xref: /openbmc/linux/arch/m68k/math-emu/multi_arith.h (revision 1c2dd16a)
1 /* multi_arith.h: multi-precision integer arithmetic functions, needed
2    to do extended-precision floating point.
3 
4    (c) 1998 David Huggins-Daines.
5 
6    Somewhat based on arch/alpha/math-emu/ieee-math.c, which is (c)
7    David Mosberger-Tang.
8 
9    You may copy, modify, and redistribute this file under the terms of
10    the GNU General Public License, version 2, or any later version, at
11    your convenience. */
12 
13 /* Note:
14 
15    These are not general multi-precision math routines.  Rather, they
16    implement the subset of integer arithmetic that we need in order to
17    multiply, divide, and normalize 128-bit unsigned mantissae.  */
18 
19 #ifndef MULTI_ARITH_H
20 #define MULTI_ARITH_H
21 
22 static inline void fp_denormalize(struct fp_ext *reg, unsigned int cnt)
23 {
24 	reg->exp += cnt;
25 
26 	switch (cnt) {
27 	case 0 ... 8:
28 		reg->lowmant = reg->mant.m32[1] << (8 - cnt);
29 		reg->mant.m32[1] = (reg->mant.m32[1] >> cnt) |
30 				   (reg->mant.m32[0] << (32 - cnt));
31 		reg->mant.m32[0] = reg->mant.m32[0] >> cnt;
32 		break;
33 	case 9 ... 32:
34 		reg->lowmant = reg->mant.m32[1] >> (cnt - 8);
35 		if (reg->mant.m32[1] << (40 - cnt))
36 			reg->lowmant |= 1;
37 		reg->mant.m32[1] = (reg->mant.m32[1] >> cnt) |
38 				   (reg->mant.m32[0] << (32 - cnt));
39 		reg->mant.m32[0] = reg->mant.m32[0] >> cnt;
40 		break;
41 	case 33 ... 39:
42 		asm volatile ("bfextu %1{%2,#8},%0" : "=d" (reg->lowmant)
43 			: "m" (reg->mant.m32[0]), "d" (64 - cnt));
44 		if (reg->mant.m32[1] << (40 - cnt))
45 			reg->lowmant |= 1;
46 		reg->mant.m32[1] = reg->mant.m32[0] >> (cnt - 32);
47 		reg->mant.m32[0] = 0;
48 		break;
49 	case 40 ... 71:
50 		reg->lowmant = reg->mant.m32[0] >> (cnt - 40);
51 		if ((reg->mant.m32[0] << (72 - cnt)) || reg->mant.m32[1])
52 			reg->lowmant |= 1;
53 		reg->mant.m32[1] = reg->mant.m32[0] >> (cnt - 32);
54 		reg->mant.m32[0] = 0;
55 		break;
56 	default:
57 		reg->lowmant = reg->mant.m32[0] || reg->mant.m32[1];
58 		reg->mant.m32[0] = 0;
59 		reg->mant.m32[1] = 0;
60 		break;
61 	}
62 }
63 
64 static inline int fp_overnormalize(struct fp_ext *reg)
65 {
66 	int shift;
67 
68 	if (reg->mant.m32[0]) {
69 		asm ("bfffo %1{#0,#32},%0" : "=d" (shift) : "dm" (reg->mant.m32[0]));
70 		reg->mant.m32[0] = (reg->mant.m32[0] << shift) | (reg->mant.m32[1] >> (32 - shift));
71 		reg->mant.m32[1] = (reg->mant.m32[1] << shift);
72 	} else {
73 		asm ("bfffo %1{#0,#32},%0" : "=d" (shift) : "dm" (reg->mant.m32[1]));
74 		reg->mant.m32[0] = (reg->mant.m32[1] << shift);
75 		reg->mant.m32[1] = 0;
76 		shift += 32;
77 	}
78 
79 	return shift;
80 }
81 
82 static inline int fp_addmant(struct fp_ext *dest, struct fp_ext *src)
83 {
84 	int carry;
85 
86 	/* we assume here, gcc only insert move and a clr instr */
87 	asm volatile ("add.b %1,%0" : "=d,g" (dest->lowmant)
88 		: "g,d" (src->lowmant), "0,0" (dest->lowmant));
89 	asm volatile ("addx.l %1,%0" : "=d" (dest->mant.m32[1])
90 		: "d" (src->mant.m32[1]), "0" (dest->mant.m32[1]));
91 	asm volatile ("addx.l %1,%0" : "=d" (dest->mant.m32[0])
92 		: "d" (src->mant.m32[0]), "0" (dest->mant.m32[0]));
93 	asm volatile ("addx.l %0,%0" : "=d" (carry) : "0" (0));
94 
95 	return carry;
96 }
97 
98 static inline int fp_addcarry(struct fp_ext *reg)
99 {
100 	if (++reg->exp == 0x7fff) {
