| 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_AVX_MSMATRIX_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_AVX_MSMATRIX_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 |
IF_ARCH_X86( |
| 34 |
static const float msmatrix_const[] __lsp_aligned32 = |
| 35 |
{ |
| 36 |
LSP_DSP_VEC8(0.5f) |
| 37 |
}; |
| 38 |
) |
| 39 |
|
| 40 |
void lr_to_ms(float *m, float *s, const float *l, const float *r, size_t count) |
| 41 |
{ |
| 42 |
IF_ARCH_X86(size_t off); |
| 43 |
ARCH_X86_ASM( |
| 44 |
__ASM_EMIT("xor %[off], %[off]") |
| 45 |
__ASM_EMIT("vmovaps %[X_HALF], %%ymm7") |
| 46 |
// 16x blocks |
| 47 |
__ASM_EMIT32("subl $16, %[count]") |
| 48 |
__ASM_EMIT64("sub $16, %[count]") |
| 49 |
__ASM_EMIT("vmovaps %%ymm7, %%ymm6") |
| 50 |
__ASM_EMIT("jb 2f") |
| 51 |
__ASM_EMIT("1:") |
| 52 |
__ASM_EMIT("vmulps 0x00(%[left], %[off]), %%ymm6, %%ymm0") // ymm0 = l*0.5 |
| 53 |
__ASM_EMIT("vmulps 0x20(%[left], %[off]), %%ymm7, %%ymm1") |
| 54 |
__ASM_EMIT("vmulps 0x00(%[right], %[off]), %%ymm6, %%ymm4") // ymm4 = r*0.5 |
| 55 |
__ASM_EMIT("vmulps 0x20(%[right], %[off]), %%ymm7, %%ymm5") |
| 56 |
__ASM_EMIT("vsubps %%ymm4, %%ymm0, %%ymm2") // ymm2 = s = (l-r)*0.5 |
| 57 |
__ASM_EMIT("vsubps %%ymm5, %%ymm1, %%ymm3") |
| 58 |
__ASM_EMIT("vaddps %%ymm4, %%ymm0, %%ymm0") // ymm0 = m = (l+r)*0.5 |
| 59 |
__ASM_EMIT("vaddps %%ymm5, %%ymm1, %%ymm1") |
| 60 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[mid], %[off])") |
| 61 |
__ASM_EMIT("vmovups %%ymm1, 0x20(%[mid], %[off])") |
| 62 |
__ASM_EMIT("vmovups %%ymm2, 0x00(%[side], %[off])") |
| 63 |
__ASM_EMIT("vmovups %%ymm3, 0x20(%[side], %[off])") |
| 64 |
__ASM_EMIT("add $0x40, %[off]") |
| 65 |
__ASM_EMIT32("subl $16, %[count]") |
| 66 |
__ASM_EMIT64("sub $16, %[count]") |
| 67 |
__ASM_EMIT("jae 1b") |
| 68 |
__ASM_EMIT("2:") |
| 69 |
// 8x block |
| 70 |
__ASM_EMIT32("addl $8, %[count]") |
| 71 |
__ASM_EMIT64("add $8, %[count]") |
| 72 |
__ASM_EMIT("jl 4f") |
| 73 |
__ASM_EMIT("vmulps 0x00(%[left], %[off]), %%ymm6, %%ymm0") // ymm0 = l*0.5 |
| 74 |
__ASM_EMIT("vmulps 0x00(%[right], %[off]), %%ymm7, %%ymm4") // ymm4 = r*0.5 |
| 75 |
__ASM_EMIT("vsubps %%ymm4, %%ymm0, %%ymm2") // ymm2 = s = (l-r)*0.5 |
| 76 |
__ASM_EMIT("vaddps %%ymm4, %%ymm0, %%ymm0") // ymm0 = m = (l+r)*0.5 |
| 77 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[mid], %[off])") |
| 78 |
__ASM_EMIT("vmovups %%ymm2, 0x00(%[side], %[off])") |
| 79 |
__ASM_EMIT("add $0x20, %[off]") |
| 80 |
