| 1 |
/* |
| 2 |
* Copyright (C) 2026 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2026 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 7 авг. 2026 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_AVX_3DMATH_COMMON_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_AVX_3DMATH_COMMON_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_AVX_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX_IMPL */ |
| 28 |
|
| 29 |
// Normalize vector/point |
| 30 |
// Input: |
| 31 |
// x0 = vector/point to normalize, w = 0 |
| 32 |
// x1 = temp |
| 33 |
// x2 = temp |
| 34 |
// Output: |
| 35 |
// x0 = normalized vector/point |
| 36 |
#define NORMALIZE(x0, x1, x2) /* x0 = vector/point to normalize */ \ |
| 37 |
__ASM_EMIT("vmulps %" x0 ", %" x0 ", %" x2) /* x2 = x*x y*y z*z 0 */ \ |
| 38 |
__ASM_EMIT("vhaddps %" x2 ", %" x2 ", %" x1) /* x1 = x*x+y*y z*z x*x+y*y z*z */ \ |
| 39 |
__ASM_EMIT("vhaddps %" x1 ", %" x1 ", %" x1) /* x1 = w2 w2 w2 w2 = x*x + y*y + z*z ... */ \ |
| 40 |
__ASM_EMIT("vxorps %" x2 ", %" x2 ", %" x2) /* x2 = 0 */ \ |
| 41 |
__ASM_EMIT("vsqrtps %" x1 ", %" x1) /* x1 = w = sqrt(w2) */ \ |
| 42 |
__ASM_EMIT("vcmpps $4, %" x2 ", %" x0 ", %" x2) /* x2 = (w != 0) */ \ |
| 43 |
__ASM_EMIT("vdivps %" x1 ", %" x0 ", %" x0) /* x0 = x/w y/w z/w w/w */ \ |
| 44 |
__ASM_EMIT("vandps %" x2 ", %" x0 ", %" x0) /* x0 = (w != 0) ? x/w y/w z/w w/w : 0 */ \ |
| 45 |
|
| 46 |
// Load matrix |
| 47 |
// Input: |
| 48 |
// ptr = address of matrix |
| 49 |
// Output: |
| 50 |
// x0 = row 0 |
| 51 |
// x1 = row 1 |
| 52 |
// x2 = row 2 |
| 53 |
// x3 = row 3 |
| 54 |
#define MATRIX_LOAD(ptr, x0, x1, x2, x3) \ |
| 55 |
__ASM_EMIT("vmovups 0x00(%[" ptr "]), %" x0 ) \ |
| 56 |
__ASM_EMIT("vmovups 0x10(%[" ptr "]), %" x1 ) \ |
| 57 |
__ASM_EMIT("vmovups 0x20(%[" ptr "]), %" x2 ) \ |
| 58 |
__ASM_EMIT("vmovups 0x30(%[" ptr "]), %" x3 ) |
| 59 |
|
| 60 |
// Store matrix |
| 61 |
// Input: |
| 62 |
// ptr = address of matrix |
| 63 |
// x0 = row 0 |
| 64 |
// x1 = row 1 |
| 65 |
// x2 = row 2 |
| 66 |
// x3 = row 3 |
| 67 |
#define MATRIX_STORE(ptr, x0, x1, x2, x3) \ |
| 68 |
__ASM_EMIT("vmovups %" x0 ", 0x00(%[" ptr "])") \ |
| 69 |
__ASM_EMIT("vmovups %" x1 ", 0x10(%[" ptr "])") \ |
| 70 |
