xref: /openbmc/linux/arch/m68k/fpsp040/kernel_ex.S (revision 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2)
1|
2|	kernel_ex.sa 3.3 12/19/90
3|
4| This file contains routines to force exception status in the
5| fpu for exceptional cases detected or reported within the
6| transcendental functions.  Typically, the t_xx routine will
7| set the appropriate bits in the USER_FPSR word on the stack.
8| The bits are tested in gen_except.sa to determine if an exceptional
9| situation needs to be created on return from the FPSP.
10|
11
12|		Copyright (C) Motorola, Inc. 1990
13|			All Rights Reserved
14|
15|	THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA
16|	The copyright notice above does not evidence any
17|	actual or intended publication of such source code.
18
19KERNEL_EX:    |idnt    2,1 | Motorola 040 Floating Point Software Package
20
21	|section    8
22
23#include "fpsp.h"
24
25mns_inf:  .long 0xffff0000,0x00000000,0x00000000
26pls_inf:  .long 0x7fff0000,0x00000000,0x00000000
27nan:      .long 0x7fff0000,0xffffffff,0xffffffff
28huge:     .long 0x7ffe0000,0xffffffff,0xffffffff
29
30	|xref	  ovf_r_k
31	|xref	  unf_sub
32	|xref	  nrm_set
33
34	.global	  t_dz
35	.global      t_dz2
36	.global      t_operr
37	.global      t_unfl
38	.global      t_ovfl
39	.global      t_ovfl2
40	.global      t_inx2
41	.global	  t_frcinx
42	.global	  t_extdnrm
43	.global	  t_resdnrm
44	.global	  dst_nan
45	.global	  src_nan
46|
47|	DZ exception
48|
49|
50|	if dz trap disabled
51|		store properly signed inf (use sign of etemp) into fp0
52|		set FPSR exception status dz bit, condition code
53|		inf bit, and accrued dz bit
54|		return
55|		frestore the frame into the machine (done by unimp_hd)
56|
57|	else dz trap enabled
58|		set exception status bit & accrued bits in FPSR
59|		set flag to disable sto_res from corrupting fp register
60|		return
61|		frestore the frame into the machine (done by unimp_hd)
62|
63| t_dz2 is used by monadic functions such as flogn (from do_func).
64| t_dz is used by monadic functions such as satanh (from the
65| transcendental function).
66|
67t_dz2:
68	bsetb	#neg_bit,FPSR_CC(%a6)	|set neg bit in FPSR
69	fmovel	#0,%FPSR			|clr status bits (Z set)
70	btstb	#dz_bit,FPCR_ENABLE(%a6)	|test FPCR for dz exc enabled
71	bnes	dz_ena_end
72	bras	m_inf			|flogx always returns -inf
73t_dz:
74	fmovel	#0,%FPSR			|clr status bits (Z set)
75	btstb	#dz_bit,FPCR_ENABLE(%a6)	|test FPCR for dz exc enabled
76	bnes	dz_ena
77|
78|	dz disabled
79|
80	btstb	#sign_bit,ETEMP_EX(%a6)	|check sign for neg or pos
81	beqs	p_inf			|branch if pos sign
82
83m_inf:
84	fmovemx mns_inf,%fp0-%fp0		|load -inf
85	bsetb	#neg_bit,FPSR_CC(%a6)	|set neg bit in FPSR
86	bras	set_fpsr
87p_inf:
88	fmovemx pls_inf,%fp0-%fp0		|load +inf
89set_fpsr:
90	orl	#dzinf_mask,USER_FPSR(%a6) |set I,DZ,ADZ
91	rts
92|
93|	dz enabled
94|
95dz_ena:
96	btstb	#sign_bit,ETEMP_EX(%a6)	|check sign for neg or pos
97	beqs	dz_ena_end
98	bsetb	#neg_bit,FPSR_CC(%a6)	|set neg bit in FPSR
99dz_ena_end:
100	orl	#dzinf_mask,USER_FPSR(%a6) |set I,DZ,ADZ
101	st	STORE_FLG(%a6)
102	rts
103|
104|	OPERR exception
105|
106|	if (operr trap disabled)
107|		set FPSR exception status operr bit, condition code
108|		nan bit; Store default NAN into fp0
109|		frestore the frame into the machine (done by unimp_hd)
110|
111|	else (operr trap enabled)
112|		set FPSR exception status operr bit, accrued operr bit
113|		set flag to disable sto_res from corrupting fp register
114|		frestore the frame into the machine (done by unimp_hd)
115|
116t_operr:
117	orl	#opnan_mask,USER_FPSR(%a6) |set NaN, OPERR, AIOP
118
119	btstb	#operr_bit,FPCR_ENABLE(%a6) |test FPCR for operr enabled
120	bnes	op_ena
121
122	fmovemx nan,%fp0-%fp0		|load default nan
123	rts
124op_ena:
125	st	STORE_FLG(%a6)		|do not corrupt destination
126	rts
127
128|
129|	t_unfl --- UNFL exception
130|
131| This entry point is used by all routines requiring unfl, inex2,
132| aunfl, and ainex to be set on exit.
