| 1 |
/* |
| 2 |
* Copyright (C) 2025 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2025 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 10 апр. 2025 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_PMATH_SIN_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_SSE2_PMATH_SIN_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 |
namespace lsp |
| 30 |
{ |
| 31 |
namespace sse2 |
| 32 |
{ |
| 33 |
/* |
| 34 |
Time-optimized sinf implementation: |
| 35 |
|
| 36 |
float sinf(float x) |
| 37 |
{ |
| 38 |
// Limit x to [-pi/2 .. pi/2] |
| 39 |
x += PI/2; |
| 40 |
int32_t period = (x / 2*PI) - (x < 0.0f); |
| 41 |
x = x - period * 2*PI; |
| 42 |
x = (x < PI) ? x - PI/2 : 3*PI/2 - x; |
| 43 |
|
| 44 |
const float x2 = x*x; |
| 45 |
return x * (1.0f + x2*(-0.166666666667f + x2*(0.00833333333333f + x2 * (-0.000198412698413f + x2 * (2.7557319224e-06f + x2 * -2.50521083854e-08f))))); |
| 46 |
} |
| 47 |
*/ |
| 48 |
|
| 49 |
IF_ARCH_X86( |
| 50 |
static const float kp_gen_const[] __lsp_aligned16 = |
| 51 |
{ |
| 52 |
0.0f, 1.0f, 2.0f, 3.0f, // +0x00: Initial values 0..3 |
| 53 |
LSP_DSP_VEC4(4.0f) // +0x10: Step |
| 54 |
}; |
| 55 |
) |
| 56 |
|
| 57 |
IF_ARCH_X86( |
| 58 |
static const uint32_t sinf_const[] __lsp_aligned16 = |
| 59 |
{ |
| 60 |
LSP_DSP_VEC4(0x3fc90fdb), // +0x00: PI/2 |
| 61 |
LSP_DSP_VEC4(0x40490fdb), // +0x10: PI |
| 62 |
LSP_DSP_VEC4(0x3e22f983), // +0x20: 1/(2*PI) |
| 63 |
LSP_DSP_VEC4(0x40c90fdb), // +0x30: 2*PI |
| 64 |
LSP_DSP_VEC4(0x4096cbe4), // +0x40: 3*PI/2 |
| 65 |
LSP_DSP_VEC4(0xb2d7322b), // +0x50: C0 = -1/11! = -2.50521083854e-08 |
| 66 |
LSP_DSP_VEC4(0x3638ef1d), // +0x60: C1 = 1/9! = 2.7557319224e-06 |
| 67 |
LSP_DSP_VEC4(0xb9500d01), // +0x70: C2 = -1/7! = -0.000198412698413 |
| 68 |
LSP_DSP_VEC4(0x3c088889), // +0x80: C3 = 1/5! = 0.00833333333333 |
| 69 |
LSP_DSP_VEC4(0xbe2aaaab), // +0x90: C4 = -1/3! = -0.166666666667 |
| 70 |
LSP_DSP_VEC4(0x3f800000), // +0xa0: 1.0 |
| 71 |
}; |
| 72 |
) |
| 73 |
|
| 74 |
#define SINF_X_PLUS_PI_2_CORE_X8 \ |
| 75 |
/* xmm0 = X1 = x1 + PI/2 */ \ |
| 76 |
/* xmm4 = X2 = x2 + PI/2 */ \ |
| 77 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = X */ \ |
| 78 |
__ASM_EMIT("movaps %%xmm4, %%xmm5") \ |
| 79 |
__ASM_EMIT("movaps %%xmm0, %%xmm3") /* xmm3 = X */ \ |
| 80 |
__ASM_EMIT("movaps %%xmm4, %%xmm7") \ |
| 81 |
__ASM_EMIT("mulps 0x20 + %[S2C], %%xmm1") /* xmm1 = X / (2*PI) */ \ |
| 82 |
__ASM_EMIT("mulps 0x20 + %[S2C], %%xmm5") \ |
| 83 |
__ASM_EMIT("psrad $31, %%xmm3") /* xmm3 = [ X < 0.0 ] ? -1 : 0 */ \ |
| 84 |
__ASM_EMIT("psrad $31, %%xmm7") \ |
| 85 |
__ASM_EMIT("cvttps2dq %%xmm1, %%xmm1") /* xmm1 = int(X / (2*PI)) */ \ |
| 86 |
__ASM_EMIT("cvttps2dq %%xmm5, %%xmm5") \ |
| 87 |
__ASM_EMIT("paddd %%xmm3, %%xmm1") /* xmm1 = int(X / (2*PI)) + [ X < 0.0 ] ? -1 : 0 */ \ |
| 88 |