__ASM_EMIT("vmovups %" x2 ", 0x20(%[" ptr "])") \ |
| 71 |
__ASM_EMIT("vmovups %" x3 ", 0x30(%[" ptr "])") |
| 72 |
|
| 73 |
// Transpose 4x4 matrix |
| 74 |
// Input: |
| 75 |
// x0 = row 0 |
| 76 |
// x1 = row 1 |
| 77 |
// x2 = row 2 |
| 78 |
// x3 = row 3 |
| 79 |
// x4 = temp |
| 80 |
// Output: |
| 81 |
// x0 = row 0 |
| 82 |
// x1 = row 1 |
| 83 |
// x2 = row 2 |
| 84 |
// x3 = row 3 |
| 85 |
#define MAT4_TRANSPOSE(x0, x1, x2, x3, x4) \ |
| 86 |
__ASM_EMIT("vpunpckhdq %" x3 ", %" x2 ", %" x4) /* x4 = c3 d3 c4 d4 */ \ |
| 87 |
__ASM_EMIT("vpunpckldq %" x3 ", %" x2 ", %" x2) /* x2 = c1 d1 c2 d2 */ \ |
| 88 |
__ASM_EMIT("vpunpckhdq %" x1 ", %" x0 ", %" x3) /* x3 = a3 b3 a4 b4 */ \ |
| 89 |
__ASM_EMIT("vpunpckldq %" x1 ", %" x0 ", %" x0) /* x0 = a1 b1 a2 b2 */ \ |
| 90 |
__ASM_EMIT("vpunpckhqdq %" x2 ", %" x0 ", %" x1) /* x1 = a2 b2 c2 d2 */ \ |
| 91 |
__ASM_EMIT("vpunpcklqdq %" x2 ", %" x0 ", %" x0) /* x0 = a1 b1 c1 d1 */ \ |
| 92 |
__ASM_EMIT("vpunpcklqdq %" x4 ", %" x3 ", %" x2) /* x2 = a3 b3 c3 d3 */ \ |
| 93 |
__ASM_EMIT("vpunpckhqdq %" x4 ", %" x3 ", %" x3) /* x3 = a4 b4 c4 d4 */ |
| 94 |
|
| 95 |
// Transpose 3x3 matrix |
| 96 |
// Input: |
| 97 |
// x0 = row 0 = a0 a1 a2 ? |
| 98 |
// x1 = row 1 = b0 b1 b2 ? |
| 99 |
// x2 = row 2 = c0 c1 c2 ? |
| 100 |
// x3 = temp |
| 101 |
// Output: |
| 102 |
// x0 = row 0 = a0 b0 c0 ? |
| 103 |
// x1 = row 1 = a1 b1 c1 ? |
| 104 |
// x2 = row 2 = a2 b2 c2 ? |
| 105 |
#define MAT3_TRANSPOSE(x0, x1, x2, x3) \ |
| 106 |
__ASM_EMIT("vunpckhps %" x1 ", %" x0 ", %" x3) /* x3 = a2 b2 ? ? */ \ |
| 107 |
__ASM_EMIT("vunpcklps %" x1 ", %" x0 ", %" x0) /* x0 = a0 b0 a1 b1 */ \ |
| 108 |
__ASM_EMIT("vmovhlps %" x0 ", %" x1 ", %" x1) /* x1 = a1 b1 b2 ? */ \ |
| 109 |
__ASM_EMIT("vmovlhps %" x2 ", %" x0 ", %" x0) /* x0 = a0 b0 c0 c1 */ \ |
| 110 |
__ASM_EMIT("vshufps $0x54, %" x2 ", %" x1", %" x1) /* x1 = a1 b1 c1 c1 */ \ |
| 111 |
__ASM_EMIT("vshufps $0xa4, %" x2 ", %" x3", %" x2) /* x2 = a2 b2 c2 c2 */ |
| 112 |
|
| 113 |
// Multiply martix by vector |
| 114 |
// Input: |
| 115 |
// x0 = martix row 0 (A) |
| 116 |
// x1 = matrix row 1 (B) |
| 117 |
// x2 = matrix row 2 (C) |
| 118 |
// x3 = matrix row 3 (D) |
