[linux.git] / include / asm-generic / div64.h

/* SPDX-License-Identifier: GPL-2.0 */
#ifndef _ASM_GENERIC_DIV64_H
#define _ASM_GENERIC_DIV64_H
/*
 * Copyright (C) 2003 Bernardo Innocenti <[email protected]>
 * Based on former asm-ppc/div64.h and asm-m68knommu/div64.h
 *
 * Optimization for constant divisors on 32-bit machines:
 * Copyright (C) 2006-2015 Nicolas Pitre
 *
 * The semantics of do_div() is, in C++ notation, observing that the name
 * is a function-like macro and the n parameter has the semantics of a C++
 * reference:
 *
 * uint32_t do_div(uint64_t &n, uint32_t base)
 * {
 * 	uint32_t remainder = n % base;
 * 	n = n / base;
 * 	return remainder;
 * }
 *
 * NOTE: macro parameter n is evaluated multiple times,
 *       beware of side effects!
 */

#include <linux/types.h>
#include <linux/compiler.h>

#if BITS_PER_LONG == 64

/**
 * do_div - returns 2 values: calculate remainder and update new dividend
 * @n: uint64_t dividend (will be updated)
 * @base: uint32_t divisor
 *
 * Summary:
 * ``uint32_t remainder = n % base;``
 * ``n = n / base;``
 *
 * Return: (uint32_t)remainder
 *
 * NOTE: macro parameter @n is evaluated multiple times,
 * beware of side effects!
 */
# define do_div(n,base) ({					\
	uint32_t __base = (base);				\
	uint32_t __rem;						\
	__rem = ((uint64_t)(n)) % __base;			\
	(n) = ((uint64_t)(n)) / __base;				\
	__rem;							\
 })

#elif BITS_PER_LONG == 32

#include <linux/log2.h>

/*
 * If the divisor happens to be constant, we determine the appropriate
 * inverse at compile time to turn the division into a few inline
 * multiplications which ought to be much faster.
 *
 * (It is unfortunate that gcc doesn't perform all this internally.)
 */

#define __div64_const32(n, ___b)					\
({									\
	/*								\
	 * Multiplication by reciprocal of b: n / b = n * (p / b) / p	\
	 *								\
	 * We rely on the fact that most of this code gets optimized	\
	 * away at compile time due to constant propagation and only	\
	 * a few multiplication instructions should remain.		\
	 * Hence this monstrous macro (static inline doesn't always	\
	 * do the trick here).						\
	 */								\
	uint64_t ___res, ___x, ___t, ___m, ___n = (n);			\
	uint32_t ___p, ___bias;						\
									\
	/* determine MSB of b */					\
	___p = 1 << ilog2(___b);					\
									\
	/* compute m = ((p << 64) + b - 1) / b */			\
	___m = (~0ULL / ___b) * ___p;					\
	___m += (((~0ULL % ___b + 1) * ___p) + ___b - 1) / ___b;	\
									\
	/* one less than the dividend with highest result */		\
	___x = ~0ULL / ___b * ___b - 1;					\
									\
	/* test our ___m with res = m * x / (p << 64) */		\
	___res = ((___m & 0xffffffff) * (___x & 0xffffffff)) >> 32;	\
	___t = ___res += (___m & 0xffffffff) * (___x >> 32);		\
	___res += (___x & 0xffffffff) * (___m >> 32);			\
	___t = (___res < ___t) ? (1ULL << 32) : 0;			\
	___res = (___res >> 32) + ___t;					\
	___res += (___m >> 32) * (___x >> 32);				\
	___res /= ___p;							\
									\
	/* Now sanitize and optimize what we've got. */			\
	if (~0ULL % (___b / (___b & -___b)) == 0) {			\
		/* special case, can be simplified to ... */		\
		___n /= (___b & -___b);					\
		___m = ~0ULL / (___b / (___b & -___b));			\
		___p = 1;						\
		___bias = 1;						\
	} else if (___res != ___x / ___b) {				\
		/*							\
		 * We can't get away without a bias to compensate	\
		 * for bit truncation errors.  To avoid it we'd need an	\
		 * additional bit to represent m which would overflow	\
		 * a 64-bit variable.					\
		 *							\
		 * Instead we do m = p / b and n / b = (n * m + m) / p.	\
		 */							\
		___bias = 1;						\
		/* Compute m = (p << 64) / b */				\
		___m = (~0ULL / ___b) * ___p;				\
		___m += ((~0ULL % ___b + 1) * ___p) / ___b;		\
	} else {							\
		/*							\
		 * Reduce m / p, and try to clear bit 31 of m when	\
		 * possible, otherwise that'll need extra overflow	\
		 * handling later.					\
		 */							\
		uint32_t ___bits = -(___m & -___m);			\
		___bits |= ___m >> 32;					\
		___bits = (~___bits) << 1;				\
		/*							\
		 * If ___bits == 0 then setting bit 31 is  unavoidable.	\
		 * Simply apply the maximum possible reduction in that	\
		 * case. Otherwise the MSB of ___bits indicates the	\
		 * best reduction we should apply.			\
		 */							\
		if (!___bits) {						\
			___p /= (___m & -___m);				\
			___m /= (___m & -___m);				\
		} else {						\
			___p >>= ilog2(___bits);			\
			___m >>= ilog2(___bits);			\
		}							\
		/* No bias needed. */					\
		___bias = 0;						\
	}								\
									\
	/*								\
	 * Now we have a combination of 2 conditions:			\
	 *								\
	 * 1) whether or not we need to apply a bias, and		\
	 *								\
	 * 2) whether or not there might be an overflow in the cross	\
	 *    product determined by (___m & ((1 << 63) | (1 << 31))).	\
	 *								\
	 * Select the best way to do (m_bias + m * n) / (1 << 64).	\
	 * From now on there will be actual runtime code generated.	\
	 */								\
	___res = __arch_xprod_64(___m, ___n, ___bias);			\
									\
	___res /= ___p;							\
})