101 		if (reg->mant.m64)
102 			fp_set_sr(FPSR_EXC_INEX2);
103 		reg->mant.m64 = 0;
104 		fp_set_sr(FPSR_EXC_OVFL);
105 		return 0;
106 	}
107 	reg->lowmant = (reg->mant.m32[1] << 7) | (reg->lowmant ? 1 : 0);
108 	reg->mant.m32[1] = (reg->mant.m32[1] >> 1) |
109 			   (reg->mant.m32[0] << 31);
110 	reg->mant.m32[0] = (reg->mant.m32[0] >> 1) | 0x80000000;
111 
112 	return 1;
113 }
114 
115 static inline void fp_submant(struct fp_ext *dest, struct fp_ext *src1,
116 			      struct fp_ext *src2)
117 {
118 	/* we assume here, gcc only insert move and a clr instr */
119 	asm volatile ("sub.b %1,%0" : "=d,g" (dest->lowmant)
120 		: "g,d" (src2->lowmant), "0,0" (src1->lowmant));
121 	asm volatile ("subx.l %1,%0" : "=d" (dest->mant.m32[1])
122 		: "d" (src2->mant.m32[1]), "0" (src1->mant.m32[1]));
123 	asm volatile ("subx.l %1,%0" : "=d" (dest->mant.m32[0])
124 		: "d" (src2->mant.m32[0]), "0" (src1->mant.m32[0]));
125 }
126 
127 #define fp_mul64(desth, destl, src1, src2) ({				\
128 	asm ("mulu.l %2,%1:%0" : "=d" (destl), "=d" (desth)		\
129 		: "dm" (src1), "0" (src2));				\
130 })
131 #define fp_div64(quot, rem, srch, srcl, div)				\
132 	asm ("divu.l %2,%1:%0" : "=d" (quot), "=d" (rem)		\
133 		: "dm" (div), "1" (srch), "0" (srcl))
134 #define fp_add64(dest1, dest2, src1, src2) ({				\
135 	asm ("add.l %1,%0" : "=d,dm" (dest2)				\
136 		: "dm,d" (src2), "0,0" (dest2));			\
137 	asm ("addx.l %1,%0" : "=d" (dest1)				\
138 		: "d" (src1), "0" (dest1));				\
139 })
140 #define fp_addx96(dest, src) ({						\
141 	/* we assume here, gcc only insert move and a clr instr */	\
142 	asm volatile ("add.l %1,%0" : "=d,g" (dest->m32[2])		\
143 		: "g,d" (temp.m32[1]), "0,0" (dest->m32[2]));		\
144 	asm volatile ("addx.l %1,%0" : "=d" (dest->m32[1])		\
145 		: "d" (temp.m32[0]), "0" (dest->m32[1]));		\
146 	asm volatile ("addx.l %1,%0" : "=d" (dest->m32[0])		\
147 		: "d" (0), "0" (dest->m32[0]));				\
148 })
149 #define fp_sub64(dest, src) ({						\
150 	asm ("sub.l %1,%0" : "=d,dm" (dest.m32[1])			\
151 		: "dm,d" (src.m32[1]), "0,0" (dest.m32[1]));		\
152 	asm ("subx.l %1,%0" : "=d" (dest.m32[0])			\
153 		: "d" (src.m32[0]), "0" (dest.m32[0]));			\
154 })
155 #define fp_sub96c(dest, srch, srcm, srcl) ({				\
156 	char carry;							\
157 	asm ("sub.l %1,%0" : "=d,dm" (dest.m32[2])			\
158 		: "dm,d" (srcl), "0,0" (dest.m32[2]));			\
159 	asm ("subx.l %1,%0" : "=d" (dest.m32[1])			\
160 		: "d" (srcm), "0" (dest.m32[1]));			\
161 	asm ("subx.l %2,%1; scs %0" : "=d" (carry), "=d" (dest.m32[0])	\
162 		: "d" (srch), "1" (dest.m32[0]));			\
163 	carry;								\
164 })
165 
166 static inline void fp_multiplymant(union fp_mant128 *dest, struct fp_ext *src1,
167 				   struct fp_ext *src2)
168 {
169 	union fp_mant64 temp;
170 
171 	fp_mul64(dest->m32[0], dest->m32[1], src1->mant.m32[0], src2->mant.m32[0]);
172 	fp_mul64(dest->m32[2], dest->m32[3], src1->mant.m32[1], src2->mant.m32[1]);
173 
174 	fp_mul64(temp.m32[0], temp.m32[1], src1->mant.m32[0], src2->mant.m32[1]);
175 	fp_addx96(dest, temp);
176 
177 	fp_mul64(temp.m32[0], temp.m32[1], src1->mant.m32[1], src2->mant.m32[0]);
178 	fp_addx96(dest, temp);
179 }
180 
181 static inline void fp_dividemant(union fp_mant128 *dest, struct fp_ext *src,
182 				 struct fp_ext *div)
183 {
184 	union fp_mant128 tmp;
185 	union fp_mant64 tmp64;
186 	unsigned long *mantp = dest->m32;
