| 1 |
/* |
| 2 |
* Copyright (C) 2023 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2023 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 5 окт. 2023 г. |
| 7 |
* |
| 8 |
* lsp-dsp-lib is free software: you can redistribute it and/or modify |
| 9 |
* it under the terms of the GNU Lesser General Public License as published by |
| 10 |
* the Free Software Foundation, either version 3 of the License, or |
| 11 |
* any later version. |
| 12 |
* |
| 13 |
* lsp-dsp-lib is distributed in the hope that it will be useful, |
| 14 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 |
* GNU Lesser General Public License for more details. |
| 17 |
* |
| 18 |
* You should have received a copy of the GNU Lesser General Public License |
| 19 |
* along with lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>. |
| 20 |
*/ |
| 21 |
|
| 22 |
#ifndef PRIVATE_DSP_ARCH_X86_SSE2_DYNAMICS_COMPRESSOR_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_SSE2_DYNAMICS_COMPRESSOR_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 |
#include <private/dsp/arch/x86/sse2/pmath/exp.h> |
| 30 |
#include <private/dsp/arch/x86/sse2/pmath/log.h> |
| 31 |
|
| 32 |
namespace lsp |
| 33 |
{ |
| 34 |
namespace sse2 |
| 35 |
{ |
| 36 |
#pragma pack(push, 1) |
| 37 |
typedef struct comp_knee_t |
| 38 |
{ |
| 39 |
float start[4]; // +0x00 |
| 40 |
float end[4]; // +0x10 |
| 41 |
float gain[4]; // +0x20 |
| 42 |
float herm[12]; // +0x30 |
| 43 |
float tilt[8]; // +0x60 |
| 44 |
} comp_knee_t; |
| 45 |
#pragma pack(pop) |
| 46 |
|
| 47 |
#define UNPACK_COMP_KNEE(DST, DOFF, SRC, SOFF) \ |
| 48 |
__ASM_EMIT("movss " SOFF " + 0x00(%[" SRC "]), %%xmm0") \ |
| 49 |
__ASM_EMIT("movss " SOFF " + 0x04(%[" SRC "]), %%xmm1") \ |
| 50 |
__ASM_EMIT("movss " SOFF " + 0x08(%[" SRC "]), %%xmm2") \ |
| 51 |
__ASM_EMIT("movss " SOFF " + 0x0c(%[" SRC "]), %%xmm3") \ |
| 52 |
__ASM_EMIT("movss " SOFF " + 0x10(%[" SRC "]), %%xmm4") \ |
| 53 |
__ASM_EMIT("movss " SOFF " + 0x14(%[" SRC "]), %%xmm5") \ |
| 54 |
__ASM_EMIT("movss " SOFF " + 0x18(%[" SRC "]), %%xmm6") \ |
| 55 |
__ASM_EMIT("movss " SOFF " + 0x1c(%[" SRC "]), %%xmm7") \ |
| 56 |
__ASM_EMIT("shufps $0x00, %%xmm0, %%xmm0") \ |
| 57 |
__ASM_EMIT("shufps $0x00, %%xmm1, %%xmm1") \ |
| 58 |
__ASM_EMIT("shufps $0x00, %%xmm2, %%xmm2") \ |
| 59 |
__ASM_EMIT("shufps $0x00, %%xmm3, %%xmm3") \ |
| 60 |
__ASM_EMIT("shufps $0x00, %%xmm4, %%xmm4") \ |
| 61 |
__ASM_EMIT("shufps $0x00, %%xmm5, %%xmm5") \ |
| 62 |
