| 1 |
/* |
| 2 |
* Copyright (C) 2025 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2025 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 17 авг. 2025 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_SSE2_3DMATH_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_X86_SSE2_3DMATH_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_X86_SSE2_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE2_IMPL */ |
| 28 |
|
| 29 |
/* 1x matrix transpose |
| 30 |
* Input: |
| 31 |
* x0 = row 0 (a1 a2 a3 a4) |
| 32 |
* x1 = row 1 (b1 b2 b3 b4) |
| 33 |
* x2 = row 2 (c1 c2 c3 c4) |
| 34 |
* x3 = row 3 (d1 d2 d3 d4) |
| 35 |
* x4 = temporary |
| 36 |
* |
| 37 |
* Output: |
| 38 |
* x0 = row 0 (a1 b1 c1 d1) |
| 39 |
* x1 = row 1 (a2 b2 c2 d2) |
| 40 |
* x2 = row 2 (a3 b3 c3 d3) |
| 41 |
* x3 = row 3 (a4 b4 c4 d4) |
| 42 |
*/ |
| 43 |
#define MAT4_TRANSPOSE(x0, x1, x2, x3, x4) \ |
| 44 |
__ASM_EMIT("movaps %" x2 ", %" x4) /* xmm4 = c1 c2 c3 c4 */ \ |
| 45 |
__ASM_EMIT("punpckldq %" x3 ", %" x2) /* xmm2 = c1 d1 c2 d2 */ \ |
| 46 |
__ASM_EMIT("punpckhdq %" x3 ", %" x4) /* xmm4 = c3 d3 c4 d4 */ \ |
| 47 |
__ASM_EMIT("movaps %" x0 ", %" x3) /* xmm3 = a1 a2 a3 a4 */ \ |
| 48 |
__ASM_EMIT("punpckldq %" x1 ", %" x0) /* xmm0 = a1 b1 a2 b2 */ \ |
| 49 |
__ASM_EMIT("punpckhdq %" x1 ", %" x3) /* xmm3 = a3 b3 a4 b4 */ \ |
| 50 |
__ASM_EMIT("movaps %" x0 ", %" x1) /* xmm1 = a1 b2 a2 b2 */ \ |
| 51 |
__ASM_EMIT("punpcklqdq %" x2 ", %" x0) /* xmm0 = a1 b1 c1 d1 */ \ |
| 52 |
__ASM_EMIT("punpckhqdq %" x2 ", %" x1) /* xmm1 = a2 b2 c2 d2 */ \ |
| 53 |
__ASM_EMIT("movaps %" x3 ", %" x2) /* xmm2 = a3 b3 a4 b4 */ \ |
| 54 |
__ASM_EMIT("punpcklqdq %" x4 ", %" x2) /* xmm2 = a3 b3 c3 d3 */ \ |
| 55 |
__ASM_EMIT("punpckhqdq %" x4 ", %" x3) /* xmm3 = a4 b4 c4 d4 */ |
| 56 |
|
| 57 |
#define MATRIX_LOAD(ptr, x0, x1, x2, x3) \ |
| 58 |
__ASM_EMIT("movups 0x00(%[" ptr "]), %" x0 ) \ |
| 59 |
__ASM_EMIT("movups 0x10(%[" ptr "]), %" x1 ) \ |
| 60 |
__ASM_EMIT("movups 0x20(%[" ptr "]), %" x2 ) \ |
| 61 |
__ASM_EMIT("movups 0x30(%[" ptr "]), %" x3 ) |
| 62 |
|
| 63 |
#define MATRIX_STORE(ptr, x0, x1, x2, x3) \ |
| 64 |
__ASM_EMIT("movups %" x0 ", 0x00(%[" ptr "])") \ |
| 65 |
__ASM_EMIT("movups %" x1 ", 0x10(%[" ptr "])") \ |
| 66 |
__ASM_EMIT("movups %" x2 ", 0x20(%[" ptr "])") \ |
| 67 |
__ASM_EMIT("movups %" x3 ", 0x30(%[" ptr "])") |
| 68 |
|
| 69 |
namespace lsp |
| 70 |
{ |
| 71 |
namespace sse2 |
| 72 |
{ |
| 73 |
using namespace dsp; |
| 74 |
|
| 75 |
IF_ARCH_X86( |
| 76 |
static const float X_3D_TOLERANCE[] __lsp_aligned16 = { LSP_DSP_VEC4(DSP_3D_TOLERANCE) }; |
| 77 |
static const float X_3D_MTOLERANCE[] __lsp_aligned16 = { LSP_DSP_VEC4(-DSP_3D_TOLERANCE) }; |
| 78 |
|
| 79 |
static const uint32_t IONE[] __lsp_aligned16 = { LSP_DSP_VEC4(1) }; |
| 80 |
|
| 81 |
static const uint32_t ICULL3[] __lsp_aligned16 = { 0x01, 0x04, 0x10, 0 }; |
| 82 |
) |
| 83 |
|
| 84 |
size_t colocation_x2_v1p2(const vector3d_t *pl, const point3d_t *p0, const point3d_t *p1) |
| 85 |
{ |
| 86 |
float x0, x1, x2, x3; |
| 87 |
float res[4] __lsp_aligned16; |
| 88 |
|
| 89 |
ARCH_X86_ASM |
| 90 |
( |
| 91 |
__ASM_EMIT("movups (%[pl]), %[x2]") /* xmm3 = pl */ |
| 92 |
__ASM_EMIT("movups (%[p0]), %[x0]") /* xmm0 = p0 */ |
| 93 |
__ASM_EMIT("movups (%[p1]), %[x1]") /* xmm1 = p1 */ |
| 94 |
|
| 95 |
__ASM_EMIT("mulps %[x2], %[x0]") /* xmm0 = x0*px y0*py z0*pz w0*pw */ |
| 96 |
__ASM_EMIT("mulps %[x2], %[x1]") /* xmm1 = x1*px y1*py z1*pz w1*pw */ |
| 97 |
__ASM_EMIT("movaps %[x0], %[x3]") /* xmm3 = x0*px y0*py z0*pz w0*pw */ |
| 98 |
__ASM_EMIT("movlhps %[x1], %[x0]") /* xmm0 = x0*px y0*py x1*px y1*py */ |
| 99 |
__ASM_EMIT("movhlps %[x3], %[x1]") /* xmm1 = z0*pz w0*pw z1*pz w1*pw */ |
| 100 |
