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
|
/*
* 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_AVX_FFT_P_REPACK_H_
#define PRIVATE_DSP_ARCH_X86_AVX_FFT_P_REPACK_H_
#ifndef PRIVATE_DSP_ARCH_X86_AVX_IMPL
#error "This header should not be included directly"
#endif /* PRIVATE_DSP_ARCH_X86_AVX_IMPL */
namespace lsp
{
namespace avx
{
static inline void packed_fft_repack(float *dst, size_t rank)
{
size_t blocks = 1 << rank;
// Perform 4-element butterflies
ARCH_X86_ASM
(
// 16x blocks
__ASM_EMIT("sub $16, %[blocks]")
__ASM_EMIT("jb 2f")
__ASM_EMIT("1:")
__ASM_EMIT("vmovups 0x00(%[dst]), %%ymm0") /* ymm0 = r0 r1 r2 r3 r4 r5 r6 r7 */
__ASM_EMIT("vmovups 0x20(%[dst]), %%ymm1") /* ymm1 = i0 i1 i2 i3 i4 i5 i6 i7 */
__ASM_EMIT("vmovups 0x40(%[dst]), %%ymm2") /* ymm2 = r8 r9 r10 r11 r12 r13 r14 r15 */
__ASM_EMIT("vmovups 0x60(%[dst]), %%ymm3") /* ymm3 = i8 i9 i10 i11 i12 i13 i14 i15 */
__ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0 i0 r1 i1 r4 i4 r5 i5 */
__ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm5") /* ymm5 = r2 i2 r3 i3 r6 i6 r7 i7 */
__ASM_EMIT("vunpcklps %%ymm3, %%ymm2, %%ymm0") /* ymm0 = r8 i8 r9 i9 r12 i12 r13 i13 */
__ASM_EMIT("vunpckhps %%ymm3, %%ymm2, %%ymm1") /* ymm1 = r10 i10 r12 i12 r14 i14 r15 i15 */
__ASM_EMIT("vmovups %%xmm4, 0x00(%[dst])")
__ASM_EMIT("vmovups %%xmm5, 0x10(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm4, 0x20(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm5, 0x30(%[dst])")
__ASM_EMIT("vmovups %%xmm0, 0x40(%[dst])")
__ASM_EMIT("vmovups %%xmm1, 0x50(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm0, 0x60(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm1, 0x70(%[dst])")
__ASM_EMIT("add $0x80, %[dst]")
__ASM_EMIT("sub $16, %[blocks]")
__ASM_EMIT("jae 1b")
__ASM_EMIT("2:")
// 8x block
__ASM_EMIT("add $8, %[blocks]")
__ASM_EMIT("jl 4f")
__ASM_EMIT("vmovups 0x00(%[dst]), %%ymm0") /* ymm0 = r0 r1 r2 r3 r4 r5 r6 r7 */
__ASM_EMIT("vmovups 0x20(%[dst]), %%ymm1") /* ymm1 = i0 i1 i2 i3 i4 i5 i6 i7 */
__ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0 i0 r1 i1 r4 i4 r5 i5 */
__ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm5") /* ymm5 = r2 i2 r3 i3 r6 i6 r7 i7 */
__ASM_EMIT("vmovups %%xmm4, 0x00(%[dst])")
__ASM_EMIT("vmovups %%xmm5, 0x10(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm4, 0x20(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm5, 0x30(%[dst])")
__ASM_EMIT("4:")
: [dst] "+r"(dst), [blocks] "+r" (blocks)
:
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5"
);
}
static inline void packed_fft_repack_normalize(float *dst, size_t rank)
{
size_t blocks = 1 << rank;
float norm = 1.0f / float(blocks);
// Perform 4-element butterflies
ARCH_X86_ASM
(
__ASM_EMIT("vbroadcastss %[norm], %%ymm6")
__ASM_EMIT("vmovaps %%ymm6, %%ymm7")
// 16x blocks
__ASM_EMIT("sub $16, %[blocks]")
__ASM_EMIT("jb 2f")
__ASM_EMIT("1:")
__ASM_EMIT("vmulps 0x00(%[dst]), %%ymm6, %%ymm0") /* ymm0 = r0 r1 r2 r3 r4 r5 r6 r7 */
__ASM_EMIT("vmulps 0x20(%[dst]), %%ymm7, %%ymm1") /* ymm1 = i0 i1 i2 i3 i4 i5 i6 i7 */
__ASM_EMIT("vmulps 0x40(%[dst]), %%ymm6, %%ymm2") /* ymm2 = r8 r9 r10 r11 r12 r13 r14 r15 */
__ASM_EMIT("vmulps 0x60(%[dst]), %%ymm7, %%ymm3") /* ymm3 = i8 i9 i10 i11 i12 i13 i14 i15 */
__ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0 i0 r1 i1 r4 i4 r5 i5 */
__ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm5") /* ymm5 = r2 i2 r3 i3 r6 i6 r7 i7 */
__ASM_EMIT("vunpcklps %%ymm3, %%ymm2, %%ymm0") /* ymm0 = r8 i8 r9 i9 r12 i12 r13 i13 */
__ASM_EMIT("vunpckhps %%ymm3, %%ymm2, %%ymm1") /* ymm1 = r10 i10 r12 i12 r14 i14 r15 i15 */
__ASM_EMIT("vmovups %%xmm4, 0x00(%[dst])")
__ASM_EMIT("vmovups %%xmm5, 0x10(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm4, 0x20(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm5, 0x30(%[dst])")
__ASM_EMIT("vmovups %%xmm0, 0x40(%[dst])")
__ASM_EMIT("vmovups %%xmm1, 0x50(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm0, 0x60(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm1, 0x70(%[dst])")
__ASM_EMIT("add $0x80, %[dst]")
__ASM_EMIT("sub $16, %[blocks]")
__ASM_EMIT("jae 1b")
__ASM_EMIT("2:")
// 8x block
__ASM_EMIT("add $8, %[blocks]")
__ASM_EMIT("jl 4f")
__ASM_EMIT("vmulps 0x00(%[dst]), %%ymm6, %%ymm0") /* ymm0 = r0 r1 r2 r3 r4 r5 r6 r7 */
__ASM_EMIT("vmulps 0x20(%[dst]), %%ymm7, %%ymm1") /* ymm1 = i0 i1 i2 i3 i4 i5 i6 i7 */
__ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0 i0 r1 i1 r4 i4 r5 i5 */
__ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm5") /* ymm5 = r2 i2 r3 i3 r6 i6 r7 i7 */
__ASM_EMIT("vmovups %%xmm4, 0x00(%[dst])")
__ASM_EMIT("vmovups %%xmm5, 0x10(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm4, 0x20(%[dst])")
__ASM_EMIT("vextractf128 $1, %%ymm5, 0x30(%[dst])")
__ASM_EMIT("4:")
: [dst] "+r"(dst), [blocks] "+r" (blocks)
: [norm] "m" (norm)
: "cc", "memory",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7"
);
}
}
}
#endif /* PRIVATE_DSP_ARCH_X86_AVX_FFT_P_REPACK_H_ */
|
