| Austin Schuh | bb1338c | 2024-06-15 19:31:16 -0700 | [diff] [blame] | 1 | /* mpn_sqr -- square natural numbers. |
| 2 | |
| 3 | Copyright 1991, 1993, 1994, 1996-2003, 2005, 2008, 2009 Free Software |
| 4 | Foundation, Inc. |
| 5 | |
| 6 | This file is part of the GNU MP Library. |
| 7 | |
| 8 | The GNU MP Library is free software; you can redistribute it and/or modify |
| 9 | it under the terms of either: |
| 10 | |
| 11 | * the GNU Lesser General Public License as published by the Free |
| 12 | Software Foundation; either version 3 of the License, or (at your |
| 13 | option) any later version. |
| 14 | |
| 15 | or |
| 16 | |
| 17 | * the GNU General Public License as published by the Free Software |
| 18 | Foundation; either version 2 of the License, or (at your option) any |
| 19 | later version. |
| 20 | |
| 21 | or both in parallel, as here. |
| 22 | |
| 23 | The GNU MP Library is distributed in the hope that it will be useful, but |
| 24 | WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY |
| 25 | or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 26 | for more details. |
| 27 | |
| 28 | You should have received copies of the GNU General Public License and the |
| 29 | GNU Lesser General Public License along with the GNU MP Library. If not, |
| 30 | see https://www.gnu.org/licenses/. */ |
| 31 | |
| 32 | #include "gmp-impl.h" |
| 33 | #include "longlong.h" |
| 34 | |
| 35 | void |
| 36 | mpn_sqr (mp_ptr p, mp_srcptr a, mp_size_t n) |
| 37 | { |
| 38 | ASSERT (n >= 1); |
| 39 | ASSERT (! MPN_OVERLAP_P (p, 2 * n, a, n)); |
| 40 | |
| 41 | if (BELOW_THRESHOLD (n, SQR_BASECASE_THRESHOLD)) |
| 42 | { /* mul_basecase is faster than sqr_basecase on small sizes sometimes */ |
| 43 | mpn_mul_basecase (p, a, n, a, n); |
| 44 | } |
| 45 | else if (BELOW_THRESHOLD (n, SQR_TOOM2_THRESHOLD)) |
| 46 | { |
| 47 | mpn_sqr_basecase (p, a, n); |
| 48 | } |
| 49 | else if (BELOW_THRESHOLD (n, SQR_TOOM3_THRESHOLD)) |
| 50 | { |
| 51 | /* Allocate workspace of fixed size on stack: fast! */ |
| 52 | mp_limb_t ws[mpn_toom2_sqr_itch (SQR_TOOM3_THRESHOLD_LIMIT-1)]; |
| 53 | ASSERT (SQR_TOOM3_THRESHOLD <= SQR_TOOM3_THRESHOLD_LIMIT); |
| 54 | mpn_toom2_sqr (p, a, n, ws); |
| 55 | } |
| 56 | else if (BELOW_THRESHOLD (n, SQR_TOOM4_THRESHOLD)) |
| 57 | { |
| 58 | mp_ptr ws; |
| 59 | TMP_SDECL; |
| 60 | TMP_SMARK; |
| 61 | ws = TMP_SALLOC_LIMBS (mpn_toom3_sqr_itch (n)); |
| 62 | mpn_toom3_sqr (p, a, n, ws); |
| 63 | TMP_SFREE; |
| 64 | } |
| 65 | else if (BELOW_THRESHOLD (n, SQR_TOOM6_THRESHOLD)) |
| 66 | { |
| 67 | mp_ptr ws; |
| 68 | TMP_SDECL; |
| 69 | TMP_SMARK; |
| 70 | ws = TMP_SALLOC_LIMBS (mpn_toom4_sqr_itch (n)); |
| 71 | mpn_toom4_sqr (p, a, n, ws); |
| 72 | TMP_SFREE; |
| 73 | } |
| 74 | else if (BELOW_THRESHOLD (n, SQR_TOOM8_THRESHOLD)) |
| 75 | { |
| 76 | mp_ptr ws; |
| 77 | TMP_SDECL; |
| 78 | TMP_SMARK; |
| 79 | ws = TMP_SALLOC_LIMBS (mpn_toom6_sqr_itch (n)); |
| 80 | mpn_toom6_sqr (p, a, n, ws); |
| 81 | TMP_SFREE; |
| 82 | } |
| 83 | else if (BELOW_THRESHOLD (n, SQR_FFT_THRESHOLD)) |
| 84 | { |
| 85 | mp_ptr ws; |
| 86 | TMP_DECL; |
| 87 | TMP_MARK; |
| 88 | ws = TMP_ALLOC_LIMBS (mpn_toom8_sqr_itch (n)); |
| 89 | mpn_toom8_sqr (p, a, n, ws); |
| 90 | TMP_FREE; |
| 91 | } |
| 92 | else |
| 93 | { |
| 94 | /* The current FFT code allocates its own space. That should probably |
| 95 | change. */ |
| 96 | mpn_fft_mul (p, a, n, a, n); |
| 97 | } |
| 98 | } |