libs/newlib-cygwin
libs
/
newlib-cygwin
mirror of git://sourceware.org/git/newlib-cygwin.git
4
0
Fork 0
Code Issues Projects Releases Wiki Activity
newlib-cygwin/newlib/libm/common/log2_data.c

235 lines
9.4 KiB
C
Raw Normal View History

New log2 implementation The new implementation is provided under !__OBSOLETE_MATH, it uses ISO C99 code. With default settings the worst case error in nearest rounding mode is 0.547 ULP with inlined fma and fma contraction. It uses a 1 KB lookup table, on aarch64 .text+.rodata size of libm.a is increased by 1584 bytes. Note that the math.h header defines log2(x) to be log(x)/Ln2, this is not changed, so the new code is only used if that macro is suppressed. Improvements on Cortex-A72: latency: 2.0x thruput: 2.2x
2018-06-26 23:06:54 +08:00
/* Data for log2.
Copyright (c) 2018 Arm Ltd. All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
2. 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.
3. The name of the company may not be used to endorse or promote
products derived from this software without specific prior written
permission.
Update Arm copyright notices in new math files While working on the strstr patch I noticed several copyright headers of the new math functions are missing closing quotes after ``AS IS. I've added these. Also update spellings of Arm Ltd in a few places (but still use ARM LTD in upper case portion). Finally add SPDX identifiers to make everything consistent.
2018-09-28 01:03:32 +08:00
THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
New log2 implementation The new implementation is provided under !__OBSOLETE_MATH, it uses ISO C99 code. With default settings the worst case error in nearest rounding mode is 0.547 ULP with inlined fma and fma contraction. It uses a 1 KB lookup table, on aarch64 .text+.rodata size of libm.a is increased by 1584 bytes. Note that the math.h header defines log2(x) to be log(x)/Ln2, this is not changed, so the new code is only used if that macro is suppressed. Improvements on Cortex-A72: latency: 2.0x thruput: 2.2x
2018-06-26 23:06:54 +08:00
WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
Update Arm copyright notices in new math files While working on the strstr patch I noticed several copyright headers of the new math functions are missing closing quotes after ``AS IS. I've added these. Also update spellings of Arm Ltd in a few places (but still use ARM LTD in upper case portion). Finally add SPDX identifiers to make everything consistent.
2018-09-28 01:03:32 +08:00
IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
New log2 implementation The new implementation is provided under !__OBSOLETE_MATH, it uses ISO C99 code. With default settings the worst case error in nearest rounding mode is 0.547 ULP with inlined fma and fma contraction. It uses a 1 KB lookup table, on aarch64 .text+.rodata size of libm.a is increased by 1584 bytes. Note that the math.h header defines log2(x) to be log(x)/Ln2, this is not changed, so the new code is only used if that macro is suppressed. Improvements on Cortex-A72: latency: 2.0x thruput: 2.2x
2018-06-26 23:06:54 +08:00
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 "fdlibm.h"
#if !__OBSOLETE_MATH
#include "math_config.h"
#define N (1 << LOG2_TABLE_BITS)
const struct log2_data __log2_data = {
// First coefficient: 0x1.71547652b82fe1777d0ffda0d24p0
.invln2hi = 0x1.7154765200000p+0,
.invln2lo = 0x1.705fc2eefa200p-33,
.poly1 = {
#if LOG2_POLY1_ORDER == 11
// relative error: 0x1.2fad8188p-63
// in -0x1.5b51p-5 0x1.6ab2p-5
-0x1.71547652b82fep-1,
0x1.ec709dc3a03f7p-2,
-0x1.71547652b7c3fp-2,
0x1.2776c50f05be4p-2,
-0x1.ec709dd768fe5p-3,
0x1.a61761ec4e736p-3,
-0x1.7153fbc64a79bp-3,
0x1.484d154f01b4ap-3,
-0x1.289e4a72c383cp-3,
0x1.0b32f285aee66p-3,
#endif
},
.poly = {
#if N == 64 && LOG2_POLY_ORDER == 7
// relative error: 0x1.a72c2bf8p-58
// abs error: 0x1.67a552c8p-66
// in -0x1.f45p-8 0x1.f45p-8
-0x1.71547652b8339p-1,
0x1.ec709dc3a04bep-2,
-0x1.7154764702ffbp-2,
0x1.2776c50034c48p-2,
-0x1.ec7b328ea92bcp-3,
0x1.a6225e117f92ep-3,
#endif
},
Document the log table generation method Add comments with enough detail so the log lookup tables can be recreated.
