| 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: 26 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_AVX512_PMATH_LRAMP_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_AVX512_PMATH_LRAMP_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_AVX512_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX512_IMPL */ |
| 28 |
|
| 29 |
namespace lsp |
| 30 |
{ |
| 31 |
namespace avx512 |
| 32 |
{ |
| 33 |
IF_ARCH_X86( |
| 34 |
static const float lramp_const[] __lsp_aligned64 = |
| 35 |
{ |
| 36 |
0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, /* Initial values 0..7 */ |
| 37 |
8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, /* Initial values 8..15 */ |
| 38 |
16.0f, 17.0f, 18.0f, 19.0f, 20.0f, 21.0f, 22.0f, 23.0f, /* Initial values 16..23 */ |
| 39 |
24.0f, 25.0f, 26.0f, 27.0f, 28.0f, 29.0f, 30.0f, 31.0f, /* Initial values 8..15 */ |
| 40 |
LSP_DSP_VEC16(32.0f) /* Step */ |
| 41 |
}; |
| 42 |
) |
| 43 |
|
| 44 |
void lramp_set1(float *dst, float v1, float v2, size_t count) |
| 45 |
{ |
| 46 |
float delta = v2 - v1; |
| 47 |
if (delta == 0.0f) |
| 48 |
{ |
| 49 |
dsp::fill(dst, v1, count); |
| 50 |
return; |
| 51 |
} |
| 52 |
if (count == 0) |
| 53 |
return; |
| 54 |
|
| 55 |
delta /= count; |
| 56 |
IF_ARCH_X86( size_t off ); |
| 57 |
ARCH_X86_ASM( |
| 58 |
__ASM_EMIT("xor %[off], %[off]") |
| 59 |
__ASM_EMIT("vbroadcastss %[v1], %%zmm6") /* zmm6 = v1 */ |
| 60 |
__ASM_EMIT("vbroadcastss %[delta], %%zmm7") /* zmm7 = delta */ |
| 61 |
__ASM_EMIT("vmovaps 0x00(%[CC]), %%zmm3") /* zmm3 = i[0] */ |
| 62 |
__ASM_EMIT("vmovaps 0x40(%[CC]), %%zmm4") /* zmm4 = i[1] */ |
| 63 |
__ASM_EMIT("vmovaps 0x80(%[CC]), %%zmm5") /* zmm5 = step */ |
| 64 |
/* 32x blocks */ |
| 65 |
__ASM_EMIT("sub $32, %[count]") |
| 66 |
__ASM_EMIT("jb 2f") |
| 67 |
__ASM_EMIT("1:") |
| 68 |
__ASM_EMIT("vmulps %%zmm7, %%zmm3, %%zmm0") /* zmm0 = i[0]*delta */ |
| 69 |
__ASM_EMIT("vmulps %%zmm7, %%zmm4, %%zmm1") /* zmm1 = i[1]*delta */ |
| 70 |
__ASM_EMIT("vaddps %%zmm5, %%zmm3, %%zmm3") /* zmm3 = x[0]' = x[0] + step */ |
| 71 |
__ASM_EMIT("vaddps %%zmm5, %%zmm4, %%zmm4") /* zmm4 = x[1]' = x[1] + step */ |
| 72 |
__ASM_EMIT("vaddps %%zmm6, %%zmm0, %%zmm0") /* zmm0 = v1 + i[0]*delta */ |
| 73 |
__ASM_EMIT("vaddps %%zmm6, %%zmm1, %%zmm1") /* zmm1 = v1 + i[1]*delta */ |
| 74 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst], %[off])") |
| 75 |
__ASM_EMIT("vmovups %%zmm1, 0x40(%[dst], %[off])") |
| 76 |
__ASM_EMIT("add $0x80, %[off]") |
| 77 |
__ASM_EMIT("sub $32, %[count]") |
| 78 |
__ASM_EMIT("jae 1b") |
| 79 |
__ASM_EMIT("2:") |
| 80 |
/* 16x block */ |
| 81 |
__ASM_EMIT("add $16, %[count]") |
| 82 |
__ASM_EMIT("jl 4f") |
| 83 |
__ASM_EMIT("vmulps %%zmm7, %%zmm3, %%zmm0") /* zmm0 = i[0]*delta */ |
| 84 |
__ASM_EMIT("vmovaps %%zmm4, %%zmm3") /* zmm3 = x[0]' = x[0] + step */ |
| 85 |
__ASM_EMIT("vaddps %%zmm6, %%zmm0, %%zmm0") /* zmm0 = v1 + i[0]*delta */ |
| 86 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[dst], %[off])") |
| 87 |
__ASM_EMIT("sub $16, %[count]") |
| 88 |
__ASM_EMIT("add $0x40, %[off]") |
| 89 |
__ASM_EMIT("4:") |
| 90 |
/* 8x blocks */ |
| 91 |
__ASM_EMIT("add $8, %[count]") |
| 92 |
__ASM_EMIT("jl 4f") |
| 93 |
__ASM_EMIT("vmulps %%ymm7, %%ymm3, %%ymm0") /* ymm0 = i[0]*delta */ |
| 94 |
__ASM_EMIT("vextractf32x8 $1, %%zmm3, %%ymm3") /* ymm3 = x[0]' = x[0] + step */ |
| 95 |
