[J-u-boot.git] / include / div64.h

#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() are:
 *
 * u32 do_div(u64 *n, u32 base)
 * {
 *	u32 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

# define do_div(n,base) ({					\
	u32 __base = (base);				\
	u32 __rem;						\
	__rem = ((u64)(n)) % __base;			\
	(n) = ((u64)(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. And yet only if compiling
 * with a sufficiently recent gcc version to perform proper 64-bit constant
 * propagation.
 *
 * (It is unfortunate that gcc doesn't perform all this internally.)
 */

#ifndef __div64_const32_is_OK
#define __div64_const32_is_OK (__GNUC__ >= 4)
#endif

#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).						\
	 */								\
	u64 ___res, ___x, ___t, ___m, ___n = (n);			\
	u32 ___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.					\
		 */							\
		u32 ___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: u64 __arch_xprod_64(const u64 m, u64 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 u64 __arch_xprod_64(const u64 m, u64 n, bool bias)
{
	u32 m_lo = m;
	u32 m_hi = m >> 32;
	u32 n_lo = n;
	u32 n_hi = n >> 32;
	u64 res, tmp;

	if (!bias) {
		res = ((u64)m_lo * n_lo) >> 32;
	} else if (!(m & ((1ULL << 63) | (1ULL << 31)))) {
		/* there can't be any overflow here */
		res = (m + (u64)m_lo * n_lo) >> 32;
	} else {
		res = m + (u64)m_lo * n_lo;
		tmp = (res < m) ? (1ULL << 32) : 0;
		res = (res >> 32) + tmp;
	}

	if (!(m & ((1ULL << 63) | (1ULL << 31)))) {
		/* there can't be any overflow here */
		res += (u64)m_lo * n_hi;
		res += (u64)m_hi * n_lo;
		res >>= 32;
	} else {
		tmp = res += (u64)m_lo * n_hi;
		res += (u64)m_hi * n_lo;
		tmp = (res < tmp) ? (1ULL << 32) : 0;
		res = (res >> 32) + tmp;
	}

	res += (u64)m_hi * n_hi;

	return res;
}
#endif

#ifndef __div64_32
extern u32 __div64_32(u64 *dividend, u32 divisor);
#endif

/* The unnecessary pointer compare is there
 * to check for type safety (n must be 64bit)
 */
# define do_div(n,base) ({				\
	u32 __base = (base);			\
	u32 __rem;					\
	(void)(((typeof((n)) *)0) == ((u64 *)0));	\
	if (__builtin_constant_p(__base) &&		\
	    is_power_of_2(__base)) {			\
		__rem = (n) & (__base - 1);		\
		(n) >>= ilog2(__base);			\
	} else if (__div64_const32_is_OK &&		\
		   __builtin_constant_p(__base) &&	\
		   __base != 0) {			\
		u32 __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 = (u32)(n) % __base;		\
		(n) = (u32)(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 */

/* Wrapper for do_div(). Doesn't modify dividend and returns
 * the result, not remainder.
 */
static inline u64 lldiv(u64 dividend, u32 divisor)
{
	u64 __res = dividend;
	do_div(__res, divisor);
	return(__res);
}

