1 /* Software floating-point emulation.
2    Definitions for IEEE Quad Precision on the PowerPC.
3    Copyright (C) 2016-2018 Free Software Foundation, Inc.
4    This file is part of the GNU C Library.
5    Contributed by Michael Meissner (meissner@linux.vnet.ibm.com).
6 
7    The GNU C Library is free software; you can redistribute it and/or
8    modify it under the terms of the GNU Lesser General Public
9    License as published by the Free Software Foundation; either
10    version 2.1 of the License, or (at your option) any later version.
11 
12    In addition to the permissions in the GNU Lesser General Public
13    License, the Free Software Foundation gives you unlimited
14    permission to link the compiled version of this file into
15    combinations with other programs, and to distribute those
16    combinations without any restriction coming from the use of this
17    file.  (The Lesser General Public License restrictions do apply in
18    other respects; for example, they cover modification of the file,
19    and distribution when not linked into a combine executable.)
20 
21    The GNU C Library is distributed in the hope that it will be useful,
22    but WITHOUT ANY WARRANTY; without even the implied warranty of
23    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
24    Lesser General Public License for more details.
25 
26    You should have received a copy of the GNU Lesser General Public
27    License along with the GNU C Library; if not, see
28    <http://www.gnu.org/licenses/>.  */
29 
30 /* quad.h defines the TFtype type by:
31    typedef float TFtype __attribute__ ((mode (TF)));
32 
33    This define forces it to use KFmode (aka, ieee 128-bit floating point).
34    However, when the compiler's default is changed so that long double is IEEE
35    128-bit floating point, we need to go back to using TFmode and TCmode.  */
36 #ifndef __LONG_DOUBLE_IEEE128__
37 #define TF KF
38 
39 /* We also need TCtype to represent complex ieee 128-bit float for
40    __mulkc3 and __divkc3.  */
41 typedef __complex float TCtype __attribute__ ((mode (KC)));
42 
43 #else
44 typedef __complex float TCtype __attribute__ ((mode (TC)));
45 #endif
46 
47 /* Force the use of the VSX instruction set.  */
48 #if defined(_ARCH_PPC) && (!defined(__VSX__) || !defined(__FLOAT128__))
49 #pragma GCC target ("vsx,float128")
50 #endif
51 
52 #include <quad.h>
53 
54 #define IBM128_TYPE	__ibm128
55 
56 /* Add prototypes of the library functions created.  In case the appropriate
57    int/long types are not declared in scope by the time quad.h is included,
58    provide our own version.  */
59 typedef int	 SItype_ppc  __attribute__ ((__mode__ (__SI__)));
60 typedef int	 DItype_ppc  __attribute__ ((__mode__ (__DI__)));
61 typedef unsigned USItype_ppc __attribute__ ((__mode__ (__SI__)));
62 typedef unsigned UDItype_ppc __attribute__ ((__mode__ (__DI__)));
63 
64 #ifdef _ARCH_PPC64
65 typedef int	 TItype_ppc  __attribute__ ((__mode__ (__TI__)));
66 typedef unsigned UTItype_ppc __attribute__ ((__mode__ (__TI__)));
67 #endif
68 
69 /* Software emulation functions.  */
70 extern TFtype __addkf3_sw (TFtype, TFtype);
71 extern TFtype __subkf3_sw (TFtype, TFtype);
72 extern TFtype __mulkf3_sw (TFtype, TFtype);
73 extern TFtype __divkf3_sw (TFtype, TFtype);