__ASM_EMIT("shufps $0x00, %%xmm6, %%xmm6") \ |
| 63 |
__ASM_EMIT("shufps $0x00, %%xmm7, %%xmm7") \ |
| 64 |
__ASM_EMIT("movaps %%xmm0, " DOFF "+ 0x00 + %[" DST "]") \ |
| 65 |
__ASM_EMIT("movaps %%xmm1, " DOFF " + 0x10 + %[" DST "]") \ |
| 66 |
__ASM_EMIT("movaps %%xmm2, " DOFF " + 0x20 + %[" DST "]") \ |
| 67 |
__ASM_EMIT("movaps %%xmm3, " DOFF " + 0x30 + %[" DST "]") \ |
| 68 |
__ASM_EMIT("movaps %%xmm4, " DOFF " + 0x40 + %[" DST "]") \ |
| 69 |
__ASM_EMIT("movaps %%xmm5, " DOFF " + 0x50 + %[" DST "]") \ |
| 70 |
__ASM_EMIT("movaps %%xmm6, " DOFF " + 0x60 + %[" DST "]") \ |
| 71 |
__ASM_EMIT("movaps %%xmm7, " DOFF " + 0x70 + %[" DST "]") |
| 72 |
|
| 73 |
#define PROCESS_KNEE_SINGLE_X8(OFF) \ |
| 74 |
/* in: xmm0 = lx0, xmm4 = lx1 */ \ |
| 75 |
__ASM_EMIT("movaps " OFF " + 0x30 + %[knee], %%xmm1") /* xmm1 = herm[0] */ \ |
| 76 |
__ASM_EMIT("movaps " OFF " + 0x30 + %[knee], %%xmm5") \ |
| 77 |
__ASM_EMIT("mulps %%xmm0, %%xmm1") /* xmm1 = herm[0]*lx0 */ \ |
| 78 |
__ASM_EMIT("mulps %%xmm4, %%xmm5") \ |
| 79 |
__ASM_EMIT("addps " OFF " + 0x40 + %[knee], %%xmm1") /* xmm1 = herm[0]*lx0+herm[1] */ \ |
| 80 |
__ASM_EMIT("addps " OFF " + 0x40 + %[knee], %%xmm5") \ |
| 81 |
__ASM_EMIT("mulps %%xmm0, %%xmm1") /* xmm1 = (herm[0]*lx0+herm[1])*lx0 */ \ |
| 82 |
__ASM_EMIT("mulps %%xmm4, %%xmm5") \ |
| 83 |
__ASM_EMIT("addps " OFF " + 0x50 + %[knee], %%xmm1") /* xmm1 = KV = (herm[0]*lx0+herm[1])*lx0+herm[2] */ \ |
| 84 |
__ASM_EMIT("addps " OFF " + 0x50 + %[knee], %%xmm5") \ |
| 85 |
__ASM_EMIT("movaps " OFF " + 0x60 + %[knee], %%xmm2") /* xmm2 = tilt[0] */ \ |
| 86 |
__ASM_EMIT("movaps " OFF " + 0x60 + %[knee], %%xmm6") \ |
| 87 |
__ASM_EMIT("mulps %%xmm0, %%xmm2") /* xmm2 = tilt[0]*lx0 */ \ |
| 88 |
__ASM_EMIT("mulps %%xmm4, %%xmm6") \ |
| 89 |
__ASM_EMIT("addps " OFF " + 0x70 + %[knee], %%xmm2") /* xmm2 = TV = tilt[0]*lx0+tilt[1] */ \ |
| 90 |
__ASM_EMIT("addps " OFF " + 0x70 + %[knee], %%xmm6") \ |
| 91 |
__ASM_EMIT("movaps 0x00 + %[mem], %%xmm0") /* xmm0 = x0 */ \ |
| 92 |
__ASM_EMIT("movaps 0x10 + %[mem], %%xmm4") \ |
| 93 |
__ASM_EMIT("cmpps $5, " OFF " + 0x10 + %[knee], %%xmm0") /* xmm0 = [x0 >= end] */ \ |
| 94 |
__ASM_EMIT("cmpps $5, " OFF " + 0x10 + %[knee], %%xmm4") \ |
| 95 |
__ASM_EMIT("andps %%xmm0, %%xmm2") /* xmm2 = [x0 >= end] & TV */ \ |
| 96 |
__ASM_EMIT("andps %%xmm4, %%xmm6") \ |
| 97 |
__ASM_EMIT("andnps %%xmm1, %%xmm0") /* xmm0 = [x0 < end] & KV */ \ |
| 98 |
__ASM_EMIT("andnps %%xmm5, %%xmm4") \ |
| 99 |
__ASM_EMIT("orps %%xmm2, %%xmm0") /* xmm0 = [x0 >= end] ? TV : KV */ \ |
| 100 |
