libgcc/config/arc/ieee-754/arc600-mul64/divsf3.S

   1 /* Copyright (C) 2008-2021 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 /*
  27    - calculate 15..18 bit inverse using a table of approximating polynoms.
  28      precision is higher for polynoms used to evaluate input with larger
  29      value.
  30    - do one newton-raphson iteration step to double the precision,
  31      then multiply this with the divisor
  32         -> more time to decide if dividend is subnormal
  33      - the worst error propagation is on the side of the value range
  34        with the least initial defect, thus giving us about 30 bits precision.
  35  */
  36 #include "../arc-ieee-754.h"
  37
  38 #if 0 /* DEBUG */
  39         .global __divsf3
  40         FUNC(__divsf3)
  41         .balign 4
  42 __divsf3:
  43         push_s blink
  44         push_s r1
  45         bl.d __divsf3_c
  46         push_s r0
  47         ld_s r1,[sp,4]
  48         st_s r0,[sp,4]
  49         bl.d __divsf3_asm
  50         pop_s r0
  51         pop_s r1
  52         pop_s blink
  53         cmp r0,r1
  54 #if 1
  55         bne abort
  56         jeq_s [blink]
  57         b abort
  58 #else
  59         bne abort
  60         j_s [blink]
  61 #endif
  62         ENDFUNC(__divsf3)
  63 #define __divsf3 __divsf3_asm
  64 #endif /* DEBUG */
  65
  66         FUNC(__divsf3)
  67         .balign 4
  68 .Ldivtab:
  69         .long 0xfc0ffff0
  70         .long 0xf46ffefd
  71         .long 0xed1ffd2a
  72         .long 0xe627fa8e
  73         .long 0xdf7ff73b
  74         .long 0xd917f33b
  75         .long 0xd2f7eea3
  76         .long 0xcd1fe986
  77         .long 0xc77fe3e7
  78         .long 0xc21fdddb
  79         .long 0xbcefd760
  80         .long 0xb7f7d08c
  81         .long 0xb32fc960
  82         .long 0xae97c1ea
  83         .long 0xaa27ba26
  84         .long 0xa5e7b22e
  85         .long 0xa1cfa9fe
  86         .long 0x9ddfa1a0
  87         .long 0x9a0f990c
  88         .long 0x9667905d
  89         .long 0x92df878a
  90         .long 0x8f6f7e84
  91         .long 0x8c27757e
  92         .long 0x88f76c54
  93         .long 0x85df630c
  94         .long 0x82e759c5
  95         .long 0x8007506d
  96         .long 0x7d3f470a
  97         .long 0x7a8f3da2
  98         .long 0x77ef341e
  99         .long 0x756f2abe
 100         .long 0x72f7212d
 101         .long 0x709717ad
 102         .long 0x6e4f0e44
 103         .long 0x6c1704d6
 104         .long 0x69e6fb44
 105         .long 0x67cef1d7
 106         .long 0x65c6e872
 107         .long 0x63cedf18
 108         .long 0x61e6d5cd
 109         .long 0x6006cc6d
 110         .long 0x5e36c323
 111         .long 0x5c76b9f3
 112         .long 0x5abeb0b7
 113         .long 0x5916a79b
 114         .long 0x57769e77
 115         .long 0x55de954d
 116         .long 0x54568c4e
 117         .long 0x52d6834d
 118         .long 0x51667a7f
 119         .long 0x4ffe71b5
 120         .long 0x4e9e68f1
 121         .long 0x4d466035
 122         .long 0x4bf65784
 123         .long 0x4aae4ede
 124         .long 0x496e4646
 125         .long 0x48363dbd
 126         .long 0x47063547
 127         .long 0x45de2ce5
 128         .long 0x44be2498
 129         .long 0x43a61c64
 130         .long 0x4296144a
 131         .long 0x41860c0e
 132         .long 0x407e03ee
 133 .L7f800000:
 134         .long 0x7f800000
 135         .balign 4
 136         .global __divsf3_support
 137 __divsf3_support:
 138 .Linf_NaN:
 139         bclr.f 0,r0,31 ; 0/0 -> NaN
 140         xor_s r0,r0,r1
 141         bmsk r1,r0,30
 142         bic_s r0,r0,r1
 143         sub.eq r0,r0,1
 144         j_s.d [blink]
 145         or r0,r0,r9
 146 .Lret0:
 147         xor_s r0,r0,r1
 148         bmsk r1,r0,30
 149         j_s.d [blink]
 150         bic_s r0,r0,r1
 151 /* N.B. the spacing between divtab and the sub3 to get its address must
 152    be a multiple of 8.  */
 153 __divsf3:
 154         lsr r2,r1,17
 155         sub3 r3,pcl,37 ; (.-.Ldivtab) >> 3
