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

Subversion Repositories openrisc

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

Details | Compare with Previous | View Log

Line No. Rev Author Line
1 30 unneback
//
2
//      $Id: bindec.S,v 1.2 2001-09-27 12:01:22 chris Exp $
3
//
4
//      bindec.sa 3.4 1/3/91
5
//
6
//      bindec
7
//
8
//      Description:
9
//              Converts an input in extended precision format
10
//              to bcd format.
11
//
12
//      Input:
13
//              a0 points to the input extended precision value
14
//              value in memory; d0 contains the k-factor sign-extended
15
//              to 32-bits.  The input may be either normalized,
16
//              unnormalized, or denormalized.
17
//
18
//      Output: result in the FP_SCR1 space on the stack.
19
//
20
//      Saves and Modifies: D2-D7,A2,FP2
21
//
22
//      Algorithm:
23
//
24
//      A1.     Set RM and size ext;  Set SIGMA = sign of input.
25
//              The k-factor is saved for use in d7. Clear the
26
//              BINDEC_FLG for separating normalized/denormalized
27
//              input.  If input is unnormalized or denormalized,
28
//              normalize it.
29
//
30
//      A2.     Set X = abs(input).
31
//
32
//      A3.     Compute ILOG.
33
//              ILOG is the log base 10 of the input value.  It is
34
//              approximated by adding e + 0.f when the original
35
//              value is viewed as 2^^e * 1.f in extended precision.
36
//              This value is stored in d6.
37
//
38
//      A4.     Clr INEX bit.
39
//              The operation in A3 above may have set INEX2.
40
//
41
//      A5.     Set ICTR = 0;
42
//              ICTR is a flag used in A13.  It must be set before the
43
//              loop entry A6.
44
//
45
//      A6.     Calculate LEN.
46
//              LEN is the number of digits to be displayed.  The
47
//              k-factor can dictate either the total number of digits,
48
//              if it is a positive number, or the number of digits
49
//              after the decimal point which are to be included as
50
//              significant.  See the 68882 manual for examples.
51
//              If LEN is computed to be greater than 17, set OPERR in
52
//              USER_FPSR.  LEN is stored in d4.
53
//
54
//      A7.     Calculate SCALE.
55
//              SCALE is equal to 10^ISCALE, where ISCALE is the number
56
//              of decimal places needed to insure LEN integer digits
57
//              in the output before conversion to bcd. LAMBDA is the
58
//              sign of ISCALE, used in A9. Fp1 contains
59
//              10^^(abs(ISCALE)) using a rounding mode which is a
60
//              function of the original rounding mode and the signs
61
//              of ISCALE and X.  A table is given in the code.
62
//
63
//      A8.     Clr INEX; Force RZ.
64
//              The operation in A3 above may have set INEX2.
65
//              RZ mode is forced for the scaling operation to insure
66
//              only one rounding error.  The grs bits are collected in
67
//              the INEX flag for use in A10.
68
//
69
//      A9.     Scale X -> Y.
70
//              The mantissa is scaled to the desired number of
71
//              significant digits.  The excess digits are collected
72
//              in INEX2.
73
//
74
//      A10.    Or in INEX.
75
//              If INEX is set, round error occurred.  This is
76
//              compensated for by 'or-ing' in the INEX2 flag to
77
//              the lsb of Y.
78
//
79
//      A11.    Restore original FPCR; set size ext.
80
//              Perform FINT operation in the user's rounding mode.
81
//              Keep the size to extended.
82
//
83
//      A12.    Calculate YINT = FINT(Y) according to user's rounding
84
//              mode.  The FPSP routine sintd0 is used.  The output
85
//              is in fp0.
86
//
87
//      A13.    Check for LEN digits.
88
//              If the int operation results in more than LEN digits,
89
//              or less than LEN -1 digits, adjust ILOG and repeat from
90
//              A6.  This test occurs only on the first pass.  If the
91
//              result is exactly 10^LEN, decrement ILOG and divide
92
//              the mantissa by 10.
93
//
94
//      A14.    Convert the mantissa to bcd.
95
//              The binstr routine is used to convert the LEN digit
96
//              mantissa to bcd in memory.  The input to binstr is
97
//              to be a fraction; i.e. (mantissa)/10^LEN and adjusted
98
//              such that the decimal point is to the left of bit 63.
99
//              The bcd digits are stored in the correct position in
100
//              the final string area in memory.
101
//
102
//      A15.    Convert the exponent to bcd.
103
//              As in A14 above, the exp is converted to bcd and the
104
//              digits are stored in the final string.
105
//              Test the length of the final exponent string.  If the
106
//              length is 4, set operr.
107
//
108
//      A16.    Write sign bits to final string.
109
//
110
//      Implementation Notes:
111
//
112
//      The registers are used as follows:
113
//
114
//              d0: scratch; LEN input to binstr
115
//              d1: scratch
116
//              d2: upper 32-bits of mantissa for binstr
117
//              d3: scratch;lower 32-bits of mantissa for binstr
118
//              d4: LEN
119
//                      d5: LAMBDA/ICTR
120
//              d6: ILOG
121
//              d7: k-factor
122
//              a0: ptr for original operand/final result
123
//              a1: scratch pointer
124
//              a2: pointer to FP_X; abs(original value) in ext
125
//              fp0: scratch
126
//              fp1: scratch
127
//              fp2: scratch
128
//              F_SCR1:
129
//              F_SCR2:
130
//              L_SCR1:
131
//              L_SCR2:
132
 
133
//              Copyright (C) Motorola, Inc. 1990
134
//                      All Rights Reserved
135
//
136
//      THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA
137
//      The copyright notice above does not evidence any
138
//      actual or intended publication of such source code.
139
 
