| 1 |
/* |
| 2 |
* Copyright (C) 2024 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2024 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_SSE_PMATH_ABS_VV_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_SSE_PMATH_ABS_VV_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_SSE_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE_IMPL */ |
| 28 |
|
| 29 |
namespace lsp |
| 30 |
{ |
| 31 |
namespace sse |
| 32 |
{ |
| 33 |
IF_ARCH_X86( |
| 34 |
static const uint32_t abs_vv_const[] __lsp_aligned16 = |
| 35 |
{ |
| 36 |
LSP_DSP_VEC4(0x7fffffff) |
| 37 |
}; |
| 38 |
) |
| 39 |
|
| 40 |
#define OP_DSEL(a, b) a |
| 41 |
#define OP_RSEL(a, b) b |
| 42 |
|
| 43 |
#define ABS_CORE(DST, SRC) \ |
| 44 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 45 |
__ASM_EMIT("movaps %[X_SIGN], %%xmm0") \ |
| 46 |
__ASM_EMIT("sub $24, %[count]") \ |
| 47 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") \ |
| 48 |
__ASM_EMIT("jb 2f") \ |
| 49 |
/* 24x blocks */ \ |
| 50 |
__ASM_EMIT("1:") \ |
| 51 |
__ASM_EMIT("movups 0x00(%[" SRC "], %[off]), %%xmm2") \ |
| 52 |
__ASM_EMIT("movups 0x10(%[" SRC "], %[off]), %%xmm3") \ |
| 53 |
__ASM_EMIT("movups 0x20(%[" SRC "], %[off]), %%xmm4") \ |
| 54 |
__ASM_EMIT("movups 0x30(%[" SRC "], %[off]), %%xmm5") \ |
| 55 |
__ASM_EMIT("movups 0x40(%[" SRC "], %[off]), %%xmm6") \ |
| 56 |
__ASM_EMIT("movups 0x50(%[" SRC "], %[off]), %%xmm7") \ |
| 57 |
__ASM_EMIT("andps %%xmm0, %%xmm2") \ |
| 58 |
__ASM_EMIT("andps %%xmm1, %%xmm3") \ |
| 59 |
__ASM_EMIT("andps %%xmm0, %%xmm4") \ |
| 60 |
__ASM_EMIT("andps %%xmm1, %%xmm5") \ |
| 61 |
__ASM_EMIT("andps %%xmm0, %%xmm6") \ |
| 62 |
__ASM_EMIT("andps %%xmm1, %%xmm7") \ |
| 63 |
__ASM_EMIT("movups %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 64 |
__ASM_EMIT("movups %%xmm3, 0x10(%[" DST "], %[off])") \ |
| 65 |
__ASM_EMIT("movups %%xmm4, 0x20(%[" DST "], %[off])") \ |
| 66 |
__ASM_EMIT("movups %%xmm5, 0x30(%[" DST "], %[off])") \ |
| 67 |
__ASM_EMIT("movups %%xmm6, 0x40(%[" DST "], %[off])") \ |
| 68 |
__ASM_EMIT("movups %%xmm7, 0x50(%[" DST "], %[off])") \ |
| 69 |
__ASM_EMIT("add $0x60, %[off]") \ |
| 70 |
__ASM_EMIT("sub $24, %[count]") \ |
| 71 |
__ASM_EMIT("jae 1b") \ |
| 72 |
/* 16x block */ \ |
| 73 |
__ASM_EMIT("2:") \ |
| 74 |
__ASM_EMIT("add $8, %[count]") \ |
| 75 |
__ASM_EMIT("jl 4f") \ |
| 76 |
__ASM_EMIT("movups 0x00(%[" SRC "], %[off]), %%xmm2") \ |
| 77 |
__ASM_EMIT("movups 0x10(%[" SRC "], %[off]), %%xmm3") \ |
| 78 |
__ASM_EMIT("movups 0x20(%[" SRC "], %[off]), %%xmm4") \ |
| 79 |
__ASM_EMIT("movups 0x30(%[" SRC "], %[off]), %%xmm5") \ |
| 80 |
__ASM_EMIT("andps %%xmm0, %%xmm2") \ |
| 81 |
__ASM_EMIT("andps %%xmm1, %%xmm3") \ |
| 82 |
__ASM_EMIT("andps %%xmm0, %%xmm4") \ |
| 83 |
__ASM_EMIT("andps %%xmm1, %%xmm5") \ |
| 84 |
__ASM_EMIT("movups %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 85 |
__ASM_EMIT("movups %%xmm3, 0x10(%[" DST "], %[off])") \ |
| 86 |
