| 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 INCLUDE_PRIVATE_DSP_ARCH_X86_AVX512_PMATH_COS_H_ |
| 23 |
#define INCLUDE_PRIVATE_DSP_ARCH_X86_AVX512_PMATH_COS_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_AVX512_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX512_IMPL */ |
| 28 |
|
| 29 |
#include <private/dsp/arch/x86/avx512/pmath/sin.h> |
| 30 |
|
| 31 |
namespace lsp |
| 32 |
{ |
| 33 |
namespace avx512 |
| 34 |
{ |
| 35 |
void cosf1(float *dst, size_t count) |
| 36 |
{ |
| 37 |
IF_ARCH_X86(float *src); |
| 38 |
|
| 39 |
ARCH_X86_ASM( |
| 40 |
// x32 blocks |
| 41 |
__ASM_EMIT("sub $32, %[count]") |
| 42 |
__ASM_EMIT("jb 2f") |
| 43 |
__ASM_EMIT("1:") |
| 44 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%zmm1") // zmm1 = PI |
| 45 |
__ASM_EMIT("vsubps 0x00(%[dst]), %%zmm1, %%zmm0") // zmm0 = PI/2 - x1 + PI/2 |
| 46 |
__ASM_EMIT("vsubps 0x40(%[dst]), %%zmm1, %%zmm4") // zmm4 = PI/2 - x2 + PI/2 |
| 47 |
SINF_X_PLUS_PI_2_CORE_X32 |
| 48 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst])") |
| 49 |
__ASM_EMIT("vmovups %%zmm4, 0x40(%[dst])") |
| 50 |
__ASM_EMIT("add $0x80, %[dst]") |
| 51 |
__ASM_EMIT("sub $32, %[count]") |
| 52 |
__ASM_EMIT("jae 1b") |
| 53 |
__ASM_EMIT("2:") |
| 54 |
// x16 blocks |
| 55 |
__ASM_EMIT("add $16, %[count]") |
| 56 |
__ASM_EMIT("jl 4f") |
| 57 |
__ASM_EMIT("1:") |
| 58 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%zmm1") // zmm1 = PI |
| 59 |
__ASM_EMIT("vsubps 0x00(%[dst]), %%zmm1, %%zmm0") // zmm0 = PI/2 - x1 + PI/2 |
| 60 |
SINF_X_PLUS_PI_2_CORE_X16 |
| 61 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst])") |
| 62 |
__ASM_EMIT("add $0x40, %[dst]") |
| 63 |
__ASM_EMIT("sub $16, %[count]") |
| 64 |
__ASM_EMIT("4:") |
| 65 |
// x8 block |
| 66 |
__ASM_EMIT("add $8, %[count]") |
| 67 |
__ASM_EMIT("jl 6f") |
| 68 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%ymm1") // ymm1 = PI |
| 69 |
__ASM_EMIT("vsubps 0x00(%[dst]), %%ymm1, %%ymm0") // ymm0 = PI/2 - x1 + PI/2 |
| 70 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 71 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[dst])") |
| 72 |
__ASM_EMIT("add $0x20, %[dst]") |
| 73 |
__ASM_EMIT("sub $8, %[count]") |
| 74 |
__ASM_EMIT("6:") |
| 75 |
// x4 block |
| 76 |
__ASM_EMIT("add $4, %[count]") |
| 77 |
__ASM_EMIT("jl 8f") |
| 78 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%xmm1") // xmm1 = PI |
| 79 |
__ASM_EMIT("vsubps 0x00(%[dst]), %%xmm1, %%xmm0") // xmm0 = PI/2 - x1 + PI/2 |
| 80 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 81 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst])") |
| 82 |
__ASM_EMIT("add $0x10, %[dst]") |
| 83 |
__ASM_EMIT("sub $4, %[count]") |
| 84 |
