From 1671fbe1ca0825a77cb9137a08cf4ad8a48d0f95 Mon Sep 17 00:00:00 2001 From: Jeff Johnston Date: Fri, 28 Sep 2007 18:58:10 +0000 Subject: [PATCH] 2007-09-28 Patrick Mansfield * libm/machine/spu/headers/dom_chkd_less_than.h: New file. * libm/machine/spu/headers/dom_chkd_negone_one.h: Ditto. * libm/machine/spu/headers/dom_chkf_less_than.h: Ditto. * libm/machine/spu/headers/dom_chkf_negone_one.h: Ditto. * libm/machine/spu/headers/acos.h: Add domain check. * libm/machine/spu/headers/acosf.h: Ditto. * libm/machine/spu/headers/acosh.h: Ditto. * libm/machine/spu/headers/acoshf.h: Ditto. * libm/machine/spu/headers/asin.h: Ditto. * libm/machine/spu/headers/asinf.h: Ditto. * libm/machine/spu/headers/atanh.h: Ditto. * libm/machine/spu/headers/atanhf.h: Ditto. * libm/machine/spu/headers/fmod.h: Ditto. * libm/machine/spu/headers/fmodf.h: Ditto. * libm/machine/spu/headers/log.h: Ditto. * libm/machine/spu/headers/log10.h: Ditto. * libm/machine/spu/headers/log1p.h: Ditto. * libm/machine/spu/headers/log1pf.h: Ditto. * libm/machine/spu/headers/log2.h: Ditto. * libm/machine/spu/headers/log2f.h: Ditto. * libm/machine/spu/headers/sqrt.h: Ditto. * libm/machine/spu/headers/sqrtf.h: Ditto. * libm/machine/spu/headers/tgamma.h: Ditto. * libm/machine/spu/headers/tgammaf.h: Ditto. --- newlib/ChangeLog | 27 ++++++++ newlib/libm/machine/spu/headers/acos.h | 14 +++- newlib/libm/machine/spu/headers/acosf.h | 14 +++- newlib/libm/machine/spu/headers/acosh.h | 12 +++- newlib/libm/machine/spu/headers/acoshf.h | 12 +++- newlib/libm/machine/spu/headers/asin.h | 14 +++- newlib/libm/machine/spu/headers/asinf.h | 14 +++- newlib/libm/machine/spu/headers/atanh.h | 14 +++- newlib/libm/machine/spu/headers/atanhf.h | 14 +++- .../machine/spu/headers/dom_chkd_less_than.h | 59 ++++++++++++++++ .../machine/spu/headers/dom_chkd_negone_one.h | 69 +++++++++++++++++++ .../machine/spu/headers/dom_chkf_less_than.h | 59 ++++++++++++++++ .../machine/spu/headers/dom_chkf_negone_one.h | 69 +++++++++++++++++++ newlib/libm/machine/spu/headers/fmod.h | 14 ++++ newlib/libm/machine/spu/headers/fmodf.h | 14 ++++ newlib/libm/machine/spu/headers/log.h | 12 +++- newlib/libm/machine/spu/headers/log10.h | 12 +++- newlib/libm/machine/spu/headers/log1p.h | 12 +++- newlib/libm/machine/spu/headers/log1pf.h | 12 +++- newlib/libm/machine/spu/headers/log2.h | 14 +++- newlib/libm/machine/spu/headers/log2f.h | 9 +++ newlib/libm/machine/spu/headers/sqrt.h | 5 ++ newlib/libm/machine/spu/headers/sqrtf.h | 12 +++- newlib/libm/machine/spu/headers/tgamma.h | 26 ++++++- newlib/libm/machine/spu/headers/tgammaf.h | 25 ++++++- 25 files changed, 541 insertions(+), 17 deletions(-) create mode 100644 newlib/libm/machine/spu/headers/dom_chkd_less_than.h create mode 100644 newlib/libm/machine/spu/headers/dom_chkd_negone_one.h create mode 100644 newlib/libm/machine/spu/headers/dom_chkf_less_than.h create mode 100644 newlib/libm/machine/spu/headers/dom_chkf_negone_one.h diff --git a/newlib/ChangeLog b/newlib/ChangeLog index 1bcaec5b5..721cbc3a4 100644 --- a/newlib/ChangeLog +++ b/newlib/ChangeLog @@ -1,3 +1,30 @@ +2007-09-28 Patrick Mansfield + + * libm/machine/spu/headers/dom_chkd_less_than.h: New file. + * libm/machine/spu/headers/dom_chkd_negone_one.h: