OpenCores
URL https://opencores.org/ocsvn/openrisc/openrisc/trunk

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [rtos/] [rtems/] [c/] [src/] [lib/] [libcpu/] [m68k/] [m68040/] [fpsp/] [res_func.S] - Blame information for rev 173

Details | Compare with Previous | View Log

Line No. Rev Author Line
1 30 unneback
//
2
//      $Id: res_func.S,v 1.2 2001-09-27 12:01:22 chris Exp $
3
//
4
//      res_func.sa 3.9 7/29/91
5
//
6
// Normalizes denormalized numbers if necessary and updates the
7
// stack frame.  The function is then restored back into the
8
// machine and the 040 completes the operation.  This routine
9
// is only used by the unsupported data type/format handler.
10
// (Exception vector 55).
11
//
12
// For packed move out (fmove.p fpm,) the operation is
13
// completed here; data is packed and moved to user memory.
14
// The stack is restored to the 040 only in the case of a
15
// reportable exception in the conversion.
16
//
17
//
18
//              Copyright (C) Motorola, Inc. 1990
19
//                      All Rights Reserved
20
//
21
//      THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA
22
//      The copyright notice above does not evidence any
23
//      actual or intended publication of such source code.
24
 
25
RES_FUNC:    //idnt    2,1 | Motorola 040 Floating Point Software Package
26
 
27
        |section        8
28
 
29
#include "fpsp.defs"
30
 
31
sp_bnds:        .short  0x3f81,0x407e
32
                .short  0x3f6a,0x0000
33
dp_bnds:        .short  0x3c01,0x43fe
34
                .short  0x3bcd,0x0000
35
 
36
        |xref   mem_write
37
        |xref   bindec
38
        |xref   get_fline
39
        |xref   round
40
        |xref   denorm
41
        |xref   dest_ext
42
        |xref   dest_dbl
43
        |xref   dest_sgl
44
        |xref   unf_sub
45
        |xref   nrm_set
46
        |xref   dnrm_lp
47
        |xref   ovf_res
48
        |xref   reg_dest
49
        |xref   t_ovfl
50
        |xref   t_unfl
51
 
52
        .global res_func
53
        .global         p_move
54
 
55
res_func:
56
        clrb    DNRM_FLG(%a6)
57
        clrb    RES_FLG(%a6)
58
        clrb    CU_ONLY(%a6)
59
        tstb    DY_MO_FLG(%a6)
60
        beqs    monadic
61
dyadic:
62
        btstb   #7,DTAG(%a6)    //if dop = norm=000, zero=001,
63
//                              ;inf=010 or nan=011
64
        beqs    monadic         //then branch
65
//                              ;else denorm
66
// HANDLE DESTINATION DENORM HERE
67
//                              ;set dtag to norm
68
//                              ;write the tag & fpte15 to the fstack
69
        leal    FPTEMP(%a6),%a0
70
 
71
        bclrb   #sign_bit,LOCAL_EX(%a0)
72
        sne     LOCAL_SGN(%a0)
73
 
74
        bsr     nrm_set         //normalize number (exp will go negative)
75
        bclrb   #sign_bit,LOCAL_EX(%a0) //get rid of false sign
76
        bfclr   LOCAL_SGN(%a0){#0:#8}   //change back to IEEE ext format
77
        beqs    dpos
78
        bsetb   #sign_bit,LOCAL_EX(%a0)
79
dpos:
80
        bfclr   DTAG(%a6){#0:#4}        //set tag to normalized, FPTE15 = 0
81
        bsetb   #4,DTAG(%a6)    //set FPTE15
82
        orb     #0x0f,DNRM_FLG(%a6)
83
monadic:
84
        leal    ETEMP(%a6),%a0
85
        btstb   #direction_bit,CMDREG1B(%a6)    //check direction
86
        bne     opclass3                        //it is a mv out
87
//
88
// At this point, only opclass 0 and 2 possible
89
//
90
        btstb   #7,STAG(%a6)    //if sop = norm=000, zero=001,
91
//                              ;inf=010 or nan=011
92
        bne     mon_dnrm        //else denorm
93
        tstb    DY_MO_FLG(%a6)  //all cases of dyadic instructions would
94
        bne     normal          //require normalization of denorm
95
 
96
// At this point:
97
//      monadic instructions:   fabs  = $18  fneg   = $1a  ftst   = $3a
98
//                              fmove = $00  fsmove = $40  fdmove = $44
99
//                              fsqrt = $05* fssqrt = $41  fdsqrt = $45
100
//                              (*fsqrt reencoded to $05)
101
//
102
        movew   CMDREG1B(%a6),%d0       //get command register
103
        andil   #0x7f,%d0                       //strip to only command word
104
//
105
// At this point, fabs, fneg, fsmove, fdmove, ftst, fsqrt, fssqrt, and
106
// fdsqrt are possible.
107
// For cases fabs, fneg, fsmove, and fdmove goto spos (do not normalize)
108
// For cases fsqrt, fssqrt, and fdsqrt goto nrm_src (do normalize)
109
//
110
        btstl   #0,%d0
111
        bne     normal                  //weed out fsqrt instructions
112
//
113
// cu_norm handles fmove in instructions with normalized inputs.
114
// The routine round is used to correctly round the input for the
115
// destination precision and mode.
116
//
117
cu_norm:
118
        st      CU_ONLY(%a6)            //set cu-only inst flag
119
        movew   CMDREG1B(%a6),%d0
120
        andib   #0x3b,%d0               //isolate bits to select inst
121
        tstb    %d0
122
        beql    cu_nmove        //if zero, it is an fmove
123
        cmpib   #0x18,%d0
124
        beql    cu_nabs         //if $18, it is fabs
125
        cmpib   #0x1a,%d0
126
        beql    cu_nneg         //if $1a, it is fneg
127
//
128
// Inst is ftst.  Check the source operand and set the cc's accordingly.
129
// No write is done, so simply rts.
130
//
131
cu_ntst:
132
        movew   LOCAL_EX(%a0),%d0
133
        bclrl   #15,%d0
134
        sne     LOCAL_SGN(%a0)
135
        beqs    cu_ntpo
136
        orl     #neg_mask,USER_FPSR(%a6) //set N
137
cu_ntpo:
138
        cmpiw   #0x7fff,%d0     //test for inf/nan
139
        bnes    cu_ntcz
140
        tstl    LOCAL_HI(%a0)
141
        bnes    cu_ntn
142
        tstl    LOCAL_LO(%a0)
143
        bnes    cu_ntn
144
        orl     #inf_mask,USER_FPSR(%a6)
145
        rts
146
cu_ntn:
147
        orl     #nan_mask,USER_FPSR(%a6)
148
        movel   ETEMP_EX(%a6),FPTEMP_EX(%a6)    //set up fptemp sign for
149
//                                              ;snan handler
150
 
151
        rts
152
cu_ntcz:
153
        tstl    LOCAL_HI(%a0)
154
        bnel    cu_ntsx
155
        tstl    LOCAL_LO(%a0)
156
        bnel    cu_ntsx
157
        orl     #z_mask,USER_FPSR(%a6)
158
cu_ntsx:
159
        rts
160
//
161
// Inst is fabs.  Execute the absolute value function on the input.
162
// Branch to the fmove code.  If the operand is NaN, do nothing.
163
//
164
cu_nabs:
165
        moveb   STAG(%a6),%d0
166
        btstl   #5,%d0                  //test for NaN or zero
167
        bne     wr_etemp                //if either, simply write it
168
        bclrb   #7,LOCAL_EX(%a0)                //do abs
169
        bras    cu_nmove                //fmove code will finish
170
//
171
// Inst is fneg.  Execute the negate value function on the input.
172
// Fall though to the fmove code.  If the operand is NaN, do nothing.
173
//
174
cu_nneg:
175
        moveb   STAG(%a6),%d0
176
        btstl   #5,%d0                  //test for NaN or zero
177
        bne     wr_etemp                //if either, simply write it
178
        bchgb   #7,LOCAL_EX(%a0)                //do neg
179
//
180
// Inst is fmove.  This code also handles all result writes.
181
// If bit 2 is set, round is forced to double.  If it is clear,
182
// and bit 6 is set, round is forced to single.  If both are clear,
183
// the round precision is found in the fpcr.  If the rounding precision
184
// is double or single, round the result before the write.
185
//
186
cu_nmove:
187
        moveb   STAG(%a6),%d0
188
        andib   #0xe0,%d0                       //isolate stag bits
189
        bne     wr_etemp                //if not norm, simply write it
190
        btstb   #2,CMDREG1B+1(%a6)      //check for rd
191
        bne     cu_nmrd
192
        btstb   #6,CMDREG1B+1(%a6)      //check for rs
193
        bne     cu_nmrs
194
//
195
// The move or operation is not with forced precision.  Test for
196
// nan or inf as the input; if so, simply write it to FPn.  Use the
197
// FPCR_MODE byte to get rounding on norms and zeros.
198
//
199
cu_nmnr:
200
        bfextu  FPCR_MODE(%a6){#0:#2},%d0
201
        tstb    %d0                     //check for extended
202
        beq     cu_wrexn                //if so, just write result
203
        cmpib   #1,%d0                  //check for single
204
        beq     cu_nmrs                 //fall through to double
205
//
206
// The move is fdmove or round precision is double.
207
//
208
cu_nmrd:
209
        movel   #2,%d0                  //set up the size for denorm
210
        movew   LOCAL_EX(%a0),%d1               //compare exponent to double threshold
211
        andw    #0x7fff,%d1
212
        cmpw    #0x3c01,%d1
213
        bls     cu_nunfl
214
        bfextu  FPCR_MODE(%a6){#2:#2},%d1       //get rmode
215
        orl     #0x00020000,%d1         //or in rprec (double)
216
        clrl    %d0                     //clear g,r,s for round
217
        bclrb   #sign_bit,LOCAL_EX(%a0) //convert to internal format
218
        sne     LOCAL_SGN(%a0)
219
        bsrl    round
220
        bfclr   LOCAL_SGN(%a0){#0:#8}
221
        beqs    cu_nmrdc
222
        bsetb   #sign_bit,LOCAL_EX(%a0)
223
cu_nmrdc:
224
        movew   LOCAL_EX(%a0),%d1               //check for overflow
225
        andw    #0x7fff,%d1
226
        cmpw    #0x43ff,%d1
227
        bge     cu_novfl                //take care of overflow case
228
        bra     cu_wrexn
229
//
230
// The move is fsmove or round precision is single.
231
//
232
cu_nmrs:
233
        movel   #1,%d0
234
        movew   LOCAL_EX(%a0),%d1
235
        andw    #0x7fff,%d1
236
        cmpw    #0x3f81,%d1
237
        bls     cu_nunfl
238
        bfextu  FPCR_MODE(%a6){#2:#2},%d1
239
        orl     #0x00010000,%d1
240
        clrl    %d0
241
        bclrb   #sign_bit,LOCAL_EX(%a0)
242
        sne     LOCAL_SGN(%a0)
243
        bsrl    round
244
        bfclr   LOCAL_SGN(%a0){#0:#8}
245
        beqs    cu_nmrsc
246
        bsetb   #sign_bit,LOCAL_EX(%a0)
247
cu_nmrsc:
248
        movew   LOCAL_EX(%a0),%d1
249
        andw    #0x7FFF,%d1
250
        cmpw    #0x407f,%d1
251
        blt     cu_wrexn
252
//
253
// The operand is above precision boundaries.  Use t_ovfl to
254
// generate the correct value.
255
//
256
cu_novfl:
257
        bsr     t_ovfl
258
        bra     cu_wrexn
259
//
260
// The operand is below precision boundaries.  Use denorm to
261
// generate the correct value.
262
//
263
cu_nunfl:
264
        bclrb   #sign_bit,LOCAL_EX(%a0)
265
        sne     LOCAL_SGN(%a0)
266
        bsr     denorm
267
        bfclr   LOCAL_SGN(%a0){#0:#8}   //change back to IEEE ext format
268
        beqs    cu_nucont
269
        bsetb   #sign_bit,LOCAL_EX(%a0)
270
cu_nucont:
271
        bfextu  FPCR_MODE(%a6){#2:#2},%d1
272
        btstb   #2,CMDREG1B+1(%a6)      //check for rd
273
        bne     inst_d
274
        btstb   #6,CMDREG1B+1(%a6)      //check for rs
275
        bne     inst_s
276
        swap    %d1
277
        moveb   FPCR_MODE(%a6),%d1
278
        lsrb    #6,%d1
279
        swap    %d1
280
        bra     inst_sd
281
inst_d:
282
        orl     #0x00020000,%d1
283
        bra     inst_sd
284
inst_s:
285
        orl     #0x00010000,%d1
286
inst_sd:
287
        bclrb   #sign_bit,LOCAL_EX(%a0)
288
        sne     LOCAL_SGN(%a0)
289
        bsrl    round
290
        bfclr   LOCAL_SGN(%a0){#0:#8}
291
        beqs    cu_nuflp
292
        bsetb   #sign_bit,LOCAL_EX(%a0)
293
cu_nuflp:
294
        btstb   #inex2_bit,FPSR_EXCEPT(%a6)
295
        beqs    cu_nuninx
296
        orl     #aunfl_mask,USER_FPSR(%a6) //if the round was inex, set AUNFL
297
cu_nuninx:
298
        tstl    LOCAL_HI(%a0)           //test for zero
299
        bnes    cu_nunzro
300
        tstl    LOCAL_LO(%a0)
301
        bnes    cu_nunzro
302
//
303
// The mantissa is zero from the denorm loop.  Check sign and rmode
304
// to see if rounding should have occurred which would leave the lsb.
305
//
306
        movel   USER_FPCR(%a6),%d0
307
        andil   #0x30,%d0               //isolate rmode
308
        cmpil   #0x20,%d0
309
        blts    cu_nzro
310
        bnes    cu_nrp
311
cu_nrm:
312
        tstw    LOCAL_EX(%a0)   //if positive, set lsb
313
        bges    cu_nzro
314
        btstb   #7,FPCR_MODE(%a6) //check for double
315
        beqs    cu_nincs
316
        bras    cu_nincd
317
cu_nrp:
318
        tstw    LOCAL_EX(%a0)   //if positive, set lsb
319
        blts    cu_nzro
320
        btstb   #7,FPCR_MODE(%a6) //check for double
321
        beqs    cu_nincs
322
cu_nincd:
323
        orl     #0x800,LOCAL_LO(%a0) //inc for double
324
        bra     cu_nunzro
325
cu_nincs:
326
        orl     #0x100,LOCAL_HI(%a0) //inc for single
327
        bra     cu_nunzro
328
cu_nzro:
329
        orl     #z_mask,USER_FPSR(%a6)
330
        moveb   STAG(%a6),%d0
331
        andib   #0xe0,%d0
332
        cmpib   #0x40,%d0               //check if input was tagged zero
333
        beqs    cu_numv
334
cu_nunzro:
335
        orl     #unfl_mask,USER_FPSR(%a6) //set unfl
336
cu_numv:
337
        movel   (%a0),ETEMP(%a6)
338
        movel   4(%a0),ETEMP_HI(%a6)
339
        movel   8(%a0),ETEMP_LO(%a6)
340
//
341
// Write the result to memory, setting the fpsr cc bits.  NaN and Inf
342
// bypass cu_wrexn.
343
//
344
cu_wrexn:
345
        tstw    LOCAL_EX(%a0)           //test for zero
346
        beqs    cu_wrzero
347
        cmpw    #0x8000,LOCAL_EX(%a0)   //test for zero
348
        bnes    cu_wreon
349
cu_wrzero:
350
        orl     #z_mask,USER_FPSR(%a6)  //set Z bit
351
cu_wreon:
352
        tstw    LOCAL_EX(%a0)
353
        bpl     wr_etemp
354
        orl     #neg_mask,USER_FPSR(%a6)
355
        bra     wr_etemp
356
 