 156         bmsk_s r2,r2,5
 157         ld.as r5,[r3,r2]
 158         asl r4,r1,9
 159         ld.as r9,[pcl,-13]; [pcl,(-((.-.L7f800000) >> 2))] ; 0x7f800000
 160         mulu64 r5,r4
 161         and.f r11,r1,r9
 162         asl r6,r1,8
 163         bset r6,r6,31
 164         beq.d .Ldenorm_fp1
 165         asl r5,r5,13
 166         breq.d r11,r9,.Linf_nan_fp1
 167         and.f r2,r0,r9
 168         sub r7,r5,mhi
 169         mulu64 r7,r6
 170         beq.d .Ldenorm_fp0
 171         asl r12,r0,8
 172         breq.d r2,r9,.Linf_nan_fp0
 173         mulu64 mhi,r7
 174 .Lpast_denorm_fp1:
 175         bset r3,r12,31
 176 .Lpast_denorm_fp0:
 177         cmp_s r3,r6
 178         lsr.cc r3,r3,1
 179         add_s r2,r2, /* wait for immediate */ \
 180                 0x3f000000
 181         sub r7,r7,mhi ; u1.31 inverse, about 30 bit
 182         mulu64 r3,r7
 183         sbc r2,r2,r11
 184         xor.f 0,r0,r1
 185         and r0,r2,r9
 186         bclr r3,r9,23 ; 0x7f000000
 187         brhs.d r2,r3,.Linf_denorm
 188         bxor.mi r0,r0,31
 189 .Lpast_denorm:
 190         add r3,mhi,0x22 ; round to nearest or higher
 191         tst r3,0x3c ; check if rounding was unsafe
 192         lsr r3,r3,6
 193         jne.d [blink] ; return if rounding was safe.
 194         add_s r0,r0,r3
 195         /* work out exact rounding if we fall through here.  */
 196         /* We know that the exact result cannot be represented in single
 197            precision.  Find the mid-point between the two nearest
 198            representable values, multiply with the divisor, and check if
 199            the result is larger than the dividend.  */
 200         add_s r3,r3,r3
 201         sub_s r3,r3,1
 202         mulu64 r3,r6
 203         asr.f 0,r0,1 ; for round-to-even in case this is a denorm
 204         rsub r2,r9,25
 205         asl_s r12,r12,r2
 206         sub.f 0,r12,mlo
 207         j_s.d [blink]
 208         sub.mi r0,r0,1
 209 .Linf_nan_fp1:
 210         lsr_s r0,r0,31
 211         bmsk.f 0,r1,22
 212         asl_s r0,r0,31
 213         bne_s 0f ; inf/inf -> nan
 214         brne r2,r9,.Lsigned0 ; x/inf -> 0, but x/nan -> nan
 215 0:      j_s.d [blink]
 216         mov r0,-1
 217 .Lsigned0:
 218 .Linf_nan_fp0:
 219         tst_s r1,r1
 220         j_s.d [blink]
 221         bxor.mi r0,r0,31
 222         .balign 4
 223         .global __divsf3
 224 /* For denormal results, it is possible that an exact result needs
 225    rounding, and thus the round-to-even rule has to come into play.  */
 226 .Linf_denorm:
 227         brlo r2,0xc0000000,.Linf
 228 .Ldenorm:
 229         asr_s r2,r2,23
 230         bic r0,r0,r9
 231         neg r9,r2
 232         brlo.d r9,25,.Lpast_denorm
 233         lsr r3,mlo,r9
 234         /* Fall through: return +- 0 */
 235         j_s [blink]
 236 .Linf:
 237         j_s.d [blink]
 238         or r0,r0,r9
 239         .balign 4
 240 .Ldenorm_fp1:
 241         bclr r6,r6,31
 242         norm.f r12,r6 ; flag for x/0 -> Inf check
 243         add r6,r6,r6
 244         rsub r5,r12,16
 245         ror r5,r1,r5
 246         asl r6,r6,r12
 247         bmsk r5,r5,5
 248         ld.as r5,[r3,r5]
 249         add r4,r6,r6
 250         ; load latency
 251         mulu64 r5,r4
 252         bic.ne.f 0, \
 253                 0x60000000,r0 ; large number / denorm -> Inf
 254         asl r5,r5,13
 255         sub r7,r5,mhi
 256         beq.d .Linf_NaN
 257         mulu64 r7,r6
 258         asl_s r12,r12,23
 259         and.f r2,r0,r9
 260         add_s r2,r2,r12
 261         asl r12,r0,8
 262         bne.d .Lpast_denorm_fp1
 263 .Ldenorm_fp0: mulu64 mhi,r7
 264         bclr r12,r12,31
 265         norm.f r3,r12 ; flag for 0/x -> 0 check
 266         bic.ne.f 0,0x60000000,r1 ; denorm/large number -> 0
 267         beq_s .Lret0
 268         asl_s r12,r12,r3
 269         asl_s r3,r3,23
 270         add_s r12,r12,r12
 271         add r11,r11,r3
 272         b.d .Lpast_denorm_fp0
 273         mov_s r3,r12
 274         ENDFUNC(__divsf3)