133|
134| On entry, a0 points to the exceptional operand.  The final exceptional
135| operand is built in FP_SCR1 and only the sign from the original operand
136| is used.
137|
138t_unfl:
139	clrl	FP_SCR1(%a6)		|set exceptional operand to zero
140	clrl	FP_SCR1+4(%a6)
141	clrl	FP_SCR1+8(%a6)
142	tstb	(%a0)			|extract sign from caller's exop
143	bpls	unfl_signok
144	bset	#sign_bit,FP_SCR1(%a6)
145unfl_signok:
146	leal	FP_SCR1(%a6),%a0
147	orl	#unfinx_mask,USER_FPSR(%a6)
148|					;set UNFL, INEX2, AUNFL, AINEX
149unfl_con:
150	btstb	#unfl_bit,FPCR_ENABLE(%a6)
151	beqs	unfl_dis
152
153unfl_ena:
154	bfclr	STAG(%a6){#5:#3}		|clear wbtm66,wbtm1,wbtm0
155	bsetb	#wbtemp15_bit,WB_BYTE(%a6) |set wbtemp15
156	bsetb	#sticky_bit,STICKY(%a6)	|set sticky bit
157
158	bclrb	#E1,E_BYTE(%a6)
159
160unfl_dis:
161	bfextu	FPCR_MODE(%a6){#0:#2},%d0	|get round precision
162
163	bclrb	#sign_bit,LOCAL_EX(%a0)
164	sne	LOCAL_SGN(%a0)		|convert to internal ext format
165
166	bsr	unf_sub			|returns IEEE result at a0
167|					;and sets FPSR_CC accordingly
168
169	bfclr	LOCAL_SGN(%a0){#0:#8}	|convert back to IEEE ext format
170	beqs	unfl_fin
171
172	bsetb	#sign_bit,LOCAL_EX(%a0)
173	bsetb	#sign_bit,FP_SCR1(%a6)	|set sign bit of exc operand
174
175unfl_fin:
176	fmovemx (%a0),%fp0-%fp0		|store result in fp0
177	rts
178
179
180|
181|	t_ovfl2 --- OVFL exception (without inex2 returned)
182|
183| This entry is used by scale to force catastrophic overflow.  The
184| ovfl, aovfl, and ainex bits are set, but not the inex2 bit.
185|
186t_ovfl2:
187	orl	#ovfl_inx_mask,USER_FPSR(%a6)
188	movel	ETEMP(%a6),FP_SCR1(%a6)
189	movel	ETEMP_HI(%a6),FP_SCR1+4(%a6)
190	movel	ETEMP_LO(%a6),FP_SCR1+8(%a6)
191|
192| Check for single or double round precision.  If single, check if
193| the lower 40 bits of ETEMP are zero; if not, set inex2.  If double,
194| check if the lower 21 bits are zero; if not, set inex2.
195|
196	moveb	FPCR_MODE(%a6),%d0
197	andib	#0xc0,%d0
198	beq	t_work		|if extended, finish ovfl processing
199	cmpib	#0x40,%d0		|test for single
200	bnes	t_dbl
201t_sgl:
202	tstb	ETEMP_LO(%a6)
203	bnes	t_setinx2
204	movel	ETEMP_HI(%a6),%d0
205	andil	#0xff,%d0		|look at only lower 8 bits
206	bnes	t_setinx2
207	bra	t_work
208t_dbl:
209	movel	ETEMP_LO(%a6),%d0
210	andil	#0x7ff,%d0	|look at only lower 11 bits
211	beq	t_work
212t_setinx2:
213	orl	#inex2_mask,USER_FPSR(%a6)
214	bras	t_work
215|
216|	t_ovfl --- OVFL exception
217|
218|** Note: the exc operand is returned in ETEMP.
