| 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: 31 мар. 2020 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_SSE3_PCOMPLEX_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_SSE3_PCOMPLEX_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_SSE3_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE3_IMPL */ |
| 28 |
|
| 29 |
namespace lsp |
| 30 |
{ |
| 31 |
namespace sse3 |
| 32 |
{ |
| 33 |
IF_ARCH_X86( |
| 34 |
static const uint32_t pcomplex_div_const[] __lsp_aligned16 = |
| 35 |
{ |
| 36 |
LSP_DSP_VEC4(0x3f800000), |
| 37 |
LSP_DSP_VEC4(0x80000000) |
| 38 |
}; |
| 39 |
); |
| 40 |
|
| 41 |
#define PCOMPLEX_MUL_CORE(DST, T, B) \ |
| 42 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 43 |
/* x4 blocks */ \ |
| 44 |
__ASM_EMIT("sub $4, %[count]") \ |
| 45 |
__ASM_EMIT("jb 2f") \ |
| 46 |
__ASM_EMIT("1:") \ |
| 47 |
__ASM_EMIT("lddqu 0x00(%[" T "], %[off]), %%xmm0") /* xmm0 = ar0 ai0 ar1 ai1 */ \ |
| 48 |
__ASM_EMIT("lddqu 0x10(%[" T "], %[off]), %%xmm4") /* xmm4 = ar2 ai2 ar3 ai3 */ \ |
| 49 |
__ASM_EMIT("lddqu 0x00(%[" B "], %[off]), %%xmm1") /* xmm1 = br0 bi0 br1 bi1 */ \ |
| 50 |
__ASM_EMIT("lddqu 0x10(%[" B "], %[off]), %%xmm5") /* xmm5 = br2 bi2 br3 bi3 */ \ |
| 51 |
__ASM_EMIT("movsldup %%xmm0, %%xmm2") /* xmm2 = ar0 ar0 ar1 ar1 */ \ |
| 52 |
__ASM_EMIT("movsldup %%xmm4, %%xmm6") /* xmm6 = ar2 ar2 ar3 ar3 */ \ |
| 53 |
__ASM_EMIT("movshdup %%xmm0, %%xmm0") /* xmm0 = ai0 ai0 ai1 ai1 */ \ |
| 54 |
__ASM_EMIT("movshdup %%xmm4, %%xmm4") /* xmm4 = ai2 ai2 ai3 ai3 */ \ |
| 55 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = ai0*br0 ai0*bi0 ai1*br1 ai1*bi1 */ \ |
| 56 |
__ASM_EMIT("mulps %%xmm5, %%xmm4") /* xmm4 = ai2*br2 ai2*bi2 ai3*br3 ai3*bi3 */ \ |
| 57 |
__ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = ar0*br0 ar0*bi0 ar1*br1 ar1*bi1 */ \ |
| 58 |
__ASM_EMIT("mulps %%xmm5, %%xmm6") /* xmm6 = ar2*br2 ar2*bi2 ar3*br3 ar3*bi3 */ \ |
| 59 |
__ASM_EMIT("shufps $0xb1, %%xmm0, %%xmm0") /* xmm0 = ai0*bi0 ai0*br0 ai1*bi1 ai1*br1 */ \ |
| 60 |
__ASM_EMIT("shufps $0xb1, %%xmm4, %%xmm4") /* xmm4 = ai2*bi2 ai2*br2 ai3*bi3 ai3*br3 */ \ |
| 61 |
__ASM_EMIT("addsubps %%xmm0, %%xmm2") /* xmm2 = ar0*br0 - ai0*bi0 ar0*bi0 + ai0*br0 ar1*br1 - ai1*bi1 ar1*bi1 + ai1*br1 */ \ |
| 62 |
__ASM_EMIT("addsubps %%xmm4, %%xmm6") /* xmm6 = ar2*br2 - ai2*bi2 ar2*bi2 + ai2*br2 ar3*br3 - ai3*bi3 ar3*bi3 + ai3*br3 */ \ |
| 63 |
__ASM_EMIT("movdqu %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 64 |
__ASM_EMIT("movdqu %%xmm6, 0x10(%[" DST "], %[off])") \ |
| 65 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 66 |
__ASM_EMIT("sub $4, %[count]") \ |
| 67 |
__ASM_EMIT("jae 1b") \ |
| 68 |
__ASM_EMIT("2:") \ |
| 69 |
/* Process scalar data */ \ |
| 70 |
__ASM_EMIT("add $3, %[count]") \ |
| 71 |
__ASM_EMIT("jl 4f") \ |
| 72 |
__ASM_EMIT("3:") \ |
| 73 |
__ASM_EMIT("movlps 0x00(%[" T "], %[off]), %%xmm0") /* xmm0 = ar0 ai0 ar1 ai1 */ \ |
| 74 |
__ASM_EMIT("movlps 0x00(%[" B "], %[off]), %%xmm1") /* xmm1 = br0 bi0 br1 bi1 */ \ |
| 75 |
__ASM_EMIT("movsldup %%xmm0, %%xmm2") /* xmm2 = ar0 ar0 ar1 ar1 */ \ |
| 76 |
__ASM_EMIT("movshdup %%xmm0, %%xmm0") /* xmm0 = ai0 ai0 ai1 ai1 */ \ |
| 77 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = ai0*br0 ai0*bi0 ai1*br1 ai1*bi1 */ \ |
| 78 |
__ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = ar0*br0 ar0*bi0 ar1*br1 ar1*bi1 */ \ |
| 79 |
__ASM_EMIT("shufps $0xb1, %%xmm0, %%xmm0") /* xmm0 = ai0*bi0 ai0*br0 ai1*bi1 ai1*br1 */ \ |
| 80 |
__ASM_EMIT("addsubps %%xmm0, %%xmm2") /* xmm2 = ar0*br0 - ai0*bi0 ar0*bi0 + ai0*br0 ar1*br1 - ai1*bi1 ar1*bi1 + ai1*br1 */ \ |
| 81 |
__ASM_EMIT("movlps %%xmm2, 0x00(%[" DST "], %[off])") \ |
| 82 |
__ASM_EMIT("add $0x08, %[off]") \ |
| 83 |
__ASM_EMIT("dec %[count]") \ |
| 84 |
__ASM_EMIT("jge 3b") \ |
| 85 |
__ASM_EMIT("4:") |
| 86 |
|
| 87 |
void pcomplex_mul3(float *dst, const float *src1, const float *src2, size_t count) |
| 88 |
{ |
| 89 |
IF_ARCH_X86(size_t off); |
| 90 |
|
| 91 |
ARCH_X86_ASM |
| 92 |
( |
| 93 |
PCOMPLEX_MUL_CORE("dst", "src1", "src2") |
| 94 |
: [off] "=&r" (off), |
| 95 |
[count] "+r" (count) |
| 96 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2) |
| 97 |
: "cc", "memory", |
| 98 |
"%xmm0", "%xmm1", "%xmm2", "%xmm4", |
| 99 |
"%xmm5", "%xmm6" |
| 100 |
); |
| 101 |
} |
| 102 |
|
| 103 |
void pcomplex_mul2(float *dst, const float *src, size_t count) |
| 104 |
{ |
| 105 |
IF_ARCH_X86(size_t off); |
| 106 |
|
| 107 |
ARCH_X86_ASM |
| 108 |
( |
| 109 |
PCOMPLEX_MUL_CORE("dst", "dst", "src") |
| 110 |
: [off] "=&r" (off), |
| 111 |
[count] "+r" (count) |
| 112 |
: [dst] "r" (dst), [src] "r" (src) |
| 113 |
: "cc", "memory", |
| 114 |
"%xmm0", "%xmm1", "%xmm2", "%xmm4", |
| 115 |
"%xmm5", "%xmm6" |
| 116 |
); |
| 117 |
} |
| 118 |
|
| 119 |
#undef PCOMPLEX_MUL_CORE |
| 120 |
|
| 121 |
#define PCOMPLEX_DIV_CORE(DST, T, B) \ |
| 122 |
__ASM_EMIT("xor %[off], %[off]") \ |
| 123 |
/* x4 blocks */ \ |
| 124 |
__ASM_EMIT("sub $4, %[count]") \ |
| 125 |
__ASM_EMIT("jb 2f") \ |
| 126 |
__ASM_EMIT("1:") \ |
| 127 |
__ASM_EMIT("lddqu 0x00(%[" T "], %[off]), %%xmm0") /* xmm0 = tr0 ti0 tr1 ti1 */ \ |
| 128 |
__ASM_EMIT("lddqu 0x10(%[" T "], %[off]), %%xmm4") \ |
| 129 |
__ASM_EMIT("lddqu 0x00(%[" B "], %[off]), %%xmm1") /* xmm1 = br0 bi0 br1 bi1 */ \ |
| 130 |
__ASM_EMIT("lddqu 0x10(%[" B "], %[off]), %%xmm5") \ |
| 131 |
__ASM_EMIT("movsldup %%xmm0, %%xmm2") /* xmm2 = tr0 tr0 tr1 tr1 */ \ |
| 132 |
__ASM_EMIT("movsldup %%xmm4, %%xmm6") \ |
| 133 |
__ASM_EMIT("movshdup %%xmm0, %%xmm0") /* xmm0 = ti0 ti0 ti1 ti1 */ \ |
| 134 |
__ASM_EMIT("movshdup %%xmm4, %%xmm4") \ |
| 135 |
__ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = tr0*br0 tr0*bi0 tr1*br1 tr1*bi1 */ \ |
| 136 |
