| 1 |
/* |
| 2 |
* Copyright (C) 2024 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2024 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 29 нояб. 2024 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_AARCH64_ASIMD_PAN_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_AARCH64_ASIMD_PAN_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_AARCH64_ASIMD_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_AARCH64_ASIMD_IMPL */ |
| 28 |
|
| 29 |
namespace lsp |
| 30 |
{ |
| 31 |
namespace asimd |
| 32 |
{ |
| 33 |
IF_ARCH_AARCH64( |
| 34 |
static const float depan_lin_const_f[] __lsp_aligned32 = |
| 35 |
{ |
| 36 |
LSP_DSP_VEC8(1e-18f) |
| 37 |
}; |
| 38 |
); |
| 39 |
|
| 40 |
void depan_lin(float *dst, const float *l, const float *r, float dfl, size_t count) |
| 41 |
{ |
| 42 |
/* |
| 43 |
const float sl = fabsf(l[i]); |
| 44 |
const float sr = fabsf(r[i]); |
| 45 |
const float den = sl + sr; |
| 46 |
dst[i] = (den >= 1e-18f) ? sr / den : dfl; |
| 47 |
*/ |
| 48 |
ARCH_AARCH64_ASM |
| 49 |
( |
| 50 |
__ASM_EMIT("dup v31.4s, %[dfl].s[0]") /* v31 = dfl */ |
| 51 |
__ASM_EMIT("ldp q28, q29, [%[CC], #0x00]") /* v28, v29 = thresh */ |
| 52 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 53 |
__ASM_EMIT("b.lo 2f") |
| 54 |
/* 16x blocks */ |
| 55 |
__ASM_EMIT("1:") |
| 56 |
__ASM_EMIT("ldp q0, q1, [%[l], #0x00]") /* v0 = l */ |
| 57 |
__ASM_EMIT("ldp q2, q3, [%[l], #0x20]") |
| 58 |
__ASM_EMIT("ldp q4, q5, [%[r], #0x00]") /* v4 = r */ |
| 59 |
__ASM_EMIT("ldp q6, q7, [%[r], #0x20]") |
| 60 |
__ASM_EMIT("fabs v0.4s, v0.4s") /* v0 = fabsf(l) */ |
| 61 |
__ASM_EMIT("fabs v1.4s, v1.4s") |
| 62 |
__ASM_EMIT("fabs v2.4s, v2.4s") |
| 63 |
__ASM_EMIT("fabs v3.4s, v3.4s") |
| 64 |
__ASM_EMIT("fabs v4.4s, v4.4s") /* v4 = fabsf(r) */ |
| 65 |
__ASM_EMIT("fabs v5.4s, v5.4s") |
| 66 |
__ASM_EMIT("fabs v6.4s, v6.4s") |
| 67 |
__ASM_EMIT("fabs v7.4s, v7.4s") |
| 68 |
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") /* v0 = den = fabsf(l) + fabsf(r) */ |
| 69 |
__ASM_EMIT("fadd v1.4s, v1.4s, v5.4s") |
| 70 |
__ASM_EMIT("fadd v2.4s, v2.4s, v6.4s") |
| 71 |
__ASM_EMIT("fadd v3.4s, v3.4s, v7.4s") |
| 72 |
__ASM_EMIT("fcmge v24.4s, v0.4s, v28.4s") /* v24 = den >= thresh */ |
| 73 |
__ASM_EMIT("fcmge v25.4s, v1.4s, v29.4s") |
| 74 |
__ASM_EMIT("fcmge v26.4s, v2.4s, v28.4s") |
| 75 |
__ASM_EMIT("fcmge v27.4s, v3.4s, v29.4s") |
| 76 |
__ASM_EMIT("frecpe v16.4s, v0.4s") /* v16 = s2 */ |
| 77 |
__ASM_EMIT("frecpe v17.4s, v1.4s") |
| 78 |
__ASM_EMIT("frecpe v18.4s, v2.4s") |
| 79 |
__ASM_EMIT("frecpe v19.4s, v3.4s") |
| 80 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2) */ |
| 81 |