__ASM_EMIT("addps %[x1], %[x0]") /* xmm0 = x0*px+z0*pz y0*py+w0*pw x1*px+z1*pz y1*py+w1*pw */ |
| 101 |
__ASM_EMIT("shufps $0xd8, %[x0], %[x0]") /* xmm0 = x0*px+z0*pz x1*px+z1*pz y0*py+w0*pw y1*py+w1*pw */ |
| 102 |
__ASM_EMIT("movhlps %[x0], %[x1]") /* xmm1 = y0*py+w0*pw y1*py+w1*pw ? ? */ |
| 103 |
__ASM_EMIT("addps %[x1], %[x0]") /* xmm0 = x0*px+y0*py+z0*pz+w0*pw x1*px+y1*py+z1*pz+w1*pw ? ? = k0 k1 ? ? */ |
| 104 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = k0 k1 ? ? */ |
| 105 |
|
| 106 |
__ASM_EMIT("cmpps $2, %[PTOL], %[x0]") /* xmm0 = k0 <= +TOL k1 <= +TOL ? ? */ |
| 107 |
__ASM_EMIT("cmpps $1, %[MTOL], %[x1]") /* xmm1 = k0 < -TOL k1 < -TOL ? ? */ |
| 108 |
__ASM_EMIT("andps %[IONE], %[x0]") /* xmm0 = 1*[k0 <= +TOL] 1*[k1 <= +TOL] ? ? */ |
| 109 |
__ASM_EMIT("andps %[IONE], %[x1]") /* xmm1 = 1*[k0 < -TOL] 1*[k1 < -TOL] ? ? */ |
| 110 |
__ASM_EMIT("paddd %[x1], %[x0]") |
| 111 |
__ASM_EMIT("movdqa %[x0], (%[res])") |
| 112 |
__ASM_EMIT32("movl 0x00(%[res]), %[p0]") |
| 113 |
__ASM_EMIT32("movl 0x04(%[res]), %[p1]") |
| 114 |
__ASM_EMIT64("movl 0x00(%[res]), %k[p0]") |
| 115 |
__ASM_EMIT64("movl 0x04(%[res]), %k[p1]") |
| 116 |
__ASM_EMIT("lea (%[p0], %[p1], 4), %[p0]") |
| 117 |
: [p0] "+r" (p0), [p1] "+r" (p1), |
| 118 |
[x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3) |
| 119 |
: [pl] "r" (pl), |
| 120 |
[res] "r" (res), |
| 121 |
[PTOL] "m" (X_3D_TOLERANCE), |
| 122 |
[MTOL] "m" (X_3D_MTOLERANCE), |
| 123 |
[IONE] "m" (IONE) |
| 124 |
: "cc" |
| 125 |
); |
| 126 |
|
| 127 |
return size_t(p0); |
| 128 |
} |
| 129 |
|
| 130 |
size_t colocation_x2_v1pv(const vector3d_t *pl, const point3d_t *pv) |
| 131 |
{ |
| 132 |
float x0, x1, x2, x3; |
| 133 |
float res[4] __lsp_aligned16; |
| 134 |
|
| 135 |
ARCH_X86_ASM |
| 136 |
( |
| 137 |
__ASM_EMIT("movups (%[pl]), %[x2]") /* xmm3 = pl */ |
| 138 |
__ASM_EMIT("movups 0x00(%[pv]), %[x0]") /* xmm0 = p0 */ |
| 139 |
__ASM_EMIT("movups 0x10(%[pv]), %[x1]") /* xmm1 = p1 */ |
| 140 |
|
| 141 |
__ASM_EMIT("mulps %[x2], %[x0]") /* xmm0 = x0*px y0*py z0*pz w0*pw */ |
| 142 |
__ASM_EMIT("mulps %[x2], %[x1]") /* xmm1 = x1*px y1*py z1*pz w1*pw */ |
| 143 |
__ASM_EMIT("movaps %[x0], %[x3]") /* xmm3 = x0*px y0*py z0*pz w0*pw */ |
| 144 |
__ASM_EMIT("movlhps %[x1], %[x0]") /* xmm0 = x0*px y0*py x1*px y1*py */ |
| 145 |
__ASM_EMIT("movhlps %[x3], %[x1]") /* xmm1 = z0*pz w0*pw z1*pz w1*pw */ |
| 146 |
__ASM_EMIT("addps %[x1], %[x0]") /* xmm0 = x0*px+z0*pz y0*py+w0*pw x1*px+z1*pz y1*py+w1*pw */ |
| 147 |
__ASM_EMIT("shufps $0xd8, %[x0], %[x0]") /* xmm0 = x0*px+z0*pz x1*px+z1*pz y0*py+w0*pw y1*py+w1*pw */ |
| 148 |
__ASM_EMIT("movhlps %[x0], %[x1]") /* xmm1 = y0*py+w0*pw y1*py+w1*pw ? ? */ |
| 149 |
__ASM_EMIT("addps %[x1], %[x0]") /* xmm0 = x0*px+y0*py+z0*pz+w0*pw x1*px+y1*py+z1*pz+w1*pw ? ? = k0 k1 ? ? */ |
| 150 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = k0 k1 ? ? */ |
| 151 |
|
| 152 |
__ASM_EMIT("cmpps $2, %[PTOL], %[x0]") /* xmm0 = k0 <= +TOL k1 <= +TOL ? ? */ |
| 153 |
__ASM_EMIT("cmpps $1, %[MTOL], %[x1]") /* xmm1 = k0 < -TOL k1 < -TOL ? ? */ |
| 154 |
__ASM_EMIT("andps %[IONE], %[x0]") /* xmm0 = 1*[k0 <= +TOL] 1*[k1 <= +TOL] ? ? */ |
| 155 |
__ASM_EMIT("andps %[IONE], %[x1]") /* xmm1 = 1*[k0 < -TOL] 1*[k1 < -TOL] ? ? */ |
| 156 |
__ASM_EMIT("paddd %[x1], %[x0]") |
| 157 |
__ASM_EMIT("movdqa %[x0], (%[res])") |
| 158 |
__ASM_EMIT32("movl 0x00(%[res]), %[pl]") |
| 159 |
__ASM_EMIT32("movl 0x04(%[res]), %[pv]") |
| 160 |
__ASM_EMIT64("movl 0x00(%[res]), %k[pl]") |
| 161 |
__ASM_EMIT64("movl 0x04(%[res]), %k[pv]") |
| 162 |
__ASM_EMIT("lea (%[pl], %[pv], 4), %[pl]") |
| 163 |
: [pl] "+r" (pl), [pv] "+r" (pv), |
| 164 |
[x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3) |
| 165 |
: [res] "r" (res), |
| 166 |
[PTOL] "m" (X_3D_TOLERANCE), |
| 167 |
[MTOL] "m" (X_3D_MTOLERANCE), |