140
//BINDEC    idnt    2,1 | Motorola 040 Floating Point Software Package
141
 
142
#include "fpsp.defs"
143
 
144
        |section        8
145
 
146
// Constants in extended precision
147
LOG2:   .long   0x3FFD0000,0x9A209A84,0xFBCFF798,0x00000000
148
LOG2UP1:        .long   0x3FFD0000,0x9A209A84,0xFBCFF799,0x00000000
149
 
150
// Constants in single precision
151
FONE:   .long   0x3F800000,0x00000000,0x00000000,0x00000000
152
FTWO:   .long   0x40000000,0x00000000,0x00000000,0x00000000
153
FTEN:   .long   0x41200000,0x00000000,0x00000000,0x00000000
154
F4933:  .long   0x459A2800,0x00000000,0x00000000,0x00000000
155
 
156
RBDTBL:         .byte   0,0,0,0
157
        .byte   3,3,2,2
158
        .byte   3,2,2,3
159
        .byte   2,3,3,2
160
 
161
        |xref   binstr
162
        |xref   sintdo
163
        |xref   ptenrn,ptenrm,ptenrp
164
 
165
        .global bindec
166
        .global sc_mul
167
bindec:
168
        moveml  %d2-%d7/%a2,-(%a7)
169
        fmovemx %fp0-%fp2,-(%a7)
170
 
171
// A1. Set RM and size ext. Set SIGMA = sign input;
172
//     The k-factor is saved for use in d7.  Clear BINDEC_FLG for
173
//     separating  normalized/denormalized input.  If the input
174
//     is a denormalized number, set the BINDEC_FLG memory word
175
//     to signal denorm.  If the input is unnormalized, normalize
176
//     the input and test for denormalized result.
177
//
178
        fmovel  #rm_mode,%FPCR  //set RM and ext
179
        movel   (%a0),L_SCR2(%a6)       //save exponent for sign check
180
        movel   %d0,%d7         //move k-factor to d7
181
        clrb    BINDEC_FLG(%a6) //clr norm/denorm flag
182
        movew   STAG(%a6),%d0   //get stag
183
        andiw   #0xe000,%d0     //isolate stag bits
184
        beq     A2_str          //if zero, input is norm
185
//
186
// Normalize the denorm
187
//
188
un_de_norm:
189
        movew   (%a0),%d0
190
        andiw   #0x7fff,%d0     //strip sign of normalized exp
191
        movel   4(%a0),%d1
192
        movel   8(%a0),%d2
193
norm_loop:
194
        subw    #1,%d0
195
        lsll    #1,%d2
196
        roxll   #1,%d1
197
        tstl    %d1
198
        bges    norm_loop
199
//
200
// Test if the normalized input is denormalized
201
//
202
        tstw    %d0
203
        bgts    pos_exp         //if greater than zero, it is a norm
204
        st      BINDEC_FLG(%a6) //set flag for denorm
205
pos_exp:
206
        andiw   #0x7fff,%d0     //strip sign of normalized exp
207
        movew   %d0,(%a0)
208
        movel   %d1,4(%a0)
209
        movel   %d2,8(%a0)
210
 
211
// A2. Set X = abs(input).
212
//
213
A2_str:
214
        movel   (%a0),FP_SCR2(%a6) // move input to work space
215
        movel   4(%a0),FP_SCR2+4(%a6) // move input to work space
216
        movel   8(%a0),FP_SCR2+8(%a6) // move input to work space
217
        andil   #0x7fffffff,FP_SCR2(%a6) //create abs(X)
218
 
219
// A3. Compute ILOG.
220
//     ILOG is the log base 10 of the input value.  It is approx-
221
//     imated by adding e + 0.f when the original value is viewed
222
//     as 2^^e * 1.f in extended precision.  This value is stored
223
//     in d6.
224
//
225
// Register usage:
226
//      Input/Output
227
//      d0: k-factor/exponent
228
//      d2: x/x
229
//      d3: x/x
230
//      d4: x/x
231
//      d5: x/x
232
//      d6: x/ILOG
233
//      d7: k-factor/Unchanged
234
//      a0: ptr for original operand/final result
235
//      a1: x/x
236
//      a2: x/x
237
//      fp0: x/float(ILOG)
238
//      fp1: x/x
239
//      fp2: x/x
240
//      F_SCR1:x/x
241
//      F_SCR2:Abs(X)/Abs(X) with $3fff exponent
242
//      L_SCR1:x/x
243
//      L_SCR2:first word of X packed/Unchanged
244
 
245
        tstb    BINDEC_FLG(%a6) //check for denorm
246
        beqs    A3_cont         //if clr, continue with norm
247
        movel   #-4933,%d6      //force ILOG = -4933
248
        bras    A4_str
249
A3_cont:
250
        movew   FP_SCR2(%a6),%d0        //move exp to d0
251
        movew   #0x3fff,FP_SCR2(%a6) //replace exponent with 0x3fff
252
        fmovex  FP_SCR2(%a6),%fp0       //now fp0 has 1.f
253
        subw    #0x3fff,%d0     //strip off bias
254
        faddw   %d0,%fp0                //add in exp
255
        fsubs   FONE,%fp0       //subtract off 1.0
256
        fbge    pos_res         //if pos, branch
257
        fmulx   LOG2UP1,%fp0    //if neg, mul by LOG2UP1
258
        fmovel  %fp0,%d6                //put ILOG in d6 as a lword
259
        bras    A4_str          //go move out ILOG
260
pos_res:
261
        fmulx   LOG2,%fp0       //if pos, mul by LOG2
262
        fmovel  %fp0,%d6                //put ILOG in d6 as a lword
263
 