__ASM_EMIT("frecps v21.4s, v17.4s, v1.4s") |
| 82 |
__ASM_EMIT("frecps v22.4s, v18.4s, v2.4s") |
| 83 |
__ASM_EMIT("frecps v23.4s, v19.4s, v3.4s") |
| 84 |
__ASM_EMIT("fmul v16.4s, v20.4s, v16.4s") /* v16 = s2' = s2 * (2 - R*s2) */ |
| 85 |
__ASM_EMIT("fmul v17.4s, v21.4s, v17.4s") |
| 86 |
__ASM_EMIT("fmul v18.4s, v22.4s, v18.4s") |
| 87 |
__ASM_EMIT("fmul v19.4s, v23.4s, v19.4s") |
| 88 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2') */ |
| 89 |
__ASM_EMIT("frecps v21.4s, v17.4s, v1.4s") |
| 90 |
__ASM_EMIT("frecps v22.4s, v18.4s, v2.4s") |
| 91 |
__ASM_EMIT("frecps v23.4s, v19.4s, v3.4s") |
| 92 |
__ASM_EMIT("fmul v0.4s, v20.4s, v16.4s") /* v0 = s2" = s2' * (2 - R*s2) = 1/s2 */ |
| 93 |
__ASM_EMIT("fmul v1.4s, v21.4s, v17.4s") |
| 94 |
__ASM_EMIT("fmul v2.4s, v22.4s, v18.4s") |
| 95 |
__ASM_EMIT("fmul v3.4s, v23.4s, v19.4s") |
| 96 |
__ASM_EMIT("fmul v0.4s, v0.4s, v4.4s") /* v0 = pan = fabsf(r) / den */ |
| 97 |
__ASM_EMIT("fmul v1.4s, v1.4s, v5.4s") |
| 98 |
__ASM_EMIT("fmul v2.4s, v2.4s, v6.4s") |
| 99 |
__ASM_EMIT("fmul v3.4s, v3.4s, v7.4s") |
| 100 |
__ASM_EMIT("bif v0.16b, v31.16b, v24.16b") /* v0 = (den >= thresh) ? pan : dfl */ |
| 101 |
__ASM_EMIT("bif v1.16b, v31.16b, v25.16b") |
| 102 |
__ASM_EMIT("bif v2.16b, v31.16b, v26.16b") |
| 103 |
__ASM_EMIT("bif v3.16b, v31.16b, v27.16b") |
| 104 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 105 |
__ASM_EMIT("stp q2, q3, [%[dst], #0x20]") |
| 106 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 107 |
__ASM_EMIT("add %[l], %[l], #0x40") |
| 108 |
__ASM_EMIT("add %[r], %[r], #0x40") |
| 109 |
__ASM_EMIT("add %[dst], %[dst], #0x40") |
| 110 |
__ASM_EMIT("b.hs 1b") |
| 111 |
/* 8x block */ |
| 112 |
__ASM_EMIT("2:") |
| 113 |
__ASM_EMIT("adds %[count], %[count], #8") |
| 114 |
__ASM_EMIT("b.lt 4f") |
| 115 |
__ASM_EMIT("ldp q0, q1, [%[l], #0x00]") /* v0 = l */ |
| 116 |
__ASM_EMIT("ldp q4, q5, [%[r], #0x00]") /* v4 = r */ |
| 117 |
__ASM_EMIT("fabs v0.4s, v0.4s") /* v0 = fabsf(l) */ |
| 118 |
__ASM_EMIT("fabs v1.4s, v1.4s") |
| 119 |
__ASM_EMIT("fabs v4.4s, v4.4s") /* v4 = fabsf(r) */ |
| 120 |
__ASM_EMIT("fabs v5.4s, v5.4s") |
| 121 |
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") /* v0 = den = fabsf(l) + fabsf(r) */ |
| 122 |
__ASM_EMIT("fadd v1.4s, v1.4s, v5.4s") |
| 123 |
__ASM_EMIT("fcmge v24.4s, v0.4s, v28.4s") /* v24 = den >= thresh */ |
| 124 |
__ASM_EMIT("fcmge v25.4s, v1.4s, v29.4s") |
| 125 |
__ASM_EMIT("frecpe v16.4s, v0.4s") /* v16 = s2 */ |
| 126 |
__ASM_EMIT("frecpe v17.4s, v1.4s") |
| 127 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2) */ |
| 128 |
__ASM_EMIT("frecps v21.4s, v17.4s, v1.4s") |
| 129 |
