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
|
/*
* 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_BUTTERFLY_H_
#define PRIVATE_DSP_ARCH_X86_AVX_FFT_BUTTERFLY_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
{
#define FFT_BUTTERFLY_BODY8(add_b, add_a, FMA_SEL) \
__IF_32(float *ptr1, *ptr2);\
\
ARCH_X86_ASM \
( \
/* Prepare angle */ \
__ASM_EMIT32("mov %[fft_a], %[ptr2]") \
__ASM_EMIT32("mov %[dst_re], %[ptr1]") \
__ASM_EMIT("vmovaps 0x00(%[" __IF_32_64("ptr2", "fft_a") "]), %%ymm6") /* ymm6 = x_re */ \
__ASM_EMIT("vmovaps 0x20(%[" __IF_32_64("ptr2", "fft_a") "]), %%ymm7") /* ymm7 = x_im */ \
__ASM_EMIT32("mov %[dst_im], %[ptr2]") \
/* Start loop */ \
__ASM_EMIT("1:") \
__ASM_EMIT("vmovups 0x00(%[" __IF_32_64("ptr1", "dst_re") "], %[off1]), %%ymm0") /* ymm0 = a_re */ \
__ASM_EMIT("vmovups 0x00(%[" __IF_32_64("ptr1", "dst_re") "], %[off2]), %%ymm2") /* ymm2 = b_re */ \
__ASM_EMIT("vmovups 0x00(%[" __IF_32_64("ptr2", "dst_im") "], %[off1]), %%ymm1") /* ymm1 = a_im */ \
__ASM_EMIT("vmovups 0x00(%[" __IF_32_64("ptr2", "dst_im") "], %[off2]), %%ymm3") /* ymm3 = b_im */ \
/* Calculate complex multiplication */ \
__ASM_EMIT("vmulps %%ymm7, %%ymm2, %%ymm4") /* ymm4 = x_im * b_re */ \
__ASM_EMIT("vmulps %%ymm7, %%ymm3, %%ymm5") /* ymm5 = x_im * b_im */ \
__ASM_EMIT(FMA_SEL("vmulps %%ymm6, %%ymm2, %%ymm2", "")) /* ymm2 = x_re * b_re */ \
__ASM_EMIT(FMA_SEL("vmulps %%ymm6, %%ymm3, %%ymm3", "")) /* ymm3 = x_re * b_im */ \
__ASM_EMIT(FMA_SEL(add_b " %%ymm5, %%ymm2, %%ymm5", add_b " %%ymm6, %%ymm2, %%ymm5")) /* ymm5 = c_re = x_re * b_re +- x_im * b_im */ \
__ASM_EMIT(FMA_SEL(add_a " %%ymm4, %%ymm3, %%ymm4", add_a " %%ymm6, %%ymm3, %%ymm4")) /* ymm4 = c_im = x_re * b_im -+ x_im * b_re */ \
/* Perform butterfly */ \
__ASM_EMIT("vsubps %%ymm5, %%ymm0, %%ymm2") /* ymm2 = a_re - c_re */ \
__ASM_EMIT("vsubps %%ymm4, %%ymm1, %%ymm3") /* ymm3 = a_im - c_im */ \
__ASM_EMIT("vaddps %%ymm5, %%ymm0, %%ymm0") /* ymm0 = a_re + c_re */ \
__ASM_EMIT("vaddps %%ymm4, %%ymm1, %%ymm1") /* ymm1 = a_im + c_im */ \
/* Store values */ \
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" __IF_32_64("ptr1", "dst_re") "], %[off1])") \
__ASM_EMIT("vmovups %%ymm2, 0x00(%[" __IF_32_64("ptr1", "dst_re") "], %[off2])") \
__ASM_EMIT("vmovups %%ymm1, 0x00(%[" __IF_32_64("ptr2", "dst_im") "], %[off1])") \
__ASM_EMIT("vmovups %%ymm3, 0x00(%[" __IF_32_64("ptr2", "dst_im") "], %[off2])") \
__ASM_EMIT("add $0x20, %[off1]") \
__ASM_EMIT("add $0x20, %[off2]") \
__ASM_EMIT32("subl $8, %[np]") \
__ASM_EMIT64("subq $8, %[np]") \
__ASM_EMIT("jz 2f") \
/* Rotate angle */ \
