[PATCH 3/3] Add domain checks for SPU math functions
Jeff Johnston
jjohnstn@redhat.com
Mon Oct 1 18:23:00 GMT 2007
Patch applied.
-- Jeff J.
Patrick Mansfield wrote:
> Add domain checks for these SPU math functions:
>
> acos acosf
> acosh acoshf
> asin asinf
> atanh atanhf
> fmod fmodf
> log logf (via log2f)
> log10 log10f (via log2f)
> log1p log1pf
> log2 log2f
> sqrt sqrtf
> tgamma tgammaf
>
> Newlib ChangeLog:
>
> 2007-09-19 Patrick Mansfield <patmans@us.ibm.com>
>
> * 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.
>
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/acos.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/acos.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/dom_chkd_negone_one.h
> ===================================================================
> --- /dev/null
> +++ push-sdk3-patches-quilt-base/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 <errno.h>
> +
> +/*
> + * 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);
> +}
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/asin.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/asin.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/dom_chkf_negone_one.h
> ===================================================================
> --- /dev/null
> +++ push-sdk3-patches-quilt-base/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 <errno.h>
> +
> +/*
> + * 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);
> +}
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/acosf.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/acosf.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/asinf.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/asinf.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/atanh.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/atanh.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/atanhf.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/atanhf.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/dom_chkd_less_than.h
> ===================================================================
> --- /dev/null
> +++ push-sdk3-patches-quilt-base/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 <errno.h>
> +
> +/*
> + * 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);
> +}
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/acosh.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/acosh.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/dom_chkf_less_than.h
> ===================================================================
> --- /dev/null
> +++ push-sdk3-patches-quilt-base/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 <errno.h>
> +
> +/*
> + * 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);
> +}
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/acoshf.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/acoshf.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/log.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/log.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/log10.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/log10.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/log1p.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/log1p.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/log2.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/log2.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/sqrt.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/sqrt.h
> +++ push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/sqrt.h
> @@ -43,6 +43,7 @@
> */
> #include <spu_intrinsics.h>
> #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_ */
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/log1pf.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/log1pf.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/log2f.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/log2f.h
> +++ push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/log2f.h
> @@ -33,6 +33,9 @@
> #ifndef _LOG2F_H_
> #define _LOG2F_H_ 1
>
> +#include <spu_intrinsics.h>
> +#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);
> }
>
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/sqrtf.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/sqrtf.h
> +++ push-sdk3-patches-quilt-base/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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/fmodf.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/fmodf.h
> +++ push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/fmodf.h
> @@ -33,6 +33,7 @@
> #ifndef _FMODF_H_
> #define _FMODF_H_ 1
>
> +#include <errno.h>
> #include <spu_intrinsics.h>
> #include "headers/vec_literal.h"
>
> @@ -83,6 +84,10 @@ static __inline float _fmodf(float x, fl
> 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, fl
> 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 */
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/fmod.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/fmod.h
> +++ push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/fmod.h
> @@ -34,6 +34,7 @@
> #define _FMOD_H_ 1
>
> #include <spu_intrinsics.h>
> +#include <errno.h>
> #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, d
> 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, d
> 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_ */
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/tgamma.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/tgamma.h
> +++ push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/tgamma.h
> @@ -1,6 +1,30 @@
> +#include <errno.h>
> +#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;
> }
> Index: push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/tgammaf.h
> ===================================================================
> --- push-sdk3-patches-quilt-base.orig/newlib/libm/machine/spu/headers/tgammaf.h
> +++ push-sdk3-patches-quilt-base/newlib/libm/machine/spu/headers/tgammaf.h
> @@ -1,6 +1,29 @@
> +#include <errno.h>
> +#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;
> }
>
More information about the Newlib
mailing list