__ASM_EMIT("vaddps %%ymm6, %%ymm0, %%ymm0") /* ymm0 = v1 + i[0]*delta */ |
| 96 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[dst], %[off])") |
| 97 |
__ASM_EMIT("sub $8, %[count]") |
| 98 |
__ASM_EMIT("add $0x20, %[off]") |
| 99 |
__ASM_EMIT("4:") |
| 100 |
/* 4x block */ |
| 101 |
__ASM_EMIT("add $4, %[count]") |
| 102 |
__ASM_EMIT("jl 6f") |
| 103 |
__ASM_EMIT("vmulps %%xmm7, %%xmm3, %%xmm0") /* xmm0 = i[0]*delta */ |
| 104 |
__ASM_EMIT("vextractf128 $1, %%ymm3, %%xmm3") /* xmm3 = x[0]' = x[0] + step */ |
| 105 |
__ASM_EMIT("vaddps %%xmm6, %%xmm0, %%xmm0") /* xmm0 = v1 + i[0]*delta */ |
| 106 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[dst], %[off])") |
| 107 |
__ASM_EMIT("sub $4, %[count]") |
| 108 |
__ASM_EMIT("add $0x10, %[off]") |
| 109 |
__ASM_EMIT("6:") |
| 110 |
/* 1x blocks */ |
| 111 |
__ASM_EMIT("add $3, %[count]") |
| 112 |
__ASM_EMIT("jl 8f") |
| 113 |
__ASM_EMIT("vmulps %%xmm7, %%xmm3, %%xmm0") /* xmm0 = x[0]*delta */ |
| 114 |
__ASM_EMIT("vaddps %%xmm6, %%xmm0, %%xmm0") /* xmm0 = v1 + i[0]*delta */ |
| 115 |
__ASM_EMIT("7:") |
| 116 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[dst], %[off])") |
| 117 |
__ASM_EMIT("vshufps $0x39, %%xmm0, %%xmm0, %%xmm0") /* shift xmm0 */ |
| 118 |
__ASM_EMIT("add $0x04, %[off]") |
| 119 |
__ASM_EMIT("dec %[count]") |
| 120 |
__ASM_EMIT("jge 7b") |
| 121 |
__ASM_EMIT("8:") |
| 122 |
|
| 123 |
: [count] "+r" (count), [off] "=&r" (off) |
| 124 |
: [dst] "r" (dst), |
| 125 |
[CC] "r" (lramp_const), |
| 126 |
[delta] "m" (delta), |
| 127 |
[v1] "m" (v1) |
| 128 |
: "cc", "memory", |
| 129 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 130 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 131 |
); |
| 132 |
} |
| 133 |
|
| 134 |
#define LRAMP_CORE(DST, SRC) \ |
| 135 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 136 |
__ASM_EMIT("vbroadcastss %[v1], %%zmm6") /* zmm6 = v1 */ \ |
| 137 |
__ASM_EMIT("vbroadcastss %[delta], %%zmm7") /* zmm7 = delta */ \ |
| 138 |
__ASM_EMIT("vmovaps 0x00(%[CC]), %%zmm3") /* zmm3 = i[0] */ \ |
| 139 |
__ASM_EMIT("vmovaps 0x40(%[CC]), %%zmm4") /* zmm4 = i[1] */ \ |
| 140 |
__ASM_EMIT("vmovaps 0x80(%[CC]), %%zmm5") /* zmm5 = step */ \ |
| 141 |
/* 32x blocks */ \ |
| 142 |
__ASM_EMIT("sub $32, %[count]") \ |
| 143 |
__ASM_EMIT("jb 2f") \ |
| 144 |
__ASM_EMIT("1:") \ |
| 145 |
__ASM_EMIT("vmulps %%zmm7, %%zmm3, %%zmm0") /* zmm0 = i[0]*delta */ \ |
| 146 |
__ASM_EMIT("vmulps %%zmm7, %%zmm4, %%zmm1") /* zmm1 = i[1]*delta */ \ |
| 147 |
__ASM_EMIT("vaddps %%zmm5, %%zmm3, %%zmm3") /* zmm3 = x[0]' = x[0] + step */ \ |
| 148 |
__ASM_EMIT("vaddps %%zmm5, %%zmm4, %%zmm4") /* zmm4 = x[1]' = x[1] + step */ \ |
| 149 |
__ASM_EMIT("vaddps %%zmm6, %%zmm0, %%zmm0") /* zmm0 = v1 + i[0]*delta */ \ |
| 150 |
__ASM_EMIT("vaddps %%zmm6, %%zmm1, %%zmm1") /* zmm1 = v1 + i[1]*delta */ \ |
| 151 |
__ASM_EMIT("vmulps 0x00(%[" SRC "], %[off]), %%zmm0, %%zmm0") /* zmm0 = src[0] * (v1 + i[0]*delta) */ \ |
| 152 |
__ASM_EMIT("vmulps 0x40(%[" SRC "], %[off]), %%zmm1, %%zmm1") /* zmm1 = src[1] * (v1 + i[1]*delta) */ \ |
| 153 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[" DST "], %[off])") \ |
| 154 |
__ASM_EMIT("vmovups %%zmm1, 0x40(%[" DST "], %[off])") \ |
| 155 |
__ASM_EMIT("add $0x80, %[off]") \ |
| 156 |
__ASM_EMIT("sub $32, %[count]") \ |
| 157 |
__ASM_EMIT("jae 1b") \ |
| 158 |
__ASM_EMIT("2:") \ |
| 159 |
/* 16x blocks */ \ |
| 160 |
__ASM_EMIT("add $16, %[count]") \ |
| 161 |
__ASM_EMIT("jl 4f") \ |
| 162 |
