xref: /openbmc/linux/arch/x86/math-emu/reg_u_sub.S (revision 4f3db074)
1	.file	"reg_u_sub.S"
2/*---------------------------------------------------------------------------+
3 |  reg_u_sub.S                                                              |
4 |                                                                           |
5 | Core floating point subtraction routine.                                  |
6 |                                                                           |
7 | Copyright (C) 1992,1993,1995,1997                                         |
8 |                  W. Metzenthen, 22 Parker St, Ormond, Vic 3163, Australia |
9 |                  E-mail   billm@suburbia.net                              |
10 |                                                                           |
11 | Call from C as:                                                           |
12 |    int FPU_u_sub(FPU_REG *arg1, FPU_REG *arg2, FPU_REG *answ,             |
13 |                                                int control_w)             |
14 |    Return value is the tag of the answer, or-ed with FPU_Exception if     |
15 |    one was raised, or -1 on internal error.                               |
16 |                                                                           |
17 +---------------------------------------------------------------------------*/
18
19/*
20 |    Kernel subtraction routine FPU_u_sub(reg *arg1, reg *arg2, reg *answ).
21 |    Takes two valid reg f.p. numbers (TAG_Valid), which are
22 |    treated as unsigned numbers,
23 |    and returns their difference as a TAG_Valid or TAG_Zero f.p.
24 |    number.
25 |    The first number (arg1) must be the larger.
26 |    The returned number is normalized.
27 |    Basic checks are performed if PARANOID is defined.
28 */
29
30#include "exception.h"
31#include "fpu_emu.h"
32#include "control_w.h"
33
34.text
35ENTRY(FPU_u_sub)
36	pushl	%ebp
37	movl	%esp,%ebp
38	pushl	%esi
39	pushl	%edi
40	pushl	%ebx
41
42	movl	PARAM1,%esi	/* source 1 */
43	movl	PARAM2,%edi	/* source 2 */
44
45	movl	PARAM6,%ecx
46	subl	PARAM7,%ecx	/* exp1 - exp2 */
47
48#ifdef PARANOID
49	/* source 2 is always smaller than source 1 */
50	js	L_bugged_1
51
52	testl	$0x80000000,SIGH(%edi)	/* The args are assumed to be be normalized */
53	je	L_bugged_2
54
55	testl	$0x80000000,SIGH(%esi)
56	je	L_bugged_2
57#endif /* PARANOID */
58
59/*--------------------------------------+
60 |	Form a register holding the     |
61 |	smaller number                  |
62 +--------------------------------------*/
63	movl	SIGH(%edi),%eax	/* register ms word */
64	movl	SIGL(%edi),%ebx	/* register ls word */
65
66	movl	PARAM3,%edi	/* destination */
67	movl	PARAM6,%edx
68	movw	%dx,EXP(%edi)	/* Copy exponent to destination */
69
70	xorl	%edx,%edx	/* register extension */
71
72/*--------------------------------------+
73 |	Shift the temporary register	|
74 |      right the required number of	|
75 |	places.				|
76 +--------------------------------------*/
77
78	cmpw	$32,%cx		/* shrd only works for 0..31 bits */
79	jnc	L_more_than_31
80
81/* less than 32 bits */
82	shrd	%cl,%ebx,%edx
83	shrd	%cl,%eax,%ebx
84	shr	%cl,%eax
85	jmp	L_shift_done
86
87L_more_than_31:
88	cmpw	$64,%cx
89	jnc	L_more_than_63
90
91	subb	$32,%cl
92	jz	L_exactly_32
93
94	shrd	%cl,%eax,%edx
95	shr	%cl,%eax
96	orl	%ebx,%ebx
97	jz	L_more_31_no_low	/* none of the lowest bits is set */
98
99	orl	$1,%edx			/* record the fact in the extension */
