• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* IEEE754 floating point arithmetic
2  * double precision: common utilities
3  */
4 /*
5  * MIPS floating point support
6  * Copyright (C) 1994-2000 Algorithmics Ltd.
7  *
8  *  This program is free software; you can distribute it and/or modify it
9  *  under the terms of the GNU General Public License (Version 2) as
10  *  published by the Free Software Foundation.
11  *
12  *  This program is distributed in the hope it will be useful, but WITHOUT
13  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
15  *  for more details.
16  *
17  *  You should have received a copy of the GNU General Public License along
18  *  with this program; if not, write to the Free Software Foundation, Inc.,
19  *  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA.
20  */
21 
22 #include <linux/compiler.h>
23 
24 #include "ieee754dp.h"
25 
ieee754dp_class(union ieee754dp x)26 int ieee754dp_class(union ieee754dp x)
27 {
28 	COMPXDP;
29 	EXPLODEXDP;
30 	return xc;
31 }
32 
ieee754dp_isnan(union ieee754dp x)33 static inline int ieee754dp_isnan(union ieee754dp x)
34 {
35 	return ieee754_class_nan(ieee754dp_class(x));
36 }
37 
ieee754dp_issnan(union ieee754dp x)38 static inline int ieee754dp_issnan(union ieee754dp x)
39 {
40 	int qbit;
41 
42 	assert(ieee754dp_isnan(x));
43 	qbit = (DPMANT(x) & DP_MBIT(DP_FBITS - 1)) == DP_MBIT(DP_FBITS - 1);
44 	return ieee754_csr.nan2008 ^ qbit;
45 }
46 
47 
48 /*
49  * Raise the Invalid Operation IEEE 754 exception
50  * and convert the signaling NaN supplied to a quiet NaN.
51  */
ieee754dp_nanxcpt(union ieee754dp r)52 union ieee754dp __cold ieee754dp_nanxcpt(union ieee754dp r)
53 {
54 	assert(ieee754dp_issnan(r));
55 
56 	ieee754_setcx(IEEE754_INVALID_OPERATION);
57 	if (ieee754_csr.nan2008) {
58 		DPMANT(r) |= DP_MBIT(DP_FBITS - 1);
59 	} else {
60 		DPMANT(r) &= ~DP_MBIT(DP_FBITS - 1);
61 		if (!ieee754dp_isnan(r))
62 			DPMANT(r) |= DP_MBIT(DP_FBITS - 2);
63 	}
64 
65 	return r;
66 }
67 
ieee754dp_get_rounding(int sn,u64 xm)68 static u64 ieee754dp_get_rounding(int sn, u64 xm)
69 {
70 	/* inexact must round of 3 bits
71 	 */
72 	if (xm & (DP_MBIT(3) - 1)) {
73 		switch (ieee754_csr.rm) {
74 		case FPU_CSR_RZ:
75 			break;
76 		case FPU_CSR_RN:
77 			xm += 0x3 + ((xm >> 3) & 1);
78 			/* xm += (xm&0x8)?0x4:0x3 */
79 			break;
80 		case FPU_CSR_RU:	/* toward +Infinity */
81 			if (!sn)	/* ?? */
82 				xm += 0x8;
83 			break;
84 		case FPU_CSR_RD:	/* toward -Infinity */
85 			if (sn) /* ?? */
86 				xm += 0x8;
87 			break;
88 		}
89 	}
90 	return xm;
91 }
92 
93 
94 /* generate a normal/denormal number with over,under handling
95  * sn is sign
96  * xe is an unbiased exponent
97  * xm is 3bit extended precision value.