__ASM_EMIT("mulps %%xmm5, %%xmm6") \ |
| 137 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = ti0*br0 ti0*bi0 ti1*br1 ti1*bi1 */ \ |
| 138 |
__ASM_EMIT("mulps %%xmm5, %%xmm4") \ |
| 139 |
__ASM_EMIT("xorps 0x10 + %[CC], %%xmm2") /* xmm2 = -tr0*br0 -tr0*bi0 -tr1*br1 -tr1*bi1 */ \ |
| 140 |
__ASM_EMIT("xorps 0x10 + %[CC], %%xmm6") \ |
| 141 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") /* xmm1 = br0*br0 bi0*bi0 br1*br1 bi1*bi1 */ \ |
| 142 |
__ASM_EMIT("mulps %%xmm5, %%xmm5") \ |
| 143 |
__ASM_EMIT("shufps $0xb1, %%xmm0, %%xmm0") /* xmm0 = ti0*bi0 ti0*br0 ti1*bi1 ti1*br1 */ \ |
| 144 |
__ASM_EMIT("shufps $0xb1, %%xmm4, %%xmm4") \ |
| 145 |
__ASM_EMIT("movaps 0x00 + %[CC], %%xmm3") /* xmm3 = 1 */ \ |
| 146 |
__ASM_EMIT("haddps %%xmm5, %%xmm1") /* xmm1 = R0 R1 R2 R3 = br0*br0+bi0*bi0 br1*br1+bi1*bi1 */ \ |
| 147 |
__ASM_EMIT("divps %%xmm1, %%xmm3") /* xmm3 = 1/R0 1/R2 1/R2 1/R3 */ \ |
| 148 |
__ASM_EMIT("addsubps %%xmm2, %%xmm0") /* xmm0 = r0 i0 r1 i1 = ti0*bi0+tr0*br0 ti0*br0-tr0*bi0 ti1*bi1+tr1*br1 ti1*br1-tr1*bi1 */ \ |
| 149 |
__ASM_EMIT("addsubps %%xmm6, %%xmm4") \ |
| 150 |
__ASM_EMIT("movaps %%xmm3, %%xmm7") /* xmm7 = 1/R0 1/R2 1/R2 1/R3 */ \ |
| 151 |
__ASM_EMIT("unpcklps %%xmm3, %%xmm3") /* xmm3 = 1/R0 1/R0 1/R2 1/R2 */ \ |
| 152 |
__ASM_EMIT("unpckhps %%xmm7, %%xmm7") /* xmm7 = 1/R2 1/R2 1/R3 1/R3 */ \ |
| 153 |
__ASM_EMIT("mulps %%xmm3, %%xmm0") /* xmm0 = r0' i0' r1' i1' */ \ |
| 154 |
__ASM_EMIT("mulps %%xmm7, %%xmm4") \ |
| 155 |
__ASM_EMIT("movdqu %%xmm0, 0x00(%[" DST "], %[off])") \ |
| 156 |
__ASM_EMIT("movdqu %%xmm4, 0x10(%[" DST "], %[off])") \ |
| 157 |
__ASM_EMIT("add $0x20, %[off]") \ |
| 158 |
__ASM_EMIT("sub $4, %[count]") \ |
| 159 |
__ASM_EMIT("jae 1b") \ |
| 160 |
__ASM_EMIT("2:") \ |
| 161 |
/* Process scalar data */ \ |
| 162 |
__ASM_EMIT("add $3, %[count]") \ |
| 163 |
__ASM_EMIT("jl 4f") \ |
| 164 |
__ASM_EMIT("3:") \ |
| 165 |
__ASM_EMIT("movlps 0x00(%[" T "], %[off]), %%xmm0") /* xmm0 = tr ti */ \ |
| 166 |
__ASM_EMIT("movlps 0x00(%[" B "], %[off]), %%xmm1") /* xmm1 = br bi */ \ |
| 167 |
__ASM_EMIT("movsldup %%xmm0, %%xmm2") /* xmm2 = tr tr */ \ |
| 168 |
__ASM_EMIT("movshdup %%xmm0, %%xmm0") /* xmm0 = ti ti */ \ |
| 169 |
__ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = tr*br tr*bi */ \ |
| 170 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = ti*br ti*bi */ \ |
| 171 |
__ASM_EMIT("xorps 0x10 + %[CC], %%xmm2") /* xmm2 = -tr*br -tr*bi */ \ |
| 172 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") /* xmm1 = br*br bi*bi */ \ |
| 173 |
__ASM_EMIT("shufps $0xb1, %%xmm0, %%xmm0") /* xmm0 = ti*bi ti*br */ \ |
| 174 |
__ASM_EMIT("haddps %%xmm1, %%xmm1") /* xmm1 = R = br*br+bi*bi */ \ |
| 175 |
__ASM_EMIT("addsubps %%xmm2, %%xmm0") /* xmm0 = r i = ti*bi+tr*br ti*br-tr*bi */ \ |
| 176 |
__ASM_EMIT("unpcklps %%xmm1, %%xmm1") /* xmm1 = R R */ \ |
| 177 |
__ASM_EMIT("divps %%xmm1, %%xmm0") /* xmm0 = r' i' */ \ |
| 178 |
__ASM_EMIT("movlps %%xmm0, 0x00(%[" DST "], %[off])") \ |
| 179 |
__ASM_EMIT("add $0x08, %[off]") \ |
| 180 |
__ASM_EMIT("dec %[count]") \ |
| 181 |
__ASM_EMIT("jge 3b") \ |
| 182 |
__ASM_EMIT("4:") |
| 183 |
|
| 184 |
void pcomplex_div3(float *dst, const float *src1, const float *src2, size_t count) |
| 185 |
{ |
| 186 |
size_t off; |
| 187 |
|
| 188 |
ARCH_X86_ASM |
| 189 |
( |
| 190 |
PCOMPLEX_DIV_CORE("dst", "src1", "src2") |
| 191 |
: [off] "=&r" (off), |
| 192 |
[count] "+r" (count) |
| 193 |
: [dst] "r" (dst), [src1] "r" (src1), [src2] "r" (src2), |
| 194 |
[CC] "o" (pcomplex_div_const) |
| 195 |
: "cc", "memory", |
| 196 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 197 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 198 |
); |
| 199 |
} |
| 200 |
|
| 201 |
void pcomplex_div2(float *dst, const float *src, size_t count) |
| 202 |
{ |
| 203 |
IF_ARCH_X86(size_t off); |
| 204 |
|
| 205 |
ARCH_X86_ASM |
| 206 |
( |
| 207 |
PCOMPLEX_DIV_CORE("dst", "dst", "src") |
| 208 |
: [off] "=&r" (off), |
| 209 |
[count] "+r" (count) |
| 210 |
: [dst] "r" (dst), [src] "r" (src), |
| 211 |
[CC] "o" (pcomplex_div_const) |
| 212 |
: "cc", "memory", |
| 213 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 214 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 215 |
); |
| 216 |
} |
| 217 |
|
| 218 |
void pcomplex_rdiv2(float *dst, const float *src, size_t count) |
| 219 |
{ |
| 220 |
IF_ARCH_X86(size_t off); |
| 221 |
|
| 222 |
ARCH_X86_ASM |
| 223 |
( |
| 224 |
PCOMPLEX_DIV_CORE("dst", "src", "dst") |
| 225 |
: [off] "=&r" (off), |
| 226 |
[count] "+r" (count) |
| 227 |
: [dst] "r" (dst), [src] "r" (src), |
| 228 |
[CC] "o" (pcomplex_div_const) |
| 229 |
: "cc", "memory", |
| 230 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 231 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 232 |
); |
| 233 |
} |
| 234 |
|
| 235 |
#undef PCOMPLEX_DIV_CORE |
| 236 |
|
| 237 |
void pcomplex_mod(float *dst, const float *src, size_t count) |
| 238 |
{ |
| 239 |
IF_ARCH_X86(size_t off); |
| 240 |
|
| 241 |
ARCH_X86_ASM |
| 242 |
( |
| 243 |
__ASM_EMIT("xor %[off], %[off]") |
| 244 |
__ASM_EMIT("sub $16, %[count]") |
| 245 |
__ASM_EMIT("jb 2f") |
| 246 |
|
| 247 |
// 16x blocks |
| 248 |
__ASM_EMIT("1:") |
| 249 |
__ASM_EMIT("lddqu 0x00(%[src], %[off], 2), %%xmm0") |
| 250 |
__ASM_EMIT("lddqu 0x10(%[src], %[off], 2), %%xmm1") |
| 251 |
__ASM_EMIT("lddqu 0x20(%[src], %[off], 2), %%xmm2") |
| 252 |
__ASM_EMIT("lddqu 0x30(%[src], %[off], 2), %%xmm3") |
| 253 |
__ASM_EMIT("lddqu 0x40(%[src], %[off], 2), %%xmm4") |
| 254 |
__ASM_EMIT("lddqu 0x50(%[src], %[off], 2), %%xmm5") |
| 255 |
__ASM_EMIT("lddqu 0x60(%[src], %[off], 2), %%xmm6") |
| 256 |
