| 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: 9 сент. 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_COMPLEX_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_AVX512_COMPLEX_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 |
#define COMPLEX_MUL3_CORE(DST, SRC1, SRC2) \ |
| 34 |
__ASM_EMIT ("xor %[off], %[off]") \ |
| 35 |
/* x16 blocks */ \ |
| 36 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 37 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 38 |
__ASM_EMIT ("jb 2f") \ |
| 39 |
__ASM_EMIT ("1:") \ |
| 40 |
__ASM_EMIT32("mov %[" SRC1 "_re], %[ptr_re]") \ |
| 41 |
__ASM_EMIT32("mov %[" SRC1 "_im], %[ptr_im]") \ |
| 42 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%zmm4") /* zmm4 = ar */ \ |
| 43 |
__ASM_EMIT32("vmovups 0x00(%[ptr_im], %[off]), %%zmm5") /* zmm5 = ai */ \ |
| 44 |
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_re], %[off]), %%zmm4") /* zmm4 = ar */ \ |
| 45 |
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_im], %[off]), %%zmm5") /* zmm5 = ai */ \ |
| 46 |
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_re], %[off]), %%zmm0") /* zmm0 = br */ \ |
| 47 |
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_im], %[off]), %%zmm1") /* zmm1 = bi */ \ |
| 48 |
__ASM_EMIT ("vmulps %%zmm0, %%zmm5, %%zmm2") /* zmm2 = ai*br */ \ |
| 49 |
__ASM_EMIT ("vmulps %%zmm1, %%zmm5, %%zmm3") /* zmm3 = ai*bi */ \ |
| 50 |
__ASM_EMIT32("mov %[" DST "_re], %[ptr_re]") \ |
| 51 |
__ASM_EMIT32("mov %[" DST "_im], %[ptr_im]") \ |
| 52 |
__ASM_EMIT ("vfmsub132ps %%zmm4, %%zmm3, %%zmm0") /* ymm0 = ar*br - ai*bi = r */ \ |
| 53 |
__ASM_EMIT ("vfmadd132ps %%zmm4, %%zmm2, %%zmm1") /* ymm1 = ar*bi + ai*br = i */ \ |
| 54 |
__ASM_EMIT32("vmovups %%zmm0, 0x00(%[ptr_re], %[off])") \ |
| 55 |
__ASM_EMIT32("vmovups %%zmm1, 0x00(%[ptr_im], %[off])") \ |
| 56 |
__ASM_EMIT64("vmovups %%zmm0, 0x00(%[" DST "_re], %[off])") \ |
| 57 |
__ASM_EMIT64("vmovups %%zmm1, 0x00(%[" DST "_im], %[off])") \ |
| 58 |
__ASM_EMIT ("add $0x40, %[off]") \ |
| 59 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 60 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 61 |
__ASM_EMIT ("jae 1b") \ |
| 62 |
__ASM_EMIT ("2:") \ |
| 63 |
/* x8 blocks */ \ |
| 64 |
__ASM_EMIT32("addl $8, %[count]") \ |
| 65 |
__ASM_EMIT64("add $8, %[count]") \ |
| 66 |
__ASM_EMIT ("jl 4f") \ |
| 67 |
__ASM_EMIT32("mov %[" SRC1 "_re], %[ptr_re]") \ |
| 68 |
__ASM_EMIT32("mov %[" SRC1 "_im], %[ptr_im]") \ |
| 69 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%ymm4") /* ymm4 = ar */ \ |
| 70 |
__ASM_EMIT32("vmovups 0x00(%[ptr_im], %[off]), %%ymm5") /* ymm5 = ai */ \ |
| 71 |
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_re], %[off]), %%ymm4") /* ymm4 = ar */ \ |
| 72 |
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_im], %[off]), %%ymm5") /* ymm5 = ai */ \ |
| 73 |
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_re], %[off]), %%ymm0") /* ymm0 = br */ \ |
| 74 |
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_im], %[off]), %%ymm1") /* ymm1 = bi */ \ |
| 75 |
__ASM_EMIT ("vmulps %%ymm0, %%ymm5, %%ymm2") /* ymm2 = ai*br */ \ |
| 76 |
__ASM_EMIT ("vmulps %%ymm1, %%ymm5, %%ymm3") /* ymm3 = ai*bi */ \ |
| 77 |
__ASM_EMIT32("mov %[" DST "_re], %[ptr_re]") \ |
| 78 |
__ASM_EMIT32("mov %[" DST "_im], %[ptr_im]") \ |
| 79 |
__ASM_EMIT ("vfmsub132ps %%ymm4, %%ymm3, %%ymm0") /* ymm0 = ar*br - ai*bi = r */ \ |
| 80 |
__ASM_EMIT ("vfmadd132ps %%ymm4, %%ymm2, %%ymm1") /* ymm1 = ar*bi + ai*br = i */ \ |
| 81 |
__ASM_EMIT32("vmovups %%ymm0, 0x00(%[ptr_re], %[off])") \ |
| 82 |
__ASM_EMIT32("vmovups %%ymm1, 0x00(%[ptr_im], %[off])") \ |
| 83 |
__ASM_EMIT64("vmovups %%ymm0, 0x00(%[" DST "_re], %[off])") \ |
| 84 |
__ASM_EMIT64("vmovups %%ymm1, 0x00(%[" DST "_im], %[off])") \ |
| 85 |
__ASM_EMIT ("add $0x20, %[off]") \ |
| 86 |
__ASM_EMIT32("subl $8, %[count]") \ |
| 87 |
__ASM_EMIT64("sub $8, %[count]") \ |
| 88 |
__ASM_EMIT ("jae 1b") \ |
| 89 |
__ASM_EMIT ("4:") \ |
| 90 |
/* 4x block */ \ |
| 91 |
__ASM_EMIT32("addl $4, %[count]") \ |
| 92 |
__ASM_EMIT64("add $4, %[count]") \ |
| 93 |
__ASM_EMIT ("jl 6f") \ |
| 94 |
__ASM_EMIT32("mov %[" SRC1 "_re], %[ptr_re]") \ |
| 95 |
__ASM_EMIT32("mov %[" SRC1 "_im], %[ptr_im]") \ |
| 96 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%xmm4") /* xmm4 = ar */ \ |
| 97 |
__ASM_EMIT32("vmovups 0x00(%[ptr_im], %[off]), %%xmm5") /* xmm5 = ai */ \ |
| 98 |
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_re], %[off]), %%xmm4") /* xmm4 = ar */ \ |
| 99 |
__ASM_EMIT64("vmovups 0x00(%[" SRC1 "_im], %[off]), %%xmm5") /* xmm5 = ai */ \ |
| 100 |
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_re], %[off]), %%xmm0") /* xmm0 = br */ \ |
| 101 |
__ASM_EMIT ("vmovups 0x00(%[" SRC2 "_im], %[off]), %%xmm1") /* xmm1 = bi */ \ |
| 102 |
__ASM_EMIT ("vmulps %%xmm0, %%xmm5, %%xmm2") /* xmm2 = ai*br */ \ |
| 103 |
__ASM_EMIT ("vmulps %%xmm1, %%xmm5, %%xmm3") /* xmm3 = ai*bi */ \ |
| 104 |
__ASM_EMIT32("mov %[" DST "_re], %[ptr_re]") \ |
| 105 |
__ASM_EMIT32("mov %[" DST "_im], %[ptr_im]") \ |
| 106 |
__ASM_EMIT ("vfmsub132ps %%xmm4, %%xmm3, %%xmm0") /* xmm0 = ar*br - ai*bi = r */ \ |
| 107 |
__ASM_EMIT ("vfmadd132ps %%xmm4, %%xmm2, %%xmm1") /* xmm1 = ar*bi + ai*br = i */ \ |
| 108 |
__ASM_EMIT32("vmovups %%xmm0, 0x00(%[ptr_re], %[off])") \ |
| 109 |
__ASM_EMIT32("vmovups %%xmm1, 0x00(%[ptr_im], %[off])") \ |
| 110 |
__ASM_EMIT64("vmovups %%xmm0, 0x00(%[" DST "_re], %[off])") \ |
| 111 |
__ASM_EMIT64("vmovups %%xmm1, 0x00(%[" DST "_im], %[off])") \ |
| 112 |
__ASM_EMIT ("add $0x10, %[off]") \ |
| 113 |
__ASM_EMIT32("subl $4, %[count]") \ |
| 114 |
__ASM_EMIT64("sub $4, %[count]") \ |
| 115 |
__ASM_EMIT ("6:") \ |
| 116 |
/* 1x blocks */ \ |
| 117 |
__ASM_EMIT32("addl $3, %[count]") \ |
| 118 |
__ASM_EMIT64("add $3, %[count]") \ |
| 119 |
__ASM_EMIT ("jl 8f") \ |
| 120 |
__ASM_EMIT ("7:") \ |
| 121 |
__ASM_EMIT32("mov %[" SRC1 "_re], %[ptr_re]") \ |
| 122 |
__ASM_EMIT32("mov %[" SRC1 "_im], %[ptr_im]") \ |
| 123 |
