86 xfail-rounding:ldbl-96-intel:x86_64
cosh 0x2.c679d1f73f0fb624d358b213a7p+8
cosh 0x2.c679d1f73f0fb624d358b213a8p+8
cosh -0x2.c679d1f73f0fb624d358b213a7p+8
cosh -0x2.c679d1f73f0fb624d358b213a8p+8
cosh 0x2.c5d37700c6bb03a4p+12 no-test-inline
cosh 0x2.c5d37700c6bb03a8p+12 no-test-inline
cosh -0x2.c5d37700c6bb03a4p+12 no-test-inline
@ -712,6 +711,7 @@ exp 1000.0 xfail-rounding:dbl-64
exp 710 xfail-rounding:dbl-64
exp -1234
# Bug 16284: results on directed rounding may be incorrect.
exp 0x2.c679d1f73f0fb628p+8 xfail-rounding:dbl-64
exp 1e5 xfail-rounding:dbl-64
exp max xfail-rounding:dbl-64
exp -7.4444006192138124e+02
@ -726,15 +726,22 @@ exp10 36
exp10 -36
exp10 305
exp10 -305
exp10 4932
exp10 -4932
exp10 1e5
# Bug 16284: results on directed rounding may be incorrect.
exp10 4932 xfail-rounding:flt-32
# Bug 16361: underflow exception may be misssing
exp10 -4932 missing-underflow:ldbl-96-intel:x86 missing-underflow:ldbl-96-intel:x86_64
# Bug 16284: results on directed rounding may be incorrect.
exp10 1e5 xfail-rounding:flt-32
exp10 -1e5
exp10 1e6
# Bug 16284: results on directed rounding may be incorrect.
exp10 1e6 xfail-rounding:flt-32
exp10 -1e6
exp10 max
# Bug 16284: results on directed rounding may be incorrect.
exp10 max xfail-rounding:flt-32
exp10 -max
exp10 0.75
# Bug 16284: results on directed rounding may be incorrect.
exp10 0x1.348e45573a1dd72cp+8 xfail-rounding:flt-32 xfail-rounding:dbl-64
exp2 0
exp2 -0

1254
math/auto-libm-test-out
View File

File diff suppressed because it is too large Load Diff

60
math/libm-test.inc
View File

@ -7280,6 +7280,62 @@ exp10_test (void)
END;
}
static const struct test_f_f_data exp10_tonearest_test_data[] =
{
AUTO_TESTS_f_f (exp10, tonearest),
};
static void
exp10_test_tonearest (void)
{
START (exp10_tonearest);
RUN_TEST_LOOP_f_f (exp10, exp10_tonearest_test_data, FE_TONEAREST);
END;
}
static const struct test_f_f_data exp10_towardzero_test_data[] =
{
AUTO_TESTS_f_f (exp10, towardzero),
};
static void
exp10_test_towardzero (void)
{
START (exp10_towardzero);
RUN_TEST_LOOP_f_f (exp10, exp10_towardzero_test_data, FE_TOWARDZERO);
END;
}
static const struct test_f_f_data exp10_downward_test_data[] =
{
AUTO_TESTS_f_f (exp10, downward),
};
static void
exp10_test_downward (void)
{
START (exp10_downward);
RUN_TEST_LOOP_f_f (exp10, exp10_downward_test_data, FE_DOWNWARD);
END;
}
static const struct test_f_f_data exp10_upward_test_data[] =
{
AUTO_TESTS_f_f (exp10, upward),
};
static void
exp10_test_upward (void)
{
START (exp10_upward);
RUN_TEST_LOOP_f_f (exp10, exp10_upward_test_data, FE_UPWARD);
END;
}
static void
pow10_test (void)
{
@ -12819,6 +12875,10 @@ main (int argc, char **argv)
exp_test_downward ();
exp_test_upward ();
exp10_test ();
exp10_test_tonearest ();
exp10_test_towardzero ();
exp10_test_downward ();
exp10_test_upward ();
exp2_test ();
expm1_test ();
expm1_test_tonearest ();