__ASM_EMIT("lddqu 0x70(%[src], %[off], 2), %%xmm7") |
| 257 |
|
| 258 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") |
| 259 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") |
| 260 |
__ASM_EMIT("mulps %%xmm2, %%xmm2") |
| 261 |
__ASM_EMIT("mulps %%xmm3, %%xmm3") |
| 262 |
__ASM_EMIT("mulps %%xmm4, %%xmm4") |
| 263 |
__ASM_EMIT("mulps %%xmm5, %%xmm5") |
| 264 |
__ASM_EMIT("mulps %%xmm6, %%xmm6") |
| 265 |
__ASM_EMIT("mulps %%xmm7, %%xmm7") |
| 266 |
__ASM_EMIT("haddps %%xmm1, %%xmm0") |
| 267 |
__ASM_EMIT("haddps %%xmm3, %%xmm2") |
| 268 |
__ASM_EMIT("haddps %%xmm5, %%xmm4") |
| 269 |
__ASM_EMIT("haddps %%xmm7, %%xmm6") |
| 270 |
__ASM_EMIT("sqrtps %%xmm0, %%xmm0") |
| 271 |
__ASM_EMIT("sqrtps %%xmm2, %%xmm2") |
| 272 |
__ASM_EMIT("sqrtps %%xmm4, %%xmm4") |
| 273 |
__ASM_EMIT("sqrtps %%xmm6, %%xmm6") |
| 274 |
|
| 275 |
__ASM_EMIT("movdqu %%xmm0, 0x00(%[dst], %[off])") |
| 276 |
__ASM_EMIT("movdqu %%xmm2, 0x10(%[dst], %[off])") |
| 277 |
__ASM_EMIT("movdqu %%xmm4, 0x20(%[dst], %[off])") |
| 278 |
__ASM_EMIT("movdqu %%xmm6, 0x30(%[dst], %[off])") |
| 279 |
__ASM_EMIT("add $0x40, %[off]") |
| 280 |
__ASM_EMIT("sub $16, %[count]") |
| 281 |
__ASM_EMIT("jae 1b") |
| 282 |
|
| 283 |
// 4x blocks |
| 284 |
__ASM_EMIT("2:") |
| 285 |
__ASM_EMIT("add $12, %[count]") |
| 286 |
__ASM_EMIT("jl 4f") |
| 287 |
__ASM_EMIT("3:") |
| 288 |
__ASM_EMIT("lddqu 0x00(%[src], %[off], 2), %%xmm0") |
| 289 |
__ASM_EMIT("lddqu 0x10(%[src], %[off], 2), %%xmm1") |
| 290 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") |
| 291 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") |
| 292 |
__ASM_EMIT("haddps %%xmm1, %%xmm0") |
| 293 |
__ASM_EMIT("sqrtps %%xmm0, %%xmm0") |
| 294 |
__ASM_EMIT("movdqu %%xmm0, 0x00(%[dst], %[off])") |
| 295 |
__ASM_EMIT("add $0x10, %[off]") |
| 296 |
__ASM_EMIT("sub $4, %[count]") |
| 297 |
__ASM_EMIT("jae 3b") |
| 298 |
|
| 299 |
// 1x blocks |
| 300 |
__ASM_EMIT("4:") |
| 301 |
__ASM_EMIT("add $3, %[count]") |
| 302 |
__ASM_EMIT("jl 6f") |
| 303 |
__ASM_EMIT("5:") |
| 304 |
__ASM_EMIT("movlps 0x00(%[src], %[off], 2), %%xmm0") |
| 305 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") |
| 306 |
__ASM_EMIT("haddps %%xmm0, %%xmm0") |
| 307 |
__ASM_EMIT("sqrtss %%xmm0, %%xmm0") |
| 308 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst], %[off])") |
| 309 |
__ASM_EMIT("add $0x4, %[off]") |
| 310 |
__ASM_EMIT("dec %[count]") |
| 311 |
__ASM_EMIT("jge 5b") |
| 312 |
|
| 313 |
// End |
| 314 |
__ASM_EMIT("6:") |
| 315 |
: [dst] "+r" (dst), [src] "+r" (src), |
| 316 |
[count] "+r" (count), [off] "=&r" (off) |
| 317 |
: |
| 318 |
: "cc", "memory", |
| 319 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 320 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 321 |
); |
| 322 |
} |
| 323 |
|
| 324 |
void x64_pcomplex_mul3(float *dst, const float *src1, const float *src2, size_t count) |
| 325 |
{ |
| 326 |
IF_ARCH_X86_64(size_t off); |
| 327 |
|
| 328 |
ARCH_X86_64_ASM |
| 329 |
( |
| 330 |
/* Do block processing */ |
| 331 |
__ASM_EMIT("xor %[off], %[off]") |
| 332 |
__ASM_EMIT("sub $8, %[count]") |
| 333 |
__ASM_EMIT("jb 3f") |
| 334 |
|
| 335 |
__ASM_EMIT(".align 16") |
| 336 |
__ASM_EMIT("2:") |
| 337 |
/* Process vectorized data */ |
| 338 |
__ASM_EMIT("lddqu 0x00(%[src1], %[off]), %%xmm0") /* xmm0 = ar0 ai0 ar1 ai1 */ |
| 339 |
__ASM_EMIT("lddqu 0x10(%[src1], %[off]), %%xmm4") /* xmm4 = ar2 ai2 ar3 ai3 */ |
| 340 |
__ASM_EMIT("lddqu 0x20(%[src1], %[off]), %%xmm8") |
| 341 |
__ASM_EMIT("lddqu 0x30(%[src1], %[off]), %%xmm12") |
| 342 |
__ASM_EMIT("lddqu 0x00(%[src2], %[off]), %%xmm1") /* xmm1 = br0 bi0 br1 bi1 */ |
| 343 |
__ASM_EMIT("lddqu 0x10(%[src2], %[off]), %%xmm5") /* xmm5 = br2 bi2 br3 bi3 */ |
| 344 |
__ASM_EMIT("lddqu 0x20(%[src2], %[off]), %%xmm9") |
| 345 |
__ASM_EMIT("lddqu 0x30(%[src2], %[off]), %%xmm13") |
| 346 |
/* Calc multiplication */ |
| 347 |
__ASM_EMIT("movsldup %%xmm0, %%xmm2") /* xmm2 = ar0 ar0 ar1, ar1 */ |
| 348 |
__ASM_EMIT("movsldup %%xmm4, %%xmm6") /* xmm6 = ar2 ar2 ar3, ar3 */ |
| 349 |
__ASM_EMIT("movsldup %%xmm8, %%xmm10") |
| 350 |
__ASM_EMIT("movsldup %%xmm12, %%xmm14") |
| 351 |
__ASM_EMIT("movshdup %%xmm0, %%xmm0") /* xmm0 = ai0 ai0 ai1, ai1 */ |
| 352 |
__ASM_EMIT("movshdup %%xmm4, %%xmm4") /* xmm4 = ai2 ai2 ai3, ai3 */ |
| 353 |
__ASM_EMIT("movshdup %%xmm8, %%xmm8") |
| 354 |
__ASM_EMIT("movshdup %%xmm12, %%xmm12") |
| 355 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = ai0*br0 ai0*bi0 ai1*br1 ai1*bi1 */ |
| 356 |
__ASM_EMIT("mulps %%xmm5, %%xmm4") /* xmm4 = ai2*br2 ai2*bi2 ai3*br3 ai3*bi3 */ |
| 357 |
__ASM_EMIT("mulps %%xmm9, %%xmm8") |
| 358 |
__ASM_EMIT("mulps %%xmm13, %%xmm12") |
| 359 |
__ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = ar0*br0 ar0*bi0 ar1*br1 ar1*bi1 */ |
| 360 |
__ASM_EMIT("mulps %%xmm5, %%xmm6") /* xmm6 = ar2*br2 ar2*bi2 ar3*br3 ar3*bi3 */ |
| 361 |
__ASM_EMIT("mulps %%xmm9, %%xmm10") |
| 362 |
__ASM_EMIT("mulps %%xmm13, %%xmm14") |
| 363 |
__ASM_EMIT("shufps $0xb1, %%xmm0, %%xmm0") /* xmm0 = ai0*bi0 ai0*br0 ai1*bi1 ai1*br1 */ |
| 364 |
__ASM_EMIT("shufps $0xb1, %%xmm4, %%xmm4") /* xmm4 = ai2*bi2 ai2*br2 ai3*bi3 ai3*br3 */ |
| 365 |
__ASM_EMIT("shufps $0xb1, %%xmm8, %%xmm8") |
| 366 |
__ASM_EMIT("shufps $0xb1, %%xmm12, %%xmm12") |
| 367 |
__ASM_EMIT("addsubps %%xmm0, %%xmm2") /* xmm2 = ar0*br0 - ai0*bi0 ar0*bi0 + ai0*br0 ar1*br1 - ai1*bi1 ar1*bi1 + ai1*br1 */ |
| 368 |
__ASM_EMIT("addsubps %%xmm4, %%xmm6") /* xmm6 = ar2*br2 - ai2*bi2 ar2*bi2 + ai2*br2 ar3*br3 - ai3*bi3 ar3*bi3 + ai3*br3 */ |
| 369 |
__ASM_EMIT("addsubps %%xmm8, %%xmm10") |
| 370 |
__ASM_EMIT("addsubps %%xmm12, %%xmm14") |
| 371 |
/* Store */ |
| 372 |
__ASM_EMIT("movdqu %%xmm2, 0x00(%[dst], %[off])") |
| 373 |