__ASM_EMIT32("vmovss 0x00(%[ptr_re], %[off]), %%xmm4") /* xmm4 = ar */ \ |
| 124 |
__ASM_EMIT32("vmovss 0x00(%[ptr_im], %[off]), %%xmm5") /* xmm5 = ai */ \ |
| 125 |
__ASM_EMIT64("vmovss 0x00(%[" SRC1 "_re], %[off]), %%xmm4") /* xmm4 = ar */ \ |
| 126 |
__ASM_EMIT64("vmovss 0x00(%[" SRC1 "_im], %[off]), %%xmm5") /* xmm5 = ai */ \ |
| 127 |
__ASM_EMIT ("vmovss 0x00(%[" SRC2 "_re], %[off]), %%xmm0") /* xmm0 = br */ \ |
| 128 |
__ASM_EMIT ("vmovss 0x00(%[" SRC2 "_im], %[off]), %%xmm1") /* xmm1 = bi */ \ |
| 129 |
__ASM_EMIT ("vmulps %%xmm0, %%xmm5, %%xmm2") /* xmm2 = ai*br */ \ |
| 130 |
__ASM_EMIT ("vmulps %%xmm1, %%xmm5, %%xmm3") /* xmm3 = ai*bi */ \ |
| 131 |
__ASM_EMIT32("mov %[" DST "_re], %[ptr_re]") \ |
| 132 |
__ASM_EMIT32("mov %[" DST "_im], %[ptr_im]") \ |
| 133 |
__ASM_EMIT ("vfmsub132ss %%xmm4, %%xmm3, %%xmm0") /* xmm0 = ar*br - ai*bi = r */ \ |
| 134 |
__ASM_EMIT ("vfmadd132ss %%xmm4, %%xmm2, %%xmm1") /* xmm1 = ar*bi + ai*br = i */ \ |
| 135 |
__ASM_EMIT32("vmovss %%xmm0, 0x00(%[ptr_re], %[off])") \ |
| 136 |
__ASM_EMIT32("vmovss %%xmm1, 0x00(%[ptr_im], %[off])") \ |
| 137 |
__ASM_EMIT64("vmovss %%xmm0, 0x00(%[" DST "_re], %[off])") \ |
| 138 |
__ASM_EMIT64("vmovss %%xmm1, 0x00(%[" DST "_im], %[off])") \ |
| 139 |
__ASM_EMIT ("add $0x04, %[off]") \ |
| 140 |
__ASM_EMIT32("decl %[count]") \ |
| 141 |
__ASM_EMIT64("dec %[count]") \ |
| 142 |
__ASM_EMIT ("jge 7b") \ |
| 143 |
__ASM_EMIT ("8:") |
| 144 |
|
| 145 |
void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count) |
| 146 |
{ |
| 147 |
IF_ARCH_I386(float *ptr_re, *ptr_im); |
| 148 |
IF_ARCH_X86(size_t off); |
| 149 |
ARCH_X86_ASM |
| 150 |
( |
| 151 |
COMPLEX_MUL3_CORE("dst", "src1", "src2") |
| 152 |
: __IF_32( |
| 153 |
[ptr_re] "=&r" (ptr_re), [ptr_im] "=&r" (ptr_im), |
| 154 |
) |
| 155 |
[count] X86_PGREG (count), |
| 156 |
[off] "=&r" (off) |
| 157 |
: [dst_re] X86_GREG (dst_re), [dst_im] X86_GREG (dst_im), |
| 158 |
[src1_re] X86_GREG (src1_re), [src1_im] X86_GREG (src1_im), |
| 159 |
[src2_re] "r" (src2_re), [src2_im] "r" (src2_im) |
| 160 |
: "cc", "memory", |
| 161 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 162 |
"%xmm4", "%xmm5" |
| 163 |
); |
| 164 |
} |
| 165 |
|
| 166 |
#define COMPLEX_MUL2_CORE(DST, SRC, SEL) \ |
| 167 |
__ASM_EMIT ("xor %[off], %[off]") \ |
| 168 |
/* x16 blocks */ \ |
| 169 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 170 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 171 |
__ASM_EMIT ("jb 2f") \ |
| 172 |
__ASM_EMIT ("1:") \ |
| 173 |
__ASM_EMIT ("vmovups 0x00(%[" DST "_re], %[off]), %%zmm4") /* zmm4 = ar */ \ |
| 174 |
__ASM_EMIT ("vmovups 0x00(%[" DST "_im], %[off]), %%zmm5") /* zmm5 = ai */ \ |
| 175 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%zmm0") /* zmm0 = br */ \ |
| 176 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%zmm1") /* zmm1 = bi */ \ |
| 177 |
__ASM_EMIT ("vmulps %%zmm0, %%zmm5, %%zmm2") /* ymm2 = ai*br */ \ |
| 178 |
__ASM_EMIT ("vmulps %%zmm1, %%zmm5, %%zmm3") /* ymm3 = ai*bi */ \ |
| 179 |
__ASM_EMIT ("vfmsub132ps %%zmm4, %%zmm3, %%zmm0") /* ymm0 = ar*br - ai*bi = r */ \ |
| 180 |
__ASM_EMIT ("vfmadd132ps %%zmm4, %%zmm2, %%zmm1") /* ymm1 = ar*bi + ai*br = i */ \ |
| 181 |
__ASM_EMIT ("vmovups %%zmm0, 0x00(%[" DST "_re], %[off])") \ |
| 182 |
__ASM_EMIT ("vmovups %%zmm1, 0x00(%[" DST "_im], %[off])") \ |
| 183 |
__ASM_EMIT ("add $0x40, %[off]") \ |
| 184 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 185 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 186 |
__ASM_EMIT ("jae 1b") \ |
| 187 |
__ASM_EMIT ("2:") \ |
| 188 |
/* x8 blocks */ \ |
| 189 |
__ASM_EMIT32("addl $8, %[count]") \ |
| 190 |
__ASM_EMIT64("add $8, %[count]") \ |
| 191 |
__ASM_EMIT ("jl 4f") \ |
| 192 |
__ASM_EMIT ("vmovups 0x00(%[" DST "_re], %[off]), %%ymm4") /* ymm4 = ar */ \ |
| 193 |
__ASM_EMIT ("vmovups 0x00(%[" DST "_im], %[off]), %%ymm5") /* ymm5 = ai */ \ |
| 194 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = br */ \ |
| 195 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm1") /* ymm1 = bi */ \ |
| 196 |
__ASM_EMIT ("vmulps %%ymm0, %%ymm5, %%ymm2") /* ymm2 = ai*br */ \ |
| 197 |
__ASM_EMIT ("vmulps %%ymm1, %%ymm5, %%ymm3") /* ymm3 = ai*bi */ \ |
| 198 |
__ASM_EMIT ("vfmsub132ps %%ymm4, %%ymm3, %%ymm0") /* ymm0 = ar*br - ai*bi = r */ \ |
| 199 |
__ASM_EMIT ("vfmadd132ps %%ymm4, %%ymm2, %%ymm1") /* ymm1 = ar*bi + ai*br = i */ \ |
| 200 |
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "_re], %[off])") \ |
| 201 |
__ASM_EMIT ("vmovups %%ymm1, 0x00(%[" DST "_im], %[off])") \ |
| 202 |
__ASM_EMIT ("add $0x20, %[off]") \ |
| 203 |
__ASM_EMIT32("subl $8, %[count]") \ |
| 204 |
__ASM_EMIT64("sub $8, %[count]") \ |
| 205 |
__ASM_EMIT ("jae 1b") \ |
| 206 |
__ASM_EMIT ("4:") \ |
| 207 |
/* 4x block */ \ |
| 208 |
__ASM_EMIT32("addl $4, %[count]") \ |
| 209 |
__ASM_EMIT64("add $4, %[count]") \ |
| 210 |
__ASM_EMIT ("jl 6f") \ |
| 211 |
__ASM_EMIT ("vmovups 0x00(%[" DST "_re], %[off]), %%xmm4") /* xmm4 = ar */ \ |
| 212 |
__ASM_EMIT ("vmovups 0x00(%[" DST "_im], %[off]), %%xmm5") /* xmm5 = ai */ \ |
| 213 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = br */ \ |
| 214 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%xmm1") /* xmm1 = bi */ \ |
| 215 |
__ASM_EMIT ("vmulps %%xmm0, %%xmm5, %%xmm2") /* xmm2 = ai*br */ \ |
| 216 |
__ASM_EMIT ("vmulps %%xmm1, %%xmm5, %%xmm3") /* xmm3 = ai*bi */ \ |
| 217 |
__ASM_EMIT ("vfmsub132ps %%xmm4, %%xmm3, %%xmm0") /* xmm0 = ar*br - ai*bi = r */ \ |
| 218 |
__ASM_EMIT ("vfmadd132ps %%xmm4, %%xmm2, %%xmm1") /* xmm1 = ar*bi + ai*br = i */ \ |
| 219 |
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DST "_re], %[off])") \ |
| 220 |
__ASM_EMIT ("vmovups %%xmm1, 0x00(%[" DST "_im], %[off])") \ |
| 221 |
__ASM_EMIT ("add $0x10, %[off]") \ |
| 222 |
__ASM_EMIT32("subl $4, %[count]") \ |
| 223 |
__ASM_EMIT64("sub $4, %[count]") \ |
| 224 |
__ASM_EMIT ("6:") \ |
| 225 |
/* 1x blocks */ \ |
| 226 |
__ASM_EMIT32("addl $3, %[count]") \ |
| 227 |