__ASM_EMIT("fmul v16.4s, v20.4s, v16.4s") /* v16 = s2' = s2 * (2 - R*s2) */ |
| 130 |
__ASM_EMIT("fmul v17.4s, v21.4s, v17.4s") |
| 131 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2') */ |
| 132 |
__ASM_EMIT("frecps v21.4s, v17.4s, v1.4s") |
| 133 |
__ASM_EMIT("fmul v0.4s, v20.4s, v16.4s") /* v0 = s2" = s2' * (2 - R*s2) = 1/s2 */ |
| 134 |
__ASM_EMIT("fmul v1.4s, v21.4s, v17.4s") |
| 135 |
__ASM_EMIT("fmul v0.4s, v0.4s, v4.4s") /* v0 = pan = fabsf(r) / den */ |
| 136 |
__ASM_EMIT("fmul v1.4s, v1.4s, v5.4s") |
| 137 |
__ASM_EMIT("bif v0.16b, v31.16b, v24.16b") /* v0 = (den >= thresh) ? pan : dfl */ |
| 138 |
__ASM_EMIT("bif v1.16b, v31.16b, v25.16b") |
| 139 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 140 |
__ASM_EMIT("sub %[count], %[count], #8") |
| 141 |
__ASM_EMIT("add %[l], %[l], #0x20") |
| 142 |
__ASM_EMIT("add %[r], %[r], #0x20") |
| 143 |
__ASM_EMIT("add %[dst], %[dst], #0x20") |
| 144 |
/* 4x block */ |
| 145 |
__ASM_EMIT("4:") |
| 146 |
__ASM_EMIT("adds %[count], %[count], #4") |
| 147 |
__ASM_EMIT("b.lt 6f") |
| 148 |
__ASM_EMIT("ldr q0, [%[l], #0x00]") /* v0 = l */ |
| 149 |
__ASM_EMIT("ldr q4, [%[r], #0x00]") /* v4 = r */ |
| 150 |
__ASM_EMIT("fabs v0.4s, v0.4s") /* v0 = fabsf(l) */ |
| 151 |
__ASM_EMIT("fabs v4.4s, v4.4s") /* v4 = fabsf(r) */ |
| 152 |
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") /* v0 = den = fabsf(l) + fabsf(r) */ |
| 153 |
__ASM_EMIT("fcmge v24.4s, v0.4s, v28.4s") /* v24 = den >= thresh */ |
| 154 |
__ASM_EMIT("frecpe v16.4s, v0.4s") /* v16 = s2 */ |
| 155 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2) */ |
| 156 |
__ASM_EMIT("fmul v16.4s, v20.4s, v16.4s") /* v16 = s2' = s2 * (2 - R*s2) */ |
| 157 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2') */ |
| 158 |
__ASM_EMIT("fmul v0.4s, v20.4s, v16.4s") /* v0 = s2" = s2' * (2 - R*s2) = 1/s2 */ |
| 159 |
__ASM_EMIT("fmul v0.4s, v0.4s, v4.4s") /* v0 = pan = fabsf(r) / den */ |
| 160 |
__ASM_EMIT("bif v0.16b, v31.16b, v24.16b") /* v0 = (den >= thresh) ? pan : dfl */ |
| 161 |
__ASM_EMIT("str q0, [%[dst], #0x00]") |
| 162 |
__ASM_EMIT("sub %[count], %[count], #4") |
| 163 |
__ASM_EMIT("add %[l], %[l], #0x10") |
| 164 |
__ASM_EMIT("add %[r], %[r], #0x10") |
| 165 |
__ASM_EMIT("add %[dst], %[dst], #0x10") |
| 166 |
/* 1x blocks */ |
| 167 |
__ASM_EMIT("6:") |
| 168 |
__ASM_EMIT("adds %[count], %[count], #3") |
| 169 |
__ASM_EMIT("b.lt 8f") |
| 170 |
__ASM_EMIT("7:") |
| 171 |
__ASM_EMIT("ld1r {v0.4s}, [%[l]]") /* v0 = l */ |
| 172 |
__ASM_EMIT("ld1r {v4.4s}, [%[r]]") /* v4 = r */ |
| 173 |
__ASM_EMIT("fabs v0.4s, v0.4s") /* v0 = fabsf(l) */ |
| 174 |
__ASM_EMIT("fabs v4.4s, v4.4s") /* v4 = fabsf(r) */ |
| 175 |
