| 1 |
/* |
| 2 |
* Copyright (C) 2020 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2020 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_SSE2_GRAPHICS_EFFECTS_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_SSE2_GRAPHICS_EFFECTS_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_SSE2_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE2_IMPL */ |
| 28 |
|
| 29 |
#define X4_TRANSPOSE \ |
| 30 |
__ASM_EMIT("movaps %%xmm2, %%xmm4") \ |
| 31 |
__ASM_EMIT("punpckldq %%xmm3, %%xmm2") \ |
| 32 |
__ASM_EMIT("punpckhdq %%xmm3, %%xmm4") \ |
| 33 |
__ASM_EMIT("movaps %%xmm0, %%xmm3") \ |
| 34 |
__ASM_EMIT("punpckldq %%xmm1, %%xmm0") \ |
| 35 |
__ASM_EMIT("punpckhdq %%xmm1, %%xmm3") \ |
| 36 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") \ |
| 37 |
__ASM_EMIT("punpcklqdq %%xmm2, %%xmm0") \ |
| 38 |
__ASM_EMIT("punpckhqdq %%xmm2, %%xmm1") \ |
| 39 |
__ASM_EMIT("movaps %%xmm3, %%xmm2") \ |
| 40 |
__ASM_EMIT("punpcklqdq %%xmm4, %%xmm2") \ |
| 41 |
__ASM_EMIT("punpckhqdq %%xmm4, %%xmm3") |
| 42 |
|
| 43 |
namespace lsp |
| 44 |
{ |
| 45 |
namespace sse2 |
| 46 |
{ |
| 47 |
IF_ARCH_X86( |
| 48 |
static const float EFF_HSLA_HUE_XC[] __lsp_aligned16 = |
| 49 |
{ |
| 50 |
LSP_DSP_VEC4(1.0f) |
| 51 |
}; |
| 52 |
) |
| 53 |
|
| 54 |
#define EFF_HSLA_HUE_CORE \ |
| 55 |
/* xmm0 = v, xmm6 = T, xmm7 = KT */ \ |
| 56 |
__ASM_EMIT("movaps %[XC], %%xmm5") /* xmm5 = 1 */ \ |
| 57 |
__ASM_EMIT("xorps %%xmm4, %%xmm4") /* xmm4 = 0 */ \ |
| 58 |
__ASM_EMIT("movaps %%xmm5, %%xmm1") /* xmm1 = 1 */ \ |
| 59 |
__ASM_EMIT("cmpps $6, %%xmm0, %%xmm4") /* xmm4 = [0 > v] */ \ |
| 60 |
__ASM_EMIT("movups 0x00(%[eff]), %%xmm2") /* xmm2 = h s l a */ \ |
| 61 |
__ASM_EMIT("subps %%xmm0, %%xmm1") /* xmm1 = 1 - v */ \ |
| 62 |
__ASM_EMIT("shufps $0x00, %%xmm2, %%xmm2") /* xmm2 = EH */ \ |
| 63 |
__ASM_EMIT("addps %%xmm5, %%xmm0") /* xmm0 = 1 + v */ \ |
| 64 |
__ASM_EMIT("andps %%xmm4, %%xmm0") /* xmm0 = (1+v) & [0 > v] */ \ |
| 65 |
__ASM_EMIT("andnps %%xmm1, %%xmm4") /* xmm4 = (1-v) & [0 <= v] */ \ |
| 66 |
__ASM_EMIT("orps %%xmm4, %%xmm0") /* xmm0 = V = ((1+v) & [0 > v]) | ((1-v) & [0 <= v]) */ \ |
| 67 |
/* xmm0 = V */ \ |
| 68 |
/* xmm2 = EH */ \ |
| 69 |
/* xmm6 = T */ \ |
| 70 |
/* xmm7 = KT */ \ |
| 71 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = V */ \ |
| 72 |
__ASM_EMIT("xorps %%xmm4, %%xmm4") /* xmm4 = 0 */ \ |
| 73 |
__ASM_EMIT("subps %%xmm6, %%xmm1") /* xmm1 = V - T */ \ |
| 74 |
__ASM_EMIT("addps %%xmm2, %%xmm0") /* xmm0 = EH + V */ \ |
| 75 |
__ASM_EMIT("movaps %%xmm1, %%xmm3") /* xmm2 = V - T */ \ |
| 76 |
__ASM_EMIT("addps %%xmm6, %%xmm2") /* xmm3 = EH + T */ \ |
| 77 |