__ASM_EMIT32("subl $8, %[count]") |
| 81 |
__ASM_EMIT64("sub $8, %[count]") |
| 82 |
__ASM_EMIT("4:") |
| 83 |
// 4x block |
| 84 |
__ASM_EMIT32("addl $4, %[count]") |
| 85 |
__ASM_EMIT64("add $4, %[count]") |
| 86 |
__ASM_EMIT("jl 6f") |
| 87 |
__ASM_EMIT("vmulps 0x00(%[left], %[off]), %%xmm6, %%xmm0") // xmm0 = l*0.5 |
| 88 |
__ASM_EMIT("vmulps 0x00(%[right], %[off]), %%xmm7, %%xmm4") // xmm4 = r*0.5 |
| 89 |
__ASM_EMIT("vsubps %%xmm4, %%xmm0, %%xmm2") // xmm2 = s = (l-r)*0.5 |
| 90 |
__ASM_EMIT("vaddps %%xmm4, %%xmm0, %%xmm0") // xmm0 = m = (l+r)*0.5 |
| 91 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[mid], %[off])") |
| 92 |
__ASM_EMIT("vmovups %%xmm2, 0x00(%[side], %[off])") |
| 93 |
__ASM_EMIT("add $0x10, %[off]") |
| 94 |
__ASM_EMIT32("subl $4, %[count]") |
| 95 |
__ASM_EMIT64("sub $4, %[count]") |
| 96 |
__ASM_EMIT("6:") |
| 97 |
// 1x blocks |
| 98 |
__ASM_EMIT32("addl $3, %[count]") |
| 99 |
__ASM_EMIT64("add $3, %[count]") |
| 100 |
__ASM_EMIT("jl 8f") |
| 101 |
__ASM_EMIT("7:") |
| 102 |
__ASM_EMIT("vmulss 0x00(%[left], %[off]), %%xmm6, %%xmm0") // xmm0 = l*0.5 |
| 103 |
__ASM_EMIT("vmulss 0x00(%[right], %[off]), %%xmm7, %%xmm4") // xmm4 = r*0.5 |
| 104 |
__ASM_EMIT("vsubss %%xmm4, %%xmm0, %%xmm2") // xmm2 = s = (l-r)*0.5 |
| 105 |
__ASM_EMIT("vaddss %%xmm4, %%xmm0, %%xmm0") // xmm0 = m = (l+r)*0.5 |
| 106 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[mid], %[off])") |
| 107 |
__ASM_EMIT("vmovss %%xmm2, 0x00(%[side], %[off])") |
| 108 |
__ASM_EMIT("add $0x04, %[off]") |
| 109 |
__ASM_EMIT32("decl %[count]") |
| 110 |
__ASM_EMIT64("dec %[count]") |
| 111 |
__ASM_EMIT("jge 7b") |
| 112 |
__ASM_EMIT("8:") |
| 113 |
|
| 114 |
: [off] "=&r" (off), |
| 115 |
[count] X86_PGREG (count) |
| 116 |
: [left] "r"(l), [right] "r" (r), |
| 117 |
[mid] "r" (m), [side] "r" (s), |
| 118 |
[X_HALF] "m" (msmatrix_const) |
| 119 |
: "cc", "memory", |
| 120 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 121 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 122 |
); |
| 123 |
} |
| 124 |
|
| 125 |
#define LR_TO_PART(P, L, R, OP) \ |
| 126 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 127 |
__ASM_EMIT("vmovaps %[X_HALF], %%ymm7") \ |
| 128 |
/* 32x blocks */ \ |
| 129 |
__ASM_EMIT("sub $32, %[count]") \ |
| 130 |
__ASM_EMIT("vmovaps %%ymm7, %%ymm6") \ |
| 131 |
__ASM_EMIT("jb 2f") \ |
| 132 |
__ASM_EMIT("1:") \ |
| 133 |
__ASM_EMIT("vmovups 0x00(%[" L "], %[off]), %%ymm0") /* ymm0 = l */ \ |
| 134 |
