diff options
author | Adhemerval Zanella <azanella@linux.vnet.ibm.com> | 2011-06-17 13:30:52 -0500 |
---|---|---|
committer | Ryan S. Arnold <rsa@us.ibm.com> | 2011-06-17 14:12:07 -0500 |
commit | 6559b01648c2a08ea862ae923ce7d86c6e8cca45 (patch) | |
tree | 456119f629e086a9a279edc57a440e912c20846f | |
parent | 4749a0058b27274a95c5a798e339c7299cdf890e (diff) | |
download | glibc-6559b01648c2a08ea862ae923ce7d86c6e8cca45.zip glibc-6559b01648c2a08ea862ae923ce7d86c6e8cca45.tar.gz glibc-6559b01648c2a08ea862ae923ce7d86c6e8cca45.tar.bz2 |
Trigonometric optimizations for POWER cpus
These optimizations remove most of the FP->INT conversions making
the computation done with FP operations instead. This eliminates
Load-Hit-Store (LHS) stalls on POWER, increasing performance of
hypot/hypotf (about 50% on POWER7, 25% on POWER6, and 30% on POWER5)
and sinf/cosf (30% on POWER7, 15% on POWER6, and 10% on POWER5).
(cherry picked from commit 5939fc0867e1616a537c26d84ca1612b53b0303e)
-rw-r--r-- | ChangeLog | 22 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/Makefile | 5 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/e_hypot.c | 86 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/e_hypotf.c | 86 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/e_rem_pio2f.c | 431 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/k_cosf.c | 65 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/k_sinf.c | 53 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/s_cosf.c | 71 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/s_float_bitwise.h | 35 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/s_scalbnf.c | 135 | ||||
-rw-r--r-- | sysdeps/powerpc/fpu/s_sinf.c | 70 | ||||
-rw-r--r-- | sysdeps/powerpc/powerpc32/fpu/s_float_bitwise.S | 96 | ||||
-rw-r--r-- | sysdeps/powerpc/powerpc32/power7/fpu/s_float_bitwise.S | 149 | ||||
-rw-r--r-- | sysdeps/powerpc/powerpc64/fpu/s_float_bitwise.S | 87 | ||||
-rw-r--r-- | sysdeps/powerpc/powerpc64/power7/fpu/s_float_bitwise.S | 83 |
15 files changed, 1474 insertions, 0 deletions
@@ -1,3 +1,25 @@ +2011-06-15 Adhemerval Zanella <azanella@linux.vnet.ibm.com> + + * sysdeps/powerpc/fpu/Makefile: Added new objects. + * sysdeps/powerpc/fpu/e_hypot.c: New file: hypot optimized for POWER. + * sysdeps/powerpc/fpu/e_hypotf.c: New file: hypotf optimized for + POWER. + * sysdeps/powerpc/fpu/e_rem_pio2f.c: New file: optimized for POWER. + * sysdeps/powerpc/fpu/k_cosf.c: New file: optimized for POWER. + * sysdeps/powerpc/fpu/k_sinf.c: New file: optimized for POWER. + * sysdeps/powerpc/fpu/s_cosf.c: New file: sinf optimized for POWER. + * sysdeps/powerpc/fpu/s_sinf.c: New file: cosf optimized for POWER. + * sysdeps/powerpc/fpu/s_float_bitwise.h: New file: bitwise operation + over floats. + * sysdeps/powerpc/powerpc32/fpu/s_float_bitwise.S: New file: bitwise + operation over float, PPC32 implementation. + * sysdeps/powerpc/powerpc32/power7/fpu/s_float_bitwise.S: New file: + bitwise operation over floats, PPC32 implementation using VSX. + * sysdeps/powerpc/powerpc64/fpu/s_float_bitwise.S: New file: bitwise + operation over float, PPC64 implementation. + * sysdeps/powerpc/powerpc64/power7/fpu/s_float_bitwise.S: New file: + bitwise operation over floats, PPC64 implementation using VSX. + 2011-05-18 Ryan S. Arnold <rsa@us.ibm.com> * sysdeps/powerpc/powerpc64/Makefile (no-special-regs): Add -mno-vsx diff --git a/sysdeps/powerpc/fpu/Makefile b/sysdeps/powerpc/fpu/Makefile index ffacf1a..1c98ca6 100644 --- a/sysdeps/powerpc/fpu/Makefile +++ b/sysdeps/powerpc/fpu/Makefile @@ -4,6 +4,11 @@ libm-tests += test-powerpc-snan # libm needs ld.so to access dl_hwcap $(objpfx)libm.so: $(elfobjdir)/ld.so + +duplicated-routines = s_float_bitwise + +libm-sysdep_routines += $(duplicated-routines:s_%=m_%) +sysdep_routines += $(duplicated-routines) endif ifeq ($(subdir),stdlib) diff --git a/sysdeps/powerpc/fpu/e_hypot.c b/sysdeps/powerpc/fpu/e_hypot.c new file mode 100644 index 0000000..c09c0c8 --- /dev/null +++ b/sysdeps/powerpc/fpu/e_hypot.c @@ -0,0 +1,86 @@ +/* Pythagorean addition using doubles + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + 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, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include "math.h" + +static const double two60 = 1.152921504606847e+18; +static const double two500 = 3.2733906078961419e+150; +static const double two600 = 4.149515568880993e+180; +static const double two1022 = 4.49423283715579e+307; +static const double twoM500 = 3.054936363499605e-151; +static const double twoM600 = 4.616489308892868e-128; +static const double pdnum = 2.225073858507201e-308; + +/* __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) +{ + double j; + + if (isinf(x) || isinf(y)) + { + return INFINITY; + } + if (isnan(x) || isnan(y)) + { + return NAN; + } + x = __builtin_fabs (x); + y = __builtin_fabs (y); + if (y > x) + { + j = x; + x = y; + y = j; + } + if (y == 0.0 || (x / y) > two60) + { + return x + y; + } + if (x > two500) + { + x *= twoM600; + y *= twoM600; + return __builtin_sqrt (x * x + y * y) / twoM600; + } + if (y < twoM500) + { + if (y <= pdnum) + { + x *= two1022; + y *= two1022; + return __builtin_sqrt (x * x + y * y) / two1022; + } + else + { + x *= two600; + y *= two600; + return __builtin_sqrt (x * x + y * y) / two600; + } + } + return __builtin_sqrt (x * x + y * y); +} diff --git a/sysdeps/powerpc/fpu/e_hypotf.c b/sysdeps/powerpc/fpu/e_hypotf.c new file mode 100644 index 0000000..55eb2cb --- /dev/null +++ b/sysdeps/powerpc/fpu/e_hypotf.c @@ -0,0 +1,86 @@ +/* Pythagorean addition using floats + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + 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, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include "math.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 well scaling when it + * is needed. + */ + +static const float two30 = 1.0737418e09; +static const float two50 = 1.1259000e15; +static const float two60 = 1.1529221e18; +static const float two126 = 8.5070592e+37; +static const float twoM50 = 8.8817842e-16; +static const float twoM60 = 6.7762644e-21; +static const float pdnum = 1.1754939e-38; + +float +__ieee754_hypotf (float x, float y) +{ + float j; + + if (isinff(x) || isinff(y)) + { + return INFINITY; + } + if (isnanf(x) || isnanf(y)) + { + return NAN; + } + x = __builtin_fabsf (x); + y = __builtin_fabsf (y); + if (y > x) + { + j = x; + x = y; + y = j; + } + if (y == 0.0 || (x / y) > two30) + { + return x + y; + } + if (x > two50) + { + x *= twoM60; + y *= twoM60; + return __builtin_sqrtf (x * x + y * y) / twoM60; + } + if (y < twoM50) + { + if (y <= pdnum) + { + x *= two126; + y *= two126; + return __builtin_sqrtf (x * x + y * y) / two126; + } + else + { + x *= two60; + y *= two60; + return __builtin_sqrtf (x * x + y * y) / two60; + } + } + return __builtin_sqrtf (x * x + y * y); +} diff --git a/sysdeps/powerpc/fpu/e_rem_pio2f.c b/sysdeps/powerpc/fpu/e_rem_pio2f.c new file mode 100644 index 0000000..5b6e7b6 --- /dev/null +++ b/sysdeps/powerpc/fpu/e_rem_pio2f.c @@ -0,0 +1,431 @@ +/* e_rem_pio2f.c -- float version of e_rem_pio2.c + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + 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, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <math.h> +#include <stdint.h> + +#include "s_float_bitwise.h" + +static int32_t __fp_kernel_rem_pio2f (float *, float *, float, int32_t); + +/* __ieee754_rem_pio2f(x,y) + * + * return the remainder of x rem pi/2 in y[0]+y[1] + */ + +static const float two_over_pi[] = { + 1.62000000e+02, 2.49000000e+02, 1.31000000e+02, 1.10000000e+02, + 7.80000000e+01, 6.80000000e+01, 2.10000000e+01, 4.10000000e+01, + 2.52000000e+02, 3.90000000e+01, 8.70000000e+01, 2.09000000e+02, + 2.45000000e+02, 5.20000000e+01, 2.21000000e+02, 1.92000000e+02, + 2.19000000e+02, 9.80000000e+01, 1.49000000e+02, 1.53000000e+02, + 6.00000000e+01, 6.70000000e+01, 1.44000000e+02, 6.50000000e+01, + 2.54000000e+02, 8.10000000e+01, 9.90000000e+01, 1.71000000e+02, + 2.22000000e+02, 1.87000000e+02, 1.97000000e+02, 9.70000000e+01, + 1.83000000e+02, 3.60000000e+01, 1.10000000e+02, 5.80000000e+01, + 6.60000000e+01, 7.70000000e+01, 2.10000000e+02, 2.24000000e+02, + 6.00000000e+00, 7.30000000e+01, 4.60000000e+01, 2.34000000e+02, + 9.00000000e+00, 2.09000000e+02, 1.46000000e+02, 2.80000000e+01, + 2.54000000e+02, 2.90000000e+01, 2.35000000e+02, 2.80000000e+01, + 1.77000000e+02, 4.10000000e+01, 1.67000000e+02, 6.20000000e+01, + 2.32000000e+02, 1.30000000e+02, 5.30000000e+01, 2.45000000e+02, + 4.60000000e+01, 1.87000000e+02, 6.80000000e+01, 1.32000000e+02, + 2.33000000e+02, 1.56000000e+02, 1.12000000e+02, 3.80000000e+01, + 1.80000000e+02, 9.50000000e+01, 1.26000000e+02, 6.50000000e+01, + 5.70000000e+01, 1.45000000e+02, 2.14000000e+02, 5.70000000e+01, + 1.31000000e+02, 8.30000000e+01, 5.70000000e+01, 2.44000000e+02, + 1.56000000e+02, 1.32000000e+02, 9.50000000e+01, 1.39000000e+02, + 1.89000000e+02, 2.49000000e+02, 4.00000000e+01, 5.90000000e+01, + 3.10000000e+01, 2.48000000e+02, 1.51000000e+02, 2.55000000e+02, + 2.22000000e+02, 5.00000000e+00, 1.52000000e+02, 1.50000000e+01, + 2.39000000e+02, 4.70000000e+01, 1.70000000e+01, 1.39000000e+02, + 9.00000000e+01, 1.00000000e+01, 1.09000000e+02, 3.10000000e+01, + 1.09000000e+02, 5.40000000e+01, 1.26000000e+02, 2.07000000e+02, + 3.90000000e+01, 2.03000000e+02, 9.00000000e+00, 1.83000000e+02, + 7.90000000e+01, 7.00000000e+01, 6.30000000e+01, 1.02000000e+02, + 1.58000000e+02, 9.50000000e+01, 2.34000000e+02, 4.50000000e+01, + 1.17000000e+02, 3.90000000e+01, 1.86000000e+02, 1.99000000e+02, + 2.35000000e+02, 2.29000000e+02, 2.41000000e+02, 1.23000000e+02, + 6.10000000e+01, 7.00000000e+00, 5.70000000e+01, 2.47000000e+02, + 1.38000000e+02, 8.20000000e+01, 1.46000000e+02, 2.34000000e+02, + 1.07000000e+02, 