__ASM_EMIT("mulps %%xmm7, %%xmm3") /* xmm3 = (V-T)*KT */ \ |
| 78 |
__ASM_EMIT("cmpps $2, %%xmm1, %%xmm4") /* xmm4 = [(V - T) >= 0] */ \ |
| 79 |
__ASM_EMIT("andps %%xmm4, %%xmm3") /* xmm3 = A = ((V-T)*KT) & [(V - T) >= 0] */ \ |
| 80 |
__ASM_EMIT("andps %%xmm4, %%xmm2") /* xmm2 = (EH + T) & [(V - T) >= 0] */ \ |
| 81 |
__ASM_EMIT("andnps %%xmm0, %%xmm4") /* xmm4 = (EH + V) & [(V - T) < 0] */ \ |
| 82 |
__ASM_EMIT("orps %%xmm2, %%xmm4") /* xmm4 = NH = ((EH + T) & [(V - T) >= 0]) | ((EH + V) & [(V - T) < 0]) */ \ |
| 83 |
__ASM_EMIT("movaps %[XC], %%xmm0") /* xmm0 = 1 */ \ |
| 84 |
__ASM_EMIT("movaps %%xmm4, %%xmm1") /* xmm1 = NH */ \ |
| 85 |
__ASM_EMIT("movaps %%xmm0, %%xmm5") /* xmm5 = 1 */ \ |
| 86 |
__ASM_EMIT("cmpps $6, %%xmm4, %%xmm0") /* xmm0 = [1 > NH] */ \ |
| 87 |
__ASM_EMIT("subps %%xmm5, %%xmm4") /* xmm4 = NH - 1 */ \ |
| 88 |
__ASM_EMIT("movups 0x00(%[eff]), %%xmm2") /* xmm2 = h s l a */ \ |
| 89 |
__ASM_EMIT("andps %%xmm0, %%xmm1") /* xmm1 = NH & [1 > H] */ \ |
| 90 |
__ASM_EMIT("andnps %%xmm4, %%xmm0") /* xmm0 = (NH-1) & [1 <= H] */ \ |
| 91 |
__ASM_EMIT("orps %%xmm1, %%xmm0") /* xmm0 = H = (NH & [1 > H]) | ((NH-1) & [1 <= H]) */ \ |
| 92 |
__ASM_EMIT("movaps %%xmm2, %%xmm1") /* xmm1 = h s l a */ \ |
| 93 |
__ASM_EMIT("shufps $0x55, %%xmm1, %%xmm1") /* xmm1 = S */ \ |
| 94 |
__ASM_EMIT("shufps $0xaa, %%xmm2, %%xmm2") /* xmm2 = L */ \ |
| 95 |
\ |
| 96 |
X4_TRANSPOSE |
| 97 |
|
| 98 |
|
| 99 |
void eff_hsla_hue(float *dst, const float *v, const dsp::hsla_hue_eff_t *eff, size_t count) |
| 100 |
{ |
| 101 |
ARCH_X86_ASM( |
| 102 |
__ASM_EMIT("movaps %[XC], %%xmm6") /* xmm6 = 1 */ |
| 103 |
__ASM_EMIT("movss 0x10(%[eff]), %%xmm4") /* xmm4 = t 0 0 0 */ |
| 104 |
__ASM_EMIT("movaps %%xmm6, %%xmm7") /* xmm7 = 1 */ |
| 105 |
__ASM_EMIT("shufps $0x00, %%xmm4, %%xmm4") /* xmm4 = t */ |
| 106 |
__ASM_EMIT("subps %%xmm4, %%xmm6") /* xmm6 = T = 1 - t */ |
| 107 |
__ASM_EMIT("divps %%xmm4, %%xmm7") /* xmm7 = KT = 1 / t */ |
| 108 |
|
| 109 |
__ASM_EMIT("sub $4, %[count]") |
| 110 |
__ASM_EMIT("jb 2f") |
| 111 |
|
| 112 |
//----------------------------------------------------------------- |
| 113 |
// 4x blocks |
| 114 |
__ASM_EMIT("1:") |
| 115 |
|
| 116 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm0") /* xmm0 = v */ |
| 117 |
EFF_HSLA_HUE_CORE |
| 118 |
|
| 119 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 120 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst])") |
| 121 |
__ASM_EMIT("movups %%xmm2, 0x20(%[dst])") |
| 122 |
__ASM_EMIT("movups %%xmm3, 0x30(%[dst])") |
| 123 |
|
| 124 |
__ASM_EMIT("add $0x10, %[src]") |
| 125 |
__ASM_EMIT("add $0x40, %[dst]") |
| 126 |
__ASM_EMIT("sub $4, %[count]") |
| 127 |