4
sysdeps/i386/fpu/e_expl.S
View File

@ -130,7 +130,6 @@ ENTRY(IEEE754_EXPL)
#endif
3: FLDLOG /* 1 log2(base) */
fmul %st(1), %st /* 1 x log2(base) */
#ifdef USE_AS_EXPM1L
/* Set round-to-nearest temporarily. */
subl $8, %esp
cfi_adjust_cfa_offset (8)
@ -139,15 +138,12 @@ ENTRY(IEEE754_EXPL)
andl 4(%esp), %edx
movl %edx, (%esp)
fldcw (%esp)
#endif
frndint /* 1 i */
fld %st(1) /* 2 x */
frndint /* 2 xi */
#ifdef USE_AS_EXPM1L
fldcw 4(%esp)
addl $8, %esp
cfi_adjust_cfa_offset (-8)
#endif
fld %st(1) /* 3 i */
fldt MO(c0) /* 4 c0 */
fld %st(2) /* 5 xi */

114
sysdeps/i386/fpu/libm-test-ulps
View File

@ -7095,6 +7095,10 @@ Test "cosh_downward (-0x2.c5e3bp+8)":
ildouble: 1
Test "cosh_downward (-0x2.c679d1f73f0fap+8)":
ildouble: 2
Test "cosh_downward (-0x2.c679d1f73f0fb624p+8)":
ildouble: 2
Test "cosh_downward (-0x2.c679d1f73f0fb628p+8)":
ildouble: 2
Test "cosh_downward (-0x2.c679d1f73f0fcp+8)":
ildouble: 1
Test "cosh_downward (-0x2.c679dp+8)":
@ -7181,15 +7185,25 @@ Test "cosh_towardzero (-0x2.c5d37700c6bbp+12)":
ldouble: 2
Test "cosh_towardzero (-0x2.c5e3acd2922a6p+8)":
ildouble: 1
Test "cosh_towardzero (-0x2.c679d1f73f0fap+8)":
ildouble: 2
Test "cosh_towardzero (-0x2.c679d1f73f0fb624p+8)":
ildouble: 2
Test "cosh_towardzero (-0x2.c679d1f73f0fb628p+8)":
ildouble: 2
Test "cosh_towardzero (-0x2.c679dp+8)":
double: 1
ildouble: 1
Test "cosh_towardzero (-0x5.96a7ep+4)":
double: 1
ildouble: 2
ldouble: 1
Test "cosh_towardzero (0x1.6p+4)":
ildouble: 1
ldouble: 2
Test "cosh_towardzero (0x1.8p+4)":
ildouble: 1
ldouble: 1
Test "cosh_towardzero (0x2.c5d374p+12)":
ldouble: 1
Test "cosh_towardzero (0x2.c5d37700c6bb03a4p+12)":
@ -7240,6 +7254,11 @@ ildouble: 1
Test "cosh_upward (-0x2.c679d1f73f0fap+8)":
double: 1
ildouble: 1
Test "cosh_upward (-0x2.c679d1f73f0fb624p+8)":
ildouble: 1
Test "cosh_upward (-0x2.c679d1f73f0fb628p+8)":
ildouble: 1
ldouble: 1
Test "cosh_upward (-0x2.c679d1f73f0fcp+8)":
ildouble: 1
Test "cosh_upward (-0x2p-16384)":
@ -7292,6 +7311,11 @@ ildouble: 1
Test "cosh_upward (0x2.c679d1f73f0fap+8)":
double: 1
ildouble: 1
Test "cosh_upward (0x2.c679d1f73f0fb624p+8)":
ildouble: 1
Test "cosh_upward (0x2.c679d1f73f0fb628p+8)":
ildouble: 2
ldouble: 1
Test "cosh_upward (0x2.c679d1f73f0fcp+8)":
ildouble: 1
Test "cosh_upward (0x2.c679d4p+8)":
@ -8323,11 +8347,13 @@ Test "Real part of: ctan_towardzero (0x1p+0 + 0x1.63p+8 i)":
ildouble: 1
ldouble: 1
Test "Real part of: ctan_towardzero (0x1p+0 + 0x1.6dp+8 i)":
ildouble: 1
ldouble: 1
ildouble: 2
ldouble: 2