__ASM_EMIT("movups %%xmm4, 0x20(%[" DST "], %[off])") \ |
| 87 |
__ASM_EMIT("movups %%xmm5, 0x30(%[" DST "], %[off])") \ |
| 88 |
__ASM_EMIT("sub $16, %[count]") \ |
| 89 |
__ASM_EMIT("add $0x40, %[off]") \ |
| 90 |
/* 8x block */ \ |
| 91 |
__ASM_EMIT("4:") \ |
| 92 |
__ASM_EMIT("add $8, %[count]") \ |
| 93 |
__ASM_EMIT("jl 6f") \ |
| 94 |
__ASM_EMIT("movups 0x00(%[" SRC "], %[off]), %%xmm2") \ |
| 95 |
__ASM_EMIT("movups 0x10(%[" SRC "], %[off]), %%xmm3") \ |
| 96 |
__ASM_EMIT("andps %%xmm0, %%xmm2") \ |
| 97 |
__ASM_EMIT("andps %%xmm1, %%xmm3") \ |
| 98 |
__ASM_EMIT("movups %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 99 |
__ASM_EMIT("movups %%xmm3, 0x10(%[" DST "], %[off])") \ |
| 100 |
__ASM_EMIT("sub $8, %[count]") \ |
| 101 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 102 |
/* 4x block */ \ |
| 103 |
__ASM_EMIT("6:") \ |
| 104 |
__ASM_EMIT("add $4, %[count]") \ |
| 105 |
__ASM_EMIT("jl 8f") \ |
| 106 |
__ASM_EMIT("movups 0x00(%[" SRC "], %[off]), %%xmm2") \ |
| 107 |
__ASM_EMIT("andps %%xmm0, %%xmm2") \ |
| 108 |
__ASM_EMIT("movups %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 109 |
__ASM_EMIT("sub $4, %[count]") \ |
| 110 |
__ASM_EMIT("add $0x10, %[off]") \ |
| 111 |
/* 1x block */ \ |
| 112 |
__ASM_EMIT("8:") \ |
| 113 |
__ASM_EMIT("add $3, %[count]") \ |
| 114 |
__ASM_EMIT("jl 10f") \ |
| 115 |
__ASM_EMIT("9:") \ |
| 116 |
__ASM_EMIT("movss 0x00(%[" SRC "], %[off]), %%xmm2") \ |
| 117 |
__ASM_EMIT("andps %%xmm0, %%xmm2") \ |
| 118 |
__ASM_EMIT("movss %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 119 |
__ASM_EMIT("add $0x04, %[off]") \ |
| 120 |
__ASM_EMIT("dec %[count]") \ |
| 121 |
__ASM_EMIT("jge 9b") \ |
| 122 |
__ASM_EMIT("10:") |
| 123 |
|
| 124 |
void abs1(float *dst, size_t count) |
| 125 |
{ |
| 126 |
IF_ARCH_X86(size_t off); |
| 127 |
ARCH_X86_ASM |
| 128 |
( |
| 129 |
ABS_CORE("dst", "dst") |
| 130 |
: [off] "=&r" (off), [count] "+r" (count) |
| 131 |
: [dst] "r" (dst), |
| 132 |
[X_SIGN] "m" (abs_vv_const) |
| 133 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 134 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 135 |
); |
| 136 |
} |
| 137 |
|
| 138 |
void abs2(float *dst, const float *src, size_t count) |
| 139 |
{ |
| 140 |
IF_ARCH_X86(size_t off); |
| 141 |
ARCH_X86_ASM |
| 142 |
( |
| 143 |
ABS_CORE("dst", "src") |
| 144 |
: [off] "=&r" (off), [count] "+r" (count) |
| 145 |
: [dst] "r" (dst), [src] "r" (src), |
| 146 |
[X_SIGN] "m" (abs_vv_const) |
| 147 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 148 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 149 |
); |
| 150 |
} |
| 151 |
|
| 152 |
#undef ABS_CORE |
| 153 |
|
| 154 |
#define ABS_OP2_CORE(OP, DST, SRC1, SRC2, SEL) \ |
| 155 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 156 |
__ASM_EMIT("movaps %[X_SIGN], %%xmm0") \ |
| 157 |
__ASM_EMIT("sub $12, %[count]") \ |
| 158 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") \ |
| 159 |
__ASM_EMIT("jb 2f") \ |
| 160 |