__ASM_EMIT("movdqu %%xmm6, 0x10(%[dst], %[off])") |
| 374 |
__ASM_EMIT("movdqu %%xmm10, 0x20(%[dst], %[off])") |
| 375 |
__ASM_EMIT("movdqu %%xmm14, 0x30(%[dst], %[off])") |
| 376 |
/* Repeat loop */ |
| 377 |
__ASM_EMIT("add $0x40, %[off]") |
| 378 |
__ASM_EMIT("sub $8, %[count]") |
| 379 |
__ASM_EMIT("jae 2b") |
| 380 |
|
| 381 |
/* 4-element block */ |
| 382 |
__ASM_EMIT("3:") |
| 383 |
__ASM_EMIT("add $4, %[count]") |
| 384 |
__ASM_EMIT("jl 4f") |
| 385 |
/* Process vectorized data */ |
| 386 |
__ASM_EMIT("lddqu 0x00(%[src1], %[off]), %%xmm0") /* xmm0 = ar0 ai0 ar1 ai1 */ |
| 387 |
__ASM_EMIT("lddqu 0x10(%[src1], %[off]), %%xmm4") /* xmm4 = ar2 ai2 ar3 ai3 */ |
| 388 |
__ASM_EMIT("lddqu 0x00(%[src2], %[off]), %%xmm1") /* xmm1 = br0 bi0 br1 bi1 */ |
| 389 |
__ASM_EMIT("lddqu 0x10(%[src2], %[off]), %%xmm5") /* xmm5 = br2 bi2 br3 bi3 */ |
| 390 |
/* Calc multiplication */ |
| 391 |
__ASM_EMIT("movsldup %%xmm0, %%xmm2") /* xmm2 = ar0 ar0 ar1, ar1 */ |
| 392 |
__ASM_EMIT("movsldup %%xmm4, %%xmm6") /* xmm6 = ar2 ar2 ar3, ar3 */ |
| 393 |
__ASM_EMIT("movshdup %%xmm0, %%xmm0") /* xmm0 = ai0 ai0 ai1, ai1 */ |
| 394 |
__ASM_EMIT("movshdup %%xmm4, %%xmm4") /* xmm4 = ai2 ai2 ai3, ai3 */ |
| 395 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = ai0*br0 ai0*bi0 ai1*br1 ai1*bi1 */ |
| 396 |
__ASM_EMIT("mulps %%xmm5, %%xmm4") /* xmm4 = ai2*br2 ai2*bi2 ai3*br3 ai3*bi3 */ |
| 397 |
__ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = ar0*br0 ar0*bi0 ar1*br1 ar1*bi1 */ |
| 398 |
__ASM_EMIT("mulps %%xmm5, %%xmm6") /* xmm6 = ar2*br2 ar2*bi2 ar3*br3 ar3*bi3 */ |
| 399 |
__ASM_EMIT("shufps $0xb1, %%xmm0, %%xmm0") /* xmm0 = ai0*bi0 ai0*br0 ai1*bi1 ai1*br1 */ |
| 400 |
__ASM_EMIT("shufps $0xb1, %%xmm4, %%xmm4") /* xmm4 = ai2*bi2 ai2*br2 ai3*bi3 ai3*br3 */ |
| 401 |
__ASM_EMIT("addsubps %%xmm0, %%xmm2") /* xmm2 = ar0*br0 - ai0*bi0 ar0*bi0 + ai0*br0 ar1*br1 - ai1*bi1 ar1*bi1 + ai1*br1 */ |
| 402 |
__ASM_EMIT("addsubps %%xmm4, %%xmm6") /* xmm6 = ar2*br2 - ai2*bi2 ar2*bi2 + ai2*br2 ar3*br3 - ai3*bi3 ar3*bi3 + ai3*br3 */ |
| 403 |
/* Store */ |
| 404 |
__ASM_EMIT("movdqu %%xmm2, 0x00(%[dst], %[off])") |
| 405 |
__ASM_EMIT("movdqu %%xmm6, 0x10(%[dst], %[off])") |
| 406 |
/* Repeat loop */ |
| 407 |
__ASM_EMIT("add $0x20, %[off]") |
| 408 |
__ASM_EMIT("sub $4, %[count]") |
| 409 |
|
| 410 |
/* Check count again */ |
| 411 |
__ASM_EMIT("4:") |
| 412 |
__ASM_EMIT("add $4, %[count]") |
| 413 |
__ASM_EMIT("jle 6f") |
| 414 |
/* Process scalar data */ |
| 415 |
__ASM_EMIT("5:") |
| 416 |
/* Load */ |
| 417 |
__ASM_EMIT("movss 0x00(%[src1], %[off]), %%xmm0") /* xmm0 = ar */ |
| 418 |
__ASM_EMIT("movss 0x04(%[src1], %[off]), %%xmm2") /* xmm2 = ai */ |
| 419 |
__ASM_EMIT("movss 0x00(%[src2], %[off]), %%xmm4") /* xmm4 = br */ |
| 420 |
__ASM_EMIT("movss 0x04(%[src2], %[off]), %%xmm6") /* xmm6 = bi */ |
| 421 |
/* Calculate multiplication */ |
| 422 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = ar */ |
| 423 |
__ASM_EMIT("mulss %%xmm4, %%xmm0") /* xmm0 = ar*br */ |
| 424 |
__ASM_EMIT("mulss %%xmm6, %%xmm1") /* xmm1 = ar*bi */ |
| 425 |
__ASM_EMIT("mulss %%xmm2, %%xmm6") /* xmm6 = ai*bi */ |
| 426 |
__ASM_EMIT("mulss %%xmm2, %%xmm4") /* xmm4 = ai*br */ |
| 427 |
__ASM_EMIT("addss %%xmm4, %%xmm1") /* xmm1 = ar*bi + ai*br = i */ |
| 428 |
__ASM_EMIT("subss %%xmm6, %%xmm0") /* xmm0 = ar*br - ai*bi = r */ |
| 429 |
/* Store */ |
| 430 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst], %[off])") |
| 431 |
__ASM_EMIT("movss %%xmm1, 0x04(%[dst], %[off])") |
| 432 |
/* Repeat loop */ |
| 433 |
__ASM_EMIT("add $0x08, %[off]") |
| 434 |
__ASM_EMIT("dec %[count]") |
| 435 |
__ASM_EMIT("jnz 5b") |
| 436 |
|
| 437 |
/* Exit */ |
| 438 |
__ASM_EMIT("6:") |
| 439 |
|
| 440 |
: [dst] "+r" (dst), [src1] "+r" (src1), [src2] "+r" (src2), |
| 441 |
[off] "=&r" (off), |
| 442 |
[count] "+r" (count) |
| 443 |
: |
| 444 |
: "cc", "memory", |
| 445 |
"%xmm0", "%xmm1", "%xmm2", |
| 446 |
"%xmm4", "%xmm5", "%xmm6", |
| 447 |
"%xmm8", "%xmm9", "%xmm10", |
| 448 |
"%xmm12", "%xmm13", "%xmm14" |
| 449 |
); |
| 450 |
} |
| 451 |
|
| 452 |
void x64_pcomplex_mul2(float *dst, const float *src, size_t count) |
| 453 |
{ |
| 454 |
IF_ARCH_X86_64(size_t off); |
| 455 |
|
| 456 |
ARCH_X86_64_ASM |
| 457 |
( |
| 458 |
/* Do block processing */ |
| 459 |
__ASM_EMIT("xor %[off], %[off]") |
| 460 |
__ASM_EMIT("sub $8, %[count]") |
| 461 |
__ASM_EMIT("jb 3f") |
| 462 |
|
| 463 |
__ASM_EMIT(".align 16") |
| 464 |
__ASM_EMIT("2:") |
| 465 |
/* Process vectorized data */ |
| 466 |
__ASM_EMIT("lddqu 0x00(%[dst], %[off]), %%xmm0") /* xmm0 = ar0 ai0 ar1 ai1 */ |
| 467 |
__ASM_EMIT("lddqu 0x10(%[dst], %[off]), %%xmm4") /* xmm4 = ar2 ai2 ar3 ai3 */ |
| 468 |
__ASM_EMIT("lddqu 0x20(%[dst], %[off]), %%xmm8") |
| 469 |
__ASM_EMIT("lddqu 0x30(%[dst], %[off]), %%xmm12") |
| 470 |
__ASM_EMIT("lddqu 0x00(%[src], %[off]), %%xmm1") /* xmm1 = br0 bi0 br1 bi1 */ |
| 471 |
__ASM_EMIT("lddqu 0x10(%[src], %[off]), %%xmm5") /* xmm5 = br2 bi2 br3 bi3 */ |
| 472 |
__ASM_EMIT("lddqu 0x20(%[src], %[off]), %%xmm9") |
| 473 |
__ASM_EMIT("lddqu 0x30(%[src], %[off]), %%xmm13") |
| 474 |
/* Calc multiplication */ |
| 475 |
__ASM_EMIT("movsldup %%xmm0, %%xmm2") /* xmm2 = ar0 ar0 ar1, ar1 */ |
| 476 |
__ASM_EMIT("movsldup %%xmm4, %%xmm6") /* xmm6 = ar2 ar2 ar3, ar3 */ |
| 477 |
__ASM_EMIT("movsldup %%xmm8, %%xmm10") |
| 478 |
__ASM_EMIT("movsldup %%xmm12, %%xmm14") |
| 479 |
__ASM_EMIT("movshdup %%xmm0, %%xmm0") /* xmm0 = ai0 ai0 ai1, ai1 */ |
| 480 |
__ASM_EMIT("movshdup %%xmm4, %%xmm4") /* xmm4 = ai2 ai2 ai3, ai3 */ |
| 481 |
__ASM_EMIT("movshdup %%xmm8, %%xmm8") |
| 482 |