__ASM_EMIT("paddd %%xmm7, %%xmm5") \ |
| 89 |
__ASM_EMIT("cvtdq2ps %%xmm1, %%xmm1") /* xmm1 = period = int(X / (2*PI)) [ X < 0.0 ] ? -1 : 0 */ \ |
| 90 |
__ASM_EMIT("cvtdq2ps %%xmm5, %%xmm5") \ |
| 91 |
__ASM_EMIT("mulps 0x30 + %[S2C], %%xmm1") /* xmm1 = period * 2 * PI */ \ |
| 92 |
__ASM_EMIT("mulps 0x30 + %[S2C], %%xmm5") \ |
| 93 |
__ASM_EMIT("subps %%xmm1, %%xmm0") /* xmm0 = Y = X - period * 2 * PI */ \ |
| 94 |
__ASM_EMIT("subps %%xmm5, %%xmm4") \ |
| 95 |
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = Y */ \ |
| 96 |
__ASM_EMIT("movaps %%xmm4, %%xmm6") \ |
| 97 |
__ASM_EMIT("movaps 0x40 + %[S2C], %%xmm1") /* xmm1 = 3*PI/2 */ \ |
| 98 |
__ASM_EMIT("movaps 0x40 + %[S2C], %%xmm5") \ |
| 99 |
__ASM_EMIT("cmpps $1, 0x10 + %[S2C], %%xmm2") /* xmm2 = [ Y < PI ] */ \ |
| 100 |
__ASM_EMIT("cmpps $1, 0x10 + %[S2C], %%xmm6") \ |
| 101 |
__ASM_EMIT("subps %%xmm0, %%xmm1") /* xmm1 = 3*PI/2 - Y */ \ |
| 102 |
__ASM_EMIT("subps %%xmm4, %%xmm5") \ |
| 103 |
__ASM_EMIT("subps 0x00 + %[S2C], %%xmm0") /* xmm0 = Y - PI/2 */ \ |
| 104 |
__ASM_EMIT("subps 0x00 + %[S2C], %%xmm4") \ |
| 105 |
__ASM_EMIT("andps %%xmm2, %%xmm0") /* xmm0 = (Y - PI/2) & [ Y < PI ] */ \ |
| 106 |
__ASM_EMIT("andps %%xmm6, %%xmm4") \ |
| 107 |
__ASM_EMIT("andnps %%xmm1, %%xmm2") /* xmm2 = (3*PI/2 - Y) & [ Y >= PI ] */ \ |
| 108 |
__ASM_EMIT("andnps %%xmm5, %%xmm6") \ |
| 109 |
__ASM_EMIT("orps %%xmm2, %%xmm0") /* xmm0 = XX = [ Y < PI ] ? (Y - PI/2) : (3*PI/2 - Y) */ \ |
| 110 |
__ASM_EMIT("orps %%xmm6, %%xmm4") \ |
| 111 |
/* xmm0 = XX1 */ \ |
| 112 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = XX */ \ |
| 113 |
__ASM_EMIT("movaps %%xmm4, %%xmm5") \ |
| 114 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") /* xmm0 = X2 = XX*XX */ \ |
| 115 |
__ASM_EMIT("mulps %%xmm4, %%xmm4") \ |
| 116 |
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = X2 */ \ |
| 117 |
__ASM_EMIT("movaps %%xmm4, %%xmm6") \ |
| 118 |
__ASM_EMIT("mulps 0x50 + %[S2C], %%xmm0") /* xmm0 = X2*C0 */ \ |
| 119 |
__ASM_EMIT("mulps 0x50 + %[S2C], %%xmm4") \ |
| 120 |
__ASM_EMIT("addps 0x60 + %[S2C], %%xmm0") /* xmm0 = C1 + X2*C0 */ \ |
| 121 |
__ASM_EMIT("addps 0x60 + %[S2C], %%xmm4") \ |
| 122 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = X2*(C1 + X2*C0) */ \ |
| 123 |
__ASM_EMIT("mulps %%xmm6, %%xmm4") \ |
| 124 |
__ASM_EMIT("addps 0x70 + %[S2C], %%xmm0") /* xmm0 = C2 + X2*(C1 + X2*C0) */ \ |
| 125 |
__ASM_EMIT("addps 0x70 + %[S2C], %%xmm4") \ |
| 126 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 127 |
__ASM_EMIT("mulps %%xmm6, %%xmm4") \ |
| 128 |
__ASM_EMIT("addps 0x80 + %[S2C], %%xmm0") /* xmm0 = C3 + X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 129 |
__ASM_EMIT("addps 0x80 + %[S2C], %%xmm4") \ |
| 130 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 131 |
__ASM_EMIT("mulps %%xmm6, %%xmm4") \ |
| 132 |