| 119 |
// x4 = in vector, modified |
| 120 |
// x5 = tmp |
| 121 |
// x6 = tmp |
| 122 |
// x7 = tmp |
| 123 |
// Output: |
| 124 |
// x8 = out vector |
| 125 |
#define MATMUL_CORE(x0, x1, x2, x3, x4, x5, x6, x7, x8) \ |
| 126 |
__ASM_EMIT("vshufps $0xff, %" x4 ", %" x4 ", %" x7) /* x7 = v3 v3 v3 v3 */ \ |
| 127 |
__ASM_EMIT("vshufps $0xaa, %" x4 ", %" x4 ", %" x6) /* x6 = v2 v2 v2 v2 */ \ |
| 128 |
__ASM_EMIT("vshufps $0x55, %" x4 ", %" x4 ", %" x5) /* x5 = v1 v1 v1 v1 */ \ |
| 129 |
__ASM_EMIT("vshufps $0x00, %" x4 ", %" x4 ", %" x4) /* x4 = v0 v0 v0 v0 */ \ |
| 130 |
__ASM_EMIT("vmulps %" x7 ", %" x3 ", %" x7) /* x7 = d0*v3 d1*v3 d2*v3 d3*v3 */ \ |
| 131 |
__ASM_EMIT("vmulps %" x6 ", %" x2 ", %" x6) /* x6 = c0*v2 c1*v2 c2*v2 c3*v2 */ \ |
| 132 |
__ASM_EMIT("vmulps %" x5 ", %" x1 ", %" x5) /* x5 = b0*v1 d1*v1 d2*v1 d3*v1 */ \ |
| 133 |
__ASM_EMIT("vmulps %" x4 ", %" x0 ", %" x4) /* x4 = a0*v0 c1*v0 c2*v0 c3*v0 */ \ |
| 134 |
__ASM_EMIT("vaddps %" x7 ", %" x6 ", %" x6) \ |
| 135 |
__ASM_EMIT("vaddps %" x5 ", %" x4 ", %" x4) \ |
| 136 |
__ASM_EMIT("vaddps %" x6 ", %" x4 ", %" x8) |
| 137 |
|
| 138 |
// Multiply martix by vector (FMA3) |
| 139 |
// Input: |
| 140 |
// x0 = martix row 0 (A) |
| 141 |
// x1 = matrix row 1 (B) |
| 142 |
// x2 = matrix row 2 (C) |
| 143 |
// x3 = matrix row 3 (D) |
| 144 |
// x4 = in vector, modified |
| 145 |
// x5 = tmp |
| 146 |
// x6 = tmp |
| 147 |
// x7 = tmp |
| 148 |
// Output: |
| 149 |
// x8 = out vector |
| 150 |
#define MATMUL_CORE_FMA3(x0, x1, x2, x3, x4, x5, x6, x7, x8) \ |
| 151 |
__ASM_EMIT("vshufps $0xff, %" x4 ", %" x4 ", %" x7) /* x7 = v3 v3 v3 v3 */ \ |
| 152 |
__ASM_EMIT("vshufps $0xaa, %" x4 ", %" x4 ", %" x6) /* x6 = v2 v2 v2 v2 */ \ |
| 153 |
__ASM_EMIT("vshufps $0x55, %" x4 ", %" x4 ", %" x5) /* x5 = v1 v1 v1 v1 */ \ |
| 154 |
__ASM_EMIT("vshufps $0x00, %" x4 ", %" x4 ", %" x4) /* x4 = v0 v0 v0 v0 */ \ |
| 155 |
__ASM_EMIT("vmulps %" x6 ", %" x2 ", %" x6) /* x6 = c0*v2 c1*v2 c2*v2 c3*v2 */ \ |
| 156 |
__ASM_EMIT("vmulps %" x4 ", %" x0 ", %" x4) /* x4 = a0*v0 c1*v0 c2*v0 c3*v0 */ \ |
| 157 |
__ASM_EMIT("vfmadd231ps %" x7 ", %" x3 ", %" x6) \ |
| 158 |
__ASM_EMIT("vfmadd231ps %" x5 ", %" x1 ", %" x4) \ |
| 159 |