__ASM_EMIT("jae 1b") |
| 128 |
|
| 129 |
__ASM_EMIT("2:") |
| 130 |
__ASM_EMIT("add $4, %[count]") |
| 131 |
__ASM_EMIT("jle 10f") |
| 132 |
|
| 133 |
//----------------------------------------------------------------- |
| 134 |
// 1x - 3x block |
| 135 |
// Load last variable-sized chunk |
| 136 |
__ASM_EMIT("test $1, %[count]") |
| 137 |
__ASM_EMIT("jz 4f") |
| 138 |
__ASM_EMIT("movss 0x00(%[src]), %%xmm0") |
| 139 |
__ASM_EMIT("add $0x04, %[src]") |
| 140 |
__ASM_EMIT("movlhps %%xmm0, %%xmm0") |
| 141 |
__ASM_EMIT("4:") |
| 142 |
__ASM_EMIT("test $2, %[count]") |
| 143 |
__ASM_EMIT("jz 6f") |
| 144 |
__ASM_EMIT("movlps 0x00(%[src]), %%xmm0") |
| 145 |
__ASM_EMIT("6:") |
| 146 |
|
| 147 |
EFF_HSLA_HUE_CORE |
| 148 |
|
| 149 |
// Store last chunk |
| 150 |
__ASM_EMIT("test $1, %[count]") |
| 151 |
__ASM_EMIT("jz 8f") |
| 152 |
__ASM_EMIT("movups %%xmm2, 0x00(%[dst])") |
| 153 |
__ASM_EMIT("add $0x10, %[dst]") |
| 154 |
__ASM_EMIT("8:") |
| 155 |
__ASM_EMIT("test $2, %[count]") |
| 156 |
__ASM_EMIT("jz 10f") |
| 157 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 158 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst])") |
| 159 |
|
| 160 |
__ASM_EMIT("10:") |
| 161 |
|
| 162 |
: [dst] "+r" (dst), [src] "+r" (v), [count] "+r" (count) |
| 163 |
: [eff] "r" (eff), |
| 164 |
[XC] "m" (EFF_HSLA_HUE_XC) |
| 165 |
: "cc", "memory", |
| 166 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 167 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 168 |
); |
| 169 |
/* |
| 170 |
float value, hue, alpha; |
| 171 |
float t = 1.0f - eff->thresh; |
| 172 |
float kt = 1.0f / eff->thresh; |
| 173 |
|
| 174 |
for (size_t i=0; i<count; ++i, dst += 4) |
| 175 |
{ |
| 176 |
value = v[i]; |
| 177 |
value = (0 > value) ? 1.0f + value : 1.0f - value; |
| 178 |
|
| 179 |
if ((value - t) >= 0) |
| 180 |
{ |
| 181 |
hue = eff->h + t; |
| 182 |
alpha = ((value - t) * kt); |
| 183 |
} |
| 184 |
else |
| 185 |
{ |
| 186 |
hue = eff->h + value; |
| 187 |
alpha = 0.0f; |
| 188 |
} |
| 189 |
|
| 190 |
dst[0] = (hue < 1.0f) ? hue : hue - 1.0f; |
| 191 |
dst[1] = eff->s; |
| 192 |
dst[2] = eff->l; |
| 193 |
dst[3] = alpha; |
| 194 |
}*/ |
| 195 |
} |
| 196 |
|
| 197 |
#undef EFF_HSLA_HUE_CORE |
| 198 |
|
| 199 |
IF_ARCH_X86( |
| 200 |
static const float EFF_HSLA_ALPHA_XC[] __lsp_aligned16 = |
| 201 |
{ |
| 202 |
LSP_DSP_VEC4(1.0f) |
| 203 |
}; |
| 204 |
) |
| 205 |
|
| 206 |
#define EFF_HSLA_ALPHA_CORE \ |
| 207 |
/* xmm3 = v */ \ |
| 208 |
__ASM_EMIT("xorps %%xmm4, %%xmm4") /* xmm4 = 0 */ \ |
| 209 |
__ASM_EMIT("movaps %[XC], %%xmm5") /* xmm5 = 1 */ \ |
| 210 |
__ASM_EMIT("cmpps $6, %%xmm3, %%xmm4") /* xmm4 = [0 > v] */ \ |
| 211 |