/* 12x blocks */ \ |
| 161 |
__ASM_EMIT("1:") \ |
| 162 |
__ASM_EMIT("movups 0x00(%[" SEL(SRC2, SRC1) "], %[off]), %%xmm2") \ |
| 163 |
__ASM_EMIT("movups 0x10(%[" SEL(SRC2, SRC1) "], %[off]), %%xmm3") \ |
| 164 |
__ASM_EMIT("movups 0x20(%[" SEL(SRC2, SRC1) "], %[off]), %%xmm4") \ |
| 165 |
__ASM_EMIT("movups 0x00(%[" SEL(SRC1, SRC2) "], %[off]), %%xmm5") \ |
| 166 |
__ASM_EMIT("movups 0x10(%[" SEL(SRC1, SRC2) "], %[off]), %%xmm6") \ |
| 167 |
__ASM_EMIT("movups 0x20(%[" SEL(SRC1, SRC2) "], %[off]), %%xmm7") \ |
| 168 |
__ASM_EMIT("andps %%xmm0, " SEL("%%xmm2", "%%xmm5")) \ |
| 169 |
__ASM_EMIT("andps %%xmm1, " SEL("%%xmm3", "%%xmm6")) \ |
| 170 |
__ASM_EMIT("andps %%xmm0, " SEL("%%xmm4", "%%xmm7")) \ |
| 171 |
__ASM_EMIT(OP "ps %%xmm2, %%xmm5") \ |
| 172 |
__ASM_EMIT(OP "ps %%xmm3, %%xmm6") \ |
| 173 |
__ASM_EMIT(OP "ps %%xmm4, %%xmm7") \ |
| 174 |
__ASM_EMIT("movups %%xmm5, 0x00(%[" DST "], %[off])") \ |
| 175 |
__ASM_EMIT("movups %%xmm6, 0x10(%[" DST "], %[off])") \ |
| 176 |
__ASM_EMIT("movups %%xmm7, 0x20(%[" DST "], %[off])") \ |
| 177 |
__ASM_EMIT("add $0x30, %[off]") \ |
| 178 |
__ASM_EMIT("sub $12, %[count]") \ |
| 179 |
__ASM_EMIT("jae 1b") \ |
| 180 |
/* 8x block */ \ |
| 181 |
__ASM_EMIT("2:") \ |
| 182 |
__ASM_EMIT("add $4, %[count]") \ |
| 183 |
__ASM_EMIT("jl 4f") \ |
| 184 |
__ASM_EMIT("movups 0x00(%[" SEL(SRC2, SRC1) "], %[off]), %%xmm2") \ |
| 185 |
__ASM_EMIT("movups 0x10(%[" SEL(SRC2, SRC1) "], %[off]), %%xmm3") \ |
| 186 |
__ASM_EMIT("movups 0x00(%[" SEL(SRC1, SRC2) "], %[off]), %%xmm4") \ |
| 187 |
__ASM_EMIT("movups 0x10(%[" SEL(SRC1, SRC2) "], %[off]), %%xmm5") \ |
| 188 |
__ASM_EMIT("andps %%xmm0, " SEL("%%xmm2", "%%xmm4")) \ |
| 189 |
__ASM_EMIT("andps %%xmm1, " SEL("%%xmm3", "%%xmm5")) \ |
| 190 |
__ASM_EMIT(OP "ps %%xmm2, %%xmm4") \ |
| 191 |
__ASM_EMIT(OP "ps %%xmm3, %%xmm5") \ |
| 192 |
__ASM_EMIT("movups %%xmm4, 0x00(%[" DST "], %[off])") \ |
| 193 |
__ASM_EMIT("movups %%xmm5, 0x10(%[" DST "], %[off])") \ |
| 194 |
__ASM_EMIT("sub $8, %[count]") \ |
| 195 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 196 |
/* 4x block */ \ |
| 197 |
__ASM_EMIT("4:") \ |
| 198 |
__ASM_EMIT("add $4, %[count]") \ |
| 199 |
__ASM_EMIT("jl 6f") \ |
| 200 |
__ASM_EMIT("movups 0x00(%[" SEL(SRC2, SRC1) "], %[off]), %%xmm2") \ |
| 201 |
__ASM_EMIT("movups 0x00(%[" SEL(SRC1, SRC2) "], %[off]), %%xmm4") \ |
| 202 |
__ASM_EMIT("andps %%xmm0, " SEL("%%xmm2", "%%xmm4")) \ |
| 203 |
__ASM_EMIT(OP "ps %%xmm2, %%xmm4") \ |
| 204 |
__ASM_EMIT("movups %%xmm4, 0x00(%[" DST "], %[off])") \ |
| 205 |
__ASM_EMIT("sub $4, %[count]") \ |
| 206 |
__ASM_EMIT("add $0x10, %[off]") \ |
| 207 |
/* 1x blocks */ \ |
| 208 |
__ASM_EMIT("6:") \ |
| 209 |
__ASM_EMIT("add $3, %[count]") \ |
| 210 |
__ASM_EMIT("jl 8f") \ |
| 211 |
__ASM_EMIT("7:") \ |
| 212 |
__ASM_EMIT("movss 0x00(%[" SEL(SRC2, SRC1) "], %[off]), %%xmm2") \ |
| 213 |
__ASM_EMIT("movss 0x00(%[" SEL(SRC1, SRC2) "], %[off]), %%xmm4") \ |
| 214 |