__ASM_EMIT64("add $3, %[count]") \ |
| 228 |
__ASM_EMIT ("jl 8f") \ |
| 229 |
__ASM_EMIT ("7:") \ |
| 230 |
__ASM_EMIT ("vmovss 0x00(%[" DST "_re], %[off]), %%xmm4") /* xmm4 = ar */ \ |
| 231 |
__ASM_EMIT ("vmovss 0x00(%[" DST "_im], %[off]), %%xmm5") /* xmm5 = ai */ \ |
| 232 |
__ASM_EMIT ("vmovss 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = br */ \ |
| 233 |
__ASM_EMIT ("vmovss 0x00(%[" SRC "_im], %[off]), %%xmm1") /* xmm1 = bi */ \ |
| 234 |
__ASM_EMIT ("vmulps %%xmm0, %%xmm5, %%xmm2") /* xmm2 = ai*br */ \ |
| 235 |
__ASM_EMIT ("vmulps %%xmm1, %%xmm5, %%xmm3") /* xmm3 = ai*bi */ \ |
| 236 |
__ASM_EMIT ("vfmsub132ss %%xmm4, %%xmm3, %%xmm0") /* xmm0 = ar*br - ai*bi = r */ \ |
| 237 |
__ASM_EMIT ("vfmadd132ss %%xmm4, %%xmm2, %%xmm1") /* xmm1 = ar*bi + ai*br = i */ \ |
| 238 |
__ASM_EMIT ("vmovss %%xmm0, 0x00(%[" DST "_re], %[off])") \ |
| 239 |
__ASM_EMIT ("vmovss %%xmm1, 0x00(%[" DST "_im], %[off])") \ |
| 240 |
__ASM_EMIT ("add $0x04, %[off]") \ |
| 241 |
__ASM_EMIT32("decl %[count]") \ |
| 242 |
__ASM_EMIT64("dec %[count]") \ |
| 243 |
__ASM_EMIT ("jge 7b") \ |
| 244 |
__ASM_EMIT ("8:") |
| 245 |
|
| 246 |
void complex_mul2(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count) |
| 247 |
{ |
| 248 |
IF_ARCH_X86(size_t off); |
| 249 |
ARCH_X86_ASM |
| 250 |
( |
| 251 |
COMPLEX_MUL2_CORE("dst", "src", FMA_OFF) |
| 252 |
: [count] X86_PGREG (count), |
| 253 |
[off] "=&r" (off) |
| 254 |
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im), |
| 255 |
[src_re] "r" (src_re), [src_im] "r" (src_im) |
| 256 |
: "cc", "memory", |
| 257 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 258 |
"%xmm4", "%xmm5" |
| 259 |
); |
| 260 |
} |
| 261 |
|
| 262 |
#define COMPLEX_MOD_CORE(DST, SRC) \ |
| 263 |
__ASM_EMIT ("xor %[off], %[off]") \ |
| 264 |
/* x64 blocks */ \ |
| 265 |
__ASM_EMIT ("sub $64, %[count]") \ |
| 266 |
__ASM_EMIT ("jb 2f") \ |
| 267 |
__ASM_EMIT ("1:") \ |
| 268 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%zmm0") /* zmm0 = r */ \ |
| 269 |
__ASM_EMIT ("vmovups 0x40(%[" SRC "_re], %[off]), %%zmm1") \ |
| 270 |
__ASM_EMIT ("vmovups 0x80(%[" SRC "_re], %[off]), %%zmm2") \ |
| 271 |
__ASM_EMIT ("vmovups 0xc0(%[" SRC "_re], %[off]), %%zmm3") \ |
| 272 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%zmm4") /* zmm4 = i */ \ |
| 273 |
__ASM_EMIT ("vmovups 0x40(%[" SRC "_im], %[off]), %%zmm5") \ |
| 274 |
__ASM_EMIT ("vmovups 0x80(%[" SRC "_im], %[off]), %%zmm6") \ |
| 275 |
__ASM_EMIT ("vmovups 0xc0(%[" SRC "_im], %[off]), %%zmm7") \ |
| 276 |
__ASM_EMIT ("vmulps %%zmm0, %%zmm0, %%zmm0") /* zmm0 = r*r */ \ |
| 277 |
__ASM_EMIT ("vmulps %%zmm1, %%zmm1, %%zmm1") \ |
| 278 |
__ASM_EMIT ("vmulps %%zmm2, %%zmm2, %%zmm2") \ |
| 279 |
__ASM_EMIT ("vmulps %%zmm3, %%zmm3, %%zmm3") \ |
| 280 |
__ASM_EMIT ("vfmadd231ps %%zmm4, %%zmm4, %%zmm0") /* zmm0 = r*r + i*i */ \ |
| 281 |
__ASM_EMIT ("vfmadd231ps %%zmm5, %%zmm5, %%zmm1") \ |
| 282 |
__ASM_EMIT ("vfmadd231ps %%zmm6, %%zmm6, %%zmm2") \ |
| 283 |
__ASM_EMIT ("vfmadd231ps %%zmm7, %%zmm7, %%zmm3") \ |
| 284 |
__ASM_EMIT ("vsqrtps %%zmm0, %%zmm0") /* zmm0 = sqrt(r*r + i*i) */ \ |
| 285 |
__ASM_EMIT ("vsqrtps %%zmm1, %%zmm1") \ |
| 286 |
__ASM_EMIT ("vsqrtps %%zmm2, %%zmm2") \ |
| 287 |
__ASM_EMIT ("vsqrtps %%zmm3, %%zmm3") \ |
| 288 |
__ASM_EMIT ("vmovups %%zmm0, 0x00(%[" DST "], %[off])") \ |
| 289 |
__ASM_EMIT ("vmovups %%zmm1, 0x40(%[" DST "], %[off])") \ |
| 290 |
__ASM_EMIT ("vmovups %%zmm2, 0x80(%[" DST "], %[off])") \ |
| 291 |
__ASM_EMIT ("vmovups %%zmm3, 0xc0(%[" DST "], %[off])") \ |
| 292 |
__ASM_EMIT ("add $0x100, %[off]") \ |
| 293 |
__ASM_EMIT ("sub $64, %[count]") \ |
| 294 |
__ASM_EMIT ("jae 1b") \ |
| 295 |
__ASM_EMIT ("2:") \ |
| 296 |
/* x32 block */ \ |
| 297 |
__ASM_EMIT ("add $32, %[count]") \ |
| 298 |
__ASM_EMIT ("jl 4f") \ |
| 299 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%zmm0") /* zmm0 = r */ \ |
| 300 |
__ASM_EMIT ("vmovups 0x40(%[" SRC "_re], %[off]), %%zmm1") \ |
| 301 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%zmm4") /* zmm4 = i */ \ |
| 302 |
__ASM_EMIT ("vmovups 0x40(%[" SRC "_im], %[off]), %%zmm5") \ |
| 303 |
__ASM_EMIT ("vmulps %%zmm0, %%zmm0, %%zmm0") /* zmm0 = r*r */ \ |
| 304 |
__ASM_EMIT ("vmulps %%zmm1, %%zmm1, %%zmm1") \ |
| 305 |
__ASM_EMIT ("vfmadd231ps %%zmm4, %%zmm4, %%zmm0") /* zmm0 = r*r + i*i */ \ |
| 306 |
__ASM_EMIT ("vfmadd231ps %%zmm5, %%zmm5, %%zmm1") \ |
| 307 |
__ASM_EMIT ("vsqrtps %%zmm0, %%zmm0") /* zmm0 = sqrt(r*r + i*i) */ \ |
| 308 |
__ASM_EMIT ("vsqrtps %%zmm1, %%zmm1") \ |
| 309 |
__ASM_EMIT ("vmovups %%zmm0, 0x00(%[" DST "], %[off])") \ |
| 310 |
__ASM_EMIT ("vmovups %%zmm1, 0x40(%[" DST "], %[off])") \ |
| 311 |
__ASM_EMIT ("sub $32, %[count]") \ |
| 312 |
__ASM_EMIT ("add $0x80, %[off]") \ |
| 313 |
__ASM_EMIT ("4:") \ |
| 314 |
/* x16 blocks */ \ |
| 315 |
__ASM_EMIT ("add $16, %[count]") \ |
| 316 |
__ASM_EMIT ("jl 6f") \ |
| 317 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = r */ \ |
| 318 |
__ASM_EMIT ("vmovups 0x20(%[" SRC "_re], %[off]), %%ymm1") \ |
| 319 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm2") /* ymm2 = i */ \ |
| 320 |
__ASM_EMIT ("vmovups 0x20(%[" SRC "_im], %[off]), %%ymm3") \ |
| 321 |
__ASM_EMIT ("vmulps %%ymm0, %%ymm0, %%ymm0") /* ymm0 = r*r */ \ |
| 322 |
__ASM_EMIT ("vmulps %%ymm1, %%ymm1, %%ymm1") \ |
| 323 |
__ASM_EMIT ("vfmadd231ps %%ymm2, %%ymm2, %%ymm0") /* ymm0 = r*r + i*i */ \ |
| 324 |
__ASM_EMIT ("vfmadd231ps %%ymm3, %%ymm3, %%ymm1") \ |
| 325 |
__ASM_EMIT ("vsqrtps %%ymm0, %%ymm0") /* ymm0 = sqrt(r*r + i*i) */ \ |
| 326 |
__ASM_EMIT ("vsqrtps %%ymm1, %%ymm1") \ |
| 327 |
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "], %[off])") \ |
| 328 |
__ASM_EMIT ("vmovups %%ymm1, 0x20(%[" DST "], %[off])") \ |
| 329 |
__ASM_EMIT ("sub $16, %[count]") \ |
| 330 |
__ASM_EMIT ("add $0x40, %[off]") \ |
| 331 |
__ASM_EMIT ("6:") \ |
| 332 |
/* x8 block */ \ |
| 333 |
__ASM_EMIT ("add $8, %[count]") \ |
| 334 |