187 	unsigned long fix, rem, first, dummy;
188 	int i;
189 
190 	/* the algorithm below requires dest to be smaller than div,
191 	   but both have the high bit set */
192 	if (src->mant.m64 >= div->mant.m64) {
193 		fp_sub64(src->mant, div->mant);
194 		*mantp = 1;
195 	} else
196 		*mantp = 0;
197 	mantp++;
198 
199 	/* basic idea behind this algorithm: we can't divide two 64bit numbers
200 	   (AB/CD) directly, but we can calculate AB/C0, but this means this
201 	   quotient is off by C0/CD, so we have to multiply the first result
202 	   to fix the result, after that we have nearly the correct result
203 	   and only a few corrections are needed. */
204 
205 	/* C0/CD can be precalculated, but it's an 64bit division again, but
206 	   we can make it a bit easier, by dividing first through C so we get
207 	   10/1D and now only a single shift and the value fits into 32bit. */
208 	fix = 0x80000000;
209 	dummy = div->mant.m32[1] / div->mant.m32[0] + 1;
210 	dummy = (dummy >> 1) | fix;
211 	fp_div64(fix, dummy, fix, 0, dummy);
212 	fix--;
213 
214 	for (i = 0; i < 3; i++, mantp++) {
215 		if (src->mant.m32[0] == div->mant.m32[0]) {
216 			fp_div64(first, rem, 0, src->mant.m32[1], div->mant.m32[0]);
217 
218 			fp_mul64(*mantp, dummy, first, fix);
219 			*mantp += fix;
220 		} else {
221 			fp_div64(first, rem, src->mant.m32[0], src->mant.m32[1], div->mant.m32[0]);
222 
223 			fp_mul64(*mantp, dummy, first, fix);
224 		}
225 
226 		fp_mul64(tmp.m32[0], tmp.m32[1], div->mant.m32[0], first - *mantp);
227 		fp_add64(tmp.m32[0], tmp.m32[1], 0, rem);
228 		tmp.m32[2] = 0;
229 
230 		fp_mul64(tmp64.m32[0], tmp64.m32[1], *mantp, div->mant.m32[1]);
231 		fp_sub96c(tmp, 0, tmp64.m32[0], tmp64.m32[1]);
232 
233 		src->mant.m32[0] = tmp.m32[1];
234 		src->mant.m32[1] = tmp.m32[2];
235 
236 		while (!fp_sub96c(tmp, 0, div->mant.m32[0], div->mant.m32[1])) {
237 			src->mant.m32[0] = tmp.m32[1];
238 			src->mant.m32[1] = tmp.m32[2];
239 			*mantp += 1;
240 		}
241 	}
242 }
243 
244 static inline void fp_putmant128(struct fp_ext *dest, union fp_mant128 *src,
245 				 int shift)
246 {
247 	unsigned long tmp;
248 
249 	switch (shift) {
250 	case 0:
251 		dest->mant.m64 = src->m64[0];
252 		dest->lowmant = src->m32[2] >> 24;
253 		if (src->m32[3] || (src->m32[2] << 8))
254 			dest->lowmant |= 1;
255 		break;
256 	case 1:
257 		asm volatile ("lsl.l #1,%0"
258 			: "=d" (tmp) : "0" (src->m32[2]));
259 		asm volatile ("roxl.l #1,%0"
260 			: "=d" (dest->mant.m32[1]) : "0" (src->m32[1]));
261 		asm volatile ("roxl.l #1,%0"
262 			: "=d" (dest->mant.m32[0]) : "0" (src->m32[0]));
263 		dest->lowmant = tmp >> 24;
264 		if (src->m32[3] || (tmp << 8))
265 			dest->lowmant |= 1;
266 		break;
267 	case 31:
268 		asm volatile ("lsr.l #1,%1; roxr.l #1,%0"
269 			: "=d" (dest->mant.m32[0])
270 			: "d" (src->m32[0]), "0" (src->m32[1]));
271 		asm volatile ("roxr.l #1,%0"
272 			: "=d" (dest->mant.m32[1]) : "0" (src->m32[2]));
273 		asm volatile ("roxr.l #1,%0"
274 			: "=d" (tmp) : "0" (src->m32[3]));
275 		dest->lowmant = tmp >> 24;
276 		if (src->m32[3] << 7)
277 			dest->lowmant |= 1;
278 		break;
279 	case 32:
280 		dest->mant.m32[0] = src->m32[1];
281 		dest->mant.m32[1] = src->m32[2];
282 		dest->lowmant = src->m32[3] >> 24;
283 		if (src->m32[3] << 8)
284 			dest->lowmant |= 1;
285 		break;
286 	}
287 }
288 
289 #endif	/* MULTI_ARITH_H */
290