__ASM_EMIT("vmulps %%zmm7, %%zmm3, %%zmm0") /* zmm0 = i[0]*delta */ \ |
| 163 |
__ASM_EMIT("vmovaps %%zmm4, %%zmm3") /* zmm3 = x[0]' = x[0] + step */ \ |
| 164 |
__ASM_EMIT("vaddps %%zmm6, %%zmm0, %%zmm0") /* zmm0 = v1 + i[0]*delta */ \ |
| 165 |
__ASM_EMIT("vmulps 0x00(%[" SRC "], %[off]), %%zmm0, %%zmm0") /* zmm0 = src[0] * (v1 + i[0]*delta) */ \ |
| 166 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[" DST "], %[off])") \ |
| 167 |
__ASM_EMIT("sub $16, %[count]") \ |
| 168 |
__ASM_EMIT("add $0x40, %[off]") \ |
| 169 |
__ASM_EMIT("4:") \ |
| 170 |
/* 8x blocks */ \ |
| 171 |
__ASM_EMIT("add $8, %[count]") \ |
| 172 |
__ASM_EMIT("jl 6f") \ |
| 173 |
__ASM_EMIT("vmulps %%ymm7, %%ymm3, %%ymm0") /* ymm0 = i[0]*delta */ \ |
| 174 |
__ASM_EMIT("vextractf32x8 $1, %%zmm3, %%ymm3") /* ymm3 = x[0]' = x[0] + step */ \ |
| 175 |
__ASM_EMIT("vaddps %%ymm6, %%ymm0, %%ymm0") /* ymm0 = v1 + i[0]*delta */ \ |
| 176 |
__ASM_EMIT("vmulps 0x00(%[" SRC "], %[off]), %%ymm0, %%ymm0") /* ymm0 = src[0] * (v1 + i[0]*delta) */ \ |
| 177 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" DST "], %[off])") \ |
| 178 |
__ASM_EMIT("sub $8, %[count]") \ |
| 179 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 180 |
__ASM_EMIT("6:") \ |
| 181 |
/* 4x block */ \ |
| 182 |
__ASM_EMIT("add $4, %[count]") \ |
| 183 |
__ASM_EMIT("jl 8f") \ |
| 184 |
__ASM_EMIT("vmulps %%xmm7, %%xmm3, %%xmm0") /* xmm0 = i[0]*delta */ \ |
| 185 |
__ASM_EMIT("vextractf128 $1, %%ymm3, %%xmm3") /* xmm3 = x[0]' = x[0] + step */ \ |
| 186 |
__ASM_EMIT("vaddps %%xmm6, %%xmm0, %%xmm0") /* xmm0 = v1 + i[0]*delta */ \ |
| 187 |
__ASM_EMIT("vmulps 0x00(%[" SRC "], %[off]), %%xmm0, %%xmm0") /* ymm0 = src[0] * (v1 + i[0]*delta) */ \ |
| 188 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[" DST "], %[off])") \ |
| 189 |
__ASM_EMIT("sub $4, %[count]") \ |
| 190 |
__ASM_EMIT("add $0x10, %[off]") \ |
| 191 |
__ASM_EMIT("8:") \ |
| 192 |
/* 1x blocks */ \ |
| 193 |
__ASM_EMIT("add $3, %[count]") \ |
| 194 |
__ASM_EMIT("jl 10f") \ |
| 195 |
__ASM_EMIT("9:") \ |
| 196 |
__ASM_EMIT("vmulss %%xmm7, %%xmm3, %%xmm0") /* xmm0 = x[0]*delta */ \ |
| 197 |
__ASM_EMIT("vshufps $0x39, %%xmm3, %%xmm3, %%xmm3") /* shift xmm3 */ \ |
| 198 |
__ASM_EMIT("vaddss %%xmm6, %%xmm0, %%xmm0") /* xmm0 = v1 + i[0]*delta */ \ |
| 199 |
__ASM_EMIT("vmulss 0x00(%[" SRC "], %[off]), %%xmm0, %%xmm0") /* ymm0 = src[0] * (v1 + i[0]*delta) */ \ |
| 200 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[" DST "], %[off])") \ |
| 201 |
__ASM_EMIT("add $0x04, %[off]") \ |
| 202 |
__ASM_EMIT("dec %[count]") \ |
| 203 |
__ASM_EMIT("jge 9b") \ |
| 204 |
__ASM_EMIT("10:") |
| 205 |
|
| 206 |
|
| 207 |
void lramp1(float *dst, float v1, float v2, size_t count) |
| 208 |
{ |
| 209 |
float delta = v2 - v1; |
| 210 |
if (delta == 0.0f) |
| 211 |
{ |
| 212 |
dsp::mul_k2(dst, v1, count); |
| 213 |
return; |
| 214 |
} |
| 215 |
if (count == 0) |
| 216 |
return; |
| 217 |
|
| 218 |
delta /= count; |
| 219 |
IF_ARCH_X86( size_t off ); |
| 220 |
ARCH_X86_ASM( |
| 221 |
LRAMP_CORE("dst", "dst") |
| 222 |
: [count] "+r" (count), [off] "=&r" (off) |
| 223 |
: [dst] "r" (dst), |
| 224 |
[CC] "r" (lramp_const), |
| 225 |
[delta] "m" (delta), |
| 226 |
[v1] "m" (v1) |
| 227 |
: "cc", "memory", |
| 228 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 229 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 230 |
); |
| 231 |
} |
| 232 |
|
| 233 |