__ASM_EMIT("vmovups 0x20(%[" L "], %[off]), %%ymm1") \ |
| 135 |
__ASM_EMIT("vmovups 0x40(%[" L "], %[off]), %%ymm2") \ |
| 136 |
__ASM_EMIT("vmovups 0x60(%[" L "], %[off]), %%ymm3") \ |
| 137 |
__ASM_EMIT(OP "ps 0x00(%[" R "], %[off]), %%ymm0, %%ymm0") /* ymm0 = l op r */ \ |
| 138 |
__ASM_EMIT(OP "ps 0x20(%[" R "], %[off]), %%ymm1, %%ymm1") \ |
| 139 |
__ASM_EMIT(OP "ps 0x40(%[" R "], %[off]), %%ymm2, %%ymm2") \ |
| 140 |
__ASM_EMIT(OP "ps 0x60(%[" R "], %[off]), %%ymm3, %%ymm3") \ |
| 141 |
__ASM_EMIT("vmulps %%ymm6, %%ymm0, %%ymm0") /* ymm0 = (l op r) * 0.5 */ \ |
| 142 |
__ASM_EMIT("vmulps %%ymm7, %%ymm1, %%ymm1") \ |
| 143 |
__ASM_EMIT("vmulps %%ymm6, %%ymm2, %%ymm2") \ |
| 144 |
__ASM_EMIT("vmulps %%ymm7, %%ymm3, %%ymm3") \ |
| 145 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" P "], %[off])") \ |
| 146 |
__ASM_EMIT("vmovups %%ymm1, 0x20(%[" P "], %[off])") \ |
| 147 |
__ASM_EMIT("vmovups %%ymm2, 0x40(%[" P "], %[off])") \ |
| 148 |
__ASM_EMIT("vmovups %%ymm3, 0x60(%[" P "], %[off])") \ |
| 149 |
__ASM_EMIT("add $0x80, %[off]") \ |
| 150 |
__ASM_EMIT("sub $32, %[count]") \ |
| 151 |
__ASM_EMIT("jae 1b") \ |
| 152 |
__ASM_EMIT("2:") \ |
| 153 |
/* 16x block */ \ |
| 154 |
__ASM_EMIT("add $16, %[count]") \ |
| 155 |
__ASM_EMIT("jl 4f") \ |
| 156 |
__ASM_EMIT("vmovups 0x00(%[" L "], %[off]), %%ymm0") /* ymm0 = l */ \ |
| 157 |
__ASM_EMIT("vmovups 0x20(%[" L "], %[off]), %%ymm1") \ |
| 158 |
__ASM_EMIT(OP "ps 0x00(%[" R "], %[off]), %%ymm0, %%ymm0") /* ymm0 = l op r */ \ |
| 159 |
__ASM_EMIT(OP "ps 0x20(%[" R "], %[off]), %%ymm1, %%ymm1") \ |
| 160 |
__ASM_EMIT("vmulps %%ymm6, %%ymm0, %%ymm0") /* ymm0 = (l op r) * 0.5 */ \ |
| 161 |
__ASM_EMIT("vmulps %%ymm7, %%ymm1, %%ymm1") \ |
| 162 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" P "], %[off])") \ |
| 163 |
__ASM_EMIT("vmovups %%ymm1, 0x20(%[" P "], %[off])") \ |
| 164 |
__ASM_EMIT("sub $16, %[count]") \ |
| 165 |
__ASM_EMIT("add $0x40, %[off]") \ |
| 166 |
__ASM_EMIT("4:") \ |
| 167 |
/* 8x block */ \ |
| 168 |
__ASM_EMIT("add $8, %[count]") \ |
| 169 |
__ASM_EMIT("jl 6f") \ |
| 170 |
__ASM_EMIT("vmovups 0x00(%[" L "], %[off]), %%ymm0") /* ymm0 = l */ \ |
| 171 |
__ASM_EMIT(OP "ps 0x00(%[" R "], %[off]), %%ymm0, %%ymm0") /* ymm0 = l op r */ \ |
| 172 |
__ASM_EMIT("vmulps %%ymm6, %%ymm0, %%ymm0") /* ymm0 = (l op r) * 0.5 */ \ |
| 173 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" P "], %[off])") \ |
