| 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: 27 нояб. 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_SSE_PMATH_PMIX_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_SSE_PMATH_PMIX_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_SSE_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE_IMPL */ |
| 28 |
|
| 29 |
namespace lsp |
| 30 |
{ |
| 31 |
namespace sse |
| 32 |
{ |
| 33 |
|
| 34 |
void pmix_v1(float *dst, const float *src, const float *k, size_t count) |
| 35 |
{ |
| 36 |
IF_ARCH_X86(size_t off); |
| 37 |
ARCH_X86_ASM |
| 38 |
( |
| 39 |
__ASM_EMIT("xor %[off], %[off]") |
| 40 |
// 8x blocks |
| 41 |
__ASM_EMIT("sub $8, %[count]") |
| 42 |
__ASM_EMIT("jb 2f") |
| 43 |
__ASM_EMIT("1:") |
| 44 |
__ASM_EMIT("movups 0x00(%[src],%[off]), %%xmm0") /* xmm0 = s */ |
| 45 |
__ASM_EMIT("movups 0x10(%[src],%[off]), %%xmm1") |
| 46 |
__ASM_EMIT("movups 0x00(%[dst],%[off]), %%xmm2") /* xmm2 = d */ |
| 47 |
__ASM_EMIT("movups 0x10(%[dst],%[off]), %%xmm3") |
| 48 |
__ASM_EMIT("movups 0x00(%[k],%[off]), %%xmm4") /* xmm4 = k */ |
| 49 |
__ASM_EMIT("movups 0x10(%[k],%[off]), %%xmm5") |
| 50 |
__ASM_EMIT("subps %%xmm2, %%xmm0") /* xmm0 = s - d */ |
| 51 |
__ASM_EMIT("subps %%xmm3, %%xmm1") |
| 52 |
__ASM_EMIT("mulps %%xmm4, %%xmm0") /* xmm0 = (s-d) * k */ |
| 53 |
__ASM_EMIT("mulps %%xmm5, %%xmm1") |
| 54 |
__ASM_EMIT("addps %%xmm2, %%xmm0") /* xmm0 = mix = d + (s-d) * k */ |
| 55 |
__ASM_EMIT("addps %%xmm3, %%xmm1") |
| 56 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst],%[off])") |
| 57 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst],%[off])") |
| 58 |
__ASM_EMIT("add $0x20, %[off]") |
| 59 |
__ASM_EMIT("sub $8, %[count]") |
| 60 |
__ASM_EMIT("jae 1b") |
| 61 |
// 4x block |
| 62 |
__ASM_EMIT("2:") |
| 63 |
__ASM_EMIT("add $4, %[count]") |
| 64 |
__ASM_EMIT("jl 4f") |
| 65 |
__ASM_EMIT("movups 0x00(%[src],%[off]), %%xmm0") /* xmm0 = s */ |
| 66 |
__ASM_EMIT("movups 0x00(%[dst],%[off]), %%xmm2") /* xmm2 = d */ |
| 67 |
__ASM_EMIT("movups 0x00(%[k],%[off]), %%xmm4") /* xmm4 = k */ |
| 68 |
__ASM_EMIT("subps %%xmm2, %%xmm0") /* xmm0 = s - d */ |
| 69 |
__ASM_EMIT("mulps %%xmm4, %%xmm0") /* xmm0 = (s-d) * k */ |
| 70 |
__ASM_EMIT("addps %%xmm2, %%xmm0") /* xmm0 = mix = d + (s-d) * k */ |
| 71 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst],%[off])") |
| 72 |
__ASM_EMIT("sub $4, %[count]") |
| 73 |
__ASM_EMIT("add $0x10, %[off]") |
| 74 |
// 1x blocks |
| 75 |
__ASM_EMIT("4:") |
| 76 |
__ASM_EMIT("add $3, %[count]") |
| 77 |
__ASM_EMIT("jl 6f") |
| 78 |
__ASM_EMIT("5:") |
| 79 |
__ASM_EMIT("movss 0x00(%[src],%[off]), %%xmm0") /* xmm0 = s */ |
| 80 |
__ASM_EMIT("movss 0x00(%[dst],%[off]), %%xmm2") /* xmm2 = d */ |
| 81 |
__ASM_EMIT("movss 0x00(%[k],%[off]), %%xmm4") /* xmm4 = k */ |
| 82 |
__ASM_EMIT("subss %%xmm2, %%xmm0") /* xmm0 = s - d */ |
| 83 |
__ASM_EMIT("mulss %%xmm4, %%xmm0") /* xmm0 = (s-d) * k */ |
| 84 |
__ASM_EMIT("addss %%xmm2, %%xmm0") /* xmm0 = mix = d + (s-d) * k */ |
| 85 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst],%[off])") |
| 86 |
__ASM_EMIT("add $0x04, %[off]") |
| 87 |
__ASM_EMIT("dec %[count]") |
| 88 |
__ASM_EMIT("jge 5b") |
| 89 |
// End |
| 90 |
__ASM_EMIT("6:") |
| 91 |
: [off] "=&r" (off), [count] "+r" (count) |