__ASM_EMIT("8:") |
| 85 |
// Tail: 1x-3x block |
| 86 |
__ASM_EMIT("add $4, %[count]") |
| 87 |
__ASM_EMIT("jle 16f") |
| 88 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%xmm1") // xmm1 = PI |
| 89 |
__ASM_EMIT("mov %[dst], %[src]") |
| 90 |
__ASM_EMIT("test $1, %[count]") |
| 91 |
__ASM_EMIT("jz 10f") |
| 92 |
__ASM_EMIT("vmovss 0x00(%[src]), %%xmm0") |
| 93 |
__ASM_EMIT("add $4, %[src]") |
| 94 |
__ASM_EMIT("10:") |
| 95 |
__ASM_EMIT("test $2, %[count]") |
| 96 |
__ASM_EMIT("jz 12f") |
| 97 |
__ASM_EMIT("vmovhps 0x00(%[src]), %%xmm0, %%xmm0") |
| 98 |
__ASM_EMIT("12:") |
| 99 |
__ASM_EMIT("vsubps %%xmm0, %%xmm1, %%xmm0") // xmm0 = PI/2 - x1 + PI/2 |
| 100 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 101 |
__ASM_EMIT("test $1, %[count]") |
| 102 |
__ASM_EMIT("jz 14f") |
| 103 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[dst])") |
| 104 |
__ASM_EMIT("add $4, %[dst]") |
| 105 |
__ASM_EMIT("14:") |
| 106 |
__ASM_EMIT("test $2, %[count]") |
| 107 |
__ASM_EMIT("jz 16f") |
| 108 |
__ASM_EMIT("vmovhps %%xmm0, 0x00(%[dst])") |
| 109 |
// End |
| 110 |
__ASM_EMIT("16:") |
| 111 |
|
| 112 |
: [dst] "+r" (dst), [src] "=&r" (src), [count] "+r" (count) |
| 113 |
: [S2C] "o" (sinf_const) |
| 114 |
: "cc", "memory", |
| 115 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 116 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 117 |
"%k1", "%k2" |
| 118 |
); |
| 119 |
} |
| 120 |
|
| 121 |
void cosf2(float *dst, const float *src, size_t count) |
| 122 |
{ |
| 123 |
ARCH_X86_ASM( |
| 124 |
// x32 blocks |
| 125 |
__ASM_EMIT("sub $32, %[count]") |
| 126 |
__ASM_EMIT("jb 2f") |
| 127 |
__ASM_EMIT("1:") |
| 128 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%zmm1") // zmm1 = PI |
| 129 |
__ASM_EMIT("vsubps 0x00(%[src]), %%zmm1, %%zmm0") // zmm0 = PI/2 - x1 + PI/2 |
| 130 |
__ASM_EMIT("vsubps 0x40(%[src]), %%zmm1, %%zmm4") // zmm4 = PI/2 - x2 + PI/2 |
| 131 |
SINF_X_PLUS_PI_2_CORE_X32 |
| 132 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst])") |
| 133 |
__ASM_EMIT("vmovups %%zmm4, 0x40(%[dst])") |
| 134 |
__ASM_EMIT("add $0x80, %[src]") |
| 135 |
__ASM_EMIT("add $0x80, %[dst]") |
| 136 |
__ASM_EMIT("sub $32, %[count]") |
| 137 |
__ASM_EMIT("jae 1b") |
| 138 |
__ASM_EMIT("2:") |
| 139 |
// x16 block |
| 140 |
__ASM_EMIT("add $16, %[count]") |
| 141 |
__ASM_EMIT("jl 4f") |
| 142 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%zmm1") // zmm1 = PI |
| 143 |
__ASM_EMIT("vsubps 0x00(%[src]), %%zmm1, %%zmm0") // zmm0 = x1 + PI/2 |
| 144 |
SINF_X_PLUS_PI_2_CORE_X16 |
| 145 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst])") |
| 146 |
__ASM_EMIT("add $0x40, %[src]") |
| 147 |