__ASM_EMIT("movups 0x00(%[eff]), %%xmm0") /* xmm0 = hsla */ \ |
| 212 |
__ASM_EMIT("movaps %%xmm3, %%xmm6") /* xmm6 = v */ \ |
| 213 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = hsla */ \ |
| 214 |
__ASM_EMIT("addps %%xmm5, %%xmm3") /* xmm3 = 1 + v */ \ |
| 215 |
__ASM_EMIT("movaps %%xmm1, %%xmm2") /* xmm2 = hsla */ \ |
| 216 |
__ASM_EMIT("subps %%xmm6, %%xmm5") /* xmm5 = 1 - v */ \ |
| 217 |
__ASM_EMIT("shufps $0x00, %%xmm0, %%xmm0") /* xmm0 = H */ \ |
| 218 |
__ASM_EMIT("andps %%xmm4, %%xmm3") /* xmm4 = (1+v) & [0 > v] */ \ |
| 219 |
__ASM_EMIT("shufps $0x55, %%xmm1, %%xmm1") /* xmm1 = S */ \ |
| 220 |
__ASM_EMIT("andnps %%xmm5, %%xmm4") /* xmm4 = (1-v) & [0 <= v] */ \ |
| 221 |
__ASM_EMIT("shufps $0xaa, %%xmm2, %%xmm2") /* xmm1 = L */ \ |
| 222 |
__ASM_EMIT("orps %%xmm4, %%xmm3") /* xmm3 = A = ((1+v) & [0 > v]) | ((1-v) & [0 <= v]) */ \ |
| 223 |
\ |
| 224 |
X4_TRANSPOSE |
| 225 |
|
| 226 |
/* |
| 227 |
value = v[i]; |
| 228 |
value = (0.0f > value) ? 1.0f + value : 1.0f - value; |
| 229 |
|
| 230 |
dst[0] = eff->h; |
| 231 |
dst[1] = eff->s; |
| 232 |
dst[2] = eff->l; |
| 233 |
dst[3] = value; // Fill alpha channel |
| 234 |
*/ |
| 235 |
|
| 236 |
void eff_hsla_alpha(float *dst, const float *v, const dsp::hsla_alpha_eff_t *eff, size_t count) |
| 237 |
{ |
| 238 |
ARCH_X86_ASM( |
| 239 |
__ASM_EMIT("sub $4, %[count]") |
| 240 |
__ASM_EMIT("jb 2f") |
| 241 |
|
| 242 |
//----------------------------------------------------------------- |
| 243 |
// 4x blocks |
| 244 |
__ASM_EMIT("1:") |
| 245 |
|
| 246 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm3") /* xmm3 = v */ |
| 247 |
EFF_HSLA_ALPHA_CORE |
| 248 |
|
| 249 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 250 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst])") |
| 251 |
__ASM_EMIT("movups %%xmm2, 0x20(%[dst])") |
| 252 |
__ASM_EMIT("movups %%xmm3, 0x30(%[dst])") |
| 253 |
|
| 254 |
__ASM_EMIT("add $0x10, %[src]") |
| 255 |
__ASM_EMIT("add $0x40, %[dst]") |
| 256 |
__ASM_EMIT("sub $4, %[count]") |
| 257 |
__ASM_EMIT("jae 1b") |
| 258 |
|
| 259 |
__ASM_EMIT("2:") |
| 260 |
__ASM_EMIT("add $4, %[count]") |
| 261 |
__ASM_EMIT("jle 10f") |
| 262 |
|
| 263 |
//----------------------------------------------------------------- |
| 264 |
// 1x - 3x block |
| 265 |
// Load last variable-sized chunk |
| 266 |
__ASM_EMIT("test $1, %[count]") |
| 267 |
__ASM_EMIT("jz 4f") |
| 268 |
__ASM_EMIT("movss 0x00(%[src]), %%xmm3") |
| 269 |
__ASM_EMIT("add $0x04, %[src]") |
| 270 |
__ASM_EMIT("movlhps %%xmm3, %%xmm3") |
| 271 |
__ASM_EMIT("4:") |
| 272 |
__ASM_EMIT("test $2, %[count]") |
| 273 |
__ASM_EMIT("jz 6f") |
| 274 |
