| 1 |
/* |
| 2 |
* Copyright (C) 2023 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2023 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 28 окт. 2023 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_AVX512_PMATH_OP_VV_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_AVX512_PMATH_OP_VV_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 |
namespace lsp |
| 30 |
{ |
| 31 |
namespace avx512 |
| 32 |
{ |
| 33 |
#define OP_VV_CORE(DST, SRC1, SRC2, OP) \ |
| 34 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 35 |
/* 128x blocks */ \ |
| 36 |
__ASM_EMIT("sub $128, %[count]") \ |
| 37 |
__ASM_EMIT("jb 2f") \ |
| 38 |
__ASM_EMIT("1:") \ |
| 39 |
__ASM_EMIT("vmovups 0x000(%[" SRC1 "], %[off]), %%zmm0") \ |
| 40 |
__ASM_EMIT("vmovups 0x040(%[" SRC1 "], %[off]), %%zmm1") \ |
| 41 |
__ASM_EMIT("vmovups 0x080(%[" SRC1 "], %[off]), %%zmm2") \ |
| 42 |
__ASM_EMIT("vmovups 0x0c0(%[" SRC1 "], %[off]), %%zmm3") \ |
| 43 |
__ASM_EMIT("vmovups 0x100(%[" SRC1 "], %[off]), %%zmm4") \ |
| 44 |
__ASM_EMIT("vmovups 0x140(%[" SRC1 "], %[off]), %%zmm5") \ |
| 45 |
__ASM_EMIT("vmovups 0x180(%[" SRC1 "], %[off]), %%zmm6") \ |
| 46 |
__ASM_EMIT("vmovups 0x1c0(%[" SRC1 "], %[off]), %%zmm7") \ |
| 47 |
__ASM_EMIT(OP "ps 0x000(%[" SRC2 "], %[off]), %%zmm0, %%zmm0") \ |
| 48 |
__ASM_EMIT(OP "ps 0x040(%[" SRC2 "], %[off]), %%zmm1, %%zmm1") \ |
| 49 |
__ASM_EMIT(OP "ps 0x080(%[" SRC2 "], %[off]), %%zmm2, %%zmm2") \ |
| 50 |
__ASM_EMIT(OP "ps 0x0c0(%[" SRC2 "], %[off]), %%zmm3, %%zmm3") \ |
| 51 |
__ASM_EMIT(OP "ps 0x100(%[" SRC2 "], %[off]), %%zmm4, %%zmm4") \ |
| 52 |
__ASM_EMIT(OP "ps 0x140(%[" SRC2 "], %[off]), %%zmm5, %%zmm5") \ |
| 53 |
__ASM_EMIT(OP "ps 0x180(%[" SRC2 "], %[off]), %%zmm6, %%zmm6") \ |
| 54 |
__ASM_EMIT(OP "ps 0x1c0(%[" SRC2 "], %[off]), %%zmm7, %%zmm7") \ |
| 55 |
__ASM_EMIT("vmovups %%zmm0, 0x000(%[" DST "], %[off])") \ |
| 56 |
__ASM_EMIT("vmovups %%zmm1, 0x040(%[" DST "], %[off])") \ |
| 57 |
__ASM_EMIT("vmovups %%zmm2, 0x080(%[" DST "], %[off])") \ |
| 58 |
__ASM_EMIT("vmovups %%zmm3, 0x0c0(%[" DST "], %[off])") \ |
| 59 |
__ASM_EMIT("vmovups %%zmm4, 0x100(%[" DST "], %[off])") \ |
| 60 |
__ASM_EMIT("vmovups %%zmm5, 0x140(%[" DST "], %[off])") \ |
| 61 |
__ASM_EMIT("vmovups %%zmm6, 0x180(%[" DST "], %[off])") \ |
| 62 |
__ASM_EMIT("vmovups %%zmm7, 0x1c0(%[" DST "], %[off])") \ |
| 63 |
__ASM_EMIT("add $0x200, %[off]") \ |
| 64 |
__ASM_EMIT("sub $128, %[count]") \ |
| 65 |
__ASM_EMIT("jae 1b") \ |
| 66 |
__ASM_EMIT("2:") \ |
| 67 |
/* 64x blocks */ \ |
| 68 |
__ASM_EMIT("add $64, %[count]") \ |
| 69 |
__ASM_EMIT("jl 4f") \ |
| 70 |
__ASM_EMIT("vmovups 0x000(%[" SRC1 "], %[off]), %%ymm0") \ |
| 71 |
