| 1 |
/* |
| 2 |
* Copyright (C) 2025 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2025 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 14 апр. 2025 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_ARM_NEON_D32_PMATH_COS_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_ARM_NEON_D32_PMATH_COS_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_ARM_NEON_D32_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_ARM_NEON_D32_IMPL */ |
| 28 |
|
| 29 |
#include <private/dsp/arch/arm/neon-d32/pmath/sin.h> |
| 30 |
|
| 31 |
namespace lsp |
| 32 |
{ |
| 33 |
namespace neon_d32 |
| 34 |
{ |
| 35 |
void cosf1(float *dst, size_t count) |
| 36 |
{ |
| 37 |
ARCH_ARM_ASM( |
| 38 |
__ASM_EMIT("vldm %[S2C], {q8-q15}") |
| 39 |
// x8 blocks |
| 40 |
__ASM_EMIT("subs %[count], #8") |
| 41 |
__ASM_EMIT("blo 2f") |
| 42 |
__ASM_EMIT("1:") |
| 43 |
__ASM_EMIT("vldm %[dst], {q0-q1}") |
| 44 |
__ASM_EMIT("vsub.f32 q0, q9, q0") // q0 = PI/2 - x1 + PI/2 |
| 45 |
__ASM_EMIT("vsub.f32 q1, q9, q1") // q1 = PI/2 - x2 + PI/2 |
| 46 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 47 |
__ASM_EMIT("subs %[count], #8") |
| 48 |
__ASM_EMIT("vstm %[dst]!, {q0-q1}") |
| 49 |
__ASM_EMIT("bhs 1b") |
| 50 |
__ASM_EMIT("2:") |
| 51 |
// x4 block |
| 52 |
__ASM_EMIT("adds %[count], #4") |
| 53 |
__ASM_EMIT("blt 4f") |
| 54 |
__ASM_EMIT("vld1.32 {q0}, [%[dst]]") |
| 55 |
__ASM_EMIT("vsub.f32 q0, q9, q0") // q0 = PI/2 - x1 + PI/2 |
| 56 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 57 |
__ASM_EMIT("sub %[count], #4") |
| 58 |
__ASM_EMIT("vst1.32 {q0}, [%[dst]]!") |
| 59 |
__ASM_EMIT("4:") |
| 60 |
// Tail: 1x-3x block |
| 61 |
__ASM_EMIT("adds %[count], #4") |
| 62 |
__ASM_EMIT("bls 12f") |
| 63 |
__ASM_EMIT("tst %[count], #1") |
| 64 |
__ASM_EMIT("beq 6f") |
| 65 |
__ASM_EMIT("vld1.32 {d0[0]}, [%[dst]]!") |
| 66 |
__ASM_EMIT("6:") |
| 67 |
__ASM_EMIT("tst %[count], #2") |
| 68 |
__ASM_EMIT("beq 8f") |
| 69 |
__ASM_EMIT("vld1.32 {d1}, [%[dst]]") |
| 70 |
__ASM_EMIT("8:") |
| 71 |
__ASM_EMIT("vsub.f32 q0, q9, q0") // q0 = PI/2 - x1 + PI/2 |
| 72 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 73 |
__ASM_EMIT("tst %[count], #1") |
| 74 |
__ASM_EMIT("beq 10f") |
| 75 |
__ASM_EMIT("sub %[dst], #0x04") |
| 76 |
__ASM_EMIT("vst1.32 {d0[0]}, [%[dst]]!") |
| 77 |
__ASM_EMIT("10:") |
| 78 |
__ASM_EMIT("tst %[count], #2") |
| 79 |