Test "Real part of: ctan_towardzero (0x1p+0 + 0x2.dp+4 i)":
double: 1
idouble: 1
ildouble: 1
ldouble: 1
Test "Real part of: ctan_towardzero (0x1p+0 + 0x2.fp+4 i)":
double: 1
idouble: 1
@ -8361,6 +8387,9 @@ double: 1
idouble: 1
ildouble: 1
ldouble: 1
Test "Imaginary part of: ctan_towardzero (0xcp-4 + 0x1.4p+0 i)":
ildouble: 2
ldouble: 2
Test "Real part of: ctan_towardzero (0xf.ffffffffffff8p+1020 + 0x1p+0 i)":
double: 1
idouble: 1
@ -9188,8 +9217,8 @@ Test "Imaginary part of: ctanh_towardzero (0x1.63p+8 + 0x1p+0 i)":
ildouble: 1
ldouble: 1
Test "Imaginary part of: ctanh_towardzero (0x1.6dp+8 + 0x1p+0 i)":
ildouble: 1
ldouble: 1
ildouble: 2
ldouble: 2
Test "Imaginary part of: ctanh_towardzero (0x1p+0 + 0x8p+1020 i)":
double: 1
idouble: 1
@ -9235,6 +9264,8 @@ ldouble: 1
Test "Imaginary part of: ctanh_towardzero (0x2.dp+4 + 0x1p+0 i)":
double: 1
idouble: 1
ildouble: 1
ldouble: 1
Test "Imaginary part of: ctanh_towardzero (0x2.fp+4 + 0x1p+0 i)":
double: 1
idouble: 1
@ -9697,6 +9728,48 @@ Test "exp10 (0xcp-4)":
ildouble: 1
ldouble: 1
# exp10_downward
Test "exp10_downward (0x1.348e45573a1dd72cp+8)":
ildouble: 2
ldouble: 2
Test "exp10_downward (0x3p+0)":
double: 1
float: 1
idouble: 1
ifloat: 1
ildouble: 1
ldouble: 1
# exp10_tonearest
Test "exp10_tonearest (0xcp-4)":
ildouble: 1
ldouble: 1
# exp10_towardzero
Test "exp10_towardzero (-0x1p+0)":
ildouble: 1
ldouble: 1
Test "exp10_towardzero (0x1.344p+12)":
ildouble: 1
ldouble: 1
Test "exp10_towardzero (0x3p+0)":
double: 1
float: 1
idouble: 1
ifloat: 1
# exp10_upward
Test "exp10_upward (0x1.344p+12)":
ildouble: 1
ldouble: 1
Test "exp10_upward (0x3p+0)":
double: 1
float: 1
idouble: 1
ifloat: 1
ildouble: 1
ldouble: 1
# exp_downward
Test "exp_downward (0x2.c5cp+8)":
double: 1
@ -9788,6 +9861,9 @@ ldouble: 1
Test "exp_upward (-0x2.e870a7e5e88cp+8)":
ildouble: 1
ldouble: 1
Test "exp_upward (0x2.c679d1f73f0fb628p+8)":
ildouble: 1
ldouble: 1
Test "exp_upward (1)":
double: 1
float: 1
@ -14500,7 +14576,7 @@ double: 1
float: 1
idouble: 1
ifloat: 1
ildouble: 1
ildouble: 2
ldouble: 3
Function: "cosh_upward":
@ -14740,6 +14816,34 @@ Function: "exp10":
ildouble: 1
ldouble: 1
Function: "exp10_downward":
double: 1
float: 1
idouble: 1
ifloat: 1
ildouble: 2
ldouble: 2
Function: "exp10_tonearest":
ildouble: 1
ldouble: 1
Function: "exp10_towardzero":
double: 1
float: 1
idouble: 1
ifloat: 1
ildouble: 1
ldouble: 1
Function: "exp10_upward":
double: 1
float: 1
idouble: 1
ifloat: 1
ildouble: 1
ldouble: 1
Function: "exp_downward":
double: 1
float: 1