__ASM_EMIT("vmovups 0x020(%[" SRC1 "], %[off]), %%ymm1") \ |
| 72 |
__ASM_EMIT("vmovups 0x040(%[" SRC1 "], %[off]), %%ymm2") \ |
| 73 |
__ASM_EMIT("vmovups 0x060(%[" SRC1 "], %[off]), %%ymm3") \ |
| 74 |
__ASM_EMIT("vmovups 0x080(%[" SRC1 "], %[off]), %%ymm4") \ |
| 75 |
__ASM_EMIT("vmovups 0x0a0(%[" SRC1 "], %[off]), %%ymm5") \ |
| 76 |
__ASM_EMIT("vmovups 0x0c0(%[" SRC1 "], %[off]), %%ymm6") \ |
| 77 |
__ASM_EMIT("vmovups 0x0e0(%[" SRC1 "], %[off]), %%ymm7") \ |
| 78 |
__ASM_EMIT(OP "ps 0x000(%[" SRC2 "], %[off]), %%ymm0, %%ymm0") \ |
| 79 |
__ASM_EMIT(OP "ps 0x020(%[" SRC2 "], %[off]), %%ymm1, %%ymm1") \ |
| 80 |
__ASM_EMIT(OP "ps 0x040(%[" SRC2 "], %[off]), %%ymm2, %%ymm2") \ |
| 81 |
__ASM_EMIT(OP "ps 0x060(%[" SRC2 "], %[off]), %%ymm3, %%ymm3") \ |
| 82 |
__ASM_EMIT(OP "ps 0x080(%[" SRC2 "], %[off]), %%ymm4, %%ymm4") \ |
| 83 |
__ASM_EMIT(OP "ps 0x0a0(%[" SRC2 "], %[off]), %%ymm5, %%ymm5") \ |
| 84 |
__ASM_EMIT(OP "ps 0x0c0(%[" SRC2 "], %[off]), %%ymm6, %%ymm6") \ |
| 85 |
__ASM_EMIT(OP "ps 0x0e0(%[" SRC2 "], %[off]), %%ymm7, %%ymm7") \ |
| 86 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[" DST "], %[off])") \ |
| 87 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[" DST "], %[off])") \ |
| 88 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[" DST "], %[off])") \ |
| 89 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[" DST "], %[off])") \ |
| 90 |
__ASM_EMIT("vmovups %%ymm4, 0x080(%[" DST "], %[off])") \ |
| 91 |
__ASM_EMIT("vmovups %%ymm5, 0x0a0(%[" DST "], %[off])") \ |
| 92 |
__ASM_EMIT("vmovups %%ymm6, 0x0c0(%[" DST "], %[off])") \ |
| 93 |
__ASM_EMIT("vmovups %%ymm7, 0x0e0(%[" DST "], %[off])") \ |
| 94 |
__ASM_EMIT("sub $64, %[count]") \ |
| 95 |
__ASM_EMIT("add $0x100, %[off]") \ |
| 96 |
__ASM_EMIT("4:") \ |
| 97 |
/* 32x block */ \ |
| 98 |
__ASM_EMIT("add $32, %[count]") \ |
| 99 |
__ASM_EMIT("jl 6f") \ |
| 100 |
__ASM_EMIT("vmovups 0x000(%[" SRC1 "], %[off]), %%ymm0") \ |
| 101 |
__ASM_EMIT("vmovups 0x020(%[" SRC1 "], %[off]), %%ymm1") \ |
| 102 |
__ASM_EMIT("vmovups 0x040(%[" SRC1 "], %[off]), %%ymm2") \ |
| 103 |
__ASM_EMIT("vmovups 0x060(%[" SRC1 "], %[off]), %%ymm3") \ |
| 104 |
__ASM_EMIT(OP "ps 0x000(%[" SRC2 "], %[off]), %%ymm0, %%ymm0") \ |
| 105 |
__ASM_EMIT(OP "ps 0x020(%[" SRC2 "], %[off]), %%ymm1, %%ymm1") \ |
| 106 |
__ASM_EMIT(OP "ps 0x040(%[" SRC2 "], %[off]), %%ymm2, %%ymm2") \ |
| 107 |
__ASM_EMIT(OP "ps 0x060(%[" SRC2 "], %[off]), %%ymm3, %%ymm3") \ |
| 108 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[" DST "], %[off])") \ |
| 109 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[" DST "], %[off])") \ |