2018-09-05 19:15:55 +08:00
/* Algorithm:
x = 2^k z
log2(x) = k + log2(c) + log2(z/c)
log2(z/c) = poly(z/c - 1)
where z is in [1.6p-1; 1.6p0] which is split into N subintervals and z falls
into the ith one, then table entries are computed as
tab[i].invc = 1/c
tab[i].logc = (double)log2(c)
tab2[i].chi = (double)c
tab2[i].clo = (double)(c - (double)c)
where c is near the center of the subinterval and is chosen by trying +-2^29
floating point invc candidates around 1/center and selecting one for which
1) the rounding error in 0x1.8p10 + logc is 0,
2) the rounding error in z - chi - clo is < 0x1p-64 and
3) the rounding error in (double)log2(c) is minimized (< 0x1p-68).
Note: 1) ensures that k + logc can be computed without rounding error, 2)
ensures that z/c - 1 can be computed as (z - chi - clo)*invc with close to a
single rounding error when there is no fast fma for z*invc - 1, 3) ensures
that logc + poly(z/c - 1) has small error, however near x == 1 when
|log2(x)| < 0x1p-4, this is not enough so that is special cased. */
New log2 implementation The new implementation is provided under !__OBSOLETE_MATH, it uses ISO C99 code. With default settings the worst case error in nearest rounding mode is 0.547 ULP with inlined fma and fma contraction. It uses a 1 KB lookup table, on aarch64 .text+.rodata size of libm.a is increased by 1584 bytes. Note that the math.h header defines log2(x) to be log(x)/Ln2, this is not changed, so the new code is only used if that macro is suppressed. Improvements on Cortex-A72: latency: 2.0x thruput: 2.2x
2018-06-26 23:06:54 +08:00
.tab = {
#if N == 64
{0x1.724286bb1acf8p+0, -0x1.1095feecdb000p-1},
{0x1.6e1f766d2cca1p+0, -0x1.08494bd76d000p-1},
{0x1.6a13d0e30d48ap+0, -0x1.00143aee8f800p-1},
{0x1.661ec32d06c85p+0, -0x1.efec5360b4000p-2},
{0x1.623fa951198f8p+0, -0x1.dfdd91ab7e000p-2},
{0x1.5e75ba4cf026cp+0, -0x1.cffae0cc79000p-2},
{0x1.5ac055a214fb8p+0, -0x1.c043811fda000p-2},
{0x1.571ed0f166e1ep+0, -0x1.b0b67323ae000p-2},
{0x1.53909590bf835p+0, -0x1.a152f5a2db000p-2},
{0x1.5014fed61adddp+0, -0x1.9217f5af86000p-2},
{0x1.4cab88e487bd0p+0, -0x1.8304db0719000p-2},
{0x1.49539b4334feep+0, -0x1.74189f9a9e000p-2},
{0x1.460cbdfafd569p+0, -0x1.6552bb5199000p-2},
{0x1.42d664ee4b953p+0, -0x1.56b23a29b1000p-2},
{0x1.3fb01111dd8a6p+0, -0x1.483650f5fa000p-2},
{0x1.3c995b70c5836p+0, -0x1.39de937f6a000p-2},
{0x1.3991c4ab6fd4ap+0, -0x1.2baa1538d6000p-2},