357
//
358
// HANDLE SOURCE DENORM HERE
359
//
360
//                              ;clear denorm stag to norm
361
//                              ;write the new tag & ete15 to the fstack
362
mon_dnrm:
363
//
364
// At this point, check for the cases in which normalizing the
365
// denorm produces incorrect results.
366
//
367
        tstb    DY_MO_FLG(%a6)  //all cases of dyadic instructions would
368
        bnes    nrm_src         //require normalization of denorm
369
 
370
// At this point:
371
//      monadic instructions:   fabs  = $18  fneg   = $1a  ftst   = $3a
372
//                              fmove = $00  fsmove = $40  fdmove = $44
373
//                              fsqrt = $05* fssqrt = $41  fdsqrt = $45
374
//                              (*fsqrt reencoded to $05)
375
//
376
        movew   CMDREG1B(%a6),%d0       //get command register
377
        andil   #0x7f,%d0                       //strip to only command word
378
//
379
// At this point, fabs, fneg, fsmove, fdmove, ftst, fsqrt, fssqrt, and
380
// fdsqrt are possible.
381
// For cases fabs, fneg, fsmove, and fdmove goto spos (do not normalize)
382
// For cases fsqrt, fssqrt, and fdsqrt goto nrm_src (do normalize)
383
//
384
        btstl   #0,%d0
385
        bnes    nrm_src         //weed out fsqrt instructions
386
        st      CU_ONLY(%a6)    //set cu-only inst flag
387
        bra     cu_dnrm         //fmove, fabs, fneg, ftst
388
//                              ;cases go to cu_dnrm
389
nrm_src:
390
        bclrb   #sign_bit,LOCAL_EX(%a0)
391
        sne     LOCAL_SGN(%a0)
392
        bsr     nrm_set         //normalize number (exponent will go
393
//                              ; negative)
394
        bclrb   #sign_bit,LOCAL_EX(%a0) //get rid of false sign
395
 
396
        bfclr   LOCAL_SGN(%a0){#0:#8}   //change back to IEEE ext format
397
        beqs    spos
398
        bsetb   #sign_bit,LOCAL_EX(%a0)
399
spos:
400
        bfclr   STAG(%a6){#0:#4}        //set tag to normalized, FPTE15 = 0
401
        bsetb   #4,STAG(%a6)    //set ETE15
402
        orb     #0xf0,DNRM_FLG(%a6)
403
normal:
404
        tstb    DNRM_FLG(%a6)   //check if any of the ops were denorms
405
        bne     ck_wrap         //if so, check if it is a potential
406
//                              ;wrap-around case
407
fix_stk:
408
        moveb   #0xfe,CU_SAVEPC(%a6)
409
        bclrb   #E1,E_BYTE(%a6)
410
 
411
        clrw    NMNEXC(%a6)
412
 
413
        st      RES_FLG(%a6)    //indicate that a restore is needed
414
        rts
415
 
416
//
417
// cu_dnrm handles all cu-only instructions (fmove, fabs, fneg, and
418
// ftst) completely in software without an frestore to the 040.
419
//
420
cu_dnrm:
421
        st      CU_ONLY(%a6)
422
        movew   CMDREG1B(%a6),%d0
423
        andib   #0x3b,%d0               //isolate bits to select inst
424
        tstb    %d0
425
        beql    cu_dmove        //if zero, it is an fmove
426
        cmpib   #0x18,%d0
427
        beql    cu_dabs         //if $18, it is fabs
428
        cmpib   #0x1a,%d0
429
        beql    cu_dneg         //if $1a, it is fneg
430
//
431
// Inst is ftst.  Check the source operand and set the cc's accordingly.
432
// No write is done, so simply rts.
433
//
434
cu_dtst:
435
        movew   LOCAL_EX(%a0),%d0
436
        bclrl   #15,%d0
437
        sne     LOCAL_SGN(%a0)
438
        beqs    cu_dtpo
439
        orl     #neg_mask,USER_FPSR(%a6) //set N
440
cu_dtpo:
441
        cmpiw   #0x7fff,%d0     //test for inf/nan
442
        bnes    cu_dtcz
443
        tstl    LOCAL_HI(%a0)
444
        bnes    cu_dtn
445
        tstl    LOCAL_LO(%a0)
446
        bnes    cu_dtn
447
        orl     #inf_mask,USER_FPSR(%a6)
448
        rts
449
cu_dtn:
450
        orl     #nan_mask,USER_FPSR(%a6)
451
        movel   ETEMP_EX(%a6),FPTEMP_EX(%a6)    //set up fptemp sign for
452
//                                              ;snan handler
453
        rts
454
cu_dtcz:
455
        tstl    LOCAL_HI(%a0)
456
        bnel    cu_dtsx
457
        tstl    LOCAL_LO(%a0)
458
        bnel    cu_dtsx
459
        orl     #z_mask,USER_FPSR(%a6)
460
cu_dtsx:
461
        rts
462
//
463
// Inst is fabs.  Execute the absolute value function on the input.
464
// Branch to the fmove code.
465
//
466
cu_dabs:
467
        bclrb   #7,LOCAL_EX(%a0)                //do abs
468
        bras    cu_dmove                //fmove code will finish
469
//
470
// Inst is fneg.  Execute the negate value function on the input.
471
// Fall though to the fmove code.
472
//
473
cu_dneg:
474
        bchgb   #7,LOCAL_EX(%a0)                //do neg
475
//
476
// Inst is fmove.  This code also handles all result writes.
477
// If bit 2 is set, round is forced to double.  If it is clear,
478
// and bit 6 is set, round is forced to single.  If both are clear,
479
// the round precision is found in the fpcr.  If the rounding precision
480
// is double or single, the result is zero, and the mode is checked
481
// to determine if the lsb of the result should be set.
482
//
483
cu_dmove:
484
        btstb   #2,CMDREG1B+1(%a6)      //check for rd
485
        bne     cu_dmrd
486
        btstb   #6,CMDREG1B+1(%a6)      //check for rs
487
        bne     cu_dmrs
488
//
489
// The move or operation is not with forced precision.  Use the
490
// FPCR_MODE byte to get rounding.
491
//
492
cu_dmnr:
493
        bfextu  FPCR_MODE(%a6){#0:#2},%d0
494
        tstb    %d0                     //check for extended
495
        beq     cu_wrexd                //if so, just write result
496
        cmpib   #1,%d0                  //check for single
497
        beq     cu_dmrs                 //fall through to double
498
//
499
// The move is fdmove or round precision is double.  Result is zero.
500
// Check rmode for rp or rm and set lsb accordingly.
501
//
502
cu_dmrd:
503
        bfextu  FPCR_MODE(%a6){#2:#2},%d1       //get rmode
504
        tstw    LOCAL_EX(%a0)           //check sign
505
        blts    cu_dmdn
506
        cmpib   #3,%d1                  //check for rp
507
        bne     cu_dpd                  //load double pos zero
508
        bra     cu_dpdr                 //load double pos zero w/lsb
509
cu_dmdn:
510
        cmpib   #2,%d1                  //check for rm
511
        bne     cu_dnd                  //load double neg zero
512
        bra     cu_dndr                 //load double neg zero w/lsb
513
//
514
// The move is fsmove or round precision is single.  Result is zero.
515
// Check for rp or rm and set lsb accordingly.
516
//
517
cu_dmrs:
518
        bfextu  FPCR_MODE(%a6){#2:#2},%d1       //get rmode
519
        tstw    LOCAL_EX(%a0)           //check sign
520
        blts    cu_dmsn
521
        cmpib   #3,%d1                  //check for rp
522
        bne     cu_spd                  //load single pos zero
523
        bra     cu_spdr                 //load single pos zero w/lsb
524
cu_dmsn:
525
        cmpib   #2,%d1                  //check for rm
526
        bne     cu_snd                  //load single neg zero
527
        bra     cu_sndr                 //load single neg zero w/lsb
528
//
529
// The precision is extended, so the result in etemp is correct.
530
// Simply set unfl (not inex2 or aunfl) and write the result to
531
// the correct fp register.
532
cu_wrexd:
533
        orl     #unfl_mask,USER_FPSR(%a6)
534
        tstw    LOCAL_EX(%a0)
535
        beq     wr_etemp
536
        orl     #neg_mask,USER_FPSR(%a6)
537
        bra     wr_etemp
538
//
539
// These routines write +/- zero in double format.  The routines
540
// cu_dpdr and cu_dndr set the double lsb.
541
//
542
cu_dpd:
543
        movel   #0x3c010000,LOCAL_EX(%a0)       //force pos double zero
544
        clrl    LOCAL_HI(%a0)
545
        clrl    LOCAL_LO(%a0)
546
        orl     #z_mask,USER_FPSR(%a6)
547
        orl     #unfinx_mask,USER_FPSR(%a6)
548
        bra     wr_etemp
549
cu_dpdr:
550
        movel   #0x3c010000,LOCAL_EX(%a0)       //force pos double zero
551
        clrl    LOCAL_HI(%a0)
552
        movel   #0x800,LOCAL_LO(%a0)    //with lsb set
553
        orl     #unfinx_mask,USER_FPSR(%a6)
554
        bra     wr_etemp
555
cu_dnd:
556
        movel   #0xbc010000,LOCAL_EX(%a0)       //force pos double zero
557
        clrl    LOCAL_HI(%a0)
558
        clrl    LOCAL_LO(%a0)
559
        orl     #z_mask,USER_FPSR(%a6)
560
        orl     #neg_mask,USER_FPSR(%a6)
561
        orl     #unfinx_mask,USER_FPSR(%a6)
562
        bra     wr_etemp
563
cu_dndr:
564
        movel   #0xbc010000,LOCAL_EX(%a0)       //force pos double zero
565
        clrl    LOCAL_HI(%a0)
566
        movel   #0x800,LOCAL_LO(%a0)    //with lsb set
567
        orl     #neg_mask,USER_FPSR(%a6)
568
        orl     #unfinx_mask,USER_FPSR(%a6)
569
        bra     wr_etemp
570
//
571
// These routines write +/- zero in single format.  The routines
572
// cu_dpdr and cu_dndr set the single lsb.
573
//
574
cu_spd:
575
        movel   #0x3f810000,LOCAL_EX(%a0)       //force pos single zero
576
        clrl    LOCAL_HI(%a0)
577
        clrl    LOCAL_LO(%a0)
578
        orl     #z_mask,USER_FPSR(%a6)
579
        orl     #unfinx_mask,USER_FPSR(%a6)
580
        bra     wr_etemp
581
cu_spdr:
582
        movel   #0x3f810000,LOCAL_EX(%a0)       //force pos single zero
583
        movel   #0x100,LOCAL_HI(%a0)    //with lsb set
584
        clrl    LOCAL_LO(%a0)
585
        orl     #unfinx_mask,USER_FPSR(%a6)
586
        bra     wr_etemp
587
cu_snd:
588
        movel   #0xbf810000,LOCAL_EX(%a0)       //force pos single zero
589
        clrl    LOCAL_HI(%a0)
590
        clrl    LOCAL_LO(%a0)
591
        orl     #z_mask,USER_FPSR(%a6)
592
        orl     #neg_mask,USER_FPSR(%a6)
593
        orl     #unfinx_mask,USER_FPSR(%a6)
594
        bra     wr_etemp
595
cu_sndr:
596
        movel   #0xbf810000,LOCAL_EX(%a0)       //force pos single zero
597
        movel   #0x100,LOCAL_HI(%a0)    //with lsb set
598
        clrl    LOCAL_LO(%a0)
599
        orl     #neg_mask,USER_FPSR(%a6)
600
        orl     #unfinx_mask,USER_FPSR(%a6)
601
        bra     wr_etemp
602
 