| 168 |
[IONE] "m" (IONE) |
| 169 |
: "cc" |
| 170 |
); |
| 171 |
|
| 172 |
return size_t(pl); |
| 173 |
} |
| 174 |
|
| 175 |
size_t colocation_x3_v1p3(const vector3d_t *pl, const point3d_t *p0, const point3d_t *p1, const point3d_t *p2) |
| 176 |
{ |
| 177 |
float x0, x1, x2, x3, x4; |
| 178 |
float res[4] __lsp_aligned16; |
| 179 |
|
| 180 |
ARCH_X86_ASM |
| 181 |
( |
| 182 |
__ASM_EMIT("movups (%[pl]), %[x3]") /* xmm3 = pl */ |
| 183 |
__ASM_EMIT("movups (%[p0]), %[x0]") /* xmm0 = p0 */ |
| 184 |
__ASM_EMIT("movups (%[p1]), %[x1]") /* xmm1 = p1 */ |
| 185 |
__ASM_EMIT("movups (%[p2]), %[x2]") /* xmm2 = p2 */ |
| 186 |
|
| 187 |
__ASM_EMIT("mulps %[x3], %[x0]") /* xmm0 = p0 * pl */ |
| 188 |
__ASM_EMIT("mulps %[x3], %[x1]") /* xmm1 = p1 * pl */ |
| 189 |
__ASM_EMIT("mulps %[x3], %[x2]") /* xmm2 = p2 * pl */ |
| 190 |
MAT4_TRANSPOSE("[x0]", "[x1]", "[x2]", "[x3]", "[x4]") |
| 191 |
/* xmm0 = x0 x1 x2 ? */ |
| 192 |
/* xmm1 = y0 y1 y2 ? */ |
| 193 |
/* xmm2 = z0 z1 z2 ? */ |
| 194 |
/* xmm3 = w0 w1 w2 ? */ |
| 195 |
__ASM_EMIT("addps %[x1], %[x0]") |
| 196 |
__ASM_EMIT("addps %[x3], %[x2]") |
| 197 |
__ASM_EMIT("addps %[x2], %[x0]") /* xmm0 = k0 k1 k2 ? */ |
| 198 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = k0 k1 k2 ? */ |
| 199 |
__ASM_EMIT("cmpps $2, %[PTOL], %[x0]") /* xmm0 = k0 <= +TOL k1 <= +TOL k2 <= +TOL ? */ |
| 200 |
__ASM_EMIT("cmpps $1, %[MTOL], %[x1]") /* xmm1 = k0 < -TOL k1 < -TOL k2 < -TOL ? */ |
| 201 |
__ASM_EMIT("andps %[IONE], %[x0]") /* xmm0 = 1*[k0 <= +TOL] 1*[k1 <= +TOL] 1*[k2 <= +TOL] ? */ |
| 202 |
__ASM_EMIT("andps %[IONE], %[x1]") /* xmm1 = 1*[k0 < -TOL] 1*[k1 < -TOL] 1*[k2 < -TOL] ? */ |
| 203 |
__ASM_EMIT("paddd %[x1], %[x0]") |
| 204 |
__ASM_EMIT("movdqa %[x0], (%[res])") |
| 205 |
__ASM_EMIT32("movl 0x00(%[res]), %[p0]") |
| 206 |
__ASM_EMIT32("movl 0x04(%[res]), %[p1]") |
| 207 |
__ASM_EMIT32("movl 0x08(%[res]), %[p2]") |
| 208 |
__ASM_EMIT64("movl 0x00(%[res]), %k[p0]") |
| 209 |
__ASM_EMIT64("movl 0x04(%[res]), %k[p1]") |
| 210 |
__ASM_EMIT64("movl 0x08(%[res]), %k[p2]") |
| 211 |
__ASM_EMIT("lea (%[p1], %[p2], 4), %[p1]") |
| 212 |
__ASM_EMIT("lea (%[p0], %[p1], 4), %[p0]") |
| 213 |
: [p0] "+r" (p0), [p1] "+r" (p1), [p2] "+r" (p2), |
| 214 |
[x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), |
| 215 |
[x4] "=&x" (x4) |
| 216 |
: [pl] "r" (pl), |
| 217 |
[res] "r" (res), |
| 218 |
[PTOL] "m" (X_3D_TOLERANCE), |
| 219 |
[MTOL] "m" (X_3D_MTOLERANCE), |
| 220 |
[IONE] "m" (IONE) |
| 221 |
: "cc" |
| 222 |
); |
| 223 |
|
| 224 |
return size_t(p0); |
| 225 |
} |
| 226 |
|
| 227 |
size_t colocation_x3_v1pv(const vector3d_t *pl, const point3d_t *pv) |
| 228 |
{ |
| 229 |
float x0, x1, x2, x3, x4; |
| 230 |
float res[4] __lsp_aligned16; |
| 231 |
size_t pt; |
| 232 |
|
| 233 |
ARCH_X86_ASM |
| 234 |
( |
| 235 |
__ASM_EMIT("movups (%[pl]), %[x3]") /* xmm3 = pl */ |
| 236 |
__ASM_EMIT("movups 0x00(%[pv]), %[x0]") /* xmm0 = p0 */ |
| 237 |
__ASM_EMIT("movups 0x10(%[pv]), %[x1]") /* xmm1 = p1 */ |
| 238 |
__ASM_EMIT("movups 0x20(%[pv]), %[x2]") /* xmm2 = p2 */ |
| 239 |
|
| 240 |
__ASM_EMIT("mulps %[x3], %[x0]") /* xmm0 = p0 * pl */ |
| 241 |
__ASM_EMIT("mulps %[x3], %[x1]") /* xmm1 = p1 * pl */ |
| 242 |
__ASM_EMIT("mulps %[x3], %[x2]") /* xmm2 = p2 * pl */ |
| 243 |
MAT4_TRANSPOSE("[x0]", "[x1]", "[x2]", "[x3]", "[x4]") |
| 244 |
/* xmm0 = x0 x1 x2 ? */ |
| 245 |
/* xmm1 = y0 y1 y2 ? */ |
| 246 |
/* xmm2 = z0 z1 z2 ? */ |
| 247 |
/* xmm3 = w0 w1 w2 ? */ |
| 248 |
__ASM_EMIT("addps %[x1], %[x0]") |
| 249 |
__ASM_EMIT("addps %[x3], %[x2]") |
| 250 |
__ASM_EMIT("addps %[x2], %[x0]") /* xmm0 = k0 k1 k2 ? */ |
| 251 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = k0 k1 k2 ? */ |
| 252 |