__ASM_EMIT("beq 12f") |
| 80 |
__ASM_EMIT("vst1.32 {d1}, [%[dst]]") |
| 81 |
__ASM_EMIT("12:") |
| 82 |
|
| 83 |
: [dst] "+r" (dst), [count] "+r" (count) |
| 84 |
: [S2C] "r" (&sinf_const[0]) |
| 85 |
: "cc", "memory", |
| 86 |
"q0", "q1", "q2", "q3", |
| 87 |
"q4", "q5", "q6", "q7", |
| 88 |
"q8", "q9", "q10", "q11", |
| 89 |
"q12", "q13", "q14", "q15" |
| 90 |
); |
| 91 |
} |
| 92 |
|
| 93 |
void cosf2(float *dst, const float *src, size_t count) |
| 94 |
{ |
| 95 |
ARCH_ARM_ASM( |
| 96 |
__ASM_EMIT("vldm %[S2C], {q8-q15}") |
| 97 |
// x8 blocks |
| 98 |
__ASM_EMIT("subs %[count], #8") |
| 99 |
__ASM_EMIT("blo 2f") |
| 100 |
__ASM_EMIT("1:") |
| 101 |
__ASM_EMIT("vldm %[src]!, {q0-q1}") |
| 102 |
__ASM_EMIT("vsub.f32 q0, q9, q0") // q0 = PI/2 - x1 + PI/2 |
| 103 |
__ASM_EMIT("vsub.f32 q1, q9, q1") // q1 = PI/2 - x2 + PI/2 |
| 104 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 105 |
__ASM_EMIT("subs %[count], #8") |
| 106 |
__ASM_EMIT("vstm %[dst]!, {q0-q1}") |
| 107 |
__ASM_EMIT("bhs 1b") |
| 108 |
__ASM_EMIT("2:") |
| 109 |
// x4 block |
| 110 |
__ASM_EMIT("adds %[count], #4") |
| 111 |
__ASM_EMIT("blt 4f") |
| 112 |
__ASM_EMIT("vld1.32 {q0}, [%[src]]!") |
| 113 |
__ASM_EMIT("vsub.f32 q0, q9, q0") // q0 = PI/2 - x1 + PI/2 |
| 114 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 115 |
__ASM_EMIT("sub %[count], #4") |
| 116 |
__ASM_EMIT("vst1.32 {q0}, [%[dst]]!") |
| 117 |
__ASM_EMIT("4:") |
| 118 |
// Tail: 1x-3x block |
| 119 |
__ASM_EMIT("adds %[count], #4") |
| 120 |
__ASM_EMIT("bls 12f") |
| 121 |
__ASM_EMIT("tst %[count], #1") |
| 122 |
__ASM_EMIT("beq 6f") |
| 123 |
__ASM_EMIT("vld1.32 {d0[0]}, [%[src]]!") |
| 124 |
__ASM_EMIT("6:") |
| 125 |
__ASM_EMIT("tst %[count], #2") |
| 126 |
__ASM_EMIT("beq 8f") |
| 127 |
__ASM_EMIT("vld1.32 {d1}, [%[src]]") |
| 128 |
__ASM_EMIT("8:") |
| 129 |
__ASM_EMIT("vsub.f32 q0, q9, q0") // q0 = PI/2 - x1 + PI/2 |
| 130 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 131 |
__ASM_EMIT("tst %[count], #1") |
| 132 |
__ASM_EMIT("beq 10f") |
| 133 |
__ASM_EMIT("vst1.32 {d0[0]}, [%[dst]]!") |
| 134 |
__ASM_EMIT("10:") |
| 135 |
__ASM_EMIT("tst %[count], #2") |
| 136 |
__ASM_EMIT("beq 12f") |
| 137 |
__ASM_EMIT("vst1.32 {d1}, [%[dst]]") |
| 138 |
__ASM_EMIT("12:") |
| 139 |
|
| 140 |
: [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) |
| 141 |
: [S2C] "r" (&sinf_const[0]) |