#ifndef __arch_xprod_64
/*
 * Default C implementation for __arch_xprod_64()
 *
 * Prototype: uint64_t __arch_xprod_64(const uint64_t m, uint64_t n, bool bias)
 * Semantic:  retval = ((bias ? m : 0) + m * n) >> 64
 *
 * The product is a 128-bit value, scaled down to 64 bits.
 * Assuming constant propagation to optimize away unused conditional code.
 * Architectures may provide their own optimized assembly implementation.
 */
static inline uint64_t __arch_xprod_64(const uint64_t m, uint64_t n, bool bias)
{
	uint32_t m_lo = m;
	uint32_t m_hi = m >> 32;
	uint32_t n_lo = n;
	uint32_t n_hi = n >> 32;
	uint64_t res;
	uint32_t res_lo, res_hi, tmp;

	if (!bias) {
		res = ((uint64_t)m_lo * n_lo) >> 32;
	} else if (!(m & ((1ULL << 63) | (1ULL << 31)))) {
		/* there can't be any overflow here */
		res = (m + (uint64_t)m_lo * n_lo) >> 32;
	} else {
		res = m + (uint64_t)m_lo * n_lo;
		res_lo = res >> 32;
		res_hi = (res_lo < m_hi);
		res = res_lo | ((uint64_t)res_hi << 32);
	}

	if (!(m & ((1ULL << 63) | (1ULL << 31)))) {
		/* there can't be any overflow here */
		res += (uint64_t)m_lo * n_hi;
		res += (uint64_t)m_hi * n_lo;
		res >>= 32;
	} else {
		res += (uint64_t)m_lo * n_hi;
		tmp = res >> 32;
		res += (uint64_t)m_hi * n_lo;
		res_lo = res >> 32;
		res_hi = (res_lo < tmp);
		res = res_lo | ((uint64_t)res_hi << 32);
	}

	res += (uint64_t)m_hi * n_hi;

	return res;
}
#endif

#ifndef __div64_32
extern uint32_t __div64_32(uint64_t *dividend, uint32_t divisor);
#endif