264
 
265
// A4. Clr INEX bit.
266
//     The operation in A3 above may have set INEX2.
267
 
268
A4_str:
269
        fmovel  #0,%FPSR                //zero all of fpsr - nothing needed
270
 
271
 
272
// A5. Set ICTR = 0;
273
//     ICTR is a flag used in A13.  It must be set before the
274
//     loop entry A6. The lower word of d5 is used for ICTR.
275
 
276
        clrw    %d5             //clear ICTR
277
 
278
 
279
// A6. Calculate LEN.
280
//     LEN is the number of digits to be displayed.  The k-factor
281
//     can dictate either the total number of digits, if it is
282
//     a positive number, or the number of digits after the
283
//     original decimal point which are to be included as
284
//     significant.  See the 68882 manual for examples.
285
//     If LEN is computed to be greater than 17, set OPERR in
286
//     USER_FPSR.  LEN is stored in d4.
287
//
288
// Register usage:
289
//      Input/Output
290
//      d0: exponent/Unchanged
291
//      d2: x/x/scratch
292
//      d3: x/x
293
//      d4: exc picture/LEN
294
//      d5: ICTR/Unchanged
295
//      d6: ILOG/Unchanged
296
//      d7: k-factor/Unchanged
297
//      a0: ptr for original operand/final result
298
//      a1: x/x
299
//      a2: x/x
300
//      fp0: float(ILOG)/Unchanged
301
//      fp1: x/x
302
//      fp2: x/x
303
//      F_SCR1:x/x
304
//      F_SCR2:Abs(X) with $3fff exponent/Unchanged
305
//      L_SCR1:x/x
306
//      L_SCR2:first word of X packed/Unchanged
307
 
308
A6_str:
309
        tstl    %d7             //branch on sign of k
310
        bles    k_neg           //if k <= 0, LEN = ILOG + 1 - k
311
        movel   %d7,%d4         //if k > 0, LEN = k
312
        bras    len_ck          //skip to LEN check
313
k_neg:
314
        movel   %d6,%d4         //first load ILOG to d4
315
        subl    %d7,%d4         //subtract off k
316
        addql   #1,%d4          //add in the 1
317
len_ck:
318
        tstl    %d4             //LEN check: branch on sign of LEN
319
        bles    LEN_ng          //if neg, set LEN = 1
320
        cmpl    #17,%d4         //test if LEN > 17
321
        bles    A7_str          //if not, forget it
322
        movel   #17,%d4         //set max LEN = 17
323
        tstl    %d7             //if negative, never set OPERR
324
        bles    A7_str          //if positive, continue
325
        orl     #opaop_mask,USER_FPSR(%a6) //set OPERR & AIOP in USER_FPSR
326
        bras    A7_str          //finished here
327
LEN_ng:
328
        moveql  #1,%d4          //min LEN is 1
329
 
330
 
331
// A7. Calculate SCALE.
332
//     SCALE is equal to 10^ISCALE, where ISCALE is the number
333
//     of decimal places needed to insure LEN integer digits
334
//     in the output before conversion to bcd. LAMBDA is the sign
335
//     of ISCALE, used in A9.  Fp1 contains 10^^(abs(ISCALE)) using
336
//     the rounding mode as given in the following table (see
337
//     Coonen, p. 7.23 as ref.; however, the SCALE variable is
338
//     of opposite sign in bindec.sa from Coonen).
339
//
340
//      Initial                                 USE
341
//      FPCR[6:5]       LAMBDA  SIGN(X)         FPCR[6:5]
342
//      ----------------------------------------------
343
//       RN     00         0       0            00/0    RN
344
//       RN     00         0       1            00/0    RN
345
//       RN     00         1       0            00/0    RN
346
//       RN     00         1       1            00/0    RN
347
//       RZ     01         0       0            11/3    RP
348
//       RZ     01         0       1            11/3    RP
349
//       RZ     01         1       0            10/2    RM
350
//       RZ     01         1       1            10/2    RM
351
//       RM     10         0       0            11/3    RP
352
//       RM     10         0       1            10/2    RM
353
//       RM     10         1       0            10/2    RM
354
//       RM     10         1       1            11/3    RP
355
//       RP     11         0       0            10/2    RM
356
//       RP     11         0       1            11/3    RP
357
//       RP     11         1       0            11/3    RP
358
//       RP     11         1       1            10/2    RM
359
//
360
// Register usage:
361
//      Input/Output
362
//      d0: exponent/scratch - final is 0
363
//      d2: x/0 or 24 for A9
364
//      d3: x/scratch - offset ptr into PTENRM array
365
//      d4: LEN/Unchanged
366
//      d5: 0/ICTR:LAMBDA
367
//      d6: ILOG/ILOG or k if ((k<=0)&(ILOG
368
//      d7: k-factor/Unchanged
369
//      a0: ptr for original operand/final result
370
//      a1: x/ptr to PTENRM array
371
//      a2: x/x
372
//      fp0: float(ILOG)/Unchanged
373
//      fp1: x/10^ISCALE
374
//      fp2: x/x
375
//      F_SCR1:x/x
376
//      F_SCR2:Abs(X) with $3fff exponent/Unchanged
377
//      L_SCR1:x/x
378
//      L_SCR2:first word of X packed/Unchanged
379
 