__ASM_EMIT ("jl 8f") \ |
| 335 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = r */ \ |
| 336 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm2") /* ymm2 = i */ \ |
| 337 |
__ASM_EMIT ("vmulps %%ymm0, %%ymm0, %%ymm0") /* ymm0 = r*r */ \ |
| 338 |
__ASM_EMIT ("vfmadd231ps %%ymm2, %%ymm2, %%ymm0") /* ymm0 = r*r + i*i */ \ |
| 339 |
__ASM_EMIT ("vsqrtps %%ymm0, %%ymm0") /* ymm0 = sqrt(r*r + i*i) */ \ |
| 340 |
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "], %[off])") \ |
| 341 |
__ASM_EMIT ("add $0x20, %[off]") \ |
| 342 |
__ASM_EMIT ("sub $8, %[count]") \ |
| 343 |
__ASM_EMIT ("8:") \ |
| 344 |
/* x4 block */ \ |
| 345 |
__ASM_EMIT ("add $4, %[count]") \ |
| 346 |
__ASM_EMIT ("jl 10f") \ |
| 347 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = r */ \ |
| 348 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = i */ \ |
| 349 |
__ASM_EMIT ("vmulps %%xmm0, %%xmm0, %%xmm0") /* xmm0 = r*r */ \ |
| 350 |
__ASM_EMIT ("vfmadd231ps %%xmm2, %%xmm2, %%xmm0") /* xmm0 = r*r + i*i */ \ |
| 351 |
__ASM_EMIT ("vsqrtps %%xmm0, %%xmm0") /* xmm0 = sqrt(r*r + i*i) */ \ |
| 352 |
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DST "], %[off])") \ |
| 353 |
__ASM_EMIT ("add $0x10, %[off]") \ |
| 354 |
__ASM_EMIT ("sub $4, %[count]") \ |
| 355 |
__ASM_EMIT ("10:") \ |
| 356 |
/* x1 block */ \ |
| 357 |
__ASM_EMIT ("add $3, %[count]") \ |
| 358 |
__ASM_EMIT ("jl 12f") \ |
| 359 |
__ASM_EMIT ("11:") \ |
| 360 |
__ASM_EMIT ("vmovss 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = r */ \ |
| 361 |
__ASM_EMIT ("vmovss 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = i */ \ |
| 362 |
__ASM_EMIT ("vmulss %%xmm0, %%xmm0, %%xmm0") /* xmm0 = r*r */ \ |
| 363 |
__ASM_EMIT ("vfmadd231ps %%xmm2, %%xmm2, %%xmm0") /* xmm0 = r*r + i*i */ \ |
| 364 |
__ASM_EMIT ("vsqrtss %%xmm0, %%xmm0, %%xmm0") /* xmm0 = sqrt(r*r + i*i) */ \ |
| 365 |
__ASM_EMIT ("vmovss %%xmm0, 0x00(%[" DST "], %[off])") \ |
| 366 |
__ASM_EMIT ("add $0x04, %[off]") \ |
| 367 |
__ASM_EMIT ("dec %[count]") \ |
| 368 |
__ASM_EMIT ("jge 11b") \ |
| 369 |
__ASM_EMIT ("12:") \ |
| 370 |
|
| 371 |
void complex_mod(float *dst, const float *src_re, const float *src_im, size_t count) |
| 372 |
{ |
| 373 |
IF_ARCH_X86(size_t off); |
| 374 |
ARCH_X86_ASM |
| 375 |
( |
| 376 |
COMPLEX_MOD_CORE("dst", "src") |
| 377 |
: [count] "+r" (count), [off] "=&r" (off) |
| 378 |
: [dst] "r" (dst), [src_re] "r" (src_re), [src_im] "r" (src_im) |
| 379 |
: "cc", "memory", |
| 380 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 381 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 382 |
); |
| 383 |
} |
| 384 |
|
| 385 |
IF_ARCH_X86( |
| 386 |
static uint32_t complex_div_const[] __lsp_aligned64 = |
| 387 |
{ |
| 388 |
LSP_DSP_VEC16(0x3f800000), // 1.0 |
| 389 |
LSP_DSP_VEC16(0x80000000) // signbit |
| 390 |
}; |
| 391 |
) |
| 392 |
|
| 393 |
#define COMPLEX_DIV3_CORE(DSTR, DSTI, TR, TI, BR, BI) \ |
| 394 |
__ASM_EMIT ("xor %[off], %[off]") \ |
| 395 |
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%zmm6") /* zmm6 = 1 */ \ |
| 396 |
/* x16 blocks */ \ |
| 397 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 398 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 399 |
__ASM_EMIT ("jb 2f") \ |
| 400 |
__ASM_EMIT ("1:") \ |
| 401 |
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%zmm2") /* zmm2 = br */ \ |
| 402 |
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%zmm3") /* zmm3 = bi */ \ |
| 403 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%zmm0") /* zmm0 = tr */ \ |
| 404 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%zmm1") /* zmm1 = ti */ \ |
| 405 |
__ASM_EMIT64("vmovups 0x00(%[" TR "], %[off]), %%zmm0") /* zmm0 = tr */ \ |
| 406 |
__ASM_EMIT64("vmovups 0x00(%[" TI "], %[off]), %%zmm1") /* zmm1 = ti */ \ |
| 407 |
__ASM_EMIT32("mov %[" DSTR "], %[ptr_re]") \ |
| 408 |
__ASM_EMIT32("mov %[" DSTI "], %[ptr_im]") \ |
| 409 |
__ASM_EMIT ("vmulps %%zmm2, %%zmm2, %%zmm7") /* zmm7 = br*br */ \ |
| 410 |
__ASM_EMIT ("vmulps %%zmm3, %%zmm1, %%zmm4") /* zmm4 = ti*bi */ \ |
| 411 |
__ASM_EMIT ("vmulps %%zmm3, %%zmm0, %%zmm5") /* zmm5 = tr*bi */ \ |
| 412 |
__ASM_EMIT ("vfmadd231ps %%zmm3, %%zmm3, %%zmm7") /* zmm7 = M = br*br + bi*bi */ \ |
| 413 |
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%zmm6") /* zmm6 = 1 */ \ |
| 414 |
__ASM_EMIT ("vdivps %%zmm7, %%zmm6, %%zmm7") /* zmm7 = R = 1/M */ \ |
| 415 |
__ASM_EMIT ("vfmadd132ps %%zmm2, %%zmm4, %%zmm0") /* zmm0 = r = tr*br + ti*bi */ \ |
| 416 |
__ASM_EMIT ("vfmsub132ps %%zmm2, %%zmm5, %%zmm1") /* zmm1 = i = tr*br - tr*bi */ \ |
| 417 |
__ASM_EMIT ("vmulps %%zmm7, %%zmm0, %%zmm0") /* zmm0 = r/R */ \ |
| 418 |
__ASM_EMIT ("vmulps %%zmm7, %%zmm1, %%zmm1") /* zmm1 = i/R */ \ |
| 419 |
__ASM_EMIT32("vmovups %%zmm0, 0x00(%[ptr_re], %[off])") \ |
| 420 |
__ASM_EMIT32("vmovups %%zmm1, 0x00(%[ptr_re], %[off])") \ |
| 421 |
__ASM_EMIT64("vmovups %%zmm0, 0x00(%[" DSTR "], %[off])") \ |
| 422 |
__ASM_EMIT64("vmovups %%zmm1, 0x00(%[" DSTI "], %[off])") \ |
| 423 |
__ASM_EMIT ("add $0x40, %[off]") \ |
| 424 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 425 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 426 |
__ASM_EMIT ("jae 1b") \ |
| 427 |
__ASM_EMIT ("2:") \ |
| 428 |
/* x8 blocks */ \ |
| 429 |
__ASM_EMIT32("addl $8, %[count]") \ |
| 430 |
__ASM_EMIT64("add $8, %[count]") \ |
| 431 |
__ASM_EMIT ("jl 4f") \ |
| 432 |
__ASM_EMIT32("mov %[" TR "], %[ptr_re]") \ |
| 433 |
__ASM_EMIT32("mov %[" TI "], %[ptr_im]") \ |
| 434 |
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%ymm2") /* ymm2 = br */ \ |
| 435 |
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%ymm3") /* ymm3 = bi */ \ |
| 436 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%ymm0") /* ymm0 = tr */ \ |
| 437 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%ymm1") /* ymm1 = ti */ \ |
| 438 |
__ASM_EMIT64("vmovups 0x00(%[" TR "], %[off]), %%ymm0") /* ymm0 = tr */ \ |
| 439 |
__ASM_EMIT64("vmovups 0x00(%[" TI "], %[off]), %%ymm1") /* ymm1 = ti */ \ |
| 440 |