__ASM_EMIT("add $0x40, %[dst]") |
| 148 |
__ASM_EMIT("sub $16, %[count]") |
| 149 |
__ASM_EMIT("4:") |
| 150 |
// x8 block |
| 151 |
__ASM_EMIT("add $8, %[count]") |
| 152 |
__ASM_EMIT("jl 6f") |
| 153 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%ymm1") // ymm1 = PI |
| 154 |
__ASM_EMIT("vsubps 0x00(%[src]), %%ymm1, %%ymm0") // ymm0 = x + PI/2 |
| 155 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 156 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[dst])") |
| 157 |
__ASM_EMIT("add $0x20, %[src]") |
| 158 |
__ASM_EMIT("add $0x20, %[dst]") |
| 159 |
__ASM_EMIT("sub $8, %[count]") |
| 160 |
__ASM_EMIT("6:") |
| 161 |
// x4 block |
| 162 |
__ASM_EMIT("add $4, %[count]") |
| 163 |
__ASM_EMIT("jl 8f") |
| 164 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%xmm1") // xmm1 = PI |
| 165 |
__ASM_EMIT("vsubps 0x00(%[src]), %%xmm1, %%xmm0") // xmm0 = x + PI/2 |
| 166 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 167 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst])") |
| 168 |
__ASM_EMIT("add $0x10, %[src]") |
| 169 |
__ASM_EMIT("add $0x10, %[dst]") |
| 170 |
__ASM_EMIT("sub $4, %[count]") |
| 171 |
__ASM_EMIT("8:") |
| 172 |
// Tail: 1x-3x block |
| 173 |
__ASM_EMIT("add $4, %[count]") |
| 174 |
__ASM_EMIT("jle 16f") |
| 175 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%xmm1") // xmm1 = PI/2 |
| 176 |
__ASM_EMIT("test $1, %[count]") |
| 177 |
__ASM_EMIT("jz 10f") |
| 178 |
__ASM_EMIT("vmovss 0x00(%[src]), %%xmm0") |
| 179 |
__ASM_EMIT("add $4, %[src]") |
| 180 |
__ASM_EMIT("10:") |
| 181 |
__ASM_EMIT("test $2, %[count]") |
| 182 |
__ASM_EMIT("jz 12f") |
| 183 |
__ASM_EMIT("vmovhps 0x00(%[src]), %%xmm0, %%xmm0") |
| 184 |
__ASM_EMIT("12:") |
| 185 |
__ASM_EMIT("vsubps %%xmm0, %%xmm1, %%xmm0") // xmm0 = x + PI/2 |
| 186 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 187 |
__ASM_EMIT("test $1, %[count]") |
| 188 |
__ASM_EMIT("jz 14f") |
| 189 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[dst])") |
| 190 |
__ASM_EMIT("add $4, %[dst]") |
| 191 |
__ASM_EMIT("14:") |
| 192 |
__ASM_EMIT("test $2, %[count]") |
| 193 |
__ASM_EMIT("jz 16f") |
| 194 |
__ASM_EMIT("vmovhps %%xmm0, 0x00(%[dst])") |
| 195 |
// End |
| 196 |
__ASM_EMIT("16:") |
| 197 |
|
| 198 |
: [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) |
| 199 |
: [S2C] "o" (sinf_const) |
| 200 |
: "cc", "memory", |
| 201 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 202 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 203 |
"%k1", "%k2" |
| 204 |
); |
| 205 |
} |
| 206 |
|
| 207 |
void cosf_kp1(float *dst, float k, float p, size_t count) |
| 208 |
{ |
| 209 |
ARCH_X86_ASM( |
| 210 |
// Prepare |
| 211 |