380
A7_str:
381
        tstl    %d7             //test sign of k
382
        bgts    k_pos           //if pos and > 0, skip this
383
        cmpl    %d6,%d7         //test k - ILOG
384
        blts    k_pos           //if ILOG >= k, skip this
385
        movel   %d7,%d6         //if ((k<0) & (ILOG < k)) ILOG = k
386
k_pos:
387
        movel   %d6,%d0         //calc ILOG + 1 - LEN in d0
388
        addql   #1,%d0          //add the 1
389
        subl    %d4,%d0         //sub off LEN
390
        swap    %d5             //use upper word of d5 for LAMBDA
391
        clrw    %d5             //set it zero initially
392
        clrw    %d2             //set up d2 for very small case
393
        tstl    %d0             //test sign of ISCALE
394
        bges    iscale          //if pos, skip next inst
395
        addqw   #1,%d5          //if neg, set LAMBDA true
396
        cmpl    #0xffffecd4,%d0 //test iscale <= -4908
397
        bgts    no_inf          //if false, skip rest
398
        addil   #24,%d0         //add in 24 to iscale
399
        movel   #24,%d2         //put 24 in d2 for A9
400
no_inf:
401
        negl    %d0             //and take abs of ISCALE
402
iscale:
403
        fmoves  FONE,%fp1       //init fp1 to 1
404
        bfextu  USER_FPCR(%a6){#26:#2},%d1 //get initial rmode bits
405
        lslw    #1,%d1          //put them in bits 2:1
406
        addw    %d5,%d1         //add in LAMBDA
407
        lslw    #1,%d1          //put them in bits 3:1
408
        tstl    L_SCR2(%a6)     //test sign of original x
409
        bges    x_pos           //if pos, don't set bit 0
410
        addql   #1,%d1          //if neg, set bit 0
411
x_pos:
412
        leal    RBDTBL,%a2      //load rbdtbl base
413
        moveb   (%a2,%d1),%d3   //load d3 with new rmode
414
        lsll    #4,%d3          //put bits in proper position
415
        fmovel  %d3,%fpcr               //load bits into fpu
416
        lsrl    #4,%d3          //put bits in proper position
417
        tstb    %d3             //decode new rmode for pten table
418
        bnes    not_rn          //if zero, it is RN
419
        leal    PTENRN,%a1      //load a1 with RN table base
420
        bras    rmode           //exit decode
421
not_rn:
422
        lsrb    #1,%d3          //get lsb in carry
423
        bccs    not_rp          //if carry clear, it is RM
424
        leal    PTENRP,%a1      //load a1 with RP table base
425
        bras    rmode           //exit decode
426
not_rp:
427
        leal    PTENRM,%a1      //load a1 with RM table base
428
rmode:
429
        clrl    %d3             //clr table index
430
e_loop:
431
        lsrl    #1,%d0          //shift next bit into carry
432
        bccs    e_next          //if zero, skip the mul
433
        fmulx   (%a1,%d3),%fp1  //mul by 10**(d3_bit_no)
434
e_next:
435
        addl    #12,%d3         //inc d3 to next pwrten table entry
436
        tstl    %d0             //test if ISCALE is zero
437
        bnes    e_loop          //if not, loop
438
 
439
 
440
// A8. Clr INEX; Force RZ.
441
//     The operation in A3 above may have set INEX2.
442
//     RZ mode is forced for the scaling operation to insure
443
//     only one rounding error.  The grs bits are collected in
444
//     the INEX flag for use in A10.
445
//
446
// Register usage:
447
//      Input/Output
448
 
449
        fmovel  #0,%FPSR                //clr INEX
450
        fmovel  #rz_mode,%FPCR  //set RZ rounding mode
451
 
452
 
453
// A9. Scale X -> Y.
454
//     The mantissa is scaled to the desired number of significant
455
//     digits.  The excess digits are collected in INEX2. If mul,
456
//     Check d2 for excess 10 exponential value.  If not zero,
457
//     the iscale value would have caused the pwrten calculation
458
//     to overflow.  Only a negative iscale can cause this, so
459
//     multiply by 10^(d2), which is now only allowed to be 24,
460
//     with a multiply by 10^8 and 10^16, which is exact since
461
//     10^24 is exact.  If the input was denormalized, we must
462
//     create a busy stack frame with the mul command and the
463
//     two operands, and allow the fpu to complete the multiply.
464
//
465
// Register usage:
466
//      Input/Output
467
//      d0: FPCR with RZ mode/Unchanged
468
//      d2: 0 or 24/unchanged
469
//      d3: x/x
470
//      d4: LEN/Unchanged
471
//      d5: ICTR:LAMBDA
472
//      d6: ILOG/Unchanged
473
//      d7: k-factor/Unchanged
474
//      a0: ptr for original operand/final result
475
//      a1: ptr to PTENRM array/Unchanged
476
//      a2: x/x
477
//      fp0: float(ILOG)/X adjusted for SCALE (Y)
478
//      fp1: 10^ISCALE/Unchanged
479
//      fp2: x/x
480
//      F_SCR1:x/x
481
//      F_SCR2:Abs(X) with $3fff exponent/Unchanged
482
//      L_SCR1:x/x
483
//      L_SCR2:first word of X packed/Unchanged
484
 
485
A9_str:
486
        fmovex  (%a0),%fp0      //load X from memory
487
        fabsx   %fp0            //use abs(X)
488
        tstw    %d5             //LAMBDA is in lower word of d5
489
        bne     sc_mul          //if neg (LAMBDA = 1), scale by mul
490
        fdivx   %fp1,%fp0               //calculate X / SCALE -> Y to fp0
491
        bras    A10_st          //branch to A10
492
 