void lramp2(float *dst, const float *src, float v1, float v2, size_t count) |
| 234 |
{ |
| 235 |
float delta = v2 - v1; |
| 236 |
if (delta == 0.0f) |
| 237 |
{ |
| 238 |
dsp::mul_k3(dst, src, v1, count); |
| 239 |
return; |
| 240 |
} |
| 241 |
if (count == 0) |
| 242 |
return; |
| 243 |
|
| 244 |
delta /= count; |
| 245 |
IF_ARCH_X86( size_t off ); |
| 246 |
ARCH_X86_ASM( |
| 247 |
LRAMP_CORE("dst", "src") |
| 248 |
: [count] "+r" (count), [off] "=&r" (off) |
| 249 |
: [dst] "r" (dst), [src] "r" (src), |
| 250 |
[CC] "r" (lramp_const), |
| 251 |
[delta] "m" (delta), |
| 252 |
[v1] "m" (v1) |
| 253 |
: "cc", "memory", |
| 254 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 255 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 256 |
); |
| 257 |
} |
| 258 |
|
| 259 |
#undef LRAMP_CORE |
| 260 |
|
| 261 |
#define OP_DSEL(a, b) a |
| 262 |
#define OP_RSEL(a, b) b |
| 263 |
|
| 264 |
#define LRAMP_OP3_CORE(DST, A, B, OP, SEL) \ |
| 265 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 266 |
__ASM_EMIT("vbroadcastss %[v1], %%zmm6") /* zmm6 = v1 */ \ |
| 267 |
__ASM_EMIT("vbroadcastss %[delta], %%zmm7") /* zmm7 = delta */ \ |
| 268 |
__ASM_EMIT("vmovaps 0x00(%[CC]), %%zmm4") /* zmm4 = i[0] */ \ |
| 269 |
__ASM_EMIT("vmovaps 0x40(%[CC]), %%zmm5") /* zmm5 = i[1] */ \ |
| 270 |
/* 32x blocks */ \ |
| 271 |
__ASM_EMIT("sub $32, %[count]") \ |
| 272 |
__ASM_EMIT("jb 2f") \ |
| 273 |
__ASM_EMIT("1:") \ |
| 274 |
__ASM_EMIT("vmulps %%zmm7, %%zmm4, %%zmm2") /* zmm2 = i[0]*delta */ \ |
| 275 |
__ASM_EMIT("vmulps %%zmm7, %%zmm5, %%zmm3") /* zmm3 = i[1]*delta */ \ |
| 276 |
__ASM_EMIT("vmovups 0x00(%[" A "], %[off]), %%zmm0") /* zmm0 = a[0] */ \ |
| 277 |
__ASM_EMIT("vmovups 0x40(%[" A "], %[off]), %%zmm1") /* zmm1 = a[1] */ \ |
| 278 |
__ASM_EMIT("vaddps 0x80(%[CC]), %%zmm4, %%zmm4") /* zmm4 = x[0]' = x[0] + step */ \ |
| 279 |
__ASM_EMIT("vaddps 0x80(%[CC]), %%zmm5, %%zmm5") /* zmm5 = x[1]' = x[1] + step */ \ |
| 280 |
__ASM_EMIT("vaddps %%zmm6, %%zmm2, %%zmm2") /* zmm2 = v1 + i[0]*delta */ \ |
| 281 |
__ASM_EMIT("vaddps %%zmm6, %%zmm3, %%zmm3") /* zmm3 = v1 + i[1]*delta */ \ |
| 282 |
__ASM_EMIT("vmulps 0x00(%[" B "], %[off]), %%zmm2, %%zmm2") /* zmm0 = b[0] * (v1 + i[0]*delta) */ \ |
| 283 |
__ASM_EMIT("vmulps 0x40(%[" B "], %[off]), %%zmm3, %%zmm3") /* zmm1 = b[1] * (v1 + i[1]*delta) */ \ |
| 284 |
__ASM_EMIT(OP "ps " SEL("%%zmm2, %%zmm0, %%zmm0", "%%zmm0, %%zmm2, %%zmm0")) /* zmm0 = a[0] OP b[0] * (v1 + i[0]*delta) */ \ |
| 285 |
__ASM_EMIT(OP "ps " SEL("%%zmm3, %%zmm1, %%zmm1", "%%zmm1, %%zmm3, %%zmm1")) /* zmm1 = a[1] OP b[1] * (v1 + i[1]*delta) */ \ |
| 286 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[" DST "], %[off])") \ |
| 287 |
__ASM_EMIT("vmovups %%zmm1, 0x40(%[" DST "], %[off])") \ |
| 288 |
__ASM_EMIT("add $0x80, %[off]") \ |
| 289 |
__ASM_EMIT("sub $32, %[count]") \ |
| 290 |
__ASM_EMIT("jae 1b") \ |
| 291 |
__ASM_EMIT("2:") \ |
| 292 |
/* 16x blocks */ \ |
| 293 |
__ASM_EMIT("add $16, %[count]") \ |
| 294 |
__ASM_EMIT("jl 4f") \ |
| 295 |
__ASM_EMIT("1:") \ |
| 296 |
__ASM_EMIT("vmulps %%zmm7, %%zmm4, %%zmm2") /* zmm2 = i[0]*delta */ \ |
| 297 |
__ASM_EMIT("vmovups 0x00(%[" A "], %[off]), %%zmm0") /* zmm0 = a[0] */ \ |
| 298 |
__ASM_EMIT("vmovaps %%zmm5, %%zmm4") /* zmm4 = x[0]' = x[0] + step */ \ |
| 299 |
__ASM_EMIT("vaddps %%zmm6, %%zmm2, %%zmm2") /* zmm2 = v1 + i[0]*delta */ \ |
| 300 |
