1dnl  IBM POWER mpn_submul_1 -- Multiply a limb vector with a limb and subtract
2dnl  the result from 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_submul_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	11
61	cax	9,9,7
62	l	7,4(3)
63	sf	8,11,7		C add res_limb
64	a	11,8,11		C invert cy (r11 is junk)
65	blt	Lneg
66Lpos:	bdz	Lend
67
68Lploop:	lu	0,4(4)
69	stu	8,4(3)
70	cmpi	0,0,0
71	mul	10,0,6
72	mfmq	0
73	ae	11,0,9		C low limb + old_cy_limb + old cy
74	l	7,4(3)
75	aze	10,10		C propagate cy to new cy_limb
76	sf	8,11,7		C add res_limb
77	a	11,8,11		C invert cy (r11 is junk)
78	bge	Lp0
79	cax	10,10,6		C adjust high limb for negative limb from s1
80Lp0:	bdz	Lend0
81	lu	0,4(4)
82	stu	8,4(3)
83	cmpi	0,0,0
84	mul	9,0,6
85	mfmq	0
86	ae	11,0,10
87	l	7,4(3)
88	aze	9,9
89	sf	8,11,7
90	a	11,8,11		C invert cy (r11 is junk)
91	bge	Lp1
92	cax	9,9,6		C adjust high limb for negative limb from s1
93Lp1:	bdn	Lploop
94
95	b	Lend
96
97Lneg:	cax	9,9,0
98	bdz	Lend
99Lnloop:	lu	0,4(4)
100	stu	8,4(3)
101	cmpi	0,0,0
102	mul	10,0,6
103	mfmq	7
104	ae	11,7,9
105	l	7,4(3)
106	ae	10,10,0		C propagate cy to new cy_limb
107	sf	8,11,7		C add res_limb
108	a	11,8,11		C invert cy (r11 is junk)
109	bge	Ln0
110	cax	10,10,6		C adjust high limb for negative limb from s1
111Ln0:	bdz	Lend0
112	lu	0,4(4)
113	stu	8,4(3)
114	cmpi	0,0,0
115	mul	9,0,6
116	mfmq	7
117	ae	11,7,10
118	l	7,4(3)
119	ae	9,9,0		C propagate cy to new cy_limb
120	sf	8,11,7		C add res_limb
121	a	11,8,11		C invert cy (r11 is junk)
122	bge	Ln1
123	cax	9,9,6		C adjust high limb for negative limb from s1
124Ln1:	bdn	Lnloop
125	b	Lend
126
127Lend0:	cal	9,0(10)
128Lend:	st	8,4(3)
129	aze	3,9
130	br
131EPILOGUE(mpn_submul_1)
132