about summary refs log tree commit diff
path: root/sysdeps/i386/fpu/s_cbrtl.S
blob: fb7b447b0f46c7a231fa519abe143e478dd2d186 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
/* Compute cubic root of long double value.
   Copyright (C) 1997-2023 Free Software Foundation, Inc.
   This file is part of the GNU C Library.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library; if not, see
   <https://www.gnu.org/licenses/>.  */

#include <libm-alias-ldouble.h>
#include <machine/asm.h>

        .section .rodata

        .align ALIGNARG(4)
        .type f8,@object
f8:	.quad   0xa57ef3d83a542839  /* 0.161617097923756032  */
	.short  0x3ffc
	ASM_SIZE_DIRECTIVE(f8)
        .align ALIGNARG(4)
        .type f7,@object
f7:	.quad   0xfd11da7820029014  /* -0.988553671195413709  */
	.short  0xbffe
	ASM_SIZE_DIRECTIVE(f7)
        .align ALIGNARG(4)
        .type f6,@object
f6:	.quad   0xa9ca93fcade3b4ad  /* 2.65298938441952296  */
	.short  0x4000
	ASM_SIZE_DIRECTIVE(f6)
        .align ALIGNARG(4)
        .type f5,@object
f5:	.quad   0x839186562c931c34  /* -4.11151425200350531  */
	.short  0xc001
	ASM_SIZE_DIRECTIVE(f5)
        .align ALIGNARG(4)
        .type f4,@object
f4:	.quad   0x830f25c9ee304594  /* 4.09559907378707839  */
	.short  0x4001
	ASM_SIZE_DIRECTIVE(f4)
        .align ALIGNARG(4)
        .type f3,@object
f3:	.quad   0xb4bedd1d5fa2f0c6  /* -2.82414939754975962  */
	.short  0xc000
	ASM_SIZE_DIRECTIVE(f3)
        .align ALIGNARG(4)
        .type f2,@object
f2:	.quad   0xd685a163b08586e3  /* 1.67595307700780102  */
	.short  0x3fff
	ASM_SIZE_DIRECTIVE(f2)
        .align ALIGNARG(4)
        .type f1,@object
f1:	.quad   0xad16073ed4ec3b45  /* 0.338058687610520237  */
	.short  0x3ffd
	ASM_SIZE_DIRECTIVE(f1)

	/* We make the entries in the following table all 16 bytes
	   wide to avoid having to implement a multiplication by 10.  */
	.type factor,@object
        .align ALIGNARG(4)
factor:	/* 1.0 / cbrt (2.0) ^ 2 = 0.629960524947436582364439673883  */
	.quad 0xa14517cc6b945711
	.short 0x3ffe
	.byte 0, 0, 0, 0, 0, 0
	/* 1.0 / cbrt (2.0) = 0.793700525984099737355196796584  */
	.quad 0xcb2ff529eb71e415
	.short 0x3ffe
	.byte 0, 0, 0, 0, 0, 0
	/* 1.0L  */
	.quad 0x8000000000000000
	.short 0x3fff
	.byte 0, 0, 0, 0, 0, 0
	/* cbrt (2.0) = 1.2599210498948731648  */
	.quad 0xa14517cc6b945711
	.short 0x3fff
	.byte 0, 0, 0, 0, 0, 0
	/* cbrt (2.0) ^ 2 = 1.5874010519681994748  */
	.quad 0xcb2ff529eb71e416
	.short 0x3fff
	ASM_SIZE_DIRECTIVE(factor)

        .type two64,@object
        .align ALIGNARG(4)
two64:  .byte 0, 0, 0, 0, 0, 0, 0xf0, 0x43
        ASM_SIZE_DIRECTIVE(two64)

#ifdef PIC
#define MO(op) op##@GOTOFF(%ebx)
#define MOX(op,x) op##@GOTOFF(%ebx,x,1)
#else
#define MO(op) op
#define MOX(op,x) op(x)
#endif

	.text
ENTRY(__cbrtl)
	movl	4(%esp), %ecx
	movl	12(%esp), %eax
	orl	8(%esp), %ecx
	movl	%eax, %edx
	andl	$0x7fff, %eax
	orl	%eax, %ecx
	jz	1f
	xorl	%ecx, %ecx
	cmpl	$0x7fff, %eax
	je	1f

