xf.li | bdd93d5 | 2023-05-12 07:10:14 -0700 | [diff] [blame] | 1 | /* Copyright (C) 1996-2016 Free Software Foundation, Inc. |
| 2 | This file is part of the GNU C Library. |
| 3 | Contributed by Richard Henderson <rth@tamu.edu>. |
| 4 | |
| 5 | The GNU C Library is free software; you can redistribute it and/or |
| 6 | modify it under the terms of the GNU Lesser General Public |
| 7 | License as published by the Free Software Foundation; either |
| 8 | version 2.1 of the License, or (at your option) any later version. |
| 9 | |
| 10 | The GNU C Library is distributed in the hope that it will be useful, |
| 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | Lesser General Public License for more details. |
| 14 | |
| 15 | You should have received a copy of the GNU Lesser General Public |
| 16 | License along with the GNU C Library. If not, see |
| 17 | <http://www.gnu.org/licenses/>. */ |
| 18 | |
| 19 | #include "div_libc.h" |
| 20 | |
| 21 | #undef FRAME |
| 22 | #ifdef __alpha_fix__ |
| 23 | #define FRAME 0 |
| 24 | #else |
| 25 | #define FRAME 16 |
| 26 | #endif |
| 27 | |
| 28 | #undef X |
| 29 | #undef Y |
| 30 | #define X $17 |
| 31 | #define Y $18 |
| 32 | |
| 33 | .set noat |
| 34 | |
| 35 | .align 4 |
| 36 | .globl ldiv |
| 37 | .ent ldiv |
| 38 | ldiv: |
| 39 | .frame sp, FRAME, ra |
| 40 | #if FRAME > 0 |
| 41 | lda sp, -FRAME(sp) |
| 42 | #endif |
| 43 | #ifdef PROF |
| 44 | .set macro |
| 45 | ldgp gp, 0(pv) |
| 46 | lda AT, _mcount |
| 47 | jsr AT, (AT), _mcount |
| 48 | .set nomacro |
| 49 | .prologue 1 |
| 50 | #else |
| 51 | .prologue 0 |
| 52 | #endif |
| 53 | |
| 54 | beq Y, $divbyzero |
| 55 | excb |
| 56 | mf_fpcr $f10 |
| 57 | |
| 58 | _ITOFT2 X, $f0, 0, Y, $f1, 8 |
| 59 | |
| 60 | .align 4 |
| 61 | cvtqt $f0, $f0 |
| 62 | cvtqt $f1, $f1 |
| 63 | divt/c $f0, $f1, $f0 |
| 64 | unop |
| 65 | |
| 66 | /* Check to see if X fit in the double as an exact value. */ |
| 67 | sll X, (64-53), AT |
| 68 | sra AT, (64-53), AT |
| 69 | cmpeq X, AT, AT |
| 70 | beq AT, $x_big |
| 71 | |
| 72 | /* If we get here, we're expecting exact results from the division. |
| 73 | Do nothing else besides convert and clean up. */ |
| 74 | cvttq/c $f0, $f0 |
| 75 | excb |
| 76 | mt_fpcr $f10 |
| 77 | _FTOIT $f0, $0, 0 |
| 78 | |
| 79 | $egress: |
| 80 | mulq $0, Y, $1 |
| 81 | subq X, $1, $1 |
| 82 | |
| 83 | stq $0, 0($16) |
| 84 | stq $1, 8($16) |
| 85 | mov $16, $0 |
| 86 | |
| 87 | #if FRAME > 0 |
| 88 | lda sp, FRAME(sp) |
| 89 | #endif |
| 90 | ret |
| 91 | |
| 92 | .align 4 |
| 93 | $x_big: |
| 94 | /* If we get here, X is large enough that we don't expect exact |
| 95 | results, and neither X nor Y got mis-translated for the fp |
| 96 | division. Our task is to take the fp result, figure out how |
| 97 | far it's off from the correct result and compute a fixup. */ |
| 98 | |
| 99 | #define Q v0 /* quotient */ |
| 100 | #define R t0 /* remainder */ |
| 101 | #define SY t1 /* scaled Y */ |
| 102 | #define S t2 /* scalar */ |
| 103 | #define QY t3 /* Q*Y */ |
| 104 | |
| 105 | /* The fixup code below can only handle unsigned values. */ |
