| 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_AVX2_GRAPHICS_EFFECTS_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_AVX2_GRAPHICS_EFFECTS_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_AVX2_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX2_IMPL */ |
| 28 |
|
| 29 |
#include <private/dsp/arch/x86/avx2/graphics/transpose.h> |
| 30 |
|
| 31 |
namespace lsp |
| 32 |
{ |
| 33 |
namespace avx2 |
| 34 |
{ |
| 35 |
IF_ARCH_X86_64( |
| 36 |
static const float EFF_HSLA_HUE_XC[] __lsp_aligned32 = |
| 37 |
{ |
| 38 |
LSP_DSP_VEC8(1.0f) |
| 39 |
}; |
| 40 |
) |
| 41 |
|
| 42 |
#define EFF_HSLA_HUE_CORE_X16 \ |
| 43 |
/* ymm0 = v, ymm10 = h, ymm11 = s, ymm12 = l, ymm13 = 1, ymm14 = T, ymm15 = KT */ \ |
| 44 |
__ASM_EMIT("vaddps %%ymm0, %%ymm13, %%ymm2") /* ymm2 = 1+v */ \ |
| 45 |
__ASM_EMIT("vaddps %%ymm4, %%ymm13, %%ymm6") \ |
| 46 |
__ASM_EMIT("vsubps %%ymm0, %%ymm13, %%ymm3") /* ymm3 = 1-v */ \ |
| 47 |
__ASM_EMIT("vsubps %%ymm4, %%ymm13, %%ymm7") \ |
| 48 |
__ASM_EMIT("vblendvps %%ymm0, %%ymm2, %%ymm3, %%ymm0") /* ymm0 = V = (v < 0) ? 1+v : 1-v */ \ |
| 49 |
__ASM_EMIT("vblendvps %%ymm4, %%ymm6, %%ymm7, %%ymm4") \ |
| 50 |
/* ymm0 = V, ymm1 = EH, ymm14 = T, ymm15 = KT */ \ |
| 51 |
__ASM_EMIT("vsubps %%ymm14, %%ymm0, %%ymm3") /* ymm3 = V-T */ \ |
| 52 |
__ASM_EMIT("vsubps %%ymm14, %%ymm4, %%ymm7") \ |
| 53 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm0, %%ymm14, %%ymm0") /* ymm0 = (V-T) < 0 ? V : T */ \ |
| 54 |
__ASM_EMIT("vblendvps %%ymm7, %%ymm4, %%ymm14, %%ymm4") \ |
| 55 |
__ASM_EMIT("vxorps %%ymm8, %%ymm8, %%ymm8") /* ymm8 = 0 */ \ |
| 56 |
__ASM_EMIT("vaddps %%ymm10, %%ymm0, %%ymm0") /* ymm0 = NH = h + ((V-T) < 0 ? V : T) */ \ |
| 57 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm8, %%ymm3, %%ymm3") /* ymm3 = (V-T) < 0 ? 0 : V-T */ \ |
| 58 |
__ASM_EMIT("vaddps %%ymm10, %%ymm4, %%ymm4") \ |
| 59 |
__ASM_EMIT("vblendvps %%ymm7, %%ymm8, %%ymm7, %%ymm7") \ |
| 60 |
__ASM_EMIT("vsubps %%ymm13, %%ymm0, %%ymm1") /* ymm1 = NH-1 */ \ |
| 61 |
__ASM_EMIT("vmulps %%ymm15, %%ymm3, %%ymm3") /* ymm3 = A = KT * ((V-T) < 0 ? 0 : V-T) */ \ |
| 62 |
__ASM_EMIT("vsubps %%ymm13, %%ymm4, %%ymm5") \ |
| 63 |
__ASM_EMIT("vmulps %%ymm15, %%ymm7, %%ymm7") \ |
| 64 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm2") /* ymm2 = L */ \ |
| 65 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm6") \ |
| 66 |
__ASM_EMIT("vblendvps %%ymm1, %%ymm0, %%ymm1, %%ymm0") /* ymm0 = H = ((NH-1)<0) ? NH : NH-1 */ \ |
| 67 |
__ASM_EMIT("vblendvps %%ymm5, %%ymm4, %%ymm5, %%ymm4") \ |
| 68 |
__ASM_EMIT("vmovaps %%ymm11, %%ymm1") /* ymm1 = S */ \ |
| 69 |
__ASM_EMIT("vmovaps %%ymm11, %%ymm5") \ |
| 70 |
MAT4_TRANSPOSE("%%ymm0", "%%ymm1", "%%ymm2", "%%ymm3", "%%ymm8", "%%ymm9") \ |
| 71 |
MAT4_TRANSPOSE("%%ymm4", "%%ymm5", "%%ymm6", "%%ymm7", "%%ymm8", "%%ymm9") \ |
| 72 |
MAT4X2_INTERLEAVE("0", "1", "2", "3", "8", "9") \ |
| 73 |
MAT4X2_INTERLEAVE("4", "5", "6", "7", "8", "9") |
| 74 |
|
| 75 |
#define EFF_HSLA_HUE_CORE_X8 \ |
| 76 |
/* ymm0 = v, ymm10 = h, ymm11 = s, ymm12 = l, ymm13 = 1, ymm14 = T, ymm15 = KT */ \ |
| 77 |
__ASM_EMIT("vaddps %%ymm0, %%ymm13, %%ymm2") /* ymm2 = 1+v */ \ |
| 78 |
__ASM_EMIT("vsubps %%ymm0, %%ymm13, %%ymm3") /* ymm3 = 1-v */ \ |
| 79 |
__ASM_EMIT("vblendvps %%ymm0, %%ymm2, %%ymm3, %%ymm0") /* ymm0 = V = (v < 0) ? 1+v : 1-v */ \ |
| 80 |
/* ymm0 = V, ymm1 = EH, ymm14 = T, ymm15 = KT */ \ |
| 81 |
__ASM_EMIT("vsubps %%ymm14, %%ymm0, %%ymm3") /* ymm3 = V-T */ \ |
| 82 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm0, %%ymm14, %%ymm0") /* ymm0 = (V-T) < 0 ? V : T */ \ |
| 83 |
__ASM_EMIT("vxorps %%ymm8, %%ymm8, %%ymm8") /* ymm8 = 0 */ \ |
| 84 |
__ASM_EMIT("vaddps %%ymm10, %%ymm0, %%ymm0") /* ymm0 = NH = h + ((V-T) < 0 ? V : T) */ \ |
| 85 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm8, %%ymm3, %%ymm3") /* ymm3 = (V-T) < 0 ? 0 : V-T */ \ |
| 86 |
__ASM_EMIT("vsubps %%ymm13, %%ymm0, %%ymm1") /* ymm1 = NH-1 */ \ |
| 87 |
__ASM_EMIT("vmulps %%ymm15, %%ymm3, %%ymm3") /* ymm3 = A = KT * ((V-T) < 0 ? 0 : V-T) */ \ |
| 88 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm2") /* ymm2 = L */ \ |
| 89 |
__ASM_EMIT("vblendvps %%ymm1, %%ymm0, %%ymm1, %%ymm0") /* ymm0 = H = ((NH-1)<0) ? NH : NH-1 */ \ |
| 90 |
__ASM_EMIT("vmovaps %%ymm11, %%ymm1") /* ymm1 = S */ \ |
| 91 |
MAT4_TRANSPOSE("%%ymm0", "%%ymm1", "%%ymm2", "%%ymm3", "%%ymm8", "%%ymm9") \ |
| 92 |
MAT4X2_INTERLEAVE("0", "1", "2", "3", "8", "9") |
| 93 |
|
| 94 |
#define EFF_HSLA_HUE_CORE_X4 \ |
| 95 |
/* xmm0 = v, xmm10 = h, xmm11 = s, xmm12 = l, xmm13 = 1, xmm14 = T, xmm15 = KT */ \ |
| 96 |
__ASM_EMIT("vaddps %%xmm0, %%xmm13, %%xmm2") /* xmm2 = 1+v */ \ |
| 97 |