219|
220t_ovfl:
221	orl	#ovfinx_mask,USER_FPSR(%a6)
222t_work:
223	btstb	#ovfl_bit,FPCR_ENABLE(%a6) |test FPCR for ovfl enabled
224	beqs	ovf_dis
225
226ovf_ena:
227	clrl	FP_SCR1(%a6)		|set exceptional operand
228	clrl	FP_SCR1+4(%a6)
229	clrl	FP_SCR1+8(%a6)
230
231	bfclr	STAG(%a6){#5:#3}		|clear wbtm66,wbtm1,wbtm0
232	bclrb	#wbtemp15_bit,WB_BYTE(%a6) |clear wbtemp15
233	bsetb	#sticky_bit,STICKY(%a6)	|set sticky bit
234
235	bclrb	#E1,E_BYTE(%a6)
236|					;fall through to disabled case
237
238| For disabled overflow call 'ovf_r_k'.  This routine loads the
239| correct result based on the rounding precision, destination
240| format, rounding mode and sign.
241|
242ovf_dis:
243	bsr	ovf_r_k			|returns unsigned ETEMP_EX
244|					;and sets FPSR_CC accordingly.
245	bfclr	ETEMP_SGN(%a6){#0:#8}	|fix sign
246	beqs	ovf_pos
247	bsetb	#sign_bit,ETEMP_EX(%a6)
248	bsetb	#sign_bit,FP_SCR1(%a6)	|set exceptional operand sign
249ovf_pos:
250	fmovemx ETEMP(%a6),%fp0-%fp0		|move the result to fp0
251	rts
252
253
254|
255|	INEX2 exception
256|
257| The inex2 and ainex bits are set.
258|
259t_inx2:
260	orl	#inx2a_mask,USER_FPSR(%a6) |set INEX2, AINEX
261	rts
262
263|
264|	Force Inex2
265|
266| This routine is called by the transcendental routines to force
267| the inex2 exception bits set in the FPSR.  If the underflow bit
268| is set, but the underflow trap was not taken, the aunfl bit in
269| the FPSR must be set.
270|
271t_frcinx:
272	orl	#inx2a_mask,USER_FPSR(%a6) |set INEX2, AINEX
273	btstb	#unfl_bit,FPSR_EXCEPT(%a6) |test for unfl bit set
274	beqs	no_uacc1		|if clear, do not set aunfl
275	bsetb	#aunfl_bit,FPSR_AEXCEPT(%a6)
276no_uacc1:
277	rts
278
279|
280|	DST_NAN
281|
282| Determine if the destination nan is signalling or non-signalling,
283| and set the FPSR bits accordingly.  See the MC68040 User's Manual
284| section 3.2.2.5 NOT-A-NUMBERS.
285|
286dst_nan:
287	btstb	#sign_bit,FPTEMP_EX(%a6) |test sign of nan
288	beqs	dst_pos			|if clr, it was positive
289	bsetb	#neg_bit,FPSR_CC(%a6)	|set N bit
290dst_pos:
291	btstb	#signan_bit,FPTEMP_HI(%a6) |check if signalling
292	beqs	dst_snan		|branch if signalling
293
294	fmovel	%d1,%fpcr			|restore user's rmode/prec
295	fmovex FPTEMP(%a6),%fp0		|return the non-signalling nan
296|
297| Check the source nan.  If it is signalling, snan will be reported.