{0x1.3698e0ce099b5p+0, -0x1.1d98340ca4000p-2},
{0x1.33ae48213e7b2p+0, -0x1.0fa853a40e000p-2},
{0x1.30d191985bdb1p+0, -0x1.01d9c32e73000p-2},
{0x1.2e025cab271d7p+0, -0x1.e857da2fa6000p-3},
{0x1.2b404cf13cd82p+0, -0x1.cd3c8633d8000p-3},
{0x1.288b02c7ccb50p+0, -0x1.b26034c14a000p-3},
{0x1.25e2263944de5p+0, -0x1.97c1c2f4fe000p-3},
{0x1.234563d8615b1p+0, -0x1.7d6023f800000p-3},
{0x1.20b46e33eaf38p+0, -0x1.633a71a05e000p-3},
{0x1.1e2eefdcda3ddp+0, -0x1.494f5e9570000p-3},
{0x1.1bb4a580b3930p+0, -0x1.2f9e424e0a000p-3},
{0x1.19453847f2200p+0, -0x1.162595afdc000p-3},
{0x1.16e06c0d5d73cp+0, -0x1.f9c9a75bd8000p-4},
{0x1.1485f47b7e4c2p+0, -0x1.c7b575bf9c000p-4},
{0x1.12358ad0085d1p+0, -0x1.960c60ff48000p-4},
{0x1.0fef00f532227p+0, -0x1.64ce247b60000p-4},
{0x1.0db2077d03a8fp+0, -0x1.33f78b2014000p-4},
{0x1.0b7e6d65980d9p+0, -0x1.0387d1a42c000p-4},
{0x1.0953efe7b408dp+0, -0x1.a6f9208b50000p-5},
{0x1.07325cac53b83p+0, -0x1.47a954f770000p-5},
{0x1.05197e40d1b5cp+0, -0x1.d23a8c50c0000p-6},
{0x1.03091c1208ea2p+0, -0x1.16a2629780000p-6},
{0x1.0101025b37e21p+0, -0x1.720f8d8e80000p-8},
{0x1.fc07ef9caa76bp-1, 0x1.6fe53b1500000p-7},
{0x1.f4465d3f6f184p-1, 0x1.11ccce10f8000p-5},
{0x1.ecc079f84107fp-1, 0x1.c4dfc8c8b8000p-5},
{0x1.e573a99975ae8p-1, 0x1.3aa321e574000p-4},
{0x1.de5d6f0bd3de6p-1, 0x1.918a0d08b8000p-4},
{0x1.d77b681ff38b3p-1, 0x1.e72e9da044000p-4},
{0x1.d0cb5724de943p-1, 0x1.1dcd2507f6000p-3},
{0x1.ca4b2dc0e7563p-1, 0x1.476ab03dea000p-3},
{0x1.c3f8ee8d6cb51p-1, 0x1.7074377e22000p-3},
{0x1.bdd2b4f020c4cp-1, 0x1.98ede8ba94000p-3},
{0x1.b7d6c006015cap-1, 0x1.c0db86ad2e000p-3},
{0x1.b20366e2e338fp-1, 0x1.e840aafcee000p-3},
{0x1.ac57026295039p-1, 0x1.0790ab4678000p-2},
{0x1.a6d01bc2731ddp-1, 0x1.1ac056801c000p-2},
{0x1.a16d3bc3ff18bp-1, 0x1.2db11d4fee000p-2},
{0x1.9c2d14967feadp-1, 0x1.406464ec58000p-2},
{0x1.970e4f47c9902p-1, 0x1.52dbe093af000p-2},
{0x1.920fb3982bcf2p-1, 0x1.651902050d000p-2},
{0x1.8d30187f759f1p-1, 0x1.771d2cdeaf000p-2},
{0x1.886e5ebb9f66dp-1, 0x1.88e9c857d9000p-2},
{0x1.83c97b658b994p-1, 0x1.9a80155e16000p-2},
{0x1.7f405ffc61022p-1, 0x1.abe186ed3d000p-2},
{0x1.7ad22181415cap-1, 0x1.bd0f2aea0e000p-2},
{0x1.767dcf99eff8cp-1, 0x1.ce0a43dbf4000p-2},
#endif
},
Move __HAVE_FAST_FMA to math_config.h Define it consistently with other HAVE_* macros that only affect code using math_config.h. This is also closer to the Arm Optimized Routines code.
