summaryrefslogblamecommitdiffstats
path: root/arch/m68k/fpsp040/decbin.S
blob: 16ed796bad87f375b98e21e5804aa375c4648858 (plain) (tree)
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505






































































                                                                               

                                                                















































































































































































































































































































































































































































                                                                                
|
|	decbin.sa 3.3 12/19/90
|
|	Description: Converts normalized packed bcd value pointed to by
|	register A6 to extended-precision value in FP0.
|
|	Input: Normalized packed bcd value in ETEMP(a6).
|
|	Output:	Exact floating-point representation of the packed bcd value.
|
|	Saves and Modifies: D2-D5
|
|	Speed: The program decbin takes ??? cycles to execute.
|
|	Object Size:
|
|	External Reference(s): None.
|
|	Algorithm:
|	Expected is a normal bcd (i.e. non-exceptional; all inf, zero,
|	and NaN operands are dispatched without entering this routine)
|	value in 68881/882 format at location ETEMP(A6).
|
|	A1.	Convert the bcd exponent to binary by successive adds and muls.
|	Set the sign according to SE. Subtract 16 to compensate
|	for the mantissa which is to be interpreted as 17 integer
|	digits, rather than 1 integer and 16 fraction digits.
|	Note: this operation can never overflow.
|
|	A2. Convert the bcd mantissa to binary by successive
|	adds and muls in FP0. Set the sign according to SM.
|	The mantissa digits will be converted with the decimal point
|	assumed following the least-significant digit.
|	Note: this operation can never overflow.
|
|	A3. Count the number of leading/trailing zeros in the
|	bcd string.  If SE is positive, count the leading zeros;
|	if negative, count the trailing zeros.  Set the adjusted
|	exponent equal to the exponent from A1 and the zero count
|	added if SM = 1 and subtracted if SM = 0.  Scale the
|	mantissa the equivalent of forcing in the bcd value:
|
|	SM = 0	a non-zero digit in the integer position
|	SM = 1	a non-zero digit in Mant0, lsd of the fraction
|
|	this will insure that any value, regardless of its
|	representation (ex. 0.1E2, 1E1, 10E0, 100E-1), is converted
|	consistently.
|
|	A4. Calculate the factor 10^exp in FP1 using a table of
|	10^(2^n) values.  To reduce the error in forming factors
|	greater than 10^27, a directed rounding scheme is used with
|	tables rounded to RN, RM, and RP, according to the table
|	in the comments of the pwrten section.
|
|	A5. Form the final binary number by scaling the mantissa by
|	the exponent factor.  This is done by multiplying the
|	mantissa in FP0 by the factor in FP1 if the adjusted
|	exponent sign is positive, and dividing FP0 by FP1 if
|	it is negative.
|
|	Clean up and return.  Check if the final mul or div resulted
|	in an inex2 exception.  If so, set inex1 in the fpsr and
|	check if the inex1 exception is enabled.  If so, set d7 upper
|	word to $0100.  This will signal unimp.sa that an enabled inex1
|	exception occurred.  Unimp will fix the stack.
|

|		Copyright (C) Motorola, Inc. 1990
|			All Rights Reserved
|
|       For details on the license for this file, please see the
|       file, README, in this same directory.

|DECBIN    idnt    2,1 | Motorola 040 Floating Point Software Package

	|section	8

#include "fpsp.h"

|
|	PTENRN, PTENRM, and PTENRP are arrays of powers of 10 rounded
|	to nearest, minus, and plus, respectively.  The tables include
|	10**{1,2,4,8,16,32,64,128,256,512,1024,2048,4096}.  No rounding
|	is required until the power is greater than 27, however, all
|	tables include the first 5 for ease of indexing.
|
	|xref	PTENRN
	|xref	PTENRM
	|xref	PTENRP

RTABLE:	.byte	0,0,0,0
	.byte	2,3,2,3
	.byte	2,3,3,2
	.byte	3,2,2,3

	.global	decbin
	.global	calc_e
	.global	pwrten
	.global	calc_m
	.global	norm
	.global	ap_st_z
	.global	ap_st_n
|
	.set	FNIBS,7
	.set	FSTRT,0
|
	.set	ESTRT,4
	.set	EDIGITS,2	|
|
| Constants in single precision
FZERO:	.long	0x00000000
FONE:	.long	0x3F800000
FTEN:	.long	0x41200000