__ASM_EMIT("vmulps 0x00(%[" B "], %[off]), %%zmm2, %%zmm2") /* zmm0 = b[0] * (v1 + i[0]*delta) */ \ |
| 301 |
__ASM_EMIT(OP "ps " SEL("%%zmm2, %%zmm0, %%zmm0", "%%zmm0, %%zmm2, %%zmm0")) /* zmm0 = a[0] OP b[0] * (v1 + i[0]*delta) */ \ |
| 302 |
__ASM_EMIT("vmovups %%zmm0, 0x00(%[" DST "], %[off])") \ |
| 303 |
__ASM_EMIT("add $0x40, %[off]") \ |
| 304 |
__ASM_EMIT("sub $16, %[count]") \ |
| 305 |
__ASM_EMIT("4:") \ |
| 306 |
/* 8x blocks */ \ |
| 307 |
__ASM_EMIT("add $8, %[count]") \ |
| 308 |
__ASM_EMIT("jl 6f") \ |
| 309 |
__ASM_EMIT("vmulps %%ymm7, %%ymm4, %%ymm2") /* ymm2 = i[0]*delta */ \ |
| 310 |
__ASM_EMIT("vmovups 0x00(%[" A "], %[off]), %%ymm0") /* ymm0 = a[0] */ \ |
| 311 |
__ASM_EMIT("vextractf32x8 $1, %%zmm4, %%ymm4") /* ymm4 = x[0]' = x[0] + step */ \ |
| 312 |
__ASM_EMIT("vaddps %%ymm6, %%ymm2, %%ymm2") /* zmm2 = v1 + i[0]*delta */ \ |
| 313 |
__ASM_EMIT("vmulps 0x00(%[" B "], %[off]), %%ymm2, %%ymm2") /* ymm0 = b[0] * (v1 + i[0]*delta) */ \ |
| 314 |
__ASM_EMIT(OP "ps " SEL("%%ymm2, %%ymm0, %%ymm0", "%%ymm0, %%ymm2, %%ymm0")) /* ymm0 = a[0] OP b[0] * (v1 + i[0]*delta) */ \ |
| 315 |
__ASM_EMIT("vmovups %%ymm0, 0x00(%[" DST "], %[off])") \ |
| 316 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 317 |
__ASM_EMIT("sub $8, %[count]") \ |
| 318 |
__ASM_EMIT("6:") \ |
| 319 |
/* 4x block */ \ |
| 320 |
__ASM_EMIT("add $4, %[count]") \ |
| 321 |
__ASM_EMIT("jl 8f") \ |
| 322 |
__ASM_EMIT("vmulps %%xmm7, %%xmm4, %%xmm2") /* xmm2 = i[0]*delta */ \ |
| 323 |
__ASM_EMIT("vmovups 0x00(%[" A "], %[off]), %%xmm0") /* xmm0 = a[0] */ \ |
| 324 |
__ASM_EMIT("vextractf128 $1, %%ymm4, %%xmm4") /* xmm4 = x[0]' = x[0] + step */ \ |
| 325 |
__ASM_EMIT("vaddps %%xmm6, %%xmm2, %%xmm2") /* xmm2 = v1 + i[0]*delta */ \ |
| 326 |
__ASM_EMIT("vmulps 0x00(%[" B "], %[off]), %%xmm2, %%xmm2") /* xmm0 = b[0] * (v1 + i[0]*delta) */ \ |
| 327 |
__ASM_EMIT(OP "ps " SEL("%%xmm2, %%xmm0, %%xmm0", "%%xmm0, %%xmm2, %%xmm0")) /* xmm0 = a[0] OP b[0] * (v1 + i[0]*delta) */ \ |
| 328 |
__ASM_EMIT("vmovups %%xmm0, 0x00(%[" DST "], %[off])") \ |
| 329 |
__ASM_EMIT("add $0x10, %[off]") \ |
| 330 |
__ASM_EMIT("sub $4, %[count]") \ |
| 331 |
__ASM_EMIT("8:") \ |
| 332 |
/* 1x blocks */ \ |
| 333 |
__ASM_EMIT("add $3, %[count]") \ |
| 334 |
__ASM_EMIT("jl 10f") \ |
| 335 |
__ASM_EMIT("9:") \ |
| 336 |
__ASM_EMIT("vmulss %%xmm7, %%xmm4, %%xmm2") /* xmm2 = i[0]*delta */ \ |
| 337 |
__ASM_EMIT("vmovss 0x00(%[" A "], %[off]), %%xmm0") /* xmm0 = a[0] */ \ |
| 338 |
__ASM_EMIT("vshufps $0x39, %%xmm4, %%xmm4, %%xmm4") /* shift xmm4 */ \ |
| 339 |
__ASM_EMIT("vaddss %%xmm6, %%xmm2, %%xmm2") /* xmm2 = v1 + i[0]*delta */ \ |
| 340 |
__ASM_EMIT("vmulss 0x00(%[" B "], %[off]), %%xmm2, %%xmm2") /* xmm0 = b[0] * (v1 + i[0]*delta) */ \ |
| 341 |
__ASM_EMIT(OP "ss " SEL("%%xmm2, %%xmm0, %%xmm0", "%%xmm0, %%xmm2, %%xmm0")) /* xmm0 = a[0] OP b[0] * (v1 + i[0]*delta) */ \ |
| 342 |
__ASM_EMIT("vmovss %%xmm0, 0x00(%[" DST "], %[off])") \ |
| 343 |
__ASM_EMIT("add $0x04, %[off]") \ |
| 344 |
__ASM_EMIT("dec %[count]") \ |
| 345 |
__ASM_EMIT("jge 9b") \ |
| 346 |
__ASM_EMIT("10:") |
| 347 |
|
| 348 |
void lramp_add2(float *dst, const float *src, float v1, float v2, size_t count) |
| 349 |
{ |
| 350 |
float delta = v2 - v1; |
| 351 |
if (delta == 0.0f) |
| 352 |
{ |
| 353 |
dsp::fmadd_k3(dst, src, v1, count); |
| 354 |
return; |
| 355 |
} |
| 356 |
if (count == 0) |
| 357 |
return; |
| 358 |
|
| 359 |
delta /= count; |
| 360 |