2.51000000e+02, 9.50000000e+01, 1.77000000e+02, + 3.10000000e+01, 1.41000000e+02, 9.30000000e+01, 8.00000000e+00, + 8.60000000e+01, 3.00000000e+00, 4.80000000e+01, 7.00000000e+01, + 2.52000000e+02, 1.23000000e+02, 1.07000000e+02, 1.71000000e+02, + 2.40000000e+02, 2.07000000e+02, 1.88000000e+02, 3.20000000e+01, + 1.54000000e+02, 2.44000000e+02, 5.40000000e+01, 2.90000000e+01, + 1.69000000e+02, 2.27000000e+02, 1.45000000e+02, 9.70000000e+01, + 9.40000000e+01, 2.30000000e+02, 2.70000000e+01, 8.00000000e+00, + 1.01000000e+02, 1.53000000e+02, 1.33000000e+02, 9.50000000e+01, + 2.00000000e+01, 1.60000000e+02, 1.04000000e+02, 6.40000000e+01, + 1.41000000e+02, 2.55000000e+02, 2.16000000e+02, 1.28000000e+02, + 7.70000000e+01, 1.15000000e+02, 3.90000000e+01, 4.90000000e+01, + 6.00000000e+00, 6.00000000e+00, 2.10000000e+01, 8.60000000e+01, + 2.02000000e+02, 1.15000000e+02, 1.68000000e+02, 2.01000000e+02, + 9.60000000e+01, 2.26000000e+02, 1.23000000e+02, 1.92000000e+02, + 1.40000000e+02, 1.07000000e+02 +}; + + +static const float npio2_hw[] = { + 1.57077026e+00, 3.14154053e+00, 4.71228027e+00, 6.28308105e+00, + 7.85388184e+00, 9.42456055e+00, 1.09953613e+01, 1.25661621e+01, + 1.41369629e+01, 1.57077637e+01, 1.72783203e+01, 1.88491211e+01, + 2.04199219e+01, 2.19907227e+01, 2.35615234e+01, 2.51323242e+01, + 2.67031250e+01, 2.82739258e+01, 2.98447266e+01, 3.14155273e+01, + 3.29863281e+01, 3.45566406e+01, 3.61279297e+01, 3.76982422e+01, + 3.92695312e+01, 4.08398438e+01, 4.24111328e+01, 4.39814453e+01, + 4.55527344e+01, 4.71230469e+01, 4.86943359e+01, 5.02646484e+01 +}; + +static const float PIo2[] = { + 1.5703125000e+00, /* 0x3fc90000 */ + 4.5776367188e-04, /* 0x39f00000 */ + 2.5987625122e-05, /* 0x37da0000 */ + 7.5437128544e-08, /* 0x33a20000 */ + 6.0026650317e-11, /* 0x2e840000 */ + 7.3896444519e-13, /* 0x2b500000 */ + 5.3845816694e-15, /* 0x27c20000 */ + 5.6378512969e-18, /* 0x22d00000 */ + 8.3009228831e-20, /* 0x1fc40000 */ + 3.2756352257e-22, /* 0x1bc60000 */ + 6.3331015649e-25, /* 0x17440000 */ +}; + +static const float zero = 0.0000000000e+00; +static const float one = 1.0000000000; +static const float two8 = 2.5600000000e+02; +static const float twon8 = 3.9062500000e-03; + +static const float half = 5.0000000000e-01; +static const float invpio2 = 6.3661980629e-01; +static const float pio2_1 = 1.5707855225e+00; +static const float pio2_1t = 1.0804334124e-05; +static const float pio2_2 = 1.0804273188e-05; +static const float pio2_2t = 6.0770999344e-11; +static const float pio2_3 = 6.0770943833e-11; +static const float pio2_3t = 6.1232342629e-17; + +static const float pio4 = 7.8539801e-01; +static const float pio3_4 = 2.3561945e+00; +static const float pio2_24b = 1.5707951e+00; +static const float pio2_2e7 = 2.0106054e+02; + + +int32_t +__ieee754_rem_pio2f (float x, float *y) +{ + float ax, z, n, r, w, t, e0; + float tx[3]; + int32_t i, nx; + + ax = __builtin_fabsf (x); + if (ax <= pio4) + { + y[0] = x; + y[1] = 0; + return 0; + } + if (ax < pio3_4) + { + if (x > 0) + { + z = x - pio2_1; + if (!__float_and_test28 (ax, pio2_24b)) + { + y[0] = z - pio2_1t; + y[1] = (z - y[0]) - pio2_1t; + } + else + { + z -= pio2_2; + y[0] = z - pio2_2t; + y[1] = (z - y[0]) - pio2_2t; + } + return 1; + } + else + { + z = x + pio2_1; + if (!__float_and_test28 (ax, pio2_24b)) + { + y[0] = z + pio2_1t; + y[1] = (z - y[0]) + pio2_1t; + } + else + { + z += pio2_2; + y[0] = z + pio2_2t; + y[1] = (z - y[0]) + pio2_2t; + } + return -1; + } + } + if (ax <= pio2_2e7) + { + n = floorf (ax * invpio2 + half); + i = (int32_t) n; + r = ax - n * pio2_1; + w = n * pio2_1t; /* 1st round good to 40 bit */ + if (i < 32 && !__float_and_test24 (ax, npio2_hw[i - 1])) + { + y[0] = r - w; + } + else + { + float i, j; + j = __float_and8 (ax); + y[0] = r - w; + i = __float_and8 (y[0]); + if (j / i > 256.0 || j / i < 3.9062500e-3) + { /* 2nd iterations needed, good to 57 */ + t = r; + w = n * pio2_2; + r = t - w; + w = n * pio2_2t - ((t - r) - w); + y[0] = r - w; + i = __float_and8 (y[0]); + if (j / i > 33554432 || j / i < 2.9802322e-8) + { /* 3rd iteration needed, 74 bits acc */ + t = r; + w = n * pio2_3; + r = t - w; + w = n * pio2_3t - ((t - r) - w); + y[0] = r - w; + } + } + } + y[1] = (r - y[0]) - w; + if (x < 0) + { + y[0] = -y[0]; + y[1] = -y[1]; + return -i; + } + else + { + return i; + } + } + + /* all other (large) arguments */ + if (isnanf (x) || isinff (x)) + { + y[0] = y[1] = x - x; + return 0; + } + + /* set z = scalbn(|x|,ilogb(x)-7) */ + e0 = __float_and8 (ax / 128.0); + z = ax / e0; + + tx[0] = floorf (z); + z = (z - tx[0]) * two8; + tx[1] = floorf (z); + z = (z - tx[1]) * two8; + tx[2] = floorf (z); + + nx = 3; + while (tx[nx - 1] == zero) + nx--; + + i = __fp_kernel_rem_pio2f (tx, y, e0, nx); + if (x < 0) + { + y[0] = -y[0]; + y[1] = -y[1]; + return -i; + } + return i; +} + +static int32_t +__fp_kernel_rem_pio2f (float *x, float *y, float