/* The unnecessary pointer compare is there
 * to check for type safety (n must be 64bit)
 */
# define do_div(n,base) ({				\
	uint32_t __base = (base);			\
	uint32_t __rem;					\
	(void)(((typeof((n)) *)0) == ((uint64_t *)0));	\
	if (__builtin_constant_p(__base) &&		\
	    is_power_of_2(__base)) {			\
		__rem = (n) & (__base - 1);		\
		(n) >>= ilog2(__base);			\
	} else if (__builtin_constant_p(__base) &&	\
		   __base != 0) {			\
		uint32_t __res_lo, __n_lo = (n);	\
		(n) = __div64_const32(n, __base);	\
		/* the remainder can be computed with 32-bit regs */ \
		__res_lo = (n);				\
		__rem = __n_lo - __res_lo * __base;	\
	} else if (likely(((n) >> 32) == 0)) {		\
		__rem = (uint32_t)(n) % __base;		\
		(n) = (uint32_t)(n) / __base;		\
	} else {					\
		__rem = __div64_32(&(n), __base);	\
	}						\
	__rem;						\
 })

#else /* BITS_PER_LONG == ?? */

# error do_div() does not yet support the C64

#endif /* BITS_PER_LONG */

#endif /* _ASM_GENERIC_DIV64_H */
Commit	Line	Data
b2441318	1	/* SPDX-License-Identifier: GPL-2.0 */
1da177e4 LT	2	#ifndef _ASM_GENERIC_DIV64_H
	3	#define _ASM_GENERIC_DIV64_H
	4	/*
	5	* Copyright (C) 2003 Bernardo Innocenti <[email protected]>
	6	* Based on former asm-ppc/div64.h and asm-m68knommu/div64.h
	7	*
461a5e51 NP	8	* Optimization for constant divisors on 32-bit machines:
	9	* Copyright (C) 2006-2015 Nicolas Pitre
	10	*
f2875832 MR	11	* The semantics of do_div() is, in C++ notation, observing that the name
	12	* is a function-like macro and the n parameter has the semantics of a C++
	13	* reference:
1da177e4	14	*
f2875832	15	* uint32_t do_div(uint64_t &n, uint32_t base)
1da177e4	16	* {
f2875832 MR	17	* uint32_t remainder = n % base;
f2875832 MR	18	* n = n / base;
1da177e4 LT	19	* return remainder;
	20	* }
	21	*
	22	* NOTE: macro parameter n is evaluated multiple times,
	23	* beware of side effects!
	24	*/
	25
	26	#include <linux/types.h>
	27	#include <linux/compiler.h>
	28
	29	#if BITS_PER_LONG == 64
	30
6ec72e61 RD	31	/**
6ec72e61 RD	32	* do_div - returns 2 values: calculate remainder and update new dividend
e8e4eb0f	33	* @n: uint64_t dividend (will be updated)
6ec72e61 RD	34	* @base: uint32_t divisor
	35	*
	36	* Summary:
e8e4eb0f JN	37	* ``uint32_t remainder = n % base;``
e8e4eb0f JN	38	* ``n = n / base;``
6ec72e61 RD	39	*
	40	* Return: (uint32_t)remainder
	41	*
	42	* NOTE: macro parameter @n is evaluated multiple times,
	43	* beware of side effects!
	44	*/
1da177e4 LT	45	# define do_div(n,base) ({ \
	46	uint32_t __base = (base); \
	47	uint32_t __rem; \
	48	__rem = ((uint64_t)(n)) % __base; \
	49	(n) = ((uint64_t)(n)) / __base; \
	50	__rem; \
	51	})
	52
	53	#elif BITS_PER_LONG == 32
	54
911918aa NP	55	#include <linux/log2.h>
911918aa NP	56
461a5e51 NP	57	/*
	58	* If the divisor happens to be constant, we determine the appropriate
	59	* inverse at compile time to turn the division into a few inline
c747ce47	60	* multiplications which ought to be much faster.
461a5e51 NP	61	*
	62	* (It is unfortunate that gcc doesn't perform all this internally.)
	63	*/
	64
461a5e51 NP	65	#define __div64_const32(n, ___b) \
	66	({ \
	67	/* \