__ASM_EMIT("andps %%xmm0, " SEL("%%xmm2", "%%xmm4")) \ |
| 215 |
__ASM_EMIT(OP "ss %%xmm2, %%xmm4") \ |
| 216 |
__ASM_EMIT("movss %%xmm4, 0x00(%[" DST "], %[off])") \ |
| 217 |
__ASM_EMIT("add $0x04, %[off]") \ |
| 218 |
__ASM_EMIT("dec %[count]") \ |
| 219 |
__ASM_EMIT("jge 7b") \ |
| 220 |
__ASM_EMIT("8:") \ |
| 221 |
|
| 222 |
void abs_add2(float *dst, const float *src, size_t count) |
| 223 |
{ |
| 224 |
IF_ARCH_X86(size_t off); |
| 225 |
ARCH_X86_ASM |
| 226 |
( |
| 227 |
ABS_OP2_CORE("add", "dst", "dst", "src", OP_DSEL) |
| 228 |
: [off] "=&r" (off), [count] "+r" (count) |
| 229 |
: [dst] "r" (dst), [src] "r" (src), |
| 230 |
[X_SIGN] "m" (abs_vv_const) |
| 231 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 232 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 233 |
); |
| 234 |
} |
| 235 |
|
| 236 |
void abs_sub2(float *dst, const float *src, size_t count) |
| 237 |
{ |
| 238 |
IF_ARCH_X86(size_t off); |
| 239 |
ARCH_X86_ASM |
| 240 |
( |
| 241 |
ABS_OP2_CORE("sub", "dst", "dst", "src", OP_DSEL) |
| 242 |
: [off] "=&r" (off), [count] "+r" (count) |
| 243 |
: [dst] "r" (dst), [src] "r" (src), |
| 244 |
[X_SIGN] "m" (abs_vv_const) |
| 245 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 246 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 247 |
); |
| 248 |
} |
| 249 |
|
| 250 |
void abs_rsub2(float *dst, const float *src, size_t count) |
| 251 |
{ |
| 252 |
IF_ARCH_X86(size_t off); |
| 253 |
ARCH_X86_ASM |
| 254 |
( |
| 255 |
ABS_OP2_CORE("sub", "dst", "dst", "src", OP_RSEL) |
| 256 |
: [off] "=&r" (off), [count] "+r" (count) |
| 257 |
: [dst] "r" (dst), [src] "r" (src), |
| 258 |
[X_SIGN] "m" (abs_vv_const) |
| 259 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 260 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 261 |
); |
| 262 |
} |
| 263 |
|
| 264 |
void abs_mul2(float *dst, const float *src, size_t count) |
| 265 |
{ |
| 266 |
IF_ARCH_X86(size_t off); |
| 267 |
ARCH_X86_ASM |
| 268 |
( |
| 269 |
ABS_OP2_CORE("mul", "dst", "dst", "src", OP_DSEL) |
| 270 |
: [off] "=&r" (off), [count] "+r" (count) |
| 271 |
: [dst] "r" (dst), [src] "r" (src), |
| 272 |
[X_SIGN] "m" (abs_vv_const) |
| 273 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 274 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 275 |
); |
| 276 |
} |
| 277 |
|
| 278 |
void abs_div2(float *dst, const float *src, size_t count) |
| 279 |
{ |
| 280 |
IF_ARCH_X86(size_t off); |
| 281 |
ARCH_X86_ASM |
| 282 |
( |
| 283 |
ABS_OP2_CORE("div", "dst", "dst", "src", OP_DSEL) |
| 284 |
: [off] "=&r" (off), [count] "+r" (count) |
| 285 |
: [dst] "r" (dst), [src] "r" (src), |
| 286 |
[X_SIGN] "m" (abs_vv_const) |
| 287 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 288 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 289 |
); |
| 290 |
} |
| 291 |
|
| 292 |
void abs_rdiv2(float *dst, const float *src, size_t count) |
| 293 |
{ |
| 294 |
IF_ARCH_X86(size_t off); |