__ASM_EMIT("vbroadcastss %[k], %%zmm6") // zmm6 = k |
| 212 |
__ASM_EMIT("vbroadcastss %[p], %%zmm7") // zmm7 = p |
| 213 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%zmm0") // zmm0 = PI |
| 214 |
__ASM_EMIT("vmovaps 0x000 + %[S2KP], %%zmm4") // zmm4 = i = 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
| 215 |
__ASM_EMIT("vmovaps 0x0c0 + %[S2KP], %%zmm5") // zmm5 = step = 16 ... |
| 216 |
__ASM_EMIT("vsubps %%zmm7, %%zmm0, %%zmm7") // zmm7 = PI/2 - p + PI/2 ... |
| 217 |
// x16 blocks |
| 218 |
__ASM_EMIT("sub $16, %[count]") |
| 219 |
__ASM_EMIT("jb 2f") |
| 220 |
__ASM_EMIT("1:") |
| 221 |
__ASM_EMIT("vmulps %%zmm4, %%zmm6, %%zmm0") // zmm0 = k*i |
| 222 |
__ASM_EMIT("vaddps %%zmm5, %%zmm4, %%zmm4") // zmm4 = i + step |
| 223 |
__ASM_EMIT("vsubps %%zmm0, %%zmm7, %%zmm0") // zmm0 = PI/2 - (k*i + p) + PI/2 |
| 224 |
SINF_X_PLUS_PI_2_CORE_X16 |
| 225 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst])") |
| 226 |
__ASM_EMIT("add $0x40, %[dst]") |
| 227 |
__ASM_EMIT("sub $16, %[count]") |
| 228 |
__ASM_EMIT("jae 1b") |
| 229 |
__ASM_EMIT("2:") |
| 230 |
// x8 blocks |
| 231 |
__ASM_EMIT("add $8, %[count]") |
| 232 |
__ASM_EMIT("jl 4f") |
| 233 |
__ASM_EMIT("vmulps %%ymm4, %%ymm6, %%ymm0") // ymm0 = k*i |
| 234 |
__ASM_EMIT("vaddps 0x100 + %[S2KP], %%ymm4, %%ymm4") // ymm4 = i + step |
| 235 |
__ASM_EMIT("vsubps %%ymm0, %%ymm7, %%ymm0") // ymm0 = PI/2 - (k*i + p) + PI/2 |
| 236 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 237 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[dst])") |
| 238 |
__ASM_EMIT("add $0x20, %[dst]") |
| 239 |
__ASM_EMIT("sub $8, %[count]") |
| 240 |
__ASM_EMIT("4:") |
| 241 |
// x4 block |
| 242 |
__ASM_EMIT("add $4, %[count]") |
| 243 |
__ASM_EMIT("jl 6f") |
| 244 |
__ASM_EMIT("vmulps %%xmm4, %%xmm6, %%xmm0") // xmm0 = k*i |
| 245 |
__ASM_EMIT("vaddps 0x120 + %[S2KP], %%xmm4, %%xmm4") // xmm4 = i + step |
| 246 |
__ASM_EMIT("vsubps %%xmm0, %%xmm7, %%xmm0") // xmm0 = PI/2 - (k*i + p) + PI/2 |
| 247 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 248 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst])") |
| 249 |
__ASM_EMIT("add $0x10, %[dst]") |
| 250 |
__ASM_EMIT("sub $4, %[count]") |
| 251 |
__ASM_EMIT("6:") |
| 252 |
// Tail: 1x-3x block |
| 253 |
__ASM_EMIT("add $4, %[count]") |
| 254 |
__ASM_EMIT("jle 10f") |
| 255 |
__ASM_EMIT("vmulps %%xmm4, %%xmm6, %%xmm0") // xmm0 = k*i |
| 256 |
__ASM_EMIT("vsubps %%xmm0, %%xmm7, %%xmm0") // xmm0 = PI/2 - (k*i + p) + PI/2 |
| 257 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 258 |
__ASM_EMIT("test $1, %[count]") |
| 259 |
__ASM_EMIT("jz 8f") |
| 260 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[dst])") |
