| 1 |
282 |
jeremybenn |
;; Copyright (C) 2001, 2004 Free Software Foundation, Inc.
|
| 2 |
|
|
;;
|
| 3 |
|
|
;; This file is part of GCC.
|
| 4 |
|
|
;;
|
| 5 |
|
|
;; GCC is free software; you can redistribute it and/or modify it under
|
| 6 |
|
|
;; the terms of the GNU General Public License as published by the Free
|
| 7 |
|
|
;; Software Foundation; either version 3, or (at your option) any later
|
| 8 |
|
|
;; version.
|
| 9 |
|
|
;;
|
| 10 |
|
|
;; GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
| 11 |
|
|
;; WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
| 12 |
|
|
;; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
| 13 |
|
|
;; for more details.
|
| 14 |
|
|
;;
|
| 15 |
|
|
;; Under Section 7 of GPL version 3, you are granted additional
|
| 16 |
|
|
;; permissions described in the GCC Runtime Library Exception, version
|
| 17 |
|
|
;; 3.1, as published by the Free Software Foundation.
|
| 18 |
|
|
;;
|
| 19 |
|
|
;; You should have received a copy of the GNU General Public License and
|
| 20 |
|
|
;; a copy of the GCC Runtime Library Exception along with this program;
|
| 21 |
|
|
;; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
| 22 |
|
|
;; .
|
| 23 |
|
|
;;
|
| 24 |
|
|
;; This code used to be expanded through interesting expansions in
|
| 25 |
|
|
;; the machine description, compiled from this code:
|
| 26 |
|
|
;;
|
| 27 |
|
|
;; #ifdef L_mulsi3
|
| 28 |
|
|
;; long __Mul (unsigned long a, unsigned long b) __attribute__ ((__const__));
|
| 29 |
|
|
;;
|
| 30 |
|
|
;; /* This must be compiled with the -mexpand-mul flag, to synthesize the
|
| 31 |
|
|
;; multiplication from the mstep instructions. The check for
|
| 32 |
|
|
;; smaller-size multiplication pays off in the order of .5-10%;
|
| 33 |
|
|
;; estimated median 1%, depending on application.
|
| 34 |
|
|
;; FIXME: It can be further optimized if we go to assembler code, as
|
| 35 |
|
|
;; gcc 2.7.2 adds a few unnecessary instructions and does not put the
|
| 36 |
|
|
;; basic blocks in optimal order. */
|
| 37 |
|
|
;; long
|
| 38 |
|
|
;; __Mul (unsigned long a, unsigned long b)
|
| 39 |
|
|
;; {
|
| 40 |
|
|
;; #if defined (__CRIS_arch_version) && __CRIS_arch_version >= 10
|
| 41 |
|
|
;; /* In case other code is compiled without -march=v10, they will
|
| 42 |
|
|
;; contain calls to __Mul, regardless of flags at link-time. The
|
| 43 |
|
|
;; "else"-code below will work, but is unnecessarily slow. This
|
| 44 |
|
|
;; sometimes cuts a few minutes off from simulation time by just
|
| 45 |
|
|
;; returning a "mulu.d". */
|
| 46 |
|
|
;; return a * b;
|
| 47 |
|
|
;; #else
|
| 48 |
|
|
;; unsigned long min;
|
| 49 |
|
|
;;
|
| 50 |
|
|
;; /* Get minimum via the bound insn. */
|
| 51 |
|
|
;; min = a < b ? a : b;
|
| 52 |
|
|
;;
|
| 53 |
|
|
;; /* Can we omit computation of the high part? */
|
| 54 |
|
|
;; if (min > 65535)
|
| 55 |
|
|
;; /* No. Perform full multiplication. */
|
| 56 |
|
|
;; return a * b;
|
| 57 |
|
|
;; else
|
| 58 |
|
|
;; {
|
| 59 |
|
|
;; /* Check if both operands are within 16 bits. */
|
| 60 |
|
|
;; unsigned long max;
|
| 61 |
|
|
;;
|
| 62 |
|
|
;; /* Get maximum, by knowing the minimum.
|
| 63 |
|
|
;; This will partition a and b into max and min.
|
| 64 |
|
|
;; This is not currently something GCC understands,
|
| 65 |
|
|
;; so do this trick by asm. */
|
| 66 |
|
|
;; __asm__ ("xor %1,%0\n\txor %2,%0"
|
| 67 |
|
|
;; : "=r" (max)
|
| 68 |
|
|
;; : "r" (b), "r" (a), "0" (min));
|
| 69 |
|
|
;;
|
| 70 |
|
|
;; if (max > 65535)
|
| 71 |
|
|
;; /* Make GCC understand that only the low part of "min" will be
|
| 72 |
|
|
;; used. */
|
| 73 |
|
|
;; return max * (unsigned short) min;
|
| 74 |
|
|
;; else
|
| 75 |
|
|
;; /* Only the low parts of both operands are necessary. */
|
| 76 |
|
|
;; return ((unsigned short) max) * (unsigned short) min;
|
| 77 |
|
|
;; }
|
| 78 |
|
|
;; #endif /* not __CRIS_arch_version >= 10 */
|
| 79 |
|
|
;; }
|
| 80 |
|
|
;; #endif /* L_mulsi3 */
|
| 81 |
|
|
;;
|
| 82 |
|
|
;; That approach was abandoned since the caveats outweighted the
|
| 83 |
|
|
;; benefits. The expand-multiplication machinery is also removed, so you
|
| 84 |
|
|
;; can't do this anymore.