#ifdef PIC
	pushl	%ebx
	cfi_adjust_cfa_offset (4)
	cfi_rel_offset (ebx, 0)
	LOAD_PIC_REG (bx)
#endif

	cmpl	$0, %eax
	jne	2f

#ifdef PIC
	fldt	8(%esp)
#else
	fldt	4(%esp)
#endif
	fmull	MO(two64)
	movl	$-64, %ecx
#ifdef PIC
	fstpt	8(%esp)
	movl	16(%esp), %eax
#else
	fstpt	4(%esp)
	movl	12(%esp), %eax
#endif
	movl	%eax, %edx
	andl	$0x7fff, %eax

2:	andl	$0x8000, %edx
	subl	$16382, %eax
	orl	$0x3ffe, %edx
	addl	%eax, %ecx
#ifdef PIC
	movl	%edx, 16(%esp)

	fldt	8(%esp)			/* xm */
#else
	movl	%edx, 12(%esp)

	fldt	4(%esp)			/* xm */
#endif
	fabs

	/* The following code has two tracks:
	    a) compute the normalized cbrt value
	    b) compute xe/3 and xe%3
	   The right track computes the value for b) and this is done
	   in an optimized way by avoiding division.

	   But why two tracks at all?  Very easy: efficiency.  Some FP
	   instruction can overlap with a certain amount of integer (and
	   FP) instructions.  So we get (except for the imull) all
	   instructions for free.  */

	fldt	MO(f8)			/* f8 : xm */
	fmul	%st(1)			/* f8*xm : xm */

	fldt	MO(f7)
	faddp				/* f7+f8*xm : xm */
	fmul	%st(1)			/* (f7+f8*xm)*xm : xm */
			movl	$1431655766, %eax
	fldt	MO(f6)
	faddp				/* f6+(f7+f8*xm)*xm : xm */
			imull	%ecx
	fmul	%st(1)			/* (f6+(f7+f8*xm)*xm)*xm : xm */
			movl	%ecx, %eax
	fldt	MO(f5)
	faddp				/* f5+(f6+(f7+f8*xm)*xm)*xm : xm */
			sarl	$31, %eax
	fmul	%st(1)			/* (f5+(f6+(f7+f8*xm)*xm)*xm)*xm : xm */
			subl	%eax, %edx
	fldt	MO(f4)
	faddp				/* f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm : xm */
	fmul	%st(1)			/* (f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm : xm */
	fldt	MO(f3)
	faddp				/* f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm : xm */
	fmul	%st(1)			/* (f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm)*xm : xm */
	fldt	MO(f2)
	faddp				/* f2+(f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm)*xm : xm */
	fmul	%st(1)			/* (f2+(f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm)*xm)*xm : xm */
	fldt	MO(f1)
	faddp				/* u:=f1+(f2+(f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm)*xm)*xm : xm */

	fld	%st			/* u : u : xm */
	fmul	%st(1)			/* u*u : u : xm */
	fld	%st(2)			/* xm : u*u : u : xm */
	fadd	%st			/* 2*xm : u*u : u : xm */
	fxch	%st(1)			/* u*u : 2*xm : u : xm */
	fmul	%st(2)			/* t2:=u*u*u : 2*xm : u : xm */
			movl	%edx, %eax
	fadd	%st, %st(1)		/* t2 : t2+2*xm : u : xm */
			leal	(%edx,%edx,2),%edx
	fadd	%st(0)			/* 2*t2 : t2+2*xm : u : xm */
			subl	%edx, %ecx
	faddp	%st, %st(3)		/* t2+2*xm : u : 2*t2+xm */
			shll	$4, %ecx
	fmulp				/* u*(t2+2*xm) : 2*t2+xm */
	fdivp	%st, %st(1)		/* u*(t2+2*xm)/(2*t2+xm) */
	fldt	MOX(32+factor,%ecx)
	fmulp				/* u*(t2+2*xm)/(2*t2+xm)*FACT */
	pushl	%eax
	cfi_adjust_cfa_offset (4)
	fildl	(%esp)			/* xe/3 : u*(t2+2*xm)/(2*t2+xm)*FACT */
	fxch				/* u*(t2+2*xm)/(2*t2+xm)*FACT : xe/3 */
	fscale				/* u*(t2+2*xm)/(2*t2+xm)*FACT*2^xe/3 */
	popl	%edx
	cfi_adjust_cfa_offset (-4)
#ifdef PIC
	movl	16(%esp), %eax
	popl	%ebx
	cfi_adjust_cfa_offset (-4)
	cfi_restore (ebx)
#else
	movl	12(%esp), %eax
#endif
	testl	$0x8000, %eax
	fstp	%st(1)
	jz	4f
	fchs
4:	ret

	/* Return the argument.  */
1:	fldt	4(%esp)
	fadd	%st
	ret
END(__cbrtl)
libm_alias_ldouble (__cbrt, cbrt)