__ASM_EMIT("vaddps %" x6 ", %" x4 ", %" x8) |
| 160 |
|
| 161 |
// 1x vector cross-product (3 coordinates) |
| 162 |
// Input: |
| 163 |
// x0 = vector1 [dx1 dy1 dz1 0 ] |
| 164 |
// x1 = vector2 [dx2 dy2 dz2 0 ] |
| 165 |
// x2 = temporary |
| 166 |
// x3 = temporary |
| 167 |
// |
| 168 |
// Output: |
| 169 |
// x0 = vector1 * vector2 [ vz vx vy 0 ] |
| 170 |
#define VECTOR_CROSS3(x0, x1, x2, x3) \ |
| 171 |
__ASM_EMIT("vshufps $0xc9, %" x0 ", %" x0 ", %" x2) /* x2 = dy1 dz1 dx1 0 */ \ |
| 172 |
__ASM_EMIT("vshufps $0xc9, %" x1 ", %" x1 ", %" x3) /* x3 = dy2 dz2 dx2 0 */ \ |
| 173 |
__ASM_EMIT("vmulps %" x2 ", %" x1 ", %" x1) /* x1 = dx2*dy1 dy2*dz1 dz2*dx1 0 */ \ |
| 174 |
__ASM_EMIT("vmulps %" x3 ", %" x0 ", %" x0) /* x0 = dx1*dy2 dy1*dz2 dz1*dx2 0 */ \ |
| 175 |
__ASM_EMIT("vsubps %" x1 ", %" x0 ", %" x0) /* x0 = dx1*dy2-dx2*dy1 dy1*dz2-dy2*dz1 dz1*dx2-dz2*dx1 0 = NY NZ NX 0 */ \ |
| 176 |
|
| 177 |
// 1x vector cross-product (3 coordinates), FMA3 |
| 178 |
// Input: |
| 179 |
// x0 = vector1 [dx1 dy1 dz1 0 ] |
| 180 |
// x1 = vector2 [dx2 dy2 dz2 0 ] |
| 181 |
// x2 = temporary |
| 182 |
// x3 = temporary |
| 183 |
// |
| 184 |
// Output: |
| 185 |
// x0 = vector1 * vector2 [ vz vx vy 0 ] |
| 186 |
#define VECTOR_CROSS3_FMA3(x0, x1, x2, x3) \ |
| 187 |
__ASM_EMIT("vshufps $0xc9, %" x0 ", %" x0 ", %" x2) /* x2 = dy1 dz1 dx1 0 */ \ |
| 188 |
__ASM_EMIT("vshufps $0xc9, %" x1 ", %" x1 ", %" x3) /* x3 = dy2 dz2 dx2 0 */ \ |
| 189 |
__ASM_EMIT("vmulps %" x2 ", %" x1 ", %" x1) /* x1 = dx2*dy1 dy2*dz1 dz2*dx1 0 */ \ |
| 190 |
__ASM_EMIT("vfmsub213ps %" x1 ", %" x3 ", %" x0) /* x0 = dx1*dy2-dx2*dy1 dy1*dz2-dy2*dz1 dz1*dx2-dz2*dx1 0 = NY NZ NX 0 */ |
| 191 |
|
| 192 |
/* 1x vector dot-product (3 coordinates) |
| 193 |
* Input: |
| 194 |
* x0 = vector1 [dx1 dy1 dz1 0 ] |
| 195 |
* x1 = vector2 [dx2 dy2 dz2 0 ] |
| 196 |
* x2 = temporary |
| 197 |
* x3 = temporary |
| 198 |
* |
| 199 |
* Output: |
| 200 |
* x0 = result (scalar) |
| 201 |
*/ |
| 202 |
#define VECTOR_DOT3(x0, x1, x2, x3) \ |
| 203 |
/* do not use DPPS as it is slow */ \ |
| 204 |
/* shuffles and adds work faster than vhaddps */ \ |
| 205 |
__ASM_EMIT("vmulps %" x1 ", %" x0 ", %" x0) /* x0 = dx1*dx2 dy1*dy2 dz1*dz2 ? */ \ |
| 206 |