| 110 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[" DST "], %[off])") \ |
| 111 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[" DST "], %[off])") \ |
| 112 |
__ASM_EMIT("sub $32, %[count]") \ |
| 113 |
__ASM_EMIT("add $0x80, %[off]") \ |
| 114 |
__ASM_EMIT("6:") \ |
| 115 |
/* 16x block */ \ |
| 116 |
__ASM_EMIT("add $16, %[count]") \ |
| 117 |
__ASM_EMIT("jl 8f") \ |
| 118 |
__ASM_EMIT("vmovups 0x000(%[" SRC1 "], %[off]), %%ymm0") \ |
| 119 |
__ASM_EMIT("vmovups 0x020(%[" SRC1 "], %[off]), %%ymm1") \ |
| 120 |
__ASM_EMIT(OP "ps 0x000(%[" SRC2 "], %[off]), %%ymm0, %%ymm0") \ |
| 121 |
__ASM_EMIT(OP "ps 0x020(%[" SRC2 "], %[off]), %%ymm1, %%ymm1") \ |
| 122 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[" DST "], %[off])") \ |
| 123 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[" DST "], %[off])") \ |
| 124 |
__ASM_EMIT("sub $16, %[count]") \ |
| 125 |
__ASM_EMIT("add $0x40, %[off]") \ |
| 126 |
__ASM_EMIT("8:") \ |
| 127 |
/* 8x block */ \ |
| 128 |
__ASM_EMIT("add $8, %[count]") \ |
| 129 |
__ASM_EMIT("jl 10f") \ |
| 130 |
__ASM_EMIT("vmovups 0x000(%[" SRC1 "], %[off]), %%ymm0") \ |
| 131 |
__ASM_EMIT(OP "ps 0x000(%[" SRC2 "], %[off]), %%ymm0, %%ymm0") \ |
| 132 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[" DST "], %[off])") \ |
| 133 |
__ASM_EMIT("sub $8, %[count]") \ |
| 134 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 135 |
__ASM_EMIT("10:") \ |
| 136 |
/* 4x block */ \ |
| 137 |
__ASM_EMIT("add $4, %[count]") \ |
| 138 |
__ASM_EMIT("jl 12f") \ |
| 139 |
__ASM_EMIT("vmovups 0x000(%[" SRC1 "], %[off]), %%xmm0") \ |
| 140 |
__ASM_EMIT(OP "ps 0x000(%[" SRC2 "], %[off]), %%xmm0, %%xmm0") \ |
| 141 |
__ASM_EMIT("vmovups %%xmm0, 0x000(%[" DST "], %[off])") \ |
| 142 |
__ASM_EMIT("add $0x10, %[off]") \ |
| 143 |
__ASM_EMIT("sub $4, %[count]") \ |
| 144 |
__ASM_EMIT("12:") \ |
| 145 |
/* 1x blocks */ \ |
| 146 |
__ASM_EMIT("add $3, %[count]") \ |
| 147 |
__ASM_EMIT("jl 14f") \ |
| 148 |
__ASM_EMIT("13:") \ |
| 149 |
__ASM_EMIT("vmovss 0x000(%[" SRC1 "], %[off]), %%xmm0") \ |
| 150 |
__ASM_EMIT(OP "ss 0x000(%[" SRC2 "], %[off]), %%xmm0, %%xmm0") \ |
| 151 |
__ASM_EMIT("vmovss %%xmm0, 0x000(%[" DST "], %[off])") \ |
| 152 |
__ASM_EMIT("add $0x04, %[off]") \ |
| 153 |
__ASM_EMIT("dec %[count]") \ |
| 154 |
__ASM_EMIT("jge 13b") \ |
| 155 |
__ASM_EMIT("14:") |
| 156 |
|
| 157 |
void add2(float *dst, const float *src, size_t count) |
| 158 |
{ |
| 159 |
IF_ARCH_X86(size_t off); |
| 160 |
ARCH_X86_ASM |
| 161 |
( |
| 162 |
OP_VV_CORE("dst", "dst", "src", "vadd") |
| 163 |
: [off] "=&r" (off), [count] "+r" (count) |
| 164 |
: [dst] "r" (dst), [src] "r" (src) |
