libgcc/config/arc/ieee-754/arc600-mul64/muldf3.S

   1 /* Copyright (C) 2008-2020 Free Software Foundation, Inc.
   2    Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
   3                 on behalf of Synopsys Inc.
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify it under
   8 the terms of the GNU General Public License as published by the Free
   9 Software Foundation; either version 3, or (at your option) any later
  10 version.
  11
  12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15 for more details.
  16
  17 Under Section 7 of GPL version 3, you are granted additional
  18 permissions described in the GCC Runtime Library Exception, version
  19 3.1, as published by the Free Software Foundation.
  20
  21 You should have received a copy of the GNU General Public License and
  22 a copy of the GCC Runtime Library Exception along with this program;
  23 see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
  24 <http://www.gnu.org/licenses/>.  */
  25
  26 #include "../arc-ieee-754.h"
  27
  28 #if 0 /* DEBUG */
  29         .global __muldf3
  30         .balign 4
  31 __muldf3:
  32         push_s blink
  33         push_s r2
  34         push_s r3
  35         push_s r0
  36         bl.d __muldf3_c
  37         push_s r1
  38         ld_s r2,[sp,12]
  39         ld_s r3,[sp,8]
  40         st_s r0,[sp,12]
  41         st_s r1,[sp,8]
  42         pop_s r1
  43         bl.d __muldf3_asm
  44         pop_s r0
  45         pop_s r3
  46         pop_s r2
  47         pop_s blink
  48         cmp r0,r2
  49         cmp.eq r1,r3
  50         jeq_s [blink]
  51         and r12,DBL0H,DBL1H
  52         bic.f 0,0x7ff80000,r12 ; both NaN -> OK
  53         jeq_s [blink]
  54         b abort
  55 #define __muldf3 __muldf3_asm
  56 #endif /* DEBUG */
  57
  58 __muldf3_support: /* This label makes debugger output saner.  */
  59         .balign 4
  60         FUNC(__muldf3)
  61 .Ldenorm_2:
  62         breq.d DBL1L,0,.Lret0_2 ; 0 input -> 0 output
  63         norm.f r12,DBL1L
  64         mov.mi r12,21
  65         add.pl r12,r12,22
  66         neg r11,r12
  67         asl_s r12,r12,20
  68         lsr.f DBL1H,DBL1L,r11
  69         ror DBL1L,DBL1L,r11
  70         sub_s DBL0H,DBL0H,r12
  71         mov.eq DBL1H,DBL1L
  72         sub_l DBL1L,DBL1L,DBL1H
  73         /* Fall through.  */
  74         .global __muldf3
  75         .balign 4
  76 __muldf3:
  77         mulu64 DBL0L,DBL1L
  78         ld.as r9,[pcl,0x68] ; ((.L7ff00000-.+2)/4)]
  79         bmsk r6,DBL0H,19
  80         bset r6,r6,20
  81         and r11,DBL0H,r9
  82         breq.d r11,0,.Ldenorm_dbl0
  83         and r12,DBL1H,r9
  84         breq.d r12,0,.Ldenorm_dbl1
  85         mov r8,mlo
  86         mov r4,mhi
  87         mulu64 r6,DBL1L
  88         breq.d r11,r9,.Linf_nan
  89         bmsk r10,DBL1H,19
  90         breq.d r12,r9,.Linf_nan
  91         bset r10,r10,20
  92         add.f r4,r4,mlo
  93         adc r5,mhi,0
  94         mulu64 r10,DBL0L
  95         add_s r12,r12,r11 ; add exponents
  96         add.f r4,r4,mlo
  97         adc r5,r5,mhi
  98         mulu64 r6,r10
  99         tst r8,r8
 100         bclr r8,r9,30 ; 0x3ff00000
 101         bset.ne r4,r4,0 ; put least significant word into sticky bit
 102         bclr r6,r9,20 ; 0x7fe00000
 103         add.f r5,r5,mlo
 104         adc r7,mhi,0 ; fraction product in r7:r5:r4
 105         lsr.f r10,r7,9
 106         rsub.eq r8,r8,r9 ; 0x40000000
 107         sub r12,r12,r8 ; subtract bias + implicit 1
 108         brhs.d r12,r6,.Linf_denorm
 109         rsub r10,r10,12
 110 .Lshift_frac:
 111         neg r8,r10
 112         asl r6,r4,r10
 113         lsr DBL0L,r4,r8
 114         add.f 0,r6,r6
 115         btst.eq DBL0L,0
 116         cmp.eq r4,r4 ; round to nearest / round to even
 117         asl r4,r5,r10
 118         lsr r5,r5,r8
 119         adc.f DBL0L,DBL0L,r4
 120         xor.f 0,DBL0H,DBL1H
 121         asl r7,r7,r10
 122         add_s r12,r12,r5
 123         adc DBL0H,r12,r7
 124         j_s.d [blink]
 125         bset.mi DBL0H,DBL0H,31
 126
 127 /* N.B. This is optimized for ARC700.