| 174 |
__ASM_EMIT("sub $8, %[count]") \ |
| 175 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 176 |
__ASM_EMIT("6:") \ |
| 177 |
/* 4x block */ \ |
| 178 |
__ASM_EMIT("add $4, %[count]") \ |
| 179 |
__ASM_EMIT("jl 8f") \ |
| 180 |
__ASM_EMIT("vmovups 0x00(%[" L "], %[off]), %%xmm0") /* xmm0 = l */ \ |
| 181 |
__ASM_EMIT(OP "ps 0x00(%[" R "], %[off]), %%xmm0, %%xmm0") /* xmm0 = l op r */ \ |
| 182 |
__ASM_EMIT("vmulps %%xmm6, %%xmm0, %%xmm0") /* xmm0 = (l op r) * 0.5 */ \ |
| 183 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[" P "], %[off])") \ |
| 184 |
__ASM_EMIT("sub $4, %[count]") \ |
| 185 |
__ASM_EMIT("add $0x10, %[off]") \ |
| 186 |
__ASM_EMIT("8:") \ |
| 187 |
/* 1x blocks */ \ |
| 188 |
__ASM_EMIT("add $3, %[count]") \ |
| 189 |
__ASM_EMIT("jl 10f") \ |
| 190 |
__ASM_EMIT("9:") \ |
| 191 |
__ASM_EMIT("vmovss 0x00(%[" L "], %[off]), %%xmm0") /* xmm0 = l */ \ |
| 192 |
__ASM_EMIT(OP "ss 0x00(%[" R "], %[off]), %%xmm0, %%xmm0") /* xmm0 = l op r */ \ |
| 193 |
__ASM_EMIT("vmulss %%xmm6, %%xmm0, %%xmm0") /* xmm0 = (l op r) * 0.5 */ \ |
| 194 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[" P "], %[off])") \ |
| 195 |
__ASM_EMIT("add $0x04, %[off]") \ |
| 196 |
__ASM_EMIT("dec %[count]") \ |
| 197 |
__ASM_EMIT("jge 9b") \ |
| 198 |
__ASM_EMIT("10:") |
| 199 |
|
| 200 |
void lr_to_mid(float *m, const float *l, const float *r, size_t count) |
| 201 |
{ |
| 202 |
IF_ARCH_X86(size_t off); |
| 203 |
ARCH_X86_ASM ( |
| 204 |
LR_TO_PART("mid", "left", "right", "vadd") |
| 205 |
: [off] "=&r" (off), [count] "+r" (count) |
| 206 |
: [left] "r" (l), [right] "r" (r), |
| 207 |
[mid] "r" (m), |
| 208 |
[X_HALF] "m" (msmatrix_const) |
| 209 |
: "cc", "memory", |
| 210 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 211 |
"%xmm6", "%xmm7" |
| 212 |
); |
| 213 |
} |
| 214 |
|
| 215 |
void lr_to_side(float *s, const float *l, const float *r, size_t count) |
| 216 |
{ |
| 217 |
IF_ARCH_X86(size_t off); |
| 218 |
ARCH_X86_ASM ( |
| 219 |
LR_TO_PART("side", "left", "right", "vsub") |
| 220 |
: [off] "=&r" (off), [count] "+r" (count) |
| 221 |
: [left] "r" (l), [right] "r" (r), |
| 222 |
[side] "r" (s), |
| 223 |
[X_HALF] "m" (msmatrix_const) |
| 224 |
: "cc", "memory", |
| 225 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 226 |
"%xmm6", "%xmm7" |
| 227 |
); |
| 228 |
} |
| 229 |
|
| 230 |
#undef LR_TO_PART |
| 231 |
|
| 232 |
void ms_to_lr(float *l, float *r, const float *m, const float *s, size_t count) |