e0, int32_t nx) +{ + int32_t jz, jx, jv, jp, jk, carry, n, iq[20], i, j, k, m, q0, ih, exp; + float z, fw, f[20], fq[20], q[20]; + + /* initialize jk */ + jp = jk = 9; + + /* determine jx,jv,q0, note that 3>q0 */ + jx = nx - 1; + exp = __float_get_exp (e0) - 127; + jv = (exp - 3) / 8; + if (jv < 0) + jv = 0; + q0 = exp - 8 * (jv + 1); + + /* set up f[0] to f[jx+jk] where f[jx+jk] = two_over_pi[jv+jk] */ + j = jv - jx; + m = jx + jk; + for (i = 0; i <= m; i++, j++) + f[i] = (j < 0) ? zero : two_over_pi[j]; + + /* compute q[0],q[1],...q[jk] */ + for (i = 0; i <= jk; i++) + { + for (j = 0, fw = 0.0; j <= jx; j++) + fw += x[j] * f[jx + i - j]; + q[i] = fw; + } + + jz = jk; +recompute: + /* distill q[] into iq[] reversingly */ + for (i = 0, j = jz, z = q[jz]; j > 0; i++, j--) + { + fw = truncf (twon8 * z); + iq[i] = (int32_t) (z - two8 * fw); + z = q[j - 1] + fw; + } + + /* compute n */ + z = __scalbnf (z, q0); /* actual value of z */ + z -= 8.0 * floorf (z * 0.125); /* trim off integer >= 8 */ + n = (int32_t) z; + z -= truncf (z); + ih = 0; + if (q0 > 0) + { /* need iq[jz-1] to determine n */ + i = (iq[jz - 1] >> (8 - q0)); + n += i; + iq[jz - 1] -= i << (8 - q0); + ih = iq[jz - 1] >> (7 - q0); + } + else if (q0 == 0) + ih = iq[jz - 1] >> 8; + else if (z >= 0.5) + ih = 2; + + if (ih > 0) + { /* q > 0.5 */ + n += 1; + carry = 0; + for (i = 0; i < jz; i++) + { /* compute 1-q */ + j = iq[i]; + if (carry == 0) + { + if (j != 0) + { + carry = 1; + iq[i] = 0x100 - j; + } + } + else + iq[i] = 0xff - j; + } + if (q0 > 0) + { /* rare case: chance is 1 in 12 */ + switch (q0) + { + case 1: + iq[jz - 1] &= 0x7f; + break; + case 2: + iq[jz - 1] &= 0x3f; + break; + } + } + if (ih == 2) + { + z = one - z; + if (carry != 0) + z -= __scalbnf (one, q0); + } + } + + /* check if recomputation is needed */ + if (z == zero) + { + j = 0; + for (i = jz - 1; i >= jk; i--) + j |= iq[i]; + if (j == 0) + { /* need recomputation */ + for (k = 1; iq[jk - k] == 0; k++); /* k = no. of terms needed */ + + for (i = jz + 1; i <= jz + k; i++) + { /* add q[jz+1] to q[jz+k] */ + f[jx + i] = two_over_pi[jv + i]; + for (j = 0, fw = 0.0; j <= jx; j++) + fw += x[j] * f[jx + i - j]; + q[i] = fw; + } + jz += k; + goto recompute; + } + } + + /* chop off zero terms */ + if (z == 0.0) + { + jz -= 1; + q0 -= 8; + while (iq[jz] == 0) + { + jz--; + q0 -= 8; + } + } + else + { /* break z into 8-bit if necessary */ + z = __scalbnf (z, -q0); + if (z >= two8) + { + fw = truncf (twon8 * z); + iq[jz] = (int32_t) (z - two8 * fw); + jz += 1; + q0 += 8; + iq[jz] = (int32_t) fw; + } + else + iq[jz] = (int32_t) z; + } + + /* convert integer "bit" chunk to floating-point value */ + fw = __scalbnf (one, q0); + for (i = jz; i >= 0; i--) + { + q[i] = fw * (float) iq[i]; + fw *= twon8; + } + + /* compute PIo2[0,...,jp]*q[jz,...,0] */ + for (i = jz; i >= 0; i--) + { + for (fw = 0.0, k = 0; k <= jp && k <= jz - i; k++) + fw += PIo2[k] * q[i + k]; + fq[jz - i] = fw; + } + + /* compress fq[] into y[] */ + fw = 0.0; + for (i = jz; i >= 0; i--) + fw += fq[i]; + y[0] = (ih == 0) ? fw : -fw; + fw = fq[0] - fw; + for (i = 1; i <= jz; i++) + fw += fq[i]; + y[1] = (ih == 0) ? fw : -fw; + + return n & 7; +} diff --git a/sysdeps/powerpc/fpu/k_cosf.c b/sysdeps/powerpc/fpu/k_cosf.c new file mode 100644 index 0000000..cb0e34e --- /dev/null +++ b/sysdeps/powerpc/fpu/k_cosf.c @@ -0,0 +1,65 @@ +/* k_cosf.c -- float version of k_cos.c + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + 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, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include "math.h" +#include "math_private.h" + +static const float twom27 = 7.4505806e-09; +static const float dot3 = 3.0000001e-01; +static const float dot78125 = 7.8125000e-01; + +static const float one = 1.0000000000e+00; +static const float C1 = 4.1666667908e-02; +static const float C2 = -1.3888889225e-03; +static const float C3 = 2.4801587642e-05; +static const float C4 = -2.7557314297e-07; +static const float C5 = 2.0875723372e-09; +static const float C6 = -1.1359647598e-11; + +float +__kernel_cosf (float x, float y) +{ + float a, hz, z, r, qx; + float ix; + ix = __builtin_fabsf (x); + if (ix < twom27) + { /* |x| < 2**-27 */ + if (x == 0.0) + return one; + } + z = x * x; + r = z * (C1 + z * (C2 + z * (C3 + z * (C4 + z * (C5 + z * C6))))); + if (ix < dot3) /* if |x| < 0.3 */ + return one - ((float) 0.5 * z - (z * r - x * y)); + else + { + if (ix > dot78125) + { /* x > 0.78125 */ + qx = (float) 0.28125; + } + else + { + qx = ix / 4.0; + } + hz = (float) 0.5 *z - qx; + a = one - qx; + return a - (hz - (z * r - x * y)); + } +} diff --git a/sysdeps/powerpc/fpu/k_sinf.c b/sysdeps/powerpc/fpu/k_sinf.c new file mode 100644 index 0000000..f93a478 --- /dev/null +++ b/sysdeps/powerpc/fpu/k_sinf.c @@ -0,0 +1,53 @@ +/* k_sinf.c -- float version of k_sin.c + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + 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, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include "math.h" +#include "math_private.h" + + +static const float twom27 = 7.4505806000e-09; +static const float half = 5.0000000000e-01; +static const float S1 = -1.6666667163e-01; +static const float S2 = 8.3333337680e-03; +static const float S3 = -1.9841270114e-04; +static const float S4 = 2.7557314297e-06; +static const float S5 = -2.5050759689e-08; +static const float S6 = 1.5896910177e-10; + + +float +__kernel_sinf (float x, float y, int iy) +{ + float z, r, v; + float ix; + ix = __builtin_fabsf (x); + if (ix < twom27) + { /* |x| < 2**-27 */ + if (x == 0.0) + return x; + } + z = x * x; + v = z * x; + r = S2 + z * (S3 + z * (S4 + z * (S5 + z * S6))); + if (iy == 0) + return x + v * (S1 + z * r); + else + return x - ((z * (half * y - v * r) - y) - v * S1); +} diff --git a/sysdeps/powerpc/fpu/s_cosf.c b/sysdeps/powerpc/fpu/s_cosf.c new file mode 100644 index 0000000..8b8778a --- /dev/null +++ b/sysdeps/powerpc/fpu/s_cosf.c @@ -0,0 +1,71 @@ +/* s_cosf.c -- float version of s_cos.c. + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + 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, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <errno.h> +#include "math.h" +#include "math_private.h" + +static const float one = 1.0; +static const float pio4 = 7.8539801e-1; + +float +__cosf (float x) +{ + float y[2], z = 0.0; + float ix; + int32_t n; + + ix = __builtin_fabsf (x); + + /* |x| ~< pi/4 */ + if (ix <= pio4) + { + return __kernel_cosf (x, z); + /* cos(Inf or NaN) is NaN */ + } + else if (isnanf (ix)) + { + return x - x; + } + else if (isinff (ix)) + { + __set_errno (EDOM); + return x - x; + } + + /* argument reduction needed */ + else + { + n = __ieee754_rem_pio2f (x, y); + switch (n & 3) + { + case 0: + return __kernel_cosf (y[0], y[1]); + case 1: + return -__kernel_sinf (y[0], y[1], 1); + case 2: + return -__kernel_cosf (y[0], y[1]); + default: + return __kernel_sinf (y[0], y[1], 1); + } + } +} + +weak_alias (__cosf, cosf) diff --git a/sysdeps/powerpc/fpu/s_float_bitwise.h b/sysdeps/powerpc/fpu/s_float_bitwise.h new file mode 100644 index 0000000..25892ee --- /dev/null +++ b/sysdeps/powerpc/fpu/s_float_bitwise.h @@ -0,0 +1,35 @@ +/* Bitwise manipulation over float. Function prototypes. + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 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 + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, write to the Free + Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA + 02111-1307 USA. */ + +#ifndef _FLOAT_BITWISE_ +#define _FLOAT_BITWISE_ 1 + +/* Returns (num & 0x7FFFFFF0 == value) */ +int __float_and_test28 (float num, float value); +/* Returns (num & 0x7FFFFF00 == value) */ +int __float_and_test24 (float num, float value); +/* Returns (num & 0x7F800000) */ +float __float_and8 (float num); +/* Returns ((int32_t)(num & 0x7F800000) >> 23) */ +int32_t __float_get_exp (float num); +/* Returns ((num & 0x807FFFFF) | exp) */ +float __float_set_exp (float num, float exp); + +#endif /* s_float_bitwise.h */ diff --git a/sysdeps/powerpc/fpu/s_scalbnf.c b/sysdeps/powerpc/fpu/s_scalbnf.c new file mode 100644 index 0000000..b49dd07 --- /dev/null +++ b/sysdeps/powerpc/fpu/s_scalbnf.c @@ -0,0 +1,135 @@ +/* s_scalbnf.c -- float version of s_scalbn.c. + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + 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, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include "math.h" +#include "math_private.h" + +#include "s_float_bitwise.h" + +static const float huge = 1.0e+30; +static const float tiny = 1.0e-30; +static const float two25 = 3.35544320e+07; +static const float twom25 = 2.98023224e-08; + +/* (float)(i << 23), 0<=i<255 - [ 2**0, 2**-126, ... , 2**127 ] */ +static const float pow2[] = { + 0.00000000e+00, 1.17549435e-38, 2.35098870e-38, 4.70197740e-38, + 9.40395481e-38, 1.88079096e-37, 3.76158192e-37, 7.52316385e-37, + 1.50463277e-36, 3.00926554e-36, 6.01853108e-36, 1.20370622e-35, + 2.40741243e-35, 4.81482486e-35, 9.62964972e-35, 1.92592994e-34, + 3.85185989e-34, 7.70371978e-34, 1.54074396e-33, 3.08148791e-33, + 6.16297582e-33, 1.23259516e-32, 2.46519033e-32, 4.93038066e-32, + 9.86076132e-32, 1.97215226e-31, 3.94430453e-31, 7.88860905e-31, + 1.57772181e-30, 3.15544362e-30, 6.31088724e-30, 1.26217745e-29, + 2.52435490e-29, 5.04870979e-29, 1.00974196e-28, 2.01948392e-28, + 4.03896783e-28, 8.07793567e-28, 1.61558713e-27, 3.23117427e-27, + 6.46234854e-27, 1.29246971e-26, 2.58493941e-26, 5.16987883e-26, + 1.03397577e-25, 2.06795153e-25, 4.13590306e-25, 8.27180613e-25, + 1.65436123e-24, 3.30872245e-24, 6.61744490e-24, 1.32348898e-23, + 2.64697796e-23, 5.29395592e-23, 1.05879118e-22, 2.11758237e-22, + 4.23516474e-22, 8.47032947e-22, 1.69406589e-21, 3.38813179e-21, + 6.77626358e-21, 1.35525272e-20, 2.71050543e-20, 