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
|
/*
* 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_IMPL
#error "This header should not be included directly"
#endif /* PRIVATE_DSP_ARCH_X86_SSE_IMPL */
#define _SSE_FFT_NAME(id) id
#define _SSE_CALL_NAME(id, bits) id ## bits
#define SSE_FFT_NAME(id) _SSE_FFT_NAME(id)
#define SSE_CALL_NAME(id, bits) _SSE_CALL_NAME(id, bits)(dst, src, rank)
namespace lsp
{
namespace sse
{
static inline void SSE_FFT_NAME(FFT_SCRAMBLE_DIRECT_NAME)(float *dst, const float *src, size_t rank)
{
// Scramble the order of samples
if (dst == src)
{
if (rank <= 8)
SSE_CALL_NAME(FFT_SCRAMBLE_SELF_DIRECT_NAME, 8);
else //if (rank <= 16)
SSE_CALL_NAME(FFT_SCRAMBLE_SELF_DIRECT_NAME, 16);
}
else
{
rank -= 3;
if (rank <= 8)
SSE_CALL_NAME(FFT_SCRAMBLE_COPY_DIRECT_NAME, 8);
else //if (rank <= 16)
SSE_CALL_NAME(FFT_SCRAMBLE_COPY_DIRECT_NAME, 16);
}
}
static inline void SSE_FFT_NAME(FFT_SCRAMBLE_REVERSE_NAME)(float *dst, const float *src, size_t rank)
{
// Scramble the order of samples
if (dst == src)
{
if (rank <= 8)
SSE_CALL_NAME(FFT_SCRAMBLE_SELF_REVERSE_NAME, 8);
else //if (rank <= 16)
SSE_CALL_NAME(FFT_SCRAMBLE_SELF_REVERSE_NAME, 16);
}
else
{
rank -= 3;
if (rank <= 8)
SSE_CALL_NAME(FFT_SCRAMBLE_COPY_REVERSE_NAME, 8);
else //if (rank <= 16)
SSE_CALL_NAME(FFT_SCRAMBLE_COPY_REVERSE_NAME, 16);
}
}
static inline void SSE_FFT_NAME(FFT_REPACK)(float *dst, size_t rank)
{
size_t blocks = 1 << (rank-3);
// Perform 4-element butterflies
ARCH_X86_ASM
(
__ASM_EMIT(".align 16")
__ASM_EMIT("1:")
/* Load data */
__ASM_EMIT("movups 0x00(%[dst]), %%xmm0") /* xmm0 = r0 r1 r2 r3 */
__ASM_EMIT("movups 0x10(%[dst]), %%xmm1") /* xmm1 = i0 i1 i2 i3 */
__ASM_EMIT("movups 0x20(%[dst]), %%xmm4") /* xmm4 = r4 r5 r6 r7 */
__ASM_EMIT("movups 0x30(%[dst]), %%xmm5") /* xmm5 = i4 i5 i6 i7 */
/* 1st step: repack pairs */
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = r0 r1 r2 r3 */
__ASM_EMIT("movaps %%xmm4, %%xmm6") /* xmm6 = r4 r5 r6 r7 */
__ASM_EMIT("unpcklps %%xmm1, %%xmm0") /* xmm0 = r0 i0 r1 i1 */
__ASM_EMIT("unpcklps %%xmm5, %%xmm4") /* xmm4 = r4 i4 r5 i5 */
__ASM_EMIT("unpckhps %%xmm1, %%xmm2") /* xmm2 = r2 i2 r3 i3 */
__ASM_EMIT("unpckhps %%xmm5, %%xmm6") /* xmm6 = r6 i6 r7 i7 */
/* Store data */
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])")
