xref: /openbmc/linux/arch/ia64/lib/idiv32.S (revision c900529f3d9161bfde5cca0754f83b4d3c3e0220)
1b2441318SGreg Kroah-Hartman/* SPDX-License-Identifier: GPL-2.0 */
21da177e4SLinus Torvalds/*
31da177e4SLinus Torvalds * Copyright (C) 2000 Hewlett-Packard Co
41da177e4SLinus Torvalds * Copyright (C) 2000 David Mosberger-Tang <davidm@hpl.hp.com>
51da177e4SLinus Torvalds *
61da177e4SLinus Torvalds * 32-bit integer division.
71da177e4SLinus Torvalds *
81da177e4SLinus Torvalds * This code is based on the application note entitled "Divide, Square Root
91da177e4SLinus Torvalds * and Remainder Algorithms for the IA-64 Architecture".  This document
101da177e4SLinus Torvalds * is available as Intel document number 248725-002 or via the web at
111da177e4SLinus Torvalds * http://developer.intel.com/software/opensource/numerics/
121da177e4SLinus Torvalds *
131da177e4SLinus Torvalds * For more details on the theory behind these algorithms, see "IA-64
141da177e4SLinus Torvalds * and Elementary Functions" by Peter Markstein; HP Professional Books
15a9eb221aSSina Hamedian * (http://www.goodreads.com/book/show/2019887.Ia_64_and_Elementary_Functions)
161da177e4SLinus Torvalds */
171da177e4SLinus Torvalds
18*ab03e604SMasahiro Yamada#include <linux/export.h>
191da177e4SLinus Torvalds#include <asm/asmmacro.h>
201da177e4SLinus Torvalds
211da177e4SLinus Torvalds#ifdef MODULO
221da177e4SLinus Torvalds# define OP	mod
231da177e4SLinus Torvalds#else
241da177e4SLinus Torvalds# define OP	div
251da177e4SLinus Torvalds#endif
261da177e4SLinus Torvalds
271da177e4SLinus Torvalds#ifdef UNSIGNED
281da177e4SLinus Torvalds# define SGN	u
291da177e4SLinus Torvalds# define EXTEND	zxt4
301da177e4SLinus Torvalds# define INT_TO_FP(a,b)	fcvt.xuf.s1 a=b
311da177e4SLinus Torvalds# define FP_TO_INT(a,b)	fcvt.fxu.trunc.s1 a=b
321da177e4SLinus Torvalds#else
331da177e4SLinus Torvalds# define SGN
341da177e4SLinus Torvalds# define EXTEND	sxt4
351da177e4SLinus Torvalds# define INT_TO_FP(a,b)	fcvt.xf a=b
361da177e4SLinus Torvalds# define FP_TO_INT(a,b)	fcvt.fx.trunc.s1 a=b
371da177e4SLinus Torvalds#endif
381da177e4SLinus Torvalds
391da177e4SLinus Torvalds#define PASTE1(a,b)	a##b
401da177e4SLinus Torvalds#define PASTE(a,b)	PASTE1(a,b)
411da177e4SLinus Torvalds#define NAME		PASTE(PASTE(__,SGN),PASTE(OP,si3))
421da177e4SLinus Torvalds
431da177e4SLinus TorvaldsGLOBAL_ENTRY(NAME)
441da177e4SLinus Torvalds	.regstk 2,0,0,0
451da177e4SLinus Torvalds	// Transfer inputs to FP registers.
461da177e4SLinus Torvalds	mov r2 = 0xffdd			// r2 = -34 + 65535 (fp reg format bias)
471da177e4SLinus Torvalds	EXTEND in0 = in0		// in0 = a
481da177e4SLinus Torvalds	EXTEND in1 = in1		// in1 = b
491da177e4SLinus Torvalds	;;
501da177e4SLinus Torvalds	setf.sig f8 = in0
511da177e4SLinus Torvalds	setf.sig f9 = in1
521da177e4SLinus Torvalds#ifdef MODULO
531da177e4SLinus Torvalds	sub in1 = r0, in1		// in1 = -b
541da177e4SLinus Torvalds#endif
551da177e4SLinus Torvalds	;;
561da177e4SLinus Torvalds	// Convert the inputs to FP, to avoid FP software-assist faults.
571da177e4SLinus Torvalds	INT_TO_FP(f8, f8)
581da177e4SLinus Torvalds	INT_TO_FP(f9, f9)
591da177e4SLinus Torvalds	;;
601da177e4SLinus Torvalds	setf.exp f7 = r2		// f7 = 2^-34
611da177e4SLinus Torvalds	frcpa.s1 f6, p6 = f8, f9	// y0 = frcpa(b)
621da177e4SLinus Torvalds	;;
631da177e4SLinus Torvalds(p6)	fmpy.s1 f8 = f8, f6		// q0 = a*y0
641da177e4SLinus Torvalds(p6)	fnma.s1 f6 = f9, f6, f1		// e0 = -b*y0 + 1
651da177e4SLinus Torvalds	;;
661da177e4SLinus Torvalds#ifdef MODULO
671da177e4SLinus Torvalds	setf.sig f9 = in1		// f9 = -b
681da177e4SLinus Torvalds#endif
691da177e4SLinus Torvalds(p6)	fma.s1 f8 = f6, f8, f8		// q1 = e0*q0 + q0
701da177e4SLinus Torvalds(p6)	fma.s1 f6 = f6, f6, f7		// e1 = e0*e0 + 2^-34
711da177e4SLinus Torvalds	;;
721da177e4SLinus Torvalds#ifdef MODULO
731da177e4SLinus Torvalds	setf.sig f7 = in0
741da177e4SLinus Torvalds#endif
751da177e4SLinus Torvalds(p6)	fma.s1 f6 = f6, f8, f8		// q2 = e1*q1 + q1
761da177e4SLinus Torvalds	;;
771da177e4SLinus Torvalds	FP_TO_INT(f6, f6)		// q = trunc(q2)
781da177e4SLinus Torvalds	;;
791da177e4SLinus Torvalds#ifdef MODULO
801da177e4SLinus Torvalds	xma.l f6 = f6, f9, f7		// r = q*(-b) + a
811da177e4SLinus Torvalds	;;
821da177e4SLinus Torvalds#endif
831da177e4SLinus Torvalds	getf.sig r8 = f6		// transfer result to result register
841da177e4SLinus Torvalds	br.ret.sptk.many rp
851da177e4SLinus TorvaldsEND(NAME)
86e007c533SAl ViroEXPORT_SYMBOL(NAME)
87