__ASM_EMIT("vsubps %%xmm0, %%xmm13, %%xmm3") /* xmm3 = 1-v */ \ |
| 98 |
__ASM_EMIT("vblendvps %%xmm0, %%xmm2, %%xmm3, %%xmm0") /* xmm0 = V = (v < 0) ? 1+v : 1-v */ \ |
| 99 |
/* xmm0 = V, xmm1 = EH, xmm14 = T, xmm15 = KT */ \ |
| 100 |
__ASM_EMIT("vsubps %%xmm14, %%xmm0, %%xmm3") /* xmm3 = V-T */ \ |
| 101 |
__ASM_EMIT("vblendvps %%xmm3, %%xmm0, %%xmm14, %%xmm0") /* xmm0 = (V-T) < 0 ? V : T */ \ |
| 102 |
__ASM_EMIT("vxorps %%xmm8, %%xmm8, %%xmm8") /* xmm8 = 0 */ \ |
| 103 |
__ASM_EMIT("vaddps %%xmm10, %%xmm0, %%xmm0") /* xmm0 = NH = h + ((V-T) < 0 ? V : T) */ \ |
| 104 |
__ASM_EMIT("vblendvps %%xmm3, %%xmm8, %%xmm3, %%xmm3") /* xmm3 = (V-T) < 0 ? 0 : V-T */ \ |
| 105 |
__ASM_EMIT("vsubps %%xmm13, %%xmm0, %%xmm1") /* xmm1 = NH-1 */ \ |
| 106 |
__ASM_EMIT("vmulps %%xmm15, %%xmm3, %%xmm3") /* xmm3 = A = KT * ((V-T) < 0 ? 0 : V-T) */ \ |
| 107 |
__ASM_EMIT("vmovaps %%xmm12, %%xmm2") /* xmm2 = L */ \ |
| 108 |
__ASM_EMIT("vblendvps %%xmm1, %%xmm0, %%xmm1, %%xmm0") /* xmm0 = H = ((NH-1)<0) ? NH : NH-1 */ \ |
| 109 |
__ASM_EMIT("vmovaps %%xmm11, %%xmm1") /* xmm1 = S */ \ |
| 110 |
MAT4_TRANSPOSE("%%xmm0", "%%xmm1", "%%xmm2", "%%xmm3", "%%xmm8", "%%xmm9") |
| 111 |
|
| 112 |
/* |
| 113 |
float value, hue, alpha; |
| 114 |
float t = 1.0f - eff->thresh; |
| 115 |
float kt = 1.0f / eff->thresh; |
| 116 |
|
| 117 |
for (size_t i=0; i<count; ++i, dst += 4) |
| 118 |
{ |
| 119 |
value = v[i]; |
| 120 |
value = (value < 0) ? 1.0f + value : 1.0f - value; |
| 121 |
|
| 122 |
if ((value - t) >= 0) |
| 123 |
{ |
| 124 |
hue = eff->h + t; |
| 125 |
alpha = ((value - t) * kt); |
| 126 |
} |
| 127 |
else |
| 128 |
{ |
| 129 |
hue = eff->h + value; |
| 130 |
alpha = 0.0f; |
| 131 |
} |
| 132 |
|
| 133 |
dst[0] = (hue < 1.0f) ? hue : hue - 1.0f; |
| 134 |
dst[1] = eff->s; |
| 135 |
dst[2] = eff->l; |
| 136 |
dst[3] = alpha; |
| 137 |
}*/ |
| 138 |
|
| 139 |
void x64_eff_hsla_hue(float *dst, const float *v, const dsp::hsla_hue_eff_t *eff, size_t count) |
| 140 |
{ |
| 141 |
ARCH_X86_64_ASM( |
| 142 |
//----------------------------------------------------------------- |
| 143 |
// Prepare |
| 144 |
__ASM_EMIT("vbroadcastf128 0x00(%[eff]), %%ymm10") /* ymm10 = h s l a h s l a */ |
| 145 |
__ASM_EMIT("vbroadcastss 0x10(%[eff]), %%ymm14") /* ymm14 = t */ |
| 146 |
__ASM_EMIT("vmovaps 0x00 + %[XC], %%ymm13") /* ymm13 = 1 */ |
| 147 |
__ASM_EMIT("vshufps $0xaa, %%ymm10, %%ymm10, %%ymm12") /* ymm12 = l */ |
| 148 |
__ASM_EMIT("vdivps %%ymm14, %%ymm13, %%ymm15") /* ymm15 = KT = 1 / t */ |
| 149 |
__ASM_EMIT("vshufps $0x55, %%ymm10, %%ymm10, %%ymm11") /* ymm12 = s */ |
| 150 |
__ASM_EMIT("vsubps %%ymm14, %%ymm13, %%ymm14") /* ymm14 = T = 1 - t */ |
| 151 |
__ASM_EMIT("vshufps $0x00, %%ymm10, %%ymm10, %%ymm10") /* ymm12 = h */ |
| 152 |
|
| 153 |
//----------------------------------------------------------------- |
| 154 |
// x16 blocks |
| 155 |
__ASM_EMIT("sub $16, %[count]") |
| 156 |
__ASM_EMIT("jb 2f") |
| 157 |
__ASM_EMIT("1:") |
| 158 |
__ASM_EMIT("vmovups 0x00(%[src]), %%ymm0") |
| 159 |
__ASM_EMIT("vmovups 0x20(%[src]), %%ymm4") |
| 160 |
EFF_HSLA_HUE_CORE_X16 |
| 161 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[dst])") |
| 162 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[dst])") |
| 163 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[dst])") |
| 164 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[dst])") |
| 165 |
__ASM_EMIT("vmovups %%ymm4, 0x080(%[dst])") |
| 166 |
__ASM_EMIT("vmovups %%ymm5, 0x0a0(%[dst])") |
| 167 |
__ASM_EMIT("vmovups %%ymm6, 0x0c0(%[dst])") |
| 168 |
__ASM_EMIT("vmovups %%ymm7, 0x0e0(%[dst])") |
| 169 |
__ASM_EMIT("add $0x40, %[src]") |
| 170 |
__ASM_EMIT("add $0x100, %[dst]") |
| 171 |
__ASM_EMIT("sub $16, %[count]") |
| 172 |
__ASM_EMIT("jae 1b") |
| 173 |
|
| 174 |
//----------------------------------------------------------------- |
| 175 |
// x8 block |
| 176 |
__ASM_EMIT("2:") |
| 177 |
__ASM_EMIT("add $8, %[count]") |
| 178 |
__ASM_EMIT("jl 4f") |
| 179 |
__ASM_EMIT("vmovups 0x00(%[src]), %%ymm0") |
| 180 |
EFF_HSLA_HUE_CORE_X8 |
| 181 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[dst])") |
| 182 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[dst])") |
| 183 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[dst])") |
| 184 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[dst])") |
| 185 |
__ASM_EMIT("sub $8, %[count]") |
| 186 |
__ASM_EMIT("add $0x20, %[src]") |
| 187 |
__ASM_EMIT("add $0x80, %[dst]") |
| 188 |
|
| 189 |
//----------------------------------------------------------------- |
| 190 |
// x4 block |
| 191 |
__ASM_EMIT("4:") |
| 192 |