#endif /* _ASM_GENERIC_DIV64_H */
Commit	Line	Data
7b64fef3 WD	1	#ifndef _ASM_GENERIC_DIV64_H
	2	#define _ASM_GENERIC_DIV64_H
	3	/*
	4	* Copyright (C) 2003 Bernardo Innocenti <[email protected]>
	5	* Based on former asm-ppc/div64.h and asm-m68knommu/div64.h
	6	*
0342e335 PF	7	* Optimization for constant divisors on 32-bit machines:
	8	* Copyright (C) 2006-2015 Nicolas Pitre
	9	*
7b64fef3 WD	10	* The semantics of do_div() are:
7b64fef3 WD	11	*
ca49b2c6	12	* u32 do_div(u64 *n, u32 base)
7b64fef3	13	* {
ca49b2c6 SG	14	* u32 remainder = *n % base;
	15	* n = n / base;
	16	* return remainder;
7b64fef3 WD	17	* }
	18	*
	19	* NOTE: macro parameter n is evaluated multiple times,
	20	* beware of side effects!
	21	*/
	22
	23	#include <linux/types.h>
0342e335 PF	24	#include <linux/compiler.h>
	25
	26	#if BITS_PER_LONG == 64
	27
	28	# define do_div(n,base) ({ \
ca49b2c6 SG	29	u32 __base = (base); \
	30	u32 __rem; \
	31	__rem = ((u64)(n)) % __base; \
	32	(n) = ((u64)(n)) / __base; \
0342e335 PF	33	__rem; \
	34	})
	35
	36	#elif BITS_PER_LONG == 32
	37
	38	#include <linux/log2.h>
	39
	40	/*
	41	* If the divisor happens to be constant, we determine the appropriate
	42	* inverse at compile time to turn the division into a few inline
	43	* multiplications which ought to be much faster. And yet only if compiling
	44	* with a sufficiently recent gcc version to perform proper 64-bit constant
	45	* propagation.
	46	*
	47	* (It is unfortunate that gcc doesn't perform all this internally.)
	48	*/
	49
	50	#ifndef __div64_const32_is_OK
	51	#define __div64_const32_is_OK (__GNUC__ >= 4)
	52	#endif
	53
	54	#define __div64_const32(n, ___b) \
	55	({ \
	56	/* \
	57	* Multiplication by reciprocal of b: n / b = n * (p / b) / p \
	58	* \
	59	* We rely on the fact that most of this code gets optimized \
	60	* away at compile time due to constant propagation and only \
	61	* a few multiplication instructions should remain. \
	62	* Hence this monstrous macro (static inline doesn't always \
	63	* do the trick here). \
	64	*/ \
ca49b2c6 SG	65	u64 ___res, ___x, ___t, ___m, ___n = (n); \
ca49b2c6 SG	66	u32 ___p, ___bias; \
0342e335 PF	67	\
	68	/* determine MSB of b */ \
	69	___p = 1 << ilog2(___b); \
	70	\
	71	/* compute m = ((p << 64) + b - 1) / b */ \
	72	___m = (~0ULL / ___b) * ___p; \
	73	___m += (((~0ULL % ___b + 1) * ___p) + ___b - 1) / ___b; \
	74	\
	75	/* one less than the dividend with highest result */ \
	76	___x = ~0ULL / ___b * ___b - 1; \
	77	\
	78	/* test our ___m with res = m * x / (p << 64) */ \
	79	___res = ((___m & 0xffffffff) * (___x & 0xffffffff)) >> 32; \
	80	___t = ___res += (___m & 0xffffffff) * (___x >> 32); \
	81	___res += (___x & 0xffffffff) * (___m >> 32); \
	82	___t = (___res < ___t) ? (1ULL << 32) : 0; \
	83	___res = (___res >> 32) + ___t; \
	84	___res += (___m >> 32) * (___x >> 32); \
	85	___res /= ___p; \
	86	\
	87	/* Now sanitize and optimize what we've got. */ \
	88	if (~0ULL % (___b / (___b & -___b)) == 0) { \
	89	/* special case, can be simplified to ... */ \
	90	___n /= (___b & -___b); \
	91	___m = ~0ULL / (___b / (___b & -___b)); \
	92	___p = 1; \
	93	___bias = 1; \
	94	} else if (___res != ___x / ___b) { \
	95	/* \
	96	* We can't get away without a bias to compensate \
	97	* for bit truncation errors. To avoid it we'd need an \
	98	* additional bit to represent m which would overflow \
	99	* a 64-bit variable. \
	100	* \
	101	* Instead we do m = p / b and n / b = (n * m + m) / p. \
	102	*/ \
	103	___bias = 1; \
	104	/* Compute m = (p << 64) / b */ \
	105	___m = (~0ULL / ___b) * ___p; \
	106	___m += ((~0ULL % ___b + 1) * ___p) / ___b; \
	107	} else { \
	108	/* \
	109	* Reduce m / p, and try to clear bit 31 of m when \
	110	* possible, otherwise that'll need extra overflow \
	111	* handling later. \
	112	*/ \
ca49b2c6	113	u32 ___bits = -(___m & -___m); \
0342e335 PF	114	___bits \|= ___m >> 32; \
	115	___bits = (~___bits) << 1; \
	116	/* \
	117	* If ___bits == 0 then setting bit 31 is unavoidable. \
	118	* Simply apply the maximum possible reduction in that \