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") /* v0 = den = fabsf(l) + fabsf(r) */ |
| 176 |
__ASM_EMIT("fcmge v24.4s, v0.4s, v28.4s") /* v24 = den >= thresh */ |
| 177 |
__ASM_EMIT("frecpe v16.4s, v0.4s") /* v16 = s2 */ |
| 178 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2) */ |
| 179 |
__ASM_EMIT("fmul v16.4s, v20.4s, v16.4s") /* v16 = s2' = s2 * (2 - R*s2) */ |
| 180 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2') */ |
| 181 |
__ASM_EMIT("fmul v0.4s, v20.4s, v16.4s") /* v0 = s2" = s2' * (2 - R*s2) = 1/s2 */ |
| 182 |
__ASM_EMIT("fmul v0.4s, v0.4s, v4.4s") /* v0 = pan = fabsf(r) / den */ |
| 183 |
__ASM_EMIT("bif v0.16b, v31.16b, v24.16b") /* v0 = (den >= thresh) ? pan : dfl */ |
| 184 |
__ASM_EMIT("st1 {v0.s}[0], [%[dst]]") |
| 185 |
__ASM_EMIT("subs %[count], %[count], #1") |
| 186 |
__ASM_EMIT("add %[l], %[l], #0x04") |
| 187 |
__ASM_EMIT("add %[r], %[r], #0x04") |
| 188 |
__ASM_EMIT("add %[dst], %[dst], #0x04") |
| 189 |
__ASM_EMIT("b.ge 7b") |
| 190 |
/* end */ |
| 191 |
__ASM_EMIT("8:") |
| 192 |
|
| 193 |
: [dst] "+r" (dst), [l] "+r" (l), [r] "+r" (r), |
| 194 |
[count] "+r" (count), |
| 195 |
[dfl] "+w" (dfl) |
| 196 |
: [CC] "r" (&depan_lin_const_f[0]) |
| 197 |
: "cc", "memory", |
| 198 |
/* "v0", */ "v1", "v2", "v3", |
| 199 |
"v4", "v5", "v6", "v7", |
| 200 |
"v16", "v17", "v18", "v19", |
| 201 |
"v20", "v21", "v22", "v23", |
| 202 |
"v24", "v25", "v26", "v27", |
| 203 |
"v28", "v29", "v30", "v31" |
| 204 |
); |
| 205 |
} |
| 206 |
|
| 207 |
IF_ARCH_AARCH64( |
| 208 |
static const float depan_eqpow_const_f[] __lsp_aligned32 = |
| 209 |
{ |
| 210 |
LSP_DSP_VEC8(1e-36f) |
| 211 |
}; |
| 212 |
); |
| 213 |
|
| 214 |
void depan_eqpow(float *dst, const float *l, const float *r, float dfl, size_t count) |
| 215 |
{ |
| 216 |
/* |
| 217 |
const float sl = l[i] * l[i]; |
| 218 |
const float sr = r[i] * r[i]; |
| 219 |
const float den = sl + sr; |
| 220 |
dst[i] = (den >= 1e-36f) ? sr / den : dfl; |
| 221 |
*/ |
| 222 |
ARCH_AARCH64_ASM |
| 223 |
( |
| 224 |
__ASM_EMIT("dup v31.4s, %[dfl].s[0]") /* v31 = dfl */ |
| 225 |
__ASM_EMIT("ldp q28, q29, [%[CC], #0x00]") /* v28, v29 = thresh */ |
| 226 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 227 |
__ASM_EMIT("b.lo 2f") |
| 228 |
/* 16x blocks */ |
| 229 |
__ASM_EMIT("1:") |
| 230 |
__ASM_EMIT("ldp q0, q1, [%[l], #0x00]") /* v0 = l */ |
| 231 |
__ASM_EMIT("ldp q2, q3, [%[l], #0x20]") |
| 232 |
__ASM_EMIT("ldp q4, q5, [%[r], #0x00]") /* v4 = r */ |
| 233 |
__ASM_EMIT("ldp q6, q7, [%[r], #0x20]") |
| 234 |
__ASM_EMIT("fmul v0.4s, v0.4s, v0.4s") /* v0 = l*l */ |
| 235 |
__ASM_EMIT("fmul v1.4s, v1.4s, v1.4s") |
| 236 |
__ASM_EMIT("fmul v2.4s, v2.4s, v2.4s") |
| 237 |