4
sysdeps/x86_64/fpu/e_expl.S
View File

@ -127,20 +127,16 @@ ENTRY(IEEE754_EXPL)
#endif
3: FLDLOG /* 1 log2(base) */
fmul %st(1), %st /* 1 x log2(base) */
#ifdef USE_AS_EXPM1L
/* Set round-to-nearest temporarily. */
fstcw -4(%rsp)
movl $0xf3ff, %edx
andl -4(%rsp), %edx
movl %edx, -8(%rsp)
fldcw -8(%rsp)
#endif
frndint /* 1 i */
fld %st(1) /* 2 x */
frndint /* 2 xi */
#ifdef USE_AS_EXPM1L
fldcw -4(%rsp)
#endif
fld %st(1) /* 3 i */
fldt MO(c0) /* 4 c0 */
fld %st(2) /* 5 xi */

122
sysdeps/x86_64/fpu/libm-test-ulps
View File

@ -8125,6 +8125,9 @@ ldouble: 2
Test "cosh_towardzero (0x1.7p+4)":
double: 1
idouble: 1
Test "cosh_towardzero (0x1.8p+4)":
ildouble: 1
ldouble: 1
Test "cosh_towardzero (0x2.c5d374p+12)":
ldouble: 1
Test "cosh_towardzero (0x2.c5d37700c6bb03a4p+12)":
@ -8180,6 +8183,9 @@ idouble: 1
Test "cosh_upward (-0x2.c679d1f73f0fap+8)":
double: 1
idouble: 1
Test "cosh_upward (-0x2.c679d1f73f0fb628p+8)":
ildouble: 1
ldouble: 1
Test "cosh_upward (-0x2.c679dp+8)":
double: 1
idouble: 1
@ -8215,6 +8221,9 @@ idouble: 1
Test "cosh_upward (0x2.c679d1f73f0fap+8)":
double: 1
idouble: 1
Test "cosh_upward (0x2.c679d1f73f0fb628p+8)":
ildouble: 1
ldouble: 1
Test "cosh_upward (0x2.c679dp+8)":
double: 1
idouble: 1
@ -9312,11 +9321,13 @@ Test "Real part of: ctan_towardzero (0x1p+0 + 0x1.63p+8 i)":
ildouble: 1
ldouble: 1
Test "Real part of: ctan_towardzero (0x1p+0 + 0x1.6dp+8 i)":
ildouble: 1
ldouble: 1
ildouble: 2
ldouble: 2
Test "Real part of: ctan_towardzero (0x1p+0 + 0x2.dp+4 i)":
double: 1
idouble: 1
ildouble: 1
ldouble: 1
Test "Real part of: ctan_towardzero (0x1p+0 + 0x2.fp+4 i)":
ildouble: 1
ldouble: 1
@ -9352,6 +9363,9 @@ float: 1
ifloat: 1
ildouble: 1
ldouble: 1
Test "Imaginary part of: ctan_towardzero (0xcp-4 + 0x1.4p+0 i)":
ildouble: 2
ldouble: 2
Test "Real part of: ctan_towardzero (0xf.ffffffffffff8p+1020 + 0x1p+0 i)":
double: 2
idouble: 2
@ -10342,8 +10356,8 @@ Test "Imaginary part of: ctanh_towardzero (0x1.63p+8 + 0x1p+0 i)":
ildouble: 1
ldouble: 1
Test "Imaginary part of: ctanh_towardzero (0x1.6dp+8 + 0x1p+0 i)":
ildouble: 1
ldouble: 1
ildouble: 2
ldouble: 2
Test "Imaginary part of: ctanh_towardzero (0x1p+0 + 0x8p+1020 i)":
double: 5
