linux/arch/c6x/lib/remu.S

;; SPDX-License-Identifier: GPL-2.0-or-later
;;  Copyright 2010  Free Software Foundation, Inc.
;;  Contributed by Bernd Schmidt <bernds@codesourcery.com>.
;;

#include <linux/linkage.h>

	;; ABI considerations for the divide functions
	;; The following registers are call-used:
	;; __c6xabi_divi A0,A1,A2,A4,A6,B0,B1,B2,B4,B5
	;; __c6xabi_divu A0,A1,A2,A4,A6,B0,B1,B2,B4
	;; __c6xabi_remi A1,A2,A4,A5,A6,B0,B1,B2,B4
	;; __c6xabi_remu A1,A4,A5,A7,B0,B1,B2,B4
	;;
	;; In our implementation, divu and remu are leaf functions,
	;; while both divi and remi call into divu.
	;; A0 is not clobbered by any of the functions.
	;; divu does not clobber B2 either, which is taken advantage of
	;; in remi.
	;; divi uses B5 to hold the original return address during
	;; the call to divu.
	;; remi uses B2 and A5 to hold the input values during the
	;; call to divu.  It stores B3 in on the stack.


	.text

ENTRY(__c6xabi_remu)
	;; The ABI seems designed to prevent these functions calling each other,
	;; so we duplicate most of the divsi3 code here.
	 mv	.s2x	A4, B1
	 lmbd	.l2	1, B4, B1
|| [!B1] b	.s2	B3	; RETURN A
|| [!B1] mvk	.d2	1, B4

	 mv	.l1x	B1, A7
||	 shl	.s2	B4, B1, B4

	 cmpltu	.l1x	A4, B4, A1
   [!A1] sub	.l1x	A4, B4, A4
	 shru	.s2	B4, 1, B4

_remu_loop:
	 cmpgt	.l2	B1, 7, B0
|| [B1]	 subc	.l1x	A4,B4,A4
|| [B1]	 add	.s2	-1, B1, B1
	;; RETURN A may happen here (note: must happen before the next branch)
   [B1]	 subc	.l1x	A4,B4,A4
|| [B1]	 add	.s2	-1, B1, B1
|| [B0]	 b	.s1	_remu_loop
   [B1]	 subc	.l1x	A4,B4,A4
|| [B1]	 add	.s2	-1, B1, B1
   [B1]	 subc	.l1x	A4,B4,A4
|| [B1]	 add	.s2	-1, B1, B1
   [B1]	 subc	.l1x	A4,B4,A4
|| [B1]	 add	.s2	-1, B1, B1
   [B1]	 subc	.l1x	A4,B4,A4
|| [B1]	 add	.s2	-1, B1, B1
   [B1]	 subc	.l1x	A4,B4,A4
|| [B1]	 add	.s2	-1, B1, B1
	;; loop backwards branch happens here

	 ret	.s2	B3
   [B1]	 subc	.l1x	A4,B4,A4
|| [B1]	 add	.s2	-1, B1, B1
   [B1]	 subc	.l1x	A4,B4,A4

	 extu	.s1	A4, A7, A4
	 nop	2
ENDPROC(__c6xabi_remu)