IF_ARCH_X86( size_t off ); |
| 361 |
ARCH_X86_ASM( |
| 362 |
LRAMP_OP3_CORE("dst", "dst", "src", "vadd", OP_DSEL) |
| 363 |
: [count] "+r" (count), [off] "=&r" (off) |
| 364 |
: [dst] "r" (dst), [src] "r" (src), |
| 365 |
[CC] "r" (lramp_const), |
| 366 |
[delta] "m" (delta), |
| 367 |
[v1] "m" (v1) |
| 368 |
: "cc", "memory", |
| 369 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 370 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 371 |
); |
| 372 |
} |
| 373 |
|
| 374 |
void lramp_sub2(float *dst, const float *src, float v1, float v2, size_t count) |
| 375 |
{ |
| 376 |
float delta = v2 - v1; |
| 377 |
if (delta == 0.0f) |
| 378 |
{ |
| 379 |
dsp::fmsub_k3(dst, src, v1, count); |
| 380 |
return; |
| 381 |
} |
| 382 |
if (count == 0) |
| 383 |
return; |
| 384 |
|
| 385 |
delta /= count; |
| 386 |
IF_ARCH_X86( size_t off ); |
| 387 |
ARCH_X86_ASM( |
| 388 |
LRAMP_OP3_CORE("dst", "dst", "src", "vsub", OP_DSEL) |
| 389 |
: [count] "+r" (count), [off] "=&r" (off) |
| 390 |
: [dst] "r" (dst), [src] "r" (src), |
| 391 |
[CC] "r" (lramp_const), |
| 392 |
[delta] "m" (delta), |
| 393 |
[v1] "m" (v1) |
| 394 |
: "cc", "memory", |
| 395 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 396 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 397 |
); |
| 398 |
} |
| 399 |
|
| 400 |
void lramp_rsub2(float *dst, const float *src, float v1, float v2, size_t count) |
| 401 |
{ |
| 402 |
float delta = v2 - v1; |
| 403 |
if (delta == 0.0f) |
| 404 |
{ |
| 405 |
dsp::fmrsub_k3(dst, src, v1, count); |
| 406 |
return; |
| 407 |
} |
| 408 |
if (count == 0) |
| 409 |
return; |
| 410 |
|
| 411 |
delta /= count; |
| 412 |
IF_ARCH_X86( size_t off ); |
| 413 |
ARCH_X86_ASM( |
| 414 |
LRAMP_OP3_CORE("dst", "dst", "src", "vsub", OP_RSEL) |
| 415 |
: [count] "+r" (count), [off] "=&r" (off) |
| 416 |
: [dst] "r" (dst), [src] "r" (src), |
| 417 |
[CC] "r" (lramp_const), |
| 418 |
[delta] "m" (delta), |
| 419 |
[v1] "m" (v1) |
| 420 |
: "cc", "memory", |
| 421 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 422 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 423 |
); |
| 424 |
} |
| 425 |
|
| 426 |
void lramp_mul2(float *dst, const float *src, float v1, float v2, size_t count) |
| 427 |
{ |
| 428 |
float delta = v2 - v1; |
| 429 |
if (delta == 0.0f) |
| 430 |
{ |
| 431 |
dsp::fmmul_k3(dst, src, v1, count); |
| 432 |
return; |
| 433 |
} |
| 434 |
if (count == 0) |
| 435 |
return; |
| 436 |
|
| 437 |
delta /= count; |
| 438 |
IF_ARCH_X86( size_t off ); |
| 439 |
ARCH_X86_ASM( |
| 440 |
LRAMP_OP3_CORE("dst", "dst", "src", "vmul", OP_DSEL) |
| 441 |
: [count] "+r" (count), [off] "=&r" (off) |
| 442 |
: [dst] "r" (dst), [src] "r" (src), |
| 443 |
[CC] "r" (lramp_const), |
| 444 |
[delta] "m" (delta), |
| 445 |
[v1] "m" (v1) |
| 446 |
: "cc", "memory", |
| 447 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 448 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 449 |
); |
| 450 |
} |
| 451 |
|
| 452 |
void lramp_div2(float *dst, const float *src, float v1, float v2, size_t count) |
| 453 |
{ |
| 454 |
float delta = v2 - v1; |
| 455 |
if (delta == 0.0f) |
| 456 |
{ |
| 457 |
dsp::fmdiv_k3(dst, src, v1, count); |
| 458 |
return; |
| 459 |
} |
| 460 |
if (count == 0) |
| 461 |
return; |
| 462 |
|
| 463 |
delta /= count; |
| 464 |
IF_ARCH_X86( size_t off ); |
| 465 |
ARCH_X86_ASM( |
| 466 |
LRAMP_OP3_CORE("dst", "dst", "src", "vdiv", OP_DSEL) |
| 467 |
