1 | /* Configure soft-fp for building sqrtf128. Based on sfp-machine.h in |
2 | libgcc, with soft-float and other irrelevant parts removed. */ |
3 | |
4 | /* The type of the result of a floating point comparison. This must |
5 | match `__libgcc_cmp_return__' in GCC for the target. */ |
6 | typedef int __gcc_CMPtype __attribute__ ((mode (__libgcc_cmp_return__))); |
7 | #define CMPtype __gcc_CMPtype |
8 | |
9 | #ifdef __x86_64__ |
10 | # define _FP_W_TYPE_SIZE 64 |
11 | # define _FP_W_TYPE unsigned long long |
12 | # define _FP_WS_TYPE signed long long |
13 | # define _FP_I_TYPE long long |
14 | |
15 | typedef int TItype __attribute__ ((mode (TI))); |
16 | typedef unsigned int UTItype __attribute__ ((mode (TI))); |
17 | |
18 | # define TI_BITS (__CHAR_BIT__ * (int) sizeof (TItype)) |
19 | |
20 | # define _FP_MUL_MEAT_Q(R,X,Y) \ |
21 | _FP_MUL_MEAT_2_wide(_FP_WFRACBITS_Q,R,X,Y,umul_ppmm) |
22 | |
23 | # define _FP_DIV_MEAT_Q(R,X,Y) _FP_DIV_MEAT_2_udiv(Q,R,X,Y) |
24 | |
25 | # define _FP_NANFRAC_S _FP_QNANBIT_S |
26 | # define _FP_NANFRAC_D _FP_QNANBIT_D |
27 | # define _FP_NANFRAC_E _FP_QNANBIT_E, 0 |
28 | # define _FP_NANFRAC_Q _FP_QNANBIT_Q, 0 |
29 | |
30 | # define FP_EX_SHIFT 7 |
31 | |
32 | # define _FP_DECL_EX \ |
33 | unsigned int _fcw __attribute__ ((unused)) = FP_RND_NEAREST; |
34 | |
35 | # define FP_RND_NEAREST 0 |
36 | # define FP_RND_ZERO 0x6000 |
37 | # define FP_RND_PINF 0x4000 |
38 | # define FP_RND_MINF 0x2000 |
39 | |
40 | # define FP_RND_MASK 0x6000 |
41 | |
42 | # define FP_INIT_ROUNDMODE \ |
43 | do { \ |
44 | __asm__ __volatile__ ("%vstmxcsr\t%0" : "=m" (_fcw)); \ |
45 | } while (0) |
46 | #else |
47 | # define _FP_W_TYPE_SIZE 32 |
48 | # define _FP_W_TYPE unsigned int |
49 | # define _FP_WS_TYPE signed int |
50 | # define _FP_I_TYPE int |
51 | |
52 | # define __FP_FRAC_ADD_4(r3,r2,r1,r0,x3,x2,x1,x0,y3,y2,y1,y0) \ |
53 | __asm__ ("add{l} {%11,%3|%3,%11}\n\t" \ |
54 | "adc{l} {%9,%2|%2,%9}\n\t" \ |
55 | "adc{l} {%7,%1|%1,%7}\n\t" \ |
56 | "adc{l} {%5,%0|%0,%5}" \ |
57 | : "=r" ((USItype) (r3)), \ |
58 | "=&r" ((USItype) (r2)), \ |
59 | "=&r" ((USItype) (r1)), \ |
60 | "=&r" ((USItype) (r0)) \ |
61 | : "%0" ((USItype) (x3)), \ |
62 | "g" ((USItype) (y3)), \ |
63 | "%1" ((USItype) (x2)), \ |
64 | "g" ((USItype) (y2)), \ |
65 | "%2" ((USItype) (x1)), \ |
66 | "g" ((USItype) (y1)), \ |
67 | "%3" ((USItype) (x0)), \ |
68 | "g" ((USItype) (y0))) |
