xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/power/mul_1.asm (revision 7d62b00eb9ad855ffcd7da46b41e23feb5476fac)
1dnl  IBM POWER mpn_mul_1 -- Multiply a limb vector with a limb and store the
2dnl  result in 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_mul_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	ai	0,0,0		C reset carry
62	cax	9,9,7
63	blt	Lneg
64Lpos:	bdz	Lend
65Lploop:	lu	0,4(4)
66	stu	8,4(3)
67	cmpi	0,0,0
68	mul	10,0,6
69	mfmq	0
70	ae	8,0,9
71	bge	Lp0
72	cax	10,10,6		C adjust high limb for negative limb from s1
73Lp0:	bdz	Lend0
74	lu	0,4(4)
75	stu	8,4(3)
76	cmpi	0,0,0
77	mul	9,0,6
78	mfmq	0
79	ae	8,0,10
80	bge	Lp1
81	cax	9,9,6		C adjust high limb for negative limb from s1
82Lp1:	bdn	Lploop
83	b	Lend
84
85Lneg:	cax	9,9,0
86	bdz	Lend
87Lnloop:	lu	0,4(4)
88	stu	8,4(3)
89	cmpi	0,0,0
90	mul	10,0,6
91	cax	10,10,0		C adjust high limb for negative s2_limb
92	mfmq	0
93	ae	8,0,9
94	bge	Ln0
95	cax	10,10,6		C adjust high limb for negative limb from s1
96Ln0:	bdz	Lend0
97	lu	0,4(4)
98	stu	8,4(3)
99	cmpi	0,0,0
100	mul	9,0,6
101	cax	9,9,0		C adjust high limb for negative s2_limb
102	mfmq	0
103	ae	8,0,10
104	bge	Ln1
105	cax	9,9,6		C adjust high limb for negative limb from s1
106Ln1:	bdn	Lnloop
107	b	Lend
108
109Lend0:	cal	9,0(10)
110Lend:	st	8,4(3)
111	aze	3,9
112	br
113EPILOGUE(mpn_mul_1)
114