Debian Codegraph @

modules/lsp-dsp-lib/include/private/dsp/arch/x86/avx/fastconv/apply.h

raw
1 /*
2 * Copyright (C) 2020 Linux Studio Plugins Project <https://lsp-plug.in/>
3 * (C) 2020 Vladimir Sadovnikov <[email protected]>
4 *
5 * This file is part of lsp-dsp-lib
6 * Created on: 31 мар. 2020 г.
7 *
8 * lsp-dsp-lib is free software: you can redistribute it and/or modify
9 * it under the terms of the GNU Lesser General Public License as published by
10 * the Free Software Foundation, either version 3 of the License, or
11 * any later version.
12 *
13 * lsp-dsp-lib is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public License
19 * along with lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>.
20 */
21
22 #ifndef PRIVATE_DSP_ARCH_X86_AVX_FASTCONV_APPLY_H_
23 #define PRIVATE_DSP_ARCH_X86_AVX_FASTCONV_APPLY_H_
24
25 #ifndef PRIVATE_DSP_ARCH_X86_AVX_IMPL
26 #error "This header should not be included directly"
27 #endif /* PRIVATE_DSP_ARCH_X86_AVX_IMPL */
28
29 namespace lsp
30 {
31 namespace avx
32 {
33 #define FMA_OFF(a, b) a
34 #define FMA_ON(a, b) b
35 #define FMA_SEL1(a, b) a
36
37 #define FASTCONV_APPLY_PREPARE_CORE(FMA_SEL) \
38 size_t off; \
39 ARCH_X86_ASM( \
40 /* 2x blocks of 4x-butterfly loop */ \
41 __ASM_EMIT("xor %[off], %[off]") \
42 __ASM_EMIT("sub $2, %[nb]") \
43 __ASM_EMIT("jb 2f") \
44 __ASM_EMIT("1:") \
45 /* Load data and apply convolution*/ \
46 __ASM_EMIT("vmovups 0x00(%[c1], %[off]), %%ymm0") /* ymm0 = r0 = r0 r4 */ \
47 __ASM_EMIT("vmovups 0x20(%[c1], %[off]), %%ymm2") /* ymm2 = i0 = i0 i4 */ \
48 __ASM_EMIT("vmovups 0x40(%[c1], %[off]), %%ymm4") /* xmm4 = r4 = r8 r12 */ \
49 __ASM_EMIT("vmovups 0x60(%[c1], %[off]), %%ymm6") /* xmm6 = i4 = i8 i12 */ \
50 /* Apply convolution */ \
51 __ASM_EMIT("vmulps 0x20(%[c2], %[off]), %%ymm0, %%ymm1") /* ymm1 = r0*I */ \
52 __ASM_EMIT("vmulps 0x60(%[c2], %[off]), %%ymm4, %%ymm5") /* ymm1 = r4*I */ \
53 __ASM_EMIT("vmulps 0x20(%[c2], %[off]), %%ymm2, %%ymm3") /* ymm3 = i0*I */ \
54 __ASM_EMIT("vmulps 0x60(%[c2], %[off]), %%ymm6, %%ymm7") /* ymm3 = i4*I */ \
55 __ASM_EMIT(FMA_SEL("vmulps 0x00(%[c2], %[off]), %%ymm0, %%ymm0", "")) /* ymm0 = r0*R */ \
56 __ASM_EMIT(FMA_SEL("vmulps 0x40(%[c2], %[off]), %%ymm4, %%ymm4", "")) /* ymm4 = r4*R */ \
57 __ASM_EMIT(FMA_SEL("vmulps 0x00(%[c2], %[off]), %%ymm2, %%ymm2", "")) /* ymm2 = i0*R */ \
58 __ASM_EMIT(FMA_SEL("vmulps 0x40(%[c2], %[off]), %%ymm6, %%ymm6", "")) /* ymm6 = i4*R */ \
59 __ASM_EMIT(FMA_SEL("vsubps %%ymm3, %%ymm0, %%ymm0", "vfmsub132ps 0x00(%[c2], %[off]), %%ymm3, %%ymm0")) /* ymm0 = r0*R - i0*I */ \
60 __ASM_EMIT(FMA_SEL("vsubps %%ymm7, %%ymm4, %%ymm4", "vfmsub132ps 0x40(%[c2], %[off]), %%ymm7, %%ymm4")) /* ymm4 = r4*R - i4*I */ \
61 __ASM_EMIT(FMA_SEL("vaddps %%ymm1, %%ymm2, %%ymm2", "vfmadd132ps 0x00(%[c2], %[off]), %%ymm1, %%ymm2")) /* ymm2 = r0*I + i0*R */ \
62 __ASM_EMIT(FMA_SEL("vaddps %%ymm5, %%ymm6, %%ymm6", "vfmadd132ps 0x40(%[c2], %[off]), %%ymm5, %%ymm6")) /* ymm2 = r0*I + i0*R */ \
63 /* Reshuffle */ \
64 __ASM_EMIT("vextractf128 $1, %%ymm0, %%xmm5") /* xmm5 = r4 */ \
65 __ASM_EMIT("vextractf128 $1, %%ymm2, %%xmm7") /* xmm7 = i4 */ \
66 __ASM_EMIT("vinsertf128 $1, %%xmm4, %%ymm0, %%ymm0") /* ymm0 = r0 r8 */ \
67 __ASM_EMIT("vinsertf128 $1, %%xmm6, %%ymm2, %%ymm2") /* ymm2 = i0 i8 */ \
68 __ASM_EMIT("vinsertf128 $0, %%xmm5, %%ymm4, %%ymm4") /* ymm4 = r4 r12 */ \
69 __ASM_EMIT("vinsertf128 $0, %%xmm7, %%ymm6, %%ymm6") /* ymm6 = i4 i12 */ \
70 /* 1st-order 4x butterfly */ \
71 __ASM_EMIT("vhsubps %%ymm4, %%ymm0, %%ymm1") /* ymm1 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7' */ \