	.set	TEN,10

|
decbin:
	| fmovel	#0,FPCR		;clr real fpcr
	moveml	%d2-%d5,-(%a7)
|
| Calculate exponent:
|  1. Copy bcd value in memory for use as a working copy.
|  2. Calculate absolute value of exponent in d1 by mul and add.
|  3. Correct for exponent sign.
|  4. Subtract 16 to compensate for interpreting the mant as all integer digits.
|     (i.e., all digits assumed left of the decimal point.)
|
| Register usage:
|
|  calc_e:
|	(*)  d0: temp digit storage
|	(*)  d1: accumulator for binary exponent
|	(*)  d2: digit count
|	(*)  d3: offset pointer
|	( )  d4: first word of bcd
|	( )  a0: pointer to working bcd value
|	( )  a6: pointer to original bcd value
|	(*)  FP_SCR1: working copy of original bcd value
|	(*)  L_SCR1: copy of original exponent word
|
calc_e:
	movel	#EDIGITS,%d2	|# of nibbles (digits) in fraction part
	moveql	#ESTRT,%d3	|counter to pick up digits
	leal	FP_SCR1(%a6),%a0	|load tmp bcd storage address
	movel	ETEMP(%a6),(%a0)	|save input bcd value
	movel	ETEMP_HI(%a6),4(%a0) |save words 2 and 3
	movel	ETEMP_LO(%a6),8(%a0) |and work with these
	movel	(%a0),%d4	|get first word of bcd
	clrl	%d1		|zero d1 for accumulator
e_gd:
	mulul	#TEN,%d1	|mul partial product by one digit place
	bfextu	%d4{%d3:#4},%d0	|get the digit and zero extend into d0
	addl	%d0,%d1		|d1 = d1 + d0
	addqb	#4,%d3		|advance d3 to the next digit
	dbf	%d2,e_gd	|if we have used all 3 digits, exit loop
	btst	#30,%d4		|get SE
	beqs	e_pos		|don't negate if pos
	negl	%d1		|negate before subtracting
e_pos:
	subl	#16,%d1		|sub to compensate for shift of mant
	bges	e_save		|if still pos, do not neg
	negl	%d1		|now negative, make pos and set SE
	orl	#0x40000000,%d4	|set SE in d4,
	orl	#0x40000000,(%a0)	|and in working bcd
e_save:
	movel	%d1,L_SCR1(%a6)	|save exp in memory
|
|
| Calculate mantissa:
|  1. Calculate absolute value of mantissa in fp0 by mul and add.
|  2. Correct for mantissa sign.
|     (i.e., all digits assumed left of the decimal point.)
|
| Register usage:
|
|  calc_m:
|	(*)  d0: temp digit storage
|	(*)  d1: lword counter
|	(*)  d2: digit count
|	(*)  d3: offset pointer
|	( )  d4: words 2 and 3 of bcd
|	( )  a0: pointer to working bcd value
|	( )  a6: pointer to original bcd value
|	(*) fp0: mantissa accumulator
|	( )  FP_SCR1: working copy of original bcd value
|	( )  L_SCR1: copy of original exponent word
|
calc_m:
	moveql	#1,%d1		|word counter, init to 1
	fmoves	FZERO,%fp0	|accumulator
|
|
|  Since the packed number has a long word between the first & second parts,
|  get the integer digit then skip down & get the rest of the
|  mantissa.  We will unroll the loop once.
|
	bfextu	(%a0){#28:#4},%d0	|integer part is ls digit in long word
	faddb	%d0,%fp0		|add digit to sum in fp0
|
|
|  Get the rest of the mantissa.
|
loadlw:
	movel	(%a0,%d1.L*4),%d4	|load mantissa longword into d4
	moveql	#FSTRT,%d3	|counter to pick up digits
	moveql	#FNIBS,%d2	|reset number of digits per a0 ptr
md2b:
	fmuls	FTEN,%fp0	|fp0 = fp0 * 10
	bfextu	%d4{%d3:#4},%d0	|get the digit and zero extend
	faddb	%d0,%fp0	|fp0 = fp0 + digit
|
|
|  If all the digits (8) in that long word have been converted (d2=0),
|  then inc d1 (=2) to point to the next long word and reset d3 to 0
|  to initialize the digit offset, and set d2 to 7 for the digit count;
|  else continue with this long word.
|
	addqb	#4,%d3		|advance d3 to the next digit
	dbf	%d2,md2b		|check for last digit in this lw
nextlw:
	addql	#1,%d1		|inc lw pointer in mantissa
	cmpl	#2,%d1		|test for last lw
	ble	loadlw		|if not, get last one

