1dnl  IA-64 mpn_cnd_add_n/mpn_cnd_sub_n.
2
3dnl  Contributed to the GNU project by Torbj��rn Granlund.
4
5dnl  Copyright 2013 Free Software Foundation, Inc.
6
7dnl  This file is part of the GNU MP Library.
8dnl
9dnl  The GNU MP Library is free software; you can redistribute it and/or modify
10dnl  it under the terms of either:
11dnl
12dnl    * the GNU Lesser General Public License as published by the Free
13dnl      Software Foundation; either version 3 of the License, or (at your
14dnl      option) any later version.
15dnl
16dnl  or
17dnl
18dnl    * the GNU General Public License as published by the Free Software
19dnl      Foundation; either version 2 of the License, or (at your option) any
20dnl      later version.
21dnl
22dnl  or both in parallel, as here.
23dnl
24dnl  The GNU MP Library is distributed in the hope that it will be useful, but
25dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
26dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
27dnl  for more details.
28dnl
29dnl  You should have received copies of the GNU General Public License and the
30dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
31dnl  see https://www.gnu.org/licenses/.
32
33include(`../config.m4')
34
35C           cycles/limb
36C Itanium:      ?
37C Itanium 2:    1.5
38
39C INPUT PARAMETERS
40define(`cnd', `r32')
41define(`rp',  `r33')
42define(`up',  `r34')
43define(`vp',  `r35')
44define(`n',   `r36')
45
46ifdef(`OPERATION_cnd_add_n',`
47  define(ADDSUB,	add)
48  define(CND,		ltu)
49  define(INCR,		1)
50  define(LIM,		-1)
51  define(func,    mpn_cnd_add_n)
52')
53ifdef(`OPERATION_cnd_sub_n',`
54  define(ADDSUB,	sub)
55  define(CND,		gtu)
56  define(INCR,		-1)
57  define(LIM,		0)
58  define(func,    mpn_cnd_sub_n)
59')
60
61define(PFDIST, 160)
62
63C Some useful aliases for registers we use
64define(`u0',`r14') define(`u1',`r15') define(`u2',`r16') define(`u3',`r17')
65define(`x0',`r20') define(`x1',`r21') define(`x2',`r22') define(`x3',`r23')
66define(`v0',`r24') define(`v1',`r25') define(`v2',`r26') define(`v3',`r27')
67define(`w0',`r28') define(`w1',`r29') define(`w2',`r30') define(`w3',`r31')
68define(`up1',`up') define(`up2',`r8') define(`upadv',`r1')
69define(`vp1',`vp') define(`vp2',`r9') define(`vpadv',`r11')
70define(`rp1',`rp') define(`rp2',`r10')
71
72MULFUNC_PROLOGUE(mpn_cnd_add_n mpn_cnd_sub_n)
73
74ASM_START()
75PROLOGUE(func)
76	.prologue
77	.save	ar.lc, r2
78	.body
79ifdef(`HAVE_ABI_32',`
80	addp4	rp = 0, rp		C				M I
81	addp4	up = 0, up		C				M I
82	nop.i	0
83	addp4	vp = 0, vp		C				M I
84	nop.m	0
85	zxt4	n = n			C				I
86	;;
87')
88 {.mmi;	and	r3 = 3, n		C				M I
89	add	n = -1, n		C				M I
90	mov	r2 = ar.lc		C				I0
91}{.mmi;	cmp.ne	p6, p7 = 0, cnd		C				M I
92	add	vp2 = 8, vp		C				M I
93	add	up2 = 8, up		C				M I
94	;;
95}{.mmi;	add	upadv = PFDIST, up	C				M I
96	add	vpadv = PFDIST, vp	C				M I
97	shr.u	n = n, 2		C				I0
98	.pred.rel "mutex", p6, p7
99}{.mmi;	add	rp2 = 8, rp		C				M I
100   (p6)	mov	cnd = -1		C				M I
101   (p7)	mov	cnd = 0			C				M I
102	;;
103}	cmp.eq	p9, p0 = 1, r3		C				M I
104	cmp.eq	p7, p0 = 2, r3		C				M I
105	cmp.eq	p8, p0 = 3, r3		C				M I
106   (p9)	br	L(b1)			C				B
107   (p7)	br	L(b2)			C				B
108   (p8)	br	L(b3)			C				B
109	;;
110L(b0):
111 {.mmi;	ld8	v2 = [vp1], 16		C				M01
112	ld8	v3 = [vp2], 16		C				M01
113	mov	ar.lc = n		C				I0
114	;;
115}	ld8	u2 = [up1], 16		C				M01
116	ld8	u3 = [up2], 16		C				M01
117	and	x2 = v2, cnd		C				M I
118	and	x3 = v3, cnd		C				M I
119	;;
120	ADDSUB	w2 = u2, x2		C				M I
121	ADDSUB	w3 = u3, x3		C				M I
122	;;
123	ld8	v0 = [vp1], 16		C				M01