493
sc_mul:
494
        tstb    BINDEC_FLG(%a6) //check for denorm
495
        beqs    A9_norm         //if norm, continue with mul
496
        fmovemx %fp1-%fp1,-(%a7)        //load ETEMP with 10^ISCALE
497
        movel   8(%a0),-(%a7)   //load FPTEMP with input arg
498
        movel   4(%a0),-(%a7)
499
        movel   (%a0),-(%a7)
500
        movel   #18,%d3         //load count for busy stack
501
A9_loop:
502
        clrl    -(%a7)          //clear lword on stack
503
        dbf     %d3,A9_loop
504
        moveb   VER_TMP(%a6),(%a7) //write current version number
505
        moveb   #BUSY_SIZE-4,1(%a7) //write current busy size
506
        moveb   #0x10,0x44(%a7) //set fcefpte[15] bit
507
        movew   #0x0023,0x40(%a7)       //load cmdreg1b with mul command
508
        moveb   #0xfe,0x8(%a7)  //load all 1s to cu savepc
509
        frestore (%a7)+         //restore frame to fpu for completion
510
        fmulx   36(%a1),%fp0    //multiply fp0 by 10^8
511
        fmulx   48(%a1),%fp0    //multiply fp0 by 10^16
512
        bras    A10_st
513
A9_norm:
514
        tstw    %d2             //test for small exp case
515
        beqs    A9_con          //if zero, continue as normal
516
        fmulx   36(%a1),%fp0    //multiply fp0 by 10^8
517
        fmulx   48(%a1),%fp0    //multiply fp0 by 10^16
518
A9_con:
519
        fmulx   %fp1,%fp0               //calculate X * SCALE -> Y to fp0
520
 
521
 
522
// A10. Or in INEX.
523
//      If INEX is set, round error occurred.  This is compensated
524
//      for by 'or-ing' in the INEX2 flag to the lsb of Y.
525
//
526
// Register usage:
527
//      Input/Output
528
//      d0: FPCR with RZ mode/FPSR with INEX2 isolated
529
//      d2: x/x
530
//      d3: x/x
531
//      d4: LEN/Unchanged
532
//      d5: ICTR:LAMBDA
533
//      d6: ILOG/Unchanged
534
//      d7: k-factor/Unchanged
535
//      a0: ptr for original operand/final result
536
//      a1: ptr to PTENxx array/Unchanged
537
//      a2: x/ptr to FP_SCR2(a6)
538
//      fp0: Y/Y with lsb adjusted
539
//      fp1: 10^ISCALE/Unchanged
540
//      fp2: x/x
541
 
542
A10_st:
543
        fmovel  %FPSR,%d0               //get FPSR
544
        fmovex  %fp0,FP_SCR2(%a6)       //move Y to memory
545
        leal    FP_SCR2(%a6),%a2        //load a2 with ptr to FP_SCR2
546
        btstl   #9,%d0          //check if INEX2 set
547
        beqs    A11_st          //if clear, skip rest
548
        oril    #1,8(%a2)       //or in 1 to lsb of mantissa
549
        fmovex  FP_SCR2(%a6),%fp0       //write adjusted Y back to fpu
550
 
551
 
552
// A11. Restore original FPCR; set size ext.
553
//      Perform FINT operation in the user's rounding mode.  Keep
554
//      the size to extended.  The sintdo entry point in the sint
555
//      routine expects the FPCR value to be in USER_FPCR for
556
//      mode and precision.  The original FPCR is saved in L_SCR1.
557
 
558
A11_st:
559
        movel   USER_FPCR(%a6),L_SCR1(%a6) //save it for later
560
        andil   #0x00000030,USER_FPCR(%a6) //set size to ext,
561
//                                      ;block exceptions
562
 
563
 
564
// A12. Calculate YINT = FINT(Y) according to user's rounding mode.
565
//      The FPSP routine sintd0 is used.  The output is in fp0.
566
//
567
// Register usage:
568
//      Input/Output
569
//      d0: FPSR with AINEX cleared/FPCR with size set to ext
570
//      d2: x/x/scratch
571
//      d3: x/x
572
//      d4: LEN/Unchanged
573
//      d5: ICTR:LAMBDA/Unchanged
574
//      d6: ILOG/Unchanged
575
//      d7: k-factor/Unchanged
576
//      a0: ptr for original operand/src ptr for sintdo
577
//      a1: ptr to PTENxx array/Unchanged
578
//      a2: ptr to FP_SCR2(a6)/Unchanged
579
//      a6: temp pointer to FP_SCR2(a6) - orig value saved and restored
580
//      fp0: Y/YINT
581
//      fp1: 10^ISCALE/Unchanged
582
//      fp2: x/x
583
//      F_SCR1:x/x
584
//      F_SCR2:Y adjusted for inex/Y with original exponent
585
//      L_SCR1:x/original USER_FPCR
586
//      L_SCR2:first word of X packed/Unchanged
587
 
588
A12_st:
589
        moveml  %d0-%d1/%a0-%a1,-(%a7)  //save regs used by sintd0
590
        movel   L_SCR1(%a6),-(%a7)
591
        movel   L_SCR2(%a6),-(%a7)
592
        leal    FP_SCR2(%a6),%a0                //a0 is ptr to F_SCR2(a6)
593
        fmovex  %fp0,(%a0)              //move Y to memory at FP_SCR2(a6)
594
        tstl    L_SCR2(%a6)             //test sign of original operand
595
        bges    do_fint                 //if pos, use Y
596
        orl     #0x80000000,(%a0)               //if neg, use -Y
597
do_fint:
598
        movel   USER_FPSR(%a6),-(%a7)
599
        bsr     sintdo                  //sint routine returns int in fp0
600
        moveb   (%a7),USER_FPSR(%a6)
601
        addl    #4,%a7
602
        movel   (%a7)+,L_SCR2(%a6)
603
        movel   (%a7)+,L_SCR1(%a6)
604
        moveml  (%a7)+,%d0-%d1/%a0-%a1  //restore regs used by sint
605
        movel   L_SCR2(%a6),FP_SCR2(%a6)        //restore original exponent
606
        movel   L_SCR1(%a6),USER_FPCR(%a6) //restore user's FPCR
607
 