__ASM_EMIT("cmpps $2, %[PTOL], %[x0]") /* xmm0 = k0 <= +TOL k1 <= +TOL k2 <= +TOL ? */ |
| 253 |
__ASM_EMIT("cmpps $1, %[MTOL], %[x1]") /* xmm1 = k0 < -TOL k1 < -TOL k2 < -TOL ? */ |
| 254 |
__ASM_EMIT("andps %[IONE], %[x0]") /* xmm0 = 1*[k0 <= +TOL] 1*[k1 <= +TOL] 1*[k2 <= +TOL] ? */ |
| 255 |
__ASM_EMIT("andps %[IONE], %[x1]") /* xmm1 = 1*[k0 < -TOL] 1*[k1 < -TOL] 1*[k2 < -TOL] ? */ |
| 256 |
__ASM_EMIT("paddd %[x1], %[x0]") |
| 257 |
__ASM_EMIT("movdqa %[x0], (%[res])") |
| 258 |
__ASM_EMIT32("movl 0x00(%[res]), %[pl]") |
| 259 |
__ASM_EMIT32("movl 0x04(%[res]), %[pv]") |
| 260 |
__ASM_EMIT32("movl 0x08(%[res]), %[pt]") |
| 261 |
__ASM_EMIT64("movl 0x00(%[res]), %k[pl]") |
| 262 |
__ASM_EMIT64("movl 0x04(%[res]), %k[pv]") |
| 263 |
__ASM_EMIT64("movl 0x08(%[res]), %k[pt]") |
| 264 |
__ASM_EMIT("lea (%[pv], %[pt], 4), %[pt]") |
| 265 |
__ASM_EMIT("lea (%[pl], %[pt], 4), %[pt]") |
| 266 |
: [pl] "+r" (pl), [pv] "+r" (pv), [pt] "=&r" (pt), |
| 267 |
[x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), |
| 268 |
[x4] "=&x" (x4) |
| 269 |
: [res] "r" (res), |
| 270 |
[PTOL] "m" (X_3D_TOLERANCE), |
| 271 |
[MTOL] "m" (X_3D_MTOLERANCE), |
| 272 |
[IONE] "m" (IONE) |
| 273 |
: "cc" |
| 274 |
); |
| 275 |
|
| 276 |
return pt; |
| 277 |
} |
| 278 |
|
| 279 |
size_t colocation_x3_v3p1(const vector3d_t *v0, const vector3d_t *v1, const vector3d_t *v2, const point3d_t *p) |
| 280 |
{ |
| 281 |
float x0, x1, x2, x3, x4; |
| 282 |
float res[4] __lsp_aligned16; |
| 283 |
|
| 284 |
ARCH_X86_ASM |
| 285 |
( |
| 286 |
__ASM_EMIT("movups (%[p]), %[x3]") /* xmm3 = p */ |
| 287 |
__ASM_EMIT("movups (%[v0]), %[x0]") /* xmm0 = v0 */ |
| 288 |
__ASM_EMIT("movups (%[v1]), %[x1]") /* xmm1 = v1 */ |
| 289 |
__ASM_EMIT("movups (%[v2]), %[x2]") /* xmm2 = v2 */ |
| 290 |
|
| 291 |
__ASM_EMIT("mulps %[x3], %[x0]") /* xmm0 = v0 * p */ |
| 292 |
__ASM_EMIT("mulps %[x3], %[x1]") /* xmm1 = v1 * p */ |
| 293 |
__ASM_EMIT("mulps %[x3], %[x2]") /* xmm2 = v2 * p */ |
| 294 |
MAT4_TRANSPOSE("[x0]", "[x1]", "[x2]", "[x3]", "[x4]") |
| 295 |
/* xmm0 = x0 x1 x2 ? */ |
| 296 |
/* xmm1 = y0 y1 y2 ? */ |
| 297 |
/* xmm2 = z0 z1 z2 ? */ |
| 298 |
/* xmm3 = w0 w1 w2 ? */ |
| 299 |
__ASM_EMIT("addps %[x1], %[x0]") |
| 300 |
__ASM_EMIT("addps %[x3], %[x2]") |
| 301 |
__ASM_EMIT("addps %[x2], %[x0]") /* xmm0 = k0 k1 k2 ? */ |
| 302 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = k0 k1 k2 ? */ |
| 303 |
__ASM_EMIT("cmpps $2, %[PTOL], %[x0]") /* xmm0 = k0 <= +TOL k1 <= +TOL k2 <= +TOL ? */ |
| 304 |
__ASM_EMIT("cmpps $1, %[MTOL], %[x1]") /* xmm1 = k0 < -TOL k1 < -TOL k2 < -TOL ? */ |
| 305 |
__ASM_EMIT("andps %[IONE], %[x0]") /* xmm0 = 1*[k0 <= +TOL] 1*[k1 <= +TOL] 1*[k2 <= +TOL] ? */ |
| 306 |
__ASM_EMIT("andps %[IONE], %[x1]") /* xmm1 = 1*[k0 < -TOL] 1*[k1 < -TOL] 1*[k2 < -TOL] ? */ |
| 307 |
__ASM_EMIT("paddd %[x1], %[x0]") |
| 308 |
__ASM_EMIT("movdqa %[x0], (%[res])") |
| 309 |
__ASM_EMIT32("movl 0x00(%[res]), %[v0]") |
| 310 |
__ASM_EMIT32("movl 0x04(%[res]), %[v1]") |
| 311 |
__ASM_EMIT32("movl 0x08(%[res]), %[v2]") |
| 312 |
__ASM_EMIT64("movl 0x00(%[res]), %k[v0]") |
| 313 |
__ASM_EMIT64("movl 0x04(%[res]), %k[v1]") |
| 314 |
__ASM_EMIT64("movl 0x08(%[res]), %k[v2]") |
| 315 |
__ASM_EMIT("lea (%[v1], %[v2], 4), %[v1]") |
| 316 |
__ASM_EMIT("lea (%[v0], %[v1], 4), %[v0]") |
| 317 |
|
| 318 |
: [v0] "+r" (v0), [v1] "+r" (v1), [v2] "+r" (v2), |
| 319 |
[x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), |
| 320 |
[x4] "=&x" (x4) |
| 321 |
: [p] "r" (p), |
| 322 |
[res] "r" (res), |
| 323 |
[PTOL] "m" (X_3D_TOLERANCE), |
| 324 |
[MTOL] "m" (X_3D_MTOLERANCE), |
| 325 |
[IONE] "m" (IONE) |
| 326 |
: "cc" |
| 327 |
); |
| 328 |
|
| 329 |
return size_t(v0); |
| 330 |
} |
| 331 |
|
| 332 |
size_t colocation_x3_vvp1(const vector3d_t *vv, const point3d_t *p) |
| 333 |