__ASM_EMIT("movshdup %%xmm12, %%xmm12") |
| 483 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = ai0*br0 ai0*bi0 ai1*br1 ai1*bi1 */ |
| 484 |
__ASM_EMIT("mulps %%xmm5, %%xmm4") /* xmm4 = ai2*br2 ai2*bi2 ai3*br3 ai3*bi3 */ |
| 485 |
__ASM_EMIT("mulps %%xmm9, %%xmm8") |
| 486 |
__ASM_EMIT("mulps %%xmm13, %%xmm12") |
| 487 |
__ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = ar0*br0 ar0*bi0 ar1*br1 ar1*bi1 */ |
| 488 |
__ASM_EMIT("mulps %%xmm5, %%xmm6") /* xmm6 = ar2*br2 ar2*bi2 ar3*br3 ar3*bi3 */ |
| 489 |
__ASM_EMIT("mulps %%xmm9, %%xmm10") |
| 490 |
__ASM_EMIT("mulps %%xmm13, %%xmm14") |
| 491 |
__ASM_EMIT("shufps $0xb1, %%xmm0, %%xmm0") /* xmm0 = ai0*bi0 ai0*br0 ai1*bi1 ai1*br1 */ |
| 492 |
__ASM_EMIT("shufps $0xb1, %%xmm4, %%xmm4") /* xmm4 = ai2*bi2 ai2*br2 ai3*bi3 ai3*br3 */ |
| 493 |
__ASM_EMIT("shufps $0xb1, %%xmm8, %%xmm8") |
| 494 |
__ASM_EMIT("shufps $0xb1, %%xmm12, %%xmm12") |
| 495 |
__ASM_EMIT("addsubps %%xmm0, %%xmm2") /* xmm2 = ar0*br0 - ai0*bi0 ar0*bi0 + ai0*br0 ar1*br1 - ai1*bi1 ar1*bi1 + ai1*br1 */ |
| 496 |
__ASM_EMIT("addsubps %%xmm4, %%xmm6") /* xmm6 = ar2*br2 - ai2*bi2 ar2*bi2 + ai2*br2 ar3*br3 - ai3*bi3 ar3*bi3 + ai3*br3 */ |
| 497 |
__ASM_EMIT("addsubps %%xmm8, %%xmm10") |
| 498 |
__ASM_EMIT("addsubps %%xmm12, %%xmm14") |
| 499 |
/* Store */ |
| 500 |
__ASM_EMIT("movdqu %%xmm2, 0x00(%[dst], %[off])") |
| 501 |
__ASM_EMIT("movdqu %%xmm6, 0x10(%[dst], %[off])") |
| 502 |
__ASM_EMIT("movdqu %%xmm10, 0x20(%[dst], %[off])") |
| 503 |
__ASM_EMIT("movdqu %%xmm14, 0x30(%[dst], %[off])") |
| 504 |
/* Repeat loop */ |
| 505 |
__ASM_EMIT("add $0x40, %[off]") |
| 506 |
__ASM_EMIT("sub $8, %[count]") |
| 507 |
__ASM_EMIT("jae 2b") |
| 508 |
|
| 509 |
/* 4-element block */ |
| 510 |
__ASM_EMIT("3:") |
| 511 |
__ASM_EMIT("add $4, %[count]") |
| 512 |
__ASM_EMIT("jl 4f") |
| 513 |
/* Process vectorized data */ |
| 514 |
__ASM_EMIT("lddqu 0x00(%[dst], %[off]), %%xmm0") /* xmm0 = ar0 ai0 ar1 ai1 */ |
| 515 |
__ASM_EMIT("lddqu 0x10(%[dst], %[off]), %%xmm4") /* xmm4 = ar2 ai2 ar3 ai3 */ |
| 516 |
__ASM_EMIT("lddqu 0x00(%[src], %[off]), %%xmm1") /* xmm1 = br0 bi0 br1 bi1 */ |
| 517 |
__ASM_EMIT("lddqu 0x10(%[src], %[off]), %%xmm5") /* xmm5 = br2 bi2 br3 bi3 */ |
| 518 |
/* Calc multiplication */ |
| 519 |
__ASM_EMIT("movsldup %%xmm0, %%xmm2") /* xmm2 = ar0 ar0 ar1, ar1 */ |
| 520 |
__ASM_EMIT("movsldup %%xmm4, %%xmm6") /* xmm6 = ar2 ar2 ar3, ar3 */ |
| 521 |
__ASM_EMIT("movshdup %%xmm0, %%xmm0") /* xmm0 = ai0 ai0 ai1, ai1 */ |
| 522 |
__ASM_EMIT("movshdup %%xmm4, %%xmm4") /* xmm4 = ai2 ai2 ai3, ai3 */ |
| 523 |
__ASM_EMIT("mulps %%xmm1, %%xmm0") /* xmm0 = ai0*br0 ai0*bi0 ai1*br1 ai1*bi1 */ |
| 524 |
__ASM_EMIT("mulps %%xmm5, %%xmm4") /* xmm4 = ai2*br2 ai2*bi2 ai3*br3 ai3*bi3 */ |
| 525 |
__ASM_EMIT("mulps %%xmm1, %%xmm2") /* xmm2 = ar0*br0 ar0*bi0 ar1*br1 ar1*bi1 */ |
| 526 |
__ASM_EMIT("mulps %%xmm5, %%xmm6") /* xmm6 = ar2*br2 ar2*bi2 ar3*br3 ar3*bi3 */ |
| 527 |
__ASM_EMIT("shufps $0xb1, %%xmm0, %%xmm0") /* xmm0 = ai0*bi0 ai0*br0 ai1*bi1 ai1*br1 */ |
| 528 |
__ASM_EMIT("shufps $0xb1, %%xmm4, %%xmm4") /* xmm4 = ai2*bi2 ai2*br2 ai3*bi3 ai3*br3 */ |
| 529 |
__ASM_EMIT("addsubps %%xmm0, %%xmm2") /* xmm2 = ar0*br0 - ai0*bi0 ar0*bi0 + ai0*br0 ar1*br1 - ai1*bi1 ar1*bi1 + ai1*br1 */ |
| 530 |
__ASM_EMIT("addsubps %%xmm4, %%xmm6") /* xmm6 = ar2*br2 - ai2*bi2 ar2*bi2 + ai2*br2 ar3*br3 - ai3*bi3 ar3*bi3 + ai3*br3 */ |
| 531 |
/* Store */ |
| 532 |
__ASM_EMIT("movdqu %%xmm2, 0x00(%[dst], %[off])") |
| 533 |
__ASM_EMIT("movdqu %%xmm6, 0x10(%[dst], %[off])") |
| 534 |
/* Repeat loop */ |
| 535 |
__ASM_EMIT("add $0x20, %[off]") |
| 536 |
__ASM_EMIT("sub $4, %[count]") |
| 537 |
|
| 538 |
/* Check count again */ |
| 539 |
__ASM_EMIT("4:") |
| 540 |
__ASM_EMIT("add $4, %[count]") |
| 541 |
__ASM_EMIT("jle 6f") |
| 542 |
/* Process scalar data */ |
| 543 |
__ASM_EMIT("5:") |
| 544 |
/* Load */ |
| 545 |
__ASM_EMIT("movss 0x00(%[dst], %[off]), %%xmm0") /* xmm0 = ar */ |
| 546 |
__ASM_EMIT("movss 0x04(%[dst], %[off]), %%xmm2") /* xmm2 = ai */ |
| 547 |
__ASM_EMIT("movss 0x00(%[src], %[off]), %%xmm4") /* xmm4 = br */ |
| 548 |
__ASM_EMIT("movss 0x04(%[src], %[off]), %%xmm6") /* xmm6 = bi */ |
| 549 |
/* Calculate multiplication */ |
| 550 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = ar */ |
| 551 |
__ASM_EMIT("mulss %%xmm4, %%xmm0") /* xmm0 = ar*br */ |
| 552 |
__ASM_EMIT("mulss %%xmm6, %%xmm1") /* xmm1 = ar*bi */ |
| 553 |
__ASM_EMIT("mulss %%xmm2, %%xmm6") /* xmm6 = ai*bi */ |
| 554 |
__ASM_EMIT("mulss %%xmm2, %%xmm4") /* xmm4 = ai*br */ |
| 555 |
__ASM_EMIT("addss %%xmm4, %%xmm1") /* xmm1 = ar*bi + ai*br = i */ |
| 556 |
__ASM_EMIT("subss %%xmm6, %%xmm0") /* xmm0 = ar*br - ai*bi = r */ |
| 557 |
/* Store */ |
| 558 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst], %[off])") |
| 559 |
__ASM_EMIT("movss %%xmm1, 0x04(%[dst], %[off])") |
| 560 |
/* Repeat loop */ |
| 561 |
__ASM_EMIT("add $0x08, %[off]") |
| 562 |
__ASM_EMIT("dec %[count]") |
| 563 |
__ASM_EMIT("jnz 5b") |
| 564 |
|
| 565 |
/* Exit */ |
| 566 |
__ASM_EMIT("6:") |
| 567 |
|
| 568 |
: [dst] "+r" (dst), [src] "+r" (src), |
| 569 |
[off] "=&r" (off), |
| 570 |
[count] "+r" (count) |
| 571 |
: |
| 572 |
: "cc", "memory", |
| 573 |
"%xmm0", "%xmm1", "%xmm2", |
| 574 |
"%xmm4", "%xmm5", "%xmm6", |
| 575 |
"%xmm8", "%xmm9", "%xmm10", |
| 576 |
"%xmm12", "%xmm13", "%xmm14" |
| 577 |
); |
| 578 |
} |
| 579 |
|
| 580 |
void x64_pcomplex_mod(float *dst, const float *src, size_t count) |
| 581 |
{ |
| 582 |
IF_ARCH_X86_64(size_t off); |
| 583 |
|
| 584 |
ARCH_X86_64_ASM |
| 585 |
( |
| 586 |
__ASM_EMIT("xor %[off], %[off]") |
| 587 |
__ASM_EMIT("sub $32, %[count]") |
| 588 |
__ASM_EMIT("jb 2f") |
| 589 |
|
| 590 |
// 32x blocks |