| 233 |
{ |
| 234 |
IF_ARCH_X86(size_t off); |
| 235 |
ARCH_X86_ASM( |
| 236 |
__ASM_EMIT("xor %[off], %[off]") |
| 237 |
// 32x blocks |
| 238 |
__ASM_EMIT32("subl $32, %[count]") |
| 239 |
__ASM_EMIT64("sub $32, %[count]") |
| 240 |
__ASM_EMIT("jb 2f") |
| 241 |
__ASM_EMIT("1:") |
| 242 |
__ASM_EMIT("vmovups 0x00(%[mid], %[off]), %%ymm0") // ymm0 = m |
| 243 |
__ASM_EMIT("vmovups 0x20(%[mid], %[off]), %%ymm1") |
| 244 |
__ASM_EMIT("vmovups 0x40(%[mid], %[off]), %%ymm2") |
| 245 |
__ASM_EMIT("vmovups 0x60(%[mid], %[off]), %%ymm3") |
| 246 |
__ASM_EMIT("vsubps 0x00(%[side], %[off]), %%ymm0, %%ymm4") // ymm4 = m - s |
| 247 |
__ASM_EMIT("vsubps 0x20(%[side], %[off]), %%ymm1, %%ymm5") |
| 248 |
__ASM_EMIT("vsubps 0x40(%[side], %[off]), %%ymm2, %%ymm6") |
| 249 |
__ASM_EMIT("vsubps 0x60(%[side], %[off]), %%ymm3, %%ymm7") |
| 250 |
__ASM_EMIT("vaddps 0x00(%[side], %[off]), %%ymm0, %%ymm0") // ymm0 = m + s |
| 251 |
__ASM_EMIT("vaddps 0x20(%[side], %[off]), %%ymm1, %%ymm1") |
| 252 |
__ASM_EMIT("vaddps 0x40(%[side], %[off]), %%ymm2, %%ymm2") |
| 253 |
__ASM_EMIT("vaddps 0x60(%[side], %[off]), %%ymm3, %%ymm3") |
| 254 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[left], %[off])") |
| 255 |
__ASM_EMIT("vmovups %%ymm1, 0x20(%[left], %[off])") |
| 256 |
__ASM_EMIT("vmovups %%ymm2, 0x40(%[left], %[off])") |
| 257 |
__ASM_EMIT("vmovups %%ymm3, 0x60(%[left], %[off])") |
| 258 |
__ASM_EMIT("vmovups %%ymm4, 0x00(%[right], %[off])") |
| 259 |
__ASM_EMIT("vmovups %%ymm5, 0x20(%[right], %[off])") |
| 260 |
__ASM_EMIT("vmovups %%ymm6, 0x40(%[right], %[off])") |
| 261 |
__ASM_EMIT("vmovups %%ymm7, 0x60(%[right], %[off])") |
| 262 |
__ASM_EMIT("add $0x80, %[off]") |
| 263 |
__ASM_EMIT32("subl $32, %[count]") |
| 264 |
__ASM_EMIT64("sub $32, %[count]") |
| 265 |
__ASM_EMIT("jae 1b") |
| 266 |
__ASM_EMIT("2:") |
| 267 |
// 16x block |
| 268 |
__ASM_EMIT32("addl $16, %[count]") |
| 269 |
__ASM_EMIT64("add $16, %[count]") |
| 270 |
__ASM_EMIT("jl 4f") |
| 271 |
__ASM_EMIT("vmovups 0x00(%[mid], %[off]), %%ymm0") // ymm0 = m |
| 272 |
__ASM_EMIT("vmovups 0x20(%[mid], %[off]), %%ymm1") |
| 273 |
__ASM_EMIT("vsubps 0x00(%[side], %[off]), %%ymm0, %%ymm4") // ymm4 = m - s |
| 274 |
__ASM_EMIT("vsubps 0x20(%[side], %[off]), %%ymm1, %%ymm5") |
| 275 |
__ASM_EMIT("vaddps 0x00(%[side], %[off]), %%ymm0, %%ymm0") // ymm0 = m + s |
| 276 |