: [count] "+r" (count), [off] "=&r" (off) |
| 468 |
: [dst] "r" (dst), [src] "r" (src), |
| 469 |
[CC] "r" (lramp_const), |
| 470 |
[delta] "m" (delta), |
| 471 |
[v1] "m" (v1) |
| 472 |
: "cc", "memory", |
| 473 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 474 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 475 |
); |
| 476 |
} |
| 477 |
|
| 478 |
void lramp_rdiv2(float *dst, const float *src, float v1, float v2, size_t count) |
| 479 |
{ |
| 480 |
float delta = v2 - v1; |
| 481 |
if (delta == 0.0f) |
| 482 |
{ |
| 483 |
dsp::fmrdiv_k3(dst, src, v1, count); |
| 484 |
return; |
| 485 |
} |
| 486 |
if (count == 0) |
| 487 |
return; |
| 488 |
|
| 489 |
delta /= count; |
| 490 |
IF_ARCH_X86( size_t off ); |
| 491 |
ARCH_X86_ASM( |
| 492 |
LRAMP_OP3_CORE("dst", "dst", "src", "vdiv", OP_RSEL) |
| 493 |
: [count] "+r" (count), [off] "=&r" (off) |
| 494 |
: [dst] "r" (dst), [src] "r" (src), |
| 495 |
[CC] "r" (lramp_const), |
| 496 |
[delta] "m" (delta), |
| 497 |
[v1] "m" (v1) |
| 498 |
: "cc", "memory", |
| 499 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 500 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 501 |
); |
| 502 |
} |
| 503 |
|
| 504 |
void lramp_add3(float *dst, const float *a, const float *b, float v1, float v2, size_t count) |
| 505 |
{ |
| 506 |
float delta = v2 - v1; |
| 507 |
if (delta == 0.0f) |
| 508 |
{ |
| 509 |
dsp::fmadd_k4(dst, a, b, v1, count); |
| 510 |
return; |
| 511 |
} |
| 512 |
if (count == 0) |
| 513 |
return; |
| 514 |
|
| 515 |
delta /= count; |
| 516 |
IF_ARCH_X86( size_t off ); |
| 517 |
ARCH_X86_ASM( |
| 518 |
LRAMP_OP3_CORE("dst", "a", "b", "vadd", OP_DSEL) |
| 519 |
: [off] "=&r" (off), |
| 520 |
__IF_32([count] "+g" (count)) |
| 521 |
__IF_64([count] "+r" (count)) |
| 522 |
: [dst] "r" (dst), [a] "r" (a), [b] "r" (b), |
| 523 |
[CC] "r" (lramp_const), |
| 524 |
[delta] "m" (delta), |
| 525 |
[v1] "m" (v1) |
| 526 |
: "cc", "memory", |
| 527 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 528 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 529 |
); |
| 530 |
} |
| 531 |
|
| 532 |
void lramp_sub3(float *dst, const float *a, const float *b, float v1, float v2, size_t count) |
| 533 |
{ |
| 534 |
float delta = v2 - v1; |
| 535 |
if (delta == 0.0f) |
| 536 |
{ |
| 537 |
dsp::fmsub_k4(dst, a, b, v1, count); |
| 538 |
return; |
| 539 |
} |
| 540 |
if (count == 0) |
| 541 |
return; |
| 542 |
|
| 543 |
delta /= count; |
| 544 |
IF_ARCH_X86( size_t off ); |
| 545 |
ARCH_X86_ASM( |
| 546 |
LRAMP_OP3_CORE("dst", "a", "b", "vsub", OP_DSEL) |
| 547 |
: [off] "=&r" (off), |
| 548 |
__IF_32([count] "+g" (count)) |
| 549 |
__IF_64([count] "+r" (count)) |
| 550 |
: [dst] "r" (dst), [a] "r" (a), [b] "r" (b), |
| 551 |
[CC] "r" (lramp_const), |
| 552 |
[delta] "m" (delta), |
| 553 |
[v1] "m" (v1) |
| 554 |
: "cc", "memory", |
| 555 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 556 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 557 |
); |
| 558 |
} |
| 559 |
|
| 560 |
void lramp_rsub3(float *dst, const float *a, const float *b, float v1, float v2, size_t count) |
| 561 |
{ |
| 562 |
float delta = v2 - v1; |
| 563 |
if (delta == 0.0f) |
| 564 |
{ |
| 565 |
dsp::fmrsub_k4(dst, a, b, v1, count); |
| 566 |
return; |
| 567 |
} |
| 568 |
if (count == 0) |
| 569 |
return; |
| 570 |
|
| 571 |
delta /= count; |
| 572 |
IF_ARCH_X86( size_t off ); |
| 573 |
ARCH_X86_ASM( |
| 574 |