603
//
604
// This code checks for 16-bit overflow conditions on dyadic
605
// operations which are not restorable into the floating-point
606
// unit and must be completed in software.  Basically, this
607
// condition exists with a very large norm and a denorm.  One
608
// of the operands must be denormalized to enter this code.
609
//
610
// Flags used:
611
//      DY_MO_FLG contains 0 for monadic op, $ff for dyadic
612
//      DNRM_FLG contains $00 for neither op denormalized
613
//                        $0f for the destination op denormalized
614
//                        $f0 for the source op denormalized
615
//                        $ff for both ops denormalized
616
//
617
// The wrap-around condition occurs for add, sub, div, and cmp
618
// when
619
//
620
//      abs(dest_exp - src_exp) >= $8000
621
//
622
// and for mul when
623
//
624
//      (dest_exp + src_exp) < $0
625
//
626
// we must process the operation here if this case is true.
627
//
628
// The rts following the frcfpn routine is the exit from res_func
629
// for this condition.  The restore flag (RES_FLG) is left clear.
630
// No frestore is done unless an exception is to be reported.
631
//
632
// For fadd:
633
//      if(sign_of(dest) != sign_of(src))
634
//              replace exponent of src with $3fff (keep sign)
635
//              use fpu to perform dest+new_src (user's rmode and X)
636
//              clr sticky
637
//      else
638
//              set sticky
639
//      call round with user's precision and mode
640
//      move result to fpn and wbtemp
641
//
642
// For fsub:
643
//      if(sign_of(dest) == sign_of(src))
644
//              replace exponent of src with $3fff (keep sign)
645
//              use fpu to perform dest+new_src (user's rmode and X)
646
//              clr sticky
647
//      else
648
//              set sticky
649
//      call round with user's precision and mode
650
//      move result to fpn and wbtemp
651
//
652
// For fdiv/fsgldiv:
653
//      if(both operands are denorm)
654
//              restore_to_fpu;
655
//      if(dest is norm)
656
//              force_ovf;
657
//      else(dest is denorm)
658
//              force_unf:
659
//
660
// For fcmp:
661
//      if(dest is norm)
662
//              N = sign_of(dest);
663
//      else(dest is denorm)
664
//              N = sign_of(src);
665
//
666
// For fmul:
667
//      if(both operands are denorm)
668
//              force_unf;
669
//      if((dest_exp + src_exp) < 0)
670
//              force_unf:
671
//      else
672
//              restore_to_fpu;
673
//
674
// local equates:
675
        .set    addcode,0x22
676
        .set    subcode,0x28
677
        .set    mulcode,0x23
678
        .set    divcode,0x20
679
        .set    cmpcode,0x38
680
ck_wrap:
681
        | tstb  DY_MO_FLG(%a6)  ;check for fsqrt
682
        beq     fix_stk         //if zero, it is fsqrt
683
        movew   CMDREG1B(%a6),%d0
684
        andiw   #0x3b,%d0               //strip to command bits
685
        cmpiw   #addcode,%d0
686
        beq     wrap_add
687
        cmpiw   #subcode,%d0
688
        beq     wrap_sub
689
        cmpiw   #mulcode,%d0
690
        beq     wrap_mul
691
        cmpiw   #cmpcode,%d0
692
        beq     wrap_cmp
693
//
694
// Inst is fdiv.
695
//
696
wrap_div:
697
        cmpb    #0xff,DNRM_FLG(%a6) //if both ops denorm,
698
        beq     fix_stk          //restore to fpu
699
//
700
// One of the ops is denormalized.  Test for wrap condition
701
// and force the result.
702
//
703
        cmpb    #0x0f,DNRM_FLG(%a6) //check for dest denorm
704
        bnes    div_srcd
705
div_destd:
706
        bsrl    ckinf_ns
707
        bne     fix_stk
708
        bfextu  ETEMP_EX(%a6){#1:#15},%d0       //get src exp (always pos)
709
        bfexts  FPTEMP_EX(%a6){#1:#15},%d1      //get dest exp (always neg)
710
        subl    %d1,%d0                 //subtract dest from src
711
        cmpl    #0x7fff,%d0
712
        blt     fix_stk                 //if less, not wrap case
713
        clrb    WBTEMP_SGN(%a6)
714
        movew   ETEMP_EX(%a6),%d0               //find the sign of the result
715
        movew   FPTEMP_EX(%a6),%d1
716
        eorw    %d1,%d0
717
        andiw   #0x8000,%d0
718
        beq     force_unf
719
        st      WBTEMP_SGN(%a6)
720
        bra     force_unf
721
 
722
ckinf_ns:
723
        moveb   STAG(%a6),%d0           //check source tag for inf or nan
724
        bra     ck_in_com
725
ckinf_nd:
726
        moveb   DTAG(%a6),%d0           //check destination tag for inf or nan
727
ck_in_com:
728
        andib   #0x60,%d0                       //isolate tag bits
729
        cmpb    #0x40,%d0                       //is it inf?
730
        beq     nan_or_inf              //not wrap case
731
        cmpb    #0x60,%d0                       //is it nan?
732
        beq     nan_or_inf              //yes, not wrap case?
733
        cmpb    #0x20,%d0                       //is it a zero?
734
        beq     nan_or_inf              //yes
735
        clrl    %d0
736
        rts                             //then ; it is either a zero of norm,
737
//                                      ;check wrap case
738
nan_or_inf:
739
        moveql  #-1,%d0
740
        rts
741
 
742
 
743
 
744
div_srcd:
745
        bsrl    ckinf_nd
746
        bne     fix_stk
747
        bfextu  FPTEMP_EX(%a6){#1:#15},%d0      //get dest exp (always pos)
748
        bfexts  ETEMP_EX(%a6){#1:#15},%d1       //get src exp (always neg)
749
        subl    %d1,%d0                 //subtract src from dest
750
        cmpl    #0x8000,%d0
751
        blt     fix_stk                 //if less, not wrap case
752
        clrb    WBTEMP_SGN(%a6)
753
        movew   ETEMP_EX(%a6),%d0               //find the sign of the result
754
        movew   FPTEMP_EX(%a6),%d1
755
        eorw    %d1,%d0
756
        andiw   #0x8000,%d0
757
        beqs    force_ovf
758
        st      WBTEMP_SGN(%a6)
759
//
760
// This code handles the case of the instruction resulting in
761
// an overflow condition.
762
//
763
force_ovf:
764
        bclrb   #E1,E_BYTE(%a6)
765
        orl     #ovfl_inx_mask,USER_FPSR(%a6)
766
        clrw    NMNEXC(%a6)
767
        leal    WBTEMP(%a6),%a0         //point a0 to memory location
768
        movew   CMDREG1B(%a6),%d0
769
        btstl   #6,%d0                  //test for forced precision
770
        beqs    frcovf_fpcr
771
        btstl   #2,%d0                  //check for double
772
        bnes    frcovf_dbl
773
        movel   #0x1,%d0                        //inst is forced single
774
        bras    frcovf_rnd
775
frcovf_dbl:
776
        movel   #0x2,%d0                        //inst is forced double
777
        bras    frcovf_rnd
778
frcovf_fpcr:
779
        bfextu  FPCR_MODE(%a6){#0:#2},%d0       //inst not forced - use fpcr prec
780
frcovf_rnd:
781
 
782
// The 881/882 does not set inex2 for the following case, so the
783
// line is commented out to be compatible with 881/882
784
//      tst.b   %d0
785
//      beq.b   frcovf_x
786
//      or.l    #inex2_mask,USER_FPSR(%a6) ;if prec is s or d, set inex2
787
 