__ASM_EMIT("vaddps 0x20(%[side], %[off]), %%ymm1, %%ymm1") |
| 277 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[left], %[off])") |
| 278 |
__ASM_EMIT("vmovups %%ymm1, 0x20(%[left], %[off])") |
| 279 |
__ASM_EMIT("vmovups %%ymm4, 0x00(%[right], %[off])") |
| 280 |
__ASM_EMIT("vmovups %%ymm5, 0x20(%[right], %[off])") |
| 281 |
__ASM_EMIT32("subl $16, %[count]") |
| 282 |
__ASM_EMIT64("sub $16, %[count]") |
| 283 |
__ASM_EMIT("add $0x40, %[off]") |
| 284 |
__ASM_EMIT("4:") |
| 285 |
// 8x block |
| 286 |
__ASM_EMIT32("addl $8, %[count]") |
| 287 |
__ASM_EMIT64("add $8, %[count]") |
| 288 |
__ASM_EMIT("jl 6f") |
| 289 |
__ASM_EMIT("vmovups 0x00(%[mid], %[off]), %%ymm0") // ymm0 = m |
| 290 |
__ASM_EMIT("vsubps 0x00(%[side], %[off]), %%ymm0, %%ymm4") // ymm4 = m - s |
| 291 |
__ASM_EMIT("vaddps 0x00(%[side], %[off]), %%ymm0, %%ymm0") // ymm0 = m + s |
| 292 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[left], %[off])") |
| 293 |
__ASM_EMIT("vmovups %%ymm4, 0x00(%[right], %[off])") |
| 294 |
__ASM_EMIT32("subl $8, %[count]") |
| 295 |
__ASM_EMIT64("sub $8, %[count]") |
| 296 |
__ASM_EMIT("add $0x20, %[off]") |
| 297 |
__ASM_EMIT("6:") |
| 298 |
// 4x block |
| 299 |
__ASM_EMIT32("addl $4, %[count]") |
| 300 |
__ASM_EMIT64("add $4, %[count]") |
| 301 |
__ASM_EMIT("jl 8f") |
| 302 |
__ASM_EMIT("vmovups 0x00(%[mid], %[off]), %%xmm0") // xmm0 = m |
| 303 |
__ASM_EMIT("vsubps 0x00(%[side], %[off]), %%xmm0, %%xmm4") // xmm4 = m - s |
| 304 |
__ASM_EMIT("vaddps 0x00(%[side], %[off]), %%xmm0, %%xmm0") // xmm0 = m + s |
| 305 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[left], %[off])") |
| 306 |
__ASM_EMIT("vmovups %%xmm4, 0x00(%[right], %[off])") |
| 307 |
__ASM_EMIT32("subl $4, %[count]") |
| 308 |
__ASM_EMIT64("sub $4, %[count]") |
| 309 |
__ASM_EMIT("add $0x10, %[off]") |
| 310 |
__ASM_EMIT("8:") |
| 311 |
// 1x blocks |
| 312 |
__ASM_EMIT32("addl $3, %[count]") |
| 313 |
__ASM_EMIT64("add $3, %[count]") |
| 314 |
__ASM_EMIT("jl 10f") |
| 315 |
__ASM_EMIT("9:") |
| 316 |
__ASM_EMIT("vmovss 0x00(%[mid], %[off]), %%xmm0") // xmm0 = m |
| 317 |
__ASM_EMIT("vsubss 0x00(%[side], %[off]), %%xmm0, %%xmm4") // xmm4 = m - s |
| 318 |
__ASM_EMIT("vaddss 0x00(%[side], %[off]), %%xmm0, %%xmm0") // xmm0 = m + s |
| 319 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[left], %[off])") |
| 320 |
__ASM_EMIT("vmovss %%xmm4, 0x00(%[right], %[off])") |
| 321 |
__ASM_EMIT("add $0x04, %[off]") |
| 322 |
__ASM_EMIT32("decl %[count]") |
| 323 |
__ASM_EMIT64("dec %[count]") |
| 324 |