__ASM_EMIT32("mov %[fft_w], %[ptr2]") \
__ASM_EMIT("vmovaps 0x00(%[" __IF_32_64("ptr2", "fft_w") "]), %%ymm4") /* xmm4 = w_re */ \
__ASM_EMIT("vmovaps 0x20(%[" __IF_32_64("ptr2", "fft_w") "]), %%ymm5") /* xmm5 = w_im */ \
__ASM_EMIT32("mov %[dst_im], %[ptr2]") \
__ASM_EMIT("vmulps %%ymm5, %%ymm6, %%ymm2") /* ymm2 = w_im * x_re */ \
__ASM_EMIT("vmulps %%ymm5, %%ymm7, %%ymm3") /* ymm3 = w_im * x_im */ \
__ASM_EMIT(FMA_SEL("vmulps %%ymm4, %%ymm6, %%ymm6", "")) /* ymm6 = w_re * x_re */ \
__ASM_EMIT(FMA_SEL("vmulps %%ymm4, %%ymm7, %%ymm7", "")) /* ymm7 = w_re * x_im */ \
__ASM_EMIT(FMA_SEL("vsubps %%ymm3, %%ymm6, %%ymm6", "vfmsub132ps %%ymm4, %%ymm3, %%ymm6")) /* ymm6 = x_re' = w_re * x_re - w_im * x_im */ \
__ASM_EMIT(FMA_SEL("vaddps %%ymm2, %%ymm7, %%ymm7", "vfmadd132ps %%ymm4, %%ymm2, %%ymm7")) /* ymm7 = x_im' = w_re * x_im + w_im * x_re */ \
/* Repeat loop */ \
__ASM_EMIT("jmp 1b") \
__ASM_EMIT("2:") \
\
: __IF_32([ptr1] "=&r" (ptr1), [ptr2] "=&r" (ptr2), ) \
[off1] "+r" (off1), [off2] "+r" (off2), \
[np] X86_PGREG (np) \
: [dst_re] X86_GREG (dst_re), [dst_im] X86_GREG (dst_im), [fft_a] X86_GREG (fft_a), [fft_w] X86_GREG (fft_w) \
: "cc", "memory", \
"%xmm0", "%xmm1", "%xmm2", "%xmm3", \
"%xmm4", "%xmm5", "%xmm6", "%xmm7" \
);
#define FMA_OFF(a, b) a
#define FMA_ON(a, b) b
static inline void butterfly_direct8p(float *dst_re, float *dst_im, size_t rank, size_t blocks)
{
size_t pairs = 1 << rank;
size_t off1 = 0, shift = 4 << rank; //1 << (rank + 2);
const float *fft_a = &FFT_A[(rank - 2) << 4];
const float *fft_w = &FFT_DW[(rank - 2) << 4];
for (size_t b=0; b<blocks; ++b)
{
size_t off2 = off1 + shift;
size_t np = pairs;
FFT_BUTTERFLY_BODY8("vaddps", "vsubps", FMA_OFF);
off1 = off2;
}
}
static inline void butterfly_reverse8p(float *dst_re, float *dst_im, size_t rank, size_t blocks)
{
size_t pairs = 1 << rank;
size_t off1 = 0, shift = 4 << rank; // 1 << (rank + 2);
const float *fft_a = &FFT_A[(rank - 2) << 4];
const float *fft_w = &FFT_DW[(rank - 2) << 4];
for (size_t b=0; b<blocks; ++b)
{
size_t off2 = off1 + shift;
size_t np = pairs;
FFT_BUTTERFLY_BODY8("vsubps", "vaddps", FMA_OFF);
off1 = off2;
}
}
static inline void butterfly_direct8p_fma3(float *dst_re, float *dst_im, size_t rank, size_t blocks)
{
size_t pairs = 1 << rank;
size_t off1 = 0, shift = 4 << rank; // 1 << (rank + 2);
const float *fft_a = &FFT_A[(rank - 2) << 4];
const float *fft_w = &FFT_DW[(rank - 2) << 4];
for (size_t b=0; b<blocks; ++b)
{
size_t off2 = off1 + shift;
size_t np = pairs;
FFT_BUTTERFLY_BODY8("vfmadd231ps", "vfmsub231ps", FMA_ON);
off1 = off2;
}
}
static inline void butterfly_reverse8p_fma3(float *dst_re, float *dst_im, size_t rank, size_t blocks)
{
size_t pairs = 1 << rank;
size_t off1 = 0, shift = 4 << rank; // 1 << (rank + 2);
const float *fft_a = &FFT_A[(rank - 2) << 4];
const float *fft_w = &FFT_DW[(rank - 2) << 4];
for (size_t b=0; b<blocks; ++b)
{
size_t off2 = off1 + shift;
size_t np = pairs;
FFT_BUTTERFLY_BODY8("vfmsub231ps", "vfmadd231ps", FMA_ON);
off1 = off2;
}
}
#undef FMA_OFF
#undef FMA_ON
#undef FFT_BUTTERFLY_BODY8
}
}
#endif /* PRIVATE_DSP_ARCH_X86_AVX_FFT_BUTTERFLY_H_ */
|
