1dnl  IBM POWER mpn_addmul_1 -- Multiply a limb vector with a limb and add the
2dnl  result to a second limb vector.
3
4dnl  Copyright 1992, 1994, 1999-2001 Free Software Foundation, Inc.
5
6dnl  This file is part of the GNU MP Library.
7dnl
8dnl  The GNU MP Library is free software; you can redistribute it and/or modify
9dnl  it under the terms of either:
10dnl
11dnl    * the GNU Lesser General Public License as published by the Free
12dnl      Software Foundation; either version 3 of the License, or (at your
13dnl      option) any later version.
14dnl
15dnl  or
16dnl
17dnl    * the GNU General Public License as published by the Free Software
18dnl      Foundation; either version 2 of the License, or (at your option) any
19dnl      later version.
20dnl
21dnl  or both in parallel, as here.
22dnl
23dnl  The GNU MP Library is distributed in the hope that it will be useful, but
24dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
25dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
26dnl  for more details.
27dnl
28dnl  You should have received copies of the GNU General Public License and the
29dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
30dnl  see https://www.gnu.org/licenses/.
31
32
33dnl  INPUT PARAMETERS
34dnl  res_ptr	r3
35dnl  s1_ptr	r4
36dnl  size	r5
37dnl  s2_limb	r6
38
39dnl  The POWER architecture has no unsigned 32x32->64 bit multiplication
40dnl  instruction.  To obtain that operation, we have to use the 32x32->64
41dnl  signed multiplication instruction, and add the appropriate compensation to
42dnl  the high limb of the result.  We add the multiplicand if the multiplier
43dnl  has its most significant bit set, and we add the multiplier if the
44dnl  multiplicand has its most significant bit set.  We need to preserve the
45dnl  carry flag between each iteration, so we have to compute the compensation
46dnl  carefully (the natural, srai+and doesn't work).  Since all POWER can
47dnl  branch in zero cycles, we use conditional branches for the compensation.
48
49include(`../config.m4')
50
51ASM_START()
52PROLOGUE(mpn_addmul_1)
53	cal	3,-4(3)
54	l	0,0(4)
55	cmpi	0,6,0
56	mtctr	5
57	mul	9,0,6
58	srai	7,0,31
59	and	7,7,6
60	mfmq	8
61	cax	9,9,7
62	l	7,4(3)
63	a	8,8,7		C add res_limb
64	blt	Lneg
65Lpos:	bdz	Lend
66
67Lploop:	lu	0,4(4)
68	stu	8,4(3)
69	cmpi	0,0,0
70	mul	10,0,6
71	mfmq	0
72	ae	8,0,9		C low limb + old_cy_limb + old cy
73	l	7,4(3)
74	aze	10,10		C propagate cy to new cy_limb
75	a	8,8,7		C add res_limb
76	bge	Lp0
77	cax	10,10,6		C adjust high limb for negative limb from s1
78Lp0:	bdz	Lend0
79	lu	0,4(4)
80	stu	8,4(3)
81	cmpi	0,0,0
82	mul	9,0,6
83	mfmq	0
84	ae	8,0,10
85	l	7,4(3)
86	aze	9,9
87	a	8,8,7
88	bge	Lp1
89	cax	9,9,6		C adjust high limb for negative limb from s1
90Lp1:	bdn	Lploop
91
92	b	Lend
93
94Lneg:	cax	9,9,0
95	bdz	Lend
96Lnloop:	lu	0,4(4)
97	stu	8,4(3)
98	cmpi	0,0,0
99	mul	10,0,6
100	mfmq	7
101	ae	8,7,9
102	l	7,4(3)
103	ae	10,10,0		C propagate cy to new cy_limb
104	a	8,8,7		C add res_limb
105	bge	Ln0
106	cax	10,10,6		C adjust high limb for negative limb from s1
107Ln0:	bdz	Lend0
108	lu	0,4(4)
109	stu	8,4(3)
110	cmpi	0,0,0
111	mul	9,0,6
112	mfmq	7
113	ae	8,7,10
114	l	7,4(3)
115	ae	9,9,0		C propagate cy to new cy_limb
116	a	8,8,7		C add res_limb
117	bge	Ln1
118	cax	9,9,6		C adjust high limb for negative limb from s1
119Ln1:	bdn	Lnloop
120	b	Lend
121
122Lend0:	cal	9,0(10)
123Lend:	st	8,4(3)
124	aze	3,9
125	br
126EPILOGUE(mpn_addmul_1)
127