| 295 |
ARCH_X86_ASM |
| 296 |
( |
| 297 |
ABS_OP2_CORE("div", "dst", "dst", "src", OP_RSEL) |
| 298 |
: [off] "=&r" (off), [count] "+r" (count) |
| 299 |
: [dst] "r" (dst), [src] "r" (src), |
| 300 |
[X_SIGN] "m" (abs_vv_const) |
| 301 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 302 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 303 |
); |
| 304 |
} |
| 305 |
|
| 306 |
void abs_max2(float *dst, const float *src, size_t count) |
| 307 |
{ |
| 308 |
IF_ARCH_X86(size_t off); |
| 309 |
ARCH_X86_ASM |
| 310 |
( |
| 311 |
ABS_OP2_CORE("max", "dst", "dst", "src", OP_DSEL) |
| 312 |
: [off] "=&r" (off), [count] "+r" (count) |
| 313 |
: [dst] "r" (dst), [src] "r" (src), |
| 314 |
[X_SIGN] "m" (abs_vv_const) |
| 315 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 316 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 317 |
); |
| 318 |
} |
| 319 |
|
| 320 |
void abs_min2(float *dst, const float *src, size_t count) |
| 321 |
{ |
| 322 |
IF_ARCH_X86(size_t off); |
| 323 |
ARCH_X86_ASM |
| 324 |
( |
| 325 |
ABS_OP2_CORE("min", "dst", "dst", "src", OP_DSEL) |
| 326 |
: [off] "=&r" (off), [count] "+r" (count) |
| 327 |
: [dst] "r" (dst), [src] "r" (src), |
| 328 |
[X_SIGN] "m" (abs_vv_const) |
| 329 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 330 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 331 |
); |
| 332 |
} |
| 333 |
|
| 334 |
void abs_add3(float *dst, const float *src1, const float *src2, size_t count) |
| 335 |
{ |
| 336 |
IF_ARCH_X86(size_t off); |
| 337 |
ARCH_X86_ASM |
| 338 |
( |
| 339 |
ABS_OP2_CORE("add", "dst", "src1", "src2", OP_DSEL) |
| 340 |
: [off] "=&r" (off), [count] "+r" (count) |
| 341 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 342 |
[X_SIGN] "m" (abs_vv_const) |
| 343 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 344 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 345 |
); |
| 346 |
} |
| 347 |
|
| 348 |
void abs_sub3(float *dst, const float *src1, const float *src2, size_t count) |
| 349 |
{ |
| 350 |
IF_ARCH_X86(size_t off); |
| 351 |
ARCH_X86_ASM |
| 352 |
( |
| 353 |
ABS_OP2_CORE("sub", "dst", "src1", "src2", OP_DSEL) |
| 354 |
: [off] "=&r" (off), [count] "+r" (count) |
| 355 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 356 |
[X_SIGN] "m" (abs_vv_const) |
| 357 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 358 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 359 |
); |
| 360 |
} |
| 361 |
|
| 362 |
void abs_rsub3(float *dst, const float *src1, const float *src2, size_t count) |
| 363 |
{ |
| 364 |
IF_ARCH_X86(size_t off); |
| 365 |
ARCH_X86_ASM |
| 366 |
( |
| 367 |
ABS_OP2_CORE("sub", "dst", "src1", "src2", OP_RSEL) |
| 368 |
: [off] "=&r" (off), [count] "+r" (count) |
| 369 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 370 |
[X_SIGN] "m" (abs_vv_const) |
| 371 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 372 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 373 |