72 __ASM_EMIT("vhsubps %%ymm6, %%ymm2, %%ymm3") /* ymm3 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7' */ \
73 __ASM_EMIT("vhaddps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6' */ \
74 __ASM_EMIT("vhaddps %%ymm6, %%ymm2, %%ymm2") /* ymm2 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6' */ \
75 /* 2nd-order 4x butterfly */ \
76 __ASM_EMIT("vblendps $0xaa, %%ymm3, %%ymm1, %%ymm4") /* ymm4 = r1' i3' r5' i7' */ \
77 __ASM_EMIT("vblendps $0xaa, %%ymm1, %%ymm3, %%ymm5") /* ymm5 = i1' r3' i5' r7' */ \
78 __ASM_EMIT("vhsubps %%ymm4, %%ymm0, %%ymm1") /* ymm1 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r1" r5" */ \
79 __ASM_EMIT("vhsubps %%ymm5, %%ymm2, %%ymm3") /* ymm3 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i3" i7" */ \
80 __ASM_EMIT("vhaddps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r3" r7" */ \
81 __ASM_EMIT("vhaddps %%ymm5, %%ymm2, %%ymm2") /* ymm2 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i1" i5" */ \
82 __ASM_EMIT("vblendps $0xcc, %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0" i4" r1" r5" */ \
83 __ASM_EMIT("vblendps $0xcc, %%ymm0, %%ymm1, %%ymm5") /* ymm5 = r2" r6" r3" r7" */ \
84 __ASM_EMIT("vshufps $0x88, %%ymm3, %%ymm2, %%ymm6") /* ymm6 = i0" i1" i2" i3" */ \
85 __ASM_EMIT("vshufps $0xdd, %%ymm3, %%ymm2, %%ymm7") /* ymm7 = i4" i5" i6" i7" */ \
86 __ASM_EMIT("vshufps $0x88, %%ymm5, %%ymm4, %%ymm2") /* ymm2 = r0" r1" r2" r3" */ \
87 __ASM_EMIT("vshufps $0xdd, %%ymm5, %%ymm4, %%ymm3") /* ymm3 = r4" r5" r6" r7" */ \
88 /* 3rd-order 8x butterfly */ \
89 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%ymm3, %%ymm4") /* ymm4 = x_im * b_re */ \
90 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%ymm7, %%ymm5") /* ymm5 = x_im * b_im */ \
91 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%ymm3, %%ymm3", "")) /* ymm3 = x_re * b_re */ \
92 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%ymm7, %%ymm7", "")) /* ymm7 = x_re * b_im */ \
93 __ASM_EMIT(FMA_SEL("vsubps %%ymm5, %%ymm3, %%ymm5", "vfmsub231ps 0x00 + %[FFT_A], %%ymm3, %%ymm5")) /* ymm5 = c_re = x_re * b_re - x_im * b_im */ \
94 __ASM_EMIT(FMA_SEL("vaddps %%ymm4, %%ymm7, %%ymm4", "vfmadd231ps 0x00 + %[FFT_A], %%ymm7, %%ymm4")) /* ymm4 = c_im = x_re * b_im + x_im * b_re */ \
95 __ASM_EMIT("vsubps %%ymm5, %%ymm2, %%ymm0") /* ymm0 = a_re - c_re */ \
96 __ASM_EMIT("vsubps %%ymm4, %%ymm6, %%ymm1") /* ymm1 = a_im - c_im */ \
97 __ASM_EMIT("vaddps %%ymm5, %%ymm2, %%ymm2") /* ymm2 = a_re + c_re */ \
98 __ASM_EMIT("vaddps %%ymm4, %%ymm6, %%ymm3") /* ymm3 = a_im + c_im */ \
99 /* Store */ \
100 __ASM_EMIT("vmovups %%xmm2, 0x00(%[dst], %[off])") \
101 __ASM_EMIT("vmovups %%xmm0, 0x10(%[dst], %[off])") \
102 __ASM_EMIT("vmovups %%xmm3, 0x20(%[dst], %[off])") \
103 __ASM_EMIT("vmovups %%xmm1, 0x30(%[dst], %[off])") \
104 __ASM_EMIT("vextractf128 $1, %%ymm2, 0x40(%[dst], %[off])") \
105 __ASM_EMIT("vextractf128 $1, %%ymm0, 0x50(%[dst], %[off])") \
106 __ASM_EMIT("vextractf128 $1, %%ymm3, 0x60(%[dst], %[off])") \
107 __ASM_EMIT("vextractf128 $1, %%ymm1, 0x70(%[dst], %[off])") \
108 __ASM_EMIT("add $0x80, %[off]") \
109 __ASM_EMIT("sub $2, %[nb]") \
110 __ASM_EMIT("jae 1b") \
111 /* 1x block of 4-butterfly */ \
112 __ASM_EMIT("2:") \
113 __ASM_EMIT("add $1, %[nb]") \
114 __ASM_EMIT("jl 4f") \
115 __ASM_EMIT("vmovups 0x00(%[c1], %[off]), %%ymm0") /* ymm0 = r0 = r0 r4 */ \
116 __ASM_EMIT("vmovups 0x20(%[c1], %[off]), %%ymm2") /* ymm2 = i0 = i0 i4 */ \
117 /* Apply convolution */ \
118 __ASM_EMIT("vmulps 0x20(%[c2], %[off]), %%ymm0, %%ymm1") /* ymm1 = r0*I */ \
119 __ASM_EMIT("vmulps 0x20(%[c2], %[off]), %%ymm2, %%ymm3") /* ymm3 = i0*I */ \
120 __ASM_EMIT(FMA_SEL("vmulps 0x00(%[c2], %[off]), %%ymm0, %%ymm0", "")) /* ymm0 = r0*R */ \