__ASM_EMIT("orps %%xmm6, %%xmm4") \ |
| 101 |
EXP_CORE_X8 /* xmm0 = EV = expf([x0 >= end] ? TV : KV) */ \ |
| 102 |
__ASM_EMIT("movaps 0x00 + %[mem], %%xmm2") /* xmm2 = x0 */ \ |
| 103 |
__ASM_EMIT("movaps 0x10 + %[mem], %%xmm6") \ |
| 104 |
__ASM_EMIT("cmpps $6, " OFF " + 0x00 + %[knee], %%xmm2") /* xmm2 = [x0 > start] */ \ |
| 105 |
__ASM_EMIT("cmpps $6, " OFF " + 0x00 + %[knee], %%xmm6") \ |
| 106 |
__ASM_EMIT("andps %%xmm2, %%xmm0") /* xmm4 = [x0 > start] & EV */ \ |
| 107 |
__ASM_EMIT("andps %%xmm6, %%xmm4") \ |
| 108 |
__ASM_EMIT("andnps " OFF " + 0x20 + %[knee], %%xmm2") /* xmm2 = [x0 <= start] & GV */ \ |
| 109 |
__ASM_EMIT("andnps " OFF " + 0x20 + %[knee], %%xmm6") \ |
| 110 |
__ASM_EMIT("orps %%xmm2, %%xmm0") /* xmm0 = ([x0 <= start]) ? GV : EV */ \ |
| 111 |
__ASM_EMIT("orps %%xmm6, %%xmm4") \ |
| 112 |
/* out: xmm0 = g0, xmm4 = g1 */ |
| 113 |
|
| 114 |
#define PROCESS_KNEE_SINGLE_X4(OFF) \ |
| 115 |
/* in: xmm0 = lx0 */ \ |
| 116 |
__ASM_EMIT("movaps " OFF " + 0x30 + %[knee], %%xmm1") /* xmm1 = herm[0] */ \ |
| 117 |
__ASM_EMIT("mulps %%xmm0, %%xmm1") /* xmm1 = herm[0]*lx0 */ \ |
| 118 |
__ASM_EMIT("addps " OFF " + 0x40 + %[knee], %%xmm1") /* xmm1 = herm[0]*lx0+herm[1] */ \ |
| 119 |
__ASM_EMIT("mulps %%xmm0, %%xmm1") /* xmm1 = (herm[0]*lx0+herm[1])*lx0 */ \ |
| 120 |
__ASM_EMIT("addps " OFF " + 0x50 + %[knee], %%xmm1") /* xmm1 = KV = (herm[0]*lx0+herm[1])*lx0+herm[2] */ \ |
| 121 |
__ASM_EMIT("movaps " OFF " + 0x60 + %[knee], %%xmm2") /* xmm2 = tilt[0] */ \ |
| 122 |
__ASM_EMIT("mulps %%xmm0, %%xmm2") /* xmm2 = tilt[0]*lx0 */ \ |
| 123 |
__ASM_EMIT("addps " OFF " + 0x70 + %[knee], %%xmm2") /* xmm2 = TV = tilt[0]*lx0+tilt[1] */ \ |
| 124 |
__ASM_EMIT("movaps 0x00 + %[mem], %%xmm0") /* xmm0 = x0 */ \ |
| 125 |
__ASM_EMIT("cmpps $5, " OFF " + 0x10 + %[knee], %%xmm0") /* xmm0 = [x0 >= end] */ \ |
| 126 |
__ASM_EMIT("andps %%xmm0, %%xmm2") /* xmm2 = [x0 >= end] & TV */ \ |
| 127 |
__ASM_EMIT("andnps %%xmm1, %%xmm0") /* xmm0 = [x0 < end] & KV */ \ |
| 128 |
__ASM_EMIT("orps %%xmm2, %%xmm0") /* xmm0 = [x0 >= end] ? TV : KV */ \ |
| 129 |
EXP_CORE_X4 /* xmm0 = EV = expf([x0 >= end] ? TV : KV) */ \ |
| 130 |
__ASM_EMIT("movaps 0x00 + %[mem], %%xmm2") /* xmm2 = x0 */ \ |
| 131 |
__ASM_EMIT("cmpps $6, " OFF " + 0x00 + %[knee], %%xmm2") /* xmm2 = [x0 > start] */ \ |
| 132 |
__ASM_EMIT("andps %%xmm2, %%xmm0") /* xmm4 = [x0 > start] & EV */ \ |
| 133 |
