1 /* UltraSPARC 64 mpn_modexact_1c_odd -- mpn by limb exact style remainder. 2 3 THE FUNCTIONS IN THIS FILE ARE FOR INTERNAL USE ONLY. THEY'RE ALMOST 4 CERTAIN TO BE SUBJECT TO INCOMPATIBLE CHANGES OR DISAPPEAR COMPLETELY IN 5 FUTURE GNU MP RELEASES. 6 7 Copyright 2000-2003 Free Software Foundation, Inc. 8 9 This file is part of the GNU MP Library. 10 11 The GNU MP Library is free software; you can redistribute it and/or modify 12 it under the terms of either: 13 14 * the GNU Lesser General Public License as published by the Free 15 Software Foundation; either version 3 of the License, or (at your 16 option) any later version. 17 18 or 19 20 * the GNU General Public License as published by the Free Software 21 Foundation; either version 2 of the License, or (at your option) any 22 later version. 23 24 or both in parallel, as here. 25 26 The GNU MP Library is distributed in the hope that it will be useful, but 27 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 28 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 29 for more details. 30 31 You should have received copies of the GNU General Public License and the 32 GNU Lesser General Public License along with the GNU MP Library. If not, 33 see https://www.gnu.org/licenses/. */ 34 35 #include "gmp.h" 36 #include "gmp-impl.h" 37 #include "longlong.h" 38 39 #include "mpn/sparc64/sparc64.h" 40 41 42 /* 64-bit divisor 32-bit divisor 43 cycles/limb cycles/limb 44 (approx) (approx) 45 Ultrasparc 2i: ? ? 46 */ 47 48 49 /* This implementation reduces the number of multiplies done, knowing that 50 on ultrasparc 1 and 2 the mulx instruction stalls the whole chip. 51 52 The key idea is to use the fact that the low limb of q*d equals l, this 53 being the whole purpose of the q calculated. It means there's no need to 54 calculate the lowest 32x32->64 part of the q*d, instead it can be 55 inferred from l and the other three 32x32->64 parts. See sparc64.h for 56 details. 57 58 When d is 32-bits, the same applies, but in this case there's only one 59 other 32x32->64 part (ie. HIGH(q)*d). 60 61 The net effect is that for 64-bit divisor each limb is 4 mulx, or for 62 32-bit divisor each is 2 mulx. 63 64 Enhancements: 65 66 No doubt this could be done in assembler, if that helped the scheduling, 67 or perhaps guaranteed good code irrespective of the compiler. 68 69 Alternatives: 70 71 It might be possibly to use floating point. The loop is dominated by 72 multiply latency, so not sure if floats would improve that. One 73 possibility would be to take two limbs at a time, with a 128 bit inverse, 74 if there's enough registers, which could effectively use float throughput 75 to reduce total latency across two limbs. */ 76 77 #define ASSERT_RETVAL(r) \ 78 ASSERT (orig_c < d ? r < d : r <= d) 79 80 mp_limb_t 81 mpn_modexact_1c_odd (mp_srcptr src, mp_size_t size, mp_limb_t d, mp_limb_t orig_c) 82 { 83 mp_limb_t c = orig_c; 84 mp_limb_t s, l, q, h, inverse; 85 86 ASSERT (size >= 1); 87 ASSERT (d & 1); 88 ASSERT_MPN (src, size); 89 ASSERT_LIMB (d); 90 ASSERT_LIMB (c); 91 92 /* udivx is faster than 10 or 12 mulx's for one limb via an inverse */ 93 if (size == 1) 94 { 95 s = src[0]; 96 if (s > c) 97 { 98 l = s-c; 99 h = l % d; 100 if (h != 0) 101 h = d - h; 102 } 103 else 104 { 105 l = c-s; 106 h = l % d; 107 } 108 return h; 109 } 110 111 binvert_limb (inverse, d); 112 113 if (d <= 0xFFFFFFFF) 114 { 115 s = *src++; 116 size--; 117 do 118 { 119 SUBC_LIMB (c, l, s, c); 120 s = *src++; 121 q = l * inverse; 122 umul_ppmm_half_lowequal (h, q, d, l); 123 c += h; 124 size--; 125 } 126 while (size != 0); 127 128 if (s <= d) 129 { 130 /* With high s <= d the final step can be a subtract and addback. 131 If c==0 then the addback will restore to l>=0. If c==d then 132 will get l==d if s==0, but that's ok per the function 133 definition. */ 134 135 l = c - s; 136 l += (l > c ? d : 0); 137 138 ASSERT_RETVAL (l); 139 return l; 140 } 141 else 142 { 143 /* Can't skip a divide, just do the loop code once more. */ 144 SUBC_LIMB (c, l, s, c); 145 q = l * inverse; 146 umul_ppmm_half_lowequal (h, q, d, l); 147 c += h; 148 149 ASSERT_RETVAL (c); 150 return c; 151 } 152 } 153 else 154 { 155 mp_limb_t dl = LOW32 (d); 156 mp_limb_t dh = HIGH32 (d); 157 long i; 158 159 s = *src++; 160 size--; 161 do 162 { 163 SUBC_LIMB (c, l, s, c); 164 s = *src++; 165 q = l * inverse; 166 umul_ppmm_lowequal (h, q, d, dh, dl, l); 167 c += h; 168 size--; 169 } 170 while (size != 0); 171 172 if (s <= d) 173 { 174 /* With high s <= d the final step can be a subtract and addback. 175 If c==0 then the addback will restore to l>=0. If c==d then 176 will get l==d if s==0, but that's ok per the function 177 definition. */ 178 179 l = c - s; 180 l += (l > c ? d : 0); 181 182 ASSERT_RETVAL (l); 183 return l; 184 } 185 else 186 { 187 /* Can't skip a divide, just do the loop code once more. */ 188 SUBC_LIMB (c, l, s, c); 189 q = l * inverse; 190 umul_ppmm_lowequal (h, q, d, dh, dl, l); 191 c += h; 192 193 ASSERT_RETVAL (c); 194 return c; 195 } 196 } 197 } 198