	119	* case. Otherwise the MSB of ___bits indicates the \
	120	* best reduction we should apply. \
	121	*/ \
	122	if (!___bits) { \
	123	___p /= (___m & -___m); \
	124	___m /= (___m & -___m); \
	125	} else { \
	126	___p >>= ilog2(___bits); \
	127	___m >>= ilog2(___bits); \
	128	} \
	129	/* No bias needed. */ \
	130	___bias = 0; \
	131	} \
	132	\
	133	/* \
	134	* Now we have a combination of 2 conditions: \
	135	* \
	136	* 1) whether or not we need to apply a bias, and \
	137	* \
	138	* 2) whether or not there might be an overflow in the cross \
	139	* product determined by (___m & ((1 << 63) \| (1 << 31))). \
	140	* \
	141	* Select the best way to do (m_bias + m * n) / (1 << 64). \
	142	* From now on there will be actual runtime code generated. \
	143	*/ \
	144	___res = __arch_xprod_64(___m, ___n, ___bias); \
	145	\
	146	___res /= ___p; \
	147	})
	148
	149	#ifndef __arch_xprod_64
	150	/*
	151	* Default C implementation for __arch_xprod_64()
	152	*
ca49b2c6	153	* Prototype: u64 __arch_xprod_64(const u64 m, u64 n, bool bias)
0342e335 PF	154	* Semantic: retval = ((bias ? m : 0) + m * n) >> 64
	155	*
	156	* The product is a 128-bit value, scaled down to 64 bits.
	157	* Assuming constant propagation to optimize away unused conditional code.
	158	* Architectures may provide their own optimized assembly implementation.
	159	*/
ca49b2c6	160	static inline u64 __arch_xprod_64(const u64 m, u64 n, bool bias)
0342e335	161	{
ca49b2c6 SG	162	u32 m_lo = m;
	163	u32 m_hi = m >> 32;
	164	u32 n_lo = n;
	165	u32 n_hi = n >> 32;
	166	u64 res, tmp;
0342e335 PF	167
0342e335 PF	168	if (!bias) {
ca49b2c6	169	res = ((u64)m_lo * n_lo) >> 32;
0342e335 PF	170	} else if (!(m & ((1ULL << 63) \| (1ULL << 31)))) {
0342e335 PF	171	/* there can't be any overflow here */
ca49b2c6	172	res = (m + (u64)m_lo * n_lo) >> 32;
0342e335	173	} else {
ca49b2c6	174	res = m + (u64)m_lo * n_lo;
0342e335 PF	175	tmp = (res < m) ? (1ULL << 32) : 0;
	176	res = (res >> 32) + tmp;
	177	}
	178
	179	if (!(m & ((1ULL << 63) \| (1ULL << 31)))) {
	180	/* there can't be any overflow here */
ca49b2c6 SG	181	res += (u64)m_lo * n_hi;
ca49b2c6 SG	182	res += (u64)m_hi * n_lo;
0342e335 PF	183	res >>= 32;
0342e335 PF	184	} else {
ca49b2c6 SG	185	tmp = res += (u64)m_lo * n_hi;
ca49b2c6 SG	186	res += (u64)m_hi * n_lo;
0342e335 PF	187	tmp = (res < tmp) ? (1ULL << 32) : 0;
	188	res = (res >> 32) + tmp;
	189	}
7b64fef3	190
ca49b2c6	191	res += (u64)m_hi * n_hi;
0342e335 PF	192
	193	return res;
	194	}
	195	#endif
	196
	197	#ifndef __div64_32
ca49b2c6	198	extern u32 __div64_32(u64 *dividend, u32 divisor);
0342e335	199	#endif
7b64fef3 WD	200
	201	/* The unnecessary pointer compare is there
	202	* to check for type safety (n must be 64bit)
	203	*/
	204	# define do_div(n,base) ({ \
ca49b2c6 SG	205	u32 __base = (base); \
	206	u32 __rem; \
	207	(void)(((typeof((n)) )0) == ((u64 )0)); \
0342e335 PF	208	if (__builtin_constant_p(__base) && \
	209	is_power_of_2(__base)) { \
	210	__rem = (n) & (__base - 1); \
	211	(n) >>= ilog2(__base); \
	212	} else if (__div64_const32_is_OK && \
	213	__builtin_constant_p(__base) && \
	214	__base != 0) { \
ca49b2c6	215	u32 __res_lo, __n_lo = (n); \
0342e335 PF	216	(n) = __div64_const32(n, __base); \
	217	/* the remainder can be computed with 32-bit regs */ \
	218	__res_lo = (n); \
	219	__rem = __n_lo - __res_lo * __base; \
	220	} else if (likely(((n) >> 32) == 0)) { \
ca49b2c6 SG	221	__rem = (u32)(n) % __base; \
ca49b2c6 SG	222	(n) = (u32)(n) / __base; \
0cf207ec	223	} else \
7b64fef3 WD	224	__rem = __div64_32(&(n), __base); \
	225	__rem; \
	226	})
	227
0342e335 PF	228	#else /* BITS_PER_LONG == ?? */
	229
	230	# error do_div() does not yet support the C64
	231
	232	#endif /* BITS_PER_LONG */
	233
3feb647f	234	/* Wrapper for do_div(). Doesn't modify dividend and returns
2121bbe4	235	* the result, not remainder.
3feb647f	236	*/
ca49b2c6	237	static inline u64 lldiv(u64 dividend, u32 divisor)
3feb647f	238	{
ca49b2c6	239	u64 __res = dividend;
3feb647f SP	240	do_div(__res, divisor);
	241	return(__res);
	242	}
	243
7b64fef3	244	#endif /* _ASM_GENERIC_DIV64_H */