|
| 85 |
|
|
;;
|
| 86 |
|
|
;; For doubters of there being any benefits, some where: insensitivity to:
|
| 87 |
|
|
;; - ABI changes (mostly for experimentation)
|
| 88 |
|
|
;; - assembler syntax differences (mostly debug format).
|
| 89 |
|
|
;; - insn scheduling issues.
|
| 90 |
|
|
;; Most ABI experiments will presumably happen with arches with mul insns,
|
| 91 |
|
|
;; so that argument doesn't really hold anymore, and it's unlikely there
|
| 92 |
|
|
;; being new arch variants needing insn scheduling and not having mul
|
| 93 |
|
|
;; insns.
|
| 94 |
|
|
|
| 95 |
|
|
;; ELF and a.out have different syntax for local labels: the "wrong"
|
| 96 |
|
|
;; one may not be omitted from the object.
|
| 97 |
|
|
#undef L
|
| 98 |
|
|
#ifdef __AOUT__
|
| 99 |
|
|
# define L(x) x
|
| 100 |
|
|
#else
|
| 101 |
|
|
# define L(x) .x
|
| 102 |
|
|
#endif
|
| 103 |
|
|
|
| 104 |
|
|
.global ___Mul
|
| 105 |
|
|
.type ___Mul,@function
|
| 106 |
|
|
___Mul:
|
| 107 |
|
|
#if defined (__CRIS_arch_version) && __CRIS_arch_version >= 10
|
| 108 |
|
|
;; Can't have the mulu.d last on a cache-line (in the delay-slot of the
|
| 109 |
|
|
;; "ret"), due to hardware bug. See documentation for -mmul-bug-workaround.
|
| 110 |
|
|
;; Not worthwhile to conditionalize here.
|
| 111 |
|
|
.p2alignw 2,0x050f
|
| 112 |
|
|
mulu.d $r11,$r10
|
| 113 |
|
|
ret
|
| 114 |
|
|
nop
|
| 115 |
|
|
#else
|
| 116 |
|
|
move.d $r10,$r12
|
| 117 |
|
|
move.d $r11,$r9
|
| 118 |
|
|
bound.d $r12,$r9
|
| 119 |
|
|
cmpu.w 65535,$r9
|
| 120 |
|
|
bls L(L3)
|
| 121 |
|
|
move.d $r12,$r13
|
| 122 |
|
|
|
| 123 |
|
|
movu.w $r11,$r9
|
| 124 |
|
|
lslq 16,$r13
|
| 125 |
|
|
mstep $r9,$r13
|
| 126 |
|
|
mstep $r9,$r13
|
| 127 |
|
|
mstep $r9,$r13
|
| 128 |
|
|
mstep $r9,$r13
|
| 129 |
|
|
mstep $r9,$r13
|
| 130 |
|
|
mstep $r9,$r13
|
| 131 |
|
|
mstep $r9,$r13
|
| 132 |
|
|
mstep $r9,$r13
|
| 133 |
|
|
mstep $r9,$r13
|
| 134 |
|
|
mstep $r9,$r13
|
| 135 |
|
|
mstep $r9,$r13
|
| 136 |
|
|
mstep $r9,$r13
|
| 137 |
|
|
mstep $r9,$r13
|
| 138 |
|
|
mstep $r9,$r13
|
| 139 |
|
|
mstep $r9,$r13
|
| 140 |
|
|
mstep $r9,$r13
|
| 141 |
|
|
clear.w $r10
|
| 142 |
|
|
test.d $r10
|
| 143 |
|
|
mstep $r9,$r10
|
| 144 |
|
|
mstep $r9,$r10
|
| 145 |
|
|
mstep $r9,$r10
|
| 146 |
|
|
mstep $r9,$r10
|
| 147 |
|
|
mstep $r9,$r10
|
| 148 |
|
|
mstep $r9,$r10
|
| 149 |
|
|
mstep $r9,$r10
|
| 150 |
|
|
mstep $r9,$r10
|
| 151 |
|
|
mstep $r9,$r10
|
| 152 |
|
|
mstep $r9,$r10
|
| 153 |
|
|
mstep $r9,$r10
|
| 154 |
|
|
mstep $r9,$r10
|
| 155 |
|
|
mstep $r9,$r10
|
| 156 |
|
|
mstep $r9,$r10
|
| 157 |
|
|
mstep $r9,$r10
|
| 158 |
|
|
mstep $r9,$r10
|
| 159 |
|
|
movu.w $r12,$r12
|
| 160 |
|
|
