projects / thirdparty / glibc.git / blame
| Commit | Line | Data |
|---|---|---|
| 1aff59a3 | 1 | /* Float floor function, sparc32 v9 version. |
| 568035b7 | 2 | Copyright (C) 2012-2013 Free Software Foundation, Inc. |
| 1aff59a3 DM |
3 | This file is part of the GNU C Library. |
| 4 | Contributed by David S. Miller <davem@davemloft.net>, 2012. | |
| 5 | ||
| 6 | The GNU C Library is free software; you can redistribute it and/or | |
| 7 | modify it under the terms of the GNU Lesser General Public | |
| 8 | License as published by the Free Software Foundation; either | |
| 9 | version 2.1 of the License, or (at your option) any later version. | |
| 10 | ||
| 11 | The GNU C Library is distributed in the hope that it will be useful, | |
| 12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 14 | Lesser General Public License for more details. | |
| 15 | ||
| 16 | You should have received a copy of the GNU Lesser General Public | |
| 17 | License along with the GNU C Library; if not, see | |
| 18 | <http://www.gnu.org/licenses/>. */ | |
| 19 | ||
| 20 | #include <sysdep.h> | |
| 21 | ||
| 22 | /* Since changing the rounding mode is extremely expensive, we | |
| 23 | try to round up using a method that is rounding mode | |
| 24 | agnostic. | |
| 25 | ||
| 26 | We add then subtract (or subtract than add if the initial | |
| 27 | value was negative) 2**23 to the value, then subtract it | |
| 28 | back out. | |
| 29 | ||
| 30 | This will clear out the fractional portion of the value. | |
| 31 | One of two things will happen for non-whole initial values. | |
| 32 | Either the rounding mode will round it up, or it will be | |
| 33 | rounded down. If the value started out whole, it will be | |
| 34 | equal after the addition and subtraction. This means we | |
| 35 | can accurately detect with one test whether we need to add | |
| 36 | another 1.0 to round it up properly. | |
| 37 | ||
| 38 | We pop constants into the FPU registers using the incoming | |
| 39 | argument stack slots, since this avoid having to use any | |
| 40 | PIC references. We also thus avoid having to allocate a | |
| 41 | register window. | |
| 42 | ||
| 43 | VIS instructions are used to facilitate the formation of | |
| 44 | easier constants, and the propagation of the sign bit. */ | |
| 45 | ||
| 46 | #define TWO_TWENTYTHREE 0x4b000000 /* 2**23 */ | |
| 47 | #define ONE_DOT_ZERO 0x3f800000 /* 1.0 */ | |
| 48 | ||
| 49 | #define ZERO %f10 /* 0.0 */ | |
| 50 | #define SIGN_BIT %f12 /* -0.0 */ | |
| 51 | ||
| 52 | ENTRY (__floorf) | |
| 53 | st %o0, [%sp + 68] | |
| 54 | sethi %hi(TWO_TWENTYTHREE), %o2 | |
| 55 | sethi %hi(ONE_DOT_ZERO), %o3 | |
| 56 | fzeros ZERO | |
| 57 | ||
| 58 | ld [%sp + 68], %f0 | |
| 59 | fnegs ZERO, SIGN_BIT | |
| 60 | ||
| 61 | st %o2, [%sp + 68] | |
| 62 | fabss %f0, %f14 | |
| 63 | ||
| 64 | ld [%sp + 68], %f16 | |
| 65 | fcmps %fcc3, %f14, %f16 | |
| 66 | ||
| 67 | fmovsuge %fcc3, ZERO, %f16 | |
| 68 | fands %f0, SIGN_BIT, SIGN_BIT | |
| 69 | ||
| 70 | fors %f16, SIGN_BIT, %f16 | |
| 71 | fadds %f0, %f16, %f1 | |
| 72 | fsubs %f1, %f16, %f1 | |
| 73 | fcmps %fcc2, %f1, %f0 | |
| 74 | st %o3, [%sp + 68] | |
| 75 | ||
| 76 | ld [%sp + 68], %f9 | |
| 77 | fmovsule %fcc2, ZERO, %f9 | |
| 78 | fsubs %f1, %f9, %f0 | |
| 79 | fabss %f0, %f0 | |
| 80 | retl | |
| 81 | fors %f0, SIGN_BIT, %f0 | |
| 82 | END (__floorf) | |
| 83 | weak_alias (__floorf, floorf) |