121 __ASM_EMIT(FMA_SEL("vmulps 0x00(%[c2], %[off]), %%ymm2, %%ymm2", "")) /* ymm2 = i0*R */ \
122 __ASM_EMIT(FMA_SEL("vsubps %%ymm3, %%ymm0, %%ymm0", "vfmsub132ps 0x00(%[c2], %[off]), %%ymm3, %%ymm0")) /* ymm0 = r0*R - i0*I */ \
123 __ASM_EMIT(FMA_SEL("vaddps %%ymm1, %%ymm2, %%ymm2", "vfmadd132ps 0x00(%[c2], %[off]), %%ymm1, %%ymm2")) /* ymm2 = r0*I + i0*R */ \
124 /* Reshuffle */ \
125 __ASM_EMIT("vextractf128 $1, %%ymm0, %%xmm4") /* xmm4 = r4 */ \
126 __ASM_EMIT("vextractf128 $1, %%ymm2, %%xmm6") /* xmm6 = i4 */ \
127 /* 1st-order 4x butterfly */ \
128 __ASM_EMIT("vhsubps %%xmm4, %%xmm0, %%xmm1") /* xmm1 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7' */ \
129 __ASM_EMIT("vhsubps %%xmm6, %%xmm2, %%xmm3") /* xmm3 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7' */ \
130 __ASM_EMIT("vhaddps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6' */ \
131 __ASM_EMIT("vhaddps %%xmm6, %%xmm2, %%xmm2") /* xmm2 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6' */ \
132 /* 2nd-order 4x butterfly */ \
133 __ASM_EMIT("vblendps $0xaa, %%xmm3, %%xmm1, %%xmm4") /* xmm4 = r1' i3' r5' i7' */ \
134 __ASM_EMIT("vblendps $0xaa, %%xmm1, %%xmm3, %%xmm5") /* xmm5 = i1' r3' i5' r7' */ \
135 __ASM_EMIT("vhsubps %%xmm4, %%xmm0, %%xmm1") /* xmm1 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r1" r5" */ \
136 __ASM_EMIT("vhsubps %%xmm5, %%xmm2, %%xmm3") /* xmm3 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i3" i7" */ \
137 __ASM_EMIT("vhaddps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r3" r7" */ \
138 __ASM_EMIT("vhaddps %%xmm5, %%xmm2, %%xmm2") /* xmm2 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i1" i5" */ \
139 __ASM_EMIT("vblendps $0xcc, %%xmm1, %%xmm0, %%xmm4") /* xmm4 = r0" i4" r1" r5" */ \
140 __ASM_EMIT("vblendps $0xcc, %%xmm0, %%xmm1, %%xmm5") /* xmm5 = r2" r6" r3" r7" */ \
141 __ASM_EMIT("vshufps $0x88, %%xmm3, %%xmm2, %%xmm6") /* xmm6 = i0" i1" i2" i3" */ \
142 __ASM_EMIT("vshufps $0xdd, %%xmm3, %%xmm2, %%xmm7") /* xmm7 = i4" i5" i6" i7" */ \
143 __ASM_EMIT("vshufps $0x88, %%xmm5, %%xmm4, %%xmm2") /* xmm2 = r0" r1" r2" r3" */ \
144 __ASM_EMIT("vshufps $0xdd, %%xmm5, %%xmm4, %%xmm3") /* xmm3 = r4" r5" r6" r7" */ \
145 /* 3rd-order 8x butterfly */ \
146 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%xmm3, %%xmm4") /* xmm4 = x_im * b_re */ \
147 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%xmm7, %%xmm5") /* xmm5 = x_im * b_im */ \
148 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%xmm3, %%xmm3", "")) /* xmm3 = x_re * b_re */ \
149 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%xmm7, %%xmm7", "")) /* xmm7 = x_re * b_im */ \
150 __ASM_EMIT(FMA_SEL("vsubps %%xmm5, %%xmm3, %%xmm5", "vfmsub231ps 0x00 + %[FFT_A], %%xmm3, %%xmm5")) /* xmm5 = c_re = x_re * b_re - x_im * b_im */ \
151 __ASM_EMIT(FMA_SEL("vaddps %%xmm4, %%xmm7, %%xmm4", "vfmadd231ps 0x00 + %[FFT_A], %%xmm7, %%xmm4")) /* xmm4 = c_im = x_re * b_im + x_im * b_re */ \
152 __ASM_EMIT("vsubps %%xmm5, %%xmm2, %%xmm0") /* xmm0 = a_re - c_re */ \
153 __ASM_EMIT("vsubps %%xmm4, %%xmm6, %%xmm1") /* xmm1 = a_im - c_im */ \
154 __ASM_EMIT("vaddps %%xmm5, %%xmm2, %%xmm2") /* xmm2 = a_re + c_re */ \
155 __ASM_EMIT("vaddps %%xmm4, %%xmm6, %%xmm3") /* xmm3 = a_im + c_im */ \
156 /* Store */ \
157 __ASM_EMIT("vmovups %%xmm2, 0x00(%[dst], %[off])") \
158 __ASM_EMIT("vmovups %%xmm0, 0x10(%[dst], %[off])") \
159 __ASM_EMIT("vmovups %%xmm3, 0x20(%[dst], %[off])") \
160 __ASM_EMIT("vmovups %%xmm1, 0x30(%[dst], %[off])") \
161 __ASM_EMIT("4:") \
162 : [off] "=&r" (off), [nb] "+r" (nb) \
163 : [dst] "r" (dst), [c1] "r" (c1), [c2] "r" (c2), \