__ASM_EMIT("addps 0x90 + %[S2C], %%xmm0") /* xmm0 = C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 133 |
__ASM_EMIT("addps 0x90 + %[S2C], %%xmm4") \ |
| 134 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0)))) */ \ |
| 135 |
__ASM_EMIT("mulps %%xmm6, %%xmm4") \ |
| 136 |
__ASM_EMIT("addps 0xa0 + %[S2C], %%xmm0") /* xmm0 = 1.0 + X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0)))) */ \ |
| 137 |
__ASM_EMIT("addps 0xa0 + %[S2C], %%xmm4") \ |
| 138 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = sinf(x) = XX * (1.0 + X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))))) */ \ |
| 139 |
__ASM_EMIT("mulps %%xmm5, %%xmm4") |
| 140 |
|
| 141 |
#define SINF_X_PLUS_PI_2_CORE_X4 \ |
| 142 |
/* xmm0 = X = x + PI/2 */ \ |
| 143 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = X */ \ |
| 144 |
__ASM_EMIT("movaps %%xmm0, %%xmm3") /* xmm3 = X */ \ |
| 145 |
__ASM_EMIT("mulps 0x20 + %[S2C], %%xmm1") /* xmm1 = X / (2*PI) */ \ |
| 146 |
__ASM_EMIT("psrad $31, %%xmm3") /* xmm3 = [ X < 0.0 ] ? -1 : 0 */ \ |
| 147 |
__ASM_EMIT("cvttps2dq %%xmm1, %%xmm1") /* xmm1 = int(X / (2*PI)) */ \ |
| 148 |
__ASM_EMIT("paddd %%xmm3, %%xmm1") /* xmm1 = int(X / (2*PI)) + [ X < 0.0 ] ? -1 : 0 */ \ |
| 149 |
__ASM_EMIT("cvtdq2ps %%xmm1, %%xmm1") /* xmm1 = period = int(X / (2*PI)) [ X < 0.0 ] ? -1 : 0 */ \ |
| 150 |
__ASM_EMIT("mulps 0x30 + %[S2C], %%xmm1") /* xmm1 = period * 2 * PI */ \ |
| 151 |
__ASM_EMIT("subps %%xmm1, %%xmm0") /* xmm0 = Y = X - period * 2 * PI */ \ |
| 152 |
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = Y */ \ |
| 153 |
__ASM_EMIT("movaps 0x40 + %[S2C], %%xmm1") /* xmm1 = 3*PI/2 */ \ |
| 154 |
__ASM_EMIT("cmpps $1, 0x10 + %[S2C], %%xmm2") /* xmm2 = [ Y < PI ] */ \ |
| 155 |
__ASM_EMIT("subps %%xmm0, %%xmm1") /* xmm1 = 3*PI/2 - Y */ \ |
| 156 |
__ASM_EMIT("subps 0x00 + %[S2C], %%xmm0") /* xmm0 = Y - PI/2 */ \ |
| 157 |
__ASM_EMIT("andps %%xmm2, %%xmm0") /* xmm0 = (Y - PI/2) & [ Y < PI ] */ \ |
| 158 |
__ASM_EMIT("andnps %%xmm1, %%xmm2") /* xmm2 = (3*PI/2 - Y) & [ Y >= PI ] */ \ |
| 159 |
__ASM_EMIT("orps %%xmm2, %%xmm0") /* xmm0 = XX = [ Y < PI ] ? (Y - PI/2) : (3*PI/2 - Y) */ \ |
| 160 |
/* xmm0 = XX1 */ \ |
| 161 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = XX */ \ |
| 162 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") /* xmm0 = X2 = XX*XX */ \ |
| 163 |
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = X2 */ \ |
| 164 |
__ASM_EMIT("mulps 0x50 + %[S2C], %%xmm0") /* xmm0 = X2*C0 */ \ |
| 165 |
__ASM_EMIT("addps 0x60 + %[S2C], %%xmm0") /* xmm0 = C1 + X2*C0 */ \ |
| 166 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = X2*(C1 + X2*C0) */ \ |
| 167 |
__ASM_EMIT("addps 0x70 + %[S2C], %%xmm0") /* xmm0 = C2 + X2*(C1 + X2*C0) */ \ |
| 168 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 169 |