| 261 |
__ASM_EMIT("vshufps $0x39, %%xmm0, %%xmm0, %%xmm0") |
| 262 |
__ASM_EMIT("add $4, %[dst]") |
| 263 |
__ASM_EMIT("8:") |
| 264 |
__ASM_EMIT("test $2, %[count]") |
| 265 |
__ASM_EMIT("jz 10f") |
| 266 |
__ASM_EMIT("vmovlps %%xmm0, 0x00(%[dst])") |
| 267 |
// End |
| 268 |
__ASM_EMIT("10:") |
| 269 |
|
| 270 |
: [dst] "+r" (dst), [count] "+r" (count) |
| 271 |
: [S2C] "o" (sinf_const), |
| 272 |
[S2KP] "o" (kp_gen_const), |
| 273 |
[k] "m" (k), |
| 274 |
[p] "m" (p) |
| 275 |
: "cc", "memory", |
| 276 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 277 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 278 |
"%k1", "%k2" |
| 279 |
); |
| 280 |
} |
| 281 |
|
| 282 |
void x64_cosf_kp1(float *dst, float k, float p, size_t count) |
| 283 |
{ |
| 284 |
ARCH_X86_64_ASM( |
| 285 |
// Prepare |
| 286 |
__ASM_EMIT("vbroadcastss %[k], %%zmm11") // zmm11 = k |
| 287 |
__ASM_EMIT("vbroadcastss %[p], %%zmm12") // zmm12 = p |
| 288 |
__ASM_EMIT("vmovaps 0x040 + %[S2C], %%zmm0") // zmm0 = PI |
| 289 |
__ASM_EMIT("vmovaps 0x000 + %[S2KP], %%zmm8") // zmm8 = i0 = 0 1 2 3 ... |
| 290 |
__ASM_EMIT("vmovaps 0x040 + %[S2KP], %%zmm9") // zmm9 = i1 = 16 17 18 19 ... |
| 291 |
__ASM_EMIT("vmovaps 0x080 + %[S2KP], %%zmm10") // zmm10 = step = 32 ... |
| 292 |
__ASM_EMIT("vsubps %%zmm12, %%zmm0, %%zmm12") // zmm12 = PI/2 - p + PI/2 ... |
| 293 |
// x32 blocks |
| 294 |
__ASM_EMIT("sub $32, %[count]") |
| 295 |
__ASM_EMIT("jb 2f") |
| 296 |
__ASM_EMIT("1:") |
| 297 |
__ASM_EMIT("vmulps %%zmm8, %%zmm11, %%zmm0") // zmm0 = k*i0 |
| 298 |
__ASM_EMIT("vmulps %%zmm9, %%zmm11, %%zmm4") // zmm4 = k*i1 |
| 299 |
__ASM_EMIT("vaddps %%zmm10, %%zmm8, %%zmm8") // zmm8 = i0 + step |
| 300 |
__ASM_EMIT("vaddps %%zmm10, %%zmm9, %%zmm9") // zmm9 = i1 + step |
| 301 |
__ASM_EMIT("vsubps %%zmm0, %%zmm12, %%zmm0") // zmm0 = PI/2 - (k*i0 + p) + PI/2 |
| 302 |
__ASM_EMIT("vsubps %%zmm4, %%zmm12, %%zmm4") // zmm4 = PI/2 - (k*i1 + p) + PI/2 |
| 303 |
SINF_X_PLUS_PI_2_CORE_X32 |
| 304 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst])") |
| 305 |
__ASM_EMIT("vmovups %%zmm4, 0x40(%[dst])") |
| 306 |
__ASM_EMIT("add $0x80, %[dst]") |
| 307 |
__ASM_EMIT("sub $32, %[count]") |
| 308 |
__ASM_EMIT("jae 1b") |
| 309 |
__ASM_EMIT("2:") |
| 310 |
// x16 blocks |
| 311 |
__ASM_EMIT("add $16, %[count]") |
| 312 |
__ASM_EMIT("jl 4f") |
| 313 |
__ASM_EMIT("vmulps %%zmm8, %%zmm11, %%zmm0") // zmm0 = k*i0 |
| 314 |
__ASM_EMIT("vaddps 0x0c0 + %[S2KP], %%zmm8, %%zmm8") // zmm8 = i0 + step |
| 315 |
__ASM_EMIT("vsubps %%zmm0, %%zmm12, %%zmm0") // zmm0 = PI/2 - (k*i0 + p) + PI/2 |
| 316 |
SINF_X_PLUS_PI_2_CORE_X16 |
| 317 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst])") |