| 142 |
: "cc", "memory", |
| 143 |
"q0", "q1", "q2", "q3", |
| 144 |
"q4", "q5", "q6", "q7", |
| 145 |
"q8", "q9", "q10", "q11", |
| 146 |
"q12", "q13", "q14", "q15" |
| 147 |
); |
| 148 |
} |
| 149 |
|
| 150 |
void cosf_kp1(float *dst, float k, float p, size_t count) |
| 151 |
{ |
| 152 |
ARCH_ARM_ASM( |
| 153 |
__ASM_EMIT("vld1.f32 {d8[], d9[]}, [%[k]]") // q4 = k |
| 154 |
__ASM_EMIT("vld1.f32 {d10[], d11[]}, [%[p]]") // q5 = p |
| 155 |
__ASM_EMIT("vldm %[S2C], {q8-q15}") |
| 156 |
__ASM_EMIT("vldm %[S2KP], {q6-q7}") // q6 = i = 0 1 2 3, q7 = 4 ... |
| 157 |
__ASM_EMIT("vsub.f32 q5, q9, q5") // q5 = PI/2 - p + PI/2 |
| 158 |
// x4 blocks |
| 159 |
__ASM_EMIT("subs %[count], #4") |
| 160 |
__ASM_EMIT("blo 2f") |
| 161 |
__ASM_EMIT("1:") |
| 162 |
__ASM_EMIT("vmul.f32 q0, q6, q4") // q0 = k * i |
| 163 |
__ASM_EMIT("vadd.f32 q6, q6, q7") // q6 = i + 4 |
| 164 |
__ASM_EMIT("vsub.f32 q0, q5, q0") // q0 = PI/2 - (k*i + p) + PI/2 |
| 165 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 166 |
__ASM_EMIT("subs %[count], #4") |
| 167 |
__ASM_EMIT("vst1.32 {q0}, [%[dst]]!") |
| 168 |
__ASM_EMIT("bhs 1b") |
| 169 |
__ASM_EMIT("2:") |
| 170 |
// Tail: 1x-3x block |
| 171 |
__ASM_EMIT("adds %[count], #4") |
| 172 |
__ASM_EMIT("bls 6f") |
| 173 |
__ASM_EMIT("vmul.f32 q0, q6, q4") // q0 = k * i |
| 174 |
__ASM_EMIT("vsub.f32 q0, q5, q0") // q0 = PI/2 - (k*i + p) + PI/2 |
| 175 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 176 |
__ASM_EMIT("tst %[count], #1") |
| 177 |
__ASM_EMIT("beq 4f") |
| 178 |
__ASM_EMIT("vst1.32 {d0[0]}, [%[dst]]!") |
| 179 |
__ASM_EMIT("vext.32 q0, q0, q0, #1") |
| 180 |
__ASM_EMIT("4:") |
| 181 |
__ASM_EMIT("tst %[count], #2") |
| 182 |
__ASM_EMIT("beq 6f") |
| 183 |
__ASM_EMIT("vst1.32 {d0}, [%[dst]]") |
| 184 |
// End |
| 185 |
__ASM_EMIT("6:") |
| 186 |
|
| 187 |
: [dst] "+r" (dst), [count] "+r" (count) |
| 188 |
: [k] "r" (&k), |
| 189 |
[p] "r" (&p), |
| 190 |
[S2C] "r" (&sinf_const[0]), |
| 191 |
[S2KP] "r" (&kp_gen_const[0]) |
| 192 |
: "cc", "memory", |
| 193 |
"q0", "q1", "q2", "q3", |
| 194 |
"q4", "q5", "q6", "q7", |
| 195 |
"q8", "q9", "q10", "q11", |
| 196 |
"q12", "q13", "q14", "q15" |
| 197 |
); |
| 198 |
} |
| 199 |
|
| 200 |
} /* namespace neon_d32 */ |
| 201 |
} /* namespace lsp */ |
| 202 |
|
| 203 |
|
| 204 |
#endif /* PRIVATE_DSP_ARCH_ARM_NEON_D32_PMATH_COS_H_ */ |