| 591 |
__ASM_EMIT("1:") |
| 592 |
__ASM_EMIT("lddqu 0x00(%[src], %[off], 2), %%xmm0") |
| 593 |
__ASM_EMIT("lddqu 0x10(%[src], %[off], 2), %%xmm1") |
| 594 |
__ASM_EMIT("lddqu 0x20(%[src], %[off], 2), %%xmm2") |
| 595 |
__ASM_EMIT("lddqu 0x30(%[src], %[off], 2), %%xmm3") |
| 596 |
__ASM_EMIT("lddqu 0x40(%[src], %[off], 2), %%xmm4") |
| 597 |
__ASM_EMIT("lddqu 0x50(%[src], %[off], 2), %%xmm5") |
| 598 |
__ASM_EMIT("lddqu 0x60(%[src], %[off], 2), %%xmm6") |
| 599 |
__ASM_EMIT("lddqu 0x70(%[src], %[off], 2), %%xmm7") |
| 600 |
__ASM_EMIT("lddqu 0x80(%[src], %[off], 2), %%xmm8") |
| 601 |
__ASM_EMIT("lddqu 0x90(%[src], %[off], 2), %%xmm9") |
| 602 |
__ASM_EMIT("lddqu 0xa0(%[src], %[off], 2), %%xmm10") |
| 603 |
__ASM_EMIT("lddqu 0xb0(%[src], %[off], 2), %%xmm11") |
| 604 |
__ASM_EMIT("lddqu 0xc0(%[src], %[off], 2), %%xmm12") |
| 605 |
__ASM_EMIT("lddqu 0xd0(%[src], %[off], 2), %%xmm13") |
| 606 |
__ASM_EMIT("lddqu 0xe0(%[src], %[off], 2), %%xmm14") |
| 607 |
__ASM_EMIT("lddqu 0xf0(%[src], %[off], 2), %%xmm15") |
| 608 |
|
| 609 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") |
| 610 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") |
| 611 |
__ASM_EMIT("mulps %%xmm2, %%xmm2") |
| 612 |
__ASM_EMIT("mulps %%xmm3, %%xmm3") |
| 613 |
__ASM_EMIT("mulps %%xmm4, %%xmm4") |
| 614 |
__ASM_EMIT("mulps %%xmm5, %%xmm5") |
| 615 |
__ASM_EMIT("mulps %%xmm6, %%xmm6") |
| 616 |
__ASM_EMIT("mulps %%xmm7, %%xmm7") |
| 617 |
__ASM_EMIT("mulps %%xmm8, %%xmm8") |
| 618 |
__ASM_EMIT("mulps %%xmm9, %%xmm9") |
| 619 |
__ASM_EMIT("mulps %%xmm10, %%xmm10") |
| 620 |
__ASM_EMIT("mulps %%xmm11, %%xmm11") |
| 621 |
__ASM_EMIT("mulps %%xmm12, %%xmm12") |
| 622 |
__ASM_EMIT("mulps %%xmm13, %%xmm13") |
| 623 |
__ASM_EMIT("mulps %%xmm14, %%xmm14") |
| 624 |
__ASM_EMIT("mulps %%xmm15, %%xmm15") |
| 625 |
|
| 626 |
__ASM_EMIT("haddps %%xmm1, %%xmm0") |
| 627 |
__ASM_EMIT("haddps %%xmm3, %%xmm2") |
| 628 |
__ASM_EMIT("haddps %%xmm5, %%xmm4") |
| 629 |
__ASM_EMIT("haddps %%xmm7, %%xmm6") |
| 630 |
__ASM_EMIT("haddps %%xmm9, %%xmm8") |
| 631 |
__ASM_EMIT("haddps %%xmm11, %%xmm10") |
| 632 |
__ASM_EMIT("haddps %%xmm13, %%xmm12") |
| 633 |
__ASM_EMIT("haddps %%xmm15, %%xmm14") |
| 634 |
|
| 635 |
__ASM_EMIT("sqrtps %%xmm0, %%xmm0") |
| 636 |
__ASM_EMIT("sqrtps %%xmm2, %%xmm2") |
| 637 |
__ASM_EMIT("sqrtps %%xmm4, %%xmm4") |
| 638 |
__ASM_EMIT("sqrtps %%xmm6, %%xmm6") |
| 639 |
__ASM_EMIT("sqrtps %%xmm8, %%xmm8") |
| 640 |
__ASM_EMIT("sqrtps %%xmm10, %%xmm10") |
| 641 |
__ASM_EMIT("sqrtps %%xmm12, %%xmm12") |
| 642 |
__ASM_EMIT("sqrtps %%xmm14, %%xmm14") |
| 643 |
|
| 644 |
__ASM_EMIT("movdqu %%xmm0, 0x00(%[dst], %[off])") |
| 645 |
__ASM_EMIT("movdqu %%xmm2, 0x10(%[dst], %[off])") |
| 646 |
__ASM_EMIT("movdqu %%xmm4, 0x20(%[dst], %[off])") |
| 647 |
__ASM_EMIT("movdqu %%xmm6, 0x30(%[dst], %[off])") |
| 648 |
__ASM_EMIT("movdqu %%xmm8, 0x40(%[dst], %[off])") |
| 649 |
__ASM_EMIT("movdqu %%xmm10, 0x50(%[dst], %[off])") |
| 650 |
__ASM_EMIT("movdqu %%xmm12, 0x60(%[dst], %[off])") |
| 651 |
__ASM_EMIT("movdqu %%xmm14, 0x70(%[dst], %[off])") |
| 652 |
|
| 653 |
__ASM_EMIT("add $0x80, %[off]") |
| 654 |
__ASM_EMIT("sub $32, %[count]") |
| 655 |
__ASM_EMIT("jae 1b") |
| 656 |
|
| 657 |
// 16x blocks |
| 658 |
__ASM_EMIT("2:") |
| 659 |
__ASM_EMIT("add $16, %[count]") |
| 660 |
__ASM_EMIT("jl 4f") |
| 661 |
|
| 662 |
__ASM_EMIT("3:") |
| 663 |
__ASM_EMIT("lddqu 0x00(%[src], %[off], 2), %%xmm0") |
| 664 |
__ASM_EMIT("lddqu 0x10(%[src], %[off], 2), %%xmm1") |
| 665 |
__ASM_EMIT("lddqu 0x20(%[src], %[off], 2), %%xmm2") |
| 666 |
__ASM_EMIT("lddqu 0x30(%[src], %[off], 2), %%xmm3") |
| 667 |
__ASM_EMIT("lddqu 0x40(%[src], %[off], 2), %%xmm4") |
| 668 |
__ASM_EMIT("lddqu 0x50(%[src], %[off], 2), %%xmm5") |
| 669 |
__ASM_EMIT("lddqu 0x60(%[src], %[off], 2), %%xmm6") |
| 670 |
__ASM_EMIT("lddqu 0x70(%[src], %[off], 2), %%xmm7") |
| 671 |
|
| 672 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") |
| 673 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") |
| 674 |
__ASM_EMIT("mulps %%xmm2, %%xmm2") |
| 675 |
__ASM_EMIT("mulps %%xmm3, %%xmm3") |
| 676 |
__ASM_EMIT("mulps %%xmm4, %%xmm4") |
| 677 |
__ASM_EMIT("mulps %%xmm5, %%xmm5") |
| 678 |
__ASM_EMIT("mulps %%xmm6, %%xmm6") |
| 679 |
__ASM_EMIT("mulps %%xmm7, %%xmm7") |
| 680 |
__ASM_EMIT("haddps %%xmm1, %%xmm0") |
| 681 |
__ASM_EMIT("haddps %%xmm3, %%xmm2") |
| 682 |
__ASM_EMIT("haddps %%xmm5, %%xmm4") |
| 683 |
__ASM_EMIT("haddps %%xmm7, %%xmm6") |
| 684 |
__ASM_EMIT("sqrtps %%xmm0, %%xmm0") |
| 685 |
__ASM_EMIT("sqrtps %%xmm2, %%xmm2") |
| 686 |
__ASM_EMIT("sqrtps %%xmm4, %%xmm4") |
| 687 |
__ASM_EMIT("sqrtps %%xmm6, %%xmm6") |
| 688 |
|
| 689 |
__ASM_EMIT("movdqu %%xmm0, 0x00(%[dst], %[off])") |
| 690 |
__ASM_EMIT("movdqu %%xmm2, 0x10(%[dst], %[off])") |
| 691 |
__ASM_EMIT("movdqu %%xmm4, 0x20(%[dst], %[off])") |
| 692 |
__ASM_EMIT("movdqu %%xmm6, 0x30(%[dst], %[off])") |
| 693 |
__ASM_EMIT("add $0x40, %[off]") |
| 694 |
__ASM_EMIT("sub $16, %[count]") |
| 695 |
__ASM_EMIT("jae 3b") |
| 696 |
|
| 697 |
// 4x blocks |
| 698 |
__ASM_EMIT("4:") |
| 699 |
__ASM_EMIT("add $12, %[count]") |
| 700 |
__ASM_EMIT("jl 6f") |
| 701 |
__ASM_EMIT("5:") |
| 702 |
__ASM_EMIT("lddqu 0x00(%[src], %[off], 2), %%xmm0") |
| 703 |
__ASM_EMIT("lddqu 0x10(%[src], %[off], 2), %%xmm1") |
| 704 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") |
| 705 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") |
| 706 |
__ASM_EMIT("haddps %%xmm1, %%xmm0") |
| 707 |
__ASM_EMIT("sqrtps %%xmm0, %%xmm0") |
| 708 |
__ASM_EMIT("movdqu %%xmm0, 0x00(%[dst], %[off])") |
| 709 |
__ASM_EMIT("add $0x10, %[off]") |
| 710 |
__ASM_EMIT("sub $4, %[count]") |
| 711 |