__ASM_EMIT("add $4, %[count]") |
| 193 |
__ASM_EMIT("jl 6f") |
| 194 |
__ASM_EMIT("vmovups 0x00(%[src]), %%xmm0") |
| 195 |
EFF_HSLA_HUE_CORE_X4 |
| 196 |
__ASM_EMIT("vmovups %%xmm0, 0x000(%[dst])") |
| 197 |
__ASM_EMIT("vmovups %%xmm1, 0x010(%[dst])") |
| 198 |
__ASM_EMIT("vmovups %%xmm2, 0x020(%[dst])") |
| 199 |
__ASM_EMIT("vmovups %%xmm3, 0x030(%[dst])") |
| 200 |
__ASM_EMIT("sub $4, %[count]") |
| 201 |
__ASM_EMIT("add $0x10, %[src]") |
| 202 |
__ASM_EMIT("add $0x40, %[dst]") |
| 203 |
|
| 204 |
//----------------------------------------------------------------- |
| 205 |
// x1-x3 block |
| 206 |
__ASM_EMIT("6:") |
| 207 |
__ASM_EMIT("add $4, %[count]") |
| 208 |
__ASM_EMIT("jle 14f") |
| 209 |
__ASM_EMIT("test $1, %[count]") |
| 210 |
__ASM_EMIT("jz 8f") |
| 211 |
__ASM_EMIT("vmovss 0x00(%[src]), %%xmm0") |
| 212 |
__ASM_EMIT("add $0x04, %[src]") |
| 213 |
__ASM_EMIT("8:") |
| 214 |
__ASM_EMIT("test $2, %[count]") |
| 215 |
__ASM_EMIT("jz 10f") |
| 216 |
__ASM_EMIT("vmovhps 0x00(%[src]), %%xmm0, %%xmm0") |
| 217 |
__ASM_EMIT("10:") |
| 218 |
|
| 219 |
EFF_HSLA_HUE_CORE_X4 |
| 220 |
|
| 221 |
// Store last chunk |
| 222 |
__ASM_EMIT("test $1, %[count]") |
| 223 |
__ASM_EMIT("jz 12f") |
| 224 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst])") |
| 225 |
__ASM_EMIT("add $0x10, %[dst]") |
| 226 |
__ASM_EMIT("12:") |
| 227 |
__ASM_EMIT("test $2, %[count]") |
| 228 |
__ASM_EMIT("jz 14f") |
| 229 |
__ASM_EMIT("vmovups %%xmm2, 0x00(%[dst])") |
| 230 |
__ASM_EMIT("vmovups %%xmm3, 0x10(%[dst])") |
| 231 |
|
| 232 |
__ASM_EMIT("14:") |
| 233 |
|
| 234 |
: [dst] "+r" (dst), [src] "+r" (v), [count] "+r" (count) |
| 235 |
: [eff] "r" (eff), |
| 236 |
[XC] "o" (EFF_HSLA_HUE_XC) |
| 237 |
: "cc", "memory", |
| 238 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 239 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 240 |
"%xmm8", "%xmm9", "%xmm10", "%xmm11", |
| 241 |
"%xmm12", "%xmm13", "%xmm14", "%xmm15" |
| 242 |
); |
| 243 |
} |
| 244 |
|
| 245 |
IF_ARCH_X86_64( |
| 246 |
static const float EFF_HSLA_ALPHA_XC[] __lsp_aligned32 = |
| 247 |
{ |
| 248 |
LSP_DSP_VEC8(1.0f) |
| 249 |
}; |
| 250 |
) |
| 251 |
|
| 252 |
#define EFF_HSLA_ALPHA_CORE_X16 \ |
| 253 |
/* ymm0 = v, ymm12 = h, ymm13 = s, ymm14 = l, ymm15 = 1 */ \ |
| 254 |
__ASM_EMIT("vaddps %%ymm0, %%ymm15, %%ymm2") /* ymm2 = 1+v */ \ |
| 255 |
__ASM_EMIT("vaddps %%ymm4, %%ymm15, %%ymm6") \ |
| 256 |
__ASM_EMIT("vsubps %%ymm0, %%ymm15, %%ymm3") /* ymm3 = 1-v */ \ |
| 257 |
__ASM_EMIT("vsubps %%ymm4, %%ymm15, %%ymm7") \ |
| 258 |
__ASM_EMIT("vblendvps %%ymm0, %%ymm2, %%ymm3, %%ymm3") /* ymm3 = (v<0) ? 1+v : 1-v */ \ |
| 259 |
__ASM_EMIT("vblendvps %%ymm4, %%ymm6, %%ymm7, %%ymm7") \ |
| 260 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm0") /* ymm0 = H */ \ |
| 261 |
__ASM_EMIT("vmovaps %%ymm13, %%ymm1") /* ymm1 = S */ \ |
| 262 |
__ASM_EMIT("vmovaps %%ymm14, %%ymm2") /* ymm2 = L */ \ |
| 263 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm4") \ |
| 264 |
__ASM_EMIT("vmovaps %%ymm13, %%ymm5") \ |
| 265 |
__ASM_EMIT("vmovaps %%ymm14, %%ymm6") \ |
| 266 |
MAT4_TRANSPOSE("%%ymm0", "%%ymm1", "%%ymm2", "%%ymm3", "%%ymm8", "%%ymm9") \ |
| 267 |
MAT4_TRANSPOSE("%%ymm4", "%%ymm5", "%%ymm6", "%%ymm7", "%%ymm8", "%%ymm9") \ |
| 268 |
MAT4X2_INTERLEAVE("0", "1", "2", "3", "8", "9") \ |
| 269 |
MAT4X2_INTERLEAVE("4", "5", "6", "7", "8", "9") |
| 270 |
|
| 271 |
|
| 272 |
#define EFF_HSLA_ALPHA_CORE_X8 \ |
| 273 |
/* ymm0 = v, ymm12 = h, ymm13 = s, ymm14 = l, ymm15 = 1 */ \ |
| 274 |
__ASM_EMIT("vaddps %%ymm0, %%ymm15, %%ymm2") /* ymm2 = 1+v */ \ |
| 275 |
__ASM_EMIT("vsubps %%ymm0, %%ymm15, %%ymm3") /* ymm3 = 1-v */ \ |
| 276 |
__ASM_EMIT("vblendvps %%ymm0, %%ymm2, %%ymm3, %%ymm3") /* ymm3 = (v<0) ? 1+v : 1-v */ \ |
| 277 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm0") /* ymm0 = H */ \ |
| 278 |
__ASM_EMIT("vmovaps %%ymm13, %%ymm1") /* ymm1 = S */ \ |
| 279 |
__ASM_EMIT("vmovaps %%ymm14, %%ymm2") /* ymm2 = L */ \ |
| 280 |
MAT4_TRANSPOSE("%%ymm0", "%%ymm1", "%%ymm2", "%%ymm3", "%%ymm8", "%%ymm9") \ |
| 281 |
MAT4X2_INTERLEAVE("0", "1", "2", "3", "8", "9") |
| 282 |
|
| 283 |
#define EFF_HSLA_ALPHA_CORE_X4 \ |
| 284 |
/* xmm0 = v, xmm12 = h, xmm13 = s, xmm14 = l, xmm15 = 1 */ \ |
| 285 |
__ASM_EMIT("vaddps %%xmm0, %%xmm15, %%xmm2") /* xmm2 = 1+v */ \ |
| 286 |
__ASM_EMIT("vsubps %%xmm0, %%xmm15, %%xmm3") /* xmm3 = 1-v */ \ |
| 287 |
__ASM_EMIT("vblendvps %%xmm0, %%xmm2, %%xmm3, %%xmm3") /* xmm3 = (v<0) ? 1+v : 1-v */ \ |
| 288 |
__ASM_EMIT("vmovaps %%xmm12, %%xmm0") /* xmm0 = H */ \ |
| 289 |
__ASM_EMIT("vmovaps %%xmm13, %%xmm1") /* xmm1 = S */ \ |
| 290 |
__ASM_EMIT("vmovaps %%xmm14, %%xmm2") /* xmm2 = L */ \ |
| 291 |