164 [FFT_A] "o" (FFT_A) \
165 : "cc", "memory", \
166 "%xmm0", "%xmm1", "%xmm2", "%xmm3", \
167 "%xmm4", "%xmm5", "%xmm6", "%xmm7" \
168 )
169
170 #define FASTCONV_APPLY_CORE(FMA_SEL) \
171 size_t off; \
172 ARCH_X86_ASM( \
173 /* 2x blocks of 4x-butterfly loop */ \
174 __ASM_EMIT("xor %[off], %[off]") \
175 __ASM_EMIT("sub $2, %[nb]") \
176 __ASM_EMIT("jb 2f") \
177 __ASM_EMIT("1:") \
178 /* Load data to registers */ \
179 __ASM_EMIT("vmovups 0x00(%[dst], %[off]), %%xmm0") /* xmm0 = r0 r1 r2 r3 */ \
180 __ASM_EMIT("vmovups 0x10(%[dst], %[off]), %%xmm2") /* xmm2 = r4 r5 r6 r7 */ \
181 __ASM_EMIT("vmovups 0x20(%[dst], %[off]), %%xmm1") /* xmm1 = i0 i1 i2 i3 */ \
182 __ASM_EMIT("vmovups 0x30(%[dst], %[off]), %%xmm3") /* xmm3 = i4 i5 i6 i7 */ \
183 __ASM_EMIT("vinsertf128 $1, 0x40(%[dst], %[off]), %%ymm0, %%ymm0") /* ymm0 = a_re = r0 r1 r2 r3 r8 r9 r10 r11 */ \
184 __ASM_EMIT("vinsertf128 $1, 0x50(%[dst], %[off]), %%ymm2, %%ymm2") /* ymm2 = b_re = r4 r5 r6 r7 r12 r13 r14 r15 */ \
185 __ASM_EMIT("vinsertf128 $1, 0x60(%[dst], %[off]), %%ymm1, %%ymm1") /* ymm1 = a_im = i0 i1 i2 i3 i8 i9 i10 i11 */ \
186 __ASM_EMIT("vinsertf128 $1, 0x70(%[dst], %[off]), %%ymm3, %%ymm3") /* ymm3 = b_im = i4 i5 i6 i7 i12 i13 i14 i15 */ \
187 /* Perform 3rd-order butterflies */ \
188 __ASM_EMIT("vsubps %%ymm2, %%ymm0, %%ymm4") /* ymm4 = c_re = a_re - b_re */ \
189 __ASM_EMIT("vsubps %%ymm3, %%ymm1, %%ymm5") /* ymm5 = c_im = a_im - b_im */ \
190 __ASM_EMIT("vaddps %%ymm2, %%ymm0, %%ymm0") /* ymm0 = a_re' = a_re + b_re */ \
191 __ASM_EMIT("vaddps %%ymm3, %%ymm1, %%ymm1") /* ymm1 = a_im' = a_im + b_im */ \
192 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%ymm4, %%ymm2") /* ymm2 = x_im * c_re */ \
193 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%ymm5, %%ymm3") /* ymm3 = x_im * c_im */ \
194 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%ymm4, %%ymm4", "")) /* ymm4 = x_re * c_re */ \
195 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%ymm5, %%ymm5", "")) /* ymm5 = x_re * c_im */ \
196 __ASM_EMIT(FMA_SEL("vaddps %%ymm3, %%ymm4, %%ymm4", "vfmadd132ps 0x00 + %[FFT_A], %%ymm3, %%ymm4")) /* ymm4 = b_re = x_re * c_re + x_im * c_im */ \
197 __ASM_EMIT(FMA_SEL("vsubps %%ymm2, %%ymm5, %%ymm5", "vfmsub132ps 0x00 + %[FFT_A], %%ymm2, %%ymm5")) /* ymm5 = b_im = x_re * c_im - x_im * c_re */ \
198 /* 2nd-order butterflies */ \
199 __ASM_EMIT("vshufps $0xd8, %%ymm0, %%ymm0, %%ymm0") /* ymm0 = r0 r2 r1 r3 */ \
200 __ASM_EMIT("vshufps $0xd8, %%ymm1, %%ymm1, %%ymm1") /* ymm1 = i0 i2 i1 i3 */ \
201 __ASM_EMIT("vshufps $0xd8, %%ymm4, %%ymm4, %%ymm4") \
202 __ASM_EMIT("vshufps $0xd8, %%ymm5, %%ymm5, %%ymm5") \
203 __ASM_EMIT("vhsubps %%ymm1, %%ymm0, %%ymm2") /* ymm2 = r0-r2 r1-r3 i0-i2 i1-i3 = r1' r3' i1' i3' */ \
204 __ASM_EMIT("vhsubps %%ymm5, %%ymm4, %%ymm3") \
205 __ASM_EMIT("vhaddps %%ymm1, %%ymm0, %%ymm0") /* ymm0 = r0+r2 r1+r3 i0+i2 i1+i3 = r0' r2' i0' i2' */ \
206 __ASM_EMIT("vhaddps %%ymm5, %%ymm4, %%ymm4") \
207 /* 1st-order 8x butterfly */ \
208 __ASM_EMIT("vshufps $0x6e, %%ymm2, %%ymm0, %%ymm1") /* ymm0 = i0' i2' i1' r3' */ \
209 __ASM_EMIT("vshufps $0x6e, %%ymm3, %%ymm4, %%ymm5") \
210 __ASM_EMIT("vshufps $0xc4, %%ymm2, %%ymm0, %%ymm0") /* ymm1 = r0' r2' r1' i3' */ \
211 __ASM_EMIT("vshufps $0xc4, %%ymm3, %%ymm4, %%ymm4") \
212 __ASM_EMIT("vhsubps %%ymm1, %%ymm0, %%ymm2") /* ymm2 = r0'-r2' r1'-i3' i0'-i2' i1'-r3' = r1" r3" i1" i2" */ \
213 __ASM_EMIT("vhsubps %%ymm5, %%ymm4, %%ymm3") \
214 __ASM_EMIT("vhaddps %%ymm1, %%ymm0, %%ymm0") /* ymm0 = r0'+r2' r1'+i3' i0'+i2' i1'+r3' = r0" r2" i0" i3" */ \
215 __ASM_EMIT("vhaddps %%ymm5, %%ymm4, %%ymm4") \