Ditto. + * libm/machine/spu/headers/dom_chkf_less_than.h: Ditto. + * libm/machine/spu/headers/dom_chkf_negone_one.h: Ditto. + * libm/machine/spu/headers/acos.h: Add domain check. + * libm/machine/spu/headers/acosf.h: Ditto. + * libm/machine/spu/headers/acosh.h: Ditto. + * libm/machine/spu/headers/acoshf.h: Ditto. + * libm/machine/spu/headers/asin.h: Ditto. + * libm/machine/spu/headers/asinf.h: Ditto. + * libm/machine/spu/headers/atanh.h: Ditto. + * libm/machine/spu/headers/atanhf.h: Ditto. + * libm/machine/spu/headers/fmod.h: Ditto. + * libm/machine/spu/headers/fmodf.h: Ditto. + * libm/machine/spu/headers/log.h: Ditto. + * libm/machine/spu/headers/log10.h: Ditto. + * libm/machine/spu/headers/log1p.h: Ditto. + * libm/machine/spu/headers/log1pf.h: Ditto. + * libm/machine/spu/headers/log2.h: Ditto. + * libm/machine/spu/headers/log2f.h: Ditto. + * libm/machine/spu/headers/sqrt.h: Ditto. + * libm/machine/spu/headers/sqrtf.h: Ditto. + * libm/machine/spu/headers/tgamma.h: Ditto. + * libm/machine/spu/headers/tgammaf.h: Ditto. + 2007-09-28 Patrick Mansfield * libm/machine/spu/Makefile.am: Add new C files. diff --git a/newlib/libm/machine/spu/headers/acos.h b/newlib/libm/machine/spu/headers/acos.h index a647ef2d5..feec8c329 100644 --- a/newlib/libm/machine/spu/headers/acos.h +++ b/newlib/libm/machine/spu/headers/acos.h @@ -1,6 +1,18 @@ #include "headers/acosd2.h" +#include "headers/dom_chkd_negone_one.h" static __inline double _acos(double x) { - return spu_extract(_acosd2(spu_promote(x, 0)), 0); + double res; + vector double vx; + + vx = spu_splats(x); + res = spu_extract(_acosd2(vx), 0); +#ifndef _IEEE_LIBM + /* + * Domain error if not in the interval [-1, +1] + */ + dom_chkd_negone_one(vx); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/acosf.h b/newlib/libm/machine/spu/headers/acosf.h index ba93fa266..49ab628fc 100644 --- a/newlib/libm/machine/spu/headers/acosf.h +++ b/newlib/libm/machine/spu/headers/acosf.h @@ -1,6 +1,18 @@ #include "headers/acosf4.h" +#include "headers/dom_chkf_negone_one.h" static __inline float _acosf(float x) { - return spu_extract(_acosf4(spu_promote(x, 0)), 0); + float res; + vector float vx; + + vx = spu_splats(x); + res = spu_extract(_acosf4(vx), 0); +#ifndef _IEEE_LIBM + /* + * Domain error if not in the interval [-1, +1] + */ + dom_chkf_negone_one(vx); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/acosh.h b/newlib/libm/machine/spu/headers/acosh.h index 960394b64..d84708a30 100644 --- a/newlib/libm/machine/spu/headers/acosh.h +++ b/newlib/libm/machine/spu/headers/acosh.h @@ -1,6 +1,16 @@ #include "headers/acoshd2.h" +#include "headers/dom_chkd_less_than.h" static __inline double _acosh(double x) { - return spu_extract(_acoshd2(spu_promote(x, 0)), 0); + double res; + vector double vx; + vector double vc = { 1.0, 1.0 }; + + vx = spu_promote(x, 0); + res = spu_extract(_acoshd2(vx), 0); +#ifndef _IEEE_LIBM + dom_chkd_less_than(vx, vc); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/acoshf.h b/newlib/libm/machine/spu/headers/acoshf.h index db25aaf51..537706c8a 100644 --- a/newlib/libm/machine/spu/headers/acoshf.h +++ b/newlib/libm/machine/spu/headers/acoshf.h @@ -1,6 +1,16 @@ #include "headers/acoshf4.h" +#include "headers/dom_chkf_less_than.h" static __inline float _acoshf(float