69 | # define __FP_FRAC_ADD_3(r2,r1,r0,x2,x1,x0,y2,y1,y0) \ |
70 | __asm__ ("add{l} {%8,%2|%2,%8}\n\t" \ |
71 | "adc{l} {%6,%1|%1,%6}\n\t" \ |
72 | "adc{l} {%4,%0|%0,%4}" \ |
73 | : "=r" ((USItype) (r2)), \ |
74 | "=&r" ((USItype) (r1)), \ |
75 | "=&r" ((USItype) (r0)) \ |
76 | : "%0" ((USItype) (x2)), \ |
77 | "g" ((USItype) (y2)), \ |
78 | "%1" ((USItype) (x1)), \ |
79 | "g" ((USItype) (y1)), \ |
80 | "%2" ((USItype) (x0)), \ |
81 | "g" ((USItype) (y0))) |
82 | # define __FP_FRAC_SUB_4(r3,r2,r1,r0,x3,x2,x1,x0,y3,y2,y1,y0) \ |
83 | __asm__ ("sub{l} {%11,%3|%3,%11}\n\t" \ |
84 | "sbb{l} {%9,%2|%2,%9}\n\t" \ |
85 | "sbb{l} {%7,%1|%1,%7}\n\t" \ |
86 | "sbb{l} {%5,%0|%0,%5}" \ |
87 | : "=r" ((USItype) (r3)), \ |
88 | "=&r" ((USItype) (r2)), \ |
89 | "=&r" ((USItype) (r1)), \ |
90 | "=&r" ((USItype) (r0)) \ |
91 | : "0" ((USItype) (x3)), \ |
92 | "g" ((USItype) (y3)), \ |
93 | "1" ((USItype) (x2)), \ |
94 | "g" ((USItype) (y2)), \ |
95 | "2" ((USItype) (x1)), \ |
96 | "g" ((USItype) (y1)), \ |
97 | "3" ((USItype) (x0)), \ |
98 | "g" ((USItype) (y0))) |
99 | # define __FP_FRAC_SUB_3(r2,r1,r0,x2,x1,x0,y2,y1,y0) \ |
100 | __asm__ ("sub{l} {%8,%2|%2,%8}\n\t" \ |
101 | "sbb{l} {%6,%1|%1,%6}\n\t" \ |
102 | "sbb{l} {%4,%0|%0,%4}" \ |
103 | : "=r" ((USItype) (r2)), \ |
104 | "=&r" ((USItype) (r1)), \ |
105 | "=&r" ((USItype) (r0)) \ |
106 | : "0" ((USItype) (x2)), \ |
107 | "g" ((USItype) (y2)), \ |
108 | "1" ((USItype) (x1)), \ |
109 | "g" ((USItype) (y1)), \ |
110 | "2" ((USItype) (x0)), \ |
111 | "g" ((USItype) (y0))) |
112 | # define __FP_FRAC_ADDI_4(x3,x2,x1,x0,i) \ |
113 | __asm__ ("add{l} {%4,%3|%3,%4}\n\t" \ |
114 | "adc{l} {$0,%2|%2,0}\n\t" \ |
115 | "adc{l} {$0,%1|%1,0}\n\t" \ |
116 | "adc{l} {$0,%0|%0,0}" \ |
117 | : "+r" ((USItype) (x3)), \ |
118 | "+&r" ((USItype) (x2)), \ |
119 | "+&r" ((USItype) (x1)), \ |
120 | "+&r" ((USItype) (x0)) \ |
121 | : "g" ((USItype) (i))) |
122 | |
123 | |
124 | # define _FP_MUL_MEAT_S(R,X,Y) \ |
125 | _FP_MUL_MEAT_1_wide(_FP_WFRACBITS_S,R,X,Y,umul_ppmm) |
126 | # define _FP_MUL_MEAT_D(R,X,Y) \ |
127 | _FP_MUL_MEAT_2_wide(_FP_WFRACBITS_D,R,X,Y,umul_ppmm) |
128 | # define _FP_MUL_MEAT_Q(R,X,Y) \ |
129 | _FP_MUL_MEAT_4_wide(_FP_WFRACBITS_Q,R,X,Y,umul_ppmm) |
130 | |
131 | # define _FP_DIV_MEAT_S(R,X,Y) _FP_DIV_MEAT_1_loop(S,R,X,Y) |
132 | # define _FP_DIV_MEAT_D(R,X,Y) _FP_DIV_MEAT_2_udiv(D,R,X,Y) |
133 | # define _FP_DIV_MEAT_Q(R,X,Y) _FP_DIV_MEAT_4_udiv(Q,R,X,Y) |
134 | |