__ASM_EMIT("vmovshdup %" x0 ", %" x3) /* x3 = dy1*dy2 dy1*dy2 ? ? */ \ |
| 207 |
__ASM_EMIT("vmovhlps %" x0 ", %" x0 ", %" x2) /* x2 = dz1*dz2 ? ? ? */ \ |
| 208 |
__ASM_EMIT("vaddss %" x3 ", %" x0 ", %" x0) /* x0 = dx1*dx2 + dy1*dy2 */ \ |
| 209 |
__ASM_EMIT("vaddss %" x2 ", %" x0 ", %" x0) /* x0 = dx1*dx2 + dy1*dy2 + dz1*dz2 */ \ |
| 210 |
|
| 211 |
/* Compute projection of the normal to the result of cross product of two vectors (mixed product) |
| 212 |
* Input: |
| 213 |
* x0 = vector a [ x0 y0 z0 0 ] |
| 214 |
* x1 = vector b [ x1 y1 z1 0 ] |
| 215 |
* x2 = vector c [ x2 y2 z2 0 ] |
| 216 |
* x3 = temporary |
| 217 |
* x4 = temporary |
| 218 |
* |
| 219 |
* Output: |
| 220 |
* x0 = cross product: a dot (b cross c) |
| 221 |
*/ |
| 222 |
#define MIXED_PRODUCT3(x0, x1, x2, x3, x4) \ |
| 223 |
__ASM_EMIT("vshufps $0xc9, %" x1 ", %" x1 ", %" x3) /* x3 = y1 z1 x1 0 */ \ |
| 224 |
__ASM_EMIT("vshufps $0xc9, %" x2 ", %" x2 ", %" x4) /* x4 = y2 z2 x2 0 */ \ |
| 225 |
__ASM_EMIT("vshufps $0xd2, %" x0 ", %" x0 ", %" x0) /* x0 = z0 x0 y0 0 */ \ |
| 226 |
__ASM_EMIT("vmulps %" x4 ", %" x1 ", %" x1) /* x1 = x1*y2 y1*z2 z1*x2 w1*w2 */ \ |
| 227 |
__ASM_EMIT("vmulps %" x3 ", %" x2 ", %" x2) /* x2 = y1*x2 z1*y2 x1*z2 w1*w2 */ \ |
| 228 |
__ASM_EMIT("vsubps %" x2 ", %" x1 ", %" x1) /* x1 = x1*y2-y1*x2 y1*z2-z1*y2 z1*x2-x1*z2 0 = vz vx vy 0 */ \ |
| 229 |
__ASM_EMIT("vmulps %" x1 ", %" x0 ", %" x0) /* x0 = z0*vz x0*vx y0*vy 0 */ \ |
| 230 |
__ASM_EMIT("vmovhlps %" x0 ", %" x2 ", %" x2) /* x2 = y0*vy 0 ? ? */ \ |
| 231 |
__ASM_EMIT("vunpcklps %" x0 ", %" x0 ", %" x0) /* x0 = z0*vz z0*vz x0*vx x0*vx */ \ |
| 232 |
__ASM_EMIT("vmovhlps %" x0 ", %" x1 ", %" x1) /* x1 = x0*vx x0*vx ? ? */ \ |
| 233 |
__ASM_EMIT("vaddss %" x2 ", %" x0 ", %" x0) /* x0 = z0*vz + y0*vy */ \ |
| 234 |
__ASM_EMIT("vaddss %" x1 ", %" x0 ", %" x0) /* x0 = z0*vz + y0*vy + x0*vx */ |
| 235 |
|
| 236 |
/* Compute projection of the normal to the result of cross product of two vectors (mixed product), FMA3 implementation |
| 237 |
* Input: |
| 238 |
* x0 = vector a [ x0 y0 z0 0 ] |
| 239 |
* x1 = vector b [ x1 y1 z1 0 ] |
| 240 |
* x2 = vector c [ x2 y2 z2 0 ] |
| 241 |
* x3 = temporary |
| 242 |