2018-07-05 19:37:25 +08:00
#if !HAVE_FAST_FMA
New log2 implementation The new implementation is provided under !__OBSOLETE_MATH, it uses ISO C99 code. With default settings the worst case error in nearest rounding mode is 0.547 ULP with inlined fma and fma contraction. It uses a 1 KB lookup table, on aarch64 .text+.rodata size of libm.a is increased by 1584 bytes. Note that the math.h header defines log2(x) to be log(x)/Ln2, this is not changed, so the new code is only used if that macro is suppressed. Improvements on Cortex-A72: latency: 2.0x thruput: 2.2x
2018-06-26 23:06:54 +08:00
.tab2 = {
# if N == 64
{0x1.6200012b90a8ep-1, 0x1.904ab0644b605p-55},
{0x1.66000045734a6p-1, 0x1.1ff9bea62f7a9p-57},
{0x1.69fffc325f2c5p-1, 0x1.27ecfcb3c90bap-55},
{0x1.6e00038b95a04p-1, 0x1.8ff8856739326p-55},
{0x1.71fffe09994e3p-1, 0x1.afd40275f82b1p-55},
{0x1.7600015590e1p-1, -0x1.2fd75b4238341p-56},
{0x1.7a00012655bd5p-1, 0x1.808e67c242b76p-56},
{0x1.7e0003259e9a6p-1, -0x1.208e426f622b7p-57},
{0x1.81fffedb4b2d2p-1, -0x1.402461ea5c92fp-55},
{0x1.860002dfafcc3p-1, 0x1.df7f4a2f29a1fp-57},
{0x1.89ffff78c6b5p-1, -0x1.e0453094995fdp-55},
{0x1.8e00039671566p-1, -0x1.a04f3bec77b45p-55},
{0x1.91fffe2bf1745p-1, -0x1.7fa34400e203cp-56},
{0x1.95fffcc5c9fd1p-1, -0x1.6ff8005a0695dp-56},
{0x1.9a0003bba4767p-1, 0x1.0f8c4c4ec7e03p-56},
{0x1.9dfffe7b92da5p-1, 0x1.e7fd9478c4602p-55},
{0x1.a1fffd72efdafp-1, -0x1.a0c554dcdae7ep-57},
{0x1.a5fffde04ff95p-1, 0x1.67da98ce9b26bp-55},
{0x1.a9fffca5e8d2bp-1, -0x1.284c9b54c13dep-55},
{0x1.adfffddad03eap-1, 0x1.812c8ea602e3cp-58},
{0x1.b1ffff10d3d4dp-1, -0x1.efaddad27789cp-55},
{0x1.b5fffce21165ap-1, 0x1.3cb1719c61237p-58},
{0x1.b9fffd950e674p-1, 0x1.3f7d94194cep-56},
{0x1.be000139ca8afp-1, 0x1.50ac4215d9bcp-56},
{0x1.c20005b46df99p-1, 0x1.beea653e9c1c9p-57},
{0x1.c600040b9f7aep-1, -0x1.c079f274a70d6p-56},
{0x1.ca0006255fd8ap-1, -0x1.a0b4076e84c1fp-56},
{0x1.cdfffd94c095dp-1, 0x1.8f933f99ab5d7p-55},
{0x1.d1ffff975d6cfp-1, -0x1.82c08665fe1bep-58},
{0x1.d5fffa2561c93p-1, -0x1.b04289bd295f3p-56},
{0x1.d9fff9d228b0cp-1, 0x1.70251340fa236p-55},
{0x1.de00065bc7e16p-1, -0x1.5011e16a4d80cp-56},
{0x1.e200002f64791p-1, 0x1.9802f09ef62ep-55},
{0x1.e600057d7a6d8p-1, -0x1.e0b75580cf7fap-56},
