1 /* Compute cubic root of float value.
2 Copyright (C) 1997-2019 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Dirk Alboth <dirka@uni-paderborn.de> and
5 Ulrich Drepper <drepper@cygnus.com>, 1997.
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.
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public
18 License along with the GNU C Library; if not, see
19 <http://www.gnu.org/licenses/>. */
21 #include <machine/asm.h>
22 #include <libm-alias-float.h>
28 f3: .double 0.191502161678719066
29 ASM_SIZE_DIRECTIVE(f3)
31 f2: .double 0.697570460207922770
32 ASM_SIZE_DIRECTIVE(f2)
34 f1: .double 0.492659620528969547
35 ASM_SIZE_DIRECTIVE(f1)
37 #define CBRT2 1.2599210498948731648
38 #define ONE_CBRT2 0.793700525984099737355196796584
39 #define SQR_CBRT2 1.5874010519681994748
40 #define ONE_SQR_CBRT2 0.629960524947436582364439673883
44 factor: .double ONE_SQR_CBRT2
49 ASM_SIZE_DIRECTIVE(factor)
52 two25: .byte 0, 0, 0, 0x4c
53 ASM_SIZE_DIRECTIVE(two25)
56 #define MO(op) op##@GOTOFF(%ebx)
57 #define MOX(op,x) op##@GOTOFF(%ebx,x,1)
60 #define MOX(op,x) op(x)
68 andl $0x7fffffff, %eax
70 cmpl $0x7f800000, %eax
75 cfi_adjust_cfa_offset (4)
76 cfi_rel_offset (ebx, 0)
80 cmpl $0x00800000, %eax
98 andl $0x7fffffff, %eax
101 andl $0x807fffff, %edx
103 orl $0x3f000000, %edx
108 flds 8(%esp) /* xm */
112 flds 4(%esp) /* xm */
116 /* The following code has two tracks:
117 a) compute the normalized cbrt value
118 b) compute xe/3 and xe%3
119 The right track computes the value for b) and this is done
120 in an optimized way by avoiding division.
122 But why two tracks at all? Very easy: efficiency. Some FP
123 instruction can overlap with a certain amount of integer (and
124 FP) instructions. So we get (except for the imull) all
125 instructions for free. */
127 fld %st(0) /* xm : xm */
128 fmull MO(f3) /* f3*xm : xm */
129 movl $1431655766, %eax
130 fsubrl MO(f2) /* f2-f3*xm : xm */
132 fmul %st(1) /* (f2-f3*xm)*xm : xm */
134 faddl MO(f1) /* u:=f1+(f2-f3*xm)*xm : xm */
136 fld %st /* u : u : xm */
138 fmul %st(1) /* u*u : u : xm */
139 fld %st(2) /* xm : u*u : u : xm */
140 fadd %st /* 2*xm : u*u : u : xm */
141 fxch %st(1) /* u*u : 2*xm : u : xm */
142 fmul %st(2) /* t2:=u*u*u : 2*xm : u : xm */
144 fadd %st, %st(1) /* t2 : t2+2*xm : u : xm */
145 leal (%edx,%edx,2),%edx
146 fadd %st(0) /* 2*t2 : t2+2*xm : u : xm */
148 faddp %st, %st(3) /* t2+2*xm : u : 2*t2+xm */
150 fmulp /* u*(t2+2*xm) : 2*t2+xm */
151 fdivp %st, %st(1) /* u*(t2+2*xm)/(2*t2+xm) */
152 fmull MOX(16+factor,%ecx) /* u*(t2+2*xm)/(2*t2+xm)*FACT */
154 cfi_adjust_cfa_offset (4)
155 fildl (%esp) /* xe/3 : u*(t2+2*xm)/(2*t2+xm)*FACT */
156 fxch /* u*(t2+2*xm)/(2*t2+xm)*FACT : xe/3 */
157 fscale /* u*(t2+2*xm)/(2*t2+xm)*FACT*2^xe/3 */
159 cfi_adjust_cfa_offset (-4)
163 cfi_adjust_cfa_offset (-4)
174 /* Return the argument. */
178 libm_alias_float (__cbrt, cbrt)