74 extern TFtype __negkf2_sw (TFtype);
75 extern TFtype __powikf2_sw (TFtype, SItype_ppc);
76 extern CMPtype __eqkf2_sw (TFtype, TFtype);
77 extern CMPtype __gekf2_sw (TFtype, TFtype);
78 extern CMPtype __lekf2_sw (TFtype, TFtype);
79 extern CMPtype __unordkf2_sw (TFtype, TFtype);
80 extern TFtype __extendsfkf2_sw (float);
81 extern TFtype __extenddfkf2_sw (double);
82 extern float __trunckfsf2_sw (TFtype);
83 extern double __trunckfdf2_sw (TFtype);
84 extern SItype_ppc __fixkfsi_sw (TFtype);
85 extern DItype_ppc __fixkfdi_sw (TFtype);
86 extern USItype_ppc __fixunskfsi_sw (TFtype);
87 extern UDItype_ppc __fixunskfdi_sw (TFtype);
88 extern TFtype __floatsikf_sw (SItype_ppc);
89 extern TFtype __floatdikf_sw (DItype_ppc);
90 extern TFtype __floatunsikf_sw (USItype_ppc);
91 extern TFtype __floatundikf_sw (UDItype_ppc);
92 extern IBM128_TYPE __extendkftf2_sw (TFtype);
93 extern TFtype __trunctfkf2_sw (IBM128_TYPE);
94 extern TCtype __mulkc3_sw (TFtype, TFtype, TFtype, TFtype);
95 extern TCtype __divkc3_sw (TFtype, TFtype, TFtype, TFtype);
96 
97 #ifdef _ARCH_PPC64
98 /* We do not provide ifunc resolvers for __fixkfti, __fixunskfti, __floattikf,
99    and __floatuntikf.  There is no ISA 3.0 instruction that converts between
100    128-bit integer types and 128-bit IEEE floating point, or vice versa.  So
101    use the emulator functions for these conversions.  */
102 
103 extern TItype_ppc __fixkfti (TFtype);
104 extern UTItype_ppc __fixunskfti (TFtype);
105 extern TFtype __floattikf (TItype_ppc);
106 extern TFtype __floatuntikf (UTItype_ppc);
107 #endif
108 
109 /* Functions using the ISA 3.0 hardware support.  If the code is compiled with
110    -mcpu=power9, it will not use these functions, but if it was compiled with
111    -mcpu=power7 or -mcpu=power8 and run on a ISA 3.0 system, it will use the
112    hardware instruction.  */
113 extern TFtype __addkf3_hw (TFtype, TFtype);
114 extern TFtype __subkf3_hw (TFtype, TFtype);
115 extern TFtype __mulkf3_hw (TFtype, TFtype);
116 extern TFtype __divkf3_hw (TFtype, TFtype);
117 extern TFtype __negkf2_hw (TFtype);
118 extern TFtype __powikf2_hw (TFtype, SItype_ppc);
119 extern CMPtype __eqkf2_hw (TFtype, TFtype);
120 extern CMPtype __gekf2_hw (TFtype, TFtype);
121 extern CMPtype __lekf2_hw (TFtype, TFtype);
122 extern CMPtype __unordkf2_hw (TFtype, TFtype);
123 extern TFtype __extendsfkf2_hw (float);
124 extern TFtype __extenddfkf2_hw (double);
125 extern float __trunckfsf2_hw (TFtype);
126 extern double __trunckfdf2_hw (TFtype);
127 extern SItype_ppc __fixkfsi_hw (TFtype);
128 extern DItype_ppc __fixkfdi_hw (TFtype);
129 extern USItype_ppc __fixunskfsi_hw (TFtype);
130 extern UDItype_ppc __fixunskfdi_hw (TFtype);
131 extern TFtype __floatsikf_hw (SItype_ppc);
132 extern TFtype __floatdikf_hw (DItype_ppc);
133 extern TFtype __floatunsikf_hw (USItype_ppc);
134 extern TFtype __floatundikf_hw (UDItype_ppc);
135 extern IBM128_TYPE __extendkftf2_hw (TFtype);
136 extern TFtype __trunctfkf2_hw (IBM128_TYPE);
137 extern TCtype __mulkc3_hw (TFtype, TFtype, TFtype, TFtype);
138 extern TCtype __divkc3_hw (TFtype, TFtype, TFtype, TFtype);
139 
140 /* Ifunc function declarations, to automatically switch between software
141    emulation and hardware support.  */
142 extern TFtype __addkf3 (TFtype, TFtype);