__ASM_EMIT32("mov %[" DSTR "], %[ptr_re]") \ |
| 441 |
__ASM_EMIT32("mov %[" DSTI "], %[ptr_im]") \ |
| 442 |
__ASM_EMIT ("vmulps %%ymm2, %%ymm2, %%ymm7") /* ymm7 = br*br */ \ |
| 443 |
__ASM_EMIT ("vmulps %%ymm3, %%ymm1, %%ymm4") /* ymm4 = ti*bi */ \ |
| 444 |
__ASM_EMIT ("vmulps %%ymm3, %%ymm0, %%ymm5") /* ymm5 = tr*bi */ \ |
| 445 |
__ASM_EMIT ("vfmadd231ps %%ymm3, %%ymm3, %%ymm7") /* ymm7 = M = br*br + bi*bi */ \ |
| 446 |
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%ymm6") /* ymm6 = 1 */ \ |
| 447 |
__ASM_EMIT ("vdivps %%ymm7, %%ymm6, %%ymm7") /* ymm7 = R = 1/M */ \ |
| 448 |
__ASM_EMIT ("vfmadd132ps %%ymm2, %%ymm4, %%ymm0") /* ymm0 = r = tr*br + ti*bi */ \ |
| 449 |
__ASM_EMIT ("vfmsub132ps %%ymm2, %%ymm5, %%ymm1") /* ymm1 = i = tr*br - tr*bi */ \ |
| 450 |
__ASM_EMIT ("vmulps %%ymm7, %%ymm0, %%ymm0") /* ymm0 = r/R */ \ |
| 451 |
__ASM_EMIT ("vmulps %%ymm7, %%ymm1, %%ymm1") /* ymm1 = i/R */ \ |
| 452 |
__ASM_EMIT32("vmovups %%ymm0, 0x00(%[ptr_re], %[off])") \ |
| 453 |
__ASM_EMIT32("vmovups %%ymm1, 0x00(%[ptr_re], %[off])") \ |
| 454 |
__ASM_EMIT64("vmovups %%ymm0, 0x00(%[" DSTR "], %[off])") \ |
| 455 |
__ASM_EMIT64("vmovups %%ymm1, 0x00(%[" DSTI "], %[off])") \ |
| 456 |
__ASM_EMIT32("subl $8, %[count]") \ |
| 457 |
__ASM_EMIT64("sub $8, %[count]") \ |
| 458 |
__ASM_EMIT ("add $0x20, %[off]") \ |
| 459 |
__ASM_EMIT ("4:") \ |
| 460 |
/* 4x block */ \ |
| 461 |
__ASM_EMIT32("addl $4, %[count]") \ |
| 462 |
__ASM_EMIT64("add $4, %[count]") \ |
| 463 |
__ASM_EMIT ("jl 6f") \ |
| 464 |
__ASM_EMIT32("mov %[" TR "], %[ptr_re]") \ |
| 465 |
__ASM_EMIT32("mov %[" TI "], %[ptr_im]") \ |
| 466 |
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%xmm2") /* xmm2 = br */ \ |
| 467 |
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%xmm3") /* xmm3 = bi */ \ |
| 468 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%xmm0") /* xmm0 = tr */ \ |
| 469 |
__ASM_EMIT32("vmovups 0x00(%[ptr_re], %[off]), %%xmm1") /* xmm1 = ti */ \ |
| 470 |
__ASM_EMIT64("vmovups 0x00(%[" TR "], %[off]), %%xmm0") /* xmm0 = tr */ \ |
| 471 |
__ASM_EMIT64("vmovups 0x00(%[" TI "], %[off]), %%xmm1") /* xmm1 = ti */ \ |
| 472 |
__ASM_EMIT32("mov %[" DSTR "], %[ptr_re]") \ |
| 473 |
__ASM_EMIT32("mov %[" DSTI "], %[ptr_im]") \ |
| 474 |
__ASM_EMIT ("vmulps %%xmm2, %%xmm2, %%xmm7") /* xmm7 = br*br */ \ |
| 475 |
__ASM_EMIT ("vmulps %%xmm3, %%xmm1, %%xmm4") /* xmm4 = ti*bi */ \ |
| 476 |
__ASM_EMIT ("vmulps %%xmm3, %%xmm0, %%xmm5") /* xmm5 = tr*bi */ \ |
| 477 |
__ASM_EMIT ("vfmadd231ps %%xmm3, %%xmm3, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \ |
| 478 |
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \ |
| 479 |
__ASM_EMIT ("vdivps %%xmm7, %%xmm6, %%xmm7") /* xmm7 = R = 1/M */ \ |
| 480 |
__ASM_EMIT ("vfmadd132ps %%xmm2, %%xmm4, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \ |
| 481 |
__ASM_EMIT ("vfmsub132ps %%xmm2, %%xmm5, %%xmm1") /* xmm1 = i = tr*br - tr*bi */ \ |
| 482 |
__ASM_EMIT ("vmulps %%xmm7, %%xmm0, %%xmm0") /* xmm0 = r/R */ \ |
| 483 |
__ASM_EMIT ("vmulps %%xmm7, %%xmm1, %%xmm1") /* xmm1 = i/R */ \ |
| 484 |
__ASM_EMIT32("vmovups %%xmm0, 0x00(%[ptr_re], %[off])") \ |
| 485 |
__ASM_EMIT32("vmovups %%xmm1, 0x00(%[ptr_re], %[off])") \ |
| 486 |
__ASM_EMIT64("vmovups %%xmm0, 0x00(%[" DSTR "], %[off])") \ |
| 487 |
__ASM_EMIT64("vmovups %%xmm1, 0x00(%[" DSTI "], %[off])") \ |
| 488 |
__ASM_EMIT32("subl $4, %[count]") \ |
| 489 |
__ASM_EMIT64("sub $4, %[count]") \ |
| 490 |
__ASM_EMIT ("add $0x10, %[off]") \ |
| 491 |
__ASM_EMIT ("6:") \ |
| 492 |
/* 1x blocks */ \ |
| 493 |
__ASM_EMIT32("addl $3, %[count]") \ |
| 494 |
__ASM_EMIT64("add $3, %[count]") \ |
| 495 |
__ASM_EMIT ("jl 8f") \ |
| 496 |
__ASM_EMIT ("7:") \ |
| 497 |
__ASM_EMIT32("mov %[" TR "], %[ptr_re]") \ |
| 498 |
__ASM_EMIT32("mov %[" TI "], %[ptr_im]") \ |
| 499 |
__ASM_EMIT ("vmovss 0x00(%[" BR "], %[off]), %%xmm2") /* xmm2 = br */ \ |
| 500 |
__ASM_EMIT ("vmovss 0x00(%[" BI "], %[off]), %%xmm3") /* xmm3 = bi */ \ |
| 501 |
__ASM_EMIT32("vmovss 0x00(%[ptr_re], %[off]), %%xmm0") /* xmm0 = tr */ \ |
| 502 |
__ASM_EMIT32("vmovss 0x00(%[ptr_re], %[off]), %%xmm1") /* xmm1 = ti */ \ |
| 503 |
__ASM_EMIT64("vmovss 0x00(%[" TR "], %[off]), %%xmm0") /* xmm0 = tr */ \ |
| 504 |
__ASM_EMIT64("vmovss 0x00(%[" TI "], %[off]), %%xmm1") /* xmm1 = ti */ \ |
| 505 |
__ASM_EMIT32("mov %[" DSTR "], %[ptr_re]") \ |
| 506 |
__ASM_EMIT32("mov %[" DSTI "], %[ptr_im]") \ |
| 507 |
__ASM_EMIT ("vmulss %%xmm2, %%xmm2, %%xmm7") /* xmm7 = br*br */ \ |
| 508 |
__ASM_EMIT ("vmulss %%xmm3, %%xmm1, %%xmm4") /* xmm4 = ti*bi */ \ |
| 509 |
__ASM_EMIT ("vmulss %%xmm3, %%xmm0, %%xmm5") /* xmm5 = tr*bi */ \ |
| 510 |
__ASM_EMIT ("vfmadd231ss %%xmm3, %%xmm3, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \ |
| 511 |
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \ |
| 512 |
__ASM_EMIT ("vdivss %%xmm7, %%xmm6, %%xmm7") /* xmm7 = R = 1/M */ \ |
| 513 |
__ASM_EMIT ("vfmadd132ss %%xmm2, %%xmm4, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \ |
| 514 |
__ASM_EMIT ("vfmsub132ss %%xmm2, %%xmm5, %%xmm1") /* xmm1 = i = tr*br - tr*bi */ \ |
| 515 |
__ASM_EMIT ("vmulss %%xmm7, %%xmm0, %%xmm0") /* xmm0 = r/R */ \ |
| 516 |
__ASM_EMIT ("vmulss %%xmm7, %%xmm1, %%xmm1") /* xmm1 = i/R */ \ |
| 517 |
__ASM_EMIT32("vmovss %%xmm0, 0x00(%[ptr_re], %[off])") \ |
| 518 |
__ASM_EMIT32("vmovss %%xmm1, 0x00(%[ptr_re], %[off])") \ |
| 519 |
__ASM_EMIT64("vmovss %%xmm0, 0x00(%[" DSTR "], %[off])") \ |
| 520 |
__ASM_EMIT64("vmovss %%xmm1, 0x00(%[" DSTI "], %[off])") \ |
| 521 |
__ASM_EMIT ("add $0x04, %[off]") \ |
| 522 |
__ASM_EMIT32("decl %[count]") \ |
| 523 |
__ASM_EMIT64("dec %[count]") \ |
| 524 |
__ASM_EMIT ("jge 7b") \ |
| 525 |
__ASM_EMIT ("8:") |
| 526 |
|
| 527 |
void complex_div3(float *dst_re, float *dst_im, |
| 528 |
const float *src1_re, const float *src1_im, |
| 529 |
const float *src2_re, const float *src2_im, |
| 530 |
size_t count) |
| 531 |
{ |
| 532 |
IF_ARCH_I386(float *ptr_re, *ptr_im); |
| 533 |
IF_ARCH_X86(size_t off); |