{ |
| 334 |
float x0, x1, x2, x3, x4; |
| 335 |
float res[4] __lsp_aligned16; |
| 336 |
size_t pt; |
| 337 |
|
| 338 |
ARCH_X86_ASM |
| 339 |
( |
| 340 |
__ASM_EMIT("movups (%[p]), %[x3]") /* xmm3 = p */ |
| 341 |
__ASM_EMIT("movups 0x00(%[vv]), %[x0]") /* xmm0 = v0 */ |
| 342 |
__ASM_EMIT("movups 0x10(%[vv]), %[x1]") /* xmm1 = v1 */ |
| 343 |
__ASM_EMIT("movups 0x20(%[vv]), %[x2]") /* xmm2 = v2 */ |
| 344 |
|
| 345 |
__ASM_EMIT("mulps %[x3], %[x0]") /* xmm0 = v0 * p */ |
| 346 |
__ASM_EMIT("mulps %[x3], %[x1]") /* xmm1 = v1 * p */ |
| 347 |
__ASM_EMIT("mulps %[x3], %[x2]") /* xmm2 = v2 * p */ |
| 348 |
MAT4_TRANSPOSE("[x0]", "[x1]", "[x2]", "[x3]", "[x4]") |
| 349 |
/* xmm0 = x0 x1 x2 ? */ |
| 350 |
/* xmm1 = y0 y1 y2 ? */ |
| 351 |
/* xmm2 = z0 z1 z2 ? */ |
| 352 |
/* xmm3 = w0 w1 w2 ? */ |
| 353 |
__ASM_EMIT("addps %[x1], %[x0]") |
| 354 |
__ASM_EMIT("addps %[x3], %[x2]") |
| 355 |
__ASM_EMIT("addps %[x2], %[x0]") /* xmm0 = k0 k1 k2 ? */ |
| 356 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = k0 k1 k2 ? */ |
| 357 |
__ASM_EMIT("cmpps $2, %[PTOL], %[x0]") /* xmm0 = k0 <= +TOL k1 <= +TOL k2 <= +TOL ? */ |
| 358 |
__ASM_EMIT("cmpps $1, %[MTOL], %[x1]") /* xmm1 = k0 < -TOL k1 < -TOL k2 < -TOL ? */ |
| 359 |
__ASM_EMIT("andps %[IONE], %[x0]") /* xmm0 = 1*[k0 <= +TOL] 1*[k1 <= +TOL] 1*[k2 <= +TOL] ? */ |
| 360 |
__ASM_EMIT("andps %[IONE], %[x1]") /* xmm1 = 1*[k0 < -TOL] 1*[k1 < -TOL] 1*[k2 < -TOL] ? */ |
| 361 |
__ASM_EMIT("paddd %[x1], %[x0]") |
| 362 |
__ASM_EMIT("movdqa %[x0], (%[res])") |
| 363 |
__ASM_EMIT32("movl 0x00(%[res]), %[vv]") |
| 364 |
__ASM_EMIT32("movl 0x04(%[res]), %[p]") |
| 365 |
__ASM_EMIT32("movl 0x08(%[res]), %[pt]") |
| 366 |
__ASM_EMIT64("movl 0x00(%[res]), %k[vv]") |
| 367 |
__ASM_EMIT64("movl 0x04(%[res]), %k[p]") |
| 368 |
__ASM_EMIT64("movl 0x08(%[res]), %k[pt]") |
| 369 |
__ASM_EMIT("lea (%[p], %[pt], 4), %[pt]") |
| 370 |
__ASM_EMIT("lea (%[vv], %[pt], 4), %[pt]") |
| 371 |
|
| 372 |
: [vv] "+r" (vv), [p] "+r" (p), [pt] "=&r" (pt), |
| 373 |
[x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), |
| 374 |
[x4] "=&x" (x4) |
| 375 |
: [res] "r" (res), |
| 376 |
[PTOL] "m" (X_3D_TOLERANCE), |
| 377 |
[MTOL] "m" (X_3D_MTOLERANCE), |
| 378 |
[IONE] "m" (IONE) |
| 379 |
: "cc" |
| 380 |
); |
| 381 |
|
| 382 |
return pt; |
| 383 |
} |
| 384 |
|
| 385 |
void transpose_matrix3d1(matrix3d_t *r) |
| 386 |
{ |
| 387 |
ARCH_X86_ASM |
| 388 |
( |
| 389 |
MATRIX_LOAD("m", "%xmm0", "%xmm1", "%xmm2", "%xmm3") |
| 390 |
MAT4_TRANSPOSE("%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4") |
| 391 |
MATRIX_STORE("m", "%xmm0", "%xmm1", "%xmm2", "%xmm3") |
| 392 |
: |
| 393 |
: [m] "r" (r) |
| 394 |
: "memory", |
| 395 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4" |
| 396 |
); |
| 397 |
} |
| 398 |
|
| 399 |
void transpose_matrix3d2(matrix3d_t *r, const matrix3d_t *m) |
| 400 |
{ |
| 401 |
ARCH_X86_ASM |
| 402 |
( |
| 403 |
MATRIX_LOAD("m", "%xmm0", "%xmm1", "%xmm2", "%xmm3") |
| 404 |
MAT4_TRANSPOSE("%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4") |
| 405 |
MATRIX_STORE("r", "%xmm0", "%xmm1", "%xmm2", "%xmm3") |
| 406 |
: |
| 407 |
: [r] "r" (r), [m] "r" (m) |
| 408 |
: "memory", |
| 409 |
"%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4" |
| 410 |
); |
| 411 |
} |
| 412 |
|
| 413 |
void split_triangle_raw( |
| 414 |
raw_triangle_t *out, |
| 415 |
size_t *n_out, |
| 416 |
raw_triangle_t *in, |
| 417 |
size_t *n_in, |
| 418 |
const vector3d_t *pl, |
| 419 |
const raw_triangle_t *pv |
| 420 |
) |
| 421 |
{ |
| 422 |
float x0, x1, x2, x3, x4, x5, x6, x7; |
| 423 |
float k[4] __lsp_aligned16; |
| 424 |
size_t t0; |
| 425 |
|
| 426 |
in += *n_in; |
| 427 |
out += *n_out; |
| 428 |
|
| 429 |
// Estimate co-location of triangle and point |