143 extern TFtype __subkf3 (TFtype, TFtype);
144 extern TFtype __mulkf3 (TFtype, TFtype);
145 extern TFtype __divkf3 (TFtype, TFtype);
146 extern TFtype __negkf2 (TFtype);
147 extern TFtype __powikf2 (TFtype, SItype_ppc);
148 extern CMPtype __eqkf2 (TFtype, TFtype);
149 extern CMPtype __nekf2 (TFtype, TFtype);
150 extern CMPtype __gekf2 (TFtype, TFtype);
151 extern CMPtype __gtkf2 (TFtype, TFtype);
152 extern CMPtype __lekf2 (TFtype, TFtype);
153 extern CMPtype __ltkf2 (TFtype, TFtype);
154 extern CMPtype __unordkf2 (TFtype, TFtype);
155 extern TFtype __extendsfkf2 (float);
156 extern TFtype __extenddfkf2 (double);
157 extern float __trunckfsf2 (TFtype);
158 extern double __trunckfdf2 (TFtype);
159 extern SItype_ppc __fixkfsi (TFtype);
160 extern DItype_ppc __fixkfdi (TFtype);
161 extern USItype_ppc __fixunskfsi (TFtype);
162 extern UDItype_ppc __fixunskfdi (TFtype);
163 extern TFtype __floatsikf (SItype_ppc);
164 extern TFtype __floatdikf (DItype_ppc);
165 extern TFtype __floatunsikf (USItype_ppc);
166 extern TFtype __floatundikf (UDItype_ppc);
167 extern IBM128_TYPE __extendkftf2 (TFtype);
168 extern TFtype __trunctfkf2 (IBM128_TYPE);
169 
170 /* Complex __float128 built on __float128 interfaces.  */
171 extern TCtype __mulkc3 (TFtype, TFtype, TFtype, TFtype);
172 extern TCtype __divkc3 (TFtype, TFtype, TFtype, TFtype);
173 
174 /* Implementation of conversions between __ibm128 and __float128, to allow the
175    same code to be used on systems with IEEE 128-bit emulation and with IEEE
176    128-bit hardware support.  */
177 
178 union ibm128_union {
179   IBM128_TYPE ibm128;
180   double dbl[2];
181 };
182 
183 #define CVT_FLOAT128_TO_IBM128(RESULT, VALUE)				\
184 {									\
185   double __high, __low;							\
186   TFtype __value = (VALUE);						\
187   union ibm128_union u;							\
188 									\
189   __high = (double) __value;						\
190   if (__builtin_isnan (__high) || __builtin_isinf (__high))		\
191     __low = 0.0;							\
192 									\
193   else									\
194     {									\
195       double __high_temp;						\
196 									\
197       __low = (double) (__value - (TFtype) __high);			\
198       /* Renormalize low/high and move them into canonical IBM long	\
199 	 double form.  */						\
200       __high_temp = __high + __low;					\
201       __low = (__high - __high_temp) + __low;				\
202       __high = __high_temp;						\
203     }									\
204 									\
205   u.dbl[0] = __high;							\
206   u.dbl[1] = __low;							\
207   RESULT = u.ibm128;							\
208 }
209 
210 #define CVT_IBM128_TO_FLOAT128(RESULT, VALUE)				\
211 {									\
212   union ibm128_union u;							\
213   double __high, __low;							\
214 									\
215   u.ibm128 = (VALUE);							\
216   __high = u.dbl[0];							\
217   __low = u.dbl[1];							\
218 									\
219   /* Handle the special cases of NAN and infinity.  */			\
220   if (__builtin_isnan (__high) || __builtin_isinf (__high))		\
221     RESULT = (TFtype) __high;						\
222 									\
223   /* If low is 0.0, there no need to do the add.  In addition,		\
224      avoiding the add produces the correct sign if high is -0.0.  */	\
225   else if (__low == 0.0)						\
226     RESULT = (TFtype) __high;						\
227 									\
228   else									\
229     RESULT = ((TFtype) __high) + ((TFtype) __low);			\
230 }
231