608
 
609
// A13. Check for LEN digits.
610
//      If the int operation results in more than LEN digits,
611
//      or less than LEN -1 digits, adjust ILOG and repeat from
612
//      A6.  This test occurs only on the first pass.  If the
613
//      result is exactly 10^LEN, decrement ILOG and divide
614
//      the mantissa by 10.  The calculation of 10^LEN cannot
615
//      be inexact, since all powers of ten upto 10^27 are exact
616
//      in extended precision, so the use of a previous power-of-ten
617
//      table will introduce no error.
618
//
619
//
620
// Register usage:
621
//      Input/Output
622
//      d0: FPCR with size set to ext/scratch final = 0
623
//      d2: x/x
624
//      d3: x/scratch final = x
625
//      d4: LEN/LEN adjusted
626
//      d5: ICTR:LAMBDA/LAMBDA:ICTR
627
//      d6: ILOG/ILOG adjusted
628
//      d7: k-factor/Unchanged
629
//      a0: pointer into memory for packed bcd string formation
630
//      a1: ptr to PTENxx array/Unchanged
631
//      a2: ptr to FP_SCR2(a6)/Unchanged
632
//      fp0: int portion of Y/abs(YINT) adjusted
633
//      fp1: 10^ISCALE/Unchanged
634
//      fp2: x/10^LEN
635
//      F_SCR1:x/x
636
//      F_SCR2:Y with original exponent/Unchanged
637
//      L_SCR1:original USER_FPCR/Unchanged
638
//      L_SCR2:first word of X packed/Unchanged
639
 
640
A13_st:
641
        swap    %d5             //put ICTR in lower word of d5
642
        tstw    %d5             //check if ICTR = 0
643
        bne     not_zr          //if non-zero, go to second test
644
//
645
// Compute 10^(LEN-1)
646
//
647
        fmoves  FONE,%fp2       //init fp2 to 1.0
648
        movel   %d4,%d0         //put LEN in d0
649
        subql   #1,%d0          //d0 = LEN -1
650
        clrl    %d3             //clr table index
651
l_loop:
652
        lsrl    #1,%d0          //shift next bit into carry
653
        bccs    l_next          //if zero, skip the mul
654
        fmulx   (%a1,%d3),%fp2  //mul by 10**(d3_bit_no)
655
l_next:
656
        addl    #12,%d3         //inc d3 to next pwrten table entry
657
        tstl    %d0             //test if LEN is zero
658
        bnes    l_loop          //if not, loop
659
//
660
// 10^LEN-1 is computed for this test and A14.  If the input was
661
// denormalized, check only the case in which YINT > 10^LEN.
662
//
663
        tstb    BINDEC_FLG(%a6) //check if input was norm
664
        beqs    A13_con         //if norm, continue with checking
665
        fabsx   %fp0            //take abs of YINT
666
        bra     test_2
667
//
668
// Compare abs(YINT) to 10^(LEN-1) and 10^LEN
669
//
670
A13_con:
671
        fabsx   %fp0            //take abs of YINT
672
        fcmpx   %fp2,%fp0               //compare abs(YINT) with 10^(LEN-1)
673
        fbge    test_2          //if greater, do next test
674
        subql   #1,%d6          //subtract 1 from ILOG
675
        movew   #1,%d5          //set ICTR
676
        fmovel  #rm_mode,%FPCR  //set rmode to RM
677
        fmuls   FTEN,%fp2       //compute 10^LEN
678
        bra     A6_str          //return to A6 and recompute YINT
679
test_2:
680
        fmuls   FTEN,%fp2       //compute 10^LEN
681
        fcmpx   %fp2,%fp0               //compare abs(YINT) with 10^LEN
682
        fblt    A14_st          //if less, all is ok, go to A14
683
        fbgt    fix_ex          //if greater, fix and redo
684
        fdivs   FTEN,%fp0       //if equal, divide by 10
685
        addql   #1,%d6          // and inc ILOG
686
        bras    A14_st          // and continue elsewhere
687
fix_ex:
688
        addql   #1,%d6          //increment ILOG by 1
689
        movew   #1,%d5          //set ICTR
690
        fmovel  #rm_mode,%FPCR  //set rmode to RM
691
        bra     A6_str          //return to A6 and recompute YINT
692
//
693
// Since ICTR <> 0, we have already been through one adjustment,
694
// and shouldn't have another; this is to check if abs(YINT) = 10^LEN
695
// 10^LEN is again computed using whatever table is in a1 since the
696
// value calculated cannot be inexact.
697
//
698
not_zr:
699
        fmoves  FONE,%fp2       //init fp2 to 1.0
700
        movel   %d4,%d0         //put LEN in d0
701
        clrl    %d3             //clr table index
702
z_loop:
703
        lsrl    #1,%d0          //shift next bit into carry
704
        bccs    z_next          //if zero, skip the mul
705
        fmulx   (%a1,%d3),%fp2  //mul by 10**(d3_bit_no)
706
z_next:
707
        addl    #12,%d3         //inc d3 to next pwrten table entry
708
        tstl    %d0             //test if LEN is zero
709
        bnes    z_loop          //if not, loop
710
        fabsx   %fp0            //get abs(YINT)
711
        fcmpx   %fp2,%fp0               //check if abs(YINT) = 10^LEN
712
        fbne    A14_st          //if not, skip this
713
        fdivs   FTEN,%fp0       //divide abs(YINT) by 10
714
        addql   #1,%d6          //and inc ILOG by 1
715
        addql   #1,%d4          // and inc LEN
716
        fmuls   FTEN,%fp2       // if LEN++, the get 10^^LEN
717
 