__ASM_EMIT("andnps " OFF " + 0x20 + %[knee], %%xmm2") /* xmm2 = [x0 <= start] & GV */ \ |
| 134 |
__ASM_EMIT("orps %%xmm2, %%xmm0") /* xmm0 = ([x0 <= start]) ? GV : EV */ \ |
| 135 |
/* out: xmm0 = g0 */ |
| 136 |
|
| 137 |
#define PROCESS_COMP_FULL_X8 \ |
| 138 |
/* in: xmm0 = x0, xmm4 = x1 */ \ |
| 139 |
__ASM_EMIT("andps 0x00 + %[C2C], %%xmm0") /* xmm0 = fabsf(x0) */ \ |
| 140 |
__ASM_EMIT("andps 0x00 + %[C2C], %%xmm4") \ |
| 141 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") \ |
| 142 |
__ASM_EMIT("movaps %%xmm4, %%xmm5") \ |
| 143 |
__ASM_EMIT("movaps %%xmm0, %%xmm2") \ |
| 144 |
__ASM_EMIT("movaps %%xmm4, %%xmm6") \ |
| 145 |
__ASM_EMIT("cmpps $6, 0x00 + 0x00 + %[knee], %%xmm1") /* xmm1 = [x0 > k0.start] */ \ |
| 146 |
__ASM_EMIT("cmpps $6, 0x00 + 0x00 + %[knee], %%xmm5") \ |
| 147 |
__ASM_EMIT("cmpps $6, 0x80 + 0x00 + %[knee], %%xmm2") /* xmm2 = [x0 > k1.start] */ \ |
| 148 |
__ASM_EMIT("cmpps $6, 0x80 + 0x00 + %[knee], %%xmm6") \ |
| 149 |
__ASM_EMIT("orps %%xmm2, %%xmm1") \ |
| 150 |
__ASM_EMIT("orps %%xmm6, %%xmm5") \ |
| 151 |
__ASM_EMIT("orps %%xmm5, %%xmm1") \ |
| 152 |
__ASM_EMIT("movmskps %%xmm1, %[mask]") \ |
| 153 |
__ASM_EMIT("test %[mask], %[mask]") \ |
| 154 |
__ASM_EMIT("jnz 100f") \ |
| 155 |
__ASM_EMIT("movaps 0x00 + 0x20 + %[knee], %%xmm0") /* xmm0 = g1 */ \ |
| 156 |
__ASM_EMIT("movaps 0x00 + 0x20 + %[knee], %%xmm4") \ |
| 157 |
__ASM_EMIT("mulps 0x80 + 0x20 + %[knee], %%xmm0") /* xmm0 = G = g0*g1 */ \ |
| 158 |
__ASM_EMIT("mulps 0x80 + 0x20 + %[knee], %%xmm4") \ |
| 159 |
__ASM_EMIT("jmp 200f") \ |
| 160 |
__ASM_EMIT("100:") \ |
| 161 |
__ASM_EMIT("movaps %%xmm0, 0x00 + %[mem]") /* store fabsf(x0) */ \ |
| 162 |
__ASM_EMIT("movaps %%xmm4, 0x10 + %[mem]") \ |
| 163 |
LOGE_CORE_X8 /* xmm0 = lx0 = logf(fabsf(x0)) */ \ |
| 164 |
__ASM_EMIT("movaps %%xmm0, 0x20 + %[mem]") /* *mem = lx0 */ \ |
| 165 |
__ASM_EMIT("movaps %%xmm4, 0x30 + %[mem]") \ |
| 166 |
PROCESS_KNEE_SINGLE_X8("0x00") /* apply knee 0 */ \ |
| 167 |
__ASM_EMIT("movaps %%xmm0, 0x40 + %[mem]") /* *mem = g1 */ \ |
| 168 |
__ASM_EMIT("movaps %%xmm4, 0x50 + %[mem]") \ |
| 169 |
__ASM_EMIT("movaps 0x20 + %[mem], %%xmm0") /* xmm0 = lx0 */ \ |
| 170 |
__ASM_EMIT("movaps 0x30 + %[mem], %%xmm4") \ |
| 171 |
PROCESS_KNEE_SINGLE_X8("0x80") /* apply knee 1 */ \ |
| 172 |
__ASM_EMIT("mulps 0x40 + %[mem], %%xmm0") /* xmm0 = G = g0*g1 */ \ |
| 173 |
__ASM_EMIT("mulps 0x50 + %[mem], %%xmm4") \ |
| 174 |
__ASM_EMIT("200:") \ |
| 175 |
/* out: xmm0 = G0, xmm4 = G1 */ |
| 176 |
|
| 177 |
#define PROCESS_COMP_FULL_X4 \ |
