diff options
author | Adhemerval Zanella <adhemerval.zanella@linaro.org> | 2021-04-03 23:52:45 -0300 |
---|---|---|
committer | Adhemerval Zanella <adhemerval.zanella@linaro.org> | 2021-12-13 09:08:07 -0300 |
commit | 2eb1cd2f47fe6568c539fa105551bb73df8368ec (patch) | |
tree | 8404bf6b82969a864d916c1bdbda844fad7231e2 /sysdeps/powerpc/fpu | |
parent | a1d3c9b64266d84736ba8eac96aa864165905b79 (diff) | |
download | glibc-2eb1cd2f47fe6568c539fa105551bb73df8368ec.tar.gz glibc-2eb1cd2f47fe6568c539fa105551bb73df8368ec.tar.xz glibc-2eb1cd2f47fe6568c539fa105551bb73df8368ec.zip |
math: Remove powerpc e_hypot
The generic implementation is shows only slight worse performance: POWER10 reciprocal-throughput latency master 8.28478 13.7253 new hypot 7.21945 13.1933 POWER9 reciprocal-throughput latency master 13.4024 14.0967 new hypot 14.8479 15.8061 POWER8 reciprocal-throughput latency master 15.5767 16.8885 new hypot 16.5371 18.4057 One way to improve might to make gcc generate xsmaxdp/xsmindp for fmax/fmin (it onl does for -ffast-math, clang does for default options). Checked on powerpc64-linux-gnu (power8) and powerpc64le-linux-gnu (power9).
Diffstat (limited to 'sysdeps/powerpc/fpu')
-rw-r--r-- | sysdeps/powerpc/fpu/e_hypot.c | 87 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/e_hypotf.c | 78 |
2 files changed, 0 insertions, 165 deletions
diff --git a/sysdeps/powerpc/fpu/e_hypot.c b/sysdeps/powerpc/fpu/e_hypot.c deleted file mode 100644 index f96c589bbd..0000000000 --- a/sysdeps/powerpc/fpu/e_hypot.c +++ /dev/null @@ -1,87 +0,0 @@ -/* Pythagorean addition using doubles - Copyright (C) 2011-2021 Free Software Foundation, Inc. - This file is part of the GNU C Library - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If - not, see <https://www.gnu.org/licenses/>. */ - -#include <math.h> -#include <math_private.h> -#include <math-underflow.h> -#include <stdint.h> -#include <libm-alias-finite.h> - -/* __ieee754_hypot(x,y) - * - * This a FP only version without any FP->INT conversion. - * It is similar to default C version, making appropriates - * overflow and underflows checks as well scaling when it - * is needed. - */ - -double -__ieee754_hypot (double x, double y) -{ - if ((isinf (x) || isinf (y)) - && !issignaling (x) && !issignaling (y)) - return INFINITY; - if (isnan (x) || isnan (y)) - return x + y; - - x = fabs (x); - y = fabs (y); - - if (y > x) - { - double t = x; - x = y; - y = t; - } - if (y == 0.0) - return x; - - /* if y is higher enough, y * 2^60 might overflow. The tests if - y >= 1.7976931348623157e+308/2^60 (two60factor) and uses the - appropriate check to avoid the overflow exception generation. */ - if (y <= 0x1.fffffffffffffp+963 && x > (y * 0x1p+60)) - return x + y; - - if (x > 0x1p+500) - { - x *= 0x1p-600; - y *= 0x1p-600; - return sqrt (x * x + y * y) / 0x1p-600; - } - if (y < 0x1p-500) - { - if (y <= 0x0.fffffffffffffp-1022) - { - x *= 0x1p+1022; - y *= 0x1p+1022; - double ret = sqrt (x * x + y * y) / 0x1p+1022; - math_check_force_underflow_nonneg (ret); - return ret; - } - else - { - x *= 0x1p+600; - y *= 0x1p+600; - return sqrt (x * x + y * y) / 0x1p+600; - } - } - return sqrt (x * x + y * y); -} -#ifndef __ieee754_hypot -libm_alias_finite (__ieee754_hypot, __hypot) -#endif diff --git a/sysdeps/powerpc/fpu/e_hypotf.c b/sysdeps/powerpc/fpu/e_hypotf.c deleted file mode 100644 index fa201dda51..0000000000 --- a/sysdeps/powerpc/fpu/e_hypotf.c +++ /dev/null @@ -1,78 +0,0 @@ -/* Pythagorean addition using floats - Copyright (C) 2011-2021 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If - not, see <https://www.gnu.org/licenses/>. */ - -#include <math.h> -#include <math_private.h> -#include <stdint.h> -#include <libm-alias-finite.h> - -/* __ieee754_hypotf(x,y) - - This a FP only version without any FP->INT conversion. - It is similar to default C version, making appropriates - overflow and underflows checks as using double precision - instead of scaling. */ - -#ifdef _ARCH_PWR7 -/* POWER7 isinf and isnan optimizations are fast. */ -# define TEST_INF_NAN(x, y) \ - if ((isinff(x) || isinff(y)) \ - && !issignaling (x) && !issignaling (y)) \ - return INFINITY; \ - if (isnanf(x) || isnanf(y)) \ - return x + y; -# else -/* For POWER6 and below isinf/isnan triggers LHS and PLT calls are - * costly (especially for POWER6). */ -# define GET_TWO_FLOAT_WORD(f1,f2,i1,i2) \ - do { \ - ieee_float_shape_type gf_u1; \ - ieee_float_shape_type gf_u2; \ - gf_u1.value = (f1); \ - gf_u2.value = (f2); \ - (i1) = gf_u1.word & 0x7fffffff; \ - (i2) = gf_u2.word & 0x7fffffff; \ - } while (0) - -# define TEST_INF_NAN(x, y) \ - do { \ - uint32_t hx, hy; \ - GET_TWO_FLOAT_WORD(x, y, hx, hy); \ - if (hy > hx) { \ - uint32_t ht = hx; hx = hy; hy = ht; \ - } \ - if (hx >= 0x7f800000) { \ - if ((hx == 0x7f800000 || hy == 0x7f800000) \ - && !issignaling (x) && !issignaling (y)) \ - return INFINITY; \ - return x + y; \ - } \ - } while (0) -#endif - - -float -__ieee754_hypotf (float x, float y) -{ - TEST_INF_NAN (x, y); - - return sqrt ((double) x * x + (double) y * y); -} -#ifndef __ieee754_hypotf -libm_alias_finite (__ieee754_hypotf, __hypotf) -#endif |