x) { - return spu_extract(_acoshf4(spu_promote(x, 0)), 0); + float res; + vector float vx; + vector float vc = { 1.0, 1.0, 1.0, 1.0 }; + + vx = spu_promote(x, 0); + res = spu_extract(_acoshf4(vx), 0); +#ifndef _IEEE_LIBM + dom_chkf_less_than(vx, vc); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/asin.h b/newlib/libm/machine/spu/headers/asin.h index 67988a45b..881e532bc 100644 --- a/newlib/libm/machine/spu/headers/asin.h +++ b/newlib/libm/machine/spu/headers/asin.h @@ -1,6 +1,18 @@ #include "headers/asind2.h" +#include "headers/dom_chkd_negone_one.h" static __inline double _asin(double x) { - return spu_extract(_asind2(spu_promote(x, 0)), 0); + double res; + vector double vx; + + vx = spu_splats(x); + res = spu_extract(_asind2(vx), 0); +#ifndef _IEEE_LIBM + /* + * Domain error if not in the interval [-1, +1] + */ + dom_chkd_negone_one(vx); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/asinf.h b/newlib/libm/machine/spu/headers/asinf.h index f60585d54..cb468ecec 100644 --- a/newlib/libm/machine/spu/headers/asinf.h +++ b/newlib/libm/machine/spu/headers/asinf.h @@ -1,6 +1,18 @@ #include "headers/asinf4.h" +#include "headers/dom_chkf_negone_one.h" static __inline float _asinf(float x) { - return spu_extract(_asinf4(spu_promote(x, 0)), 0); + float res; + vector float vx; + + vx = spu_splats(x); + res = spu_extract(_asinf4(vx), 0); +#ifndef _IEEE_LIBM + /* + * Domain error if not in the interval [-1, +1] + */ + dom_chkf_negone_one(vx); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/atanh.h b/newlib/libm/machine/spu/headers/atanh.h index ebc6f96c1..f81ad2ede 100644 --- a/newlib/libm/machine/spu/headers/atanh.h +++ b/newlib/libm/machine/spu/headers/atanh.h @@ -1,6 +1,18 @@ #include "headers/atanhd2.h" +#include "headers/dom_chkd_negone_one.h" static __inline double _atanh(double x) { - return spu_extract(_atanhd2(spu_promote(x, 0)), 0); + double res; + vector double vx; + + vx = spu_splats(x); + res = spu_extract(_atanhd2(vx), 0); +#ifndef _IEEE_LIBM + /* + * Domain error if not in the interval [-1, +1] + */ + dom_chkd_negone_one(vx); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/atanhf.h b/newlib/libm/machine/spu/headers/atanhf.h index e21dfefd1..f838c9f47 100644 --- a/newlib/libm/machine/spu/headers/atanhf.h +++ b/newlib/libm/machine/spu/headers/atanhf.h @@ -1,6 +1,18 @@ #include "headers/atanhf4.h" +#include "headers/dom_chkf_negone_one.h" static __inline float _atanhf(float x) { - return spu_extract(_atanhf4(spu_promote(x, 0)), 0); + float res; + vector float vx; + + vx = spu_splats(x); + res = spu_extract(_atanhf4(vx), 0); +#ifndef _IEEE_LIBM + /* + * Domain error if not in the interval [-1, +1] + */ + dom_chkf_negone_one(vx); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/dom_chkd_less_than.h b/newlib/libm/machine/spu/headers/dom_chkd_less_than.h new file mode 100644 index 000000000..f7018d62a --- /dev/null +++ b/newlib/libm/machine/spu/headers/dom_chkd_less_than.h @@ -0,0 +1,59 @@ +/* + Copyright 2007 + International Business Machines Corporation, + Sony Computer Entertainment, Incorporated, + Toshiba Corporation, + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + * Neither the names of the copyright holders nor the names of their + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS + IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A + PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include + +/* + * dom_chkd_less_than: "domain check double less than" + * + * Set errno to EDOM if vx < vc. + * + * This is for scalar use only, the input is a vector double, but both + * values in both vectors must be the same. + */ + +static __inline void dom_chkd_less_than (vector double vx, vector double vc) +{ + vector unsigned long long domain; + vector signed int verrno; + vector signed int fail = { EDOM, EDOM, EDOM, EDOM }; + + domain = spu_cmpgt(vc, vx); + verrno = spu_splats(errno); + /* + * domain is 2 long longs, but they have the same value. Even so, no + * special code is needed to extract the scalar errno (we have all ones + * or all zeroes for the preferred scalar slot). + */ + errno = spu_extract(spu_sel(verrno, fail, (vector unsigned int) domain), 0); +} diff --git a/newlib/libm/machine/spu/headers/dom_chkd_negone_one.h b/newlib/libm/machine/spu/headers/dom_chkd_negone_one.h new file mode 100644 index 000000000..e364f86f8 --- /dev/null +++ b/newlib/libm/machine/spu/headers/dom_chkd_negone_one.h @@ -0,0 +1,69 @@ +/* + Copyright 2007 + International Business Machines Corporation, + Sony Computer Entertainment, Incorporated, + Toshiba Corporation, + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + * Neither the names of the copyright holders nor the names of their + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS + IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A + PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include + +/* + * dom_chkd_negone_one: "domain check double negative-one and one": + * + * Set errno to EDOM if |x| > 1.0. + * + * This is for scalar use only, the input is a vector double, but both + * values in the vector must be the same. + * + * We *only* set errno, and do not bother setting the actual return value + * of any functions to a NAN. That way, we have the same method for double + * and single precision (there are no float nans for single precision so + * those can't return a nan). + * + * Note that for newlib, errno is/was a function call, so not so obviously + * we are not branchless here. Unknown if adding a branch (and avoiding a + * call to __errno) is faster than this current code. + */ + +static __inline void dom_chkd_negone_one (vector double vx) +{ + vector unsigned long long domain; + vector signed int verrno; + vector double ones = { 1.0, 1.0 }; + vector signed int fail = { EDOM, EDOM, EDOM, EDOM }; + + domain = spu_cmpabsgt(vx, ones); + verrno = spu_splats(errno); + /* + * domain is 2 long longs, but they have the same value. Even so, no + * special code is needed to extract the scalar errno (we have all ones + * or all zeroes for the preferred scalar slot). + */ + errno = spu_extract(spu_sel(verrno, fail, (vector unsigned int) domain), 0); +} diff --git a/newlib/libm/machine/spu/headers/dom_chkf_less_than.h b/newlib/libm/machine/spu/headers/dom_chkf_less_than.h new file mode 100644 index 000000000..7475b2d91 --- /dev/null +++ b/newlib/libm/machine/spu/headers/dom_chkf_less_than.h @@ -0,0 +1,59 @@ +/* + Copyright 2007 + International Business Machines