MAT4_TRANSPOSE("%%xmm0", "%%xmm1", "%%xmm2", "%%xmm3", "%%xmm8", "%%xmm9") |
| 292 |
|
| 293 |
/* |
| 294 |
value = v[i]; |
| 295 |
value = (value < 0) ? 1.0f + value : 1.0f - value; |
| 296 |
|
| 297 |
dst[0] = eff->h; |
| 298 |
dst[1] = eff->s; |
| 299 |
dst[2] = eff->l; |
| 300 |
dst[3] = value; // Fill alpha channel |
| 301 |
*/ |
| 302 |
void x64_eff_hsla_alpha(float *dst, const float *v, const dsp::hsla_alpha_eff_t *eff, size_t count) |
| 303 |
{ |
| 304 |
ARCH_X86_64_ASM( |
| 305 |
//----------------------------------------------------------------- |
| 306 |
// Prepare |
| 307 |
__ASM_EMIT("vbroadcastf128 0x00(%[eff]), %%ymm12") /* ymm12 = h s l a h s l a */ |
| 308 |
__ASM_EMIT("vmovaps 0x00 + %[XC], %%ymm15") /* ymm15 = 1 */ |
| 309 |
__ASM_EMIT("vshufps $0xaa, %%ymm12, %%ymm12, %%ymm14") /* ymm14 = l */ |
| 310 |
__ASM_EMIT("vshufps $0x55, %%ymm12, %%ymm12, %%ymm13") /* ymm13 = s */ |
| 311 |
__ASM_EMIT("vshufps $0x00, %%ymm12, %%ymm12, %%ymm12") /* ymm12 = h */ |
| 312 |
|
| 313 |
//----------------------------------------------------------------- |
| 314 |
// x16 blocks |
| 315 |
__ASM_EMIT("sub $16, %[count]") |
| 316 |
__ASM_EMIT("jb 2f") |
| 317 |
__ASM_EMIT("1:") |
| 318 |
__ASM_EMIT("vmovups 0x00(%[src]), %%ymm0") |
| 319 |
__ASM_EMIT("vmovups 0x20(%[src]), %%ymm4") |
| 320 |
EFF_HSLA_ALPHA_CORE_X16 |
| 321 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[dst])") |
| 322 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[dst])") |
| 323 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[dst])") |
| 324 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[dst])") |
| 325 |
__ASM_EMIT("vmovups %%ymm4, 0x080(%[dst])") |
| 326 |
__ASM_EMIT("vmovups %%ymm5, 0x0a0(%[dst])") |
| 327 |
__ASM_EMIT("vmovups %%ymm6, 0x0c0(%[dst])") |
| 328 |
__ASM_EMIT("vmovups %%ymm7, 0x0e0(%[dst])") |
| 329 |
__ASM_EMIT("add $0x40, %[src]") |
| 330 |
__ASM_EMIT("add $0x100, %[dst]") |
| 331 |
__ASM_EMIT("sub $16, %[count]") |
| 332 |
__ASM_EMIT("jae 1b") |
| 333 |
|
| 334 |
//----------------------------------------------------------------- |
| 335 |
// x8 block |
| 336 |
__ASM_EMIT("2:") |
| 337 |
__ASM_EMIT("add $8, %[count]") |
| 338 |
__ASM_EMIT("jl 4f") |
| 339 |
__ASM_EMIT("vmovups 0x00(%[src]), %%ymm0") |
| 340 |
EFF_HSLA_ALPHA_CORE_X8 |
| 341 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[dst])") |
| 342 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[dst])") |
| 343 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[dst])") |
| 344 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[dst])") |
| 345 |
__ASM_EMIT("sub $8, %[count]") |
| 346 |
__ASM_EMIT("add $0x20, %[src]") |
| 347 |
__ASM_EMIT("add $0x80, %[dst]") |
| 348 |
|
| 349 |
//----------------------------------------------------------------- |
| 350 |
// x4 block |
| 351 |
__ASM_EMIT("4:") |
| 352 |
__ASM_EMIT("add $4, %[count]") |
| 353 |
__ASM_EMIT("jl 6f") |
| 354 |
__ASM_EMIT("vmovups 0x00(%[src]), %%xmm0") |
| 355 |
EFF_HSLA_ALPHA_CORE_X4 |
| 356 |
__ASM_EMIT("vmovups %%xmm0, 0x000(%[dst])") |
| 357 |
__ASM_EMIT("vmovups %%xmm1, 0x010(%[dst])") |
| 358 |
__ASM_EMIT("vmovups %%xmm2, 0x020(%[dst])") |
| 359 |
__ASM_EMIT("vmovups %%xmm3, 0x030(%[dst])") |
| 360 |
__ASM_EMIT("sub $4, %[count]") |
| 361 |
__ASM_EMIT("add $0x10, %[src]") |
| 362 |
__ASM_EMIT("add $0x40, %[dst]") |
| 363 |
|
| 364 |
//----------------------------------------------------------------- |
| 365 |
// x1-x3 block |
| 366 |
__ASM_EMIT("6:") |
| 367 |
__ASM_EMIT("add $4, %[count]") |
| 368 |
__ASM_EMIT("jle 14f") |
| 369 |
__ASM_EMIT("test $1, %[count]") |
| 370 |
__ASM_EMIT("jz 8f") |
| 371 |
__ASM_EMIT("vmovss 0x00(%[src]), %%xmm0") |
| 372 |
__ASM_EMIT("add $0x04, %[src]") |
| 373 |
__ASM_EMIT("8:") |
| 374 |
__ASM_EMIT("test $2, %[count]") |
| 375 |
__ASM_EMIT("jz 10f") |
| 376 |
__ASM_EMIT("vmovhps 0x00(%[src]), %%xmm0, %%xmm0") |
| 377 |
__ASM_EMIT("10:") |
| 378 |
|
| 379 |
EFF_HSLA_ALPHA_CORE_X4 |
| 380 |
|
| 381 |
// Store last chunk |
| 382 |
__ASM_EMIT("test $1, %[count]") |
| 383 |
__ASM_EMIT("jz 12f") |
| 384 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst])") |
| 385 |
__ASM_EMIT("add $0x10, %[dst]") |
| 386 |
__ASM_EMIT("12:") |
| 387 |
__ASM_EMIT("test $2, %[count]") |
| 388 |
__ASM_EMIT("jz 14f") |
| 389 |
__ASM_EMIT("vmovups %%xmm2, 0x00(%[dst])") |
| 390 |
__ASM_EMIT("vmovups %%xmm3, 0x10(%[dst])") |
| 391 |
|
| 392 |
__ASM_EMIT("14:") |
| 393 |
|
| 394 |
: [dst] "+r" (dst), [src] "+r" (v), [count] "+r" (count) |
| 395 |
: [eff] "r" (eff), |
| 396 |
[XC] "o" (EFF_HSLA_ALPHA_XC) |
| 397 |
: "cc", "memory", |