move.d $r11,$r9
|
| 161 |
|
|
clear.w $r9
|
| 162 |
|
|
test.d $r9
|
| 163 |
|
|
mstep $r12,$r9
|
| 164 |
|
|
mstep $r12,$r9
|
| 165 |
|
|
mstep $r12,$r9
|
| 166 |
|
|
mstep $r12,$r9
|
| 167 |
|
|
mstep $r12,$r9
|
| 168 |
|
|
mstep $r12,$r9
|
| 169 |
|
|
mstep $r12,$r9
|
| 170 |
|
|
mstep $r12,$r9
|
| 171 |
|
|
mstep $r12,$r9
|
| 172 |
|
|
mstep $r12,$r9
|
| 173 |
|
|
mstep $r12,$r9
|
| 174 |
|
|
mstep $r12,$r9
|
| 175 |
|
|
mstep $r12,$r9
|
| 176 |
|
|
mstep $r12,$r9
|
| 177 |
|
|
mstep $r12,$r9
|
| 178 |
|
|
mstep $r12,$r9
|
| 179 |
|
|
add.w $r9,$r10
|
| 180 |
|
|
lslq 16,$r10
|
| 181 |
|
|
ret
|
| 182 |
|
|
add.d $r13,$r10
|
| 183 |
|
|
|
| 184 |
|
|
L(L3):
|
| 185 |
|
|
move.d $r9,$r10
|
| 186 |
|
|
xor $r11,$r10
|
| 187 |
|
|
xor $r12,$r10
|
| 188 |
|
|
cmpu.w 65535,$r10
|
| 189 |
|
|
bls L(L5)
|
| 190 |
|
|
movu.w $r9,$r13
|
| 191 |
|
|
|
| 192 |
|
|
movu.w $r13,$r13
|
| 193 |
|
|
move.d $r10,$r9
|
| 194 |
|
|
lslq 16,$r9
|
| 195 |
|
|
mstep $r13,$r9
|
| 196 |
|
|
mstep $r13,$r9
|
| 197 |
|
|
mstep $r13,$r9
|
| 198 |
|
|
mstep $r13,$r9
|
| 199 |
|
|
mstep $r13,$r9
|
| 200 |
|
|
mstep $r13,$r9
|
| 201 |
|
|
mstep $r13,$r9
|
| 202 |
|
|
mstep $r13,$r9
|
| 203 |
|
|
mstep $r13,$r9
|
| 204 |
|
|
mstep $r13,$r9
|
| 205 |
|
|
mstep $r13,$r9
|
| 206 |
|
|
mstep $r13,$r9
|
| 207 |
|
|
mstep $r13,$r9
|
| 208 |
|
|
mstep $r13,$r9
|
| 209 |
|
|
mstep $r13,$r9
|
| 210 |
|
|
mstep $r13,$r9
|
| 211 |
|
|
clear.w $r10
|
| 212 |
|
|
test.d $r10
|
| 213 |
|
|
mstep $r13,$r10
|
| 214 |
|
|
mstep $r13,$r10
|
| 215 |
|
|
mstep $r13,$r10
|
| 216 |
|
|
mstep $r13,$r10
|
| 217 |
|
|
mstep $r13,$r10
|
| 218 |
|
|
mstep $r13,$r10
|
| 219 |
|
|
mstep $r13,$r10
|
| 220 |
|
|
mstep $r13,$r10
|
| 221 |
|
|
mstep $r13,$r10
|
| 222 |
|
|
mstep $r13,$r10
|
| 223 |
|
|
mstep $r13,$r10
|
| 224 |
|
|
mstep $r13,$r10
|
| 225 |
|
|
mstep $r13,$r10
|
| 226 |
|
|
mstep $r13,$r10
|
| 227 |
|
|
mstep $r13,$r10
|
| 228 |
|
|
mstep $r13,$r10
|
| 229 |
|
|
lslq 16,$r10
|
| 230 |
|
|
ret
|
| 231 |
|
|
add.d $r9,$r10
|
| 232 |
|
|
|
| 233 |
|
|
L(L5):
|
| 234 |
|
|
movu.w $r9,$r9
|
| 235 |
|
|
lslq 16,$r10
|
| 236 |
|
|
mstep $r9,$r10
|
| 237 |
|
|
mstep $r9,$r10
|
| 238 |
|
|
mstep $r9,$r10
|
| 239 |
|
|
mstep $r9,$r10
|
| 240 |
|
|
mstep $r9,$r10
|
| 241 |
|
|
mstep $r9,$r10
|
| 242 |
|
|
mstep $r9,$r10
|
| 243 |
|
|
mstep $r9,$r10
|
| 244 |
|
|
mstep $r9,$r10
|
| 245 |
|
|
mstep $r9,$r10
|
| 246 |
|
|
mstep $r9,$r10
|
| 247 |
|
|
mstep $r9,$r10
|
| 248 |
|
|
mstep $r9,$r10
|
| 249 |
|
|
mstep $r9,$r10
|
| 250 |
|
|
mstep $r9,$r10
|
| 251 |
|
|
ret
|
| 252 |
|
|
mstep $r9,$r10
|
| 253 |
|
|
#endif
|
| 254 |
|
|
L(Lfe1):
|
| 255 |
|
|
.size ___Mul,L(Lfe1)-___Mul
|