5.42101086e-20, + 1.08420217e-19, 2.16840434e-19, 4.33680869e-19, 8.67361738e-19, + 1.73472348e-18, 3.46944695e-18, 6.93889390e-18, 1.38777878e-17, + 2.77555756e-17, 5.55111512e-17, 1.11022302e-16, 2.22044605e-16, + 4.44089210e-16, 8.88178420e-16, 1.77635684e-15, 3.55271368e-15, + 7.10542736e-15, 1.42108547e-14, 2.84217094e-14, 5.68434189e-14, + 1.13686838e-13, 2.27373675e-13, 4.54747351e-13, 9.09494702e-13, + 1.81898940e-12, 3.63797881e-12, 7.27595761e-12, 1.45519152e-11, + 2.91038305e-11, 5.82076609e-11, 1.16415322e-10, 2.32830644e-10, + 4.65661287e-10, 9.31322575e-10, 1.86264515e-09, 3.72529030e-09, + 7.45058060e-09, 1.49011612e-08, 2.98023224e-08, 5.96046448e-08, + 1.19209290e-07, 2.38418579e-07, 4.76837158e-07, 9.53674316e-07, + 1.90734863e-06, 3.81469727e-06, 7.62939453e-06, 1.52587891e-05, + 3.05175781e-05, 6.10351562e-05, 1.22070312e-04, 2.44140625e-04, + 4.88281250e-04, 9.76562500e-04, 1.95312500e-03, 3.90625000e-03, + 7.81250000e-03, 1.56250000e-02, 3.12500000e-02, 6.25000000e-02, + 1.25000000e-01, 2.50000000e-01, 5.00000000e-01, 1.00000000e+00, + 2.00000000e+00, 4.00000000e+00, 8.00000000e+00, 1.60000000e+01, + 3.20000000e+01, 6.40000000e+01, 1.28000000e+02, 2.56000000e+02, + 5.12000000e+02, 1.02400000e+03, 2.04800000e+03, 4.09600000e+03, + 8.19200000e+03, 1.63840000e+04, 3.27680000e+04, 6.55360000e+04, + 1.31072000e+05, 2.62144000e+05, 5.24288000e+05, 1.04857600e+06, + 2.09715200e+06, 4.19430400e+06, 8.38860800e+06, 1.67772160e+07, + 3.35544320e+07, 6.71088640e+07, 1.34217728e+08, 2.68435456e+08, + 5.36870912e+08, 1.07374182e+09, 2.14748365e+09, 4.29496730e+09, + 8.58993459e+09, 1.71798692e+10, 3.43597384e+10, 6.87194767e+10, + 1.37438953e+11, 2.74877907e+11, 5.49755814e+11, 1.09951163e+12, + 2.19902326e+12, 4.39804651e+12, 8.79609302e+12, 1.75921860e+13, + 3.51843721e+13, 7.03687442e+13, 1.40737488e+14, 2.81474977e+14, + 5.62949953e+14, 1.12589991e+15, 2.25179981e+15, 4.50359963e+15, + 9.00719925e+15, 1.80143985e+16, 3.60287970e+16, 7.20575940e+16, + 1.44115188e+17, 2.88230376e+17, 5.76460752e+17, 1.15292150e+18, + 2.30584301e+18, 4.61168602e+18, 9.22337204e+18, 1.84467441e+19, + 3.68934881e+19, 7.37869763e+19, 1.47573953e+20, 2.95147905e+20, + 5.90295810e+20, 1.18059162e+21, 2.36118324e+21, 4.72236648e+21, + 9.44473297e+21, 1.88894659e+22, 3.77789319e+22, 7.55578637e+22, + 1.51115727e+23, 3.02231455e+23, 6.04462910e+23, 1.20892582e+24, + 2.41785164e+24, 4.83570328e+24, 9.67140656e+24, 1.93428131e+25, + 3.86856262e+25, 7.73712525e+25, 1.54742505e+26, 3.09485010e+26, + 6.18970020e+26, 1.23794004e+27, 2.47588008e+27, 4.95176016e+27, + 9.90352031e+27, 1.98070406e+28, 3.96140813e+28, 7.92281625e+28, + 1.58456325e+29, 3.16912650e+29, 6.33825300e+29, 1.26765060e+30, + 2.53530120e+30, 5.07060240e+30, 1.01412048e+31, 2.02824096e+31, + 4.05648192e+31, 8.11296384e+31, 1.62259277e+32, 3.24518554e+32, + 6.49037107e+32, 1.29807421e+33, 2.59614843e+33, 5.19229686e+33, + 1.03845937e+34, 2.07691874e+34, 4.15383749e+34, 8.30767497e+34, + 1.66153499e+35, 3.32306999e+35, 6.64613998e+35, 1.32922800e+36, + 2.65845599e+36, 5.31691198e+36, 1.06338240e+37, 2.12676479e+37, + 4.25352959e+37, 8.50705917e+37, 1.70141183e+38, +}; + + +float +__scalbnf (float x, int n) +{ + float k; + int32_t ke; + + k = __float_and8 (x); + if (k == 0.0) + { + if (x == 0.0) + return x; + x *= two25; + k = __float_and8 (x) / two25; + } + if (isnanf (k) || isinff (k)) + return x + x; + ke = __float_get_exp (k) + n; + if (n > 50000 || ke > 0xfe) + return huge * __copysignf (huge, x); + if (n < -50000) + return tiny * __copysignf (tiny, x); + /* ke <= 254 */ + if (ke > 0) + { + x = __float_set_exp (x, pow2[ke]); + return x; + } + if (ke <= -25) + return tiny * __copysignf (tiny, x); + /* 0 <= ke <= 254 */ + k *= two25; + ke += 25; + x = __float_set_exp (x, pow2[ke]); + return x * twom25; +} +weak_alias (__scalbnf, scalbnf) diff --git a/sysdeps/powerpc/fpu/s_sinf.c b/sysdeps/powerpc/fpu/s_sinf.c new file mode 100644 index 0000000..dd724aa --- /dev/null +++ b/sysdeps/powerpc/fpu/s_sinf.c @@ -0,0 +1,70 @@ +/* s_sinf.c -- float version of s_sin.c. + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + 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, + write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. */ + +#include <errno.h> +#include "math.h" +#include "math_private.h" + +static const float pio4 = 7.8539801e-1; + +float +__sinf (float x) +{ + float y[2], z = 0.0; + float ix; + int32_t n; + + ix = __builtin_fabsf (x); + + /* |x| ~< pi/4 */ + if (ix <= pio4) + { + return __kernel_sinf (x, z, 0); + /* sin(Inf or NaN) is NaN */ + } + else if (isnanf (ix)) + { + return x - x; + } + else if (isinff (ix)) + { + __set_errno (EDOM); + return x - x; + } + + /* argument reduction needed */ + else + { + n = __ieee754_rem_pio2f (x, y); + switch (n & 3) + { + case 0: + return __kernel_sinf (y[0], y[1], 1); + case 1: + return __kernel_cosf (y[0], y[1]); + case 