| 430 |
ARCH_X86_ASM |
| 431 |
( |
| 432 |
__ASM_EMIT("movups (%[pl]), %[x3]") /* xmm3 = pl */ |
| 433 |
__ASM_EMIT("movups 0x00(%[pv]), %[x0]") /* xmm0 = p0 */ |
| 434 |
__ASM_EMIT("movups 0x10(%[pv]), %[x1]") /* xmm1 = p1 */ |
| 435 |
__ASM_EMIT("movups 0x20(%[pv]), %[x2]") /* xmm2 = p2 */ |
| 436 |
|
| 437 |
__ASM_EMIT("mulps %[x3], %[x0]") /* xmm0 = p0 * pl */ |
| 438 |
__ASM_EMIT("mulps %[x3], %[x1]") /* xmm1 = p1 * pl */ |
| 439 |
__ASM_EMIT("mulps %[x3], %[x2]") /* xmm2 = p2 * pl */ |
| 440 |
MAT4_TRANSPOSE("[x0]", "[x1]", "[x2]", "[x3]", "[x4]") |
| 441 |
|
| 442 |
__ASM_EMIT("addps %[x1], %[x0]") |
| 443 |
__ASM_EMIT("addps %[x3], %[x2]") |
| 444 |
__ASM_EMIT("addps %[x2], %[x0]") /* xmm0 = k0 k1 k2 ? */ |
| 445 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = k0 k1 k2 ? */ |
| 446 |
__ASM_EMIT("movaps %[x0], (%[k])") /* *k = xmm0 */ |
| 447 |
__ASM_EMIT("cmpps $2, %[PTOL], %[x0]") /* xmm0 = k0 <= +TOL k1 <= +TOL k2 <= +TOL ? */ |
| 448 |
__ASM_EMIT("cmpps $1, %[MTOL], %[x1]") /* xmm1 = k0 < -TOL k1 < -TOL k2 < -TOL ? */ |
| 449 |
__ASM_EMIT("pand %[ICULL3], %[x0]") /* xmm0 = 1*[k0 <= +TOL] 1*[k1 <= +TOL] 1*[k2 <= +TOL] ? */ |
| 450 |
__ASM_EMIT("pand %[ICULL3], %[x1]") /* xmm1 = 1*[k0 < -TOL] 1*[k1 < -TOL] 1*[k2 < -TOL] ? */ |
| 451 |
__ASM_EMIT("paddd %[x1], %[x0]") /* xmm0 = x0 x1 x2 0 */ |
| 452 |
__ASM_EMIT("movdqa %[x0], %[x2]") /* xmm2 = x0 x1 x2 0 */ |
| 453 |
__ASM_EMIT("movdqa %[x0], %[x1]") /* xmm1 = x0 x1 x2 0 */ |
| 454 |
__ASM_EMIT("psrldq $8, %[x2]") /* xmm2 = x2 0 0 0 */ |
| 455 |
__ASM_EMIT("psrldq $4, %[x1]") /* xmm1 = x1 x2 0 0 */ |
| 456 |
__ASM_EMIT("por %[x2], %[x0]") |
| 457 |
__ASM_EMIT("por %[x1], %[x0]") |
| 458 |
__ASM_EMIT32("movd %[x0], %[t0]") |
| 459 |
__ASM_EMIT64("movd %[x0], %k[t0]") |
| 460 |
: [t0] "=&r" (t0), |
| 461 |
[x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), |
| 462 |
[x4] "=&x" (x4) |
| 463 |
: [pl] "r" (pl), [pv] "r" (pv), |
| 464 |
[k] "r" (k), |
| 465 |
[PTOL] "m" (X_3D_TOLERANCE), |
| 466 |
[MTOL] "m" (X_3D_MTOLERANCE), |
| 467 |
[ICULL3] "m" (ICULL3) |
| 468 |
: "cc" |
| 469 |
); |
| 470 |
|
| 471 |
#define STR_COPY_TO(tgt) \ |
| 472 |
ARCH_X86_ASM( \ |
| 473 |
__ASM_EMIT("movups 0x00(%[src]), %[x0]") \ |
| 474 |
__ASM_EMIT("movups 0x10(%[src]), %[x1]") \ |
| 475 |
__ASM_EMIT("movups 0x20(%[src]), %[x2]") \ |
| 476 |
__ASM_EMIT("movups %[x0], 0x00(%[dst])") \ |
| 477 |
__ASM_EMIT("movups %[x1], 0x10(%[dst])") \ |
| 478 |
__ASM_EMIT("movups %[x2], 0x20(%[dst])") \ |
| 479 |
: [x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2) \ |
| 480 |
: [dst] "r" (tgt), \ |
| 481 |
[src] "r" (pv) \ |
| 482 |
: "cc", "memory" \ |
| 483 |
); |
| 484 |
|
| 485 |
#define STR_SPLIT_1P(off0, off1, koff, store) \ |
| 486 |
ARCH_X86_ASM( \ |
| 487 |
__ASM_EMIT("movups 0x" off0 "(%[st]), %[x0]") /* xmm0 = p0 = lx0 ly0 lz0 1 */ \ |
| 488 |
__ASM_EMIT("movups 0x" off1 "(%[st]), %[x1]") /* xmm1 = p1 = lx1 ly1 lz1 1 */ \ |
| 489 |
__ASM_EMIT("movups (%[pl]), %[x2]") /* xmm2 = pl = nx ny nz nw */ \ |
| 490 |
\ |
| 491 |
__ASM_EMIT("subps %[x0], %[x1]") /* xmm1 = d = p1 - p0 = dx dy dz 0 */ \ |
| 492 |
__ASM_EMIT("movss 0x" koff "(%[k]), %[x3]") /* xmm3 = k 0 0 0 */ \ |
| 493 |
__ASM_EMIT("mulps %[x1], %[x2]") /* xmm2 = dx*nx dy*ny dz*nz 0 */ \ |
| 494 |
__ASM_EMIT("shufps $0x00, %[x3], %[x3]") /* xmm3 = k k k k */ \ |
| 495 |
__ASM_EMIT("movhlps %[x2], %[x4]") /* xmm4 = dz*nz 0 ? ? */ \ |
| 496 |
__ASM_EMIT("addps %[x4], %[x2]") /* xmm2 = dx*nx+dz*nz dy*ny ? ? */ \ |
| 497 |
__ASM_EMIT("unpcklps %[x2], %[x2]") /* xmm2 = dx*nx+dz*nz dx*nx+dz*nz dy*ny dy*ny */ \ |
| 498 |
__ASM_EMIT("mulps %[x3], %[x1]") /* xmm1 = k*dx k*dy k*dz 0 */ \ |
| 499 |
__ASM_EMIT("movhlps %[x2], %[x4]") /* xmm4 = dy*ny dy*ny ? ? */ \ |
| 500 |