__ASM_EMIT("fmul v3.4s, v3.4s, v3.4s") |
| 238 |
__ASM_EMIT("fmul v4.4s, v4.4s, v4.4s") /* v4 = r*r */ |
| 239 |
__ASM_EMIT("fmul v5.4s, v5.4s, v5.4s") |
| 240 |
__ASM_EMIT("fmul v6.4s, v6.4s, v6.4s") |
| 241 |
__ASM_EMIT("fmul v7.4s, v7.4s, v7.4s") |
| 242 |
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") /* v0 = den = l*l + r*r */ |
| 243 |
__ASM_EMIT("fadd v1.4s, v1.4s, v5.4s") |
| 244 |
__ASM_EMIT("fadd v2.4s, v2.4s, v6.4s") |
| 245 |
__ASM_EMIT("fadd v3.4s, v3.4s, v7.4s") |
| 246 |
__ASM_EMIT("fcmge v24.4s, v0.4s, v28.4s") /* v24 = den >= thresh */ |
| 247 |
__ASM_EMIT("fcmge v25.4s, v1.4s, v29.4s") |
| 248 |
__ASM_EMIT("fcmge v26.4s, v2.4s, v28.4s") |
| 249 |
__ASM_EMIT("fcmge v27.4s, v3.4s, v29.4s") |
| 250 |
__ASM_EMIT("frecpe v16.4s, v0.4s") /* v16 = s2 */ |
| 251 |
__ASM_EMIT("frecpe v17.4s, v1.4s") |
| 252 |
__ASM_EMIT("frecpe v18.4s, v2.4s") |
| 253 |
__ASM_EMIT("frecpe v19.4s, v3.4s") |
| 254 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2) */ |
| 255 |
__ASM_EMIT("frecps v21.4s, v17.4s, v1.4s") |
| 256 |
__ASM_EMIT("frecps v22.4s, v18.4s, v2.4s") |
| 257 |
__ASM_EMIT("frecps v23.4s, v19.4s, v3.4s") |
| 258 |
__ASM_EMIT("fmul v16.4s, v20.4s, v16.4s") /* v16 = s2' = s2 * (2 - R*s2) */ |
| 259 |
__ASM_EMIT("fmul v17.4s, v21.4s, v17.4s") |
| 260 |
__ASM_EMIT("fmul v18.4s, v22.4s, v18.4s") |
| 261 |
__ASM_EMIT("fmul v19.4s, v23.4s, v19.4s") |
| 262 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2') */ |
| 263 |
__ASM_EMIT("frecps v21.4s, v17.4s, v1.4s") |
| 264 |
__ASM_EMIT("frecps v22.4s, v18.4s, v2.4s") |
| 265 |
__ASM_EMIT("frecps v23.4s, v19.4s, v3.4s") |
| 266 |
__ASM_EMIT("fmul v0.4s, v20.4s, v16.4s") /* v0 = s2" = s2' * (2 - R*s2) = 1/s2 */ |
| 267 |
__ASM_EMIT("fmul v1.4s, v21.4s, v17.4s") |
| 268 |
__ASM_EMIT("fmul v2.4s, v22.4s, v18.4s") |
| 269 |
__ASM_EMIT("fmul v3.4s, v23.4s, v19.4s") |
| 270 |
__ASM_EMIT("fmul v0.4s, v0.4s, v4.4s") /* v0 = pan = r*r / den */ |
| 271 |
__ASM_EMIT("fmul v1.4s, v1.4s, v5.4s") |
| 272 |
__ASM_EMIT("fmul v2.4s, v2.4s, v6.4s") |
| 273 |
__ASM_EMIT("fmul v3.4s, v3.4s, v7.4s") |
| 274 |
__ASM_EMIT("bif v0.16b, v31.16b, v24.16b") /* v0 = (den >= thresh) ? pan : dfl */ |
| 275 |
__ASM_EMIT("bif v1.16b, v31.16b, v25.16b") |
| 276 |
__ASM_EMIT("bif v2.16b, v31.16b, v26.16b") |
| 277 |
__ASM_EMIT("bif v3.16b, v31.16b, v27.16b") |
| 278 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 279 |
__ASM_EMIT("stp q2, q3, [%[dst], #0x20]") |
| 280 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 281 |
__ASM_EMIT("add %[l], %[l], #0x40") |
| 282 |
__ASM_EMIT("add %[r], %[r], #0x40") |
| 283 |
__ASM_EMIT("add %[dst], %[dst], #0x40") |
| 284 |