| 534 |
ARCH_X86_ASM |
| 535 |
( |
| 536 |
COMPLEX_DIV3_CORE("dst_re", "dst_im", "src1_re", "src1_im", "src2_re", "src2_im") |
| 537 |
: __IF_32( |
| 538 |
[ptr_re] "=&r" (ptr_re), [ptr_im] "=&r" (ptr_im), |
| 539 |
) |
| 540 |
[count] X86_PGREG (count), |
| 541 |
[off] "=&r" (off) |
| 542 |
: [dst_re] X86_GREG (dst_re), [dst_im] X86_GREG (dst_im), |
| 543 |
[src1_re] X86_GREG (src1_re), [src1_im] X86_GREG (src1_im), |
| 544 |
[src2_re] "r" (src2_re), [src2_im] "r" (src2_im), |
| 545 |
[CC] "o" (complex_div_const) |
| 546 |
: "cc", "memory", |
| 547 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 548 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 549 |
); |
| 550 |
} |
| 551 |
|
| 552 |
#undef COMPLEX_DIV3_CORE |
| 553 |
|
| 554 |
#define COMPLEX_DIV2_CORE(DSTR, DSTI, TR, TI, BR, BI) \ |
| 555 |
__ASM_EMIT ("xor %[off], %[off]") \ |
| 556 |
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%zmm6") /* zmm6 = 1 */ \ |
| 557 |
/* x16 blocks */ \ |
| 558 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 559 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 560 |
__ASM_EMIT ("jb 2f") \ |
| 561 |
__ASM_EMIT ("1:") \ |
| 562 |
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%zmm2") /* zmm2 = br */ \ |
| 563 |
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%zmm3") /* zmm3 = bi */ \ |
| 564 |
__ASM_EMIT ("vmovups 0x00(%[" TR "], %[off]), %%zmm0") /* zmm0 = tr */ \ |
| 565 |
__ASM_EMIT ("vmovups 0x00(%[" TI "], %[off]), %%zmm1") /* zmm1 = ti */ \ |
| 566 |
__ASM_EMIT ("vmulps %%zmm2, %%zmm2, %%zmm7") /* zmm7 = br*br */ \ |
| 567 |
__ASM_EMIT ("vmulps %%zmm3, %%zmm1, %%zmm4") /* zmm4 = ti*bi */ \ |
| 568 |
__ASM_EMIT ("vmulps %%zmm3, %%zmm0, %%zmm5") /* zmm5 = tr*bi */ \ |
| 569 |
__ASM_EMIT ("vfmadd231ps %%zmm3, %%zmm3, %%zmm7") /* zmm7 = M = br*br + bi*bi */ \ |
| 570 |
__ASM_EMIT ("vdivps %%zmm7, %%zmm6, %%zmm7") /* zmm7 = R = 1/M */ \ |
| 571 |
__ASM_EMIT ("vfmadd132ps %%zmm2, %%zmm4, %%zmm0") /* zmm0 = r = tr*br + ti*bi */ \ |
| 572 |
__ASM_EMIT ("vfmsub132ps %%zmm2, %%zmm5, %%zmm1") /* zmm1 = i = tr*br - tr*bi */ \ |
| 573 |
__ASM_EMIT ("vmulps %%zmm7, %%zmm0, %%zmm0") /* zmm0 = r/R */ \ |
| 574 |
__ASM_EMIT ("vmulps %%zmm7, %%zmm1, %%zmm1") /* zmm1 = i/R */ \ |
| 575 |
__ASM_EMIT ("vmovups %%zmm0, 0x00(%[" DSTR "], %[off])") \ |
| 576 |
__ASM_EMIT ("vmovups %%zmm1, 0x00(%[" DSTI "], %[off])") \ |
| 577 |
__ASM_EMIT ("add $0x40, %[off]") \ |
| 578 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 579 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 580 |
__ASM_EMIT ("jae 1b") \ |
| 581 |
__ASM_EMIT ("2:") \ |
| 582 |
/* x8 block */ \ |
| 583 |
__ASM_EMIT32("addl $8, %[count]") \ |
| 584 |
__ASM_EMIT64("add $8, %[count]") \ |
| 585 |
__ASM_EMIT ("jl 4f") \ |
| 586 |
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%ymm2") /* ymm2 = br */ \ |
| 587 |
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%ymm3") /* ymm3 = bi */ \ |
| 588 |
__ASM_EMIT ("vmovups 0x00(%[" TR "], %[off]), %%ymm0") /* ymm0 = tr */ \ |
| 589 |
__ASM_EMIT ("vmovups 0x00(%[" TI "], %[off]), %%ymm1") /* ymm1 = ti */ \ |
| 590 |
__ASM_EMIT ("vmulps %%ymm2, %%ymm2, %%ymm7") /* ymm7 = br*br */ \ |
| 591 |
__ASM_EMIT ("vmulps %%ymm3, %%ymm1, %%ymm4") /* ymm4 = ti*bi */ \ |
| 592 |
__ASM_EMIT ("vmulps %%ymm3, %%ymm0, %%ymm5") /* ymm5 = tr*bi */ \ |
| 593 |
__ASM_EMIT ("vfmadd231ps %%ymm3, %%ymm3, %%ymm7") /* ymm7 = M = br*br + bi*bi */ \ |
| 594 |
__ASM_EMIT ("vdivps %%ymm7, %%ymm6, %%ymm7") /* ymm7 = R = 1/M */ \ |
| 595 |
__ASM_EMIT ("vfmadd132ps %%ymm2, %%ymm4, %%ymm0") /* ymm0 = r = tr*br + ti*bi */ \ |
| 596 |
__ASM_EMIT ("vfmsub132ps %%ymm2, %%ymm5, %%ymm1") /* ymm1 = i = tr*br - tr*bi */ \ |
| 597 |
__ASM_EMIT ("vmulps %%ymm7, %%ymm0, %%ymm0") /* ymm0 = r/R */ \ |
| 598 |
__ASM_EMIT ("vmulps %%ymm7, %%ymm1, %%ymm1") /* ymm1 = i/R */ \ |
| 599 |
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DSTR "], %[off])") \ |
| 600 |
__ASM_EMIT ("vmovups %%ymm1, 0x00(%[" DSTI "], %[off])") \ |
| 601 |
__ASM_EMIT32("subl $8, %[count]") \ |
| 602 |
__ASM_EMIT64("sub $8, %[count]") \ |
| 603 |
__ASM_EMIT ("add $0x20, %[off]") \ |
| 604 |
__ASM_EMIT ("4:") \ |
| 605 |
/* 4x block */ \ |
| 606 |
__ASM_EMIT32("addl $4, %[count]") \ |
| 607 |
__ASM_EMIT64("add $4, %[count]") \ |
| 608 |
__ASM_EMIT ("jl 6f") \ |
| 609 |
__ASM_EMIT ("vmovups 0x00(%[" BR "], %[off]), %%xmm2") /* xmm2 = br */ \ |
| 610 |
__ASM_EMIT ("vmovups 0x00(%[" BI "], %[off]), %%xmm3") /* xmm3 = bi */ \ |
| 611 |
__ASM_EMIT ("vmovups 0x00(%[" TR "], %[off]), %%xmm0") /* xmm0 = tr */ \ |
| 612 |
__ASM_EMIT ("vmovups 0x00(%[" TI "], %[off]), %%xmm1") /* xmm1 = ti */ \ |
| 613 |
__ASM_EMIT ("vmulps %%xmm2, %%xmm2, %%xmm7") /* xmm7 = br*br */ \ |
| 614 |
__ASM_EMIT ("vmulps %%xmm3, %%xmm1, %%xmm4") /* xmm4 = ti*bi */ \ |
| 615 |
__ASM_EMIT ("vmulps %%xmm3, %%xmm0, %%xmm5") /* xmm5 = tr*bi */ \ |
| 616 |
__ASM_EMIT ("vfmadd231ps %%xmm3, %%xmm3, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \ |
| 617 |
__ASM_EMIT ("vdivps %%xmm7, %%xmm6, %%xmm7") /* xmm7 = R = 1/M */ \ |
| 618 |
__ASM_EMIT ("vfmadd132ps %%xmm2, %%xmm4, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \ |
| 619 |
__ASM_EMIT ("vfmsub132ps %%xmm2, %%xmm5, %%xmm1") /* xmm1 = i = tr*br - tr*bi */ \ |
| 620 |
__ASM_EMIT ("vmulps %%xmm7, %%xmm0, %%xmm0") /* xmm0 = r/R */ \ |
| 621 |
__ASM_EMIT ("vmulps %%xmm7, %%xmm1, %%xmm1") /* xmm1 = i/R */ \ |
| 622 |
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DSTR "], %[off])") \ |
| 623 |
__ASM_EMIT ("vmovups %%xmm1, 0x00(%[" DSTI "], %[off])") \ |
| 624 |
__ASM_EMIT32("subl $4, %[count]") \ |
| 625 |
__ASM_EMIT64("sub $4, %[count]") \ |
| 626 |
__ASM_EMIT ("add $0x10, %[off]") \ |
| 627 |
__ASM_EMIT ("6:") \ |
| 628 |
/* 1x blocks */ \ |
| 629 |
__ASM_EMIT32("addl $3, %[count]") \ |
| 630 |
__ASM_EMIT64("add $3, %[count]") \ |
| 631 |
__ASM_EMIT ("jl 8f") \ |
| 632 |
__ASM_EMIT ("7:") \ |
| 633 |
__ASM_EMIT ("vmovss 0x00(%[" BR "], %[off]), %%xmm2") /* xmm2 = br */ \ |
| 634 |