__ASM_EMIT("addps %[x4], %[x2]") /* xmm2 = dx*nx+dy*ny+dw*nw dx*nx+dy*ny+dw*nw ? ? = b b ? ? */ \ |
| 501 |
__ASM_EMIT("unpcklps %[x2], %[x2]") /* xmm2 = b b b b */ \ |
| 502 |
__ASM_EMIT("divps %[x2], %[x1]") /* xmm1 = k*dx/b k*dy/b k*dz/b 0 */ \ |
| 503 |
__ASM_EMIT("subps %[x1], %[x0]") /* xmm0 = lx0-k*dx/b ly0-k*dy/b lz0-k*dz/b 1 */ \ |
| 504 |
\ |
| 505 |
store \ |
| 506 |
\ |
| 507 |
: [x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), [x4] "=&x" (x4) \ |
| 508 |
: [st] "r" (pv), [pl] "r" (pl), [out] "r" (out), [in] "r" (in), \ |
| 509 |
[k] "r" (k) \ |
| 510 |
: "cc", "memory" \ |
| 511 |
); |
| 512 |
|
| 513 |
#define STR_SPLIT_2P(off0, off1, off2, koff, store) \ |
| 514 |
ARCH_X86_ASM( \ |
| 515 |
__ASM_EMIT("movups 0x" off0 "(%[st]), %[x0]") /* xmm0 = p0 = lx0 ly0 lz0 1 */ \ |
| 516 |
__ASM_EMIT("movups 0x" off1 "(%[st]), %[x2]") /* xmm2 = p1 = lx1 ly1 lz1 1 */ \ |
| 517 |
__ASM_EMIT("movups 0x" off2 "(%[st]), %[x3]") /* xmm3 = p2 = lx2 ly2 lz2 1 */ \ |
| 518 |
__ASM_EMIT("movups (%[pl]), %[x4]") /* xmm4 = pl = nx ny nz nw */ \ |
| 519 |
\ |
| 520 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = pl = nx ny nz nw */ \ |
| 521 |
__ASM_EMIT("movaps %[x4], %[x5]") /* xmm5 = pl = nx ny nz nw */ \ |
| 522 |
__ASM_EMIT("subps %[x0], %[x2]") /* xmm2 = d1 = p1 - p0 = dx1 dy1 dz1 0 */ \ |
| 523 |
__ASM_EMIT("subps %[x1], %[x3]") /* xmm3 = d2 = p2 - p0 = dx2 dy2 dz2 0 */ \ |
| 524 |
__ASM_EMIT("mulps %[x2], %[x4]") /* xmm4 = dx1*nx dy1*ny dz1*nz 0 */ \ |
| 525 |
__ASM_EMIT("mulps %[x3], %[x5]") /* xmm5 = dx2*nx dy2*ny dz2*nz 0 */ \ |
| 526 |
__ASM_EMIT("movhlps %[x4], %[x6]") /* xmm6 = dz1*nz 0 ? ? */ \ |
| 527 |
__ASM_EMIT("movhlps %[x5], %[x7]") /* xmm7 = dz2*nz 0 ? ? */ \ |
| 528 |
__ASM_EMIT("addps %[x6], %[x4]") /* xmm4 = dx1*nx+dz1*nz dy1*ny ? ? */ \ |
| 529 |
__ASM_EMIT("addps %[x7], %[x5]") /* xmm5 = dx1*nx+dz1*nz dy1*ny ? ? */ \ |
| 530 |
__ASM_EMIT("unpcklps %[x4], %[x4]") /* xmm4 = dx1*nx+dz1*nz dx1*nx+dz1*nz dy1*ny dy1*ny */ \ |
| 531 |
__ASM_EMIT("unpcklps %[x5], %[x5]") /* xmm5 = dx2*nx+dz2*nz dx2*nx+dz2*nz dy2*ny dy2*ny */ \ |
| 532 |
__ASM_EMIT("movhlps %[x4], %[x6]") /* xmm6 = dy1*ny dy1*ny ? ? */ \ |
| 533 |
__ASM_EMIT("movhlps %[x5], %[x7]") /* xmm7 = dy2*ny dy2*ny ? ? */ \ |
| 534 |
__ASM_EMIT("addps %[x6], %[x4]") /* xmm4 = dx1*nx+dy1*ny+dz1*nz dx1*nx+dy1*ny+dz1*nz ? ? = b1 b1 ? ? */ \ |
| 535 |
__ASM_EMIT("addps %[x7], %[x5]") /* xmm5 = dx2*nx+dy2*ny+dz2*nz dx2*nx+dy2*ny+dz2*nz ? ? = b2 b2 ? ? */ \ |
| 536 |
__ASM_EMIT("unpcklps %[x4], %[x4]") /* xmm4 = b1 b1 b1 b1 */ \ |
| 537 |
__ASM_EMIT("unpcklps %[x5], %[x5]") /* xmm5 = b2 b2 b2 b2 */ \ |
| 538 |
__ASM_EMIT("movss 0x" koff "(%[k]), %[x6]") /* xmm6 = k */ \ |
| 539 |
__ASM_EMIT("divps %[x4], %[x2]") /* xmm2 = dx1/b1 dy1/b1 dz1/b1 0 */ \ |
| 540 |
__ASM_EMIT("shufps $0x00, %[x6], %[x6]") /* xmm6 = k k k k */ \ |
| 541 |
__ASM_EMIT("divps %[x5], %[x3]") /* xmm3 = dx2/b2 dy2/b2 dz2/b2 0 */ \ |
| 542 |
__ASM_EMIT("mulps %[x6], %[x2]") /* xmm2 = k*dx1/b1 k*dy1/b1 k*dz1/b1 0 */ \ |
| 543 |
__ASM_EMIT("mulps %[x6], %[x3]") /* xmm3 = k*dx2/b2 k*dy2/b2 k*dz2/b2 0 */ \ |
| 544 |
__ASM_EMIT("subps %[x2], %[x0]") /* xmm0 = lx0-k*dx1/b1 ly0-k*dy1/b1 lz0-k*dz1/b1 1 */ \ |
| 545 |
__ASM_EMIT("subps %[x3], %[x1]") /* xmm1 = lx1-k*dx2/b2 ly1-k*dy2/b2 lz1-k*dz2/b2 1 */ \ |
| 546 |
\ |
| 547 |
store \ |
| 548 |
\ |
| 549 |
: [x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), \ |
| 550 |
[x4] "=&x" (x4), [x5] "=&x" (x5), [x6] "=&x" (x6), [x7] "=&x" (x7) \ |
| 551 |
: [st] "r" (pv), [pl] "r" (pl), [out] "r" (out), [in] "r" (in), \ |
| 552 |
[k] "r" (k) \ |
| 553 |
: "cc", "memory" \ |
| 554 |
); |
| 555 |
|
| 556 |
switch (t0) |
| 557 |
{ |
| 558 |
// 0 intersections, triangle is above |
| 559 |
case 0x00: // 0 0 0 |
| 560 |