| 398 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 399 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 400 |
"%xmm8", "%xmm9", "%xmm10", "%xmm11", |
| 401 |
"%xmm12", "%xmm13", "%xmm14", "%xmm15" |
| 402 |
); |
| 403 |
} |
| 404 |
|
| 405 |
IF_ARCH_X86_64( |
| 406 |
static const uint32_t EFF_HSLA_SAT_XC[] __lsp_aligned32 = |
| 407 |
{ |
| 408 |
LSP_DSP_VEC8(0x3f800000), |
| 409 |
LSP_DSP_VEC8(0x7fffffff) |
| 410 |
}; |
| 411 |
) |
| 412 |
|
| 413 |
#define EFF_HSLA_SAT_CORE_X16 \ |
| 414 |
/* ymm0 = v, ymm10 = h, ymm11 = s, ymm12 = l, ymm13 = T, ymm14 = ABS, ymm15 = KT */ \ |
| 415 |
__ASM_EMIT("vandps %%ymm14, %%ymm0, %%ymm0") /* ymm0 = V = abs(v) */ \ |
| 416 |
__ASM_EMIT("vandps %%ymm14, %%ymm4, %%ymm4") \ |
| 417 |
__ASM_EMIT("vxorps %%ymm8, %%ymm8, %%ymm8") /* ymm8 = 0 */ \ |
| 418 |
__ASM_EMIT("vsubps %%ymm0, %%ymm13, %%ymm3") /* ymm3 = T-V */ \ |
| 419 |
__ASM_EMIT("vsubps %%ymm4, %%ymm13, %%ymm7") \ |
| 420 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm0, %%ymm13, %%ymm1") /* ymm1 = ((T-V)<0) ? V : T */ \ |
| 421 |
__ASM_EMIT("vblendvps %%ymm7, %%ymm4, %%ymm13, %%ymm5") \ |
| 422 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm8, %%ymm3, %%ymm3") /* ymm3 = ((T-V)<0) ? 0 : T-V */ \ |
| 423 |
__ASM_EMIT("vblendvps %%ymm7, %%ymm8, %%ymm7, %%ymm7") \ |
| 424 |
__ASM_EMIT("vmulps %%ymm11, %%ymm1, %%ymm1") /* ymm1 = S = s * (((T-V)<0) ? V : T) */ \ |
| 425 |
__ASM_EMIT("vmulps %%ymm11, %%ymm5, %%ymm5") \ |
| 426 |
__ASM_EMIT("vmulps %%ymm15, %%ymm3, %%ymm3") /* ymm3 = A = kt * (((T-V)<0) ? 0 : T-V) */ \ |
| 427 |
__ASM_EMIT("vmulps %%ymm15, %%ymm7, %%ymm7") \ |
| 428 |
__ASM_EMIT("vmovaps %%ymm10, %%ymm0") /* ymm0 = H */ \ |
| 429 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm2") /* ymm2 = L */ \ |
| 430 |
__ASM_EMIT("vmovaps %%ymm10, %%ymm4") \ |
| 431 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm6") \ |
| 432 |
MAT4_TRANSPOSE("%%ymm0", "%%ymm1", "%%ymm2", "%%ymm3", "%%ymm8", "%%ymm9") \ |
| 433 |
MAT4_TRANSPOSE("%%ymm4", "%%ymm5", "%%ymm6", "%%ymm7", "%%ymm8", "%%ymm9") \ |
| 434 |
MAT4X2_INTERLEAVE("0", "1", "2", "3", "8", "9") \ |
| 435 |
MAT4X2_INTERLEAVE("4", "5", "6", "7", "8", "9") |
| 436 |
|
| 437 |
#define EFF_HSLA_SAT_CORE_X8 \ |
| 438 |
/* ymm0 = v, ymm10 = h, ymm11 = s, ymm12 = l, ymm13 = T, ymm14 = ABS, ymm15 = KT */ \ |
| 439 |
__ASM_EMIT("vandps %%ymm14, %%ymm0, %%ymm0") /* ymm0 = V = abs(v) */ \ |
| 440 |
__ASM_EMIT("vxorps %%ymm8, %%ymm8, %%ymm8") /* ymm8 = 0 */ \ |
| 441 |
__ASM_EMIT("vsubps %%ymm0, %%ymm13, %%ymm3") /* ymm3 = T-V */ \ |
| 442 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm0, %%ymm13, %%ymm1") /* ymm1 = ((T-V)<0) ? V : T */ \ |
| 443 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm8, %%ymm3, %%ymm3") /* ymm3 = ((T-V)<0) ? 0 : T-V */ \ |
| 444 |
__ASM_EMIT("vmulps %%ymm11, %%ymm1, %%ymm1") /* ymm1 = S = s * (((T-V)<0) ? V : T) */ \ |
| 445 |
__ASM_EMIT("vmulps %%ymm15, %%ymm3, %%ymm3") /* ymm3 = A = kt * (((T-V)<0) ? 0 : T-V) */ \ |
| 446 |
__ASM_EMIT("vmovaps %%ymm10, %%ymm0") /* ymm0 = H */ \ |
| 447 |
__ASM_EMIT("vmovaps %%ymm12, %%ymm2") /* ymm2 = L */ \ |
| 448 |
MAT4_TRANSPOSE("%%ymm0", "%%ymm1", "%%ymm2", "%%ymm3", "%%ymm8", "%%ymm9") \ |
| 449 |
MAT4X2_INTERLEAVE("0", "1", "2", "3", "8", "9") |
| 450 |
|
| 451 |
#define EFF_HSLA_SAT_CORE_X4 \ |
| 452 |
/* xmm0 = v, xmm10 = h, xmm11 = s, xmm12 = l, xmm13 = T, xmm14 = ABS, xmm15 = KT */ \ |
| 453 |
__ASM_EMIT("vandps %%xmm14, %%xmm0, %%xmm0") /* xmm0 = V = abs(v) */ \ |
| 454 |
__ASM_EMIT("vxorps %%xmm8, %%xmm8, %%xmm8") /* xmm8 = 0 */ \ |
| 455 |
__ASM_EMIT("vsubps %%xmm0, %%xmm13, %%xmm3") /* xmm3 = T-V */ \ |
| 456 |
__ASM_EMIT("vblendvps %%xmm3, %%xmm0, %%xmm13, %%xmm1") /* xmm1 = ((T-V)<0) ? V : T */ \ |
| 457 |
__ASM_EMIT("vblendvps %%xmm3, %%xmm8, %%xmm3, %%xmm3") /* xmm3 = ((T-V)<0) ? 0 : T-V */ \ |
| 458 |
__ASM_EMIT("vmulps %%xmm11, %%xmm1, %%xmm1") /* xmm1 = S = s * (((T-V)<0) ? V : T) */ \ |
| 459 |
__ASM_EMIT("vmulps %%xmm15, %%xmm3, %%xmm3") /* xmm3 = A = kt * (((T-V)<0) ? 0 : T-V) */ \ |
| 460 |
__ASM_EMIT("vmovaps %%xmm10, %%xmm0") /* xmm0 = H */ \ |
| 461 |
__ASM_EMIT("vmovaps %%xmm12, %%xmm2") /* xmm2 = L */ \ |
| 462 |
MAT4_TRANSPOSE("%%xmm0", "%%xmm1", "%%xmm2", "%%xmm3", "%%xmm8", "%%xmm9") |
| 463 |
|
| 464 |
/* |
| 465 |
kt = 1.0f / eff->thresh; |
| 466 |
value = (value >= 0.0f) ? value : -value; |
| 467 |
|
| 468 |
if ((eff->thresh - value) < 0) |
| 469 |
{ |
| 470 |
dst[1] = eff->s * value; |
| 471 |
dst[3] = 0.0f; |
| 472 |
} |
| 473 |
else |
| 474 |
{ |
| 475 |
dst[1] = eff->s * eff->thresh; |
| 476 |
dst[3] = (eff->thresh - value) * kt; |
| 477 |
} |
| 478 |
|
| 479 |
dst[0] = eff->h; |