135 | # define _FP_NANFRAC_S _FP_QNANBIT_S |
136 | # define _FP_NANFRAC_D _FP_QNANBIT_D, 0 |
137 | /* Even if XFmode is 12byte, we have to pad it to |
138 | 16byte since soft-fp emulation is done in 16byte. */ |
139 | # define _FP_NANFRAC_E _FP_QNANBIT_E, 0, 0, 0 |
140 | # define _FP_NANFRAC_Q _FP_QNANBIT_Q, 0, 0, 0 |
141 | |
142 | # define FP_EX_SHIFT 0 |
143 | |
144 | # define _FP_DECL_EX \ |
145 | unsigned short _fcw __attribute__ ((unused)) = FP_RND_NEAREST; |
146 | |
147 | # define FP_RND_NEAREST 0 |
148 | # define FP_RND_ZERO 0xc00 |
149 | # define FP_RND_PINF 0x800 |
150 | # define FP_RND_MINF 0x400 |
151 | |
152 | # define FP_RND_MASK 0xc00 |
153 | |
154 | # define FP_INIT_ROUNDMODE \ |
155 | do { \ |
156 | __asm__ __volatile__ ("fnstcw\t%0" : "=m" (_fcw)); \ |
157 | } while (0) |
158 | #endif |
159 | |
160 | #define _FP_KEEPNANFRACP 1 |
161 | #define _FP_QNANNEGATEDP 0 |
162 | |
163 | #define _FP_NANSIGN_S 1 |
164 | #define _FP_NANSIGN_D 1 |
165 | #define _FP_NANSIGN_E 1 |
166 | #define _FP_NANSIGN_Q 1 |
167 | |
168 | /* Here is something Intel misdesigned: the specs don't define |
169 | the case where we have two NaNs with same mantissas, but |
170 | different sign. Different operations pick up different NaNs. */ |
171 | #define _FP_CHOOSENAN(fs, wc, R, X, Y, OP) \ |
172 | do { \ |
173 | if (_FP_FRAC_GT_##wc(X, Y) \ |
174 | || (_FP_FRAC_EQ_##wc(X,Y) && (OP == '+' || OP == '*'))) \ |
175 | { \ |
176 | R##_s = X##_s; \ |
177 | _FP_FRAC_COPY_##wc(R,X); \ |
178 | } \ |
179 | else \ |
180 | { \ |
181 | R##_s = Y##_s; \ |
182 | _FP_FRAC_COPY_##wc(R,Y); \ |
183 | } \ |
184 | R##_c = FP_CLS_NAN; \ |
185 | } while (0) |
186 | |
187 | #define FP_EX_INVALID 0x01 |
188 | #define FP_EX_DENORM 0x02 |
189 | #define FP_EX_DIVZERO 0x04 |
190 | #define FP_EX_OVERFLOW 0x08 |
191 | #define FP_EX_UNDERFLOW 0x10 |
192 | #define FP_EX_INEXACT 0x20 |
193 | #define FP_EX_ALL \ |
194 | (FP_EX_INVALID | FP_EX_DENORM | FP_EX_DIVZERO | FP_EX_OVERFLOW \ |
195 | | FP_EX_UNDERFLOW | FP_EX_INEXACT) |
196 | |
197 | void __sfp_handle_exceptions (int); |
198 | |
199 | #define FP_HANDLE_EXCEPTIONS \ |
200 | do { \ |
201 | if (__builtin_expect (_fex, 0)) \ |
202 | __sfp_handle_exceptions (_fex); \ |
203 | } while (0); |
204 | |
205 | #define FP_TRAPPING_EXCEPTIONS ((~_fcw >> FP_EX_SHIFT) & FP_EX_ALL) |
206 | |
207 | #define FP_ROUNDMODE (_fcw & FP_RND_MASK) |
208 | |
209 | #define _FP_TININESS_AFTER_ROUNDING 1 |
210 | |