* x4 = temporary |
| 243 |
* |
| 244 |
* Output: |
| 245 |
* x0 = cross product: a dot (b cross c) |
| 246 |
*/ |
| 247 |
#define MIXED_PRODUCT3_FMA3(x0, x1, x2, x3, x4) \ |
| 248 |
__ASM_EMIT("vshufps $0xc9, %" x1 ", %" x1 ", %" x3) /* x3 = y1 z1 x1 0 */ \ |
| 249 |
__ASM_EMIT("vshufps $0xc9, %" x2 ", %" x2 ", %" x4) /* x4 = y2 z2 x2 0 */ \ |
| 250 |
__ASM_EMIT("vshufps $0xd2, %" x0 ", %" x0 ", %" x0) /* x0 = z0 x0 y0 0 */ \ |
| 251 |
__ASM_EMIT("vmulps %" x4 ", %" x1 ", %" x1) /* x1 = x1*y2 y1*z2 z1*x2 w1*w2 */ \ |
| 252 |
__ASM_EMIT("vfnmadd231ps %" x3 ", %" x2 ", %" x1) /* x1 = x1*y2-y1*x2 y1*z2-z1*y2 z1*x2-x1*z2 0 = vz vx vy 0 */ \ |
| 253 |
__ASM_EMIT("vmulps %" x1 ", %" x0 ", %" x0) /* x0 = z0*vz x0*vx y0*vy 0 */ \ |
| 254 |
__ASM_EMIT("vmovhlps %" x0 ", %" x2 ", %" x2) /* x2 = y0*vy 0 ? ? */ \ |
| 255 |
__ASM_EMIT("vunpcklps %" x0 ", %" x0 ", %" x0) /* x0 = z0*vz z0*vz x0*vx x0*vx */ \ |
| 256 |
__ASM_EMIT("vmovhlps %" x0 ", %" x1 ", %" x1) /* x1 = x0*vx x0*vx ? ? */ \ |
| 257 |
__ASM_EMIT("vaddss %" x2 ", %" x0 ", %" x0) /* x0 = z0*vz + y0*vy */ \ |
| 258 |
__ASM_EMIT("vaddss %" x1 ", %" x0 ", %" x0) /* x0 = z0*vz + y0*vy + x0*vx */ |
| 259 |
|
| 260 |
|
| 261 |
/* 3x vector multiplication |
| 262 |
* Input: |
| 263 |
* x0 = vector1 [dx0 dy0 dz0 0 ] |
| 264 |
* x1 = vector2 [dx1 dy1 dz1 0 ] |
| 265 |
* x2 = vector3 [dx2 dy2 dz2 0 ] |
| 266 |
* |
| 267 |
* Output: |
| 268 |
* x0 = vector1 * vector2 [ vz vx vy 0 ] |
| 269 |
* x1 = vector2 * vector3 [ vz vx vy 0 ] |
| 270 |
* x2 = vector3 * vector1 [ vz vx vy 0 ] |
| 271 |
*/ |
| 272 |
#define VECTOR_CROSS3_X3(x0, x1, x2, x3, x4, x5, x6, x7) \ |
| 273 |
__ASM_EMIT("vshufps $0xc9, %" x0 ", %" x0 ", %" x3) /* x3 = dy0 dz0 dx0 0 */ \ |
| 274 |
__ASM_EMIT("vshufps $0xc9, %" x1 ", %" x1 ", %" x4) /* x4 = dy1 dz1 dx1 0 */ \ |
| 275 |
__ASM_EMIT("vshufps $0xc9, %" x2 ", %" x2 ", %" x5) /* x5 = dy2 dz2 dx2 0 */ \ |
| 276 |
__ASM_EMIT("vmulps %" x4 ", %" x0 ", %" x6) /* x6 = dx0*dy1 dy0*dz1 dz0*dx1 0 */ \ |
| 277 |
__ASM_EMIT("vmulps %" x3 ", %" x1 ", %" x7) /* x7 = dy0*dx1 dz0*dy1 dx0*dz1 0 */ \ |
| 278 |
__ASM_EMIT("vmulps %" x5 ", %" x1 ", %" x1) /* x1 = dx1*dy2 dy1*dz2 dz1*dx2 0 */ \ |
| 279 |