| 178 |
/* in: xmm0 = x0 */ \ |
| 179 |
__ASM_EMIT("andps 0x00 + %[C2C], %%xmm0") /* xmm0 = fabsf(x0) */ \ |
| 180 |
__ASM_EMIT("movaps %%xmm0, 0x00 + %[mem]") /* store fabsf(x0) */ \ |
| 181 |
LOGE_CORE_X4 /* xmm0 = lx0 = logf(fabsf(x0)) */ \ |
| 182 |
__ASM_EMIT("movaps %%xmm0, 0x20 + %[mem]") /* *mem = lx0 */ \ |
| 183 |
PROCESS_KNEE_SINGLE_X4("0x00") /* apply knee 0 */ \ |
| 184 |
__ASM_EMIT("movaps %%xmm0, 0x40 + %[mem]") /* *mem = g1 */ \ |
| 185 |
__ASM_EMIT("movaps 0x20 + %[mem], %%xmm0") /* xmm0 = lx0 */ \ |
| 186 |
PROCESS_KNEE_SINGLE_X4("0x80") /* apply knee 1 */ \ |
| 187 |
__ASM_EMIT("mulps 0x40 + %[mem], %%xmm0") /* xmm0 = G = g0*g1 */ \ |
| 188 |
/* out: xmm0 = G0 */ |
| 189 |
|
| 190 |
static const uint32_t compressor_const[] __lsp_aligned16 = |
| 191 |
{ |
| 192 |
LSP_DSP_VEC4(0x7fffffff) |
| 193 |
}; |
| 194 |
|
| 195 |
void compressor_x2_gain(float *dst, const float *src, const dsp::compressor_x2_t *c, size_t count) |
| 196 |
{ |
| 197 |
IF_ARCH_X86( |
| 198 |
comp_knee_t knee[2] __lsp_aligned16; |
| 199 |
float mem[24] __lsp_aligned16; |
| 200 |
float stub[4] __lsp_aligned16; |
| 201 |
size_t mask; |
| 202 |
); |
| 203 |
|
| 204 |
ARCH_X86_ASM |
| 205 |
( |
| 206 |
// Prepare stuff |
| 207 |
UNPACK_COMP_KNEE("knee", "0x00", "comp", "0x00") |
| 208 |
UNPACK_COMP_KNEE("knee", "0x80", "comp", "0x20") |
| 209 |
|
| 210 |
// 8x blocks |
| 211 |
__ASM_EMIT("sub $8, %[count]") |
| 212 |
__ASM_EMIT("jb 2f") |
| 213 |
__ASM_EMIT("1:") |
| 214 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm0") |
| 215 |
__ASM_EMIT("movups 0x10(%[src]), %%xmm4") |
| 216 |
PROCESS_COMP_FULL_X8 |
| 217 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 218 |
__ASM_EMIT("movups %%xmm4, 0x10(%[dst])") |
| 219 |
__ASM_EMIT("add $0x20, %[src]") |
| 220 |
__ASM_EMIT("add $0x20, %[dst]") |
| 221 |
__ASM_EMIT("sub $8, %[count]") |
| 222 |
__ASM_EMIT("jae 1b") |
| 223 |
__ASM_EMIT("2:") |
| 224 |
// 4x blocks |
| 225 |
__ASM_EMIT("add $4, %[count]") |
| 226 |
__ASM_EMIT("jl 4f") |
| 227 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm0") |
| 228 |
PROCESS_COMP_FULL_X4 |
| 229 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 230 |
__ASM_EMIT("sub $4, %[count]") |
| 231 |
__ASM_EMIT("add $0x10, %[src]") |
| 232 |
__ASM_EMIT("add $0x10, %[dst]") |
| 233 |
__ASM_EMIT("4:") |
| 234 |
// Tail: 1x-3x block |
| 235 |
__ASM_EMIT("add $4, %[count]") |
| 236 |
__ASM_EMIT("jle 12f") |
| 237 |
__ASM_EMIT("test $1, %[count]") |
| 238 |