idouble: 5
@ -10396,6 +10410,8 @@ ldouble: 1
Test "Imaginary part of: ctanh_towardzero (0x2.dp+4 + 0x1p+0 i)":
double: 1
idouble: 1
ildouble: 1
ldouble: 1
Test "Imaginary part of: ctanh_towardzero (0x2.fp+4 + 0x1p+0 i)":
ildouble: 1
ldouble: 1
@ -10921,6 +10937,75 @@ Test "exp10 (36)":
double: 1
idouble: 1
# exp10_downward
Test "exp10_downward (0x1.348e45573a1dd72cp+8)":
ildouble: 2
ldouble: 2
Test "exp10_downward (0x2.4p+4)":
double: 1
idouble: 1
Test "exp10_downward (0x3p+0)":
ildouble: 1
ldouble: 1
# exp10_tonearest
Test "exp10_tonearest (-0x1.31p+8)":
double: 1
idouble: 1
Test "exp10_tonearest (-0x1p+0)":
double: 1
idouble: 1
Test "exp10_tonearest (-0x2.4p+4)":
double: 1
idouble: 1
Test "exp10_tonearest (0x2.4p+4)":
double: 1
idouble: 1
Test "exp10_tonearest (0x3p+0)":
double: 1
idouble: 1
Test "exp10_tonearest (0xcp-4)":
ildouble: 1
ldouble: 1
# exp10_towardzero
Test "exp10_towardzero (-0x1p+0)":
ildouble: 1
ldouble: 1
Test "exp10_towardzero (0x1.344p+12)":
ildouble: 1
ldouble: 1
Test "exp10_towardzero (0x2.4p+4)":
double: 1
idouble: 1
# exp10_upward
Test "exp10_upward (-0x1.344p+12)":
float: 1
ifloat: 1
Test "exp10_upward (-0x1.86ap+16)":
float: 1
ifloat: 1
Test "exp10_upward (-0xf.424p+16)":
float: 1
ifloat: 1
Test "exp10_upward (-0xf.fffffp+124)":
float: 1
ifloat: 1
Test "exp10_upward (0x1.344p+12)":
ildouble: 1
ldouble: 1
Test "exp10_upward (0x2.4p+4)":
double: 1
idouble: 1
Test "exp10_upward (0x3p+0)":
double: 1
float: 1
idouble: 1
ifloat: 1
ildouble: 1
ldouble: 1
# exp_downward
Test "exp_downward (0x2.c5cp+8)":
ildouble: 1
@ -11057,6 +11142,9 @@ idouble: 1
Test "exp_upward (0x2.c5cp+8)":
double: 1
idouble: 1
Test "exp_upward (0x2.c679d1f73f0fb628p+8)":
ildouble: 1
ldouble: 1
Test "exp_upward (0x3.2p+4)":
double: 1
idouble: 1
@ -16375,6 +16463,32 @@ idouble: 1
ildouble: 1
ldouble: 1
Function: "exp10_downward":
double: 1
idouble: 1
ildouble: 2
ldouble: 2
Function: "exp10_tonearest":
double: 1
idouble: 1
ildouble: 1
ldouble: 1
Function: "exp10_towardzero":
double: 1
idouble: 1
ildouble: 1
ldouble: 1
Function: "exp10_upward":
double: 1
float: 1
idouble: 1
ifloat: 1
ildouble: 1
ldouble: 1
Function: "exp_downward":
double: 1
float: 1