98  */
ieee754dp_format(int sn,int xe,u64 xm)99 union ieee754dp ieee754dp_format(int sn, int xe, u64 xm)
100 {
101 	assert(xm);		/* we don't gen exact zeros (probably should) */
102 
103 	assert((xm >> (DP_FBITS + 1 + 3)) == 0);	/* no excess */
104 	assert(xm & (DP_HIDDEN_BIT << 3));
105 
106 	if (xe < DP_EMIN) {
107 		/* strip lower bits */
108 		int es = DP_EMIN - xe;
109 
110 		if (ieee754_csr.nod) {
111 			ieee754_setcx(IEEE754_UNDERFLOW);
112 			ieee754_setcx(IEEE754_INEXACT);
113 
114 			switch(ieee754_csr.rm) {
115 			case FPU_CSR_RN:
116 			case FPU_CSR_RZ:
117 				return ieee754dp_zero(sn);
118 			case FPU_CSR_RU:    /* toward +Infinity */
119 				if (sn == 0)
120 					return ieee754dp_min(0);
121 				else
122 					return ieee754dp_zero(1);
123 			case FPU_CSR_RD:    /* toward -Infinity */
124 				if (sn == 0)
125 					return ieee754dp_zero(0);
126 				else
127 					return ieee754dp_min(1);
128 			}
129 		}
130 
131 		if (xe == DP_EMIN - 1 &&
132 		    ieee754dp_get_rounding(sn, xm) >> (DP_FBITS + 1 + 3))
133 		{
134 			/* Not tiny after rounding */
135 			ieee754_setcx(IEEE754_INEXACT);
136 			xm = ieee754dp_get_rounding(sn, xm);
137 			xm >>= 1;
138 			/* Clear grs bits */
139 			xm &= ~(DP_MBIT(3) - 1);
140 			xe++;
141 		}
142 		else {
143 			/* sticky right shift es bits
144 			 */
145 			xm = XDPSRS(xm, es);
146 			xe += es;
147 			assert((xm & (DP_HIDDEN_BIT << 3)) == 0);
148 			assert(xe == DP_EMIN);
149 		}
150 	}
151 	if (xm & (DP_MBIT(3) - 1)) {
152 		ieee754_setcx(IEEE754_INEXACT);
153 		if ((xm & (DP_HIDDEN_BIT << 3)) == 0) {
154 			ieee754_setcx(IEEE754_UNDERFLOW);
155 		}
156 
157 		/* inexact must round of 3 bits
158 		 */
159 		xm = ieee754dp_get_rounding(sn, xm);
160 		/* adjust exponent for rounding add overflowing
161 		 */
162 		if (xm >> (DP_FBITS + 3 + 1)) {
163 			/* add causes mantissa overflow */
164 			xm >>= 1;
165 			xe++;
166 		}
167 	}
168 	/* strip grs bits */
169 	xm >>= 3;
170 
171 	assert((xm >> (DP_FBITS + 1)) == 0);	/* no excess */
172 	assert(xe >= DP_EMIN);
173 
174 	if (xe > DP_EMAX) {
175 		ieee754_setcx(IEEE754_OVERFLOW);
176 		ieee754_setcx(IEEE754_INEXACT);
177 		/* -O can be table indexed by (rm,sn) */
178 		switch (ieee754_csr.rm) {
179 		case FPU_CSR_RN:
180 			return ieee754dp_inf(sn);
181 		case FPU_CSR_RZ:
182 			return ieee754dp_max(sn);
183 		case FPU_CSR_RU:	/* toward +Infinity */
184 			if (sn == 0)
185 				return ieee754dp_inf(0);
186 			else
187 				return ieee754dp_max(1);
188 		case FPU_CSR_RD:	/* toward -Infinity */
189 			if (sn == 0)
190 				return ieee754dp_max(0);
191 			else
192 				return ieee754dp_inf(1);
193 		}
194 	}
195 	/* gen norm/denorm/zero */
196 
197 	if ((xm & DP_HIDDEN_BIT) == 0) {
198 		/* we underflow (tiny/zero) */
199 		assert(xe == DP_EMIN);
200 		if (ieee754_csr.mx & IEEE754_UNDERFLOW)
201 			ieee754_setcx(IEEE754_UNDERFLOW);
202 		return builddp(sn, DP_EMIN - 1 + DP_EBIAS, xm);
203 	} else {
204 		assert((xm >> (DP_FBITS + 1)) == 0);	/* no excess */
205 		assert(xm & DP_HIDDEN_BIT);
206 
207 		return builddp(sn, xe + DP_EBIAS, xm & ~DP_HIDDEN_BIT);
208 	}
209 }
210