__ASM_EMIT ("vmovss 0x00(%[" BI "], %[off]), %%xmm3") /* xmm3 = bi */ \ |
| 635 |
__ASM_EMIT ("vmovss 0x00(%[" TR "], %[off]), %%xmm0") /* xmm0 = tr */ \ |
| 636 |
__ASM_EMIT ("vmovss 0x00(%[" TI "], %[off]), %%xmm1") /* xmm1 = ti */ \ |
| 637 |
__ASM_EMIT ("vmulss %%xmm2, %%xmm2, %%xmm7") /* xmm7 = br*br */ \ |
| 638 |
__ASM_EMIT ("vmulss %%xmm3, %%xmm1, %%xmm4") /* xmm4 = ti*bi */ \ |
| 639 |
__ASM_EMIT ("vmulss %%xmm3, %%xmm0, %%xmm5") /* xmm5 = tr*bi */ \ |
| 640 |
__ASM_EMIT ("vfmadd231ss %%xmm3, %%xmm3, %%xmm7") /* xmm7 = M = br*br + bi*bi */ \ |
| 641 |
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%xmm6") /* xmm6 = 1 */ \ |
| 642 |
__ASM_EMIT ("vdivss %%xmm7, %%xmm6, %%xmm7") /* xmm7 = R = 1/M */ \ |
| 643 |
__ASM_EMIT ("vfmadd132ss %%xmm2, %%xmm4, %%xmm0") /* xmm0 = r = tr*br + ti*bi */ \ |
| 644 |
__ASM_EMIT ("vfmsub132ss %%xmm2, %%xmm5, %%xmm1") /* xmm1 = i = tr*br - tr*bi */ \ |
| 645 |
__ASM_EMIT ("vmulss %%xmm7, %%xmm0, %%xmm0") /* xmm0 = r/R */ \ |
| 646 |
__ASM_EMIT ("vmulss %%xmm7, %%xmm1, %%xmm1") /* xmm1 = i/R */ \ |
| 647 |
__ASM_EMIT ("vmovss %%xmm0, 0x00(%[" DSTR "], %[off])") \ |
| 648 |
__ASM_EMIT ("vmovss %%xmm1, 0x00(%[" DSTI "], %[off])") \ |
| 649 |
__ASM_EMIT ("add $0x04, %[off]") \ |
| 650 |
__ASM_EMIT32("decl %[count]") \ |
| 651 |
__ASM_EMIT64("dec %[count]") \ |
| 652 |
__ASM_EMIT ("jge 7b") \ |
| 653 |
__ASM_EMIT ("8:") |
| 654 |
|
| 655 |
void complex_div2(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count) |
| 656 |
{ |
| 657 |
IF_ARCH_X86(size_t off); |
| 658 |
ARCH_X86_ASM |
| 659 |
( |
| 660 |
COMPLEX_DIV2_CORE("dst_re", "dst_im", "dst_re", "dst_im", "src_re", "src_im") |
| 661 |
: [count] X86_PGREG (count), |
| 662 |
[off] "=&r" (off) |
| 663 |
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im), |
| 664 |
[src_re] "r" (src_re), [src_im] "r" (src_im), |
| 665 |
[CC] "o" (complex_div_const) |
| 666 |
: "cc", "memory", |
| 667 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 668 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 669 |
); |
| 670 |
} |
| 671 |
|
| 672 |
void complex_rdiv2(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count) |
| 673 |
{ |
| 674 |
IF_ARCH_X86(size_t off); |
| 675 |
ARCH_X86_ASM |
| 676 |
( |
| 677 |
COMPLEX_DIV2_CORE("dst_re", "dst_im", "src_re", "src_im", "dst_re", "dst_im") |
| 678 |
: [count] X86_PGREG (count), |
| 679 |
[off] "=&r" (off) |
| 680 |
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im), |
| 681 |
[src_re] "r" (src_re), [src_im] "r" (src_im), |
| 682 |
[CC] "o" (complex_div_const) |
| 683 |
: "cc", "memory", |
| 684 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 685 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 686 |
); |
| 687 |
} |
| 688 |
|
| 689 |
#undef COMPLEX_DIV2_CORE |
| 690 |
|
| 691 |
#define COMPLEX_RCP_CORE(DST, SRC, SEL) \ |
| 692 |
__ASM_EMIT ("xor %[off], %[off]") \ |
| 693 |
__ASM_EMIT ("vmovaps 0x00 + %[CC], %%zmm6") /* zmm6 = 1 */ \ |
| 694 |
/* x32 blocks */ \ |
| 695 |
__ASM_EMIT32("subl $32, %[count]") \ |
| 696 |
__ASM_EMIT64("sub $32, %[count]") \ |
| 697 |
__ASM_EMIT ("jb 2f") \ |
| 698 |
__ASM_EMIT ("1:") \ |
| 699 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%zmm0") /* zmm0 = ar */ \ |
| 700 |
__ASM_EMIT ("vmovups 0x40(%[" SRC "_re], %[off]), %%zmm1") \ |
| 701 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%zmm2") /* zmm2 = ai */ \ |
| 702 |
__ASM_EMIT ("vmovups 0x40(%[" SRC "_im], %[off]), %%zmm3") \ |
| 703 |
__ASM_EMIT ("vmulps %%zmm0, %%zmm0, %%zmm4") /* zmm4 = ar*ar */ \ |
| 704 |
__ASM_EMIT ("vmulps %%zmm1, %%zmm1, %%zmm5") \ |
| 705 |
__ASM_EMIT ("vfmadd231ps %%zmm2, %%zmm2, %%zmm4") /* zmm4 = R = ar*ar+ai*ai */ \ |
| 706 |
__ASM_EMIT ("vfmadd231ps %%zmm3, %%zmm3, %%zmm5") \ |
| 707 |
__ASM_EMIT ("vdivps %%zmm4, %%zmm6, %%zmm4") /* zmm4 = 1/R */ \ |
| 708 |
__ASM_EMIT ("vdivps %%zmm5, %%zmm6, %%zmm5") \ |
| 709 |
__ASM_EMIT ("vpxord 0x40 + %[CC], %%zmm2, %%zmm2") /* zmm2 = -ai */ \ |
| 710 |
__ASM_EMIT ("vpxord 0x40 + %[CC], %%zmm3, %%zmm3") \ |
| 711 |
__ASM_EMIT ("vmulps %%zmm4, %%zmm0, %%zmm0") /* zmm0 = ar/R */ \ |
| 712 |
__ASM_EMIT ("vmulps %%zmm5, %%zmm1, %%zmm1") \ |
| 713 |
__ASM_EMIT ("vmulps %%zmm4, %%zmm2, %%zmm2") /* zmm1 = -ai/R */ \ |
| 714 |
__ASM_EMIT ("vmulps %%zmm5, %%zmm3, %%zmm3") \ |
| 715 |
__ASM_EMIT ("vmovups %%zmm0, 0x00(%[" DST "_re], %[off])") \ |
| 716 |
__ASM_EMIT ("vmovups %%zmm1, 0x40(%[" DST "_re], %[off])") \ |
| 717 |
__ASM_EMIT ("vmovups %%zmm2, 0x00(%[" DST "_im], %[off])") \ |
| 718 |
__ASM_EMIT ("vmovups %%zmm3, 0x40(%[" DST "_im], %[off])") \ |
| 719 |
__ASM_EMIT ("add $0x80, %[off]") \ |
| 720 |
__ASM_EMIT32("subl $32, %[count]") \ |
| 721 |
__ASM_EMIT64("sub $32, %[count]") \ |
| 722 |
__ASM_EMIT ("jae 1b") \ |
| 723 |
__ASM_EMIT ("2:") \ |
| 724 |
/* x16 blocks */ \ |
| 725 |
__ASM_EMIT32("addl $16, %[count]") \ |
| 726 |
__ASM_EMIT64("add $16, %[count]") \ |
| 727 |
__ASM_EMIT ("jl 4f") \ |
| 728 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%ymm0") /* ymm0 = ar */ \ |
| 729 |
__ASM_EMIT ("vmovups 0x20(%[" SRC "_re], %[off]), %%ymm1") \ |
| 730 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%ymm2") /* ymm2 = ai */ \ |
| 731 |
__ASM_EMIT ("vmovups 0x20(%[" SRC "_im], %[off]), %%ymm3") \ |
| 732 |
__ASM_EMIT ("vmulps %%ymm0, %%ymm0, %%ymm4") /* ymm4 = ar*ar */ \ |
| 733 |
__ASM_EMIT ("vmulps %%ymm1, %%ymm1, %%ymm5") \ |
| 734 |
__ASM_EMIT ("vfmadd231ps %%ymm2, %%ymm2, %%ymm4") /* ymm4 = R = ar*ar+ai*ai */ \ |
| 735 |
__ASM_EMIT ("vfmadd231ps %%ymm3, %%ymm3, %%ymm5") \ |
| 736 |
__ASM_EMIT ("vdivps %%ymm4, %%ymm6, %%ymm4") /* ymm4 = 1/R */ \ |
| 737 |
__ASM_EMIT ("vdivps %%ymm5, %%ymm6, %%ymm5") \ |
| 738 |
__ASM_EMIT ("vpxord 0x40 + %[CC], %%ymm2, %%ymm2") /* ymm2 = -ai */ \ |
| 739 |
__ASM_EMIT ("vpxord 0x40 + %[CC], %%ymm3, %%ymm3") \ |
| 740 |
__ASM_EMIT ("vmulps %%ymm4, %%ymm0, %%ymm0") /* ymm0 = ar/R */ \ |
| 741 |
__ASM_EMIT ("vmulps %%ymm5, %%ymm1, %%ymm1") \ |
| 742 |