|
|  Check the sign of the mant and make the value in fp0 the same sign.
|
m_sign:
	btst	#31,(%a0)	|test sign of the mantissa
	beq	ap_st_z		|if clear, go to append/strip zeros
	fnegx	%fp0		|if set, negate fp0

|
| Append/strip zeros:
|
|  For adjusted exponents which have an absolute value greater than 27*,
|  this routine calculates the amount needed to normalize the mantissa
|  for the adjusted exponent.  That number is subtracted from the exp
|  if the exp was positive, and added if it was negative.  The purpose
|  of this is to reduce the value of the exponent and the possibility
|  of error in calculation of pwrten.
|
|  1. Branch on the sign of the adjusted exponent.
|  2p.(positive exp)
|   2. Check M16 and the digits in lwords 2 and 3 in descending order.
|   3. Add one for each zero encountered until a non-zero digit.
|   4. Subtract the count from the exp.
|   5. Check if the exp has crossed zero in #3 above; make the exp abs
|	   and set SE.
|	6. Multiply the mantissa by 10**count.
|  2n.(negative exp)
|   2. Check the digits in lwords 3 and 2 in descending order.
|   3. Add one for each zero encountered until a non-zero digit.
|   4. Add the count to the exp.
|   5. Check if the exp has crossed zero in #3 above; clear SE.
|   6. Divide the mantissa by 10**count.
|
|  *Why 27?  If the adjusted exponent is within -28 < expA < 28, than
|   any adjustment due to append/strip zeros will drive the resultant
|   exponent towards zero.  Since all pwrten constants with a power
|   of 27 or less are exact, there is no need to use this routine to
|   attempt to lessen the resultant exponent.
|
| Register usage:
|
|  ap_st_z:
|	(*)  d0: temp digit storage
|	(*)  d1: zero count
|	(*)  d2: digit count
|	(*)  d3: offset pointer
|	( )  d4: first word of bcd
|	(*)  d5: lword counter
|	( )  a0: pointer to working bcd value
|	( )  FP_SCR1: working copy of original bcd value
|	( )  L_SCR1: copy of original exponent word
|
|
| First check the absolute value of the exponent to see if this
| routine is necessary.  If so, then check the sign of the exponent
| and do append (+) or strip (-) zeros accordingly.
| This section handles a positive adjusted exponent.
|
ap_st_z:
	movel	L_SCR1(%a6),%d1	|load expA for range test
	cmpl	#27,%d1		|test is with 27
	ble	pwrten		|if abs(expA) <28, skip ap/st zeros
	btst	#30,(%a0)	|check sign of exp
	bne	ap_st_n		|if neg, go to neg side
	clrl	%d1		|zero count reg
	movel	(%a0),%d4		|load lword 1 to d4
	bfextu	%d4{#28:#4},%d0	|get M16 in d0
	bnes	ap_p_fx		|if M16 is non-zero, go fix exp
	addql	#1,%d1		|inc zero count
	moveql	#1,%d5		|init lword counter
	movel	(%a0,%d5.L*4),%d4	|get lword 2 to d4
	bnes	ap_p_cl		|if lw 2 is zero, skip it
	addql	#8,%d1		|and inc count by 8
	addql	#1,%d5		|inc lword counter
	movel	(%a0,%d5.L*4),%d4	|get lword 3 to d4
ap_p_cl:
	clrl	%d3		|init offset reg
	moveql	#7,%d2		|init digit counter
