| 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: 16 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_LANCZOS_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_SSE2_PMATH_LANCZOS_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/sin.h> |
| 30 |
|
| 31 |
namespace lsp |
| 32 |
{ |
| 33 |
namespace sse2 |
| 34 |
{ |
| 35 |
IF_ARCH_X86( |
| 36 |
static const uint32_t lanczos_const[] __lsp_aligned16 = |
| 37 |
{ |
| 38 |
LSP_DSP_VEC4(0x7fffffff), // +0x00: Mask for fabsf |
| 39 |
LSP_DSP_VEC4(0x38d1b717), // +0x10: Sinc threshold = 1e-4 |
| 40 |
LSP_DSP_VEC4(0x3f800000), // +0x20: 1.0 |
| 41 |
}; |
| 42 |
) |
| 43 |
|
| 44 |
typedef struct lanczos_gen_t |
| 45 |
{ |
| 46 |
float k[4]; // +0x00: Multiplier |
| 47 |
float p[4]; // +0x10: Initial phase |
| 48 |
float a[4]; // +0x20: Number of lobes |
| 49 |
float i[4]; // +0x30: Indices |
| 50 |
float x1[4]; // +0x40: Computed X1 |
| 51 |
float d[4]; // +0x50: Denominator |
| 52 |
float t[4]; // +0x60: Time range |
| 53 |
} lanczos_gen_t; |
| 54 |
|
| 55 |
#define LANCZOS_GEN_FUNC_X4 \ |
| 56 |
/* xmm0 = x1 */ \ |
| 57 |
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = x1 */ \ |
| 58 |
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = x1 */ \ |
| 59 |
__ASM_EMIT("mulps 0x20(%[state]), %%xmm4") /* xmm4 = x2 = x1*a */ \ |
| 60 |
__ASM_EMIT("movaps %%xmm0, 0x40(%[state])") /* *x = x1 */ \ |
| 61 |
__ASM_EMIT("mulps %%xmm4, %%xmm2") /* xmm2 = d = x1 * x2 */ \ |
| 62 |
__ASM_EMIT("addps 0x00 + %[S2C], %%xmm0") /* xmm0 = x1 + PI/2 */ \ |
| 63 |
__ASM_EMIT("addps 0x00 + %[S2C], %%xmm4") /* xmm4 = x2 + PI/2 */ \ |
| 64 |
__ASM_EMIT("movaps %%xmm2, 0x50(%[state])") /* *d = d */ \ |
| 65 |
SINF_X_PLUS_PI_2_CORE_X8 /* xmm0 = sinf(x1), xmm4 = sinf(x2) */ \ |
| 66 |
__ASM_EMIT("movaps 0x40(%[state]), %%xmm1") /* xmm1 = x */ \ |
| 67 |
__ASM_EMIT("mulps %%xmm4, %%xmm0") /* xmm0 = sinf(x1)*sinf(x2) */ \ |
| 68 |
__ASM_EMIT("andps 0x00 + %[LC], %%xmm1") /* xmm1 = fabsf(x) */ \ |
| 69 |
__ASM_EMIT("divps 0x50(%[state]), %%xmm0") /* xmm0 = F = (sinf(x1)*sinf(x2)) / (x1 * x2) */ \ |
| 70 |
__ASM_EMIT("movaps %%xmm1, %%xmm2") /* xmm2 = fabsf(x) */ \ |
| 71 |
__ASM_EMIT("cmpps $5, 0x10 + %[LC], %%xmm1") /* xmm1 = [ fabsf(x) >= 1e-4 ] */ \ |
| 72 |
__ASM_EMIT("andps %%xmm1, %%xmm0") /* xmm0 = [ fabsf(x) >= 1e-4 ] & f */ \ |
| 73 |
__ASM_EMIT("cmpps $1, 0x60(%[state]), %%xmm2") /* xmm2 = [ fabsf(x) < t ] */ \ |
| 74 |
__ASM_EMIT("andnps 0x20 + %[LC], %%xmm1") /* xmm1 = [ fabsf(x) < 1e-4 ] & 1.0 */ \ |
| 75 |
__ASM_EMIT("orps %%xmm1, %%xmm0") /* xmm0 = [ fabsf(x) >= 1e-4 ] ? f : 1.0 */ \ |
| 76 |
__ASM_EMIT("andps %%xmm2, %%xmm0") /* xmm0 = [ fabsf(x) < t ] ? ([ fabsf(x) >= 1e-4 ] ? f : 1.0) : 0.0 */ |
| 77 |
|
| 78 |
void lanczos1(float *dst, float k, float p, float t, float a, size_t count) |
| 79 |
{ |
| 80 |
IF_ARCH_X86( |
| 81 |
lanczos_gen_t state __lsp_aligned16; |
| 82 |
float stub[4] __lsp_aligned16; |
| 83 |
); |
| 84 |
|
| 85 |
ARCH_X86_ASM( |
| 86 |
// Prepare |
| 87 |