216 __ASM_EMIT("vblendps $0x88, %%ymm0, %%ymm2, %%ymm1") /* ymm1 = r1" r3" i1" i3" */ \
217 __ASM_EMIT("vblendps $0x88, %%ymm4, %%ymm3, %%ymm5") \
218 __ASM_EMIT("vblendps $0x88, %%ymm2, %%ymm0, %%ymm0") /* ymm0 = r0" r2" i0" i2" */ \
219 __ASM_EMIT("vblendps $0x88, %%ymm3, %%ymm4, %%ymm4") \
220 __ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm2") /* ymm2 = i0" i1" i2" i3" */ \
221 __ASM_EMIT("vunpckhps %%ymm5, %%ymm4, %%ymm6") \
222 __ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm0") /* ymm0 = r0" r1" r2" r3" */ \
223 __ASM_EMIT("vunpcklps %%ymm5, %%ymm4, %%ymm4") \
224 /* Reshuffle */ \
225 __ASM_EMIT("vextractf128 $1, %%ymm0, %%xmm5") /* xmm5 = r4 */ \
226 __ASM_EMIT("vextractf128 $1, %%ymm2, %%xmm7") /* xmm7 = i4 */ \
227 __ASM_EMIT("vinsertf128 $1, %%xmm4, %%ymm0, %%ymm0") /* ymm0 = r0 r8 */ \
228 __ASM_EMIT("vinsertf128 $1, %%xmm6, %%ymm2, %%ymm2") /* ymm2 = i0 i8 */ \
229 __ASM_EMIT("vinsertf128 $0, %%xmm5, %%ymm4, %%ymm4") /* ymm4 = r4 r12 */ \
230 __ASM_EMIT("vinsertf128 $0, %%xmm7, %%ymm6, %%ymm6") /* ymm6 = i4 i12 */ \
231 /* Apply convolution */ \
232 __ASM_EMIT("vmulps 0x20(%[conv], %[off]), %%ymm0, %%ymm1") /* ymm1 = r0*I */ \
233 __ASM_EMIT("vmulps 0x60(%[conv], %[off]), %%ymm4, %%ymm5") /* ymm1 = r4*I */ \
234 __ASM_EMIT("vmulps 0x20(%[conv], %[off]), %%ymm2, %%ymm3") /* ymm3 = i0*I */ \
235 __ASM_EMIT("vmulps 0x60(%[conv], %[off]), %%ymm6, %%ymm7") /* ymm3 = i4*I */ \
236 __ASM_EMIT(FMA_SEL("vmulps 0x00(%[conv], %[off]), %%ymm0, %%ymm0", "")) /* ymm0 = r0*R */ \
237 __ASM_EMIT(FMA_SEL("vmulps 0x40(%[conv], %[off]), %%ymm4, %%ymm4", "")) /* ymm4 = r4*R */ \
238 __ASM_EMIT(FMA_SEL("vmulps 0x00(%[conv], %[off]), %%ymm2, %%ymm2", "")) /* ymm2 = i0*R */ \
239 __ASM_EMIT(FMA_SEL("vmulps 0x40(%[conv], %[off]), %%ymm6, %%ymm6", "")) /* ymm6 = i4*R */ \
240 __ASM_EMIT(FMA_SEL("vsubps %%ymm3, %%ymm0, %%ymm0", "vfmsub132ps 0x00(%[conv], %[off]), %%ymm3, %%ymm0")) /* ymm0 = r0*R - i0*I */ \
241 __ASM_EMIT(FMA_SEL("vsubps %%ymm7, %%ymm4, %%ymm4", "vfmsub132ps 0x40(%[conv], %[off]), %%ymm7, %%ymm4")) /* ymm4 = r4*R - i4*I */ \
242 __ASM_EMIT(FMA_SEL("vaddps %%ymm1, %%ymm2, %%ymm2", "vfmadd132ps 0x00(%[conv], %[off]), %%ymm1, %%ymm2")) /* ymm2 = r0*I + i0*R */ \
243 __ASM_EMIT(FMA_SEL("vaddps %%ymm5, %%ymm6, %%ymm6", "vfmadd132ps 0x40(%[conv], %[off]), %%ymm5, %%ymm6")) /* ymm2 = r0*I + i0*R */ \
244 /* Reshuffle */ \
245 __ASM_EMIT("vextractf128 $1, %%ymm0, %%xmm5") /* xmm5 = r4 */ \
246 __ASM_EMIT("vextractf128 $1, %%ymm2, %%xmm7") /* xmm7 = i4 */ \
247 __ASM_EMIT("vinsertf128 $1, %%xmm4, %%ymm0, %%ymm0") /* ymm0 = r0 r8 */ \
248 __ASM_EMIT("vinsertf128 $1, %%xmm6, %%ymm2, %%ymm2") /* ymm2 = i0 i8 */ \
249 __ASM_EMIT("vinsertf128 $0, %%xmm5, %%ymm4, %%ymm4") /* ymm4 = r4 r12 */ \
250 __ASM_EMIT("vinsertf128 $0, %%xmm7, %%ymm6, %%ymm6") /* ymm6 = i4 i12 */ \
251 /* 1st-order 4x butterfly */ \
252 __ASM_EMIT("vhsubps %%ymm4, %%ymm0, %%ymm1") /* ymm1 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7' */ \
253 __ASM_EMIT("vhsubps %%ymm6, %%ymm2, %%ymm3") /* ymm3 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7' */ \
254 __ASM_EMIT("vhaddps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6' */ \
255 __ASM_EMIT("vhaddps %%ymm6, %%ymm2, %%ymm2") /* ymm2 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6' */ \
256 /* 2nd-order 4x butterfly */ \
257 __ASM_EMIT("vblendps $0xaa, %%ymm3, %%ymm1, %%ymm4") /* ymm4 = r1' i3' r5' i7' */ \
258 __ASM_EMIT("vblendps $0xaa, %%ymm1, %%ymm3, %%ymm5") /* ymm5 = i1' r3' i5' r7' */ \
259 __ASM_EMIT("vhsubps %%ymm4, %%ymm0, %%ymm1") /* ymm1 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r1" r5" */ \