	68	* Multiplication by reciprocal of b: n / b = n * (p / b) / p \
	69	* \
	70	* We rely on the fact that most of this code gets optimized \
	71	* away at compile time due to constant propagation and only \
	72	* a few multiplication instructions should remain. \
	73	* Hence this monstrous macro (static inline doesn't always \
	74	* do the trick here). \
	75	*/ \
	76	uint64_t ___res, ___x, ___t, ___m, ___n = (n); \
f682b27c	77	uint32_t ___p, ___bias; \
461a5e51 NP	78	\
	79	/* determine MSB of b */ \
	80	___p = 1 << ilog2(___b); \
	81	\
	82	/* compute m = ((p << 64) + b - 1) / b */ \
	83	___m = (~0ULL / ___b) * ___p; \
	84	___m += (((~0ULL % ___b + 1) * ___p) + ___b - 1) / ___b; \
	85	\
	86	/* one less than the dividend with highest result */ \
	87	___x = ~0ULL / ___b * ___b - 1; \
	88	\
	89	/* test our ___m with res = m * x / (p << 64) */ \
	90	___res = ((___m & 0xffffffff) * (___x & 0xffffffff)) >> 32; \
	91	___t = ___res += (___m & 0xffffffff) * (___x >> 32); \
	92	___res += (___x & 0xffffffff) * (___m >> 32); \
	93	___t = (___res < ___t) ? (1ULL << 32) : 0; \
	94	___res = (___res >> 32) + ___t; \
	95	___res += (___m >> 32) * (___x >> 32); \
	96	___res /= ___p; \
	97	\
	98	/* Now sanitize and optimize what we've got. */ \
	99	if (~0ULL % (___b / (___b & -___b)) == 0) { \
	100	/* special case, can be simplified to ... */ \
	101	___n /= (___b & -___b); \
	102	___m = ~0ULL / (___b / (___b & -___b)); \
	103	___p = 1; \
	104	___bias = 1; \
	105	} else if (___res != ___x / ___b) { \
	106	/* \
	107	* We can't get away without a bias to compensate \
	108	* for bit truncation errors. To avoid it we'd need an \
	109	* additional bit to represent m which would overflow \
	110	* a 64-bit variable. \
	111	* \
	112	* Instead we do m = p / b and n / b = (n * m + m) / p. \
	113	*/ \
	114	___bias = 1; \
	115	/* Compute m = (p << 64) / b */ \
	116	___m = (~0ULL / ___b) * ___p; \
	117	___m += ((~0ULL % ___b + 1) * ___p) / ___b; \
	118	} else { \
	119	/* \
	120	* Reduce m / p, and try to clear bit 31 of m when \
	121	* possible, otherwise that'll need extra overflow \
	122	* handling later. \
	123	*/ \
	124	uint32_t ___bits = -(___m & -___m); \
	125	___bits \|= ___m >> 32; \
	126	___bits = (~___bits) << 1; \
	127	/* \
	128	* If ___bits == 0 then setting bit 31 is unavoidable. \
	129	* Simply apply the maximum possible reduction in that \
	130	* case. Otherwise the MSB of ___bits indicates the \
	131	* best reduction we should apply. \
	132	*/ \
	133	if (!___bits) { \
	134	___p /= (___m & -___m); \
	135	___m /= (___m & -___m); \
	136	} else { \
	137	___p >>= ilog2(___bits); \
	138	___m >>= ilog2(___bits); \
	139	} \
	140	/* No bias needed. */ \
	141	___bias = 0; \
142	} \
143	\
144	/* \
145	* Now we have a combination of 2 conditions: \
146	* \
147	* 1) whether or not we need to apply a bias, and \
148	* \
149	* 2) whether or not there might be an overflow in the cross \
150	* product determined by (___m & ((1 << 63) \| (1 << 31))). \
151	* \
f682b27c	152	* Select the best way to do (m_bias + m * n) / (1 << 64). \
461a5e51 NP	153	* From now on there will be actual runtime code generated. \
461a5e51 NP	154	*/ \