__ASM_EMIT("movlps 0x00(%[src]), %%xmm3") |
| 275 |
__ASM_EMIT("6:") |
| 276 |
|
| 277 |
EFF_HSLA_ALPHA_CORE |
| 278 |
|
| 279 |
// Store last chunk |
| 280 |
__ASM_EMIT("test $1, %[count]") |
| 281 |
__ASM_EMIT("jz 8f") |
| 282 |
__ASM_EMIT("movups %%xmm2, 0x00(%[dst])") |
| 283 |
__ASM_EMIT("add $0x10, %[dst]") |
| 284 |
__ASM_EMIT("8:") |
| 285 |
__ASM_EMIT("test $2, %[count]") |
| 286 |
__ASM_EMIT("jz 10f") |
| 287 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 288 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst])") |
| 289 |
|
| 290 |
__ASM_EMIT("10:") |
| 291 |
|
| 292 |
: [dst] "+r" (dst), [src] "+r" (v), [count] "+r" (count) |
| 293 |
: [eff] "r" (eff), |
| 294 |
[XC] "m" (EFF_HSLA_ALPHA_XC) |
| 295 |
: "cc", "memory", |
| 296 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 297 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 298 |
); |
| 299 |
|
| 300 |
} |
| 301 |
#undef EFF_HSLA_ALPHA_CORE |
| 302 |
|
| 303 |
IF_ARCH_X86( |
| 304 |
static const uint32_t EFF_HSLA_SAT_XC[] __lsp_aligned16 = |
| 305 |
{ |
| 306 |
LSP_DSP_VEC4(0x3f800000), |
| 307 |
LSP_DSP_VEC4(0x7fffffff) |
| 308 |
}; |
| 309 |
) |
| 310 |
|
| 311 |
#define EFF_HSLA_SAT_CORE \ |
| 312 |
/* xmm1 = v, xmm6 = T, xmm7 = KT */ \ |
| 313 |
__ASM_EMIT("movaps %%xmm6, %%xmm0") /* xmm0 = T */ \ |
| 314 |
__ASM_EMIT("andps 0x10(%[XC]), %%xmm1") /* xmm1 = V */ \ |
| 315 |
__ASM_EMIT("xorps %%xmm5, %%xmm5") /* xmm5 = 0 */ \ |
| 316 |
__ASM_EMIT("subps %%xmm1, %%xmm0") /* xmm0 = T-V */ \ |
| 317 |
__ASM_EMIT("cmpps $6, %%xmm0, %%xmm5") /* xmm5 = [ 0 >= T - V] */ \ |
| 318 |
__ASM_EMIT("mulps %%xmm7, %%xmm0") /* xmm0 = (T-V)*KT */ \ |
| 319 |
__ASM_EMIT("movaps %%xmm5, %%xmm3") /* xmm3 = [ 0 >= T - V] */ \ |
| 320 |
__ASM_EMIT("movups 0x00(%[eff]), %%xmm2") /* xmm2 = hsla */ \ |
| 321 |
__ASM_EMIT("andps %%xmm5, %%xmm1") /* xmm1 = V & [ 0 >= T - V] */ \ |
| 322 |
__ASM_EMIT("andnps %%xmm0, %%xmm3") /* xmm3 = A = (T-V)*KT & [ 0 < T - V] */ \ |
| 323 |
__ASM_EMIT("andnps %%xmm6, %%xmm5") /* xmm5 = T & [0 < T-V] */ \ |
| 324 |
__ASM_EMIT("movaps %%xmm2, %%xmm0") /* xmm0 = hsla */ \ |
| 325 |
__ASM_EMIT("orps %%xmm5, %%xmm1") /* xmm1 = KS = (V & [ 0 >= T - V]) | (T & [0 < T-V]) */ \ |
| 326 |
__ASM_EMIT("movaps %%xmm2, %%xmm4") /* xmm4 = hsla */ \ |
| 327 |
__ASM_EMIT("shufps $0x00, %%xmm0, %%xmm0") /* xmm0 = H */ \ |
| 328 |
__ASM_EMIT("shufps $0x55, %%xmm4, %%xmm4") /* xmm4 = s */ \ |
| 329 |
__ASM_EMIT("shufps $0xaa, %%xmm2, %%xmm2") /* xmm2 = L */ \ |
| 330 |
__ASM_EMIT("mulps %%xmm4, %%xmm1") /* xmm1 = S = s * KS */ \ |
| 331 |
\ |
| 332 |
X4_TRANSPOSE |
| 333 |
|
| 334 |
/* |
| 335 |
kt = 1.0f / eff->thresh; |
| 336 |
value = (value >= 0.0f) ? value : -value; |