__ASM_EMIT ("vmulps %%ymm4, %%ymm2, %%ymm2") /* ymm1 = -ai/R */ \ |
| 743 |
__ASM_EMIT ("vmulps %%ymm5, %%ymm3, %%ymm3") \ |
| 744 |
__ASM_EMIT ("vmovups %%ymm0, 0x00(%[" DST "_re], %[off])") \ |
| 745 |
__ASM_EMIT ("vmovups %%ymm1, 0x20(%[" DST "_re], %[off])") \ |
| 746 |
__ASM_EMIT ("vmovups %%ymm2, 0x00(%[" DST "_im], %[off])") \ |
| 747 |
__ASM_EMIT ("vmovups %%ymm3, 0x20(%[" DST "_im], %[off])") \ |
| 748 |
__ASM_EMIT ("add $0x40, %[off]") \ |
| 749 |
__ASM_EMIT32("subl $16, %[count]") \ |
| 750 |
__ASM_EMIT64("sub $16, %[count]") \ |
| 751 |
__ASM_EMIT ("jae 1b") \ |
| 752 |
__ASM_EMIT ("4:") \ |
| 753 |
/* 8x block */ \ |
| 754 |
__ASM_EMIT32("addl $8, %[count]") \ |
| 755 |
__ASM_EMIT64("add $8, %[count]") \ |
| 756 |
__ASM_EMIT ("jl 6f") \ |
| 757 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = ar */ \ |
| 758 |
__ASM_EMIT ("vmovups 0x10(%[" SRC "_re], %[off]), %%xmm1") \ |
| 759 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = ai */ \ |
| 760 |
__ASM_EMIT ("vmovups 0x10(%[" SRC "_im], %[off]), %%xmm3") \ |
| 761 |
__ASM_EMIT ("vmulps %%xmm0, %%xmm0, %%xmm4") /* xmm4 = ar*ar */ \ |
| 762 |
__ASM_EMIT ("vmulps %%xmm1, %%xmm1, %%xmm5") \ |
| 763 |
__ASM_EMIT ("vfmadd231ps %%xmm2, %%xmm2, %%xmm4") /* xmm4 = R = ar*ar+ai*ai */ \ |
| 764 |
__ASM_EMIT ("vfmadd231ps %%xmm3, %%xmm3, %%xmm5") \ |
| 765 |
__ASM_EMIT ("vdivps %%xmm4, %%xmm6, %%xmm4") /* xmm4 = 1/R */ \ |
| 766 |
__ASM_EMIT ("vdivps %%xmm5, %%xmm6, %%xmm5") \ |
| 767 |
__ASM_EMIT ("vpxord 0x40 + %[CC], %%xmm2, %%xmm2") /* xmm2 = -ai */ \ |
| 768 |
__ASM_EMIT ("vpxord 0x40 + %[CC], %%xmm3, %%xmm3") \ |
| 769 |
__ASM_EMIT ("vmulps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = ar/R */ \ |
| 770 |
__ASM_EMIT ("vmulps %%xmm5, %%xmm1, %%xmm1") \ |
| 771 |
__ASM_EMIT ("vmulps %%xmm4, %%xmm2, %%xmm2") /* xmm1 = -ai/R */ \ |
| 772 |
__ASM_EMIT ("vmulps %%xmm5, %%xmm3, %%xmm3") \ |
| 773 |
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DST "_re], %[off])") \ |
| 774 |
__ASM_EMIT ("vmovups %%xmm1, 0x10(%[" DST "_re], %[off])") \ |
| 775 |
__ASM_EMIT ("vmovups %%xmm2, 0x00(%[" DST "_im], %[off])") \ |
| 776 |
__ASM_EMIT ("vmovups %%xmm3, 0x10(%[" DST "_im], %[off])") \ |
| 777 |
__ASM_EMIT32("subl $8, %[count]") \ |
| 778 |
__ASM_EMIT64("sub $8, %[count]") \ |
| 779 |
__ASM_EMIT ("add $0x20, %[off]") \ |
| 780 |
__ASM_EMIT ("6:") \ |
| 781 |
/* 4x block */ \ |
| 782 |
__ASM_EMIT32("addl $4, %[count]") \ |
| 783 |
__ASM_EMIT64("add $4, %[count]") \ |
| 784 |
__ASM_EMIT ("jl 8f") \ |
| 785 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = ar */ \ |
| 786 |
__ASM_EMIT ("vmovups 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = ai */ \ |
| 787 |
__ASM_EMIT ("vmulps %%xmm0, %%xmm0, %%xmm4") /* xmm4 = ar*ar */ \ |
| 788 |
__ASM_EMIT ("vfmadd231ps %%xmm2, %%xmm2, %%xmm4") /* xmm4 = R = ar*ar+ai*ai */ \ |
| 789 |
__ASM_EMIT ("vdivps %%xmm4, %%xmm6, %%xmm4") /* xmm4 = 1/R */ \ |
| 790 |
__ASM_EMIT ("vpxord 0x40 + %[CC], %%xmm2, %%xmm2") /* xmm2 = -ai */ \ |
| 791 |
__ASM_EMIT ("vmulps %%xmm4, %%xmm0, %%xmm0") /* xmm0 = ar/R */ \ |
| 792 |
__ASM_EMIT ("vmulps %%xmm4, %%xmm2, %%xmm2") /* xmm1 = -ai/R */ \ |
| 793 |
__ASM_EMIT ("vmovups %%xmm0, 0x00(%[" DST "_re], %[off])") \ |
| 794 |
__ASM_EMIT ("vmovups %%xmm2, 0x00(%[" DST "_im], %[off])") \ |
| 795 |
__ASM_EMIT32("subl $4, %[count]") \ |
| 796 |
__ASM_EMIT64("sub $4, %[count]") \ |
| 797 |
__ASM_EMIT ("add $0x10, %[off]") \ |
| 798 |
__ASM_EMIT ("8:") \ |
| 799 |
/* 1x blocks */ \ |
| 800 |
__ASM_EMIT32("addl $3, %[count]") \ |
| 801 |
__ASM_EMIT64("add $3, %[count]") \ |
| 802 |
__ASM_EMIT ("jl 10f") \ |
| 803 |
__ASM_EMIT ("9:") \ |
| 804 |
__ASM_EMIT ("vmovss 0x00(%[" SRC "_re], %[off]), %%xmm0") /* xmm0 = ar */ \ |
| 805 |
__ASM_EMIT ("vmovss 0x00(%[" SRC "_im], %[off]), %%xmm2") /* xmm2 = ai */ \ |
| 806 |
__ASM_EMIT ("vmulss %%xmm0, %%xmm0, %%xmm4") /* xmm4 = ar*ar */ \ |
| 807 |
__ASM_EMIT ("vfmadd231ss %%xmm2, %%xmm2, %%xmm4") /* xmm4 = R = ar*ar+ai*ai */ \ |
| 808 |
__ASM_EMIT ("vdivss %%xmm4, %%xmm6, %%xmm4") /* xmm4 = 1/R */ \ |
| 809 |
__ASM_EMIT ("vpxord 0x40 + %[CC], %%xmm2, %%xmm2") /* xmm2 = -ai */ \ |
| 810 |
__ASM_EMIT ("vmulss %%xmm4, %%xmm0, %%xmm0") /* xmm0 = ar/R */ \ |
| 811 |
__ASM_EMIT ("vmulss %%xmm4, %%xmm2, %%xmm2") /* xmm1 = -ai/R */ \ |
| 812 |
__ASM_EMIT ("vmovss %%xmm0, 0x00(%[" DST "_re], %[off])") \ |
| 813 |
__ASM_EMIT ("vmovss %%xmm2, 0x00(%[" DST "_im], %[off])") \ |
| 814 |
__ASM_EMIT ("add $0x04, %[off]") \ |
| 815 |
__ASM_EMIT32("decl %[count]") \ |
| 816 |
__ASM_EMIT64("dec %[count]") \ |
| 817 |
__ASM_EMIT ("jge 9b") \ |
| 818 |
__ASM_EMIT ("10:") |
| 819 |
|
| 820 |
void complex_rcp1(float *dst_re, float *dst_im, size_t count) |
| 821 |
{ |
| 822 |
IF_ARCH_X86(size_t off); |
| 823 |
ARCH_X86_ASM |
| 824 |
( |
| 825 |
COMPLEX_RCP_CORE("dst", "dst", FMA_OFF) |
| 826 |
: [count] "+r" (count), [off] "=&r" (off) |
| 827 |
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im), |
| 828 |
[CC] "o" (complex_div_const) |
| 829 |
: "cc", "memory", |
| 830 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 831 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 832 |
); |
| 833 |
} |
| 834 |
|
| 835 |
void complex_rcp2(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t count) |
| 836 |
{ |
| 837 |
IF_ARCH_X86(size_t off); |
| 838 |
ARCH_X86_ASM |
| 839 |
( |
| 840 |
COMPLEX_RCP_CORE("dst", "src", FMA_OFF) |
| 841 |
: [count] X86_PGREG (count), |
| 842 |
[off] "=&r" (off) |
| 843 |
: [dst_re] "r" (dst_re), [dst_im] "r" (dst_im), |
| 844 |
[src_re] "r" (src_re), [src_im] "r" (src_im), |
| 845 |
[CC] "o" (complex_div_const) |
| 846 |
: "cc", "memory", |
| 847 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 848 |
"%xmm4", "%xmm5", "%xmm6", "%xmm7" |
| 849 |
); |
| 850 |
} |
| 851 |
|
| 852 |
#undef COMPLEX_RCP_CORE |
| 853 |
|
| 854 |
|
| 855 |
} /* namespace avx512 */ |
| 856 |
} /* namespace lsp */ |
| 857 |
|
| 858 |
|
| 859 |
|
| 860 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX512_COMPLEX_H_ */ |