__ASM_EMIT("jge 9b") |
| 325 |
__ASM_EMIT("10:") |
| 326 |
|
| 327 |
: [off] "=&r" (off), |
| 328 |
[count] X86_PGREG (count) |
| 329 |
: [left] "r"(l), [right] "r" (r), |
| 330 |
[mid] "r" (m), [side] "r" (s) |
| 331 |
: "cc", "memory", |
| 332 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 333 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 334 |
); |
| 335 |
} |
| 336 |
|
| 337 |
#define MS_TO_PART(P, M, S, OP) \ |
| 338 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 339 |
/* 32x blocks */ \ |
| 340 |
__ASM_EMIT("sub $32, %[count]") \ |
| 341 |
__ASM_EMIT("jb 2f") \ |
| 342 |
__ASM_EMIT("1:") \ |
| 343 |
__ASM_EMIT("vmovups 0x00(%[" M "], %[off]), %%ymm0") /* ymm0 = m */ \ |
| 344 |
__ASM_EMIT("vmovups 0x20(%[" M "], %[off]), %%ymm1") \ |
| 345 |
__ASM_EMIT("vmovups 0x40(%[" M "], %[off]), %%ymm2") \ |
| 346 |
__ASM_EMIT("vmovups 0x60(%[" M "], %[off]), %%ymm3") \ |
| 347 |
__ASM_EMIT(OP "ps 0x00(%[" S "], %[off]), %%ymm0, %%ymm0") /* ymm4 = m - s */ \ |
| 348 |
__ASM_EMIT(OP "ps 0x20(%[" S "], %[off]), %%ymm1, %%ymm1") \ |
| 349 |
__ASM_EMIT(OP "ps 0x40(%[" S "], %[off]), %%ymm2, %%ymm2") \ |
| 350 |
__ASM_EMIT(OP "ps 0x60(%[" S "], %[off]), %%ymm3, %%ymm3") \ |
| 351 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" P "], %[off])") \ |
| 352 |
__ASM_EMIT("vmovups %%ymm1, 0x20(%[" P "], %[off])") \ |
| 353 |
__ASM_EMIT("vmovups %%ymm2, 0x40(%[" P "], %[off])") \ |
| 354 |
__ASM_EMIT("vmovups %%ymm3, 0x60(%[" P "], %[off])") \ |
| 355 |
__ASM_EMIT("add $0x80, %[off]") \ |
| 356 |
__ASM_EMIT("sub $32, %[count]") \ |
| 357 |
__ASM_EMIT("jae 1b") \ |
| 358 |
__ASM_EMIT("2:") \ |
| 359 |
/* 16x block */ \ |
| 360 |
__ASM_EMIT("add $16, %[count]") \ |
| 361 |
__ASM_EMIT("jl 4f") \ |
| 362 |
__ASM_EMIT("vmovups 0x00(%[" M "], %[off]), %%ymm0") /* ymm0 = m */ \ |
| 363 |
__ASM_EMIT("vmovups 0x20(%[" M "], %[off]), %%ymm1") \ |
| 364 |
__ASM_EMIT(OP "ps 0x00(%[" S "], %[off]), %%ymm0, %%ymm0") /* ymm4 = m - s */ \ |
| 365 |
__ASM_EMIT(OP "ps 0x20(%[" S "], %[off]), %%ymm1, %%ymm1") \ |
| 366 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" P "], %[off])") \ |
| 367 |
__ASM_EMIT("vmovups %%ymm1, 0x20(%[" P "], %[off])") \ |
| 368 |
__ASM_EMIT("sub $16, %[count]") \ |
| 369 |
__ASM_EMIT("add $0x40, %[off]") \ |
| 370 |
__ASM_EMIT("4:") \ |
| 371 |
/* 8x block */ \ |
| 372 |
__ASM_EMIT("add $8, %[count]") \ |
| 373 |
__ASM_EMIT("jl 6f") \ |
| 374 |