ap_p_gd:
	bfextu	%d4{%d3:#4},%d0	|get digit
	bnes	ap_p_fx		|if non-zero, go to fix exp
	addql	#4,%d3		|point to next digit
	addql	#1,%d1		|inc digit counter
	dbf	%d2,ap_p_gd	|get next digit
ap_p_fx:
	movel	%d1,%d0		|copy counter to d2
	movel	L_SCR1(%a6),%d1	|get adjusted exp from memory
	subl	%d0,%d1		|subtract count from exp
	bges	ap_p_fm		|if still pos, go to pwrten
	negl	%d1		|now its neg; get abs
	movel	(%a0),%d4		|load lword 1 to d4
	orl	#0x40000000,%d4	| and set SE in d4
	orl	#0x40000000,(%a0)	| and in memory
|
| Calculate the mantissa multiplier to compensate for the striping of
| zeros from the mantissa.
|
ap_p_fm:
	movel	#PTENRN,%a1	|get address of power-of-ten table
	clrl	%d3		|init table index
	fmoves	FONE,%fp1	|init fp1 to 1
	moveql	#3,%d2		|init d2 to count bits in counter
ap_p_el:
	asrl	#1,%d0		|shift lsb into carry
	bccs	ap_p_en		|if 1, mul fp1 by pwrten factor
	fmulx	(%a1,%d3),%fp1	|mul by 10**(d3_bit_no)
ap_p_en:
	addl	#12,%d3		|inc d3 to next rtable entry
	tstl	%d0		|check if d0 is zero
	bnes	ap_p_el		|if not, get next bit
	fmulx	%fp1,%fp0		|mul mantissa by 10**(no_bits_shifted)
	bra	pwrten		|go calc pwrten
|
| This section handles a negative adjusted exponent.
|
ap_st_n:
	clrl	%d1		|clr counter
	moveql	#2,%d5		|set up d5 to point to lword 3
	movel	(%a0,%d5.L*4),%d4	|get lword 3
	bnes	ap_n_cl		|if not zero, check digits
	subl	#1,%d5		|dec d5 to point to lword 2
	addql	#8,%d1		|inc counter by 8
	movel	(%a0,%d5.L*4),%d4	|get lword 2
ap_n_cl:
	movel	#28,%d3		|point to last digit
	moveql	#7,%d2		|init digit counter
ap_n_gd:
	bfextu	%d4{%d3:#4},%d0	|get digit
	bnes	ap_n_fx		|if non-zero, go to exp fix
	subql	#4,%d3		|point to previous digit
	addql	#1,%d1		|inc digit counter
	dbf	%d2,ap_n_gd	|get next digit
ap_n_fx:
	movel	%d1,%d0		|copy counter to d0
	movel	L_SCR1(%a6),%d1	|get adjusted exp from memory
	subl	%d0,%d1		|subtract count from exp
	bgts	ap_n_fm		|if still pos, go fix mantissa
	negl	%d1		|take abs of exp and clr SE
	movel	(%a0),%d4		|load lword 1 to d4
	andl	#0xbfffffff,%d4	| and clr SE in d4
	andl	#0xbfffffff,(%a0)	| and in memory
|
| Calculate the mantissa multiplier to compensate for the appending of
| zeros to the mantissa.
|
ap_n_fm:
	movel	#PTENRN,%a1	|get address of power-of-ten table
	clrl	%d3		|init table index
	fmoves	FONE,%fp1	|init fp1 to 1
	moveql	#3,%d2		|init d2 to count bits in counter
ap_n_el:
	asrl	#1,%d0		|shift lsb into carry
	bccs	ap_n_en		|if 1, mul fp1 by pwrten factor
	fmulx	(%a1,%d3),%fp1	|mul by 10**(d3_bit_no)
ap_n_en:
	addl	#12,%d3		|inc d3 to next rtable entry
	tstl	%d0		|check if d0 is zero
	bnes	ap_n_el		|if not, get next bit
	fdivx	%fp1,%fp0		|div mantissa by 10**(no_bits_shifted)
|
|
| Calculate power-of-ten factor from adjusted and shifted exponent.
|
| Register usage:
|
|  pwrten:
|	(*)  d0: temp
|	( )  d1: exponent
|	(*)  d2: {FPCR[6:5],SM,SE} as index in RTABLE; temp