260 __ASM_EMIT("vhsubps %%ymm5, %%ymm2, %%ymm3") /* ymm3 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i3" i7" */ \
261 __ASM_EMIT("vhaddps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r3" r7" */ \
262 __ASM_EMIT("vhaddps %%ymm5, %%ymm2, %%ymm2") /* ymm2 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i1" i5" */ \
263 __ASM_EMIT("vblendps $0xcc, %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0" i4" r1" r5" */ \
264 __ASM_EMIT("vblendps $0xcc, %%ymm0, %%ymm1, %%ymm5") /* ymm5 = r2" r6" r3" r7" */ \
265 __ASM_EMIT("vshufps $0x88, %%ymm3, %%ymm2, %%ymm6") /* ymm6 = i0" i1" i2" i3" */ \
266 __ASM_EMIT("vshufps $0xdd, %%ymm3, %%ymm2, %%ymm7") /* ymm7 = i4" i5" i6" i7" */ \
267 __ASM_EMIT("vshufps $0x88, %%ymm5, %%ymm4, %%ymm2") /* ymm2 = r0" r1" r2" r3" */ \
268 __ASM_EMIT("vshufps $0xdd, %%ymm5, %%ymm4, %%ymm3") /* ymm3 = r4" r5" r6" r7" */ \
269 /* 3rd-order 8x butterfly */ \
270 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%ymm3, %%ymm4") /* ymm4 = x_im * b_re */ \
271 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%ymm7, %%ymm5") /* ymm5 = x_im * b_im */ \
272 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%ymm3, %%ymm3", "")) /* ymm3 = x_re * b_re */ \
273 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%ymm7, %%ymm7", "")) /* ymm7 = x_re * b_im */ \
274 __ASM_EMIT(FMA_SEL("vsubps %%ymm5, %%ymm3, %%ymm5", "vfmsub231ps 0x00 + %[FFT_A], %%ymm3, %%ymm5")) /* ymm5 = c_re = x_re * b_re - x_im * b_im */ \
275 __ASM_EMIT(FMA_SEL("vaddps %%ymm4, %%ymm7, %%ymm4", "vfmadd231ps 0x00 + %[FFT_A], %%ymm7, %%ymm4")) /* ymm4 = c_im = x_re * b_im + x_im * b_re */ \
276 __ASM_EMIT("vsubps %%ymm5, %%ymm2, %%ymm0") /* ymm0 = a_re - c_re */ \
277 __ASM_EMIT("vsubps %%ymm4, %%ymm6, %%ymm1") /* ymm1 = a_im - c_im */ \
278 __ASM_EMIT("vaddps %%ymm5, %%ymm2, %%ymm2") /* ymm2 = a_re + c_re */ \
279 __ASM_EMIT("vaddps %%ymm4, %%ymm6, %%ymm3") /* ymm3 = a_im + c_im */ \
280 /* Store */ \
281 __ASM_EMIT("vmovups %%xmm2, 0x00(%[dst], %[off])") \
282 __ASM_EMIT("vmovups %%xmm0, 0x10(%[dst], %[off])") \
283 __ASM_EMIT("vmovups %%xmm3, 0x20(%[dst], %[off])") \
284 __ASM_EMIT("vmovups %%xmm1, 0x30(%[dst], %[off])") \
285 __ASM_EMIT("vextractf128 $1, %%ymm2, 0x40(%[dst], %[off])") \
286 __ASM_EMIT("vextractf128 $1, %%ymm0, 0x50(%[dst], %[off])") \
287 __ASM_EMIT("vextractf128 $1, %%ymm3, 0x60(%[dst], %[off])") \
288 __ASM_EMIT("vextractf128 $1, %%ymm1, 0x70(%[dst], %[off])") \
289 __ASM_EMIT("add $0x80, %[off]") \
290 __ASM_EMIT("sub $2, %[nb]") \
291 __ASM_EMIT("jae 1b") \
292 /* 1x block of 4-butterfly */ \
293 __ASM_EMIT("2:") \
294 __ASM_EMIT("add $1, %[nb]") \
295 __ASM_EMIT("jl 4f") \
296 /* Load data to registers */ \
297 __ASM_EMIT("vmovups 0x00(%[dst], %[off]), %%xmm0") /* xmm0 = r0 r1 r2 r3 */ \
298 __ASM_EMIT("vmovups 0x10(%[dst], %[off]), %%xmm2") /* xmm2 = r4 r5 r6 r7 */ \
299 __ASM_EMIT("vmovups 0x20(%[dst], %[off]), %%xmm1") /* xmm1 = i0 i1 i2 i3 */ \
300 __ASM_EMIT("vmovups 0x30(%[dst], %[off]), %%xmm3") /* xmm3 = i4 i5 i6 i7 */ \
301 /* Perform 3rd-order butterflies */ \
302 __ASM_EMIT("vsubps %%xmm2, %%xmm0, %%xmm4") /* xmm4 = c_re = a_re - b_re */ \
303 __ASM_EMIT("vsubps %%xmm3, %%xmm1, %%xmm5") /* xmm5 = c_im = a_im - b_im */ \
304 __ASM_EMIT("vaddps %%xmm2, %%xmm0, %%xmm0") /* xmm0 = a_re' = a_re + b_re */ \
305 __ASM_EMIT("vaddps %%xmm3, %%xmm1, %%xmm1") /* xmm1 = a_im' = a_im + b_im */ \
306 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%xmm4, %%xmm2") /* xmm2 = x_im * c_re */ \
307 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%xmm5, %%xmm3") /* xmm3 = x_im * c_im */ \
308 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%xmm4, %%xmm4", "")) /* xmm4 = x_re * c_re */ \