 128   ARC600 has very different scheduling / instruction selection criteria.  */
 129
 130 /* If one number is denormal, subtract some from the exponent of the other
 131    one (if the other exponent is too small, return 0), and normalize the
 132    denormal.  Then re-run the computation.  */
 133 .Lret0_2:
 134         lsr_s DBL0H,DBL0H,31
 135         asl_s DBL0H,DBL0H,31
 136         j_s.d [blink]
 137         mov_s DBL0L,0
 138         .balign 4
 139 .Ldenorm_dbl0:
 140         mov_s r12,DBL0L
 141         mov_s DBL0L,DBL1L
 142         mov_s DBL1L,r12
 143         mov_s r12,DBL0H
 144         mov_s DBL0H,DBL1H
 145         mov_s DBL1H,r12
 146         and r11,DBL0H,r9
 147 .Ldenorm_dbl1:
 148         brhs r11,r9,.Linf_nan
 149         brhs 0x3ca00001,r11,.Lret0
 150         sub_s DBL0H,DBL0H,DBL1H
 151         bmsk.f DBL1H,DBL1H,30
 152         add_s DBL0H,DBL0H,DBL1H
 153         beq.d .Ldenorm_2
 154         norm r12,DBL1H
 155         sub_s r12,r12,10
 156         asl r5,r12,20
 157         asl_s DBL1H,DBL1H,r12
 158         sub DBL0H,DBL0H,r5
 159         neg r5,r12
 160         lsr r6,DBL1L,r5
 161         asl_s DBL1L,DBL1L,r12
 162         b.d __muldf3
 163         add_s DBL1H,DBL1H,r6
 164
 165 .Lret0: xor_s DBL0H,DBL0H,DBL1H
 166         bclr DBL1H,DBL0H,31
 167         xor_s DBL0H,DBL0H,DBL1H
 168         j_s.d [blink]
 169         mov_s DBL0L,0
 170
 171         .balign 4
 172 .Linf_nan:
 173         bclr r12,DBL1H,31
 174         xor_s DBL1H,DBL1H,DBL0H
 175         bclr_s DBL0H,DBL0H,31
 176         max r8,DBL0H,r12 ; either NaN -> NaN ; otherwise inf
 177         or.f 0,DBL0H,DBL0L
 178         mov_s DBL0L,0
 179         or.ne.f DBL1L,DBL1L,r12
 180         not_s DBL0H,DBL0L ; inf * 0 -> NaN
 181         mov.ne DBL0H,r8
 182         tst_s DBL1H,DBL1H
 183         j_s.d [blink]
 184         bset.mi DBL0H,DBL0H,31
 185
 186 /* We have checked for infinity / NaN input before, and transformed
 187    denormalized inputs into normalized inputs.  Thus, the worst case
 188    exponent overflows are:
 189        1 +     1 - 0x400 == 0xc02 : maximum underflow
 190    0x7fe + 0x7fe - 0x3ff == 0xbfd ; maximum overflow
 191    N.B. 0x7e and 0x7f are also values for overflow.
 192
 193    If (r12 <= -54), we have an underflow to zero.  */
 194         .balign 4
 195 .Linf_denorm:
 196         lsr r6,r12,28
 197         brlo.d r6,0xc,.Linf
 198         asr r6,r12,20
 199         add.f r10,r10,r6
 200         brgt.d r10,0,.Lshift_frac
 201         mov_s r12,0
 202         beq.d .Lround_frac
 203         add r10,r10,32
 204 .Lshift32_frac:
 205         tst r4,r4
 206         mov r4,r5
 207         bset.ne r4,r4,1
 208         mov r5,r7
 209         brge.d r10,1,.Lshift_frac
 210         mov r7,0
 211         breq.d r10,0,.Lround_frac
 212         add r10,r10,32
 213         brgt r10,21,.Lshift32_frac
 214         b_s .Lret0
 215
 216 .Lround_frac:
 217         add.f 0,r4,r4
 218         btst.eq r5,0
 219         mov_s DBL0L,r5
 220         mov_s DBL0H,r7
 221         adc.eq.f DBL0L,DBL0L,0
 222         j_s.d [blink]
 223         adc.eq DBL0H,DBL0H,0
 224
 225 .Linf:  mov_s DBL0L,0
 226         xor.f DBL1H,DBL1H,DBL0H
 227         mov_s DBL0H,r9
 228         j_s.d [blink]
 229         bset.mi DBL0H,DBL0H,31
 230         ENDFUNC(__muldf3)
 231
 232         .balign 4
 233 .L7ff00000:
 234         .long 0x7ff00000