__ASM_EMIT("vmulps %" x2 ", %" x4 ", %" x4) /* x4 = dy1*dx2 dz1*dy2 dx1*dz2 0 */ \ |
| 280 |
__ASM_EMIT("vmulps %" x3 ", %" x2 ", %" x2) /* x2 = dx2*dy0 dy2*dz0 dz2*dx0 0 */ \ |
| 281 |
__ASM_EMIT("vmulps %" x0 ", %" x5 ", %" x5) /* x5 = dy2*dx0 dz2*dy0 dx2*dz0 0 */ \ |
| 282 |
__ASM_EMIT("vsubps %" x7 ", %" x6 ", %" x0) /* x0 = N0 = (dx0*dy1-dy0*dx1 dy0*dz1-dz0*dy1 dz0*dx1-dx0*dz1 0) */ \ |
| 283 |
__ASM_EMIT("vsubps %" x4 ", %" x1 ", %" x1) /* x1 = N1 = (dx1*dy2-dy1*dx2 dy1*dz2-dz1*dy2 dz1*dx2-dx1*dz2 0) */ \ |
| 284 |
__ASM_EMIT("vsubps %" x5 ", %" x2 ", %" x2) /* x2 = N2 = (dx2*dy0-dy2*dx0 dy2*dz0-dz2*dy0 dz2*dx0-dx2*dz0 0)*/ |
| 285 |
|
| 286 |
#define VECTOR_CROSS3_X3_FMA3(x0, x1, x2, x3, x4, x5, x6, x7) \ |
| 287 |
__ASM_EMIT("vshufps $0xc9, %" x0 ", %" x0 ", %" x3) /* x3 = dy0 dz0 dx0 0 */ \ |
| 288 |
__ASM_EMIT("vshufps $0xc9, %" x1 ", %" x1 ", %" x4) /* x4 = dy1 dz1 dx1 0 */ \ |
| 289 |
__ASM_EMIT("vshufps $0xc9, %" x2 ", %" x2 ", %" x5) /* x5 = dy2 dz2 dx2 0 */ \ |
| 290 |
__ASM_EMIT("vmulps %" x4 ", %" x0 ", %" x6) /* x6 = dx0*dy1 dy0*dz1 dz0*dx1 0 */ \ |
| 291 |
__ASM_EMIT("vmulps %" x3 ", %" x1 ", %" x7) /* x7 = dy0*dx1 dz0*dy1 dx0*dz1 0 */ \ |
| 292 |
__ASM_EMIT("vmulps %" x2 ", %" x4 ", %" x4) /* x4 = dy1*dx2 dz1*dy2 dx1*dz2 0 */ \ |
| 293 |
__ASM_EMIT("vmulps %" x3 ", %" x2 ", %" x2) /* x2 = dx2*dy0 dy2*dz0 dz2*dx0 0 */ \ |
| 294 |
__ASM_EMIT("vfmsub213ps %" x4 ", %" x5 ", %" x1) /* x1 = N1 = (dx1*dy2-dy1*dx2 dy1*dz2-dz1*dy2 dz1*dx2-dx1*dz2 0) */ \ |
| 295 |
__ASM_EMIT("vfnmadd231ps %" x0 ", %" x5 ", %" x2) /* x2 = N2 = (dx2*dy0-dy2*dx0 dy2*dz0-dz2*dy0 dz2*dx0-dx2*dz0 0)*/ \ |
| 296 |
__ASM_EMIT("vsubps %" x7 ", %" x6 ", %" x0) /* x0 = N0 = (dx0*dy1-dy0*dx1 dy0*dz1-dz0*dy1 dz0*dx1-dx0*dz1 0) */ |
| 297 |
|
| 298 |
|
| 299 |
/* 3x scalar multiplication |
| 300 |
* Input: |
| 301 |
* x0 = vector1 [dx0 dy0 dz0 0 ] |
| 302 |
* x1 = vector2 [dx1 dy1 dz1 0 ] |
| 303 |
* x2 = vector3 [dx2 dy2 dz2 0 ] |
| 304 |
* |
| 305 |
* Output: |
| 306 |
* x0 = vector1 * vector2 [ a ? ? ? ] |
| 307 |
* x1 = vector2 * vector3 [ b ? ? ? ] |
| 308 |
* x2 = vector3 * vector1 [ c ? ? ? ] |
| 309 |
*/ |
| 310 |
#define VECTOR_DOT3_X3(x0, x1, x2, x3) \ |
| 311 |