__ASM_EMIT("movups %%xmm2, 0x10(%[dst])")
__ASM_EMIT("movups %%xmm4, 0x20(%[dst])")
__ASM_EMIT("movups %%xmm6, 0x30(%[dst])")
/* Move pointers and repeat cycle */
__ASM_EMIT("add $0x40, %[dst]")
__ASM_EMIT("dec %[blocks]")
__ASM_EMIT("jnz 1b")
: [dst] "+r"(dst), [blocks] "+r"(blocks)
:
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2",
"%xmm4", "%xmm5", "%xmm6"
);
}
static inline void SSE_FFT_NAME(FFT_REPACK_NORMALIZE)(float *dst, size_t rank)
{
size_t blocks = 1 << (rank-3);
float k = 0.125f/blocks;
// Perform 4-element butterflies
ARCH_X86_ASM
(
__ASM_EMIT("shufps $0x00, %%xmm0, %%xmm0") /* xmm0 = k k k k */
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = k k k k */
__ASM_EMIT(".align 16")
__ASM_EMIT("1:")
/* Load data */
__ASM_EMIT("movups 0x00(%[dst]), %%xmm2") /* xmm2 = r0 r1 r2 r3 */
__ASM_EMIT("movups 0x10(%[dst]), %%xmm3") /* xmm3 = i0 i1 i2 i3 */
__ASM_EMIT("movups 0x20(%[dst]), %%xmm5") /* xmm5 = r4 r5 r6 r7 */
__ASM_EMIT("movups 0x30(%[dst]), %%xmm6") /* xmm6 = i4 i5 i6 i7 */
/* 1st step: repack pairs */
__ASM_EMIT("movaps %%xmm2, %%xmm4") /* xmm4 = r0 r1 r2 r3 */
__ASM_EMIT("movaps %%xmm5, %%xmm7") /* xmm7 = r4 r5 r6 r7 */
__ASM_EMIT("unpcklps %%xmm3, %%xmm2") /* xmm2 = r0 i0 r1 i1 */
__ASM_EMIT("unpcklps %%xmm6, %%xmm5") /* xmm5 = r4 i4 r5 i5 */
__ASM_EMIT("unpckhps %%xmm3, %%xmm4") /* xmm4 = r2 i2 r3 i3 */
__ASM_EMIT("unpckhps %%xmm6, %%xmm7") /* xmm7 = r6 i6 r7 i7 */
/* 3rd step: apply normalization */
__ASM_EMIT("mulps %%xmm0, %%xmm2")
__ASM_EMIT("mulps %%xmm1, %%xmm4")
__ASM_EMIT("mulps %%xmm0, %%xmm5")
__ASM_EMIT("mulps %%xmm1, %%xmm7")
/* Store data */
__ASM_EMIT("movups %%xmm2, 0x00(%[dst])")
__ASM_EMIT("movups %%xmm4, 0x10(%[dst])")
__ASM_EMIT("movups %%xmm5, 0x20(%[dst])")
__ASM_EMIT("movups %%xmm7, 0x30(%[dst])")
/* Move pointers and repeat cycle */
__ASM_EMIT("add $0x40, %[dst]")
__ASM_EMIT("dec %[blocks]")
__ASM_EMIT("jnz 1b")
: [dst] "+r"(dst), [blocks] "+r"(blocks), "+Yz"(k)
:
: "cc", "memory",
"%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
}
}
#undef SSE_FFT_NAME
#undef SSE_CALL_NAME
#undef _SSE_FFT_NAME
#undef _SSE_CALL_NAME
#undef FFT_SCRAMBLE_SELF_DIRECT_NAME
#undef FFT_SCRAMBLE_COPY_DIRECT_NAME
#undef FFT_SCRAMBLE_SELF_REVERSE_NAME
#undef FFT_SCRAMBLE_COPY_REVERSE_NAME
#undef LS_RE
#undef FFT_SCRAMBLE_DIRECT_NAME
#undef FFT_SCRAMBLE_REVERSE_NAME
#undef FFT_REPACK
#undef FFT_REPACK_NORMALIZED
#undef FFT_MODE
|