124	ld8	v1 = [vp2], 16		C				M01
125	cmp.CND	p8, p0 = w2, u2		C				M I
126	cmp.CND	p9, p0 = w3, u3		C				M I
127	br	L(lo0)
128
129L(b1):	ld8	v1 = [vp1], 8		C				M01
130	add	vp2 = 8, vp2		C				M I
131	add	rp2 = 8, rp2		C				M I
132	;;
133	ld8	u1 = [up1], 8		C				M01
134	add	up2 = 8, up2		C				M I
135	and	x1 = v1, cnd		C				M I
136	;;
137	ADDSUB	w1 = u1, x1		C				M I
138	cmp.ne	p10, p0 = 0, n
139	add	n = -1, n
140	;;
141	cmp.CND	p7, p0 = w1, u1		C				M I
142	st8	[rp1] = w1, 8		C				M23
143  (p10)	br	L(b0)
144	;;
145	mov	r8 = 0			C				M I
146	br	L(e1)
147
148L(b3):	ld8	v3 = [vp1], 8		C				M01
149	add	vp2 = 8, vp2		C				M I
150	add	rp2 = 8, rp2		C				M I
151	;;
152	ld8	u3 = [up1], 8		C				M01
153	add	up2 = 8, up2		C				M I
154	and	x3 = v3, cnd		C				M I
155	;;
156	ADDSUB	w3 = u3, x3		C				M I
157	;;
158	cmp.CND	p9, p0 = w3, u3		C				M I
159	st8	[rp1] = w3, 8		C				M23
160	C fall through
161
162L(b2):
163 {.mmi;	ld8	v0 = [vp1], 16		C				M01
164	ld8	v1 = [vp2], 16		C				M01
165	mov	ar.lc = n		C				I0
166	;;
167}	ld8	u0 = [up1], 16		C				M01
168	ld8	u1 = [up2], 16		C				M01
169	and	x0 = v0, cnd		C				M I
170	and	x1 = v1, cnd		C				M I
171	;;
172	ADDSUB	w0 = u0, x0		C				M I
173	ADDSUB	w1 = u1, x1		C				M I
174	br.cloop.dptk	L(gt2)		C				B
175	;;
176	cmp.CND	p6, p0 = w0, u0		C				M I
177	br		L(e2)		C				B
178L(gt2):
179	ld8	v2 = [vp1], 16		C				M01
180	ld8	v3 = [vp2], 16		C				M01
181	cmp.CND	p6, p0 = w0, u0		C				M I
182	cmp.CND	p7, p0 = w1, u1		C				M I
183	br		L(lo2)		C				B
184
185
186C *** MAIN LOOP START ***
187C	ALIGN(32)
188L(top):
189 {.mmi;	ld8	v2 = [vp1], 16		C				M01
190	ld8	v3 = [vp2], 16		C				M01
191	cmp.CND	p6, p0 = w0, u0		C				M I
192}{.mmi;	st8	[rp1] = w2, 16		C				M23
193	st8	[rp2] = w3, 16		C				M23
194	cmp.CND	p7, p0 = w1, u1		C				M I
195	;;
196}
197L(lo2):
198 {.mmi;	ld8	u2 = [up1], 16		C				M01
199	ld8	u3 = [up2], 16		C				M01
200   (p9)	cmpeqor	p6, p0 = LIM, w0	C				M I
201}{.mmi;	and	x2 = v2, cnd		C				M I
202	and	x3 = v3, cnd		C				M I
203   (p9)	add	w0 = INCR, w0		C				M I
204	;;
205}{.mmi;	ADDSUB	w2 = u2, x2		C				M I
206   (p6)	cmpeqor	p7, p0 = LIM, w1	C				M I
207   (p6)	add	w1 = INCR, w1		C				M I
208}{.mmi;	ADDSUB	w3 = u3, x3		C				M I
209	lfetch	[upadv], 32
210	nop	0
211	;;
212}{.mmi;	ld8	v0 = [vp1], 16		C				M01
213	ld8	v1 = [vp2], 16		C				M01
214	cmp.CND	p8, p0 = w2, u2		C				M I
215}{.mmi;	st8	[rp1] = w0, 16		C				M23
216	st8	[rp2] = w1, 16		C				M23
217	cmp.CND	p9, p0 = w3, u3		C				M I
218	;;
219}
220L(lo0):
221 {.mmi;	ld8	u0 = [up1], 16		C				M01
222	ld8	u1 = [up2], 16		C				M01
223   (p7)	cmpeqor	p8, p0 = LIM, w2	C				M I
224}{.mmi;	and	x0 = v0, cnd		C				M I
225	and	x1 = v1, cnd		C				M I
226   (p7)	add	w2 = INCR, w2		C				M I
227	;;
228}{.mmi;	ADDSUB	w0 = u0, x0		C				M I
229   (p8)	cmpeqor	p9, p0 = LIM, w3	C				M I
230   (p8)	add	w3 = INCR, w3		C				M I
231}{.mmb;	ADDSUB	w1 = u1, x1		C				M I
232	lfetch	[vpadv], 32
233	br.cloop.dptk	L(top)		C				B
234	;;
235}
236C *** MAIN LOOP END ***
237
238
239L(end):
240 {.mmi;	st8	[rp1] = w2, 16		C				M23
241	st8	[rp2] = w3, 16		C				M23
242	cmp.CND	p6, p0 = w0, u0		C				M I
243	;;
244}
245L(e2):
246 {.mmi;	cmp.CND	p7, p0 = w1, u1		C				M I
247   (p9)	cmpeqor	p6, p0 = LIM, w0	C				M I
248   (p9)	add	w0 = INCR, w0		C				M I
249	;;
250}{.mmi;	mov	r8 = 0			C				M I
251   (p6)	cmpeqor	p7, p0 = LIM, w1	C				M I
252   (p6)	add	w1 = INCR, w1		C				M I
253	;;
254}{.mmi;	st8	[rp1] = w0, 16		C				M23
255	st8	[rp2] = w1, 16		C				M23
256	mov	ar.lc = r2		C				I0
257}
258L(e1):
259 {.mmb;	nop	0
260   (p7)	mov	r8 = 1			C				M I
261	br.ret.sptk.many b0		C				B
262}
263EPILOGUE()
264ASM_END()
265