788
//frcovf_x:
789
        bsrl    ovf_res                 //get correct result based on
790
//                                      ;round precision/mode.  This
791
//                                      ;sets FPSR_CC correctly
792
//                                      ;returns in external format
793
        bfclr   WBTEMP_SGN(%a6){#0:#8}
794
        beq     frcfpn
795
        bsetb   #sign_bit,WBTEMP_EX(%a6)
796
        bra     frcfpn
797
//
798
// Inst is fadd.
799
//
800
wrap_add:
801
        cmpb    #0xff,DNRM_FLG(%a6) //if both ops denorm,
802
        beq     fix_stk          //restore to fpu
803
//
804
// One of the ops is denormalized.  Test for wrap condition
805
// and complete the instruction.
806
//
807
        cmpb    #0x0f,DNRM_FLG(%a6) //check for dest denorm
808
        bnes    add_srcd
809
add_destd:
810
        bsrl    ckinf_ns
811
        bne     fix_stk
812
        bfextu  ETEMP_EX(%a6){#1:#15},%d0       //get src exp (always pos)
813
        bfexts  FPTEMP_EX(%a6){#1:#15},%d1      //get dest exp (always neg)
814
        subl    %d1,%d0                 //subtract dest from src
815
        cmpl    #0x8000,%d0
816
        blt     fix_stk                 //if less, not wrap case
817
        bra     add_wrap
818
add_srcd:
819
        bsrl    ckinf_nd
820
        bne     fix_stk
821
        bfextu  FPTEMP_EX(%a6){#1:#15},%d0      //get dest exp (always pos)
822
        bfexts  ETEMP_EX(%a6){#1:#15},%d1       //get src exp (always neg)
823
        subl    %d1,%d0                 //subtract src from dest
824
        cmpl    #0x8000,%d0
825
        blt     fix_stk                 //if less, not wrap case
826
//
827
// Check the signs of the operands.  If they are unlike, the fpu
828
// can be used to add the norm and 1.0 with the sign of the
829
// denorm and it will correctly generate the result in extended
830
// precision.  We can then call round with no sticky and the result
831
// will be correct for the user's rounding mode and precision.  If
832
// the signs are the same, we call round with the sticky bit set
833
// and the result will be correct for the user's rounding mode and
834
// precision.
835
//
836
add_wrap:
837
        movew   ETEMP_EX(%a6),%d0
838
        movew   FPTEMP_EX(%a6),%d1
839
        eorw    %d1,%d0
840
        andiw   #0x8000,%d0
841
        beq     add_same
842
//
843
// The signs are unlike.
844
//
845
        cmpb    #0x0f,DNRM_FLG(%a6) //is dest the denorm?
846
        bnes    add_u_srcd
847
        movew   FPTEMP_EX(%a6),%d0
848
        andiw   #0x8000,%d0
849
        orw     #0x3fff,%d0     //force the exponent to +/- 1
850
        movew   %d0,FPTEMP_EX(%a6) //in the denorm
851
        movel   USER_FPCR(%a6),%d0
852
        andil   #0x30,%d0
853
        fmovel  %d0,%fpcr               //set up users rmode and X
854
        fmovex  ETEMP(%a6),%fp0
855
        faddx   FPTEMP(%a6),%fp0
856
        leal    WBTEMP(%a6),%a0 //point a0 to wbtemp in frame
857
        fmovel  %fpsr,%d1
858
        orl     %d1,USER_FPSR(%a6) //capture cc's and inex from fadd
859
        fmovex  %fp0,WBTEMP(%a6)        //write result to memory
860
        lsrl    #4,%d0          //put rmode in lower 2 bits
861
        movel   USER_FPCR(%a6),%d1
862
        andil   #0xc0,%d1
863
        lsrl    #6,%d1          //put precision in upper word
864
        swap    %d1
865
        orl     %d0,%d1         //set up for round call
866
        clrl    %d0             //force sticky to zero
867
        bclrb   #sign_bit,WBTEMP_EX(%a6)
868
        sne     WBTEMP_SGN(%a6)
869
        bsrl    round           //round result to users rmode & prec
870
        bfclr   WBTEMP_SGN(%a6){#0:#8}  //convert back to IEEE ext format
871
        beq     frcfpnr
872
        bsetb   #sign_bit,WBTEMP_EX(%a6)
873
        bra     frcfpnr
874
add_u_srcd:
875
        movew   ETEMP_EX(%a6),%d0
876
        andiw   #0x8000,%d0
877
        orw     #0x3fff,%d0     //force the exponent to +/- 1
878
        movew   %d0,ETEMP_EX(%a6) //in the denorm
879
        movel   USER_FPCR(%a6),%d0
880
        andil   #0x30,%d0
881
        fmovel  %d0,%fpcr               //set up users rmode and X
882
        fmovex  ETEMP(%a6),%fp0
883
        faddx   FPTEMP(%a6),%fp0
884
        fmovel  %fpsr,%d1
885
        orl     %d1,USER_FPSR(%a6) //capture cc's and inex from fadd
886
        leal    WBTEMP(%a6),%a0 //point a0 to wbtemp in frame
887
        fmovex  %fp0,WBTEMP(%a6)        //write result to memory
888
        lsrl    #4,%d0          //put rmode in lower 2 bits
889
        movel   USER_FPCR(%a6),%d1
890
        andil   #0xc0,%d1
891
        lsrl    #6,%d1          //put precision in upper word
892
        swap    %d1
893
        orl     %d0,%d1         //set up for round call
894
        clrl    %d0             //force sticky to zero
895
        bclrb   #sign_bit,WBTEMP_EX(%a6)
896
        sne     WBTEMP_SGN(%a6) //use internal format for round
897
        bsrl    round           //round result to users rmode & prec
898
        bfclr   WBTEMP_SGN(%a6){#0:#8}  //convert back to IEEE ext format
899
        beq     frcfpnr
900
        bsetb   #sign_bit,WBTEMP_EX(%a6)
901
        bra     frcfpnr
902
//
903
// Signs are alike:
904
//
905
add_same:
906
        cmpb    #0x0f,DNRM_FLG(%a6) //is dest the denorm?
907
        bnes    add_s_srcd
908
add_s_destd:
909
        leal    ETEMP(%a6),%a0
910
        movel   USER_FPCR(%a6),%d0
911
        andil   #0x30,%d0
912
        lsrl    #4,%d0          //put rmode in lower 2 bits
913
        movel   USER_FPCR(%a6),%d1
914
        andil   #0xc0,%d1
915
        lsrl    #6,%d1          //put precision in upper word
916
        swap    %d1
917
        orl     %d0,%d1         //set up for round call
918
        movel   #0x20000000,%d0 //set sticky for round
919
        bclrb   #sign_bit,ETEMP_EX(%a6)
920
        sne     ETEMP_SGN(%a6)
921
        bsrl    round           //round result to users rmode & prec
922
        bfclr   ETEMP_SGN(%a6){#0:#8}   //convert back to IEEE ext format
923
        beqs    add_s_dclr
924
        bsetb   #sign_bit,ETEMP_EX(%a6)
925
add_s_dclr:
926
        leal    WBTEMP(%a6),%a0
927
        movel   ETEMP(%a6),(%a0)        //write result to wbtemp
928
        movel   ETEMP_HI(%a6),4(%a0)
929
        movel   ETEMP_LO(%a6),8(%a0)
930
        tstw    ETEMP_EX(%a6)
931
        bgt     add_ckovf
932
        orl     #neg_mask,USER_FPSR(%a6)
933
        bra     add_ckovf
934
add_s_srcd:
935
        leal    FPTEMP(%a6),%a0
936
        movel   USER_FPCR(%a6),%d0
937
        andil   #0x30,%d0
938
        lsrl    #4,%d0          //put rmode in lower 2 bits
939
        movel   USER_FPCR(%a6),%d1
940
        andil   #0xc0,%d1
941
        lsrl    #6,%d1          //put precision in upper word
942
        swap    %d1
943
        orl     %d0,%d1         //set up for round call
944
        movel   #0x20000000,%d0 //set sticky for round
945
        bclrb   #sign_bit,FPTEMP_EX(%a6)
946
        sne     FPTEMP_SGN(%a6)
947
        bsrl    round           //round result to users rmode & prec
948
        bfclr   FPTEMP_SGN(%a6){#0:#8}  //convert back to IEEE ext format
949
        beqs    add_s_sclr
950
        bsetb   #sign_bit,FPTEMP_EX(%a6)
951
add_s_sclr:
952
        leal    WBTEMP(%a6),%a0
953
        movel   FPTEMP(%a6),(%a0)       //write result to wbtemp
954
        movel   FPTEMP_HI(%a6),4(%a0)
955
        movel   FPTEMP_LO(%a6),8(%a0)
956
        tstw    FPTEMP_EX(%a6)
957
        bgt     add_ckovf
958
        orl     #neg_mask,USER_FPSR(%a6)
959
add_ckovf:
960
        movew   WBTEMP_EX(%a6),%d0
961
        andiw   #0x7fff,%d0
962
        cmpiw   #0x7fff,%d0
963
        bne     frcfpnr
964
//
965
// The result has overflowed to $7fff exponent.  Set I, ovfl,
966
// and aovfl, and clr the mantissa (incorrectly set by the
967
// round routine.)
968
//
969
        orl     #inf_mask+ovfl_inx_mask,USER_FPSR(%a6)
970
        clrl    4(%a0)
971
        bra     frcfpnr
972
//
973
// Inst is fsub.
974
//
975
wrap_sub:
976
        cmpb    #0xff,DNRM_FLG(%a6) //if both ops denorm,
977
        beq     fix_stk          //restore to fpu
978
//
979
// One of the ops is denormalized.  Test for wrap condition
980
// and complete the instruction.
981
//
982
        cmpb    #0x0f,DNRM_FLG(%a6) //check for dest denorm
983
        bnes    sub_srcd
984
sub_destd:
985
        bsrl    ckinf_ns
986
        bne     fix_stk
987
        bfextu  ETEMP_EX(%a6){#1:#15},%d0       //get src exp (always pos)
988
        bfexts  FPTEMP_EX(%a6){#1:#15},%d1      //get dest exp (always neg)
989
        subl    %d1,%d0                 //subtract src from dest
990
        cmpl    #0x8000,%d0
991
        blt     fix_stk                 //if less, not wrap case
992
        bra     sub_wrap
993
sub_srcd:
994
        bsrl    ckinf_nd
995
        bne     fix_stk
996
        bfextu  FPTEMP_EX(%a6){#1:#15},%d0      //get dest exp (always pos)
997
        bfexts  ETEMP_EX(%a6){#1:#15},%d1       //get src exp (always neg)
998
        subl    %d1,%d0                 //subtract dest from src
999
        cmpl    #0x8000,%d0
1000
        blt     fix_stk                 //if less, not wrap case
1001
//
1002
// Check the signs of the operands.  If they are alike, the fpu
1003
// can be used to subtract from the norm 1.0 with the sign of the
1004
// denorm and it will correctly generate the result in extended
1005
// precision.  We can then call round with no sticky and the result
1006
// will be correct for the user's rounding mode and precision.  If
1007
// the signs are unlike, we call round with the sticky bit set
1008
// and the result will be correct for the user's rounding mode and
1009
// precision.
1010
//
1011
sub_wrap:
1012
        movew   ETEMP_EX(%a6),%d0
1013
        movew   FPTEMP_EX(%a6),%d1
1014
        eorw    %d1,%d0
1015
        andiw   #0x8000,%d0
1016
        bne     sub_diff
1017
//
1018
// The signs are alike.
1019
//
1020
        cmpb    #0x0f,DNRM_FLG(%a6) //is dest the denorm?
1021
        bnes    sub_u_srcd
1022
        movew   FPTEMP_EX(%a6),%d0
1023
        andiw   #0x8000,%d0
1024
        orw     #0x3fff,%d0     //force the exponent to +/- 1
1025
        movew   %d0,FPTEMP_EX(%a6) //in the denorm
1026
        movel   USER_FPCR(%a6),%d0
1027
        andil   #0x30,%d0
1028
        fmovel  %d0,%fpcr               //set up users rmode and X
1029
        fmovex  FPTEMP(%a6),%fp0
1030
        fsubx   ETEMP(%a6),%fp0
1031
        fmovel  %fpsr,%d1
1032
        orl     %d1,USER_FPSR(%a6) //capture cc's and inex from fadd
1033
        leal    WBTEMP(%a6),%a0 //point a0 to wbtemp in frame
1034
        fmovex  %fp0,WBTEMP(%a6)        //write result to memory
1035
        lsrl    #4,%d0          //put rmode in lower 2 bits
1036
        movel   USER_FPCR(%a6),%d1
1037
        andil   #0xc0,%d1
1038
        lsrl    #6,%d1          //put precision in upper word
1039
        swap    %d1
1040
        orl     %d0,%d1         //set up for round call
1041
        clrl    %d0             //force sticky to zero
1042
        bclrb   #sign_bit,WBTEMP_EX(%a6)
1043
        sne     WBTEMP_SGN(%a6)
1044
        bsrl    round           //round result to users rmode & prec
1045
        bfclr   WBTEMP_SGN(%a6){#0:#8}  //convert back to IEEE ext format
1046
        beq     frcfpnr
1047
        bsetb   #sign_bit,WBTEMP_EX(%a6)
1048
        bra     frcfpnr
1049
sub_u_srcd:
1050
        movew   ETEMP_EX(%a6),%d0
1051
        andiw   #0x8000,%d0
1052
        orw     #0x3fff,%d0     //force the exponent to +/- 1
1053
        movew   %d0,ETEMP_EX(%a6) //in the denorm
1054
        movel   USER_FPCR(%a6),%d0
1055
        andil   #0x30,%d0
1056
        fmovel  %d0,%fpcr               //set up users rmode and X
1057
        fmovex  FPTEMP(%a6),%fp0
1058
        fsubx   ETEMP(%a6),%fp0
1059
        fmovel  %fpsr,%d1
1060
        orl     %d1,USER_FPSR(%a6) //capture cc's and inex from fadd
1061
        leal    WBTEMP(%a6),%a0 //point a0 to wbtemp in frame
1062
        fmovex  %fp0,WBTEMP(%a6)        //write result to memory
1063
        lsrl    #4,%d0          //put rmode in lower 2 bits
1064
        movel   USER_FPCR(%a6),%d1
1065
        andil   #0xc0,%d1
1066
        lsrl    #6,%d1          //put precision in upper word
1067
        swap    %d1
1068
        orl     %d0,%d1         //set up for round call
1069
        clrl    %d0             //force sticky to zero
1070
        bclrb   #sign_bit,WBTEMP_EX(%a6)
1071
        sne     WBTEMP_SGN(%a6)
1072
        bsrl    round           //round result to users rmode & prec
1073
        bfclr   WBTEMP_SGN(%a6){#0:#8}  //convert back to IEEE ext format
1074
        beq     frcfpnr
1075
        bsetb   #sign_bit,WBTEMP_EX(%a6)
1076
        bra     frcfpnr
1077
//
1078
// Signs are unlike:
1079
//
1080
sub_diff:
1081
        cmpb    #0x0f,DNRM_FLG(%a6) //is dest the denorm?
1082
        bnes    sub_s_srcd
1083
sub_s_destd:
1084
        leal    ETEMP(%a6),%a0
1085
        movel   USER_FPCR(%a6),%d0
1086
        andil   #0x30,%d0
1087
        lsrl    #4,%d0          //put rmode in lower 2 bits
1088
        movel   USER_FPCR(%a6),%d1
1089
        andil   #0xc0,%d1
1090
        lsrl    #6,%d1          //put precision in upper word
1091
        swap    %d1
1092
        orl     %d0,%d1         //set up for round call
1093
        movel   #0x20000000,%d0 //set sticky for round
1094
//
1095
// Since the dest is the denorm, the sign is the opposite of the
1096
// norm sign.
1097
//
1098
        eoriw   #0x8000,ETEMP_EX(%a6)   //flip sign on result
1099
        tstw    ETEMP_EX(%a6)
1100
        bgts    sub_s_dwr
1101
        orl     #neg_mask,USER_FPSR(%a6)
1102
sub_s_dwr:
1103
        bclrb   #sign_bit,ETEMP_EX(%a6)
1104
        sne     ETEMP_SGN(%a6)
1105
        bsrl    round           //round result to users rmode & prec
1106
        bfclr   ETEMP_SGN(%a6){#0:#8}   //convert back to IEEE ext format
1107
        beqs    sub_s_dclr
1108
        bsetb   #sign_bit,ETEMP_EX(%a6)
1109
sub_s_dclr:
1110
        leal    WBTEMP(%a6),%a0
1111
        movel   ETEMP(%a6),(%a0)        //write result to wbtemp
1112
        movel   ETEMP_HI(%a6),4(%a0)
1113
        movel   ETEMP_LO(%a6),8(%a0)
1114
        bra     sub_ckovf
1115
sub_s_srcd:
1116
        leal    FPTEMP(%a6),%a0
1117
        movel   USER_FPCR(%a6),%d0
1118
        andil   #0x30,%d0
1119
        lsrl    #4,%d0          //put rmode in lower 2 bits
1120
        movel   USER_FPCR(%a6),%d1
1121
        andil   #0xc0,%d1
1122
        lsrl    #6,%d1          //put precision in upper word
1123
        swap    %d1
1124
        orl     %d0,%d1         //set up for round call
1125
        movel   #0x20000000,%d0 //set sticky for round
1126
        bclrb   #sign_bit,FPTEMP_EX(%a6)
1127
        sne     FPTEMP_SGN(%a6)
1128
        bsrl    round           //round result to users rmode & prec
1129
        bfclr   FPTEMP_SGN(%a6){#0:#8}  //convert back to IEEE ext format
1130
        beqs    sub_s_sclr
1131
        bsetb   #sign_bit,FPTEMP_EX(%a6)
1132
sub_s_sclr:
1133
        leal    WBTEMP(%a6),%a0
1134
        movel   FPTEMP(%a6),(%a0)       //write result to wbtemp
1135
        movel   FPTEMP_HI(%a6),4(%a0)
1136
        movel   FPTEMP_LO(%a6),8(%a0)
1137
        tstw    FPTEMP_EX(%a6)
1138
        bgt     sub_ckovf
1139
        orl     #neg_mask,USER_FPSR(%a6)
1140
sub_ckovf:
1141
        movew   WBTEMP_EX(%a6),%d0
1142
        andiw   #0x7fff,%d0
1143
        cmpiw   #0x7fff,%d0
1144
        bne     frcfpnr
1145
//
1146
// The result has overflowed to $7fff exponent.  Set I, ovfl,
1147
// and aovfl, and clr the mantissa (incorrectly set by the
1148
// round routine.)
1149
//
1150
        orl     #inf_mask+ovfl_inx_mask,USER_FPSR(%a6)
1151
        clrl    4(%a0)
1152
        bra     frcfpnr
1153
//
1154
// Inst is fcmp.
1155
//
1156
wrap_cmp:
1157
        cmpb    #0xff,DNRM_FLG(%a6) //if both ops denorm,
1158
        beq     fix_stk          //restore to fpu
1159
//
1160
// One of the ops is denormalized.  Test for wrap condition
1161
// and complete the instruction.
1162
//
1163
        cmpb    #0x0f,DNRM_FLG(%a6) //check for dest denorm
1164
        bnes    cmp_srcd
1165
cmp_destd:
1166
        bsrl    ckinf_ns
1167
        bne     fix_stk
1168
        bfextu  ETEMP_EX(%a6){#1:#15},%d0       //get src exp (always pos)
1169
        bfexts  FPTEMP_EX(%a6){#1:#15},%d1      //get dest exp (always neg)
1170
        subl    %d1,%d0                 //subtract dest from src
1171
        cmpl    #0x8000,%d0
1172
        blt     fix_stk                 //if less, not wrap case
1173
        tstw    ETEMP_EX(%a6)           //set N to ~sign_of(src)
1174
        bge     cmp_setn
1175
        rts
1176
cmp_srcd:
1177
        bsrl    ckinf_nd
1178
        bne     fix_stk
1179
        bfextu  FPTEMP_EX(%a6){#1:#15},%d0      //get dest exp (always pos)
1180
        bfexts  ETEMP_EX(%a6){#1:#15},%d1       //get src exp (always neg)
1181
        subl    %d1,%d0                 //subtract src from dest
1182
        cmpl    #0x8000,%d0
1183
        blt     fix_stk                 //if less, not wrap case
1184
        tstw    FPTEMP_EX(%a6)          //set N to sign_of(dest)
1185
        blt     cmp_setn
1186
        rts
1187
cmp_setn:
1188
        orl     #neg_mask,USER_FPSR(%a6)
1189
        rts
1190
 