2: + return -__kernel_sinf (y[0], y[1], 1); + default: + return -__kernel_cosf (y[0], y[1]); + } + } +} + +weak_alias (__sinf, sinf) diff --git a/sysdeps/powerpc/powerpc32/fpu/s_float_bitwise.S b/sysdeps/powerpc/powerpc32/fpu/s_float_bitwise.S new file mode 100644 index 0000000..2a7b64e --- /dev/null +++ b/sysdeps/powerpc/powerpc32/fpu/s_float_bitwise.S @@ -0,0 +1,96 @@ +/* Bitwise manipulation over float. PowerPC32 version. + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 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 + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, write to the Free + Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA + 02111-1307 USA. */ + +#include <sysdep.h> + +/* int [r3] __float_and_test28(float [f1] num, float [f2] value) + Returns (num & 0xFFFFFFF0 == value) */ +ENTRY(__float_and_test28) + stwu r1,-32(r1) + cfi_adjust_cfa_offset(32) + stfs fp1,16(r1) + stfs fp2,8(r1) + lwz r0,16(r1) + rlwinm r3,r0,0,0,27 + lwz r0,8(r1) + xor r3,r3,r0 + cntlzw r3,r3 + srwi r3,r3,5 + extsw r3,r3 + addi r1,r1,32 + blr +END(__float_and_test28) + +/* int [r3] __float_and_test24(float [f1] num, float [f2] value) + Returns (num & 0xFFFFFF00 == value) */ +ENTRY(__float_and_test24) + stwu r1,-32(r1) + cfi_adjust_cfa_offset(32) + stfs fp1,16(r1) + stfs fp2,8(r1) + lwz r0,16(r1) + rlwinm r3,r0,0,0,23 + lwz r0,8(r1) + xor r3,r3,r0 + cntlzw r3,r3 + srwi r3,r3,5 + extsw r3,r3 + addi r1,r1,32 + blr +END(__float_and_test24) + +/* float [f1] __float_and8(float [f1] num) + Returns (num & 0x7F800000) */ +ENTRY(__float_and8) + stfs fp1,-16(r1) + lwz r0,-16(r1) + rlwinm r0,r0,0,1,8 + stw r0,-16(r1) + lfs fp1,-16(r1) + blr +END(__float_and8) + +/* int32_t [r2] __float_get_exp(float [f1] num) + Returns ((int32_t)(num & 0x7F800000) >> 23) */ +ENTRY(__float_get_exp) + stfs fp1,-16(r1) + lwz r0,-16(r1) + rlwinm r3,r0,0,1,8 + rldicl r3,r3,41,56 + blr +END(__float_get_exp) + +/* float [fp1] __float_set_exp(float [fp1] num, float [fp2] exp) + Returns ((num & 0x807FFFFF) | exp) */ +ENTRY(__float_set_exp) + stwu r1,-32(r1) + cfi_adjust_cfa_offset(32) + stfs fp1,16(r1) + stfs fp2,8(r1) + lwz r9,16(r1) + rlwinm r0,r9,1,8,31 + rlwinm r0,r0,31,0xffffffff + lwz r9,8(r1) + or r9,r0,r9 + stw r9,16(r1) + lfs fp1,16(r1) + addi r1,r1,32 + blr +END(__float_set_exp) diff --git a/sysdeps/powerpc/powerpc32/power7/fpu/s_float_bitwise.S b/sysdeps/powerpc/powerpc32/power7/fpu/s_float_bitwise.S new file mode 100644 index 0000000..02644d9 --- /dev/null +++ b/sysdeps/powerpc/powerpc32/power7/fpu/s_float_bitwise.S @@ -0,0 +1,149 @@ +/* Bitwise manipulation over float. PowerPC32 version using VSX. + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 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 + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License a4byte with the GNU C Library; if not, write to the Free + Software Foundation, Inc., 1 Franklin Street, Fifth Floor, Boston MA + 02110-1301 USA. */ + +#include <sysdep.h> + + .section .rodata.cst4,"aM",@progbits,8 + .align 2 +.LC0: /* 28-bits mask extended to double */ + .4byte 0x7ffffffe + .4byte 0x00000000 +.LC1: /* 24-bits mask extended to double */ + .4byte 0x7fffffe0 + .4byte 0x00000000 +.LC2: /* 9-bits mask extended to double */ + .4byte 0x7ff00000 + .4byte 0x00000000 +.LC3: /* mask used to get the float exp */ + .4byte 0x800FFFFF + .4byte 0xFFFFFFFF + + .section ".text" + .machine power7 +/* int [r3] __float_and_test28(float [f1] num, float [f2] value) + Returns (num & 0x7FFFFFF0 == value) */ +ENTRY(__float_and_test28) +#ifdef SHARED + mflr r11 + cfi_register(lr,r11) + bcl 20,31,1f +1: mflr r9 + addis r9,r9,.LC0-1b@ha + lfd fp3,.LC0-1b@l(r9) + mtlr r11 + cfi_same_value (lr) +#else + lis r9,.LC0@ha + lfd fp3,.LC0@l(r9) +#endif + xxland v4,v1,v3 + fcmpu cr7,fp4,fp2 + mfcr r3,cr1 + rlwinm r3,r3,31,1 + blr +END(__float_and_test28) + +/* int [r3] __float_and_test24(float [f1] num, float [f2] value) + Returns (num & 0x7FFFFF00 == value) */ +ENTRY(__float_and_test24) +#ifdef SHARED + mflr r11 + cfi_register(lr,r11) + bcl 20,31,1f +1: mflr r9 + addis r9,r9,.LC1-1b@ha + lfd fp3,.LC1-1b@l(r9) + mtlr r11 + cfi_same_value (lr) +#else + lis r9,.LC1@ha + lfd fp3,.LC1@l(r9) +#endif + xxland v4,v1,v3 + fcmpu cr7,fp4,fp2 + mfcr r3,cr1 + rlwinm r3,r3,31,1 + blr +END(__float_and_test24) + +/* float [f1] __float_and8(float [f1] num) + Returns (num & 0x7F800000) */ +EALIGN(__float_and8, 4, 1) +#ifdef SHARED + mflr r11 + cfi_register(lr,r11) + bcl 20,31,1f +1: mflr r9 + addis r9,r9,.LC2-1b@ha + lfd fp3,.LC2-1b@l(r9) + mtlr r11 + cfi_same_value (lr) +#else + lis r9,.LC2@ha + lfd fp3,.LC2@l(r9) +#endif + xxland v1,v1,v3 + blr +END(__float_and8) + +/* int32_t [r2] __float_get_exp(float [f1] num + Returns ((int32_t)(num & 0x7F800000) >> 23) */ +ENTRY(__float_get_exp) +#ifdef SHARED + mflr r11 + cfi_register(lr,r11) + bcl 20,31,1f +1: mflr r9 + addis r9,r9,.LC2-1b@ha + lfd