| 337 |
|
| 338 |
if (0 >= (eff->thresh - value)) |
| 339 |
{ |
| 340 |
dst[1] = eff->s * value; |
| 341 |
dst[3] = 0.0f; |
| 342 |
} |
| 343 |
else |
| 344 |
{ |
| 345 |
dst[1] = eff->s * eff->thresh; |
| 346 |
dst[3] = (eff->thresh - value) * kt; |
| 347 |
} |
| 348 |
|
| 349 |
dst[0] = eff->h; |
| 350 |
dst[2] = eff->l; |
| 351 |
*/ |
| 352 |
|
| 353 |
void eff_hsla_sat(float *dst, const float *v, const dsp::hsla_sat_eff_t *eff, size_t count) |
| 354 |
{ |
| 355 |
ARCH_X86_ASM( |
| 356 |
__ASM_EMIT("movss 0x10(%[eff]), %%xmm6") /* xmm6 = t 0 0 0 */ |
| 357 |
__ASM_EMIT("movaps 0x00(%[XC]), %%xmm7") /* xmm7 = 1 */ |
| 358 |
__ASM_EMIT("shufps $0x00, %%xmm6, %%xmm6") /* xmm6 = T */ |
| 359 |
__ASM_EMIT("divps %%xmm6, %%xmm7") /* xmm7 = KT = 1 / t */ |
| 360 |
|
| 361 |
__ASM_EMIT("sub $4, %[count]") |
| 362 |
__ASM_EMIT("jb 2f") |
| 363 |
|
| 364 |
//----------------------------------------------------------------- |
| 365 |
// 4x blocks |
| 366 |
__ASM_EMIT("1:") |
| 367 |
|
| 368 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm1") /* xmm1 = v */ |
| 369 |
EFF_HSLA_SAT_CORE |
| 370 |
|
| 371 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 372 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst])") |
| 373 |
__ASM_EMIT("movups %%xmm2, 0x20(%[dst])") |
| 374 |
__ASM_EMIT("movups %%xmm3, 0x30(%[dst])") |
| 375 |
|
| 376 |
__ASM_EMIT("add $0x10, %[src]") |
| 377 |
__ASM_EMIT("add $0x40, %[dst]") |
| 378 |
__ASM_EMIT("sub $4, %[count]") |
| 379 |
__ASM_EMIT("jae 1b") |
| 380 |
|
| 381 |
__ASM_EMIT("2:") |
| 382 |
__ASM_EMIT("add $4, %[count]") |
| 383 |
__ASM_EMIT("jle 10f") |
| 384 |
|
| 385 |
//----------------------------------------------------------------- |
| 386 |
// 1x - 3x block |
| 387 |
// Load last variable-sized chunk |
| 388 |
__ASM_EMIT("test $1, %[count]") |
| 389 |
__ASM_EMIT("jz 4f") |
| 390 |
__ASM_EMIT("movss 0x00(%[src]), %%xmm1") |
| 391 |
__ASM_EMIT("add $0x04, %[src]") |
| 392 |
__ASM_EMIT("movlhps %%xmm1, %%xmm1") |
| 393 |
__ASM_EMIT("4:") |
| 394 |
__ASM_EMIT("test $2, %[count]") |
| 395 |
__ASM_EMIT("jz 6f") |
| 396 |
__ASM_EMIT("movlps 0x00(%[src]), %%xmm1") |
| 397 |
__ASM_EMIT("6:") |
| 398 |
|
| 399 |
EFF_HSLA_SAT_CORE |
| 400 |
|
| 401 |
// Store last chunk |
| 402 |
__ASM_EMIT("test $1, %[count]") |
| 403 |
__ASM_EMIT("jz 8f") |
| 404 |
__ASM_EMIT("movups %%xmm2, 0x00(%[dst])") |
| 405 |
__ASM_EMIT("add $0x10, %[dst]") |
| 406 |
__ASM_EMIT("8:") |
| 407 |
__ASM_EMIT("test $2, %[count]") |
| 408 |
__ASM_EMIT("jz 10f") |
| 409 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 410 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst])") |
| 411 |
