libgcc/config/arc/ieee-754/mulsf3.S

   1 /* Copyright (C) 2008-2019 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 /* XMAC schedule: directly back-to-back multiplies stall; the third
  27    instruction after a multiply stalls unless it is also a multiply.  */
  28 #include "arc-ieee-754.h"
  29
  30 #if 0 /* DEBUG */
  31         .global __mulsf3
  32         FUNC(__mulsf3)
  33         .balign 4
  34 __mulsf3:
  35         push_s blink
  36         push_s r1
  37         bl.d __mulsf3_c
  38         push_s r0
  39         ld_s r1,[sp,4]
  40         st_s r0,[sp,4]
  41         bl.d __mulsf3_asm
  42         pop_s r0
  43         pop_s r1
  44         pop_s blink
  45         cmp r0,r1
  46         jeq_s [blink]
  47         and r12,r0,r1
  48         bic.f 0,0x7f800000,r12
  49         bne 0f
  50         bmsk.f 0,r0,22
  51         bmsk.ne.f r1,r1,22
  52         jne_s [blink] ; both NaN -> OK
  53 0:      bl abort
  54         ENDFUNC(__mulsf3)
  55 #define __mulsf3 __mulsf3_asm
  56 #endif /* DEBUG */
  57
  58         .balign 4
  59         .global __mulsf3
  60         FUNC(__mulsf3)
  61 __mulsf3:
  62         ld.as   r9,[pcl,79]; [pcl,((.L7f800000-.+2)/4)]
  63         bmsk    r4,r1,22
  64         bset    r2,r0,23
  65         asl_s   r2,r2,8
  66         bset    r3,r4,23
  67         MPYHU   r6,r2,r3
  68         and     r11,r0,r9
  69         breq    r11,0,.Ldenorm_dbl0
  70         mpyu    r7,r2,r3
  71         breq    r11,r9,.Linf_nan_dbl0
  72         and     r12,r1,r9
  73         asl.f   0,r6,8
  74         breq    r12,0,.Ldenorm_dbl1
  75 .Lpast_denorm:
  76         xor_s   r0,r0,r1
  77 .Lpast_denorm_dbl1:
  78         add.pl  r6,r6,r6
  79         bclr.pl r6,r6,23
  80         add.pl.f r7,r7,r7
  81         ld.as   r4,[pcl,64]; [pcl,((.L7fffffff-.+2)/4)]
  82         add.cs  r6,r6,1
  83         lsr.f   0,r6,1
  84         breq    r12,r9,.Linf_nan_dbl1
  85         add_s   r12,r12,r11
  86         adc.f   0,r7,r4
  87         add_s   r12,r12, \
  88                 -0x3f800000
  89         adc.f   r8,r6,r12
  90         bic     r0,r0,r4
  91         tst.pl  r8,r9
  92         min     r3,r8,r9
  93         jpnz.d  [blink]
  94         add.pnz r0,r0,r3
  95 ; infinity or denormal number
  96         add.ne.f r3,r3,r3
  97         bpnz    .Linfinity
  98         asr_s   r3,r3,23+1
  99         bset    r6,r6,23
 100         sub_s   r3,r3,1
 101         neg_s   r2,r3
 102         brhi    r2,24,.Lret_r0 ; right shift shift > 24 -> return +-0
 103         lsr     r2,r6,r2
 104         asl     r9,r6,r3
 105         lsr.f   0,r2,1
 106         tst     r7,r7
 107         add_s   r0,r0,r2
 108         bset.ne r9,r9,0
 109         adc.f   0,r9,r4
 110         j_s.d   [blink]
 111         add.cs  r0,r0,1
 112 .Linfinity:
 113         j_s.d   [blink]
 114         add_s   r0,r0,r9
 115
 116 .Lret_r0: j_s [blink]
 117
 118         .balign 4
 119 .Linf_nan_dbl0:
 120         sub_s   r2,r1,1 ; inf/nan * 0 -> nan; inf * nan -> nan (use |r2| >= inf)
 121         bic.f   0,r9,r2
 122         xor_s   r0,r0,r1
 123         bclr_s  r1,r1,31
 124         xor_s   r0,r0,r1
 125         jne_s   [blink]
 126 .Lretnan:
 127         j_s.d   [blink]
 128         mov     r0,-1
 129 .Ldenorm_dbl0_inf_nan_dbl1:
 130         bmsk.f  0,r0,30
 131         beq_s   .Lretnan
 132         xor_s   r0,r0,r1
 133 .Linf_nan_dbl1:
 134         xor_s   r1,r1,r0
 135         bclr_s  r1,r1,31
 136         j_s.d   [blink]
 137         xor_s   r0,r0,r1
 138
 139         .balign 4
 140 .Ldenorm_dbl0:
 141         bclr_s  r2,r2,31
 142         norm.f  r4,r2
 143         and     r12,r1,r9
 144         add_s   r2,r2,r2
 145         asl     r2,r2,r4
 146         asl     r4,r4,23
 147         MPYHU   r6,r2,r3
 148         breq    r12,r9,.Ldenorm_dbl0_inf_nan_dbl1
 149         sub.ne.f r12,r12,r4
 150         mpyu    r7,r2,r3
 151         bhi.d   .Lpast_denorm
 152         asl.f   0,r6,8
 153         xor_s   r0,r0,r1
 154         bmsk    r1,r0,30
 155         j_s.d   [blink]
 156         bic_l   r0,r0,r1
 157
 158         .balign 4
 159 .Ldenorm_dbl1:
 160         norm.f  r3,r4
 161         xor_s   r0,r0,r1
 162         sub_s   r3,r3,7
 163         asl     r4,r4,r3
 164         sub_s   r3,r3,1
 165         asl_s   r3,r3,23
 166         MPYHU   r6,r2,r4
 167         sub.ne.f r11,r11,r3
 168         bmsk    r8,r0,30
 169         mpyu    r7,r2,r4
 170         bhi.d   .Lpast_denorm_dbl1
 171         asl.f   0,r6,8
 172         j_s.d   [blink]
 173         bic     r0,r0,r8
 174
 175         .balign 4
 176 .L7f800000:
 177         .long   0x7f800000
 178 .L7fffffff:
 179         .long   0x7fffffff
 180         ENDFUNC(__mulsf3)