309 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%xmm5, %%xmm5", "")) /* xmm5 = x_re * c_im */ \
310 __ASM_EMIT(FMA_SEL("vaddps %%xmm3, %%xmm4, %%xmm4", "vfmadd132ps 0x00 + %[FFT_A], %%xmm3, %%xmm4")) /* xmm4 = b_re = x_re * c_re + x_im * c_im */ \
311 __ASM_EMIT(FMA_SEL("vsubps %%xmm2, %%xmm5, %%xmm5", "vfmsub132ps 0x00 + %[FFT_A], %%xmm2, %%xmm5")) /* xmm5 = b_im = x_re * c_im - x_im * c_re */ \
312 /* 2nd-order butterflies */ \
313 __ASM_EMIT("vshufps $0xd8, %%xmm0, %%xmm0, %%xmm0") /* xmm0 = r0 r2 r1 r3 */ \
314 __ASM_EMIT("vshufps $0xd8, %%xmm1, %%xmm1, %%xmm1") /* xmm1 = i0 i2 i1 i3 */ \
315 __ASM_EMIT("vshufps $0xd8, %%xmm4, %%xmm4, %%xmm4") \
316 __ASM_EMIT("vshufps $0xd8, %%xmm5, %%xmm5, %%xmm5") \
317 __ASM_EMIT("vhsubps %%xmm1, %%xmm0, %%xmm2") /* xmm2 = r0-r2 r1-r3 i0-i2 i1-i3 = r1' r3' i1' i3' */ \
318 __ASM_EMIT("vhsubps %%xmm5, %%xmm4, %%xmm3") \
319 __ASM_EMIT("vhaddps %%xmm1, %%xmm0, %%xmm0") /* xmm0 = r0+r2 r1+r3 i0+i2 i1+i3 = r0' r2' i0' i2' */ \
320 __ASM_EMIT("vhaddps %%xmm5, %%xmm4, %%xmm4") \
321 /* 1st-order 8x butterfly */ \
322 __ASM_EMIT("vshufps $0x6e, %%xmm2, %%xmm0, %%xmm1") /* xmm0 = i0' i2' i1' r3' */ \
323 __ASM_EMIT("vshufps $0x6e, %%xmm3, %%xmm4, %%xmm5") \
324 __ASM_EMIT("vshufps $0xc4, %%xmm2, %%xmm0, %%xmm0") /* xmm1 = r0' r2' r1' i3' */ \
325 __ASM_EMIT("vshufps $0xc4, %%xmm3, %%xmm4, %%xmm4") \
326 __ASM_EMIT("vhsubps %%xmm1, %%xmm0, %%xmm2") /* xmm2 = r0'-r2' r1'-i3' i0'-i2' i1'-r3' = r1" r3" i1" i2" */ \
327 __ASM_EMIT("vhsubps %%xmm5, %%xmm4, %%xmm3") \
328 __ASM_EMIT("vhaddps %%xmm1, %%xmm0, %%xmm0") /* xmm0 = r0'+r2' r1'+i3' i0'+i2' i1'+r3' = r0" r2" i0" i3" */ \
329 __ASM_EMIT("vhaddps %%xmm5, %%xmm4, %%xmm4") \
330 __ASM_EMIT("vblendps $0x88, %%xmm0, %%xmm2, %%xmm1") /* xmm1 = r1" r3" i1" i3" */ \
331 __ASM_EMIT("vblendps $0x88, %%xmm4, %%xmm3, %%xmm5") \
332 __ASM_EMIT("vblendps $0x88, %%xmm2, %%xmm0, %%xmm0") /* xmm0 = r0" r2" i0" i2" */ \
333 __ASM_EMIT("vblendps $0x88, %%xmm3, %%xmm4, %%xmm4") \
334 __ASM_EMIT("vunpckhps %%xmm1, %%xmm0, %%xmm2") /* xmm2 = i0" i1" i2" i3" */ \
335 __ASM_EMIT("vunpckhps %%xmm5, %%xmm4, %%xmm6") \
336 __ASM_EMIT("vunpcklps %%xmm1, %%xmm0, %%xmm0") /* xmm0 = r0" r1" r2" r3" */ \
337 __ASM_EMIT("vunpcklps %%xmm5, %%xmm4, %%xmm4") \
338 /* Reshuffle */ \
339 __ASM_EMIT("vinsertf128 $1, %%xmm4, %%ymm0, %%ymm0") /* ymm0 = r0 r4 */ \
340 __ASM_EMIT("vinsertf128 $1, %%xmm6, %%ymm2, %%ymm2") /* ymm2 = i0 i4 */ \
341 /* Apply convolution */ \
342 __ASM_EMIT("vmulps 0x20(%[conv], %[off]), %%ymm0, %%ymm1") /* ymm1 = r0*I */ \
343 __ASM_EMIT("vmulps 0x20(%[conv], %[off]), %%ymm2, %%ymm3") /* ymm3 = i0*I */ \
344 __ASM_EMIT(FMA_SEL("vmulps 0x00(%[conv], %[off]), %%ymm0, %%ymm0", "")) /* ymm0 = r0*R */ \
345 __ASM_EMIT(FMA_SEL("vmulps 0x00(%[conv], %[off]), %%ymm2, %%ymm2", "")) /* ymm2 = i0*R */ \
346 __ASM_EMIT(FMA_SEL("vsubps %%ymm3, %%ymm0, %%ymm0", "vfmsub132ps 0x00(%[conv], %[off]), %%ymm3, %%ymm0")) /* ymm0 = r0*R - i0*I */ \
347 __ASM_EMIT(FMA_SEL("vaddps %%ymm1, %%ymm2, %%ymm2", "vfmadd132ps 0x00(%[conv], %[off]), %%ymm1, %%ymm2")) /* ymm2 = r0*I + i0*R */ \
348 /* Reshuffle */ \
349 __ASM_EMIT("vextractf128 $1, %%ymm0, %%xmm4") /* xmm4 = r4 */ \
350 __ASM_EMIT("vextractf128 $1, %%ymm2, %%xmm6") /* xmm6 = i4 */ \
351 /* 1st-order 4x butterfly */ \
352 __ASM_EMIT("vhsubps %%xmm4, %%xmm0, %%xmm1") /* xmm1 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7' */ \
353 __ASM_EMIT("vhsubps %%xmm6, %%xmm2, %%xmm3") /* xmm3 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7' */ \
354 __ASM_EMIT("vhaddps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6' */ \