|
| 412 |
__ASM_EMIT("10:") |
| 413 |
|
| 414 |
: [dst] "+r" (dst), [src] "+r" (v), [count] "+r" (count) |
| 415 |
: [eff] "r" (eff), |
| 416 |
[XC] "r" (EFF_HSLA_SAT_XC) |
| 417 |
: "cc", "memory", |
| 418 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 419 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 420 |
); |
| 421 |
} |
| 422 |
|
| 423 |
#undef EFF_HSLA_SAT_CORE |
| 424 |
|
| 425 |
IF_ARCH_X86( |
| 426 |
static const uint32_t EFF_HSLA_LIGHT_XC[] __lsp_aligned16 = |
| 427 |
{ |
| 428 |
LSP_DSP_VEC4(0x3f800000), |
| 429 |
LSP_DSP_VEC4(0x7fffffff) |
| 430 |
}; |
| 431 |
) |
| 432 |
|
| 433 |
#define EFF_HSLA_LIGHT_CORE \ |
| 434 |
/* xmm2 = v, xmm6 = T, xmm7 = KT */ \ |
| 435 |
__ASM_EMIT("movaps %%xmm6, %%xmm0") /* xmm0 = T */ \ |
| 436 |
__ASM_EMIT("andps 0x10(%[XC]), %%xmm2") /* xmm2 = V */ \ |
| 437 |
__ASM_EMIT("xorps %%xmm5, %%xmm5") /* xmm5 = 0 */ \ |
| 438 |
__ASM_EMIT("subps %%xmm2, %%xmm0") /* xmm0 = T-V */ \ |
| 439 |
__ASM_EMIT("cmpps $6, %%xmm0, %%xmm5") /* xmm5 = [ 0 >= T - V] */ \ |
| 440 |
__ASM_EMIT("mulps %%xmm7, %%xmm0") /* xmm0 = (T-V)*KT */ \ |
| 441 |
__ASM_EMIT("movaps %%xmm5, %%xmm3") /* xmm3 = [ 0 >= T - V] */ \ |
| 442 |
__ASM_EMIT("movups 0x00(%[eff]), %%xmm1") /* xmm1 = hsla */ \ |
| 443 |
__ASM_EMIT("andps %%xmm5, %%xmm2") /* xmm2 = V & [ 0 >= T - V] */ \ |
| 444 |
__ASM_EMIT("andnps %%xmm0, %%xmm3") /* xmm3 = A = (T-V)*KT & [ 0 < T - V] */ \ |
| 445 |
__ASM_EMIT("andnps %%xmm6, %%xmm5") /* xmm5 = T & [0 < T-V] */ \ |
| 446 |
__ASM_EMIT("movaps %%xmm1, %%xmm0") /* xmm0 = hsla */ \ |
| 447 |
__ASM_EMIT("orps %%xmm5, %%xmm2") /* xmm2 = KL = (V & [ 0 >= T - V]) | (T & [0 < T-V]) */ \ |
| 448 |
__ASM_EMIT("movaps %%xmm1, %%xmm4") /* xmm4 = hsla */ \ |
| 449 |
__ASM_EMIT("shufps $0x00, %%xmm0, %%xmm0") /* xmm0 = H */ \ |
| 450 |
__ASM_EMIT("shufps $0xaa, %%xmm4, %%xmm4") /* xmm4 = l */ \ |
| 451 |
__ASM_EMIT("shufps $0x55, %%xmm1, %%xmm1") /* xmm1 = S */ \ |
| 452 |
__ASM_EMIT("mulps %%xmm4, %%xmm2") /* xmm2 = L = l * KL */ \ |
| 453 |
\ |
| 454 |
X4_TRANSPOSE |
| 455 |
|
| 456 |
/* |
| 457 |
value = (value >= 0.0f) ? value : -value; |
| 458 |
|
| 459 |
if (0 >= (eff->thresh - value)) |
| 460 |
{ |
| 461 |
dst[2] = eff->l * value; |
| 462 |
dst[3] = 0.0f; |
| 463 |
} |
| 464 |
else |
| 465 |
{ |
| 466 |
dst[2] = eff->l * eff->thresh; |
| 467 |
dst[3] = (eff->thresh - value) * kt; |
| 468 |
} |
| 469 |
|
| 470 |
dst[0] = eff->h; |
| 471 |
dst[1] = eff->s; |
| 472 |
*/ |
| 473 |
|
| 474 |
void eff_hsla_light(float *dst, const float *v, const dsp::hsla_light_eff_t *eff, size_t count) |
| 475 |
{ |
| 476 |
ARCH_X86_ASM( |
| 477 |
__ASM_EMIT("movss 0x10(%[eff]), %%xmm6") /* xmm6 = t 0 0 0 */ |
| 478 |