__ASM_EMIT("addps 0x80 + %[S2C], %%xmm0") /* xmm0 = C3 + X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 170 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 171 |
__ASM_EMIT("addps 0x90 + %[S2C], %%xmm0") /* xmm0 = C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 172 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0)))) */ \ |
| 173 |
__ASM_EMIT("addps 0xa0 + %[S2C], %%xmm0") /* xmm0 = 1.0 + X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0)))) */ \ |
| 174 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = sinf(x) = XX * (1.0 + X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))))) */ |
| 175 |
|
| 176 |
void sinf1(float *dst, size_t count) |
| 177 |
{ |
| 178 |
IF_ARCH_X86(float *src); |
| 179 |
|
| 180 |
ARCH_X86_ASM( |
| 181 |
// x8 blocks |
| 182 |
__ASM_EMIT("sub $8, %[count]") |
| 183 |
__ASM_EMIT("jb 2f") |
| 184 |
__ASM_EMIT("1:") |
| 185 |
__ASM_EMIT("movaps 0x00 + %[S2C], %%xmm1") // xmm1 = PI/2 |
| 186 |
__ASM_EMIT("movups 0x00(%[dst]), %%xmm0") // xmm0 = x1 |
| 187 |
__ASM_EMIT("movups 0x10(%[dst]), %%xmm4") // xmm4 = x2 |
| 188 |
__ASM_EMIT("addps %%xmm1, %%xmm0") // xmm0 = x1 + PI/2 |
| 189 |
__ASM_EMIT("addps %%xmm1, %%xmm4") // xmm4 = x2 + PI/2 |
| 190 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 191 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 192 |
__ASM_EMIT("movups %%xmm4, 0x10(%[dst])") |
| 193 |
__ASM_EMIT("add $0x20, %[dst]") |
| 194 |
__ASM_EMIT("sub $8, %[count]") |
| 195 |
__ASM_EMIT("jae 1b") |
| 196 |
__ASM_EMIT("2:") |
| 197 |
// x4 block |
| 198 |
__ASM_EMIT("add $4, %[count]") |
| 199 |
__ASM_EMIT("jl 4f") |
| 200 |
__ASM_EMIT("movups 0x00(%[dst]), %%xmm0") // xmm0 = x |
| 201 |
__ASM_EMIT("addps 0x00 + %[S2C], %%xmm0") // xmm0 = x + PI/2 |
| 202 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 203 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 204 |
__ASM_EMIT("add $0x10, %[dst]") |
| 205 |
__ASM_EMIT("sub $4, %[count]") |
| 206 |
__ASM_EMIT("4:") |
| 207 |
// Tail: 1x-3x block |
| 208 |
__ASM_EMIT("add $4, %[count]") |
| 209 |
__ASM_EMIT("jle 12f") |
| 210 |
__ASM_EMIT("mov %[dst], %[src]") |
| 211 |
__ASM_EMIT("test $1, %[count]") |
| 212 |
__ASM_EMIT("jz 6f") |
| 213 |
__ASM_EMIT("movss 0x00(%[src]), %%xmm0") |
| 214 |
__ASM_EMIT("add $4, %[src]") |
| 215 |
__ASM_EMIT("6:") |
| 216 |
__ASM_EMIT("test $2, %[count]") |
| 217 |
__ASM_EMIT("jz 8f") |
| 218 |
__ASM_EMIT("movhps 0x00(%[src]), %%xmm0") |
| 219 |
__ASM_EMIT("8:") |
| 220 |
__ASM_EMIT("addps 0x00 + %[S2C], %%xmm0") // xmm0 = x + PI/2 |
| 221 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 222 |
__ASM_EMIT("test $1, %[count]") |
| 223 |
__ASM_EMIT("jz 10f") |
| 224 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst])") |
| 225 |
__ASM_EMIT("add $4, %[dst]") |
| 226 |
__ASM_EMIT("10:") |
| 227 |
__ASM_EMIT("test $2, %[count]") |
| 228 |
