| 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_FFT_P_REPACK_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_AVX_FFT_P_REPACK_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 |
static inline void packed_fft_repack(float *dst, size_t rank) |
| 34 |
{ |
| 35 |
size_t blocks = 1 << rank; |
| 36 |
|
| 37 |
// Perform 4-element butterflies |
| 38 |
ARCH_X86_ASM |
| 39 |
( |
| 40 |
// 16x blocks |
| 41 |
__ASM_EMIT("sub $16, %[blocks]") |
| 42 |
__ASM_EMIT("jb 2f") |
| 43 |
__ASM_EMIT("1:") |
| 44 |
__ASM_EMIT("vmovups 0x00(%[dst]), %%ymm0") /* ymm0 = r0 r1 r2 r3 r4 r5 r6 r7 */ |
| 45 |
__ASM_EMIT("vmovups 0x20(%[dst]), %%ymm1") /* ymm1 = i0 i1 i2 i3 i4 i5 i6 i7 */ |
| 46 |
__ASM_EMIT("vmovups 0x40(%[dst]), %%ymm2") /* ymm2 = r8 r9 r10 r11 r12 r13 r14 r15 */ |
| 47 |
__ASM_EMIT("vmovups 0x60(%[dst]), %%ymm3") /* ymm3 = i8 i9 i10 i11 i12 i13 i14 i15 */ |
| 48 |
__ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0 i0 r1 i1 r4 i4 r5 i5 */ |
| 49 |
__ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm5") /* ymm5 = r2 i2 r3 i3 r6 i6 r7 i7 */ |
| 50 |
__ASM_EMIT("vunpcklps %%ymm3, %%ymm2, %%ymm0") /* ymm0 = r8 i8 r9 i9 r12 i12 r13 i13 */ |
| 51 |
__ASM_EMIT("vunpckhps %%ymm3, %%ymm2, %%ymm1") /* ymm1 = r10 i10 r12 i12 r14 i14 r15 i15 */ |
| 52 |
__ASM_EMIT("vmovups %%xmm4, 0x00(%[dst])") |
| 53 |
__ASM_EMIT("vmovups %%xmm5, 0x10(%[dst])") |
| 54 |
__ASM_EMIT("vextractf128 $1, %%ymm4, 0x20(%[dst])") |
| 55 |
__ASM_EMIT("vextractf128 $1, %%ymm5, 0x30(%[dst])") |
| 56 |
__ASM_EMIT("vmovups %%xmm0, 0x40(%[dst])") |
| 57 |
__ASM_EMIT("vmovups %%xmm1, 0x50(%[dst])") |
| 58 |
__ASM_EMIT("vextractf128 $1, %%ymm0, 0x60(%[dst])") |
| 59 |
__ASM_EMIT("vextractf128 $1, %%ymm1, 0x70(%[dst])") |
| 60 |
__ASM_EMIT("add $0x80, %[dst]") |
| 61 |
__ASM_EMIT("sub $16, %[blocks]") |
| 62 |
__ASM_EMIT("jae 1b") |
| 63 |
__ASM_EMIT("2:") |
| 64 |
// 8x block |
| 65 |
__ASM_EMIT("add $8, %[blocks]") |
| 66 |
__ASM_EMIT("jl 4f") |
| 67 |
__ASM_EMIT("vmovups 0x00(%[dst]), %%ymm0") /* ymm0 = r0 r1 r2 r3 r4 r5 r6 r7 */ |
| 68 |
__ASM_EMIT("vmovups 0x20(%[dst]), %%ymm1") /* ymm1 = i0 i1 i2 i3 i4 i5 i6 i7 */ |
| 69 |
__ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0 i0 r1 i1 r4 i4 r5 i5 */ |
| 70 |
__ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm5") /* ymm5 = r2 i2 r3 i3 r6 i6 r7 i7 */ |
| 71 |
__ASM_EMIT("vmovups %%xmm4, 0x00(%[dst])") |
| 72 |
__ASM_EMIT("vmovups %%xmm5, 0x10(%[dst])") |
| 73 |
__ASM_EMIT("vextractf128 $1, %%ymm4, 0x20(%[dst])") |
| 74 |
__ASM_EMIT("vextractf128 $1, %%ymm5, 0x30(%[dst])") |
| 75 |
__ASM_EMIT("4:") |
| 76 |
|
| 77 |
: [dst] "+r"(dst), [blocks] "+r" (blocks) |
| 78 |
: |
| 79 |
: "cc", "memory", |
| 80 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 81 |
"%xmm4", "%xmm5" |
| 82 |
); |
| 83 |
} |
| 84 |
|
| 85 |
static inline void packed_fft_repack_normalize(float *dst, size_t rank) |
| 86 |
{ |