__ASM_EMIT("vmulps %" x1 ", %" x0 ", %" x3) /* x3 = A = dx0*dx1 dy0*dy1 dz0*dz1 0 */ \ |
| 312 |
__ASM_EMIT("vmulps %" x2 ", %" x1 ", %" x1) /* x1 = B = dx1*dx2 dy1*dy2 dz1*dz2 0 */ \ |
| 313 |
__ASM_EMIT("vmulps %" x0 ", %" x2 ", %" x2) /* x2 = C = dx2*dx0 dy2*dy0 dz2*dz0 0 */ \ |
| 314 |
__ASM_EMIT("vhaddps %" x3 ", %" x3 ", %" x0) /* x0 = ax+ay az ax+ay az */ \ |
| 315 |
__ASM_EMIT("vhaddps %" x1 ", %" x1 ", %" x1) /* x1 = bx+by bz bx+by bz */ \ |
| 316 |
__ASM_EMIT("vhaddps %" x2 ", %" x2 ", %" x2) /* x2 = cx+cy cz cx+cy cz */ \ |
| 317 |
__ASM_EMIT("vhaddps %" x0 ", %" x0 ", %" x0) /* x0 = A = ax+ay+az ... */ \ |
| 318 |
__ASM_EMIT("vhaddps %" x1 ", %" x1 ", %" x1) /* x1 = B = bx+by+bz ... */ \ |
| 319 |
__ASM_EMIT("vhaddps %" x2 ", %" x2 ", %" x2) /* x2 = C = cx+cy+cz ... */ |
| 320 |
|
| 321 |
namespace lsp |
| 322 |
{ |
| 323 |
namespace avx |
| 324 |
{ |
| 325 |
using namespace dsp; |
| 326 |
|
| 327 |
IF_ARCH_X86( |
| 328 |
static const float IDENTITY[16] __lsp_aligned32 = |
| 329 |
{ |
| 330 |
1.0f, 0.0f, 0.0f, 0.0f, |
| 331 |
0.0f, 1.0f, 0.0f, 0.0f, |
| 332 |
0.0f, 0.0f, 1.0f, 0.0f, |
| 333 |
0.0f, 0.0f, 0.0f, 1.0f |
| 334 |
}; |
| 335 |
static const float ONE[] __lsp_aligned32 = { LSP_DSP_VEC8(1.0f) }; |
| 336 |
static const uint32_t IONE[] __lsp_aligned16 = { LSP_DSP_VEC4(1) }; |
| 337 |
|
| 338 |
static const float X_3D_MTOLERANCE[] __lsp_aligned16 = { LSP_DSP_VEC4(-DSP_3D_TOLERANCE) }; |
| 339 |
static const float X_3D_TOLERANCE[] __lsp_aligned16 = { LSP_DSP_VEC4(DSP_3D_TOLERANCE) }; |
| 340 |
static const uint32_t X_MASK0111[] __lsp_aligned16 = { 0xffffffff, 0xffffffff, 0xffffffff, 0 }; |
| 341 |
static const uint32_t X_SMASK0001[] __lsp_aligned16 = { 0x80000000, 0, 0, 0 }; |
| 342 |
static const uint32_t X_SMASK0010[] __lsp_aligned16 = { 0, 0x80000000, 0, 0 }; |
| 343 |
static const uint32_t X_SMASK0100[] __lsp_aligned16 = { 0, 0, 0x80000000, 0 }; |
| 344 |
static const uint32_t X_SMASK0111[] __lsp_aligned16 = { 0x80000000, 0x80000000, 0x80000000, 0 }; |
| 345 |
|
| 346 |
static const uint32_t ICULL3[] __lsp_aligned16 = { 0x01, 0x04, 0x10, 0 }; |
| 347 |
) |
| 348 |
} /* namespace avx */ |
| 349 |
} /* naemspace lsp */ |
| 350 |
|
| 351 |
#endif /* PRIVATE_DSP_ARCH_X86_AVX_3DMATH_COMMON_H_ */ |