| 318 |
__ASM_EMIT("add $0x40, %[dst]") |
| 319 |
__ASM_EMIT("sub $16, %[count]") |
| 320 |
__ASM_EMIT("4:") |
| 321 |
// x8 block |
| 322 |
__ASM_EMIT("add $8, %[count]") |
| 323 |
__ASM_EMIT("jl 6f") |
| 324 |
__ASM_EMIT("vmulps %%ymm8, %%ymm11, %%ymm0") // ymm0 = k*i0 |
| 325 |
__ASM_EMIT("vaddps 0x100 + %[S2KP], %%ymm8, %%ymm8") // ymm8 = i0 + step |
| 326 |
__ASM_EMIT("vsubps %%ymm0, %%ymm12, %%ymm0") // ymm0 = PI/2 - (k*i0 + p) + PI/2 |
| 327 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 328 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[dst])") |
| 329 |
__ASM_EMIT("add $0x20, %[dst]") |
| 330 |
__ASM_EMIT("sub $8, %[count]") |
| 331 |
__ASM_EMIT("6:") |
| 332 |
// x4 block |
| 333 |
__ASM_EMIT("add $4, %[count]") |
| 334 |
__ASM_EMIT("jl 8f") |
| 335 |
__ASM_EMIT("vmulps %%xmm8, %%xmm11, %%xmm0") // xmm0 = k*i0 |
| 336 |
__ASM_EMIT("vaddps 0x120 + %[S2KP], %%xmm8, %%xmm8") // xmm8 = i0 + step |
| 337 |
__ASM_EMIT("vsubps %%xmm0, %%xmm12, %%xmm0") // xmm0 = PI/2 - (k*i0 + p) + PI/2 |
| 338 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 339 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst])") |
| 340 |
__ASM_EMIT("add $0x10, %[dst]") |
| 341 |
__ASM_EMIT("sub $4, %[count]") |
| 342 |
__ASM_EMIT("8:") |
| 343 |
// Tail: 1x-3x block |
| 344 |
__ASM_EMIT("add $4, %[count]") |
| 345 |
__ASM_EMIT("jle 12f") |
| 346 |
__ASM_EMIT("vmulps %%xmm8, %%xmm11, %%xmm0") // xmm0 = k*i0 |
| 347 |
__ASM_EMIT("vsubps %%xmm0, %%xmm12, %%xmm0") // xmm0 = PI/2 - (k*i0 + p) + PI/2 |
| 348 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 349 |
__ASM_EMIT("test $1, %[count]") |
| 350 |
__ASM_EMIT("jz 10f") |
| 351 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[dst])") |
| 352 |
__ASM_EMIT("vshufps $0x39, %%xmm0, %%xmm0, %%xmm0") |
| 353 |
__ASM_EMIT("add $4, %[dst]") |
| 354 |
__ASM_EMIT("10:") |
| 355 |
__ASM_EMIT("test $2, %[count]") |
| 356 |
__ASM_EMIT("jz 12f") |
| 357 |
__ASM_EMIT("vmovlps %%xmm0, 0x00(%[dst])") |
| 358 |
// End |
| 359 |
__ASM_EMIT("12:") |
| 360 |
|
| 361 |
: [dst] "+r" (dst), [count] "+r" (count) |
| 362 |
: [S2C] "o" (sinf_const), |
| 363 |
[S2KP] "o" (kp_gen_const), |
| 364 |
[k] "m" (k), |
| 365 |
[p] "m" (p) |
| 366 |
: "cc", "memory", |
| 367 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 368 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 369 |
"%xmm8", "%xmm9", "%xmm10", "%xmm11", |
| 370 |
"%xmm12", |
| 371 |
"%k1", "%k2" |
| 372 |
); |
| 373 |
} |
| 374 |
|
| 375 |
} /* namespace avx512 */ |
| 376 |
} /* namespace lsp */ |
| 377 |
|
| 378 |
|
| 379 |
|
| 380 |
|
| 381 |
#endif /* INCLUDE_PRIVATE_DSP_ARCH_X86_AVX512_PMATH_COS_H_ */ |