__ASM_EMIT("jge 5b") |
| 712 |
|
| 713 |
// 1x blocks |
| 714 |
__ASM_EMIT("6:") |
| 715 |
__ASM_EMIT("add $3, %[count]") |
| 716 |
__ASM_EMIT("jl 8f") |
| 717 |
__ASM_EMIT("7:") |
| 718 |
__ASM_EMIT("movlps 0x00(%[src], %[off], 2), %%xmm0") |
| 719 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") |
| 720 |
__ASM_EMIT("haddps %%xmm0, %%xmm0") |
| 721 |
__ASM_EMIT("sqrtss %%xmm0, %%xmm0") |
| 722 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst], %[off])") |
| 723 |
__ASM_EMIT("add $0x4, %[off]") |
| 724 |
__ASM_EMIT("dec %[count]") |
| 725 |
__ASM_EMIT("jge 7b") |
| 726 |
|
| 727 |
// End |
| 728 |
__ASM_EMIT("8:") |
| 729 |
: [dst] "+r" (dst), [src] "+r" (src), |
| 730 |
[count] "+r" (count), [off] "=&r" (off) |
| 731 |
: |
| 732 |
: "cc", "memory", |
| 733 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 734 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 735 |
"%xmm8", "%xmm9", "%xmm10", |
| 736 |
"%xmm12", "%xmm13", "%xmm14" |
| 737 |
); |
| 738 |
} |
| 739 |
|
| 740 |
IF_ARCH_X86_64( |
| 741 |
static const float pcomplex_corr_const[] __lsp_aligned16 = |
| 742 |
{ |
| 743 |
LSP_DSP_VEC4(1e-36f), |
| 744 |
}; |
| 745 |
); |
| 746 |
|
| 747 |
void x64_pcomplex_corr(float *dst_corr, const float *src1, const float *src2, size_t count) |
| 748 |
{ |
| 749 |
IF_ARCH_X86_64(size_t off); |
| 750 |
|
| 751 |
/* |
| 752 |
* src1[i] = a + j*b, src2[i] = c + j*d |
| 753 |
* den = (a*a + b*b)*(c*c + d*d) |
| 754 |
* nom = a*c + b*d |
| 755 |
* corr[i] = (den > threshold) ? nom / sqrt(den) : 0.0 |
| 756 |
*/ |
| 757 |
ARCH_X86_64_ASM ( |
| 758 |
__ASM_EMIT("xor %[off], %[off]") |
| 759 |
/* x8 blocks */ |
| 760 |
__ASM_EMIT("sub $8, %[count]") /* count -= 8 */ |
| 761 |
__ASM_EMIT("movaps %[CC], %%xmm7") /* xmm7 = threshold */ |
| 762 |
__ASM_EMIT("jb 2f") |
| 763 |
__ASM_EMIT("1:") |
| 764 |
__ASM_EMIT("movups 0x00(%[src1], %[off], 2), %%xmm0") /* xmm0 = a0 b0 a1 b1 */ |
| 765 |
__ASM_EMIT("movups 0x10(%[src1], %[off], 2), %%xmm4") /* xmm4 = a2 b2 a3 b3 */ |
| 766 |
__ASM_EMIT("movups 0x20(%[src1], %[off], 2), %%xmm8") /* xmm8 = a4 b4 a5 b5 */ |
| 767 |
__ASM_EMIT("movups 0x30(%[src1], %[off], 2), %%xmm12") /* xmm12 = a6 b6 a7 b7 */ |
| 768 |
__ASM_EMIT("movups 0x00(%[src2], %[off], 2), %%xmm2") /* xmm2 = c0 d0 c1 d1 */ |
| 769 |
__ASM_EMIT("movups 0x10(%[src2], %[off], 2), %%xmm6") /* xmm6 = c2 d2 c3 d3 */ |
| 770 |
__ASM_EMIT("movups 0x20(%[src2], %[off], 2), %%xmm10") /* xmm10 = c4 d4 c5 d5 */ |
| 771 |
__ASM_EMIT("movups 0x30(%[src2], %[off], 2), %%xmm14") /* xmm14 = c6 d6 c7 d7 */ |
| 772 |
/* Do shuffle */ |
| 773 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = a0 b0 a1 b1 */ |
| 774 |
__ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = c0 d0 c1 d1 */ |
| 775 |
__ASM_EMIT("movaps %%xmm8, %%xmm9") /* xmm9 = a4 b4 a5 b5 */ |
| 776 |
__ASM_EMIT("movaps %%xmm10, %%xmm11") /* xmm11 = c4 d4 c5 d5 */ |
| 777 |
__ASM_EMIT("shufps $0x88, %%xmm4, %%xmm0") /* xmm0 = A0 = a0 a1 a2 a3 */ |
| 778 |
__ASM_EMIT("shufps $0x88, %%xmm6, %%xmm2") /* xmm2 = C0 = c0 c1 c2 c3 */ |
| 779 |
__ASM_EMIT("shufps $0xdd, %%xmm4, %%xmm1") /* xmm1 = B0 = b0 b1 b2 b3 */ |
| 780 |
__ASM_EMIT("shufps $0xdd, %%xmm6, %%xmm3") /* xmm3 = D0 = d0 d1 d2 d3 */ |
| 781 |
__ASM_EMIT("shufps $0x88, %%xmm12, %%xmm8") /* xmm8 = A1 = a0 a1 a2 a3 */ |
| 782 |
__ASM_EMIT("shufps $0x88, %%xmm14, %%xmm10") /* xmm10 = C1 = c0 c1 c2 c3 */ |
| 783 |
__ASM_EMIT("shufps $0xdd, %%xmm12, %%xmm9") /* xmm9 = B1 = b0 b1 b2 b3 */ |
| 784 |
__ASM_EMIT("shufps $0xdd, %%xmm14, %%xmm11") /* xmm11 = D1 = d0 d1 d2 d3 */ |
| 785 |
/* xmm0 = A0, xmm1 = B0, xmm2 = C0, xmm3 = D0 */ |
| 786 |
/* xmm8 = A1, xmm9 = B1, xmm10 = C1, xmm11 = D1 */ |
| 787 |
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = A0 */ |
| 788 |
__ASM_EMIT("movaps %%xmm1, %%xmm5") /* xmm5 = B0 */ |
| 789 |
__ASM_EMIT("movaps %%xmm8, %%xmm12") /* xmm12 = A1 */ |
| 790 |
__ASM_EMIT("movaps %%xmm9, %%xmm13") /* xmm13 = B1 */ |
| 791 |
__ASM_EMIT("mulps %%xmm2, %%xmm4") /* xmm4 = A0*C0 */ |
| 792 |
__ASM_EMIT("mulps %%xmm3, %%xmm5") /* xmm5 = B0*D0 */ |
| 793 |
__ASM_EMIT("mulps %%xmm10, %%xmm12") /* xmm10 = A1*C1 */ |
| 794 |
__ASM_EMIT("mulps %%xmm11, %%xmm13") /* xmm13 = B1*D1 */ |
| 795 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") /* xmm0 = A0*A0 */ |
| 796 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") /* xmm1 = B0*B0 */ |
| 797 |
__ASM_EMIT("mulps %%xmm2, %%xmm2") /* xmm2 = C0*C0 */ |
| 798 |
__ASM_EMIT("mulps %%xmm3, %%xmm3") /* xmm3 = D0*D0 */ |
| 799 |
__ASM_EMIT("mulps %%xmm8, %%xmm8") /* xmm8 = A1*A1 */ |
| 800 |
__ASM_EMIT("mulps %%xmm9, %%xmm9") /* xmm9 = B1*B1 */ |
| 801 |
__ASM_EMIT("mulps %%xmm10, %%xmm10") /* xmm10 = C1*C1 */ |
| 802 |
__ASM_EMIT("mulps %%xmm11, %%xmm11") /* xmm11 = D1*D1 */ |
| 803 |
__ASM_EMIT("addps %%xmm1, %%xmm0") /* xmm0 = A0*A0 + B0*B0 */ |
| 804 |
__ASM_EMIT("addps %%xmm3, %%xmm2") /* xmm2 = C0*C0 + D0*D0 */ |
| 805 |
__ASM_EMIT("addps %%xmm9, %%xmm8") /* xmm8 = A1*A1 + B1*B1 */ |
| 806 |
__ASM_EMIT("addps %%xmm11, %%xmm10") /* xmm10 = C1*C1 + D1*D1 */ |
| 807 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = den0 = (A0*A0 + B0*B0)*(C0*C0 + D0*D0) */ |
| 808 |
__ASM_EMIT("mulps %%xmm10, %%xmm8") /* xmm8 = den1 = (A1*A1 + B1*B1)*(C1*C1 + D1*D1) */ |
| 809 |
__ASM_EMIT("addps %%xmm5, %%xmm4") /* xmm4 = nom0 = A0*C0 + B0*D0 */ |
| 810 |
__ASM_EMIT("addps %%xmm13, %%xmm12") /* xmm12 = nom1 = A1*C1 + B1*D1 */ |
| 811 |
__ASM_EMIT("sqrtps %%xmm0, %%xmm1") /* xmm1 = sqrt(den0) */ |
| 812 |
__ASM_EMIT("sqrtps %%xmm8, %%xmm9") /* xmm9 = sqrt(den1) */ |
| 813 |
__ASM_EMIT("divps %%xmm1, %%xmm4") /* xmm4 = nom0 / sqrt(den0) */ |
| 814 |
__ASM_EMIT("divps %%xmm9, %%xmm12") /* xmm12 = nom1 / sqrt(den1) */ |
| 815 |