fp3,.LC2-1b@l(r9) + mtlr r11 + cfi_same_value (lr) +#else + lis r9,.LC2@ha + lfd fp3,.LC2@l(r9) +#endif + xxland v4,v1,v3 + stfs fp4,-16(r1) + lwz r3,-16(r1) + srawi r3,r3,23 + clrldi r3,r3,32 + blr +END(__float_get_exp) + +/* float [fp1] __float_set_exp(float [fp1] num, float [fp2] exp) + Returns ((num & 0x807FFFFF) | exp) */ +EALIGN(__float_set_exp, 4, 1) +#ifdef SHARED + mflr r11 + cfi_register(lr,r11) + bcl 20,31,1f +1: mflr r9 + addis r9,r9,.LC3-1b@ha + lfd fp3,.LC3-1b@l(r9) + mtlr r11 + cfi_same_value (lr) +#else + lis r9,.LC3@ha + lfd fp3,.LC3@l(r9) +#endif + xxland v4,v1,v3 + xxlor v1,v4,v2 + blr +END(__float_set_exp) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_float_bitwise.S b/sysdeps/powerpc/powerpc64/fpu/s_float_bitwise.S new file mode 100644 index 0000000..6cea572 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/fpu/s_float_bitwise.S @@ -0,0 +1,87 @@ +/* Bitwise manipulation over float. PowerPC64 version using VSX. + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 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 + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, write to the Free + Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA + 02111-1307 USA. */ + +#include <sysdep.h> + +/* int [r3] __float_and_test28(float [f1] num, float [f2] value) + Returns (num & 0x7FFFFFF0 == value) */ +ENTRY(__float_and_test28) + stfs fp1,-32(r1) + stfs fp2,-16(r1) + lwz r0,-32(r1) + rlwinm r3,r0,0,0,27 + lwz r0,-16(r1) + xor r3,r3,r0 + cntlzw r3,r3 + srwi r3,r3,5 + extsw r3,r3 + blr +END(__float_and_test28) + +/* int [r3] __float_and_test24(float [f1] num, float [f2] value) + Returns (num & 0x7FFFFF00 == value) */ +ENTRY(__float_and_test24) + stfs fp1,-32(r1) + stfs fp2,-16(r1) + lwz r0,-32(r1) + rlwinm r3,r0,0,0,23 + lwz r0,-16(r1) + xor r3,r3,r0 + cntlzw r3,r3 + srwi r3,r3,5 + extsw r3,r3 + blr +END(__float_and_test24) + +/* float [f1] __float_and8(float [f1] num) + Returns (num & 0x7F800000) */ +ENTRY(__float_and8) + stfs fp1,-16(r1) + lwz r0,-16(r1) + rlwinm r0,r0,0,1,8 + stw r0,-16(r1) + lfs fp1,-16(r1) + blr +END(__float_and8) + +/* int32_t [r2] __float_get_exp(float [f1] num) + Returns ((int32_t)(num & 0x7F800000) >> 23) */ +ENTRY(__float_get_exp) + stfs fp1,-16(r1) + lwz r0,-16(r1) + rlwinm r3,r0,0,1,8 + rldicl r3,r3,41,56 + blr +END(__float_get_exp) + +/* float [fp1] __float_set_exp(float [fp1] num, float [fp2] exp) + Returns ((num & 0x807FFFFF) | exp) */ +ENTRY(__float_set_exp) + stfs fp1,-32(r1) + stfs fp2,-16(r1) + lwz r9,-32(r1) + rlwinm r0,r9,1,8,31 + rlwinm r0,r0,31,0xffffffff + lwz r9,-16(r1) + or r9,r0,r9 + stw r9,-16(r1) + lfs fp1,-16(r1) + blr +END(__float_set_exp) diff --git a/sysdeps/powerpc/powerpc64/power7/fpu/s_float_bitwise.S b/sysdeps/powerpc/powerpc64/power7/fpu/s_float_bitwise.S new file mode 100644 index 0000000..fb3e9f7 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/power7/fpu/s_float_bitwise.S @@ -0,0 +1,83 @@ +/* Bitwise manipulation over float. PowerPC64 version using VSX. + Copyright (C) 2011 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Adhemerval Zanella <azanella@br.ibm.com>, 2011 + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 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 + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, write to the Free + Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA + 02111-1307 USA. */ + +#include <sysdep.h> + + .section ".toc","aw" +.LC0: + .tc FS_28MASK_0[TC],0x7ffffffe00000000 /* 28-bits mask */ +.LC1: + .tc FS_24MASK_0[TC],0x7fffffe000000000 /* 24-bits mask */ +.LC2: + .tc FL_09MASK_0[TC],0x7ff0000000000000 /* 9-bits mask */ +.LC3: + .tc FL_EXP_MASK[TC],0x800FFFFFFFFFFFFF /* float exponent mask */ + .section ".text" + .machine power7 +/* int [r3] __float_and_test28(float [f1] num, float [f2] value) + Returns (num & 0x7FFFFF00 == value) */ +ENTRY(__float_and_test28) + lfd fp3,.LC0@toc(2) + xxland v4,v1,v3 + fcmpu cr7,fp4,fp2 + mfcr r3,cr1 + rlwinm r3,r3,31,1 + blr +END(__float_and_test28) + +/* int [r3] __float_and_test24(float [f1] num, float [f2] value) + Returns (num & 0x7FFFFFF0 == value) */ +ENTRY(__float_and_test24) + lfd fp3,.LC1@toc(2) + xxland v4,v1,v3 + fcmpu cr7,fp4,fp2 + mfcr r3,cr1 + rlwinm r3,r3,31,1 + blr +END(__float_and_test24) + +/* float [f1] __float_and8(float [f1] num) + Returns (num & 0x7F800000) */ +EALIGN(__float_and8, 4, 2) + lfd fp3,.LC2@toc(2) + xxland v1,v1,v3 + blr +END(__float_and8) + +/* int32_t [r2] __float_get_exp(float [f1] num) + Returns ((int32_t)(num & 0x7F800000) >> 23) */ +ENTRY(__float_get_exp) + lfd fp3,.LC2@toc(2) + xxland v4,v1,v3 + stfs fp4,-16(r1) + lwz r3,-16(r1) + srawi r3,r3,23 + clrldi r3,r3,32 + blr +END(__float_get_exp) + +/* float [fp1] __float_set_exp(float [fp1] num, float [fp2] exp) + Returns ((num & 0x807FFFFF) | exp) */ +EALIGN(__float_set_exp, 4, 1) + lfd fp3,.LC3@toc(2) + xxland v4,v1,v3 + xxlor v1,v4,v2 + blr +END(__float_set_exp) |