__ASM_EMIT("b.hs 1b") |
| 285 |
/* 8x block */ |
| 286 |
__ASM_EMIT("2:") |
| 287 |
__ASM_EMIT("adds %[count], %[count], #8") |
| 288 |
__ASM_EMIT("b.lt 4f") |
| 289 |
__ASM_EMIT("ldp q0, q1, [%[l], #0x00]") /* v0 = l */ |
| 290 |
__ASM_EMIT("ldp q4, q5, [%[r], #0x00]") /* v4 = r */ |
| 291 |
__ASM_EMIT("fmul v0.4s, v0.4s, v0.4s") /* v0 = l*l */ |
| 292 |
__ASM_EMIT("fmul v1.4s, v1.4s, v1.4s") |
| 293 |
__ASM_EMIT("fmul v4.4s, v4.4s, v4.4s") /* v4 = r*r */ |
| 294 |
__ASM_EMIT("fmul v5.4s, v5.4s, v5.4s") |
| 295 |
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") /* v0 = den = l*l + r*r */ |
| 296 |
__ASM_EMIT("fadd v1.4s, v1.4s, v5.4s") |
| 297 |
__ASM_EMIT("fcmge v24.4s, v0.4s, v28.4s") /* v24 = den >= thresh */ |
| 298 |
__ASM_EMIT("fcmge v25.4s, v1.4s, v29.4s") |
| 299 |
__ASM_EMIT("frecpe v16.4s, v0.4s") /* v16 = s2 */ |
| 300 |
__ASM_EMIT("frecpe v17.4s, v1.4s") |
| 301 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2) */ |
| 302 |
__ASM_EMIT("frecps v21.4s, v17.4s, v1.4s") |
| 303 |
__ASM_EMIT("fmul v16.4s, v20.4s, v16.4s") /* v16 = s2' = s2 * (2 - R*s2) */ |
| 304 |
__ASM_EMIT("fmul v17.4s, v21.4s, v17.4s") |
| 305 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2') */ |
| 306 |
__ASM_EMIT("frecps v21.4s, v17.4s, v1.4s") |
| 307 |
__ASM_EMIT("fmul v0.4s, v20.4s, v16.4s") /* v0 = s2" = s2' * (2 - R*s2) = 1/s2 */ |
| 308 |
__ASM_EMIT("fmul v1.4s, v21.4s, v17.4s") |
| 309 |
__ASM_EMIT("fmul v0.4s, v0.4s, v4.4s") /* v0 = pan = r*r / den */ |
| 310 |
__ASM_EMIT("fmul v1.4s, v1.4s, v5.4s") |
| 311 |
__ASM_EMIT("bif v0.16b, v31.16b, v24.16b") /* v0 = (den >= thresh) ? pan : dfl */ |
| 312 |
__ASM_EMIT("bif v1.16b, v31.16b, v25.16b") |
| 313 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 314 |
__ASM_EMIT("sub %[count], %[count], #8") |
| 315 |
__ASM_EMIT("add %[l], %[l], #0x20") |
| 316 |
__ASM_EMIT("add %[r], %[r], #0x20") |
| 317 |
__ASM_EMIT("add %[dst], %[dst], #0x20") |
| 318 |
/* 4x block */ |
| 319 |
__ASM_EMIT("4:") |
| 320 |
__ASM_EMIT("adds %[count], %[count], #4") |
| 321 |
__ASM_EMIT("b.lt 6f") |
| 322 |
__ASM_EMIT("ldr q0, [%[l], #0x00]") /* v0 = l */ |
| 323 |
__ASM_EMIT("ldr q4, [%[r], #0x00]") /* v4 = r */ |
| 324 |
__ASM_EMIT("fmul v0.4s, v0.4s, v0.4s") /* v0 = l*l */ |
| 325 |
__ASM_EMIT("fmul v4.4s, v4.4s, v4.4s") /* v4 = r*r */ |
| 326 |
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") /* v0 = den = l*l + r*r */ |
| 327 |
__ASM_EMIT("fcmge v24.4s, v0.4s, v28.4s") /* v24 = den >= thresh */ |
| 328 |
__ASM_EMIT("frecpe v16.4s, v0.4s") /* v16 = s2 */ |
| 329 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2) */ |
| 330 |