__ASM_EMIT("cmpps $5, %%xmm7, %%xmm0") /* xmm0 = [den0 >= threshold] */ |
| 816 |
__ASM_EMIT("cmpps $5, %%xmm7, %%xmm8") /* xmm8 = [den1 >= threshold] */ |
| 817 |
__ASM_EMIT("andps %%xmm4, %%xmm0") /* xmm0 = (den0 >= threshold) ? nom0 / sqrt(den0) : 0.0f */ |
| 818 |
__ASM_EMIT("andps %%xmm12, %%xmm8") /* xmm8 = (den0 >= threshold) ? nom0 / sqrt(den0) : 0.0f */ |
| 819 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst], %[off])") /* dst[0] = v0 v1 v2 v3 */ |
| 820 |
__ASM_EMIT("movups %%xmm8, 0x10(%[dst], %[off])") /* dst[1] = v4 v5 v6 v7 */ |
| 821 |
__ASM_EMIT("add $0x20, %[off]") |
| 822 |
__ASM_EMIT("sub $8, %[count]") /* count -= 8 */ |
| 823 |
__ASM_EMIT("jae 1b") |
| 824 |
/* x4 block */ |
| 825 |
__ASM_EMIT("2:") |
| 826 |
__ASM_EMIT("add $4, %[count]") /* count += 4 */ |
| 827 |
__ASM_EMIT("jl 4f") |
| 828 |
__ASM_EMIT("movups 0x00(%[src1], %[off], 2), %%xmm0") /* xmm0 = a0 b0 a1 b1 */ |
| 829 |
__ASM_EMIT("movups 0x10(%[src1], %[off], 2), %%xmm4") /* xmm4 = a2 b2 a3 b3 */ |
| 830 |
__ASM_EMIT("movups 0x00(%[src2], %[off], 2), %%xmm2") /* xmm2 = c0 d0 c1 d1 */ |
| 831 |
__ASM_EMIT("movups 0x10(%[src2], %[off], 2), %%xmm6") /* xmm6 = c2 d2 c3 d3 */ |
| 832 |
/* Do shuffle */ |
| 833 |
__ASM_EMIT("movaps %%xmm0, %%xmm1") /* xmm1 = a0 b0 a1 b1 */ |
| 834 |
__ASM_EMIT("movaps %%xmm2, %%xmm3") /* xmm3 = c0 d0 c1 d1 */ |
| 835 |
__ASM_EMIT("shufps $0x88, %%xmm4, %%xmm0") /* xmm0 = A0 = a0 a1 a2 a3 */ |
| 836 |
__ASM_EMIT("shufps $0x88, %%xmm6, %%xmm2") /* xmm2 = C0 = c0 c1 c2 c3 */ |
| 837 |
__ASM_EMIT("shufps $0xdd, %%xmm4, %%xmm1") /* xmm1 = B0 = b0 b1 b2 b3 */ |
| 838 |
__ASM_EMIT("shufps $0xdd, %%xmm6, %%xmm3") /* xmm3 = D0 = d0 d1 d2 d3 */ |
| 839 |
/* xmm0 = A0, xmm1 = B0, xmm2 = C0, xmm3 = D0 */ |
| 840 |
/* xmm8 = A1, xmm9 = B1, xmm10 = C1, xmm11 = D1 */ |
| 841 |
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = A0 */ |
| 842 |
__ASM_EMIT("movaps %%xmm1, %%xmm5") /* xmm5 = B0 */ |
| 843 |
__ASM_EMIT("mulps %%xmm2, %%xmm4") /* xmm4 = A0*C0 */ |
| 844 |
__ASM_EMIT("mulps %%xmm3, %%xmm5") /* xmm5 = B0*D0 */ |
| 845 |
__ASM_EMIT("mulps %%xmm0, %%xmm0") /* xmm0 = A0*A0 */ |
| 846 |
__ASM_EMIT("mulps %%xmm1, %%xmm1") /* xmm1 = B0*B0 */ |
| 847 |
__ASM_EMIT("mulps %%xmm2, %%xmm2") /* xmm2 = C0*C0 */ |
| 848 |
__ASM_EMIT("mulps %%xmm3, %%xmm3") /* xmm3 = D0*D0 */ |
| 849 |
__ASM_EMIT("addps %%xmm1, %%xmm0") /* xmm0 = A0*A0 + B0*B0 */ |
| 850 |
__ASM_EMIT("addps %%xmm3, %%xmm2") /* xmm2 = C0*C0 + D0*D0 */ |
| 851 |
__ASM_EMIT("mulps %%xmm2, %%xmm0") /* xmm0 = den0 = (A0*A0 + B0*B0)*(C0*C0 + D0*D0) */ |
| 852 |
__ASM_EMIT("addps %%xmm5, %%xmm4") /* xmm4 = nom0 = A0*C0 + B0*D0 */ |
| 853 |
__ASM_EMIT("sqrtps %%xmm0, %%xmm1") /* xmm1 = sqrt(den0) */ |
| 854 |
__ASM_EMIT("divps %%xmm1, %%xmm4") /* xmm4 = nom0 / sqrt(den0) */ |
| 855 |
__ASM_EMIT("cmpps $5, %%xmm7, %%xmm0") /* xmm0 = [den0 >= threshold] */ |
| 856 |
__ASM_EMIT("andps %%xmm4, %%xmm0") /* xmm0 = (den0 >= threshold) ? nom0 / sqrt(den0) : 0.0f */ |
| 857 |
__ASM_EMIT("movups %%xmm0, 0x00(%[dst], %[off])") /* dst[0] = v0 v1 v2 v3 */ |
| 858 |
__ASM_EMIT("add $0x10, %[off]") |
| 859 |
__ASM_EMIT("sub $4, %[count]") /* count -= 4 */ |
| 860 |
/* x1 blocks */ |
| 861 |
__ASM_EMIT("4:") |
| 862 |
__ASM_EMIT("add $3, %[count]") /* count += 3 */ |
| 863 |
__ASM_EMIT("jl 6f") |
| 864 |
__ASM_EMIT("5:") |
| 865 |
__ASM_EMIT("movss 0x00(%[src1], %[off], 2), %%xmm0") /* xmm0 = a */ |
| 866 |
__ASM_EMIT("movss 0x04(%[src1], %[off], 2), %%xmm1") /* xmm1 = b */ |
| 867 |
__ASM_EMIT("movss 0x00(%[src2], %[off], 2), %%xmm2") /* xmm2 = c */ |
| 868 |
__ASM_EMIT("movss 0x04(%[src2], %[off], 2), %%xmm3") /* xmm3 = d */ |
| 869 |
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = a */ |
| 870 |
__ASM_EMIT("movaps %%xmm1, %%xmm5") /* xmm5 = b */ |
| 871 |
__ASM_EMIT("mulss %%xmm2, %%xmm4") /* xmm4 = a*c */ |
| 872 |
__ASM_EMIT("mulss %%xmm3, %%xmm5") /* xmm5 = b*d */ |
| 873 |
__ASM_EMIT("mulss %%xmm0, %%xmm0") /* xmm0 = a*a */ |
| 874 |
__ASM_EMIT("mulss %%xmm1, %%xmm1") /* xmm1 = b*b */ |
| 875 |
__ASM_EMIT("mulss %%xmm2, %%xmm2") /* xmm2 = c*c */ |
| 876 |
__ASM_EMIT("mulss %%xmm3, %%xmm3") /* xmm3 = d*d */ |
| 877 |
__ASM_EMIT("addss %%xmm1, %%xmm0") /* xmm0 = a*a + b*b */ |
| 878 |
__ASM_EMIT("addss %%xmm3, %%xmm2") /* xmm2 = c*c + d*d */ |
| 879 |
__ASM_EMIT("mulss %%xmm2, %%xmm0") /* xmm0 = den = (a*a + b*b)*(c*c + d*d) */ |
| 880 |
__ASM_EMIT("addss %%xmm5, %%xmm4") /* xmm4 = nom = a*c + b*d */ |
| 881 |
__ASM_EMIT("sqrtss %%xmm0, %%xmm1") /* xmm1 = sqrt(den) */ |
| 882 |
__ASM_EMIT("divss %%xmm1, %%xmm4") /* xmm4 = nom / sqrt(den) */ |
| 883 |
__ASM_EMIT("cmpps $5, %%xmm7, %%xmm0") /* xmm0 = [den >= threshold] */ |
| 884 |
__ASM_EMIT("andps %%xmm4, %%xmm0") /* xmm0 = (den >= threshold) ? nom / sqrt(den) : 0.0f */ |
| 885 |
__ASM_EMIT("movss %%xmm0, 0x00(%[dst], %[off])") /* dst[0] = v0 */ |
| 886 |
__ASM_EMIT("add $0x04, %[off]") |
| 887 |
__ASM_EMIT("dec %[count]") |
| 888 |
__ASM_EMIT("jge 5b") |
| 889 |
/* End */ |
| 890 |
__ASM_EMIT("6:") |
| 891 |
|
| 892 |
: [count] "+r" (count), [off] "=&r" (off) |
| 893 |
: [dst] "r" (dst_corr), [src1] "r" (src1), [src2] "r" (src2), |
| 894 |
[CC] "m" (pcomplex_corr_const) |
| 895 |
: "cc", "memory", |
| 896 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 897 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7", |
| 898 |
"%xmm8", "%xmm9", "%xmm10", "%xmm11", |
| 899 |
"%xmm12", "%xmm13", "%xmm14", "%xmm15" |
| 900 |
); |
| 901 |
} |
| 902 |
|
| 903 |
|
| 904 |
} /* namespace sse3 */ |
| 905 |
} /* namespace lsp */ |
| 906 |
|
| 907 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE3_PCOMPLEX_H_ */ |