__ASM_EMIT("jz 12f") |
| 229 |
__ASM_EMIT("movhps %%xmm0, 0x00(%[dst])") |
| 230 |
// End |
| 231 |
__ASM_EMIT("12:") |
| 232 |
|
| 233 |
: [dst] "+r" (dst), [src] "=&r" (src), [count] "+r" (count) |
| 234 |
: [S2C] "o" (sinf_const) |
| 235 |
: "cc", "memory", |
| 236 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 237 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 238 |
); |
| 239 |
} |
| 240 |
|
| 241 |
void sinf2(float *dst, const float *src, size_t count) |
| 242 |
{ |
| 243 |
ARCH_X86_ASM( |
| 244 |
// x8 blocks |
| 245 |
__ASM_EMIT("sub $8, %[count]") |
| 246 |
__ASM_EMIT("jb 2f") |
| 247 |
__ASM_EMIT("1:") |
| 248 |
__ASM_EMIT("movaps 0x00 + %[S2C], %%xmm1") // xmm1 = PI/2 |
| 249 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm0") // xmm0 = x1 |
| 250 |
__ASM_EMIT("movups 0x10(%[src]), %%xmm4") // xmm4 = x2 |
| 251 |
__ASM_EMIT("addps %%xmm1, %%xmm0") // xmm0 = x1 + PI/2 |
| 252 |
__ASM_EMIT("addps %%xmm1, %%xmm4") // xmm4 = x2 + PI/2 |
| 253 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 254 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 255 |
__ASM_EMIT("movups %%xmm4, 0x10(%[dst])") |
| 256 |
__ASM_EMIT("add $0x20, %[src]") |
| 257 |
__ASM_EMIT("add $0x20, %[dst]") |
| 258 |
__ASM_EMIT("sub $8, %[count]") |
| 259 |
__ASM_EMIT("jae 1b") |
| 260 |
__ASM_EMIT("2:") |
| 261 |
// x4 block |
| 262 |
__ASM_EMIT("add $4, %[count]") |
| 263 |
__ASM_EMIT("jl 4f") |
| 264 |
__ASM_EMIT("movups 0x00(%[src]), %%xmm0") // xmm0 = x |
| 265 |
__ASM_EMIT("addps 0x00 + %[S2C], %%xmm0") // xmm0 = x + PI/2 |
| 266 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 267 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 268 |
__ASM_EMIT("add $0x10, %[src]") |
| 269 |
__ASM_EMIT("add $0x10, %[dst]") |
| 270 |
__ASM_EMIT("sub $4, %[count]") |
| 271 |
__ASM_EMIT("4:") |
| 272 |
// Tail: 1x-3x block |
| 273 |
__ASM_EMIT("add $4, %[count]") |
| 274 |
__ASM_EMIT("jle 12f") |
| 275 |
__ASM_EMIT("test $1, %[count]") |
| 276 |
__ASM_EMIT("jz 6f") |
| 277 |
__ASM_EMIT("movss 0x00(%[src]), %%xmm0") |
| 278 |
__ASM_EMIT("add $4, %[src]") |
| 279 |
__ASM_EMIT("6:") |
| 280 |
__ASM_EMIT("test $2, %[count]") |
| 281 |
__ASM_EMIT("jz 8f") |
| 282 |
__ASM_EMIT("movhps 0x00(%[src]), %%xmm0") |
| 283 |
__ASM_EMIT("8:") |
| 284 |
__ASM_EMIT("addps 0x00 + %[S2C], %%xmm0") // xmm0 = x + PI/2 |
| 285 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 286 |
__ASM_EMIT("test $1, %[count]") |
| 287 |
__ASM_EMIT("jz 10f") |
| 288 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst])") |
| 289 |
__ASM_EMIT("add $4, %[dst]") |
| 290 |
__ASM_EMIT("10:") |
| 291 |
__ASM_EMIT("test $2, %[count]") |
| 292 |
__ASM_EMIT("jz 12f") |
| 293 |
__ASM_EMIT("movhps %%xmm0, 0x00(%[dst])") |
| 294 |
// End |
| 295 |
__ASM_EMIT("12:") |
| 296 |
|
| 297 |
