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
|
/*
* Copyright (C) 2020 Linux Studio Plugins Project <https://lsp-plug.in/>
* (C) 2020 Vladimir Sadovnikov <sadko4u@gmail.com>
*
* This file is part of lsp-dsp-lib
* Created on: 31 мар. 2020 г.
*
* lsp-dsp-lib 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 3 of the License, or
* any later version.
*
* lsp-dsp-lib 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 lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>.
*/
#ifndef PRIVATE_DSP_ARCH_X86_SSE_SMATH_H_
#define PRIVATE_DSP_ARCH_X86_SSE_SMATH_H_
#ifndef PRIVATE_DSP_ARCH_X86_SSE_IMPL
#error "This header should not be included directly"
#endif /* PRIVATE_DSP_ARCH_X86_SSE_IMPL */
namespace lsp
{
namespace sse
{
float ipowf(float x, int deg)
{
float res = 1.0f;
float one;
ARCH_X86_ASM
(
__ASM_EMIT("cmp $0, %[deg]")
__ASM_EMIT("jz 100f")
__ASM_EMIT("jl 10f")
__ASM_EMIT("1:")
__ASM_EMIT("test $1, %[deg]")
__ASM_EMIT("jz 2f")
__ASM_EMIT("mulss %[x], %[res]")
__ASM_EMIT("dec %[deg]")
__ASM_EMIT("jnz 1b")
__ASM_EMIT("jmp 100f")
__ASM_EMIT("2:")
__ASM_EMIT("mulss %[x], %[x]")
__ASM_EMIT("shr $1, %[deg]")
__ASM_EMIT("jnz 1b")
__ASM_EMIT("jmp 100f")
__ASM_EMIT("10:")
__ASM_EMIT("neg %[deg]")
__ASM_EMIT("movaps %[res], %[one]")
__ASM_EMIT("1:")
__ASM_EMIT("test $1, %[deg]")
__ASM_EMIT("jz 2f")
__ASM_EMIT("mulss %[x], %[res]")
__ASM_EMIT("dec %[deg]")
__ASM_EMIT("jnz 1b")
__ASM_EMIT("jmp 20f")
__ASM_EMIT("2:")
__ASM_EMIT("mulss %[x], %[x]")
__ASM_EMIT("shr $1, %[deg]")
__ASM_EMIT("jnz 1b")
__ASM_EMIT("20:")
__ASM_EMIT("divss %[res], %[one]")
__ASM_EMIT("movaps %[one], %[res]")
__ASM_EMIT("100:")
: [x] "+x" (x), [res] "+x" (res), [deg] "+r" (deg), [one] "=&x" (one)
:
: "cc"
);
return res;
}
IF_ARCH_X86(
static const uint32_t irootf_const[] __lsp_aligned16 =
{
LSP_DSP_VEC4(0x3f800000),
LSP_DSP_VEC4(0x7fffffff),
LSP_DSP_VEC4(0x3727c5ac)
};
);
float irootf(float x, int deg)
{
float a, tmp, k, sign, xp;
float tres, tx;
int tdeg;
ARCH_X86_ASM
(
__ASM_EMIT("cmp $1, %[deg]")
__ASM_EMIT("jl 100f")
// While deg is odd, just calculate sqrt function
__ASM_EMIT("test $1, %[deg]")
__ASM_EMIT("jne 2f")
__ASM_EMIT("1:")
__ASM_EMIT("shr $1, %[deg]")
__ASM_EMIT("sqrtss %[x], %[x]")
__ASM_EMIT("test $1, %[deg]")
__ASM_EMIT("jz 1b")
__ASM_EMIT("cmp $1, %[deg]")
__ASM_EMIT("jbe 100f")
// Prepare calculations
__ASM_EMIT("2:")
__ASM_EMIT("movaps 0x00(%[CC]), %[a]")
__ASM_EMIT("cvtsi2ss %[deg], %[tmp]")
__ASM_EMIT("movaps 0x10(%[CC]), %[sign]")
__ASM_EMIT("divss %[tmp], %[a]")
__ASM_EMIT("dec %[deg]")
__ASM_EMIT("cvtsi2ss %[deg], %[k]")
__ASM_EMIT("mulss %[a], %[k]")
__ASM_EMIT("mulss %[x], %[a]")
// Newton method
__ASM_EMIT("1:")
__ASM_EMIT("movaps %[x], %[xp]") // xp = x
__ASM_EMIT("movaps 0x00(%[CC]), %[tres]") // tres = 1
__ASM_EMIT("movaps %[x], %[tx]") // tx = x
__ASM_EMIT("mov %[deg], %[tdeg]") // tdeg = deg
// Calculate tres = x ^ deg
__ASM_EMIT("10:")
__ASM_EMIT("test $1, %[tdeg]") // if (tdeg & 1)
__ASM_EMIT("jz 2f")
__ASM_EMIT("mulss %[tx], %[tres]") // tres *= tx
__ASM_EMIT("dec %[tdeg]") // tdeg--
__ASM_EMIT("jnz 10b")
__ASM_EMIT("jmp 20f")
__ASM_EMIT("2:")
__ASM_EMIT("mulss %[tx], %[tx]") // tx = tx * tx
__ASM_EMIT("shr $1, %[tdeg]") // tdeg >>= 1
__ASM_EMIT("jnz 10b")
__ASM_EMIT("20:")
// Calculate x = k*x + a/tres
__ASM_EMIT("movaps %[a], %[tx]") // tx = a
__ASM_EMIT("mulss %[k], %[x]") // x = x*k
__ASM_EMIT("divss %[tres], %[tx]") // tx = a/tres
__ASM_EMIT("addss %[tx], %[x]") // x = x*k + a/tres
// Estimate tolerance
__ASM_EMIT("movaps %[x], %[tx]") // tx = x
__ASM_EMIT("subss %[x], %[xp]") // xp = xp - x
__ASM_EMIT("mulss 0x20(%[CC]), %[tx]") // tx = x * TOL
__ASM_EMIT("andps %[sign], %[x]") // xp = abs(xp - x)
__ASM_EMIT("andps %[sign], %[tx]") // xp = abs(xp - x)
__ASM_EMIT("ucomiss %[tx], %[xp]") // abs(xp - x) <> x * TOL
__ASM_EMIT("ja 1b") // Repeat loop if tolerance is not enough
__ASM_EMIT("100:")
: [x] "+x" (x), [a] "=&x" (a), [tmp] "=&x" (tmp), [k] "=&x" (k),
[sign] "=&x" (sign), [xp] "=&x" (xp), [tres] "=&x" (tres), [tx] "=&x" (tx),
[deg] "+r" (deg), [tdeg] "=&r" (tdeg)
: [CC] "r" (irootf_const)
: "cc"
);
return x;
// // Check validity of arguments
// if (deg <= 1)
// return x;
//
// // While root is odd, simple calc square root
// if (!(deg & 1))
// {
// do
// {
// x = sqrtf(x);
// deg >>= 1;
// } while (!(deg & 1));
//
// if (deg <= 1)
// return x;
// }
//
// // Newton method
// float xp;
// float a = 1.0f / float(deg);
// float k = float(--deg) * a;
// a *= x;
//
// do
// {
// xp = x;
// x = k*x + a / ipospowf(x, deg);
// } while (fabs(x - xp) > (1e-5 * x));
//
// return x;
}
}
}
#endif /* PRIVATE_DSP_ARCH_X86_SSE_SMATH_H_ */
|