355 __ASM_EMIT("vhaddps %%xmm6, %%xmm2, %%xmm2") /* xmm2 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6' */ \
356 /* 2nd-order 4x butterfly */ \
357 __ASM_EMIT("vblendps $0xaa, %%xmm3, %%xmm1, %%xmm4") /* xmm4 = r1' i3' r5' i7' */ \
358 __ASM_EMIT("vblendps $0xaa, %%xmm1, %%xmm3, %%xmm5") /* xmm5 = i1' r3' i5' r7' */ \
359 __ASM_EMIT("vhsubps %%xmm4, %%xmm0, %%xmm1") /* xmm1 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r1" r5" */ \
360 __ASM_EMIT("vhsubps %%xmm5, %%xmm2, %%xmm3") /* xmm3 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i3" i7" */ \
361 __ASM_EMIT("vhaddps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r3" r7" */ \
362 __ASM_EMIT("vhaddps %%xmm5, %%xmm2, %%xmm2") /* xmm2 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i1" i5" */ \
363 __ASM_EMIT("vblendps $0xcc, %%xmm1, %%xmm0, %%xmm4") /* xmm4 = r0" i4" r1" r5" */ \
364 __ASM_EMIT("vblendps $0xcc, %%xmm0, %%xmm1, %%xmm5") /* xmm5 = r2" r6" r3" r7" */ \
365 __ASM_EMIT("vshufps $0x88, %%xmm3, %%xmm2, %%xmm6") /* xmm6 = i0" i1" i2" i3" */ \
366 __ASM_EMIT("vshufps $0xdd, %%xmm3, %%xmm2, %%xmm7") /* xmm7 = i4" i5" i6" i7" */ \
367 __ASM_EMIT("vshufps $0x88, %%xmm5, %%xmm4, %%xmm2") /* xmm2 = r0" r1" r2" r3" */ \
368 __ASM_EMIT("vshufps $0xdd, %%xmm5, %%xmm4, %%xmm3") /* xmm3 = r4" r5" r6" r7" */ \
369 /* 3rd-order 8x butterfly */ \
370 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%xmm3, %%xmm4") /* xmm4 = x_im * b_re */ \
371 __ASM_EMIT("vmulps 0x20 + %[FFT_A], %%xmm7, %%xmm5") /* xmm5 = x_im * b_im */ \
372 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%xmm3, %%xmm3", "")) /* xmm3 = x_re * b_re */ \
373 __ASM_EMIT(FMA_SEL("vmulps 0x00 + %[FFT_A], %%xmm7, %%xmm7", "")) /* xmm7 = x_re * b_im */ \
374 __ASM_EMIT(FMA_SEL("vsubps %%xmm5, %%xmm3, %%xmm5", "vfmsub231ps 0x00 + %[FFT_A], %%xmm3, %%xmm5")) /* xmm5 = c_re = x_re * b_re - x_im * b_im */ \
375 __ASM_EMIT(FMA_SEL("vaddps %%xmm4, %%xmm7, %%xmm4", "vfmadd231ps 0x00 + %[FFT_A], %%xmm7, %%xmm4")) /* xmm4 = c_im = x_re * b_im + x_im * b_re */ \
376 __ASM_EMIT("vsubps %%xmm5, %%xmm2, %%xmm0") /* xmm0 = a_re - c_re */ \
377 __ASM_EMIT("vsubps %%xmm4, %%xmm6, %%xmm1") /* xmm1 = a_im - c_im */ \
378 __ASM_EMIT("vaddps %%xmm5, %%xmm2, %%xmm2") /* xmm2 = a_re + c_re */ \
379 __ASM_EMIT("vaddps %%xmm4, %%xmm6, %%xmm3") /* xmm3 = a_im + c_im */ \
380 /* Store */ \
381 __ASM_EMIT("vmovups %%xmm2, 0x00(%[dst], %[off])") \
382 __ASM_EMIT("vmovups %%xmm0, 0x10(%[dst], %[off])") \
383 __ASM_EMIT("vmovups %%xmm3, 0x20(%[dst], %[off])") \
384 __ASM_EMIT("vmovups %%xmm1, 0x30(%[dst], %[off])") \
385 __ASM_EMIT("4:") \
386 : [off] "=&r" (off), [nb] "+r" (nb) \
387 : [dst] "r" (dst), [conv] "r" (conv), \
388 [FFT_A] "o" (FFT_A) \
389 : "cc", "memory", \
390 "%xmm0", "%xmm1", "%xmm2", "%xmm3", \
391 "%xmm4", "%xmm5", "%xmm6", "%xmm7" \
392 )
393
394 static inline void fastconv_apply_prepare(float *dst, const float *c1, const float *c2, size_t nb)
395 {
396 FASTCONV_APPLY_PREPARE_CORE(FMA_OFF);
397 }
398
399 static inline void fastconv_apply_prepare_fma3(float *dst, const float *c1, const float *c2, size_t nb)
400 {
401 FASTCONV_APPLY_PREPARE_CORE(FMA_ON);
402 }
403
404 static inline void fastconv_apply_internal(float *dst, const float *conv, size_t nb)
405 {
406 FASTCONV_APPLY_CORE(FMA_OFF);
407 }
408
409 static inline void fastconv_apply_internal_fma3(float *dst, const float *conv, size_t nb)
410 {
411 FASTCONV_APPLY_CORE(FMA_ON);
412 }
413
414 #undef FASTCONV_APPLY_PREPARE_CORE
415 #undef FASTCONV_APPLY_CORE
416 #undef FMA_ON
417 #undef FMA_OFF
418 }
419 }
420
421 #endif /* PRIVATE_DSP_ARCH_X86_AVX_FASTCONV_APPLY_H_ */