| 92 |
: [dst] "r" (dst), [src] "r" (src), [k] "r" (k) |
| 93 |
: "cc", "memory", |
| 94 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 95 |
"%xmm4", "%xmm5" |
| 96 |
); |
| 97 |
} |
| 98 |
|
| 99 |
void pmix_k1(float *dst, const float *src, float k, size_t count) |
| 100 |
{ |
| 101 |
IF_ARCH_X86(size_t off); |
| 102 |
ARCH_X86_ASM |
| 103 |
( |
| 104 |
__ASM_EMIT("shufps $0x00, %[k], %[k]") |
| 105 |
__ASM_EMIT("xor %[off], %[off]") |
| 106 |
__ASM_EMIT("movaps %[k], %%xmm4") |
| 107 |
// 8x blocks |
| 108 |
__ASM_EMIT("sub $8, %[count]") |
| 109 |
__ASM_EMIT("jb 2f") |
| 110 |
__ASM_EMIT("1:") |
| 111 |
__ASM_EMIT("movups 0x00(%[src],%[off]), %%xmm0") /* xmm0 = s */ |
| 112 |
__ASM_EMIT("movups 0x10(%[src],%[off]), %%xmm1") |
| 113 |
__ASM_EMIT("movups 0x00(%[dst],%[off]), %%xmm2") /* xmm2 = d */ |
| 114 |
__ASM_EMIT("movups 0x10(%[dst],%[off]), %%xmm3") |
| 115 |
__ASM_EMIT("subps %%xmm2, %%xmm0") /* xmm0 = s - d */ |
| 116 |
__ASM_EMIT("subps %%xmm3, %%xmm1") |
| 117 |
__ASM_EMIT("mulps %%xmm4, %%xmm0") /* xmm0 = (s-d) * k */ |
| 118 |
__ASM_EMIT("mulps %%xmm4, %%xmm1") |
| 119 |
__ASM_EMIT("addps %%xmm2, %%xmm0") /* xmm0 = mix = d + (s-d) * k */ |
| 120 |
__ASM_EMIT("addps %%xmm3, %%xmm1") |
| 121 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst],%[off])") |
| 122 |
__ASM_EMIT("movups %%xmm1, 0x10(%[dst],%[off])") |
| 123 |
__ASM_EMIT("add $0x20, %[off]") |
| 124 |
__ASM_EMIT("sub $8, %[count]") |
| 125 |
__ASM_EMIT("jae 1b") |
| 126 |
// 4x block |
| 127 |
__ASM_EMIT("2:") |
| 128 |
__ASM_EMIT("add $4, %[count]") |
| 129 |
__ASM_EMIT("jl 4f") |
| 130 |
__ASM_EMIT("movups 0x00(%[src],%[off]), %%xmm0") /* xmm0 = s */ |
| 131 |
__ASM_EMIT("movups 0x00(%[dst],%[off]), %%xmm2") /* xmm2 = d */ |
| 132 |
__ASM_EMIT("subps %%xmm2, %%xmm0") /* xmm0 = s - d */ |
| 133 |
__ASM_EMIT("mulps %%xmm4, %%xmm0") /* xmm0 = (s-d) * k */ |
| 134 |
__ASM_EMIT("addps %%xmm2, %%xmm0") /* xmm0 = mix = d + (s-d) * k */ |
| 135 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst],%[off])") |
| 136 |
__ASM_EMIT("sub $4, %[count]") |
| 137 |
__ASM_EMIT("add $0x10, %[off]") |
| 138 |
// 1x blocks |
| 139 |
__ASM_EMIT("4:") |
| 140 |
__ASM_EMIT("add $3, %[count]") |
| 141 |
__ASM_EMIT("jl 6f") |
| 142 |
__ASM_EMIT("5:") |
| 143 |
__ASM_EMIT("movss 0x00(%[src],%[off]), %%xmm0") /* xmm0 = s */ |
| 144 |
__ASM_EMIT("movss 0x00(%[dst],%[off]), %%xmm2") /* xmm2 = d */ |
| 145 |
__ASM_EMIT("subss %%xmm2, %%xmm0") /* xmm0 = s - d */ |
| 146 |
__ASM_EMIT("mulss %%xmm4, %%xmm0") /* xmm0 = (s-d) * k */ |
| 147 |
__ASM_EMIT("addss %%xmm2, %%xmm0") /* xmm0 = mix = d + (s-d) * k */ |
| 148 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst],%[off])") |
| 149 |
__ASM_EMIT("add $0x04, %[off]") |
| 150 |
__ASM_EMIT("dec %[count]") |
| 151 |
__ASM_EMIT("jge 5b") |
| 152 |
// End |
| 153 |
__ASM_EMIT("6:") |
| 154 |
: [off] "=&r" (off), [count] "+r" (count), [k] "+Yz" (k) |
| 155 |
: [dst] "r" (dst), [src] "r" (src) |
| 156 |
: "cc", "memory", |
| 157 |
"%xmm1", "%xmm2", "%xmm3", |
| 158 |
"%xmm4" |
| 159 |
); |
| 160 |
} |
| 161 |
|
| 162 |
} /* namespace sse */ |
| 163 |
} /* namespace lsp */ |
| 164 |
|
| 165 |
|
| 166 |
|
| 167 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE_PMATH_PMIX_H_ */ |