| 480 |
dst[2] = eff->l; |
| 481 |
*/ |
| 482 |
|
| 483 |
void x64_eff_hsla_sat(float *dst, const float *v, const dsp::hsla_sat_eff_t *eff, size_t count) |
| 484 |
{ |
| 485 |
ARCH_X86_64_ASM( |
| 486 |
//----------------------------------------------------------------- |
| 487 |
// Prepare |
| 488 |
__ASM_EMIT("vbroadcastf128 0x00(%[eff]), %%ymm10") /* ymm12 = h s l a h s l a */ |
| 489 |
__ASM_EMIT("vmovaps 0x00 + %[XC], %%ymm15") /* ymm15 = 1 */ |
| 490 |
__ASM_EMIT("vshufps $0xaa, %%ymm10, %%ymm10, %%ymm12") /* ymm12 = l */ |
| 491 |
__ASM_EMIT("vmovaps 0x20 + %[XC], %%ymm14") /* ymm14 = ABS */ |
| 492 |
__ASM_EMIT("vshufps $0x55, %%ymm10, %%ymm10, %%ymm11") /* ymm11 = s */ |
| 493 |
__ASM_EMIT("vbroadcastss 0x10(%[eff]), %%ymm13") /* ymm13 = T */ |
| 494 |
__ASM_EMIT("vshufps $0x00, %%ymm10, %%ymm10, %%ymm10") /* ymm10 = h */ |
| 495 |
__ASM_EMIT("vdivps %%ymm13, %%ymm15, %%ymm15") /* ymm15 = KT = 1/T */ |
| 496 |
|
| 497 |
//----------------------------------------------------------------- |
| 498 |
// x16 blocks |
| 499 |
__ASM_EMIT("sub $16, %[count]") |
| 500 |
__ASM_EMIT("jb 2f") |
| 501 |
__ASM_EMIT("1:") |
| 502 |
__ASM_EMIT("vmovups 0x00(%[src]), %%ymm0") |
| 503 |
__ASM_EMIT("vmovups 0x20(%[src]), %%ymm4") |
| 504 |
EFF_HSLA_SAT_CORE_X16 |
| 505 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[dst])") |
| 506 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[dst])") |
| 507 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[dst])") |
| 508 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[dst])") |
| 509 |
__ASM_EMIT("vmovups %%ymm4, 0x080(%[dst])") |
| 510 |
__ASM_EMIT("vmovups %%ymm5, 0x0a0(%[dst])") |
| 511 |
__ASM_EMIT("vmovups %%ymm6, 0x0c0(%[dst])") |
| 512 |
__ASM_EMIT("vmovups %%ymm7, 0x0e0(%[dst])") |
| 513 |
__ASM_EMIT("add $0x40, %[src]") |
| 514 |
__ASM_EMIT("add $0x100, %[dst]") |
| 515 |
__ASM_EMIT("sub $16, %[count]") |
| 516 |
__ASM_EMIT("jae 1b") |
| 517 |
|
| 518 |
//----------------------------------------------------------------- |
| 519 |
// x8 block |
| 520 |
__ASM_EMIT("2:") |
| 521 |
__ASM_EMIT("add $8, %[count]") |
| 522 |
__ASM_EMIT("jl 4f") |
| 523 |
__ASM_EMIT("vmovups 0x00(%[src]), %%ymm0") |
| 524 |
EFF_HSLA_SAT_CORE_X8 |
| 525 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[dst])") |
| 526 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[dst])") |
| 527 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[dst])") |
| 528 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[dst])") |
| 529 |
__ASM_EMIT("sub $8, %[count]") |
| 530 |
__ASM_EMIT("add $0x20, %[src]") |
| 531 |
__ASM_EMIT("add $0x80, %[dst]") |
| 532 |
|
| 533 |
//----------------------------------------------------------------- |
| 534 |
// x4 block |
| 535 |
__ASM_EMIT("4:") |
| 536 |
__ASM_EMIT("add $4, %[count]") |
| 537 |
__ASM_EMIT("jl 6f") |
| 538 |
__ASM_EMIT("vmovups 0x00(%[src]), %%xmm0") |
| 539 |
EFF_HSLA_SAT_CORE_X4 |
| 540 |
__ASM_EMIT("vmovups %%xmm0, 0x000(%[dst])") |
| 541 |
__ASM_EMIT("vmovups %%xmm1, 0x010(%[dst])") |
| 542 |
__ASM_EMIT("vmovups %%xmm2, 0x020(%[dst])") |
| 543 |
__ASM_EMIT("vmovups %%xmm3, 0x030(%[dst])") |
| 544 |
__ASM_EMIT("sub $4, %[count]") |
| 545 |
__ASM_EMIT("add $0x10, %[src]") |
| 546 |
__ASM_EMIT("add $0x40, %[dst]") |
| 547 |
|
| 548 |
//----------------------------------------------------------------- |
| 549 |
// x1-x3 block |
| 550 |
__ASM_EMIT("6:") |
| 551 |
__ASM_EMIT("add $4, %[count]") |
| 552 |
__ASM_EMIT("jle 14f") |
| 553 |
__ASM_EMIT("test $1, %[count]") |
| 554 |
__ASM_EMIT("jz 8f") |
| 555 |
__ASM_EMIT("vmovss 0x00(%[src]), %%xmm0") |
| 556 |
__ASM_EMIT("add $0x04, %[src]") |
| 557 |
__ASM_EMIT("8:") |
| 558 |
__ASM_EMIT("test $2, %[count]") |
| 559 |
__ASM_EMIT("jz 10f") |
| 560 |
__ASM_EMIT("vmovhps 0x00(%[src]), %%xmm0, %%xmm0") |
| 561 |
__ASM_EMIT("10:") |
| 562 |
|
| 563 |
EFF_HSLA_SAT_CORE_X4 |
| 564 |
|
| 565 |
// Store last chunk |
| 566 |
__ASM_EMIT("test $1, %[count]") |
| 567 |
__ASM_EMIT("jz 12f") |
| 568 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst])") |
| 569 |
__ASM_EMIT("add $0x10, %[dst]") |
| 570 |
__ASM_EMIT("12:") |
| 571 |
__ASM_EMIT("test $2, %[count]") |
| 572 |
__ASM_EMIT("jz 14f") |
| 573 |
__ASM_EMIT("vmovups %%xmm2, 0x00(%[dst])") |
| 574 |
__ASM_EMIT("vmovups %%xmm3, 0x10(%[dst])") |
| 575 |
|
| 576 |
__ASM_EMIT("14:") |
| 577 |
|
| 578 |
: [dst] "+r" (dst), [src] "+r" (v), [count] "+r" (count) |
| 579 |
: [eff] "r" (eff), |
| 580 |
[XC] "o" (EFF_HSLA_SAT_XC) |
| 581 |
: "cc", "memory", |
| 582 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 583 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 584 |