__ASM_EMIT("movss %[k], %%xmm4") // xmm4 = k |
| 88 |
__ASM_EMIT("movss %[p], %%xmm5") // xmm5 = p |
| 89 |
__ASM_EMIT("movss %[a], %%xmm6") // xmm6 = a |
| 90 |
__ASM_EMIT("movss %[t], %%xmm7") // xmm7 = t |
| 91 |
__ASM_EMIT("shufps $0x00, %%xmm4, %%xmm4") // xmm4 = k k k k |
| 92 |
__ASM_EMIT("shufps $0x00, %%xmm5, %%xmm5") // xmm5 = p p p p |
| 93 |
__ASM_EMIT("shufps $0x00, %%xmm6, %%xmm6") // xmm6 = a a a a |
| 94 |
__ASM_EMIT("shufps $0x00, %%xmm7, %%xmm7") // xmm7 = t t t t |
| 95 |
__ASM_EMIT("movaps %%xmm4, 0x00(%[state])") |
| 96 |
__ASM_EMIT("movaps %%xmm5, 0x10(%[state])") |
| 97 |
__ASM_EMIT("movaps %%xmm6, 0x20(%[state])") |
| 98 |
__ASM_EMIT("movaps %%xmm7, 0x60(%[state])") |
| 99 |
__ASM_EMIT("movaps 0x00 + %[LGEN], %%xmm1") // xmm1 = i = 0 1 2 3 |
| 100 |
// x4 blocks |
| 101 |
__ASM_EMIT("sub $4, %[count]") |
| 102 |
__ASM_EMIT("jb 2f") |
| 103 |
__ASM_EMIT("1:") |
| 104 |
__ASM_EMIT("movaps %%xmm1, %%xmm0") // xmm0 = i |
| 105 |
__ASM_EMIT("mulps 0x00(%[state]), %%xmm0") // xmm0 = k*i |
| 106 |
__ASM_EMIT("addps 0x10 + %[LGEN], %%xmm1") // xmm1 = i' = i + step |
| 107 |
__ASM_EMIT("subps 0x10(%[state]), %%xmm0") // xmm0 = x1 = k*i - p |
| 108 |
__ASM_EMIT("movaps %%xmm1, 0x30(%[state])") // *i = i' |
| 109 |
LANCZOS_GEN_FUNC_X4 |
| 110 |
__ASM_EMIT("movaps 0x30(%[state]), %%xmm1") // xmm1 = i' |
| 111 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst])") |
| 112 |
__ASM_EMIT("add $0x10, %[dst]") |
| 113 |
__ASM_EMIT("sub $4, %[count]") |
| 114 |
__ASM_EMIT("jae 1b") |
| 115 |
__ASM_EMIT("2:") |
| 116 |
// Tail: 1x-3x block |
| 117 |
__ASM_EMIT("add $4, %[count]") |
| 118 |
__ASM_EMIT("jle 6f") |
| 119 |
__ASM_EMIT("movaps %%xmm1, %%xmm0") // xmm0 = i |
| 120 |
__ASM_EMIT("mulps 0x00(%[state]), %%xmm0") // xmm0 = k*i |
| 121 |
__ASM_EMIT("subps 0x10(%[state]), %%xmm0") // xmm0 = x1 = k*i - p |
| 122 |
LANCZOS_GEN_FUNC_X4 |
| 123 |
__ASM_EMIT("test $1, %[count]") |
| 124 |
__ASM_EMIT("jz 4f") |
| 125 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst])") |
| 126 |
__ASM_EMIT("shufps $0x39, %%xmm0, %%xmm0") // xmm0 = S1 S2 S3 S0 |
| 127 |
__ASM_EMIT("add $4, %[dst]") |
| 128 |
__ASM_EMIT("4:") |
| 129 |
__ASM_EMIT("test $2, %[count]") |
| 130 |
__ASM_EMIT("jz 6f") |
| 131 |
__ASM_EMIT("movlps %%xmm0, 0x00(%[dst])") |
| 132 |
// End |
| 133 |
__ASM_EMIT("6:") |
| 134 |
|
| 135 |
: [dst] "+r" (dst), [count] "+r" (count) |
| 136 |
: [state] "r" (&state), |
| 137 |
[S2C] "o" (sinf_const), |
| 138 |
[LGEN] "o" (kp_gen_const), |
| 139 |
[LC] "o" (lanczos_const), |
| 140 |
[stub] "o" (stub), |
| 141 |
[k] "m" (k), |
| 142 |
[p] "m" (p), |
| 143 |
[a] "m" (a), |
| 144 |
[t] "m" (t) |
| 145 |
: "cc", "memory", |
| 146 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 147 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 148 |
); |
| 149 |
} |
| 150 |
|
| 151 |
} /* namespace sse2 */ |
| 152 |
} /* namespace lsp */ |
| 153 |
|
| 154 |
|
| 155 |
|
| 156 |
|
| 157 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE2_PMATH_LANCZOS_H_ */ |