| 165 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 166 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 167 |
); |
| 168 |
} |
| 169 |
|
| 170 |
void sub2(float *dst, const float *src, size_t count) |
| 171 |
{ |
| 172 |
IF_ARCH_X86(size_t off); |
| 173 |
ARCH_X86_ASM |
| 174 |
( |
| 175 |
OP_VV_CORE("dst", "dst", "src", "vsub") |
| 176 |
: [off] "=&r" (off), [count] "+r" (count) |
| 177 |
: [dst] "r" (dst), [src] "r" (src) |
| 178 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 179 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 180 |
); |
| 181 |
} |
| 182 |
|
| 183 |
void rsub2(float *dst, const float *src, size_t count) |
| 184 |
{ |
| 185 |
IF_ARCH_X86(size_t off); |
| 186 |
ARCH_X86_ASM |
| 187 |
( |
| 188 |
OP_VV_CORE("dst", "src", "dst", "vsub") |
| 189 |
: [off] "=&r" (off), [count] "+r" (count) |
| 190 |
: [dst] "r" (dst), [src] "r" (src) |
| 191 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 192 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 193 |
); |
| 194 |
} |
| 195 |
|
| 196 |
void mul2(float *dst, const float *src, size_t count) |
| 197 |
{ |
| 198 |
IF_ARCH_X86(size_t off); |
| 199 |
ARCH_X86_ASM |
| 200 |
( |
| 201 |
OP_VV_CORE("dst", "dst", "src", "vmul") |
| 202 |
: [off] "=&r" (off), [count] "+r" (count) |
| 203 |
: [dst] "r" (dst), [src] "r" (src) |
| 204 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 205 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 206 |
); |
| 207 |
} |
| 208 |
|
| 209 |
void div2(float *dst, const float *src, size_t count) |
| 210 |
{ |
| 211 |
IF_ARCH_X86(size_t off); |
| 212 |
ARCH_X86_ASM |
| 213 |
( |
| 214 |
OP_VV_CORE("dst", "dst", "src", "vdiv") |
| 215 |
: [off] "=&r" (off), [count] "+r" (count) |
| 216 |
: [dst] "r" (dst), [src] "r" (src) |
| 217 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 218 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 219 |
); |
| 220 |
} |
| 221 |
|
| 222 |
void rdiv2(float *dst, const float *src, size_t count) |
| 223 |
{ |
| 224 |
IF_ARCH_X86(size_t off); |
| 225 |
ARCH_X86_ASM |
| 226 |
( |
| 227 |
OP_VV_CORE("dst", "src", "dst", "vdiv") |
| 228 |
: [off] "=&r" (off), [count] "+r" (count) |
| 229 |
: [dst] "r" (dst), [src] "r" (src) |
| 230 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 231 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 232 |
); |
| 233 |
} |
| 234 |
|
| 235 |
void add3(float *dst, const float *src1, const float *src2, size_t count) |
| 236 |
{ |
| 237 |
IF_ARCH_X86(size_t off); |
| 238 |
ARCH_X86_ASM |
| 239 |
( |
| 240 |
OP_VV_CORE("dst", "src1", "src2", "vadd") |
| 241 |
: [off] "=&r" (off), [count] "+r" (count) |