298|
299	moveb	STAG(%a6),%d0
300	andib	#0xe0,%d0
301	cmpib	#0x60,%d0
302	bnes	no_snan
303	btstb	#signan_bit,ETEMP_HI(%a6) |check if signalling
304	bnes	no_snan
305	orl	#snaniop_mask,USER_FPSR(%a6) |set NAN, SNAN, AIOP
306no_snan:
307	rts
308
309dst_snan:
310	btstb	#snan_bit,FPCR_ENABLE(%a6) |check if trap enabled
311	beqs	dst_dis			|branch if disabled
312
313	orb	#nan_tag,DTAG(%a6)	|set up dtag for nan
314	st	STORE_FLG(%a6)		|do not store a result
315	orl	#snaniop_mask,USER_FPSR(%a6) |set NAN, SNAN, AIOP
316	rts
317
318dst_dis:
319	bsetb	#signan_bit,FPTEMP_HI(%a6) |set SNAN bit in sop
320	fmovel	%d1,%fpcr			|restore user's rmode/prec
321	fmovex FPTEMP(%a6),%fp0		|load non-sign. nan
322	orl	#snaniop_mask,USER_FPSR(%a6) |set NAN, SNAN, AIOP
323	rts
324
325|
326|	SRC_NAN
327|
328| Determine if the source nan is signalling or non-signalling,
329| and set the FPSR bits accordingly.  See the MC68040 User's Manual
330| section 3.2.2.5 NOT-A-NUMBERS.
331|
332src_nan:
333	btstb	#sign_bit,ETEMP_EX(%a6) |test sign of nan
334	beqs	src_pos			|if clr, it was positive
335	bsetb	#neg_bit,FPSR_CC(%a6)	|set N bit
336src_pos:
337	btstb	#signan_bit,ETEMP_HI(%a6) |check if signalling
338	beqs	src_snan		|branch if signalling
339	fmovel	%d1,%fpcr			|restore user's rmode/prec
340	fmovex ETEMP(%a6),%fp0		|return the non-signalling nan
341	rts
342
343src_snan:
344	btstb	#snan_bit,FPCR_ENABLE(%a6) |check if trap enabled
345	beqs	src_dis			|branch if disabled
346	bsetb	#signan_bit,ETEMP_HI(%a6) |set SNAN bit in sop
347	orb	#norm_tag,DTAG(%a6)	|set up dtag for norm
348	orb	#nan_tag,STAG(%a6)	|set up stag for nan
349	st	STORE_FLG(%a6)		|do not store a result
350	orl	#snaniop_mask,USER_FPSR(%a6) |set NAN, SNAN, AIOP
351	rts
352
353src_dis:
354	bsetb	#signan_bit,ETEMP_HI(%a6) |set SNAN bit in sop
355	fmovel	%d1,%fpcr			|restore user's rmode/prec
356	fmovex ETEMP(%a6),%fp0		|load non-sign. nan
357	orl	#snaniop_mask,USER_FPSR(%a6) |set NAN, SNAN, AIOP
358	rts
359
360|
361| For all functions that have a denormalized input and that f(x)=x,
362| this is the entry point
363|
364t_extdnrm:
365	orl	#unfinx_mask,USER_FPSR(%a6)
366|					;set UNFL, INEX2, AUNFL, AINEX
367	bras	xdnrm_con
368|
369| Entry point for scale with extended denorm.  The function does
370| not set inex2, aunfl, or ainex.
371|
372t_resdnrm:
373	orl	#unfl_mask,USER_FPSR(%a6)
374
375xdnrm_con:
376	btstb	#unfl_bit,FPCR_ENABLE(%a6)
377	beqs	xdnrm_dis
378
379|
380| If exceptions are enabled, the additional task of setting up WBTEMP
381| is needed so that when the underflow exception handler is entered,
382| the user perceives no difference between what the 040 provides vs.
383| what the FPSP provides.