718
 
719
// A14. Convert the mantissa to bcd.
720
//      The binstr routine is used to convert the LEN digit
721
//      mantissa to bcd in memory.  The input to binstr is
722
//      to be a fraction; i.e. (mantissa)/10^LEN and adjusted
723
//      such that the decimal point is to the left of bit 63.
724
//      The bcd digits are stored in the correct position in
725
//      the final string area in memory.
726
//
727
//
728
// Register usage:
729
//      Input/Output
730
//      d0: x/LEN call to binstr - final is 0
731
//      d1: x/0
732
//      d2: x/ms 32-bits of mant of abs(YINT)
733
//      d3: x/ls 32-bits of mant of abs(YINT)
734
//      d4: LEN/Unchanged
735
//      d5: ICTR:LAMBDA/LAMBDA:ICTR
736
//      d6: ILOG
737
//      d7: k-factor/Unchanged
738
//      a0: pointer into memory for packed bcd string formation
739
//          /ptr to first mantissa byte in result string
740
//      a1: ptr to PTENxx array/Unchanged
741
//      a2: ptr to FP_SCR2(a6)/Unchanged
742
//      fp0: int portion of Y/abs(YINT) adjusted
743
//      fp1: 10^ISCALE/Unchanged
744
//      fp2: 10^LEN/Unchanged
745
//      F_SCR1:x/Work area for final result
746
//      F_SCR2:Y with original exponent/Unchanged
747
//      L_SCR1:original USER_FPCR/Unchanged
748
//      L_SCR2:first word of X packed/Unchanged
749
 
750
A14_st:
751
        fmovel  #rz_mode,%FPCR  //force rz for conversion
752
        fdivx   %fp2,%fp0               //divide abs(YINT) by 10^LEN
753
        leal    FP_SCR1(%a6),%a0
754
        fmovex  %fp0,(%a0)      //move abs(YINT)/10^LEN to memory
755
        movel   4(%a0),%d2      //move 2nd word of FP_RES to d2
756
        movel   8(%a0),%d3      //move 3rd word of FP_RES to d3
757
        clrl    4(%a0)          //zero word 2 of FP_RES
758
        clrl    8(%a0)          //zero word 3 of FP_RES
759
        movel   (%a0),%d0               //move exponent to d0
760
        swap    %d0             //put exponent in lower word
761
        beqs    no_sft          //if zero, don't shift
762
        subil   #0x3ffd,%d0     //sub bias less 2 to make fract
763
        tstl    %d0             //check if > 1
764
        bgts    no_sft          //if so, don't shift
765
        negl    %d0             //make exp positive
766
m_loop:
767
        lsrl    #1,%d2          //shift d2:d3 right, add 0s
768
        roxrl   #1,%d3          //the number of places
769
        dbf     %d0,m_loop      //given in d0
770
no_sft:
771
        tstl    %d2             //check for mantissa of zero
772
        bnes    no_zr           //if not, go on
773
        tstl    %d3             //continue zero check
774
        beqs    zer_m           //if zero, go directly to binstr
775
no_zr:
776
        clrl    %d1             //put zero in d1 for addx
777
        addil   #0x00000080,%d3 //inc at bit 7
778
        addxl   %d1,%d2         //continue inc
779
        andil   #0xffffff80,%d3 //strip off lsb not used by 882
780
zer_m:
781
        movel   %d4,%d0         //put LEN in d0 for binstr call
782
        addql   #3,%a0          //a0 points to M16 byte in result
783
        bsr     binstr          //call binstr to convert mant
784
 
785
 
786
// A15. Convert the exponent to bcd.
787
//      As in A14 above, the exp is converted to bcd and the
788
//      digits are stored in the final string.
789
//
790
//      Digits are stored in L_SCR1(a6) on return from BINDEC as:
791
//
792
//       32               16 15                0
793
//      -----------------------------------------
794
//      |  0 | e3 | e2 | e1 | e4 |  X |  X |  X |
795
//      -----------------------------------------
796
//
797
// And are moved into their proper places in FP_SCR1.  If digit e4
798
// is non-zero, OPERR is signaled.  In all cases, all 4 digits are
799
// written as specified in the 881/882 manual for packed decimal.
800
//
801
// Register usage:
802
//      Input/Output
803
//      d0: x/LEN call to binstr - final is 0
804
//      d1: x/scratch (0);shift count for final exponent packing
805
//      d2: x/ms 32-bits of exp fraction/scratch
806
//      d3: x/ls 32-bits of exp fraction
807
//      d4: LEN/Unchanged
808
//      d5: ICTR:LAMBDA/LAMBDA:ICTR
809
//      d6: ILOG
810
//      d7: k-factor/Unchanged
811
//      a0: ptr to result string/ptr to L_SCR1(a6)
812
//      a1: ptr to PTENxx array/Unchanged
813
//      a2: ptr to FP_SCR2(a6)/Unchanged
814
//      fp0: abs(YINT) adjusted/float(ILOG)
815
//      fp1: 10^ISCALE/Unchanged
816
//      fp2: 10^LEN/Unchanged
817
//      F_SCR1:Work area for final result/BCD result
818
//      F_SCR2:Y with original exponent/ILOG/10^4
819
//      L_SCR1:original USER_FPCR/Exponent digits on return from binstr
820
//      L_SCR2:first word of X packed/Unchanged
821
 
