xref: /OK3568_Linux_fs/kernel/arch/c6x/lib/remu.S (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1*4882a593Smuzhiyun;; SPDX-License-Identifier: GPL-2.0-or-later
2*4882a593Smuzhiyun;;  Copyright 2010  Free Software Foundation, Inc.
3*4882a593Smuzhiyun;;  Contributed by Bernd Schmidt <bernds@codesourcery.com>.
4*4882a593Smuzhiyun;;
5*4882a593Smuzhiyun
6*4882a593Smuzhiyun#include <linux/linkage.h>
7*4882a593Smuzhiyun
8*4882a593Smuzhiyun	;; ABI considerations for the divide functions
9*4882a593Smuzhiyun	;; The following registers are call-used:
10*4882a593Smuzhiyun	;; __c6xabi_divi A0,A1,A2,A4,A6,B0,B1,B2,B4,B5
11*4882a593Smuzhiyun	;; __c6xabi_divu A0,A1,A2,A4,A6,B0,B1,B2,B4
12*4882a593Smuzhiyun	;; __c6xabi_remi A1,A2,A4,A5,A6,B0,B1,B2,B4
13*4882a593Smuzhiyun	;; __c6xabi_remu A1,A4,A5,A7,B0,B1,B2,B4
14*4882a593Smuzhiyun	;;
15*4882a593Smuzhiyun	;; In our implementation, divu and remu are leaf functions,
16*4882a593Smuzhiyun	;; while both divi and remi call into divu.
17*4882a593Smuzhiyun	;; A0 is not clobbered by any of the functions.
18*4882a593Smuzhiyun	;; divu does not clobber B2 either, which is taken advantage of
19*4882a593Smuzhiyun	;; in remi.
20*4882a593Smuzhiyun	;; divi uses B5 to hold the original return address during
21*4882a593Smuzhiyun	;; the call to divu.
22*4882a593Smuzhiyun	;; remi uses B2 and A5 to hold the input values during the
23*4882a593Smuzhiyun	;; call to divu.  It stores B3 in on the stack.
24*4882a593Smuzhiyun
25*4882a593Smuzhiyun
26*4882a593Smuzhiyun	.text
27*4882a593Smuzhiyun
28*4882a593SmuzhiyunENTRY(__c6xabi_remu)
29*4882a593Smuzhiyun	;; The ABI seems designed to prevent these functions calling each other,
30*4882a593Smuzhiyun	;; so we duplicate most of the divsi3 code here.
31*4882a593Smuzhiyun	 mv	.s2x	A4, B1
32*4882a593Smuzhiyun	 lmbd	.l2	1, B4, B1
33*4882a593Smuzhiyun|| [!B1] b	.s2	B3	; RETURN A
34*4882a593Smuzhiyun|| [!B1] mvk	.d2	1, B4
35*4882a593Smuzhiyun
36*4882a593Smuzhiyun	 mv	.l1x	B1, A7
37*4882a593Smuzhiyun||	 shl	.s2	B4, B1, B4
38*4882a593Smuzhiyun
39*4882a593Smuzhiyun	 cmpltu	.l1x	A4, B4, A1
40*4882a593Smuzhiyun   [!A1] sub	.l1x	A4, B4, A4
41*4882a593Smuzhiyun	 shru	.s2	B4, 1, B4
42*4882a593Smuzhiyun
43*4882a593Smuzhiyun_remu_loop:
44*4882a593Smuzhiyun	 cmpgt	.l2	B1, 7, B0
45*4882a593Smuzhiyun|| [B1]	 subc	.l1x	A4,B4,A4
46*4882a593Smuzhiyun|| [B1]	 add	.s2	-1, B1, B1
47*4882a593Smuzhiyun	;; RETURN A may happen here (note: must happen before the next branch)
48*4882a593Smuzhiyun   [B1]	 subc	.l1x	A4,B4,A4
49*4882a593Smuzhiyun|| [B1]	 add	.s2	-1, B1, B1
50*4882a593Smuzhiyun|| [B0]	 b	.s1	_remu_loop
51*4882a593Smuzhiyun   [B1]	 subc	.l1x	A4,B4,A4
52*4882a593Smuzhiyun|| [B1]	 add	.s2	-1, B1, B1
53*4882a593Smuzhiyun   [B1]	 subc	.l1x	A4,B4,A4
54*4882a593Smuzhiyun|| [B1]	 add	.s2	-1, B1, B1
55*4882a593Smuzhiyun   [B1]	 subc	.l1x	A4,B4,A4
56*4882a593Smuzhiyun|| [B1]	 add	.s2	-1, B1, B1
57*4882a593Smuzhiyun   [B1]	 subc	.l1x	A4,B4,A4
58*4882a593Smuzhiyun|| [B1]	 add	.s2	-1, B1, B1
59*4882a593Smuzhiyun   [B1]	 subc	.l1x	A4,B4,A4
60*4882a593Smuzhiyun|| [B1]	 add	.s2	-1, B1, B1
61*4882a593Smuzhiyun	;; loop backwards branch happens here
62*4882a593Smuzhiyun
63*4882a593Smuzhiyun	 ret	.s2	B3
64*4882a593Smuzhiyun   [B1]	 subc	.l1x	A4,B4,A4
65*4882a593Smuzhiyun|| [B1]	 add	.s2	-1, B1, B1
66*4882a593Smuzhiyun   [B1]	 subc	.l1x	A4,B4,A4
67*4882a593Smuzhiyun
68*4882a593Smuzhiyun	 extu	.s1	A4, A7, A4
69*4882a593Smuzhiyun	 nop	2
70*4882a593SmuzhiyunENDPROC(__c6xabi_remu)
71