| 242 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2) |
| 243 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 244 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 245 |
); |
| 246 |
} |
| 247 |
|
| 248 |
void sub3(float *dst, const float *src1, const float *src2, size_t count) |
| 249 |
{ |
| 250 |
IF_ARCH_X86(size_t off); |
| 251 |
ARCH_X86_ASM |
| 252 |
( |
| 253 |
OP_VV_CORE("dst", "src1", "src2", "vsub") |
| 254 |
: [off] "=&r" (off), [count] "+r" (count) |
| 255 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2) |
| 256 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 257 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 258 |
); |
| 259 |
} |
| 260 |
|
| 261 |
void mul3(float *dst, const float *src1, const float *src2, size_t count) |
| 262 |
{ |
| 263 |
IF_ARCH_X86(size_t off); |
| 264 |
ARCH_X86_ASM |
| 265 |
( |
| 266 |
OP_VV_CORE("dst", "src1", "src2", "vmul") |
| 267 |
: [off] "=&r" (off), [count] "+r" (count) |
| 268 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2) |
| 269 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 270 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 271 |
); |
| 272 |
} |
| 273 |
|
| 274 |
void div3(float *dst, const float *src1, const float *src2, size_t count) |
| 275 |
{ |
| 276 |
IF_ARCH_X86(size_t off); |
| 277 |
ARCH_X86_ASM |
| 278 |
( |
| 279 |
OP_VV_CORE("dst", "src1", "src2", "vdiv") |
| 280 |
: [off] "=&r" (off), [count] "+r" (count) |
| 281 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2) |
| 282 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 283 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 284 |
); |
| 285 |
} |
| 286 |
|
| 287 |
#undef OP_VV_CORE |
| 288 |
|
| 289 |
#define FMOD_VV_CORE(DST, SRC1, SRC2) \ |
| 290 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 291 |
/* x32 blocks */ \ |
| 292 |
__ASM_EMIT("sub $32, %[count]") \ |
| 293 |
__ASM_EMIT("jb 2f") \ |
| 294 |
__ASM_EMIT("1:") \ |
| 295 |
__ASM_EMIT("vmovups 0x00(%[" SRC1 "], %[off]), %%zmm2") \ |
| 296 |
__ASM_EMIT("vmovups 0x40(%[" SRC1 "], %[off]), %%zmm3") \ |
| 297 |
__ASM_EMIT("vmovups 0x00(%[" SRC2 "], %[off]), %%zmm0") \ |
| 298 |
__ASM_EMIT("vmovups 0x40(%[" SRC2 "], %[off]), %%zmm1") \ |
| 299 |
__ASM_EMIT("vdivps %%zmm0, %%zmm2, %%zmm4") \ |
| 300 |
__ASM_EMIT("vdivps %%zmm1, %%zmm3, %%zmm5") \ |
| 301 |
__ASM_EMIT("vcvttps2dq %%zmm4, %%zmm4") \ |
| 302 |
__ASM_EMIT("vcvttps2dq %%zmm5, %%zmm5") \ |
| 303 |
__ASM_EMIT("vcvtdq2ps %%zmm4, %%zmm4") \ |
| 304 |
__ASM_EMIT("vcvtdq2ps %%zmm5, %%zmm5") \ |
| 305 |
__ASM_EMIT("vfnmadd231ps %%zmm0, %%zmm4, %%zmm2") \ |
| 306 |
__ASM_EMIT("vfnmadd231ps %%zmm1, %%zmm5, %%zmm3") \ |
| 307 |