822
A15_st:
823
        tstb    BINDEC_FLG(%a6) //check for denorm
824
        beqs    not_denorm
825
        ftstx   %fp0            //test for zero
826
        fbeq    den_zero        //if zero, use k-factor or 4933
827
        fmovel  %d6,%fp0                //float ILOG
828
        fabsx   %fp0            //get abs of ILOG
829
        bras    convrt
830
den_zero:
831
        tstl    %d7             //check sign of the k-factor
832
        blts    use_ilog        //if negative, use ILOG
833
        fmoves  F4933,%fp0      //force exponent to 4933
834
        bras    convrt          //do it
835
use_ilog:
836
        fmovel  %d6,%fp0                //float ILOG
837
        fabsx   %fp0            //get abs of ILOG
838
        bras    convrt
839
not_denorm:
840
        ftstx   %fp0            //test for zero
841
        fbne    not_zero        //if zero, force exponent
842
        fmoves  FONE,%fp0       //force exponent to 1
843
        bras    convrt          //do it
844
not_zero:
845
        fmovel  %d6,%fp0                //float ILOG
846
        fabsx   %fp0            //get abs of ILOG
847
convrt:
848
        fdivx   24(%a1),%fp0    //compute ILOG/10^4
849
        fmovex  %fp0,FP_SCR2(%a6)       //store fp0 in memory
850
        movel   4(%a2),%d2      //move word 2 to d2
851
        movel   8(%a2),%d3      //move word 3 to d3
852
        movew   (%a2),%d0               //move exp to d0
853
        beqs    x_loop_fin      //if zero, skip the shift
854
        subiw   #0x3ffd,%d0     //subtract off bias
855
        negw    %d0             //make exp positive
856
x_loop:
857
        lsrl    #1,%d2          //shift d2:d3 right
858
        roxrl   #1,%d3          //the number of places
859
        dbf     %d0,x_loop      //given in d0
860
x_loop_fin:
861
        clrl    %d1             //put zero in d1 for addx
862
        addil   #0x00000080,%d3 //inc at bit 6
863
        addxl   %d1,%d2         //continue inc
864
        andil   #0xffffff80,%d3 //strip off lsb not used by 882
865
        movel   #4,%d0          //put 4 in d0 for binstr call
866
        leal    L_SCR1(%a6),%a0 //a0 is ptr to L_SCR1 for exp digits
867
        bsr     binstr          //call binstr to convert exp
868
        movel   L_SCR1(%a6),%d0 //load L_SCR1 lword to d0
869
        movel   #12,%d1         //use d1 for shift count
870
        lsrl    %d1,%d0         //shift d0 right by 12
871
        bfins   %d0,FP_SCR1(%a6){#4:#12} //put e3:e2:e1 in FP_SCR1
872
        lsrl    %d1,%d0         //shift d0 right by 12
873
        bfins   %d0,FP_SCR1(%a6){#16:#4} //put e4 in FP_SCR1
874
        tstb    %d0             //check if e4 is zero
875
        beqs    A16_st          //if zero, skip rest
876
        orl     #opaop_mask,USER_FPSR(%a6) //set OPERR & AIOP in USER_FPSR
877
 
878
 
879
// A16. Write sign bits to final string.
880
//         Sigma is bit 31 of initial value; RHO is bit 31 of d6 (ILOG).
881
//
882
// Register usage:
883
//      Input/Output
884
//      d0: x/scratch - final is x
885
//      d2: x/x
886
//      d3: x/x
887
//      d4: LEN/Unchanged
888
//      d5: ICTR:LAMBDA/LAMBDA:ICTR
889
//      d6: ILOG/ILOG adjusted
890
//      d7: k-factor/Unchanged
891
//      a0: ptr to L_SCR1(a6)/Unchanged
892
//      a1: ptr to PTENxx array/Unchanged
893
//      a2: ptr to FP_SCR2(a6)/Unchanged
894
//      fp0: float(ILOG)/Unchanged
895
//      fp1: 10^ISCALE/Unchanged
896
//      fp2: 10^LEN/Unchanged
897
//      F_SCR1:BCD result with correct signs
898
//      F_SCR2:ILOG/10^4
899
//      L_SCR1:Exponent digits on return from binstr
900
//      L_SCR2:first word of X packed/Unchanged
901
 
902
A16_st:
903
        clrl    %d0             //clr d0 for collection of signs
904
        andib   #0x0f,FP_SCR1(%a6) //clear first nibble of FP_SCR1
905
        tstl    L_SCR2(%a6)     //check sign of original mantissa
906
        bges    mant_p          //if pos, don't set SM
907
        moveql  #2,%d0          //move 2 in to d0 for SM
908
mant_p:
909
        tstl    %d6             //check sign of ILOG
910
        bges    wr_sgn          //if pos, don't set SE
911
        addql   #1,%d0          //set bit 0 in d0 for SE
912
wr_sgn:
913
        bfins   %d0,FP_SCR1(%a6){#0:#2} //insert SM and SE into FP_SCR1
914
 
915
// Clean up and restore all registers used.
916
 
917
        fmovel  #0,%FPSR                //clear possible inex2/ainex bits
918
        fmovemx (%a7)+,%fp0-%fp2
919
        moveml  (%a7)+,%d2-%d7/%a2
920
        rts
921
 
922
        |end

powered by: WebSVN 2.1.0

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