"%xmm8", "%xmm9", "%xmm10", "%xmm11", |
| 585 |
"%xmm12", "%xmm13", "%xmm14", "%xmm15" |
| 586 |
); |
| 587 |
} |
| 588 |
|
| 589 |
IF_ARCH_X86_64( |
| 590 |
static const uint32_t EFF_HSLA_LIGHT_XC[] __lsp_aligned32 = |
| 591 |
{ |
| 592 |
LSP_DSP_VEC8(0x3f800000), |
| 593 |
LSP_DSP_VEC8(0x7fffffff) |
| 594 |
}; |
| 595 |
) |
| 596 |
|
| 597 |
#define EFF_HSLA_LIGHT_CORE_X16 \ |
| 598 |
/* ymm0 = v, ymm10 = h, ymm11 = s, ymm12 = l, ymm13 = T, ymm14 = ABS, ymm15 = KT */ \ |
| 599 |
__ASM_EMIT("vandps %%ymm14, %%ymm0, %%ymm0") /* ymm0 = V = abs(v) */ \ |
| 600 |
__ASM_EMIT("vandps %%ymm14, %%ymm4, %%ymm4") \ |
| 601 |
__ASM_EMIT("vxorps %%ymm8, %%ymm8, %%ymm8") /* ymm8 = 0 */ \ |
| 602 |
__ASM_EMIT("vsubps %%ymm0, %%ymm13, %%ymm3") /* ymm3 = T-V */ \ |
| 603 |
__ASM_EMIT("vsubps %%ymm4, %%ymm13, %%ymm7") \ |
| 604 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm0, %%ymm13, %%ymm2") /* ymm2 = ((T-V)<0) ? V : T */ \ |
| 605 |
__ASM_EMIT("vblendvps %%ymm7, %%ymm4, %%ymm13, %%ymm6") \ |
| 606 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm8, %%ymm3, %%ymm3") /* ymm3 = ((T-V)<0) ? 0 : T-V */ \ |
| 607 |
__ASM_EMIT("vblendvps %%ymm7, %%ymm8, %%ymm7, %%ymm7") \ |
| 608 |
__ASM_EMIT("vmulps %%ymm12, %%ymm2, %%ymm2") /* ymm2 = L = l * (((T-V)<0) ? V : T) */ \ |
| 609 |
__ASM_EMIT("vmulps %%ymm12, %%ymm6, %%ymm6") \ |
| 610 |
__ASM_EMIT("vmulps %%ymm15, %%ymm3, %%ymm3") /* ymm3 = A = kt * (((T-V)<0) ? 0 : T-V) */ \ |
| 611 |
__ASM_EMIT("vmulps %%ymm15, %%ymm7, %%ymm7") \ |
| 612 |
__ASM_EMIT("vmovaps %%ymm10, %%ymm0") /* ymm0 = H */ \ |
| 613 |
__ASM_EMIT("vmovaps %%ymm10, %%ymm4") \ |
| 614 |
__ASM_EMIT("vmovaps %%ymm11, %%ymm1") /* ymm1 = S */ \ |
| 615 |
__ASM_EMIT("vmovaps %%ymm11, %%ymm5") \ |
| 616 |
MAT4_TRANSPOSE("%%ymm0", "%%ymm1", "%%ymm2", "%%ymm3", "%%ymm8", "%%ymm9") \ |
| 617 |
MAT4_TRANSPOSE("%%ymm4", "%%ymm5", "%%ymm6", "%%ymm7", "%%ymm8", "%%ymm9") \ |
| 618 |
MAT4X2_INTERLEAVE("0", "1", "2", "3", "8", "9") \ |
| 619 |
MAT4X2_INTERLEAVE("4", "5", "6", "7", "8", "9") |
| 620 |
|
| 621 |
#define EFF_HSLA_LIGHT_CORE_X8 \ |
| 622 |
/* ymm0 = v, ymm10 = h, ymm11 = s, ymm12 = l, ymm13 = T, ymm14 = ABS, ymm15 = KT */ \ |
| 623 |
__ASM_EMIT("vandps %%ymm14, %%ymm0, %%ymm0") /* ymm0 = V = abs(v) */ \ |
| 624 |
__ASM_EMIT("vxorps %%ymm8, %%ymm8, %%ymm8") /* ymm8 = 0 */ \ |
| 625 |
__ASM_EMIT("vsubps %%ymm0, %%ymm13, %%ymm3") /* ymm3 = T-V */ \ |
| 626 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm0, %%ymm13, %%ymm2") /* ymm2 = ((T-V)<0) ? V : T */ \ |
| 627 |
__ASM_EMIT("vblendvps %%ymm3, %%ymm8, %%ymm3, %%ymm3") /* ymm3 = ((T-V)<0) ? 0 : T-V */ \ |
| 628 |
__ASM_EMIT("vmulps %%ymm12, %%ymm2, %%ymm2") /* ymm2 = L = l * (((T-V)<0) ? V : T) */ \ |
| 629 |
__ASM_EMIT("vmulps %%ymm15, %%ymm3, %%ymm3") /* ymm3 = A = kt * (((T-V)<0) ? 0 : T-V) */ \ |
| 630 |
__ASM_EMIT("vmovaps %%ymm10, %%ymm0") /* ymm0 = H */ \ |
| 631 |
__ASM_EMIT("vmovaps %%ymm11, %%ymm1") /* ymm1 = S */ \ |
| 632 |
MAT4_TRANSPOSE("%%ymm0", "%%ymm1", "%%ymm2", "%%ymm3", "%%ymm8", "%%ymm9") \ |
| 633 |
MAT4X2_INTERLEAVE("0", "1", "2", "3", "8", "9") |
| 634 |
|
| 635 |
#define EFF_HSLA_LIGHT_CORE_X4 \ |
| 636 |
/* xmm0 = v, xmm10 = h, xmm11 = s, xmm12 = l, xmm13 = T, xmm14 = ABS, xmm15 = KT */ \ |
| 637 |
__ASM_EMIT("vandps %%xmm14, %%xmm0, %%xmm0") /* xmm0 = V = abs(v) */ \ |
| 638 |
__ASM_EMIT("vxorps %%xmm8, %%xmm8, %%xmm8") /* xmm8 = 0 */ \ |
| 639 |
__ASM_EMIT("vsubps %%xmm0, %%xmm13, %%xmm3") /* xmm3 = T-V */ \ |
| 640 |
__ASM_EMIT("vblendvps %%xmm3, %%xmm0, %%xmm13, %%xmm2") /* xmm2 = ((T-V)<0) ? V : T */ \ |
| 641 |
__ASM_EMIT("vblendvps %%xmm3, %%xmm8, %%xmm3, %%xmm3") /* xmm3 = ((T-V)<0) ? 0 : T-V */ \ |
| 642 |
__ASM_EMIT("vmulps %%xmm12, %%xmm2, %%xmm2") /* xmm2 = L = l * (((T-V)<0) ? V : T) */ \ |
| 643 |
__ASM_EMIT("vmulps %%xmm15, %%xmm3, %%xmm3") /* xmm3 = A = kt * (((T-V)<0) ? 0 : T-V) */ \ |
| 644 |
__ASM_EMIT("vmovaps %%xmm10, %%xmm0") /* xmm0 = H */ \ |
| 645 |
__ASM_EMIT("vmovaps %%xmm11, %%xmm1") /* xmm1 = S */ \ |
| 646 |
MAT4_TRANSPOSE("%%xmm0", "%%xmm1", "%%xmm2", "%%xmm3", "%%xmm8", "%%xmm9") |
| 647 |
|
| 648 |
/* |
| 649 |
value = (value >= 0.0f) ? value : -value; |
| 650 |
|
| 651 |
if ((eff->thresh - value) < 0) |
| 652 |
{ |
| 653 |
dst[2] = eff->l * value; |
| 654 |
dst[3] = 0.0f; |
| 655 |
} |
| 656 |
else |
| 657 |
{ |
| 658 |
dst[2] = eff->l * eff->thresh; |
| 659 |
dst[3] = (eff->thresh - value) * kt; |
| 660 |
} |
| 661 |
|
| 662 |
dst[0] = eff->h; |
| 663 |
dst[1] = eff->s; |
| 664 |
*/ |
| 665 |
|
| 666 |
void x64_eff_hsla_light(float *dst, const float *v, const dsp::hsla_light_eff_t *eff, size_t count) |
| 667 |
{ |
| 668 |