__ASM_EMIT("vmovups %%zmm2, 0x00(%[" DST "], %[off])") \ |
| 308 |
__ASM_EMIT("vmovups %%zmm3, 0x40(%[" DST "], %[off])") \ |
| 309 |
__ASM_EMIT("add $0x80, %[off]") \ |
| 310 |
__ASM_EMIT("sub $32, %[count]") \ |
| 311 |
__ASM_EMIT("jae 1b") \ |
| 312 |
__ASM_EMIT("2:") \ |
| 313 |
/* x16 blocks */ \ |
| 314 |
__ASM_EMIT("add $16, %[count]") \ |
| 315 |
__ASM_EMIT("jl 4f") \ |
| 316 |
__ASM_EMIT("vmovups 0x00(%[" SRC1 "], %[off]), %%ymm2") \ |
| 317 |
__ASM_EMIT("vmovups 0x20(%[" SRC1 "], %[off]), %%ymm3") \ |
| 318 |
__ASM_EMIT("vmovups 0x00(%[" SRC2 "], %[off]), %%ymm0") \ |
| 319 |
__ASM_EMIT("vmovups 0x20(%[" SRC2 "], %[off]), %%ymm1") \ |
| 320 |
__ASM_EMIT("vdivps %%ymm0, %%ymm2, %%ymm4") \ |
| 321 |
__ASM_EMIT("vdivps %%ymm1, %%ymm3, %%ymm5") \ |
| 322 |
__ASM_EMIT("vcvttps2dq %%ymm4, %%ymm4") \ |
| 323 |
__ASM_EMIT("vcvttps2dq %%ymm5, %%ymm5") \ |
| 324 |
__ASM_EMIT("vcvtdq2ps %%ymm4, %%ymm4") \ |
| 325 |
__ASM_EMIT("vcvtdq2ps %%ymm5, %%ymm5") \ |
| 326 |
__ASM_EMIT("vfnmadd231ps %%ymm0, %%ymm4, %%ymm2") \ |
| 327 |
__ASM_EMIT("vfnmadd231ps %%ymm1, %%ymm5, %%ymm3") \ |
| 328 |
__ASM_EMIT("vmovups %%ymm2, 0x00(%[" DST "], %[off])") \ |
| 329 |
__ASM_EMIT("vmovups %%ymm3, 0x20(%[" DST "], %[off])") \ |
| 330 |
__ASM_EMIT("sub $16, %[count]") \ |
| 331 |
__ASM_EMIT("add $0x40, %[off]") \ |
| 332 |
__ASM_EMIT("4:") \ |
| 333 |
/* x8 block */ \ |
| 334 |
__ASM_EMIT("add $8, %[count]") \ |
| 335 |
__ASM_EMIT("jl 6f") \ |
| 336 |
__ASM_EMIT("vmovups 0x00(%[" SRC1 "], %[off]), %%xmm2") \ |
| 337 |
__ASM_EMIT("vmovups 0x10(%[" SRC1 "], %[off]), %%xmm3") \ |
| 338 |
__ASM_EMIT("vmovups 0x00(%[" SRC2 "], %[off]), %%xmm0") \ |
| 339 |
__ASM_EMIT("vmovups 0x10(%[" SRC2 "], %[off]), %%xmm1") \ |
| 340 |
__ASM_EMIT("vdivps %%xmm0, %%xmm2, %%xmm4") \ |
| 341 |
__ASM_EMIT("vdivps %%xmm1, %%xmm3, %%xmm5") \ |
| 342 |
__ASM_EMIT("vcvttps2dq %%xmm4, %%xmm4") \ |
| 343 |
__ASM_EMIT("vcvttps2dq %%xmm5, %%xmm5") \ |
| 344 |
__ASM_EMIT("vcvtdq2ps %%xmm4, %%xmm4") \ |
| 345 |
__ASM_EMIT("vcvtdq2ps %%xmm5, %%xmm5") \ |
| 346 |
__ASM_EMIT("vfnmadd231ps %%xmm0, %%xmm4, %%xmm2") \ |
| 347 |
__ASM_EMIT("vfnmadd231ps %%xmm1, %%xmm5, %%xmm3") \ |
| 348 |
__ASM_EMIT("vmovups %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 349 |
__ASM_EMIT("vmovups %%xmm3, 0x10(%[" DST "], %[off])") \ |
| 350 |
__ASM_EMIT("sub $8, %[count]") \ |
| 351 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 352 |
__ASM_EMIT("6:") \ |
| 353 |
/* x4 block */ \ |
| 354 |
__ASM_EMIT("add $4, %[count]") \ |
| 355 |
__ASM_EMIT("jl 8f") \ |
| 356 |
__ASM_EMIT("vmovups 0x00(%[" SRC1 "], %[off]), %%xmm2") \ |
| 357 |