__ASM_EMIT("jz 6f") |
| 239 |
__ASM_EMIT("movss 0x00(%[src]), %%xmm0") |
| 240 |
__ASM_EMIT("add $4, %[src]") |
| 241 |
__ASM_EMIT("6:") |
| 242 |
__ASM_EMIT("test $2, %[count]") |
| 243 |
__ASM_EMIT("jz 8f") |
| 244 |
__ASM_EMIT("movhps 0x00(%[src]), %%xmm0") |
| 245 |
__ASM_EMIT("8:") |
| 246 |
PROCESS_COMP_FULL_X4 |
| 247 |
__ASM_EMIT("test $1, %[count]") |
| 248 |
__ASM_EMIT("jz 10f") |
| 249 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst])") |
| 250 |
__ASM_EMIT("add $4, %[dst]") |
| 251 |
__ASM_EMIT("10:") |
| 252 |
__ASM_EMIT("test $2, %[count]") |
| 253 |
__ASM_EMIT("jz 12f") |
| 254 |
__ASM_EMIT("movhps %%xmm0, 0x00(%[dst])") |
| 255 |
__ASM_EMIT("12:") |
| 256 |
|
| 257 |
: [dst] "+r" (dst), [src] "+r" (src), |
| 258 |
[count] "+r" (count), |
| 259 |
[mask] "=&r" (mask) |
| 260 |
: [comp] "r" (c), |
| 261 |
[knee] "o" (knee), |
| 262 |
[mem] "o" (mem), |
| 263 |
[stub] "o" (stub), |
| 264 |
[C2C] "o" (compressor_const), |
| 265 |
[L2C] "o" (LOG2_CONST), |
| 266 |
[LOGC] "o" (LOGE_C), |
| 267 |
[E2C] "o" (EXP2_CONST), |
| 268 |
[LOG2E] "m" (EXP_LOG2E) |
| 269 |
: "cc", "memory", |
| 270 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 271 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 272 |
); |
| 273 |
} |
| 274 |
|
| 275 |
void compressor_x2_curve(float *dst, const float *src, const dsp::compressor_x2_t *c, size_t count) |
| 276 |
{ |
| 277 |
IF_ARCH_X86( |
| 278 |
comp_knee_t knee[2] __lsp_aligned16; |
| 279 |
float mem[24] __lsp_aligned16; |
| 280 |
float stub[4] __lsp_aligned16; |
| 281 |
size_t mask; |
| 282 |
); |
| 283 |
|
| 284 |
ARCH_X86_ASM |
| 285 |
( |
| 286 |
// Prepare stuff |
| 287 |
UNPACK_COMP_KNEE("knee", "0x00", "comp", "0x00") |
| 288 |
UNPACK_COMP_KNEE("knee", "0x80", "comp", "0x20") |
| 289 |
|
| 290 |
// 8x blocks |
| 291 |
__ASM_EMIT("sub $8, %[count]") |
| 292 |
__ASM_EMIT("jb 2f") |
| 293 |
__ASM_EMIT("1:") |
| 294 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm0") |
| 295 |
__ASM_EMIT("movups 0x10(%[src]), %%xmm4") |
| 296 |
PROCESS_COMP_FULL_X8 |
| 297 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm1") |
| 298 |
__ASM_EMIT("movups 0x10(%[src]), %%xmm5") |
| 299 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") |
| 300 |
__ASM_EMIT("mulps %%xmm5, %%xmm4") |
| 301 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 302 |
__ASM_EMIT("movups %%xmm4, 0x10(%[dst])") |
| 303 |
__ASM_EMIT("add $0x20, %[src]") |
| 304 |
__ASM_EMIT("add $0x20, %[dst]") |
| 305 |