1191
//
1192
// Inst is fmul.
1193
//
1194
wrap_mul:
1195
        cmpb    #0xff,DNRM_FLG(%a6) //if both ops denorm,
1196
        beq     force_unf       //force an underflow (really!)
1197
//
1198
// One of the ops is denormalized.  Test for wrap condition
1199
// and complete the instruction.
1200
//
1201
        cmpb    #0x0f,DNRM_FLG(%a6) //check for dest denorm
1202
        bnes    mul_srcd
1203
mul_destd:
1204
        bsrl    ckinf_ns
1205
        bne     fix_stk
1206
        bfextu  ETEMP_EX(%a6){#1:#15},%d0       //get src exp (always pos)
1207
        bfexts  FPTEMP_EX(%a6){#1:#15},%d1      //get dest exp (always neg)
1208
        addl    %d1,%d0                 //subtract dest from src
1209
        bgt     fix_stk
1210
        bra     force_unf
1211
mul_srcd:
1212
        bsrl    ckinf_nd
1213
        bne     fix_stk
1214
        bfextu  FPTEMP_EX(%a6){#1:#15},%d0      //get dest exp (always pos)
1215
        bfexts  ETEMP_EX(%a6){#1:#15},%d1       //get src exp (always neg)
1216
        addl    %d1,%d0                 //subtract src from dest
1217
        bgt     fix_stk
1218
 
1219
//
1220
// This code handles the case of the instruction resulting in
1221
// an underflow condition.
1222
//
1223
force_unf:
1224
        bclrb   #E1,E_BYTE(%a6)
1225
        orl     #unfinx_mask,USER_FPSR(%a6)
1226
        clrw    NMNEXC(%a6)
1227
        clrb    WBTEMP_SGN(%a6)
1228
        movew   ETEMP_EX(%a6),%d0               //find the sign of the result
1229
        movew   FPTEMP_EX(%a6),%d1
1230
        eorw    %d1,%d0
1231
        andiw   #0x8000,%d0
1232
        beqs    frcunfcont
1233
        st      WBTEMP_SGN(%a6)
1234
frcunfcont:
1235
        lea     WBTEMP(%a6),%a0         //point a0 to memory location
1236
        movew   CMDREG1B(%a6),%d0
1237
        btstl   #6,%d0                  //test for forced precision
1238
        beqs    frcunf_fpcr
1239
        btstl   #2,%d0                  //check for double
1240
        bnes    frcunf_dbl
1241
        movel   #0x1,%d0                        //inst is forced single
1242
        bras    frcunf_rnd
1243
frcunf_dbl:
1244
        movel   #0x2,%d0                        //inst is forced double
1245
        bras    frcunf_rnd
1246
frcunf_fpcr:
1247
        bfextu  FPCR_MODE(%a6){#0:#2},%d0       //inst not forced - use fpcr prec
1248
frcunf_rnd:
1249
        bsrl    unf_sub                 //get correct result based on
1250
//                                      ;round precision/mode.  This
1251
//                                      ;sets FPSR_CC correctly
1252
        bfclr   WBTEMP_SGN(%a6){#0:#8}  //convert back to IEEE ext format
1253
        beqs    frcfpn
1254
        bsetb   #sign_bit,WBTEMP_EX(%a6)
1255
        bra     frcfpn
1256
 