__ASM_EMIT("vmovups 0x00(%[" SRC2 "], %[off]), %%xmm0") \ |
| 358 |
__ASM_EMIT("vdivps %%xmm0, %%xmm2, %%xmm4") \ |
| 359 |
__ASM_EMIT("vcvttps2dq %%xmm4, %%xmm4") \ |
| 360 |
__ASM_EMIT("vcvtdq2ps %%xmm4, %%xmm4") \ |
| 361 |
__ASM_EMIT("vfnmadd231ps %%xmm0, %%xmm4, %%xmm2") \ |
| 362 |
__ASM_EMIT("vmovups %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 363 |
__ASM_EMIT("sub $4, %[count]") \ |
| 364 |
__ASM_EMIT("add $0x10, %[off]") \ |
| 365 |
__ASM_EMIT("8:") \ |
| 366 |
/* x1 blocks */ \ |
| 367 |
__ASM_EMIT("add $3, %[count]") \ |
| 368 |
__ASM_EMIT("jl 10f") \ |
| 369 |
__ASM_EMIT("9:") \ |
| 370 |
__ASM_EMIT("vmovss 0x00(%[" SRC1 "], %[off]), %%xmm2") \ |
| 371 |
__ASM_EMIT("vmovss 0x00(%[" SRC2 "], %[off]), %%xmm0") \ |
| 372 |
__ASM_EMIT("vdivps %%xmm0, %%xmm2, %%xmm4") \ |
| 373 |
__ASM_EMIT("vcvttps2dq %%xmm4, %%xmm4") \ |
| 374 |
__ASM_EMIT("vcvtdq2ps %%xmm4, %%xmm4") \ |
| 375 |
__ASM_EMIT("vfnmadd231ps %%xmm0, %%xmm4, %%xmm2") \ |
| 376 |
__ASM_EMIT("vmovss %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 377 |
__ASM_EMIT("add $0x04, %[off]") \ |
| 378 |
__ASM_EMIT("dec %[count]") \ |
| 379 |
__ASM_EMIT("jge 9b") \ |
| 380 |
__ASM_EMIT("10:") |
| 381 |
|
| 382 |
void mod2(float *dst, const float *src, size_t count) |
| 383 |
{ |
| 384 |
IF_ARCH_X86(size_t off); |
| 385 |
ARCH_X86_ASM |
| 386 |
( |
| 387 |
FMOD_VV_CORE("dst", "dst", "src") |
| 388 |
: [off] "=&r" (off), [count] "+r" (count) |
| 389 |
: [dst] "r" (dst), [src] "r" (src) |
| 390 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 391 |
"%xmm4", "%xmm5" |
| 392 |
); |
| 393 |
} |
| 394 |
|
| 395 |
void rmod2(float *dst, const float *src, size_t count) |
| 396 |
{ |
| 397 |
IF_ARCH_X86(size_t off); |
| 398 |
ARCH_X86_ASM |
| 399 |
( |
| 400 |
FMOD_VV_CORE("dst", "src", "dst") |
| 401 |
: [off] "=&r" (off), [count] "+r" (count) |
| 402 |
: [dst] "r" (dst), [src] "r" (src) |
| 403 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 404 |
"%xmm4", "%xmm5" |
| 405 |
); |
| 406 |
} |
| 407 |
|
| 408 |
void mod3(float *dst, const float *src1, const float *src2, size_t count) |
| 409 |
{ |
| 410 |
IF_ARCH_X86(size_t off); |
| 411 |
ARCH_X86_ASM |
| 412 |
( |
| 413 |
FMOD_VV_CORE("dst", "src1", "src2") |
| 414 |
: [off] "=&r" (off), [count] "+r" (count) |
| 415 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2) |
| 416 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 417 |
"%xmm4", "%xmm5" |
| 418 |
); |
| 419 |
} |
| 420 |
|
| 421 |
#undef FMOD_VV_CORE |
| 422 |
} /* namespace avx512 */ |
| 423 |
} /* namespace lsp */ |
| 424 |
|
| 425 |
|
| 426 |
|
| 427 |
|
| 428 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX512_PMATH_OP_VV_H_ */ |