__ASM_EMIT("sub $8, %[count]") |
| 306 |
__ASM_EMIT("jae 1b") |
| 307 |
__ASM_EMIT("2:") |
| 308 |
// 4x blocks |
| 309 |
__ASM_EMIT("add $4, %[count]") |
| 310 |
__ASM_EMIT("jl 4f") |
| 311 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm0") |
| 312 |
PROCESS_COMP_FULL_X4 |
| 313 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm1") |
| 314 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") |
| 315 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 316 |
__ASM_EMIT("sub $4, %[count]") |
| 317 |
__ASM_EMIT("add $0x10, %[src]") |
| 318 |
__ASM_EMIT("add $0x10, %[dst]") |
| 319 |
__ASM_EMIT("4:") |
| 320 |
// Tail: 1x-3x block |
| 321 |
__ASM_EMIT("add $4, %[count]") |
| 322 |
__ASM_EMIT("jle 12f") |
| 323 |
__ASM_EMIT("test $1, %[count]") |
| 324 |
__ASM_EMIT("jz 6f") |
| 325 |
__ASM_EMIT("movss 0x00(%[src]), %%xmm0") |
| 326 |
__ASM_EMIT("add $4, %[src]") |
| 327 |
__ASM_EMIT("6:") |
| 328 |
__ASM_EMIT("test $2, %[count]") |
| 329 |
__ASM_EMIT("jz 8f") |
| 330 |
__ASM_EMIT("movhps 0x00(%[src]), %%xmm0") |
| 331 |
__ASM_EMIT("8:") |
| 332 |
__ASM_EMIT("movaps %%xmm0, %%xmm4") |
| 333 |
PROCESS_COMP_FULL_X4 |
| 334 |
__ASM_EMIT("mulps %%xmm4, %%xmm0") |
| 335 |
__ASM_EMIT("test $1, %[count]") |
| 336 |
__ASM_EMIT("jz 10f") |
| 337 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst])") |
| 338 |
__ASM_EMIT("add $4, %[dst]") |
| 339 |
__ASM_EMIT("10:") |
| 340 |
__ASM_EMIT("test $2, %[count]") |
| 341 |
__ASM_EMIT("jz 12f") |
| 342 |
__ASM_EMIT("movhps %%xmm0, 0x00(%[dst])") |
| 343 |
__ASM_EMIT("12:") |
| 344 |
|
| 345 |
: [dst] "+r" (dst), [src] "+r" (src), |
| 346 |
[count] "+r" (count), |
| 347 |
[mask] "=&r" (mask) |
| 348 |
: [comp] "r" (c), |
| 349 |
[knee] "o" (knee), |
| 350 |
[mem] "o" (mem), |
| 351 |
[stub] "o" (stub), |
| 352 |
[C2C] "o" (compressor_const), |
| 353 |
[L2C] "o" (LOG2_CONST), |
| 354 |
[LOGC] "o" (LOGE_C), |
| 355 |
[E2C] "o" (EXP2_CONST), |
| 356 |
[LOG2E] "m" (EXP_LOG2E) |
| 357 |
: "cc", "memory", |
| 358 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 359 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 360 |
); |
| 361 |
} |
| 362 |
|
| 363 |
#undef PROCESS_KNEE_SINGLE_X4 |
| 364 |
#undef PROCESS_KNEE_SINGLE_X8 |
| 365 |
#undef PROCESS_COMP_FULL_X4 |
| 366 |
#undef PROCESS_COMP_FULL_X8 |
| 367 |
#undef UNPACK_COMP_KNEE |
| 368 |
|
| 369 |
} /* namespace sse2 */ |
| 370 |
} /* namespace lsp */ |
| 371 |
|
| 372 |
|
| 373 |
#endif /* DSP_ARCH_X86_SSE2_DYNAMICS_COMPRESSOR_H_ */ |