1257
//
1258
// Write the result to the user's fpn.  All results must be HUGE to be
1259
// written; otherwise the results would have overflowed or underflowed.
1260
// If the rounding precision is single or double, the ovf_res routine
1261
// is needed to correctly supply the max value.
1262
//
1263
frcfpnr:
1264
        movew   CMDREG1B(%a6),%d0
1265
        btstl   #6,%d0                  //test for forced precision
1266
        beqs    frcfpn_fpcr
1267
        btstl   #2,%d0                  //check for double
1268
        bnes    frcfpn_dbl
1269
        movel   #0x1,%d0                        //inst is forced single
1270
        bras    frcfpn_rnd
1271
frcfpn_dbl:
1272
        movel   #0x2,%d0                        //inst is forced double
1273
        bras    frcfpn_rnd
1274
frcfpn_fpcr:
1275
        bfextu  FPCR_MODE(%a6){#0:#2},%d0       //inst not forced - use fpcr prec
1276
        tstb    %d0
1277
        beqs    frcfpn                  //if extended, write what you got
1278
frcfpn_rnd:
1279
        bclrb   #sign_bit,WBTEMP_EX(%a6)
1280
        sne     WBTEMP_SGN(%a6)
1281
        bsrl    ovf_res                 //get correct result based on
1282
//                                      ;round precision/mode.  This
1283
//                                      ;sets FPSR_CC correctly
1284
        bfclr   WBTEMP_SGN(%a6){#0:#8}  //convert back to IEEE ext format
1285
        beqs    frcfpn_clr
1286
        bsetb   #sign_bit,WBTEMP_EX(%a6)
1287
frcfpn_clr:
1288
        orl     #ovfinx_mask,USER_FPSR(%a6)
1289
//
1290
// Perform the write.
1291
//
1292
frcfpn:
1293
        bfextu  CMDREG1B(%a6){#6:#3},%d0        //extract fp destination register
1294
        cmpib   #3,%d0
1295
        bles    frc0123                 //check if dest is fp0-fp3
1296
        movel   #7,%d1
1297
        subl    %d0,%d1
1298
        clrl    %d0
1299
        bsetl   %d1,%d0
1300
        fmovemx WBTEMP(%a6),%d0
1301
        rts
1302
frc0123:
1303
        cmpib   #0,%d0
1304
        beqs    frc0_dst
1305
        cmpib   #1,%d0
1306
        beqs    frc1_dst
1307
        cmpib   #2,%d0
1308
        beqs    frc2_dst
1309
frc3_dst:
1310
        movel   WBTEMP_EX(%a6),USER_FP3(%a6)
1311
        movel   WBTEMP_HI(%a6),USER_FP3+4(%a6)
1312
        movel   WBTEMP_LO(%a6),USER_FP3+8(%a6)
1313
        rts
1314
frc2_dst:
1315
        movel   WBTEMP_EX(%a6),USER_FP2(%a6)
1316
        movel   WBTEMP_HI(%a6),USER_FP2+4(%a6)
1317
        movel   WBTEMP_LO(%a6),USER_FP2+8(%a6)
1318
        rts
1319
frc1_dst:
1320
        movel   WBTEMP_EX(%a6),USER_FP1(%a6)
1321
        movel   WBTEMP_HI(%a6),USER_FP1+4(%a6)
1322
        movel   WBTEMP_LO(%a6),USER_FP1+8(%a6)
1323
        rts
1324
frc0_dst:
1325
        movel   WBTEMP_EX(%a6),USER_FP0(%a6)
1326
        movel   WBTEMP_HI(%a6),USER_FP0+4(%a6)
1327
        movel   WBTEMP_LO(%a6),USER_FP0+8(%a6)
1328
        rts
1329
 
1330
//
1331
// Write etemp to fpn.
1332
// A check is made on enabled and signalled snan exceptions,
1333
// and the destination is not overwritten if this condition exists.
1334
// This code is designed to make fmoveins of unsupported data types
1335
// faster.
1336
//
1337
wr_etemp:
1338
        btstb   #snan_bit,FPSR_EXCEPT(%a6)      //if snan is set, and
1339
        beqs    fmoveinc                //enabled, force restore
1340
        btstb   #snan_bit,FPCR_ENABLE(%a6) //and don't overwrite
1341
        beqs    fmoveinc                //the dest
1342
        movel   ETEMP_EX(%a6),FPTEMP_EX(%a6)    //set up fptemp sign for
1343
//                                              ;snan handler
1344
        tstb    ETEMP(%a6)              //check for negative
1345
        blts    snan_neg
1346
        rts
1347
snan_neg:
1348
        orl     #neg_bit,USER_FPSR(%a6) //snan is negative; set N
1349
        rts
1350
fmoveinc:
1351
        clrw    NMNEXC(%a6)
1352
        bclrb   #E1,E_BYTE(%a6)
1353
        moveb   STAG(%a6),%d0           //check if stag is inf
1354
        andib   #0xe0,%d0
1355
        cmpib   #0x40,%d0
1356
        bnes    fminc_cnan
1357
        orl     #inf_mask,USER_FPSR(%a6) //if inf, nothing yet has set I
1358
        tstw    LOCAL_EX(%a0)           //check sign
1359
        bges    fminc_con
1360
        orl     #neg_mask,USER_FPSR(%a6)
1361
        bra     fminc_con
1362
fminc_cnan:
1363
        cmpib   #0x60,%d0                       //check if stag is NaN
1364
        bnes    fminc_czero
1365
        orl     #nan_mask,USER_FPSR(%a6) //if nan, nothing yet has set NaN
1366
        movel   ETEMP_EX(%a6),FPTEMP_EX(%a6)    //set up fptemp sign for
1367
//                                              ;snan handler
1368
        tstw    LOCAL_EX(%a0)           //check sign
1369
        bges    fminc_con
1370
        orl     #neg_mask,USER_FPSR(%a6)
1371
        bra     fminc_con
1372
fminc_czero:
1373
        cmpib   #0x20,%d0                       //check if zero
1374
        bnes    fminc_con
1375
        orl     #z_mask,USER_FPSR(%a6)  //if zero, set Z
1376
        tstw    LOCAL_EX(%a0)           //check sign
1377
        bges    fminc_con
1378
        orl     #neg_mask,USER_FPSR(%a6)
1379
fminc_con:
1380
        bfextu  CMDREG1B(%a6){#6:#3},%d0        //extract fp destination register
1381
        cmpib   #3,%d0
1382
        bles    fp0123                  //check if dest is fp0-fp3
1383
        movel   #7,%d1
1384
        subl    %d0,%d1
1385
        clrl    %d0
1386
        bsetl   %d1,%d0
1387
        fmovemx ETEMP(%a6),%d0
1388
        rts
1389
 
1390
fp0123:
1391
        cmpib   #0,%d0
1392
        beqs    fp0_dst
1393
        cmpib   #1,%d0
1394
        beqs    fp1_dst
1395
        cmpib   #2,%d0
1396
        beqs    fp2_dst
1397
fp3_dst:
1398
        movel   ETEMP_EX(%a6),USER_FP3(%a6)
1399
        movel   ETEMP_HI(%a6),USER_FP3+4(%a6)
1400
        movel   ETEMP_LO(%a6),USER_FP3+8(%a6)
1401
        rts
1402
fp2_dst:
1403
        movel   ETEMP_EX(%a6),USER_FP2(%a6)
1404
        movel   ETEMP_HI(%a6),USER_FP2+4(%a6)
1405
        movel   ETEMP_LO(%a6),USER_FP2+8(%a6)
1406
        rts
1407
fp1_dst:
1408
        movel   ETEMP_EX(%a6),USER_FP1(%a6)
1409
        movel   ETEMP_HI(%a6),USER_FP1+4(%a6)
1410
        movel   ETEMP_LO(%a6),USER_FP1+8(%a6)
1411
        rts
1412
fp0_dst:
1413
        movel   ETEMP_EX(%a6),USER_FP0(%a6)
1414
        movel   ETEMP_HI(%a6),USER_FP0+4(%a6)
1415
        movel   ETEMP_LO(%a6),USER_FP0+8(%a6)
1416
        rts
1417
 
1418
opclass3:
1419
        st      CU_ONLY(%a6)
1420
        movew   CMDREG1B(%a6),%d0       //check if packed moveout
1421
        andiw   #0x0c00,%d0     //isolate last 2 bits of size field
1422
        cmpiw   #0x0c00,%d0     //if size is 011 or 111, it is packed
1423
        beq     pack_out        //else it is norm or denorm
1424
        bra     mv_out
1425
 
1426
 
1427
//
1428
//      MOVE OUT
1429
//
1430
 
1431
mv_tbl:
1432
        .long   li
1433
        .long   sgp
1434
        .long   xp
1435
        .long   mvout_end       //should never be taken
1436
        .long   wi
1437
        .long   dp
1438
        .long   bi
1439
        .long   mvout_end       //should never be taken
1440
mv_out:
1441
        bfextu  CMDREG1B(%a6){#3:#3},%d1        //put source specifier in d1
1442
        leal    mv_tbl,%a0
1443
        movel   %a0@(%d1:l:4),%a0
1444
        jmp     (%a0)
1445
 
1446
//
1447
// This exit is for move-out to memory.  The aunfl bit is
1448
// set if the result is inex and unfl is signalled.
1449
//
1450
mvout_end:
1451
        btstb   #inex2_bit,FPSR_EXCEPT(%a6)
1452
        beqs    no_aufl
1453
        btstb   #unfl_bit,FPSR_EXCEPT(%a6)
1454
        beqs    no_aufl
1455
        bsetb   #aunfl_bit,FPSR_AEXCEPT(%a6)
1456
no_aufl:
1457
        clrw    NMNEXC(%a6)
1458
        bclrb   #E1,E_BYTE(%a6)
1459
        fmovel  #0,%FPSR                        //clear any cc bits from res_func
1460
//
1461
// Return ETEMP to extended format from internal extended format so
1462
// that gen_except will have a correctly signed value for ovfl/unfl
1463
// handlers.
1464
//
1465
        bfclr   ETEMP_SGN(%a6){#0:#8}
1466
        beqs    mvout_con
1467
        bsetb   #sign_bit,ETEMP_EX(%a6)
1468
mvout_con:
1469
        rts
1470
//
1471
// This exit is for move-out to int register.  The aunfl bit is
1472
// not set in any case for this move.
1473
//
1474
mvouti_end:
1475
        clrw    NMNEXC(%a6)
1476
        bclrb   #E1,E_BYTE(%a6)
1477
        fmovel  #0,%FPSR                        //clear any cc bits from res_func
1478
//
1479
// Return ETEMP to extended format from internal extended format so
1480
// that gen_except will have a correctly signed value for ovfl/unfl
1481
// handlers.
1482
//
1483
        bfclr   ETEMP_SGN(%a6){#0:#8}
1484
        beqs    mvouti_con
1485
        bsetb   #sign_bit,ETEMP_EX(%a6)
1486
mvouti_con:
1487
        rts
1488
//
1489
// li is used to handle a long integer source specifier
1490
//
1491
 
1492
li:
1493
        moveql  #4,%d0          //set byte count
1494
 
1495
        btstb   #7,STAG(%a6)    //check for extended denorm
1496
        bne     int_dnrm        //if so, branch
1497
 
1498
        fmovemx ETEMP(%a6),%fp0-%fp0
1499
        fcmpd   #0x41dfffffffc00000,%fp0
1500
// 41dfffffffc00000 in dbl prec = 401d0000fffffffe00000000 in ext prec
1501
        fbge    lo_plrg
1502
        fcmpd   #0xc1e0000000000000,%fp0
1503
// c1e0000000000000 in dbl prec = c01e00008000000000000000 in ext prec
1504
        fble    lo_nlrg
1505
//
1506
// at this point, the answer is between the largest pos and neg values
1507
//
1508
        movel   USER_FPCR(%a6),%d1      //use user's rounding mode
1509
        andil   #0x30,%d1
1510
        fmovel  %d1,%fpcr
1511
        fmovel  %fp0,L_SCR1(%a6)        //let the 040 perform conversion
1512
        fmovel %fpsr,%d1
1513
        orl     %d1,USER_FPSR(%a6)      //capture inex2/ainex if set
1514
        bra     int_wrt
1515
 