__ASM_EMIT("vmovups 0x00(%[" M "], %[off]), %%ymm0") /* ymm0 = m */ \ |
| 375 |
__ASM_EMIT(OP "ps 0x00(%[" S "], %[off]), %%ymm0, %%ymm0") /* ymm4 = m - s */ \ |
| 376 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" P "], %[off])") \ |
| 377 |
__ASM_EMIT("sub $8, %[count]") \ |
| 378 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 379 |
__ASM_EMIT("6:") \ |
| 380 |
/* 4x block */ \ |
| 381 |
__ASM_EMIT("add $4, %[count]") \ |
| 382 |
__ASM_EMIT("jl 8f") \ |
| 383 |
__ASM_EMIT("vmovups 0x00(%[" M "], %[off]), %%xmm0") /* xmm0 = m */ \ |
| 384 |
__ASM_EMIT(OP "ps 0x00(%[" S "], %[off]), %%xmm0, %%xmm0") /* xmm4 = m - s */ \ |
| 385 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[" P "], %[off])") \ |
| 386 |
__ASM_EMIT("sub $4, %[count]") \ |
| 387 |
__ASM_EMIT("add $0x10, %[off]") \ |
| 388 |
__ASM_EMIT("8:") \ |
| 389 |
/* 1x blocks */ \ |
| 390 |
__ASM_EMIT("add $3, %[count]") \ |
| 391 |
__ASM_EMIT("jl 10f") \ |
| 392 |
__ASM_EMIT("9:") \ |
| 393 |
__ASM_EMIT("vmovss 0x00(%[" M "], %[off]), %%xmm0") /* xmm0 = m */ \ |
| 394 |
__ASM_EMIT(OP "ss 0x00(%[" S "], %[off]), %%xmm0, %%xmm0") /* xmm4 = m - s */ \ |
| 395 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[" P "], %[off])") \ |
| 396 |
__ASM_EMIT("add $0x04, %[off]") \ |
| 397 |
__ASM_EMIT("dec %[count]") \ |
| 398 |
__ASM_EMIT("jge 9b") \ |
| 399 |
__ASM_EMIT("10:") |
| 400 |
|
| 401 |
void ms_to_left(float *l, const float *m, const float *s, size_t count) |
| 402 |
{ |
| 403 |
size_t off; |
| 404 |
|
| 405 |
ARCH_X86_ASM ( |
| 406 |
MS_TO_PART("left", "mid", "side", "vadd") |
| 407 |
: [off] "=&r" (off), [count] "+r" (count) |
| 408 |
: [left] "r" (l), |
| 409 |
[mid] "r" (m), [side] "r" (s) |
| 410 |
: "cc", "memory", |
| 411 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3" |
| 412 |
); |
| 413 |
} |
| 414 |
|
| 415 |
void ms_to_right(float *r, const float *m, const float *s, size_t count) |
| 416 |
{ |
| 417 |
size_t off; |
| 418 |
|
| 419 |
ARCH_X86_ASM ( |
| 420 |
MS_TO_PART("right", "mid", "side", "vsub") |
| 421 |
: [off] "=&r" (off), [count] "+r" (count) |
| 422 |
: [right] "r" (r), |
| 423 |
[mid] "r" (m), [side] "r" (s) |
| 424 |
: "cc", "memory", |
| 425 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3" |
| 426 |
); |
| 427 |
} |
| 428 |
|
| 429 |
#undef MS_TO_PART |
| 430 |
} /* namespace avx */ |
| 431 |
} /* namespace lsp */ |
| 432 |
|
| 433 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX_MSMATRIX_H_ */ |