{0x1.ea00027edc00cp-1, -0x1.c848309459811p-55},
{0x1.ee0006cf5cb7cp-1, -0x1.f8027951576f4p-55},
{0x1.f2000782b7dccp-1, -0x1.f81d97274538fp-55},
{0x1.f6000260c450ap-1, -0x1.071002727ffdcp-59},
{0x1.f9fffe88cd533p-1, -0x1.81bdce1fda8bp-58},
{0x1.fdfffd50f8689p-1, 0x1.7f91acb918e6ep-55},
{0x1.0200004292367p+0, 0x1.b7ff365324681p-54},
{0x1.05fffe3e3d668p+0, 0x1.6fa08ddae957bp-55},
{0x1.0a0000a85a757p+0, -0x1.7e2de80d3fb91p-58},
{0x1.0e0001a5f3fccp+0, -0x1.1823305c5f014p-54},
{0x1.11ffff8afbaf5p+0, -0x1.bfabb6680bac2p-55},
{0x1.15fffe54d91adp+0, -0x1.d7f121737e7efp-54},
{0x1.1a00011ac36e1p+0, 0x1.c000a0516f5ffp-54},
{0x1.1e00019c84248p+0, -0x1.082fbe4da5dap-54},
{0x1.220000ffe5e6ep+0, -0x1.8fdd04c9cfb43p-55},
{0x1.26000269fd891p+0, 0x1.cfe2a7994d182p-55},
{0x1.2a00029a6e6dap+0, -0x1.00273715e8bc5p-56},
{0x1.2dfffe0293e39p+0, 0x1.b7c39dab2a6f9p-54},
{0x1.31ffff7dcf082p+0, 0x1.df1336edc5254p-56},
{0x1.35ffff05a8b6p+0, -0x1.e03564ccd31ebp-54},
{0x1.3a0002e0eaeccp+0, 0x1.5f0e74bd3a477p-56},
{0x1.3e000043bb236p+0, 0x1.c7dcb149d8833p-54},
{0x1.4200002d187ffp+0, 0x1.e08afcf2d3d28p-56},
{0x1.460000d387cb1p+0, 0x1.20837856599a6p-55},
{0x1.4a00004569f89p+0, -0x1.9fa5c904fbcd2p-55},
{0x1.4e000043543f3p+0, -0x1.81125ed175329p-56},
{0x1.51fffcc027f0fp+0, 0x1.883d8847754dcp-54},
{0x1.55ffffd87b36fp+0, -0x1.709e731d02807p-55},
{0x1.59ffff21df7bap+0, 0x1.7f79f68727b02p-55},
{0x1.5dfffebfc3481p+0, -0x1.180902e30e93ep-54},
# endif
},
Move __HAVE_FAST_FMA to math_config.h Define it consistently with other HAVE_* macros that only affect code using math_config.h. This is also closer to the Arm Optimized Routines code.
2018-07-05 19:37:25 +08:00
#endif /* !HAVE_FAST_FMA */
New log2 implementation The new implementation is provided under !__OBSOLETE_MATH, it uses ISO C99 code. With default settings the worst case error in nearest rounding mode is 0.547 ULP with inlined fma and fma contraction. It uses a 1 KB lookup table, on aarch64 .text+.rodata size of libm.a is increased by 1584 bytes. Note that the math.h header defines log2(x) to be log(x)/Ln2, this is not changed, so the new code is only used if that macro is suppressed. Improvements on Cortex-A72: latency: 2.0x thruput: 2.2x
2018-06-26 23:06:54 +08:00
};
#endif /* __OBSOLETE_MATH */