1516
 
1517
lo_plrg:
1518
        movel   #0x7fffffff,L_SCR1(%a6) //answer is largest positive int
1519
        fbeq    int_wrt                 //exact answer
1520
        fcmpd   #0x41dfffffffe00000,%fp0
1521
// 41dfffffffe00000 in dbl prec = 401d0000ffffffff00000000 in ext prec
1522
        fbge    int_operr               //set operr
1523
        bra     int_inx                 //set inexact
1524
 
1525
lo_nlrg:
1526
        movel   #0x80000000,L_SCR1(%a6)
1527
        fbeq    int_wrt                 //exact answer
1528
        fcmpd   #0xc1e0000000100000,%fp0
1529
// c1e0000000100000 in dbl prec = c01e00008000000080000000 in ext prec
1530
        fblt    int_operr               //set operr
1531
        bra     int_inx                 //set inexact
1532
 
1533
//
1534
// wi is used to handle a word integer source specifier
1535
//
1536
 
1537
wi:
1538
        moveql  #2,%d0          //set byte count
1539
 
1540
        btstb   #7,STAG(%a6)    //check for extended denorm
1541
        bne     int_dnrm        //branch if so
1542
 
1543
        fmovemx ETEMP(%a6),%fp0-%fp0
1544
        fcmps   #0x46fffe00,%fp0
1545
// 46fffe00 in sgl prec = 400d0000fffe000000000000 in ext prec
1546
        fbge    wo_plrg
1547
        fcmps   #0xc7000000,%fp0
1548
// c7000000 in sgl prec = c00e00008000000000000000 in ext prec
1549
        fble    wo_nlrg
1550
 
1551
//
1552
// at this point, the answer is between the largest pos and neg values
1553
//
1554
        movel   USER_FPCR(%a6),%d1      //use user's rounding mode
1555
        andil   #0x30,%d1
1556
        fmovel  %d1,%fpcr
1557
        fmovew  %fp0,L_SCR1(%a6)        //let the 040 perform conversion
1558
        fmovel %fpsr,%d1
1559
        orl     %d1,USER_FPSR(%a6)      //capture inex2/ainex if set
1560
        bra     int_wrt
1561
 
1562
wo_plrg:
1563
        movew   #0x7fff,L_SCR1(%a6)     //answer is largest positive int
1564
        fbeq    int_wrt                 //exact answer
1565
        fcmps   #0x46ffff00,%fp0
1566
// 46ffff00 in sgl prec = 400d0000ffff000000000000 in ext prec
1567
        fbge    int_operr               //set operr
1568
        bra     int_inx                 //set inexact
1569
 
1570
wo_nlrg:
1571
        movew   #0x8000,L_SCR1(%a6)
1572
        fbeq    int_wrt                 //exact answer
1573
        fcmps   #0xc7000080,%fp0
1574
// c7000080 in sgl prec = c00e00008000800000000000 in ext prec
1575
        fblt    int_operr               //set operr
1576
        bra     int_inx                 //set inexact
1577
 
1578
//
1579
// bi is used to handle a byte integer source specifier
1580
//
1581
 
1582
bi:
1583
        moveql  #1,%d0          //set byte count
1584
 
1585
        btstb   #7,STAG(%a6)    //check for extended denorm
1586
        bne     int_dnrm        //branch if so
1587
 
1588
        fmovemx ETEMP(%a6),%fp0-%fp0
1589
        fcmps   #0x42fe0000,%fp0
1590
// 42fe0000 in sgl prec = 40050000fe00000000000000 in ext prec
1591
        fbge    by_plrg
1592
        fcmps   #0xc3000000,%fp0
1593
// c3000000 in sgl prec = c00600008000000000000000 in ext prec
1594
        fble    by_nlrg
1595
 
1596
//
1597
// at this point, the answer is between the largest pos and neg values
1598
//
1599
        movel   USER_FPCR(%a6),%d1      //use user's rounding mode
1600
        andil   #0x30,%d1
1601
        fmovel  %d1,%fpcr
1602
        fmoveb  %fp0,L_SCR1(%a6)        //let the 040 perform conversion
1603
        fmovel %fpsr,%d1
1604
        orl     %d1,USER_FPSR(%a6)      //capture inex2/ainex if set
1605
        bra     int_wrt
1606
 
1607
by_plrg:
1608
        moveb   #0x7f,L_SCR1(%a6)               //answer is largest positive int
1609
        fbeq    int_wrt                 //exact answer
1610
        fcmps   #0x42ff0000,%fp0
1611
// 42ff0000 in sgl prec = 40050000ff00000000000000 in ext prec
1612
        fbge    int_operr               //set operr
1613
        bra     int_inx                 //set inexact
1614
 
1615
by_nlrg:
1616
        moveb   #0x80,L_SCR1(%a6)
1617
        fbeq    int_wrt                 //exact answer
1618
        fcmps   #0xc3008000,%fp0
1619
// c3008000 in sgl prec = c00600008080000000000000 in ext prec
1620
        fblt    int_operr               //set operr
1621
        bra     int_inx                 //set inexact
1622
 
1623
//
1624
// Common integer routines
1625
//
1626
// int_drnrm---account for possible nonzero result for round up with positive
1627
// operand and round down for negative answer.  In the first case (result = 1)
1628
// byte-width (store in d0) of result must be honored.  In the second case,
1629
// -1 in L_SCR1(a6) will cover all contingencies (FMOVE.B/W/L out).
1630
 
1631
int_dnrm:
1632
        movel   #0,L_SCR1(%a6)  // initialize result to 0
1633
        bfextu  FPCR_MODE(%a6){#2:#2},%d1       // d1 is the rounding mode
1634
        cmpb    #2,%d1
1635
        bmis    int_inx         // if RN or RZ, done
1636
        bnes    int_rp          // if RP, continue below
1637
        tstw    ETEMP(%a6)      // RM: store -1 in L_SCR1 if src is negative
1638
        bpls    int_inx         // otherwise result is 0
1639
        movel   #-1,L_SCR1(%a6)
1640
        bras    int_inx
1641
int_rp:
1642
        tstw    ETEMP(%a6)      // RP: store +1 of proper width in L_SCR1 if
1643
//                              ; source is greater than 0
1644
        bmis    int_inx         // otherwise, result is 0
1645
        lea     L_SCR1(%a6),%a1 // a1 is address of L_SCR1
1646
        addal   %d0,%a1         // offset by destination width -1
1647
        subal   #1,%a1
1648
        bsetb   #0,(%a1)                // set low bit at a1 address
1649
int_inx:
1650
        oril    #inx2a_mask,USER_FPSR(%a6)
1651
        bras    int_wrt
1652
int_operr:
1653
        fmovemx %fp0-%fp0,FPTEMP(%a6)   //FPTEMP must contain the extended
1654
//                              ;precision source that needs to be
1655
//                              ;converted to integer this is required
1656
//                              ;if the operr exception is enabled.
1657
//                              ;set operr/aiop (no inex2 on int ovfl)
1658
 
1659
        oril    #opaop_mask,USER_FPSR(%a6)
1660
//                              ;fall through to perform int_wrt
1661
int_wrt:
1662
        movel   EXC_EA(%a6),%a1 //load destination address
1663
        tstl    %a1             //check to see if it is a dest register
1664
        beqs    wrt_dn          //write data register
1665
        lea     L_SCR1(%a6),%a0 //point to supervisor source address
1666
        bsrl    mem_write
1667
        bra     mvouti_end
1668
 
1669
wrt_dn:
1670
        movel   %d0,-(%sp)      //d0 currently contains the size to write
1671
        bsrl    get_fline       //get_fline returns Dn in d0
1672
        andiw   #0x7,%d0                //isolate register
1673
        movel   (%sp)+,%d1      //get size
1674
        cmpil   #4,%d1          //most frequent case
1675
        beqs    sz_long
1676
        cmpil   #2,%d1
1677
        bnes    sz_con
1678
        orl     #8,%d0          //add 'word' size to register#
1679
        bras    sz_con
1680
sz_long:
1681
        orl     #0x10,%d0               //add 'long' size to register#
1682
sz_con:
1683
        movel   %d0,%d1         //reg_dest expects size:reg in d1
1684
        bsrl    reg_dest        //load proper data register
1685
        bra     mvouti_end
1686
xp:
1687
        lea     ETEMP(%a6),%a0
1688
        bclrb   #sign_bit,LOCAL_EX(%a0)
1689
        sne     LOCAL_SGN(%a0)
1690
        btstb   #7,STAG(%a6)    //check for extended denorm
1691
        bne     xdnrm
1692
        clrl    %d0
1693
        bras    do_fp           //do normal case
1694
sgp:
1695
        lea     ETEMP(%a6),%a0
1696
        bclrb   #sign_bit,LOCAL_EX(%a0)
1697
        sne     LOCAL_SGN(%a0)
1698
        btstb   #7,STAG(%a6)    //check for extended denorm
1699
        bne     sp_catas        //branch if so
1700
        movew   LOCAL_EX(%a0),%d0
1701
        lea     sp_bnds,%a1
1702
        cmpw    (%a1),%d0
1703
        blt     sp_under
1704
        cmpw    2(%a1),%d0
1705
        bgt     sp_over
1706
        movel   #1,%d0          //set destination format to single
1707
        bras    do_fp           //do normal case
1708
dp:
1709
        lea     ETEMP(%a6),%a0
1710
        bclrb   #sign_bit,LOCAL_EX(%a0)
1711
        sne     LOCAL_SGN(%a0)
1712
 
1713
        btstb   #7,STAG(%a6)    //check for extended denorm
1714
        bne     dp_catas        //branch if so
1715
 
1716
        movew   LOCAL_EX(%a0),%d0
1717
        lea     dp_bnds,%a1
1718
 
1719
        cmpw    (%a1),%d0
1720
        blt     dp_under
1721
        cmpw    2(%a1),%d0
1722
        bgt     dp_over
1723
 
1724
        movel   #2,%d0          //set destination format to double
1725
//                              ;fall through to do_fp
1726
//
1727
do_fp:
1728
        bfextu  FPCR_MODE(%a6){#2:#2},%d1       //rnd mode in d1
1729
        swap    %d0                     //rnd prec in upper word
1730
        addl    %d0,%d1                 //d1 has PREC/MODE info
1731
 
1732
        clrl    %d0                     //clear g,r,s
1733
 
1734
        bsrl    round                   //round
1735
 
1736
        movel   %a0,%a1
1737
        movel   EXC_EA(%a6),%a0
1738
 
1739
        bfextu  CMDREG1B(%a6){#3:#3},%d1        //extract destination format
1740
//                                      ;at this point only the dest
1741
//                                      ;formats sgl, dbl, ext are
1742
//                                      ;possible
1743
        cmpb    #2,%d1
1744
        bgts    ddbl                    //double=5, extended=2, single=1
1745
        bnes    dsgl
1746
//                                      ;fall through to dext
1747
dext:
1748
        bsrl    dest_ext
1749
        bra     mvout_end
1750
dsgl:
1751
        bsrl    dest_sgl
1752
        bra     mvout_end
1753
ddbl:
1754
        bsrl    dest_dbl
1755
        bra     mvout_end
1756
 
1757
//
1758
// Handle possible denorm or catastrophic underflow cases here
1759
//
1760
xdnrm:
1761
        bsr     set_xop         //initialize WBTEMP
1762
        bsetb   #wbtemp15_bit,WB_BYTE(%a6) //set wbtemp15
1763
 