| 106 | or X, Y, AT |
| 107 | mov $31, t5 |
| 108 | blt AT, $fix_sign_in |
| 109 | $fix_sign_in_ret1: |
| 110 | cvttq/c $f0, $f0 |
| 111 | |
| 112 | _FTOIT $f0, Q, 8 |
| 113 | $fix_sign_in_ret2: |
| 114 | mulq Q, Y, QY |
| 115 | excb |
| 116 | mt_fpcr $f10 |
| 117 | |
| 118 | .align 4 |
| 119 | subq QY, X, R |
| 120 | mov Y, SY |
| 121 | mov 1, S |
| 122 | bgt R, $q_high |
| 123 | |
| 124 | $q_high_ret: |
| 125 | subq X, QY, R |
| 126 | mov Y, SY |
| 127 | mov 1, S |
| 128 | bgt R, $q_low |
| 129 | |
| 130 | $q_low_ret: |
| 131 | negq Q, t4 |
| 132 | cmovlbs t5, t4, Q |
| 133 | br $egress |
| 134 | |
| 135 | .align 4 |
| 136 | /* The quotient that we computed was too large. We need to reduce |
| 137 | it by S such that Y*S >= R. Obviously the closer we get to the |
| 138 | correct value the better, but overshooting high is ok, as we'll |
| 139 | fix that up later. */ |
| 140 | 0: |
| 141 | addq SY, SY, SY |
| 142 | addq S, S, S |
| 143 | $q_high: |
| 144 | cmpult SY, R, AT |
| 145 | bne AT, 0b |
| 146 | |
| 147 | subq Q, S, Q |
| 148 | unop |
| 149 | subq QY, SY, QY |
| 150 | br $q_high_ret |
| 151 | |
| 152 | .align 4 |
| 153 | /* The quotient that we computed was too small. Divide Y by the |
| 154 | current remainder (R) and add that to the existing quotient (Q). |
| 155 | The expectation, of course, is that R is much smaller than X. */ |
| 156 | /* Begin with a shift-up loop. Compute S such that Y*S >= R. We |
| 157 | already have a copy of Y in SY and the value 1 in S. */ |
| 158 | 0: |
| 159 | addq SY, SY, SY |
| 160 | addq S, S, S |
| 161 | $q_low: |
| 162 | cmpult SY, R, AT |
| 163 | bne AT, 0b |
| 164 | |
| 165 | /* Shift-down and subtract loop. Each iteration compares our scaled |
| 166 | Y (SY) with the remainder (R); if SY <= R then X is divisible by |
| 167 | Y's scalar (S) so add it to the quotient (Q). */ |
| 168 | 2: addq Q, S, t3 |
| 169 | srl S, 1, S |
| 170 | cmpule SY, R, AT |
| 171 | subq R, SY, t4 |
| 172 | |
| 173 | cmovne AT, t3, Q |
| 174 | cmovne AT, t4, R |
| 175 | srl SY, 1, SY |
| 176 | bne S, 2b |
| 177 | |
| 178 | br $q_low_ret |
| 179 | |
| 180 | .align 4 |
| 181 | $fix_sign_in: |
| 182 | /* If we got here, then X|Y is negative. Need to adjust everything |
| 183 | such that we're doing unsigned division in the fixup loop. */ |
| 184 | /* T5 is true if result should be negative. */ |
| 185 | xor X, Y, AT |
| 186 | cmplt AT, 0, t5 |
| 187 | cmplt X, 0, AT |
| 188 | negq X, t0 |
| 189 | |
| 190 | cmovne AT, t0, X |
| 191 | cmplt Y, 0, AT |
| 192 | negq Y, t0 |
| 193 | |
| 194 | cmovne AT, t0, Y |
| 195 | blbc t5, $fix_sign_in_ret1 |
| 196 | |
| 197 | cvttq/c $f0, $f0 |
| 198 | _FTOIT $f0, Q, 8 |
| 199 | .align 3 |
| 200 | negq Q, Q |
| 201 | br $fix_sign_in_ret2 |
| 202 | |
| 203 | $divbyzero: |
| 204 | mov a0, v0 |
| 205 | lda a0, GEN_INTDIV |
| 206 | call_pal PAL_gentrap |
| 207 | stq zero, 0(v0) |
| 208 | stq zero, 8(v0) |
| 209 | |
| 210 | #if FRAME > 0 |
| 211 | lda sp, FRAME(sp) |
| 212 | #endif |
| 213 | ret |
| 214 | |
| 215 | .end ldiv |
| 216 | |
| 217 | weak_alias (ldiv, lldiv) |
| 218 | weak_alias (ldiv, imaxdiv) |