162306a36Sopenharmony_ci// SPDX-License-Identifier: GPL-2.0
262306a36Sopenharmony_ci#include <linux/export.h>
362306a36Sopenharmony_ci
462306a36Sopenharmony_ci#include "libgcc.h"
562306a36Sopenharmony_ci
662306a36Sopenharmony_ci/*
762306a36Sopenharmony_ci * GCC 7 & older can suboptimally generate __multi3 calls for mips64r6, so for
862306a36Sopenharmony_ci * that specific case only we implement that intrinsic here.
962306a36Sopenharmony_ci *
1062306a36Sopenharmony_ci * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82981
1162306a36Sopenharmony_ci */
1262306a36Sopenharmony_ci#if defined(CONFIG_64BIT) && defined(CONFIG_CPU_MIPSR6) && (__GNUC__ < 8)
1362306a36Sopenharmony_ci
1462306a36Sopenharmony_ci/* multiply 64-bit values, low 64-bits returned */
1562306a36Sopenharmony_cistatic inline long long notrace dmulu(long long a, long long b)
1662306a36Sopenharmony_ci{
1762306a36Sopenharmony_ci	long long res;
1862306a36Sopenharmony_ci
1962306a36Sopenharmony_ci	asm ("dmulu %0,%1,%2" : "=r" (res) : "r" (a), "r" (b));
2062306a36Sopenharmony_ci	return res;
2162306a36Sopenharmony_ci}
2262306a36Sopenharmony_ci
2362306a36Sopenharmony_ci/* multiply 64-bit unsigned values, high 64-bits of 128-bit result returned */
2462306a36Sopenharmony_cistatic inline long long notrace dmuhu(long long a, long long b)
2562306a36Sopenharmony_ci{
2662306a36Sopenharmony_ci	long long res;
2762306a36Sopenharmony_ci
2862306a36Sopenharmony_ci	asm ("dmuhu %0,%1,%2" : "=r" (res) : "r" (a), "r" (b));
2962306a36Sopenharmony_ci	return res;
3062306a36Sopenharmony_ci}
3162306a36Sopenharmony_ci
3262306a36Sopenharmony_ci/* multiply 128-bit values, low 128-bits returned */
3362306a36Sopenharmony_citi_type notrace __multi3(ti_type a, ti_type b)
3462306a36Sopenharmony_ci{
3562306a36Sopenharmony_ci	TWunion res, aa, bb;
3662306a36Sopenharmony_ci
3762306a36Sopenharmony_ci	aa.ti = a;
3862306a36Sopenharmony_ci	bb.ti = b;
3962306a36Sopenharmony_ci
4062306a36Sopenharmony_ci	/*
4162306a36Sopenharmony_ci	 * a * b =           (a.lo * b.lo)
4262306a36Sopenharmony_ci	 *         + 2^64  * (a.hi * b.lo + a.lo * b.hi)
4362306a36Sopenharmony_ci	 *        [+ 2^128 * (a.hi * b.hi)]
4462306a36Sopenharmony_ci	 */
4562306a36Sopenharmony_ci	res.s.low = dmulu(aa.s.low, bb.s.low);
4662306a36Sopenharmony_ci	res.s.high = dmuhu(aa.s.low, bb.s.low);
4762306a36Sopenharmony_ci	res.s.high += dmulu(aa.s.high, bb.s.low);
4862306a36Sopenharmony_ci	res.s.high += dmulu(aa.s.low, bb.s.high);
4962306a36Sopenharmony_ci
5062306a36Sopenharmony_ci	return res.ti;
5162306a36Sopenharmony_ci}
5262306a36Sopenharmony_ciEXPORT_SYMBOL(__multi3);
5362306a36Sopenharmony_ci
5462306a36Sopenharmony_ci#endif /* 64BIT && CPU_MIPSR6 && GCC7 */
55