__ASM_EMIT("movaps 0x00(%[XC]), %%xmm7") /* xmm7 = 1 */ |
| 479 |
__ASM_EMIT("shufps $0x00, %%xmm6, %%xmm6") /* xmm6 = T */ |
| 480 |
__ASM_EMIT("divps %%xmm6, %%xmm7") /* xmm7 = KT = 1 / t */ |
| 481 |
|
| 482 |
__ASM_EMIT("sub $4, %[count]") |
| 483 |
__ASM_EMIT("jb 2f") |
| 484 |
|
| 485 |
//----------------------------------------------------------------- |
| 486 |
// 4x blocks |
| 487 |
__ASM_EMIT("1:") |
| 488 |
|
| 489 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm2") /* xmm1 = v */ |
| 490 |
EFF_HSLA_LIGHT_CORE |
| 491 |
|
| 492 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 493 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst])") |
| 494 |
__ASM_EMIT("movups %%xmm2, 0x20(%[dst])") |
| 495 |
__ASM_EMIT("movups %%xmm3, 0x30(%[dst])") |
| 496 |
|
| 497 |
__ASM_EMIT("add $0x10, %[src]") |
| 498 |
__ASM_EMIT("add $0x40, %[dst]") |
| 499 |
__ASM_EMIT("sub $4, %[count]") |
| 500 |
__ASM_EMIT("jae 1b") |
| 501 |
|
| 502 |
__ASM_EMIT("2:") |
| 503 |
__ASM_EMIT("add $4, %[count]") |
| 504 |
__ASM_EMIT("jle 10f") |
| 505 |
|
| 506 |
//----------------------------------------------------------------- |
| 507 |
// 1x - 3x block |
| 508 |
// Load last variable-sized chunk |
| 509 |
__ASM_EMIT("test $1, %[count]") |
| 510 |
__ASM_EMIT("jz 4f") |
| 511 |
__ASM_EMIT("movss 0x00(%[src]), %%xmm2") |
| 512 |
__ASM_EMIT("add $0x04, %[src]") |
| 513 |
__ASM_EMIT("movlhps %%xmm2, %%xmm2") |
| 514 |
__ASM_EMIT("4:") |
| 515 |
__ASM_EMIT("test $2, %[count]") |
| 516 |
__ASM_EMIT("jz 6f") |
| 517 |
__ASM_EMIT("movlps 0x00(%[src]), %%xmm2") |
| 518 |
__ASM_EMIT("6:") |
| 519 |
|
| 520 |
EFF_HSLA_LIGHT_CORE |
| 521 |
|
| 522 |
// Store last chunk |
| 523 |
__ASM_EMIT("test $1, %[count]") |
| 524 |
__ASM_EMIT("jz 8f") |
| 525 |
__ASM_EMIT("movups %%xmm2, 0x00(%[dst])") |
| 526 |
__ASM_EMIT("add $0x10, %[dst]") |
| 527 |
__ASM_EMIT("8:") |
| 528 |
__ASM_EMIT("test $2, %[count]") |
| 529 |
__ASM_EMIT("jz 10f") |
| 530 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 531 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst])") |
| 532 |
|
| 533 |
__ASM_EMIT("10:") |
| 534 |
|
| 535 |
: [dst] "+r" (dst), [src] "+r" (v), [count] "+r" (count) |
| 536 |
: [eff] "r" (eff), |
| 537 |
[XC] "r" (EFF_HSLA_LIGHT_XC) |
| 538 |
: "cc", "memory", |
| 539 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 540 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 541 |
); |
| 542 |
} |
| 543 |
|
| 544 |
#undef EFF_HSLA_LIGHT_CORE |
| 545 |
} |
| 546 |
} |
| 547 |
|
| 548 |
#undef X4_TRANSPOSE |
| 549 |
|
| 550 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE2_GRAPHICS_EFFECTS_H_ */ |