ARCH_X86_64_ASM( |
| 669 |
//----------------------------------------------------------------- |
| 670 |
// Prepare |
| 671 |
__ASM_EMIT("vbroadcastf128 0x00(%[eff]), %%ymm10") /* ymm12 = h s l a h s l a */ |
| 672 |
__ASM_EMIT("vmovaps 0x00 + %[XC], %%ymm15") /* ymm15 = 1 */ |
| 673 |
__ASM_EMIT("vshufps $0xaa, %%ymm10, %%ymm10, %%ymm12") /* ymm12 = l */ |
| 674 |
__ASM_EMIT("vmovaps 0x20 + %[XC], %%ymm14") /* ymm14 = ABS */ |
| 675 |
__ASM_EMIT("vshufps $0x55, %%ymm10, %%ymm10, %%ymm11") /* ymm11 = s */ |
| 676 |
__ASM_EMIT("vbroadcastss 0x10(%[eff]), %%ymm13") /* ymm13 = T */ |
| 677 |
__ASM_EMIT("vshufps $0x00, %%ymm10, %%ymm10, %%ymm10") /* ymm10 = h */ |
| 678 |
__ASM_EMIT("vdivps %%ymm13, %%ymm15, %%ymm15") /* ymm15 = KT = 1/T */ |
| 679 |
|
| 680 |
//----------------------------------------------------------------- |
| 681 |
// x16 blocks |
| 682 |
__ASM_EMIT("sub $16, %[count]") |
| 683 |
__ASM_EMIT("jb 2f") |
| 684 |
__ASM_EMIT("1:") |
| 685 |
__ASM_EMIT("vmovups 0x00(%[src]), %%ymm0") |
| 686 |
__ASM_EMIT("vmovups 0x20(%[src]), %%ymm4") |
| 687 |
EFF_HSLA_LIGHT_CORE_X16 |
| 688 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[dst])") |
| 689 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[dst])") |
| 690 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[dst])") |
| 691 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[dst])") |
| 692 |
__ASM_EMIT("vmovups %%ymm4, 0x080(%[dst])") |
| 693 |
__ASM_EMIT("vmovups %%ymm5, 0x0a0(%[dst])") |
| 694 |
__ASM_EMIT("vmovups %%ymm6, 0x0c0(%[dst])") |
| 695 |
__ASM_EMIT("vmovups %%ymm7, 0x0e0(%[dst])") |
| 696 |
__ASM_EMIT("add $0x40, %[src]") |
| 697 |
__ASM_EMIT("add $0x100, %[dst]") |
| 698 |
__ASM_EMIT("sub $16, %[count]") |
| 699 |
__ASM_EMIT("jae 1b") |
| 700 |
|
| 701 |
//----------------------------------------------------------------- |
| 702 |
// x8 block |
| 703 |
__ASM_EMIT("2:") |
| 704 |
__ASM_EMIT("add $8, %[count]") |
| 705 |
__ASM_EMIT("jl 4f") |
| 706 |
__ASM_EMIT("vmovups 0x00(%[src]), %%ymm0") |
| 707 |
EFF_HSLA_LIGHT_CORE_X8 |
| 708 |
__ASM_EMIT("vmovups %%ymm0, 0x000(%[dst])") |
| 709 |
__ASM_EMIT("vmovups %%ymm1, 0x020(%[dst])") |
| 710 |
__ASM_EMIT("vmovups %%ymm2, 0x040(%[dst])") |
| 711 |
__ASM_EMIT("vmovups %%ymm3, 0x060(%[dst])") |
| 712 |
__ASM_EMIT("sub $8, %[count]") |
| 713 |
__ASM_EMIT("add $0x20, %[src]") |
| 714 |
__ASM_EMIT("add $0x80, %[dst]") |
| 715 |
|
| 716 |
//----------------------------------------------------------------- |
| 717 |
// x4 block |
| 718 |
__ASM_EMIT("4:") |
| 719 |
__ASM_EMIT("add $4, %[count]") |
| 720 |
__ASM_EMIT("jl 6f") |
| 721 |
__ASM_EMIT("vmovups 0x00(%[src]), %%xmm0") |
| 722 |
EFF_HSLA_LIGHT_CORE_X4 |
| 723 |
__ASM_EMIT("vmovups %%xmm0, 0x000(%[dst])") |
| 724 |
__ASM_EMIT("vmovups %%xmm1, 0x010(%[dst])") |
| 725 |
__ASM_EMIT("vmovups %%xmm2, 0x020(%[dst])") |
| 726 |
__ASM_EMIT("vmovups %%xmm3, 0x030(%[dst])") |
| 727 |
__ASM_EMIT("sub $4, %[count]") |
| 728 |
__ASM_EMIT("add $0x10, %[src]") |
| 729 |
__ASM_EMIT("add $0x40, %[dst]") |
| 730 |
|
| 731 |
//----------------------------------------------------------------- |
| 732 |
// x1-x3 block |
| 733 |
__ASM_EMIT("6:") |
| 734 |
__ASM_EMIT("add $4, %[count]") |
| 735 |
__ASM_EMIT("jle 14f") |
| 736 |
__ASM_EMIT("test $1, %[count]") |
| 737 |
__ASM_EMIT("jz 8f") |
| 738 |
__ASM_EMIT("vmovss 0x00(%[src]), %%xmm0") |
| 739 |
__ASM_EMIT("add $0x04, %[src]") |
| 740 |
__ASM_EMIT("8:") |
| 741 |
__ASM_EMIT("test $2, %[count]") |
| 742 |
__ASM_EMIT("jz 10f") |
| 743 |
__ASM_EMIT("vmovhps 0x00(%[src]), %%xmm0, %%xmm0") |
| 744 |
__ASM_EMIT("10:") |
| 745 |
|
| 746 |
EFF_HSLA_LIGHT_CORE_X4 |
| 747 |
|
| 748 |
// Store last chunk |
| 749 |
__ASM_EMIT("test $1, %[count]") |
| 750 |
__ASM_EMIT("jz 12f") |
| 751 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst])") |
| 752 |
__ASM_EMIT("add $0x10, %[dst]") |
| 753 |
__ASM_EMIT("12:") |
| 754 |
__ASM_EMIT("test $2, %[count]") |
| 755 |
__ASM_EMIT("jz 14f") |
| 756 |
__ASM_EMIT("vmovups %%xmm2, 0x00(%[dst])") |
| 757 |
__ASM_EMIT("vmovups %%xmm3, 0x10(%[dst])") |
| 758 |
|
| 759 |
__ASM_EMIT("14:") |
| 760 |
|
| 761 |
: [dst] "+r" (dst), [src] "+r" (v), [count] "+r" (count) |
| 762 |
: [eff] "r" (eff), |
| 763 |
[XC] "o" (EFF_HSLA_LIGHT_XC) |
| 764 |
: "cc", "memory", |
| 765 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 766 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 767 |
"%xmm8", "%xmm9", "%xmm10", "%xmm11", |
| 768 |
"%xmm12", "%xmm13", "%xmm14", "%xmm15" |
| 769 |
); |
| 770 |
} |
| 771 |
} |
| 772 |
} |
| 773 |
|
| 774 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX2_GRAPHICS_EFFECTS_H_ */ |