100
101L_more_31_no_low:
102	movl	%eax,%ebx
103	xorl	%eax,%eax
104	jmp	L_shift_done
105
106L_exactly_32:
107	movl	%ebx,%edx
108	movl	%eax,%ebx
109	xorl	%eax,%eax
110	jmp	L_shift_done
111
112L_more_than_63:
113	cmpw	$65,%cx
114	jnc	L_more_than_64
115
116	/* Shift right by 64 bits */
117	movl	%eax,%edx
118	orl	%ebx,%ebx
119	jz	L_more_63_no_low
120
121	orl	$1,%edx
122	jmp	L_more_63_no_low
123
124L_more_than_64:
125	jne	L_more_than_65
126
127	/* Shift right by 65 bits */
128	/* Carry is clear if we get here */
129	movl	%eax,%edx
130	rcrl	%edx
131	jnc	L_shift_65_nc
132
133	orl	$1,%edx
134	jmp	L_more_63_no_low
135
136L_shift_65_nc:
137	orl	%ebx,%ebx
138	jz	L_more_63_no_low
139
140	orl	$1,%edx
141	jmp	L_more_63_no_low
142
143L_more_than_65:
144	movl	$1,%edx		/* The shifted nr always at least one '1' */
145
146L_more_63_no_low:
147	xorl	%ebx,%ebx
148	xorl	%eax,%eax
149
150L_shift_done:
151L_subtr:
152/*------------------------------+
153 |	Do the subtraction	|
154 +------------------------------*/
155	xorl	%ecx,%ecx
156	subl	%edx,%ecx
157	movl	%ecx,%edx
158	movl	SIGL(%esi),%ecx
159	sbbl	%ebx,%ecx
160	movl	%ecx,%ebx
161	movl	SIGH(%esi),%ecx
162	sbbl	%eax,%ecx
163	movl	%ecx,%eax
164
165#ifdef PARANOID
166	/* We can never get a borrow */
167	jc	L_bugged
168#endif /* PARANOID */
169
170/*--------------------------------------+
171 |	Normalize the result		|
172 +--------------------------------------*/
173	testl	$0x80000000,%eax
174	jnz	L_round		/* no shifting needed */
175
176	orl	%eax,%eax
177	jnz	L_shift_1	/* shift left 1 - 31 bits */
178
179	orl	%ebx,%ebx
180	jnz	L_shift_32	/* shift left 32 - 63 bits */
181
182/*
183 *	 A rare case, the only one which is non-zero if we got here
184 *         is:           1000000 .... 0000
185 *                      -0111111 .... 1111 1
186 *                       --------------------
187 *                       0000000 .... 0000 1
188 */
189
190	cmpl	$0x80000000,%edx
191	jnz	L_must_be_zero
192
193	/* Shift left 64 bits */
194	subw	$64,EXP(%edi)
195	xchg	%edx,%eax
196	jmp	fpu_reg_round
197
198L_must_be_zero:
199#ifdef PARANOID
200	orl	%edx,%edx
201	jnz	L_bugged_3
202#endif /* PARANOID */
203
204	/* The result is zero */
205	movw	$0,EXP(%edi)		/* exponent */
206	movl	$0,SIGL(%edi)
207	movl	$0,SIGH(%edi)
208	movl	TAG_Zero,%eax
209	jmp	L_exit
210
211L_shift_32:
212	movl	%ebx,%eax
213	movl	%edx,%ebx
214	movl	$0,%edx
215	subw	$32,EXP(%edi)	/* Can get underflow here */
216
217/* We need to shift left by 1 - 31 bits */
218L_shift_1:
219	bsrl	%eax,%ecx	/* get the required shift in %ecx */
220	subl	$31,%ecx
221	negl	%ecx
222	shld	%cl,%ebx,%eax
223	shld	%cl,%edx,%ebx
224	shl	%cl,%edx
225	subw	%cx,EXP(%edi)	/* Can get underflow here */
226
227L_round:
228	jmp	fpu_reg_round	/* Round the result */
229
230
231#ifdef PARANOID
232L_bugged_1:
233	pushl	EX_INTERNAL|0x206
234	call	EXCEPTION
235	pop	%ebx
236	jmp	L_error_exit
237
238L_bugged_2:
239	pushl	EX_INTERNAL|0x209
240	call	EXCEPTION
241	pop	%ebx
242	jmp	L_error_exit
243
244L_bugged_3:
245	pushl	EX_INTERNAL|0x210
246	call	EXCEPTION
247	pop	%ebx
248	jmp	L_error_exit
249
250L_bugged_4:
251	pushl	EX_INTERNAL|0x211
252	call	EXCEPTION
253	pop	%ebx
254	jmp	L_error_exit
255
256L_bugged:
257	pushl	EX_INTERNAL|0x212
258	call	EXCEPTION
259	pop	%ebx
260	jmp	L_error_exit
261
262L_error_exit:
263	movl	$-1,%eax
264
265#endif /* PARANOID */
266
267L_exit:
268	popl	%ebx
269	popl	%edi
270	popl	%esi
271	leave
272	ret
273