384|
385xdnrm_ena:
386	movel	%a0,-(%a7)
387
388	movel	LOCAL_EX(%a0),FP_SCR1(%a6)
389	movel	LOCAL_HI(%a0),FP_SCR1+4(%a6)
390	movel	LOCAL_LO(%a0),FP_SCR1+8(%a6)
391
392	lea	FP_SCR1(%a6),%a0
393
394	bclrb	#sign_bit,LOCAL_EX(%a0)
395	sne	LOCAL_SGN(%a0)		|convert to internal ext format
396	tstw	LOCAL_EX(%a0)		|check if input is denorm
397	beqs	xdnrm_dn		|if so, skip nrm_set
398	bsr	nrm_set			|normalize the result (exponent
399|					;will be negative
400xdnrm_dn:
401	bclrb	#sign_bit,LOCAL_EX(%a0)	|take off false sign
402	bfclr	LOCAL_SGN(%a0){#0:#8}	|change back to IEEE ext format
403	beqs	xdep
404	bsetb	#sign_bit,LOCAL_EX(%a0)
405xdep:
406	bfclr	STAG(%a6){#5:#3}		|clear wbtm66,wbtm1,wbtm0
407	bsetb	#wbtemp15_bit,WB_BYTE(%a6) |set wbtemp15
408	bclrb	#sticky_bit,STICKY(%a6)	|clear sticky bit
409	bclrb	#E1,E_BYTE(%a6)
410	movel	(%a7)+,%a0
411xdnrm_dis:
412	bfextu	FPCR_MODE(%a6){#0:#2},%d0	|get round precision
413	bnes	not_ext			|if not round extended, store
414|					;IEEE defaults
415is_ext:
416	btstb	#sign_bit,LOCAL_EX(%a0)
417	beqs	xdnrm_store
418
419	bsetb	#neg_bit,FPSR_CC(%a6)	|set N bit in FPSR_CC
420
421	bras	xdnrm_store
422
423not_ext:
424	bclrb	#sign_bit,LOCAL_EX(%a0)
425	sne	LOCAL_SGN(%a0)		|convert to internal ext format
426	bsr	unf_sub			|returns IEEE result pointed by
427|					;a0; sets FPSR_CC accordingly
428	bfclr	LOCAL_SGN(%a0){#0:#8}	|convert back to IEEE ext format
429	beqs	xdnrm_store
430	bsetb	#sign_bit,LOCAL_EX(%a0)
431xdnrm_store:
432	fmovemx (%a0),%fp0-%fp0		|store result in fp0
433	rts
434
435|
436| This subroutine is used for dyadic operations that use an extended
437| denorm within the kernel. The approach used is to capture the frame,
438| fix/restore.
439|
440	.global	t_avoid_unsupp
441t_avoid_unsupp:
442	link	%a2,#-LOCAL_SIZE		|so that a2 fpsp.h negative
443|					;offsets may be used
444	fsave	-(%a7)
445	tstb	1(%a7)			|check if idle, exit if so
446	beq	idle_end
447	btstb	#E1,E_BYTE(%a2)		|check for an E1 exception if
448|					;enabled, there is an unsupp
449	beq	end_avun		|else, exit
450	btstb	#7,DTAG(%a2)		|check for denorm destination
451	beqs	src_den			|else, must be a source denorm
452|
453| handle destination denorm
454|
455	lea	FPTEMP(%a2),%a0
456	btstb	#sign_bit,LOCAL_EX(%a0)
457	sne	LOCAL_SGN(%a0)		|convert to internal ext format
458	bclrb	#7,DTAG(%a2)		|set DTAG to norm
459	bsr	nrm_set			|normalize result, exponent
460|					;will become negative
461	bclrb	#sign_bit,LOCAL_EX(%a0)	|get rid of fake sign
462	bfclr	LOCAL_SGN(%a0){#0:#8}	|convert back to IEEE ext format
463	beqs	ck_src_den		|check if source is also denorm
464	bsetb	#sign_bit,LOCAL_EX(%a0)
465ck_src_den:
466	btstb	#7,STAG(%a2)
467	beqs	end_avun
468src_den:
469	lea	ETEMP(%a2),%a0
470	btstb	#sign_bit,LOCAL_EX(%a0)
471	sne	LOCAL_SGN(%a0)		|convert to internal ext format
472	bclrb	#7,STAG(%a2)		|set STAG to norm
473	bsr	nrm_set			|normalize result, exponent
474|					;will become negative
475	bclrb	#sign_bit,LOCAL_EX(%a0)	|get rid of fake sign
476	bfclr	LOCAL_SGN(%a0){#0:#8}	|convert back to IEEE ext format
477	beqs	den_com
478	bsetb	#sign_bit,LOCAL_EX(%a0)
479den_com:
480	moveb	#0xfe,CU_SAVEPC(%a2)	|set continue frame
481	clrw	NMNEXC(%a2)		|clear NMNEXC
482	bclrb	#E1,E_BYTE(%a2)
483|	fmove.l	%FPSR,FPSR_SHADOW(%a2)
484|	bset.b	#SFLAG,E_BYTE(%a2)
485|	bset.b	#XFLAG,T_BYTE(%a2)
486end_avun:
487	frestore (%a7)+
488	unlk	%a2
489	rts
490idle_end:
491	addl	#4,%a7
492	unlk	%a2
493	rts
494	|end
495