| 87 |
size_t blocks = 1 << rank; |
| 88 |
float norm = 1.0f / float(blocks); |
| 89 |
|
| 90 |
// Perform 4-element butterflies |
| 91 |
ARCH_X86_ASM |
| 92 |
( |
| 93 |
__ASM_EMIT("vbroadcastss %[norm], %%ymm6") |
| 94 |
__ASM_EMIT("vmovaps %%ymm6, %%ymm7") |
| 95 |
// 16x blocks |
| 96 |
__ASM_EMIT("sub $16, %[blocks]") |
| 97 |
__ASM_EMIT("jb 2f") |
| 98 |
__ASM_EMIT("1:") |
| 99 |
__ASM_EMIT("vmulps 0x00(%[dst]), %%ymm6, %%ymm0") /* ymm0 = r0 r1 r2 r3 r4 r5 r6 r7 */ |
| 100 |
__ASM_EMIT("vmulps 0x20(%[dst]), %%ymm7, %%ymm1") /* ymm1 = i0 i1 i2 i3 i4 i5 i6 i7 */ |
| 101 |
__ASM_EMIT("vmulps 0x40(%[dst]), %%ymm6, %%ymm2") /* ymm2 = r8 r9 r10 r11 r12 r13 r14 r15 */ |
| 102 |
__ASM_EMIT("vmulps 0x60(%[dst]), %%ymm7, %%ymm3") /* ymm3 = i8 i9 i10 i11 i12 i13 i14 i15 */ |
| 103 |
__ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0 i0 r1 i1 r4 i4 r5 i5 */ |
| 104 |
__ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm5") /* ymm5 = r2 i2 r3 i3 r6 i6 r7 i7 */ |
| 105 |
__ASM_EMIT("vunpcklps %%ymm3, %%ymm2, %%ymm0") /* ymm0 = r8 i8 r9 i9 r12 i12 r13 i13 */ |
| 106 |
__ASM_EMIT("vunpckhps %%ymm3, %%ymm2, %%ymm1") /* ymm1 = r10 i10 r12 i12 r14 i14 r15 i15 */ |
| 107 |
__ASM_EMIT("vmovups %%xmm4, 0x00(%[dst])") |
| 108 |
__ASM_EMIT("vmovups %%xmm5, 0x10(%[dst])") |
| 109 |
__ASM_EMIT("vextractf128 $1, %%ymm4, 0x20(%[dst])") |
| 110 |
__ASM_EMIT("vextractf128 $1, %%ymm5, 0x30(%[dst])") |
| 111 |
__ASM_EMIT("vmovups %%xmm0, 0x40(%[dst])") |
| 112 |
__ASM_EMIT("vmovups %%xmm1, 0x50(%[dst])") |
| 113 |
__ASM_EMIT("vextractf128 $1, %%ymm0, 0x60(%[dst])") |
| 114 |
__ASM_EMIT("vextractf128 $1, %%ymm1, 0x70(%[dst])") |
| 115 |
__ASM_EMIT("add $0x80, %[dst]") |
| 116 |
__ASM_EMIT("sub $16, %[blocks]") |
| 117 |
__ASM_EMIT("jae 1b") |
| 118 |
__ASM_EMIT("2:") |
| 119 |
// 8x block |
| 120 |
__ASM_EMIT("add $8, %[blocks]") |
| 121 |
__ASM_EMIT("jl 4f") |
| 122 |
__ASM_EMIT("vmulps 0x00(%[dst]), %%ymm6, %%ymm0") /* ymm0 = r0 r1 r2 r3 r4 r5 r6 r7 */ |
| 123 |
__ASM_EMIT("vmulps 0x20(%[dst]), %%ymm7, %%ymm1") /* ymm1 = i0 i1 i2 i3 i4 i5 i6 i7 */ |
| 124 |
__ASM_EMIT("vunpcklps %%ymm1, %%ymm0, %%ymm4") /* ymm4 = r0 i0 r1 i1 r4 i4 r5 i5 */ |
| 125 |
__ASM_EMIT("vunpckhps %%ymm1, %%ymm0, %%ymm5") /* ymm5 = r2 i2 r3 i3 r6 i6 r7 i7 */ |
| 126 |
__ASM_EMIT("vmovups %%xmm4, 0x00(%[dst])") |
| 127 |
__ASM_EMIT("vmovups %%xmm5, 0x10(%[dst])") |
| 128 |
__ASM_EMIT("vextractf128 $1, %%ymm4, 0x20(%[dst])") |
| 129 |
__ASM_EMIT("vextractf128 $1, %%ymm5, 0x30(%[dst])") |
| 130 |
__ASM_EMIT("4:") |
| 131 |
|
| 132 |
: [dst] "+r"(dst), [blocks] "+r" (blocks) |
| 133 |
: [norm] "m" (norm) |
| 134 |
: "cc", "memory", |
| 135 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 136 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 137 |
); |
| 138 |
} |
| 139 |
} |
| 140 |
} |
| 141 |
|
| 142 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX_FFT_P_REPACK_H_ */ |