|	(*)  d3: FPCR work copy
|	( )  d4: first word of bcd
|	(*)  a1: RTABLE pointer
|  calc_p:
|	(*)  d0: temp
|	( )  d1: exponent
|	(*)  d3: PWRTxx table index
|	( )  a0: pointer to working copy of bcd
|	(*)  a1: PWRTxx pointer
|	(*) fp1: power-of-ten accumulator
|
| Pwrten calculates the exponent factor in the selected rounding mode
| according to the following table:
|
|	Sign of Mant  Sign of Exp  Rounding Mode  PWRTEN Rounding Mode
|
|	ANY	  ANY	RN	RN
|
|	 +	   +	RP	RP
|	 -	   +	RP	RM
|	 +	   -	RP	RM
|	 -	   -	RP	RP
|
|	 +	   +	RM	RM
|	 -	   +	RM	RP
|	 +	   -	RM	RP
|	 -	   -	RM	RM
|
|	 +	   +	RZ	RM
|	 -	   +	RZ	RM
|	 +	   -	RZ	RP
|	 -	   -	RZ	RP
|
|
pwrten:
	movel	USER_FPCR(%a6),%d3 |get user's FPCR
	bfextu	%d3{#26:#2},%d2	|isolate rounding mode bits
	movel	(%a0),%d4		|reload 1st bcd word to d4
	asll	#2,%d2		|format d2 to be
	bfextu	%d4{#0:#2},%d0	| {FPCR[6],FPCR[5],SM,SE}
	addl	%d0,%d2		|in d2 as index into RTABLE
	leal	RTABLE,%a1	|load rtable base
	moveb	(%a1,%d2),%d0	|load new rounding bits from table
	clrl	%d3			|clear d3 to force no exc and extended
	bfins	%d0,%d3{#26:#2}	|stuff new rounding bits in FPCR
	fmovel	%d3,%FPCR		|write new FPCR
	asrl	#1,%d0		|write correct PTENxx table
	bccs	not_rp		|to a1
	leal	PTENRP,%a1	|it is RP
	bras	calc_p		|go to init section
not_rp:
	asrl	#1,%d0		|keep checking
	bccs	not_rm
	leal	PTENRM,%a1	|it is RM
	bras	calc_p		|go to init section
not_rm:
	leal	PTENRN,%a1	|it is RN
calc_p:
	movel	%d1,%d0		|copy exp to d0;use d0
	bpls	no_neg		|if exp is negative,
	negl	%d0		|invert it
	orl	#0x40000000,(%a0)	|and set SE bit
no_neg:
	clrl	%d3		|table index
	fmoves	FONE,%fp1	|init fp1 to 1
e_loop:
	asrl	#1,%d0		|shift next bit into carry
	bccs	e_next		|if zero, skip the mul
	fmulx	(%a1,%d3),%fp1	|mul by 10**(d3_bit_no)
e_next:
	addl	#12,%d3		|inc d3 to next rtable entry
	tstl	%d0		|check if d0 is zero
	bnes	e_loop		|not zero, continue shifting
|
|
|  Check the sign of the adjusted exp and make the value in fp0 the
|  same sign. If the exp was pos then multiply fp1*fp0;
|  else divide fp0/fp1.
|
| Register Usage:
|  norm:
|	( )  a0: pointer to working bcd value
|	(*) fp0: mantissa accumulator
|	( ) fp1: scaling factor - 10**(abs(exp))
|
norm:
	btst	#30,(%a0)	|test the sign of the exponent
	beqs	mul		|if clear, go to multiply
div:
	fdivx	%fp1,%fp0		|exp is negative, so divide mant by exp
	bras	end_dec
mul:
	fmulx	%fp1,%fp0		|exp is positive, so multiply by exp
|
|
| Clean up and return with result in fp0.
|
| If the final mul/div in decbin incurred an inex exception,
| it will be inex2, but will be reported as inex1 by get_op.
|
end_dec:
	fmovel	%FPSR,%d0		|get status register
	bclrl	#inex2_bit+8,%d0	|test for inex2 and clear it
	fmovel	%d0,%FPSR		|return status reg w/o inex2
	beqs	no_exc		|skip this if no exc
	orl	#inx1a_mask,USER_FPSR(%a6) |set inex1/ainex
no_exc:
	moveml	(%a7)+,%d2-%d5
	rts
	|end