: [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) |
| 298 |
: [S2C] "o" (sinf_const) |
| 299 |
: "cc", "memory", |
| 300 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 301 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 302 |
); |
| 303 |
} |
| 304 |
|
| 305 |
void sinf_kp1(float *dst, float k, float p, size_t count) |
| 306 |
{ |
| 307 |
ARCH_X86_ASM( |
| 308 |
// Prepare |
| 309 |
__ASM_EMIT("movss %[k], %%xmm6") // xmm6 = k |
| 310 |
__ASM_EMIT("movss %[p], %%xmm7") // xmm7 = p |
| 311 |
__ASM_EMIT("shufps $0x00, %%xmm6, %%xmm6") // xmm6 = k k k k |
| 312 |
__ASM_EMIT("shufps $0x00, %%xmm7, %%xmm7") // xmm7 = p p p p |
| 313 |
__ASM_EMIT("movaps 0x00 + %[S2KP], %%xmm4") // xmm4 = i = 0 1 2 3 |
| 314 |
__ASM_EMIT("movaps 0x10 + %[S2KP], %%xmm5") // xmm5 = step = 4 4 4 4 |
| 315 |
__ASM_EMIT("addps 0x00 + %[S2C], %%xmm7") // xmm7 = p+PI/2 p+PI/2 p+PI/2 p+PI/2 |
| 316 |
// x4 blocks |
| 317 |
__ASM_EMIT("sub $4, %[count]") |
| 318 |
__ASM_EMIT("jb 2f") |
| 319 |
__ASM_EMIT("1:") |
| 320 |
__ASM_EMIT("movaps %%xmm4, %%xmm0") // xmm0 = i |
| 321 |
__ASM_EMIT("mulps %%xmm6, %%xmm0") // xmm0 = k*i |
| 322 |
__ASM_EMIT("addps %%xmm5, %%xmm4") // xmm4 = i + step |
| 323 |
__ASM_EMIT("addps %%xmm7, %%xmm0") // xmm0 = k*i + p + PI/2 |
| 324 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 325 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 326 |
__ASM_EMIT("add $0x10, %[dst]") |
| 327 |
__ASM_EMIT("sub $4, %[count]") |
| 328 |
__ASM_EMIT("jae 1b") |
| 329 |
__ASM_EMIT("2:") |
| 330 |
// Tail: 1x-3x block |
| 331 |
__ASM_EMIT("add $4, %[count]") |
| 332 |
__ASM_EMIT("jle 6f") |
| 333 |
__ASM_EMIT("movaps %%xmm4, %%xmm0") // xmm0 = i |
| 334 |
__ASM_EMIT("mulps %%xmm6, %%xmm0") // xmm0 = k*i |
| 335 |
__ASM_EMIT("addps %%xmm7, %%xmm0") // xmm0 = k*i + p + PI/2 |
| 336 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 337 |
__ASM_EMIT("test $1, %[count]") |
| 338 |
__ASM_EMIT("jz 4f") |
| 339 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst])") |
| 340 |
__ASM_EMIT("shufps $0x39, %%xmm0, %%xmm0") // xmm0 = S1 S2 S3 S0 |
| 341 |
__ASM_EMIT("add $4, %[dst]") |
| 342 |
__ASM_EMIT("4:") |
| 343 |
__ASM_EMIT("test $2, %[count]") |
| 344 |
__ASM_EMIT("jz 6f") |
| 345 |
__ASM_EMIT("movlps %%xmm0, 0x00(%[dst])") |
| 346 |
// End |
| 347 |
__ASM_EMIT("6:") |
| 348 |
|
| 349 |
: [dst] "+r" (dst), [count] "+r" (count) |
| 350 |
: [S2C] "o" (sinf_const), |
| 351 |
[S2KP] "o" (kp_gen_const), |
| 352 |
[k] "m" (k), |
| 353 |
[p] "m" (p) |
| 354 |
: "cc", "memory", |
| 355 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 356 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 357 |
); |
| 358 |
} |
| 359 |
|
| 360 |
} /* namespace sse2 */ |
| 361 |
} /* namespace lsp */ |
| 362 |
|
| 363 |
|
| 364 |
|
| 365 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE2_PMATH_SIN_H_ */ |