__ASM_EMIT("fmul v16.4s, v20.4s, v16.4s") /* v16 = s2' = s2 * (2 - R*s2) */ |
| 331 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2') */ |
| 332 |
__ASM_EMIT("fmul v0.4s, v20.4s, v16.4s") /* v0 = s2" = s2' * (2 - R*s2) = 1/s2 */ |
| 333 |
__ASM_EMIT("fmul v0.4s, v0.4s, v4.4s") /* v0 = pan = r*r / den */ |
| 334 |
__ASM_EMIT("bif v0.16b, v31.16b, v24.16b") /* v0 = (den >= thresh) ? pan : dfl */ |
| 335 |
__ASM_EMIT("str q0, [%[dst], #0x00]") |
| 336 |
__ASM_EMIT("sub %[count], %[count], #4") |
| 337 |
__ASM_EMIT("add %[l], %[l], #0x10") |
| 338 |
__ASM_EMIT("add %[r], %[r], #0x10") |
| 339 |
__ASM_EMIT("add %[dst], %[dst], #0x10") |
| 340 |
/* 1x blocks */ |
| 341 |
__ASM_EMIT("6:") |
| 342 |
__ASM_EMIT("adds %[count], %[count], #3") |
| 343 |
__ASM_EMIT("b.lt 8f") |
| 344 |
__ASM_EMIT("7:") |
| 345 |
__ASM_EMIT("ld1r {v0.4s}, [%[l]]") /* v0 = l */ |
| 346 |
__ASM_EMIT("ld1r {v4.4s}, [%[r]]") /* v4 = r */ |
| 347 |
__ASM_EMIT("fmul v0.4s, v0.4s, v0.4s") /* v0 = l*l */ |
| 348 |
__ASM_EMIT("fmul v4.4s, v4.4s, v4.4s") /* v4 = r*r */ |
| 349 |
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") /* v0 = den = l*l + r*r */ |
| 350 |
__ASM_EMIT("fcmge v24.4s, v0.4s, v28.4s") /* v24 = den >= thresh */ |
| 351 |
__ASM_EMIT("frecpe v16.4s, v0.4s") /* v16 = s2 */ |
| 352 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2) */ |
| 353 |
__ASM_EMIT("fmul v16.4s, v20.4s, v16.4s") /* v16 = s2' = s2 * (2 - R*s2) */ |
| 354 |
__ASM_EMIT("frecps v20.4s, v16.4s, v0.4s") /* v20 = (2 - R*s2') */ |
| 355 |
__ASM_EMIT("fmul v0.4s, v20.4s, v16.4s") /* v0 = s2" = s2' * (2 - R*s2) = 1/s2 */ |
| 356 |
__ASM_EMIT("fmul v0.4s, v0.4s, v4.4s") /* v0 = pan = r*r / den */ |
| 357 |
__ASM_EMIT("bif v0.16b, v31.16b, v24.16b") /* v0 = (den >= thresh) ? pan : dfl */ |
| 358 |
__ASM_EMIT("st1 {v0.s}[0], [%[dst]]") |
| 359 |
__ASM_EMIT("subs %[count], %[count], #1") |
| 360 |
__ASM_EMIT("add %[l], %[l], #0x04") |
| 361 |
__ASM_EMIT("add %[r], %[r], #0x04") |
| 362 |
__ASM_EMIT("add %[dst], %[dst], #0x04") |
| 363 |
__ASM_EMIT("b.ge 7b") |
| 364 |
/* end */ |
| 365 |
__ASM_EMIT("8:") |
| 366 |
|
| 367 |
: [dst] "+r" (dst), [l] "+r" (l), [r] "+r" (r), |
| 368 |
[count] "+r" (count), |
| 369 |
[dfl] "+w" (dfl) |
| 370 |
: [CC] "r" (&depan_eqpow_const_f[0]) |
| 371 |
: "cc", "memory", |
| 372 |
/* "v0", */ "v1", "v2", "v3", |
| 373 |
"v4", "v5", "v6", "v7", |
| 374 |
"v16", "v17", "v18", "v19", |
| 375 |
"v20", "v21", "v22", "v23", |
| 376 |
"v24", "v25", "v26", "v27", |
| 377 |
"v28", "v29", "v30", "v31" |
| 378 |
); |
| 379 |
} |
| 380 |
|
| 381 |
} /* namespace asimd */ |
| 382 |
} /* namespace lsp */ |
| 383 |
|
| 384 |
|
| 385 |
#endif /* PRIVATE_DSP_ARCH_AARCH64_ASIMD_PAN_H_ */ |