); |
| 374 |
} |
| 375 |
|
| 376 |
void abs_mul3(float *dst, const float *src1, const float *src2, size_t count) |
| 377 |
{ |
| 378 |
IF_ARCH_X86(size_t off); |
| 379 |
ARCH_X86_ASM |
| 380 |
( |
| 381 |
ABS_OP2_CORE("mul", "dst", "src1", "src2", OP_DSEL) |
| 382 |
: [off] "=&r" (off), [count] "+r" (count) |
| 383 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 384 |
[X_SIGN] "m" (abs_vv_const) |
| 385 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 386 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 387 |
); |
| 388 |
} |
| 389 |
|
| 390 |
void abs_div3(float *dst, const float *src1, const float *src2, size_t count) |
| 391 |
{ |
| 392 |
IF_ARCH_X86(size_t off); |
| 393 |
ARCH_X86_ASM |
| 394 |
( |
| 395 |
ABS_OP2_CORE("div", "dst", "src1", "src2", OP_DSEL) |
| 396 |
: [off] "=&r" (off), [count] "+r" (count) |
| 397 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 398 |
[X_SIGN] "m" (abs_vv_const) |
| 399 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 400 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 401 |
); |
| 402 |
} |
| 403 |
|
| 404 |
void abs_rdiv3(float *dst, const float *src1, const float *src2, size_t count) |
| 405 |
{ |
| 406 |
IF_ARCH_X86(size_t off); |
| 407 |
ARCH_X86_ASM |
| 408 |
( |
| 409 |
ABS_OP2_CORE("div", "dst", "src1", "src2", OP_RSEL) |
| 410 |
: [off] "=&r" (off), [count] "+r" (count) |
| 411 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 412 |
[X_SIGN] "m" (abs_vv_const) |
| 413 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 414 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 415 |
); |
| 416 |
} |
| 417 |
|
| 418 |
void abs_max3(float *dst, const float *src1, const float *src2, size_t count) |
| 419 |
{ |
| 420 |
IF_ARCH_X86(size_t off); |
| 421 |
ARCH_X86_ASM |
| 422 |
( |
| 423 |
ABS_OP2_CORE("max", "dst", "src1", "src2", OP_DSEL) |
| 424 |
: [off] "=&r" (off), [count] "+r" (count) |
| 425 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 426 |
[X_SIGN] "m" (abs_vv_const) |
| 427 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 428 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 429 |
); |
| 430 |
} |
| 431 |
|
| 432 |
void abs_min3(float *dst, const float *src1, const float *src2, size_t count) |
| 433 |
{ |
| 434 |
IF_ARCH_X86(size_t off); |
| 435 |
ARCH_X86_ASM |
| 436 |
( |
| 437 |
ABS_OP2_CORE("min", "dst", "src1", "src2", OP_DSEL) |
| 438 |
: [off] "=&r" (off), [count] "+r" (count) |
| 439 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 440 |
[X_SIGN] "m" (abs_vv_const) |
| 441 |
: "%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 442 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 443 |
); |
| 444 |
} |
| 445 |
|
| 446 |
#undef ABS_OP2_CORE |
| 447 |
|
| 448 |
#undef OP_DSEL |
| 449 |
#undef OP_RSEL |
| 450 |
} /* namespace sse */ |
| 451 |
} /* namespace lsp */ |
| 452 |
|
| 453 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE_PMATH_ABS_VV_H_ */ |