LRAMP_OP3_CORE("dst", "a", "b", "vsub", OP_RSEL) |
| 575 |
: [off] "=&r" (off), |
| 576 |
__IF_32([count] "+g" (count)) |
| 577 |
__IF_64([count] "+r" (count)) |
| 578 |
: [dst] "r" (dst), [a] "r" (a), [b] "r" (b), |
| 579 |
[CC] "r" (lramp_const), |
| 580 |
[delta] "m" (delta), |
| 581 |
[v1] "m" (v1) |
| 582 |
: "cc", "memory", |
| 583 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 584 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 585 |
); |
| 586 |
} |
| 587 |
|
| 588 |
void lramp_mul3(float *dst, const float *a, const float *b, float v1, float v2, size_t count) |
| 589 |
{ |
| 590 |
float delta = v2 - v1; |
| 591 |
if (delta == 0.0f) |
| 592 |
{ |
| 593 |
dsp::fmmul_k4(dst, a, b, v1, count); |
| 594 |
return; |
| 595 |
} |
| 596 |
if (count == 0) |
| 597 |
return; |
| 598 |
|
| 599 |
delta /= count; |
| 600 |
IF_ARCH_X86( size_t off ); |
| 601 |
ARCH_X86_ASM( |
| 602 |
LRAMP_OP3_CORE("dst", "a", "b", "vmul", OP_DSEL) |
| 603 |
: [off] "=&r" (off), |
| 604 |
__IF_32([count] "+g" (count)) |
| 605 |
__IF_64([count] "+r" (count)) |
| 606 |
: [dst] "r" (dst), [a] "r" (a), [b] "r" (b), |
| 607 |
[CC] "r" (lramp_const), |
| 608 |
[delta] "m" (delta), |
| 609 |
[v1] "m" (v1) |
| 610 |
: "cc", "memory", |
| 611 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 612 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 613 |
); |
| 614 |
} |
| 615 |
|
| 616 |
void lramp_div3(float *dst, const float *a, const float *b, float v1, float v2, size_t count) |
| 617 |
{ |
| 618 |
float delta = v2 - v1; |
| 619 |
if (delta == 0.0f) |
| 620 |
{ |
| 621 |
dsp::fmdiv_k4(dst, a, b, v1, count); |
| 622 |
return; |
| 623 |
} |
| 624 |
if (count == 0) |
| 625 |
return; |
| 626 |
|
| 627 |
delta /= count; |
| 628 |
IF_ARCH_X86( size_t off ); |
| 629 |
ARCH_X86_ASM( |
| 630 |
LRAMP_OP3_CORE("dst", "a", "b", "vdiv", OP_DSEL) |
| 631 |
: [off] "=&r" (off), |
| 632 |
__IF_32([count] "+g" (count)) |
| 633 |
__IF_64([count] "+r" (count)) |
| 634 |
: [dst] "r" (dst), [a] "r" (a), [b] "r" (b), |
| 635 |
[CC] "r" (lramp_const), |
| 636 |
[delta] "m" (delta), |
| 637 |
[v1] "m" (v1) |
| 638 |
: "cc", "memory", |
| 639 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 640 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 641 |
); |
| 642 |
} |
| 643 |
|
| 644 |
void lramp_rdiv3(float *dst, const float *a, const float *b, float v1, float v2, size_t count) |
| 645 |
{ |
| 646 |
float delta = v2 - v1; |
| 647 |
if (delta == 0.0f) |
| 648 |
{ |
| 649 |
dsp::fmrdiv_k4(dst, a, b, v1, count); |
| 650 |
return; |
| 651 |
} |
| 652 |
if (count == 0) |
| 653 |
return; |
| 654 |
|
| 655 |
delta /= count; |
| 656 |
IF_ARCH_X86( size_t off ); |
| 657 |
ARCH_X86_ASM( |
| 658 |
LRAMP_OP3_CORE("dst", "a", "b", "vdiv", OP_RSEL) |
| 659 |
: [off] "=&r" (off), |
| 660 |
__IF_32([count] "+g" (count)) |
| 661 |
__IF_64([count] "+r" (count)) |
| 662 |
: [dst] "r" (dst), [a] "r" (a), [b] "r" (b), |
| 663 |
[CC] "r" (lramp_const), |
| 664 |
[delta] "m" (delta), |
| 665 |
[v1] "m" (v1) |
| 666 |
: "cc", "memory", |
| 667 |
"xmm0", "xmm1", "xmm2", "xmm3", |
| 668 |
"xmm4", "xmm5", "xmm6", "xmm7" |
| 669 |
); |
| 670 |
} |
| 671 |
|
| 672 |
#undef LRAMP_OP3_CORE |
| 673 |
#undef OP_DSEL |
| 674 |
#undef OP_RSEL |
| 675 |
|
| 676 |
} /* namespace avx512 */ |
| 677 |
} /* namespace lsp */ |
| 678 |
|
| 679 |
|
| 680 |
|
| 681 |
|
| 682 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX512_IMPL */ |