f682b27c	155	___res = __arch_xprod_64(___m, ___n, ___bias); \
461a5e51 NP	156	\
	157	___res /= ___p; \
	158	})
	159
f682b27c NP	160	#ifndef __arch_xprod_64
	161	/*
	162	* Default C implementation for __arch_xprod_64()
	163	*
	164	* Prototype: uint64_t __arch_xprod_64(const uint64_t m, uint64_t n, bool bias)
	165	* Semantic: retval = ((bias ? m : 0) + m * n) >> 64
	166	*
	167	* The product is a 128-bit value, scaled down to 64 bits.
	168	* Assuming constant propagation to optimize away unused conditional code.
	169	* Architectures may provide their own optimized assembly implementation.
	170	*/
	171	static inline uint64_t __arch_xprod_64(const uint64_t m, uint64_t n, bool bias)
	172	{
	173	uint32_t m_lo = m;
	174	uint32_t m_hi = m >> 32;
	175	uint32_t n_lo = n;
	176	uint32_t n_hi = n >> 32;
602828c1 NP	177	uint64_t res;
602828c1 NP	178	uint32_t res_lo, res_hi, tmp;
f682b27c NP	179
	180	if (!bias) {
	181	res = ((uint64_t)m_lo * n_lo) >> 32;
	182	} else if (!(m & ((1ULL << 63) \| (1ULL << 31)))) {
	183	/* there can't be any overflow here */
	184	res = (m + (uint64_t)m_lo * n_lo) >> 32;
	185	} else {
	186	res = m + (uint64_t)m_lo * n_lo;
602828c1 NP	187	res_lo = res >> 32;
	188	res_hi = (res_lo < m_hi);
	189	res = res_lo \| ((uint64_t)res_hi << 32);
f682b27c NP	190	}
	191
	192	if (!(m & ((1ULL << 63) \| (1ULL << 31)))) {
	193	/* there can't be any overflow here */
	194	res += (uint64_t)m_lo * n_hi;
	195	res += (uint64_t)m_hi * n_lo;
	196	res >>= 32;
	197	} else {
602828c1 NP	198	res += (uint64_t)m_lo * n_hi;
602828c1 NP	199	tmp = res >> 32;
f682b27c	200	res += (uint64_t)m_hi * n_lo;
602828c1 NP	201	res_lo = res >> 32;
	202	res_hi = (res_lo < tmp);
	203	res = res_lo \| ((uint64_t)res_hi << 32);
f682b27c NP	204	}
	205
	206	res += (uint64_t)m_hi * n_hi;
	207
	208	return res;
	209	}
	210	#endif
	211
dce1eb93	212	#ifndef __div64_32
1da177e4	213	extern uint32_t __div64_32(uint64_t *dividend, uint32_t divisor);
dce1eb93	214	#endif
1da177e4 LT	215
	216	/* The unnecessary pointer compare is there
	217	* to check for type safety (n must be 64bit)
	218	*/
	219	# define do_div(n,base) ({ \
	220	uint32_t __base = (base); \
	221	uint32_t __rem; \
	222	(void)(((typeof((n)) )0) == ((uint64_t )0)); \
911918aa NP	223	if (__builtin_constant_p(__base) && \
	224	is_power_of_2(__base)) { \
	225	__rem = (n) & (__base - 1); \
	226	(n) >>= ilog2(__base); \
c747ce47	227	} else if (__builtin_constant_p(__base) && \
461a5e51 NP	228	__base != 0) { \
	229	uint32_t __res_lo, __n_lo = (n); \
	230	(n) = __div64_const32(n, __base); \
	231	/* the remainder can be computed with 32-bit regs */ \
	232	__res_lo = (n); \
	233	__rem = __n_lo - __res_lo * __base; \
911918aa	234	} else if (likely(((n) >> 32) == 0)) { \
1da177e4 LT	235	__rem = (uint32_t)(n) % __base; \
1da177e4 LT	236	(n) = (uint32_t)(n) / __base; \
c747ce47	237	} else { \
1da177e4	238	__rem = __div64_32(&(n), __base); \
c747ce47	239	} \
1da177e4 LT	240	__rem; \
	241	})
	242
	243	#else /* BITS_PER_LONG == ?? */
	244
	245	# error do_div() does not yet support the C64
	246
	247	#endif /* BITS_PER_LONG */
	248
	249	#endif /* _ASM_GENERIC_DIV64_H */