Corporation, + Sony Computer Entertainment, Incorporated, + Toshiba Corporation, + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + * Neither the names of the copyright holders nor the names of their + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS + IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A + PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include + +/* + * dom_chkf_less_than: "domain check float less than" + * + * Set errno to EDOM if vx < vc. + * + * This is for scalar use only, the input is a vector float, but both + * values in both vectors must be the same. + */ + +static __inline void dom_chkf_less_than (vector float vx, vector float vc) +{ + vector unsigned int domain; + vector signed int verrno; + vector signed int fail = { EDOM, EDOM, EDOM, EDOM }; + + domain = spu_cmpgt(vc, vx); + verrno = spu_splats(errno); + /* + * domain is 2 long longs, but they have the same value. Even so, no + * special code is needed to extract the scalar errno (we have all ones + * or all zeroes for the preferred scalar slot). + */ + errno = spu_extract(spu_sel(verrno, fail, (vector unsigned int) domain), 0); +} diff --git a/newlib/libm/machine/spu/headers/dom_chkf_negone_one.h b/newlib/libm/machine/spu/headers/dom_chkf_negone_one.h new file mode 100644 index 000000000..f2671a9df --- /dev/null +++ b/newlib/libm/machine/spu/headers/dom_chkf_negone_one.h @@ -0,0 +1,69 @@ +/* + Copyright 2007 + International Business Machines Corporation, + Sony Computer Entertainment, Incorporated, + Toshiba Corporation, + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + * Neither the names of the copyright holders nor the names of their + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS + IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A + PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include + +/* + * dom_chkf_negone_one: "domain check float negative-one and one": + * + * Set errno to EDOM if |x| > 1.0. + * + * This is for scalar use only, the input is a vector float, but all + * values in the vector must be the same. + * + * We *only* set errno, and do not bother setting the actual return value + * of any functions to a NAN. That way, we have the same method for float + * and single precision (there are no float nans for single precision so + * those can't return a nan). + * + * Note that for newlib, errno is/was a function call, so not so obviously + * we are not branchless here. Unknown if adding a branch (and avoiding a + * call to __errno) is faster than this current code. + */ + +static __inline void dom_chkf_negone_one (vector float vx) +{ + vector unsigned int domain; + vector signed int verrno; + vector float ones = { 1.0, 1.0, 1.0, 1.0 }; + vector signed int fail = { EDOM, EDOM, EDOM, EDOM }; + + domain = spu_cmpabsgt(vx, ones); + verrno = spu_splats(errno); + /* + * domain is 4 ints, but they have the same value, even so no special + * code is needed to extract the scalar errno (we have all ones or all + * zeroes for the preferred scalar slot). + */ + errno = spu_extract(spu_sel(verrno, fail, (vector unsigned int) domain), 0); +} diff --git a/newlib/libm/machine/spu/headers/fmod.h b/newlib/libm/machine/spu/headers/fmod.h index 1506fb8dc..0236da187 100644 --- a/newlib/libm/machine/spu/headers/fmod.h +++ b/newlib/libm/machine/spu/headers/fmod.h @@ -34,6 +34,7 @@ #define _FMOD_H_ 1 #include +#include #include "headers/vec_literal.h" /* This implementation returns zero if y is a denorm or zero. @@ -57,6 +58,10 @@ static __inline double _fmod(double x, double y) vec_uint4 sign_mask = (vec_uint4)(VEC_SPLAT_U64(0x8000000000000000ULL)); vec_uint4 implied_1 = (vec_uint4)(VEC_SPLAT_U64(0x0010000000000000ULL)); vec_uint4 mant_mask = (vec_uint4)(VEC_SPLAT_U64(0x000FFFFFFFFFFFFFULL)); + vec_ullong2 domain; + vec_int4 verrno; + vec_double2 vc = { 0.0, 0.0 }; + vec_int4 fail = { EDOM, EDOM, EDOM, EDOM }; vx = (vec_uint4)spu_promote(x, 0); vy = (vec_uint4)spu_promote(y, 0); @@ -147,6 +152,15 @@ static __inline double _fmod(double x, double y) result = spu_sel(spu_andc(result, spu_rlmask(result0, -1)), vx, resultx); +#ifndef _IEEE_LIBM + /* + * If y is zero, set errno to EDOM + */ + domain = spu_cmpeq(vc, (vec_double2) vy); + verrno = spu_splats(errno); + errno = spu_extract(spu_sel(verrno, fail, (vector unsigned int) domain), 0); +#endif + return (spu_extract((vec_double2)result, 0)); } #endif /* _FMOD_H_ */ diff --git a/newlib/libm/machine/spu/headers/fmodf.h b/newlib/libm/machine/spu/headers/fmodf.h index 496ada490..a62141f97 100644 --- a/newlib/libm/machine/spu/headers/fmodf.h +++ b/newlib/libm/machine/spu/headers/fmodf.h @@ -33,6 +33,7 @@ #ifndef _FMODF_H_ #define _FMODF_H_ 1 +#include #include #include "headers/vec_literal.h" @@ -83,6 +84,10 @@ static __inline float _fmodf(float x, float y) vec_uint4 sign_mask = VEC_SPLAT_U32(0x80000000); vec_uint4 implied_1 = VEC_SPLAT_U32(0x00800000); vec_uint4 mant_mask = VEC_SPLAT_U32(0x007FFFFF); + vec_uint4 domain; + vec_int4 verrno; + vec_float4 vc = { 0.0, 0.0, 0.0, 0.0 }; + vec_int4 fail = { EDOM, EDOM, EDOM, EDOM }; vx = (vec_uint4)spu_promote(x, 0); vy = (vec_uint4)spu_promote(y, 0); @@ -150,6 +155,15 @@ static __inline float _fmodf(float x, float y) result = spu_sel(spu_andc(result, spu_rlmask(result0, -1)), vx, resultx); +#ifndef _IEEE_LIBM + /* + * If y is zero, set errno to EDOM + */ + domain = spu_cmpeq(vc, (vec_float4) vy); + verrno = spu_splats(errno); + errno = spu_extract(spu_sel(verrno, fail, (vector unsigned int) domain), 0); +#endif + return (spu_extract((vec_float4)result, 0)); #endif /* FMODF_INTEGER_RANGE */ } diff --git a/newlib/libm/machine/spu/headers/log.h b/newlib/libm/machine/spu/headers/log.h index 842abbf5e..9bb5e3bff 100644 --- a/newlib/libm/machine/spu/headers/log.h +++ b/newlib/libm/machine/spu/headers/log.h @@ -1,6 +1,16 @@ #include "headers/logd2.h" +#include "headers/dom_chkd_less_than.h" static __inline double _log(double x) { - return spu_extract(_logd2(spu_promote(x, 0)), 0); + double res; + vector double vx; + vector double vc = { 0.0, 0.0 }; + + vx = spu_promote(x, 0); + res = spu_extract(_logd2(vx), 0); +#ifndef _IEEE_LIBM + dom_chkd_less_than(vx, vc); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/log10.h b/newlib/libm/machine/spu/headers/log10.h index bc9007139..c3c6f9f80 100644 --- a/newlib/libm/machine/spu/headers/log10.h +++ b/newlib/libm/machine/spu/headers/log10.h @@ -1,6 +1,16 @@ #include "headers/log10d2.h" +#include "headers/dom_chkd_less_than.h" static __inline double _log10(double x) { - return spu_extract(_log10d2(spu_promote(x, 0)), 0); + double res; + vector double vx; + vector double vc = { 0.0, 0.0 }; + + vx = spu_promote(x, 0); + res = spu_extract(_log10d2(vx), 0); +#ifndef _IEEE_LIBM + dom_chkd_less_than(vx, vc); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/log1p.h b/newlib/libm/machine/spu/headers/log1p.h index 46244fbfd..758be8d37 100644 --- a/newlib/libm/machine/spu/headers/log1p.h +++ b/newlib/libm/machine/spu/headers/log1p.h @@ -1,6 +1,16 @@ #include "headers/log1pd2.h" +#include "headers/dom_chkd_less_than.h" static __inline double _log1p(double x) { - return spu_extract(_log1pd2(spu_promote(x, 0)), 0); + double res; + vector double vx; + vector double vc = { -1.0, -1.0 }; + + vx = spu_promote(x, 0); + res = spu_extract(_log1pd2(vx), 0); +#ifndef _IEEE_LIBM + dom_chkd_less_than(vx, vc); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/log1pf.h b/newlib/libm/machine/spu/headers/log1pf.h index 1eab9c9d2..b2af48669 100644 --- a/newlib/libm/machine/spu/headers/log1pf.h +++ b/newlib/libm/machine/spu/headers/log1pf.h @@ -1,6 +1,16 @@ #include "headers/log1pf4.h" +#include "headers/dom_chkf_less_than.h" static __inline float _log1pf(float x) { - return spu_extract(_log1pf4(spu_promote(x, 0)), 0); + float res; + vector float vx; + vector float vc = { -1.0, -1.0, -1.0, -1.0 }; + + vx = spu_promote(x, 0); + res = spu_extract(_log1pf4(vx), 0); +#ifndef _IEEE_LIBM + dom_chkf_less_than(vx, vc); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/log2.h b/newlib/libm/machine/spu/headers/log2.h index 50ab477c5..696bf9fae 100644 --- a/newlib/libm/machine/spu/headers/log2.h +++ b/newlib/libm/machine/spu/headers/log2.h @@ -1,6 +1,16 @@ #include "headers/log2d2.h" +#include "headers/dom_chkd_less_than.h" -static __inline double _log2(double vx) +static __inline double _log2(double x) { - return spu_extract(_log2d2(spu_promote(vx, 0)), 0); + double res; + vector double vx; + vector double vc = { 0.0, 0.0 }; + + vx = spu_promote(x, 0); + res = spu_extract(_log2d2(vx), 0); +#ifndef _IEEE_LIBM + dom_chkd_less_than(vx, vc); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/log2f.h b/newlib/libm/machine/spu/headers/log2f.h index b72665fd4..26717c7c0 100644 --- a/newlib/libm/machine/spu/headers/log2f.h +++ b/newlib/libm/machine/spu/headers/log2f.h @@ -33,6 +33,9 @@ #ifndef _LOG2F_H_ #define _LOG2F_H_ 1 +#include +#include "headers/dom_chkf_less_than.h" + /* * FUNCTION * float _log2f(float x) @@ -73,6 +76,8 @@ static __inline float _log2f(float x) float result; float x2, x4; float hi, lo; + vector float vx; + vector float vc = { 0.0, 0.0, 0.0, 0.0 }; in.f = x; @@ -109,6 +114,10 @@ static __inline float _log2f(float x) */ result += (float)(exponent); +#ifndef _IEEE_LIBM + vx = spu_promote(x, 0); + dom_chkf_less_than(vx, vc); +#endif return (result); } diff --git a/newlib/libm/machine/spu/headers/sqrt.h b/newlib/libm/machine/spu/headers/sqrt.h index 064c47def..63b034dd9 100644 --- a/newlib/libm/machine/spu/headers/sqrt.h +++ b/newlib/libm/machine/spu/headers/sqrt.h @@ -43,6 +43,7 @@ */ #include #include "headers/vec_literal.h" +#include "headers/dom_chkd_less_than.h" static __inline double _sqrt(double in) { @@ -53,6 +54,7 @@ static __inline double _sqrt(double in) vec_ullong2 mask = VEC_SPLAT_U64(0x7FE0000000000000ULL); vec_double2 x, dx, de, dd, dy, dg, dy2, dhalf; vec_double2 denorm, neg; + vec_double2 vc = { 0.0, 0.0 }; fhalf = VEC_SPLAT_F32(0.5f); dhalf = VEC_SPLAT_F64(0.5); @@ -124,6 +126,9 @@ static __inline double _sqrt(double in) dg = spu_sel(spu_andc(spu_or(dg, neg), denorm), x, nochange); +#ifndef _IEEE_LIBM + dom_chkd_less_than(spu_splats(in), vc); +#endif return (spu_extract(dg, 0)); } #endif /* _SQRT_H_ */ diff --git a/newlib/libm/machine/spu/headers/sqrtf.h b/newlib/libm/machine/spu/headers/sqrtf.h index 4ef91428b..c1f5ab223 100644 --- a/newlib/libm/machine/spu/headers/sqrtf.h +++ b/newlib/libm/machine/spu/headers/sqrtf.h @@ -1,6 +1,16 @@ #include "headers/sqrtf4.h" +#include "headers/dom_chkf_less_than.h" static __inline float _sqrtf(float in) { - return spu_extract(_sqrtf4(spu_promote(in, 0)), 0); + float res; + vector float vx; + vector float vc = { 0.0, 0.0, 0.0, 0.0 }; + + vx = spu_promote(in, 0); + res = spu_extract(_sqrtf4(vx), 0); +#ifndef _IEEE_LIBM + dom_chkf_less_than(vx, vc); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/tgamma.h b/newlib/libm/machine/spu/headers/tgamma.h index ea0c32e0f..bd2f31f27 100644 --- a/newlib/libm/machine/spu/headers/tgamma.h +++ b/newlib/libm/machine/spu/headers/tgamma.h @@ -1,6 +1,30 @@ +#include +#include "headers/truncd2.h" #include "headers/tgammad2.h" static __inline double _tgamma(double x) { - return spu_extract(_tgammad2(spu_promote(x, 0)), 0); + double res; + vector double vx; + vector double truncx; + vector double vc = { 0.0, 0.0 }; + vector unsigned long long cmpres; + vector signed int verrno, ferrno; + vector signed int fail = { EDOM, EDOM, EDOM, EDOM }; + + vx = spu_promote(x, 0); + res = spu_extract(_tgammad2(vx), 0); + +#ifndef _IEEE_LIBM + /* + * use vector truncd2 rather than splat x, and splat truncx. + */ + truncx = _truncd2(vx); + cmpres = spu_cmpeq(truncx, vx); + verrno = spu_splats(errno); + ferrno = spu_sel(verrno, fail, (vector unsigned int) cmpres); + cmpres = spu_cmpgt(vc, vx); + errno = spu_extract(spu_sel(verrno, ferrno, (vector unsigned int) cmpres), 0); +#endif + return res; } diff --git a/newlib/libm/machine/spu/headers/tgammaf.h b/newlib/libm/machine/spu/headers/tgammaf.h index c555d35b2..a30928c60 100644 --- a/newlib/libm/machine/spu/headers/tgammaf.h +++ b/newlib/libm/machine/spu/headers/tgammaf.h @@ -1,6 +1,29 @@ +#include +#include "headers/truncf4.h" #include "headers/tgammaf4.h" static __inline float _tgammaf(float x) { - return spu_extract(_tgammaf4(spu_promote(x, 0)), 0); + float res; + vector float vx; + vector float truncx; + vector float vc = { 0.0, 0.0 }; + vector unsigned int cmpres; + vector signed int verrno, ferrno; + vector signed int fail = { EDOM, EDOM, EDOM, EDOM }; + + vx = spu_promote(x, 0); + res = spu_extract(_tgammaf4(vx), 0); +#ifndef _IEEE_LIBM + /* + * use vector truncf4 rather than splat x, and splat truncx. + */ + truncx = _truncf4(vx); + cmpres = spu_cmpeq(truncx, vx); + verrno = spu_splats(errno); + ferrno = spu_sel(verrno, fail, (vector unsigned int) cmpres); + cmpres = spu_cmpgt(vc, vx); + errno = spu_extract(spu_sel(verrno, ferrno, (vector unsigned int) cmpres), 0); +#endif + return res; }