1764
        movel   %a0,%a1
1765
        movel   EXC_EA(%a6),%a0 //a0 has the destination pointer
1766
        bsrl    dest_ext        //store to memory
1767
        bsetb   #unfl_bit,FPSR_EXCEPT(%a6)
1768
        bra     mvout_end
1769
 
1770
sp_under:
1771
        bsetb   #etemp15_bit,STAG(%a6)
1772
 
1773
        cmpw    4(%a1),%d0
1774
        blts    sp_catas        //catastrophic underflow case
1775
 
1776
        movel   #1,%d0          //load in round precision
1777
        movel   #sgl_thresh,%d1 //load in single denorm threshold
1778
        bsrl    dpspdnrm        //expects d1 to have the proper
1779
//                              ;denorm threshold
1780
        bsrl    dest_sgl        //stores value to destination
1781
        bsetb   #unfl_bit,FPSR_EXCEPT(%a6)
1782
        bra     mvout_end       //exit
1783
 
1784
dp_under:
1785
        bsetb   #etemp15_bit,STAG(%a6)
1786
 
1787
        cmpw    4(%a1),%d0
1788
        blts    dp_catas        //catastrophic underflow case
1789
 
1790
        movel   #dbl_thresh,%d1 //load in double precision threshold
1791
        movel   #2,%d0
1792
        bsrl    dpspdnrm        //expects d1 to have proper
1793
//                              ;denorm threshold
1794
//                              ;expects d0 to have round precision
1795
        bsrl    dest_dbl        //store value to destination
1796
        bsetb   #unfl_bit,FPSR_EXCEPT(%a6)
1797
        bra     mvout_end       //exit
1798
 
1799
//
1800
// Handle catastrophic underflow cases here
1801
//
1802
sp_catas:
1803
// Temp fix for z bit set in unf_sub
1804
        movel   USER_FPSR(%a6),-(%a7)
1805
 
1806
        movel   #1,%d0          //set round precision to sgl
1807
 
1808
        bsrl    unf_sub         //a0 points to result
1809
 
1810
        movel   (%a7)+,USER_FPSR(%a6)
1811
 
1812
        movel   #1,%d0
1813
        subw    %d0,LOCAL_EX(%a0) //account for difference between
1814
//                              ;denorm/norm bias
1815
 
1816
        movel   %a0,%a1         //a1 has the operand input
1817
        movel   EXC_EA(%a6),%a0 //a0 has the destination pointer
1818
 
1819
        bsrl    dest_sgl        //store the result
1820
        oril    #unfinx_mask,USER_FPSR(%a6)
1821
        bra     mvout_end
1822
 
1823
dp_catas:
1824
// Temp fix for z bit set in unf_sub
1825
        movel   USER_FPSR(%a6),-(%a7)
1826
 
1827
        movel   #2,%d0          //set round precision to dbl
1828
        bsrl    unf_sub         //a0 points to result
1829
 
1830
        movel   (%a7)+,USER_FPSR(%a6)
1831
 
1832
        movel   #1,%d0
1833
        subw    %d0,LOCAL_EX(%a0) //account for difference between
1834
//                              ;denorm/norm bias
1835
 
1836
        movel   %a0,%a1         //a1 has the operand input
1837
        movel   EXC_EA(%a6),%a0 //a0 has the destination pointer
1838
 
1839
        bsrl    dest_dbl        //store the result
1840
        oril    #unfinx_mask,USER_FPSR(%a6)
1841
        bra     mvout_end
1842
 
1843
//
1844
// Handle catastrophic overflow cases here
1845
//
1846
sp_over:
1847
// Temp fix for z bit set in unf_sub
1848
        movel   USER_FPSR(%a6),-(%a7)
1849
 
1850
        movel   #1,%d0
1851
        leal    FP_SCR1(%a6),%a0        //use FP_SCR1 for creating result
1852
        movel   ETEMP_EX(%a6),(%a0)
1853
        movel   ETEMP_HI(%a6),4(%a0)
1854
        movel   ETEMP_LO(%a6),8(%a0)
1855
        bsrl    ovf_res
1856
 
1857
        movel   (%a7)+,USER_FPSR(%a6)
1858
 
1859
        movel   %a0,%a1
1860
        movel   EXC_EA(%a6),%a0
1861
        bsrl    dest_sgl
1862
        orl     #ovfinx_mask,USER_FPSR(%a6)
1863
        bra     mvout_end
1864
 
1865
dp_over:
1866
// Temp fix for z bit set in ovf_res
1867
        movel   USER_FPSR(%a6),-(%a7)
1868
 
1869
        movel   #2,%d0
1870
        leal    FP_SCR1(%a6),%a0        //use FP_SCR1 for creating result
1871
        movel   ETEMP_EX(%a6),(%a0)
1872
        movel   ETEMP_HI(%a6),4(%a0)
1873
        movel   ETEMP_LO(%a6),8(%a0)
1874
        bsrl    ovf_res
1875
 
1876
        movel   (%a7)+,USER_FPSR(%a6)
1877
 
1878
        movel   %a0,%a1
1879
        movel   EXC_EA(%a6),%a0
1880
        bsrl    dest_dbl
1881
        orl     #ovfinx_mask,USER_FPSR(%a6)
1882
        bra     mvout_end
1883
 
1884
//
1885
//      DPSPDNRM
1886
//
1887
// This subroutine takes an extended normalized number and denormalizes
1888
// it to the given round precision. This subroutine also decrements
1889
// the input operand's exponent by 1 to account for the fact that
1890
// dest_sgl or dest_dbl expects a normalized number's bias.
1891
//
1892
// Input: a0  points to a normalized number in internal extended format
1893
//       d0  is the round precision (=1 for sgl; =2 for dbl)
1894
//       d1  is the the single precision or double precision
1895
//           denorm threshold
1896
//
1897
// Output: (In the format for dest_sgl or dest_dbl)
1898
//       a0   points to the destination
1899
//       a1   points to the operand
1900
//
1901
// Exceptions: Reports inexact 2 exception by setting USER_FPSR bits
1902
//
1903
dpspdnrm:
1904
        movel   %d0,-(%a7)      //save round precision
1905
        clrl    %d0             //clear initial g,r,s
1906
        bsrl    dnrm_lp         //careful with d0, it's needed by round
1907
 
1908
        bfextu  FPCR_MODE(%a6){#2:#2},%d1 //get rounding mode
1909
        swap    %d1
1910
        movew   2(%a7),%d1      //set rounding precision
1911
        swap    %d1             //at this point d1 has PREC/MODE info
1912
        bsrl    round           //round result, sets the inex bit in
1913
//                              ;USER_FPSR if needed
1914
 
1915
        movew   #1,%d0
1916
        subw    %d0,LOCAL_EX(%a0) //account for difference in denorm
1917
//                              ;vs norm bias
1918
 
1919
        movel   %a0,%a1         //a1 has the operand input
1920
        movel   EXC_EA(%a6),%a0 //a0 has the destination pointer
1921
        addw    #4,%a7          //pop stack
1922
        rts
1923
//
1924
// SET_XOP initialized WBTEMP with the value pointed to by a0
1925
// input: a0 points to input operand in the internal extended format
1926
//
1927
set_xop:
1928
        movel   LOCAL_EX(%a0),WBTEMP_EX(%a6)
1929
        movel   LOCAL_HI(%a0),WBTEMP_HI(%a6)
1930
        movel   LOCAL_LO(%a0),WBTEMP_LO(%a6)
1931
        bfclr   WBTEMP_SGN(%a6){#0:#8}
1932
        beqs    sxop
1933
        bsetb   #sign_bit,WBTEMP_EX(%a6)
1934
sxop:
1935
        bfclr   STAG(%a6){#5:#4}        //clear wbtm66,wbtm1,wbtm0,sbit
1936
        rts
1937
//
1938
//      P_MOVE
1939
//
1940
p_movet:
1941
        .long   p_move
1942
        .long   p_movez
1943
        .long   p_movei
1944
        .long   p_moven
1945
        .long   p_move
1946
p_regd:
1947
        .long   p_dyd0
1948
        .long   p_dyd1
1949
        .long   p_dyd2
1950
        .long   p_dyd3
1951
        .long   p_dyd4
1952
        .long   p_dyd5
1953
        .long   p_dyd6
1954
        .long   p_dyd7
1955
 
1956
pack_out:
1957
        leal    p_movet,%a0     //load jmp table address
1958
        movew   STAG(%a6),%d0   //get source tag
1959
        bfextu  %d0{#16:#3},%d0 //isolate source bits
1960
        movel   (%a0,%d0.w*4),%a0       //load a0 with routine label for tag
1961
        jmp     (%a0)           //go to the routine
1962
 
1963
p_write:
1964
        movel   #0x0c,%d0       //get byte count
1965
        movel   EXC_EA(%a6),%a1 //get the destination address
1966
        bsr     mem_write       //write the user's destination
1967
        moveb   #0,CU_SAVEPC(%a6) //set the cu save pc to all 0's
1968
 
1969
//
1970
// Also note that the dtag must be set to norm here - this is because
1971
// the 040 uses the dtag to execute the correct microcode.
1972
//
1973
        bfclr    DTAG(%a6){#0:#3}  //set dtag to norm
1974
 
1975
        rts
1976
 
1977
// Notes on handling of special case (zero, inf, and nan) inputs:
1978
//      1. Operr is not signalled if the k-factor is greater than 18.
1979
//      2. Per the manual, status bits are not set.
1980
//
1981
 
1982
p_move:
1983
        movew   CMDREG1B(%a6),%d0
1984
        btstl   #kfact_bit,%d0  //test for dynamic k-factor
1985
        beqs    statick         //if clear, k-factor is static
1986
dynamick:
1987
        bfextu  %d0{#25:#3},%d0 //isolate register for dynamic k-factor
1988
        lea     p_regd,%a0
1989
        movel   %a0@(%d0:l:4),%a0
1990
        jmp     (%a0)
1991
statick:
1992
        andiw   #0x007f,%d0     //get k-factor
1993
        bfexts  %d0{#25:#7},%d0 //sign extend d0 for bindec
1994
        leal    ETEMP(%a6),%a0  //a0 will point to the packed decimal
1995
        bsrl    bindec          //perform the convert; data at a6
1996
        leal    FP_SCR1(%a6),%a0        //load a0 with result address
1997
        bral    p_write
1998
p_movez:
1999
        leal    ETEMP(%a6),%a0  //a0 will point to the packed decimal
2000
        clrw    2(%a0)          //clear lower word of exp
2001
        clrl    4(%a0)          //load second lword of ZERO
2002
        clrl    8(%a0)          //load third lword of ZERO
2003
        bra     p_write         //go write results
2004
p_movei:
2005
        fmovel  #0,%FPSR                //clear aiop
2006
        leal    ETEMP(%a6),%a0  //a0 will point to the packed decimal
2007
        clrw    2(%a0)          //clear lower word of exp
2008
        bra     p_write         //go write the result
2009
p_moven:
2010
        leal    ETEMP(%a6),%a0  //a0 will point to the packed decimal
2011
        clrw    2(%a0)          //clear lower word of exp
2012
        bra     p_write         //go write the result
2013
 
2014
//
2015
// Routines to read the dynamic k-factor from Dn.
2016
//
2017
p_dyd0:
2018
        movel   USER_D0(%a6),%d0
2019
        bras    statick
2020
p_dyd1:
2021
        movel   USER_D1(%a6),%d0
2022
        bras    statick
2023
p_dyd2:
2024
        movel   %d2,%d0
2025
        bras    statick
2026
p_dyd3:
2027
        movel   %d3,%d0
2028
        bras    statick
2029
p_dyd4:
2030
        movel   %d4,%d0
2031
        bras    statick
2032
p_dyd5:
2033
        movel   %d5,%d0
2034
        bras    statick
2035
p_dyd6:
2036
        movel   %d6,%d0
2037
        bra     statick
2038
p_dyd7:
2039
        movel   %d7,%d0
2040
        bra     statick
2041
 
2042
        |end

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.