case 0x01: // 0 0 1 |
| 561 |
case 0x04: // 0 1 0 |
| 562 |
case 0x05: // 0 1 1 |
| 563 |
case 0x10: // 1 0 0 |
| 564 |
case 0x11: // 1 0 1 |
| 565 |
case 0x14: // 1 1 0 |
| 566 |
case 0x15: // 1 1 1 |
| 567 |
STR_COPY_TO(out); |
| 568 |
++(*n_out); |
| 569 |
break; |
| 570 |
|
| 571 |
// 0 intersections, triangle is below |
| 572 |
case 0x16: // 1 1 2 |
| 573 |
case 0x19: // 1 2 1 |
| 574 |
case 0x1a: // 1 2 2 |
| 575 |
case 0x25: // 2 1 1 |
| 576 |
case 0x26: // 2 1 2 |
| 577 |
case 0x29: // 2 2 1 |
| 578 |
case 0x2a: // 2 2 2 |
| 579 |
STR_COPY_TO(in); |
| 580 |
++(*n_in); |
| 581 |
break; |
| 582 |
|
| 583 |
// 1 intersection, 1 triangle above, 1 triangle below, counter-clockwise |
| 584 |
case 0x06: // 0 1 2 |
| 585 |
STR_SPLIT_1P("00", "20", "00", |
| 586 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 587 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 588 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 589 |
__ASM_EMIT("movups %[x1], 0x00(%[in])") |
| 590 |
__ASM_EMIT("movups %[x2], 0x10(%[in])") |
| 591 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 592 |
__ASM_EMIT("movups %[x2], 0x00(%[out])") |
| 593 |
__ASM_EMIT("movups %[x3], 0x10(%[out])") |
| 594 |
__ASM_EMIT("movups %[x0], 0x20(%[out])") |
| 595 |
); |
| 596 |
++(*n_out); ++(*n_in); |
| 597 |
break; |
| 598 |
case 0x21: // 2 0 1 |
| 599 |
STR_SPLIT_1P("00", "10", "00", |
| 600 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 601 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 602 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 603 |
__ASM_EMIT("movups %[x3], 0x00(%[in])") |
| 604 |
__ASM_EMIT("movups %[x1], 0x10(%[in])") |
| 605 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 606 |
__ASM_EMIT("movups %[x2], 0x00(%[out])") |
| 607 |
__ASM_EMIT("movups %[x0], 0x10(%[out])") |
| 608 |
__ASM_EMIT("movups %[x1], 0x20(%[out])") |
| 609 |
); |
| 610 |
++(*n_out); ++(*n_in); |
| 611 |
break; |
| 612 |
case 0x18: // 1 2 0 |
| 613 |
STR_SPLIT_1P("10", "20", "04", |
| 614 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 615 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 616 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 617 |
__ASM_EMIT("movups %[x2], 0x00(%[in])") |
| 618 |
__ASM_EMIT("movups %[x3], 0x10(%[in])") |
| 619 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 620 |
__ASM_EMIT("movups %[x1], 0x00(%[out])") |
| 621 |
__ASM_EMIT("movups %[x0], 0x10(%[out])") |
| 622 |
__ASM_EMIT("movups %[x3], 0x20(%[out])") |
| 623 |
); |
| 624 |
++(*n_out); ++(*n_in); |
| 625 |
break; |
| 626 |
|
| 627 |
// 1 intersection, 1 triangle above, 1 triangle below, clockwise |
| 628 |
case 0x24: // 2 1 0 |
| 629 |
STR_SPLIT_1P("00", "20", "00", |
| 630 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 631 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 632 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 633 |
__ASM_EMIT("movups %[x3], 0x00(%[in])") |
| 634 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 635 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 636 |
__ASM_EMIT("movups %[x1], 0x00(%[out])") |
| 637 |
__ASM_EMIT("movups %[x2], 0x10(%[out])") |
| 638 |
__ASM_EMIT("movups %[x0], 0x20(%[out])") |
| 639 |
); |
| 640 |
++(*n_out); ++(*n_in); |
| 641 |
break; |
| 642 |
case 0x12: // 1 0 2 |
| 643 |
STR_SPLIT_1P("00", "10", "00", |
| 644 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 645 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 646 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 647 |
__ASM_EMIT("movups %[x1], 0x00(%[in])") |
| 648 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 649 |
__ASM_EMIT("movups %[x3], 0x20(%[in])") |
| 650 |
__ASM_EMIT("movups %[x2], 0x00(%[out])") |
| 651 |
__ASM_EMIT("movups %[x3], 0x10(%[out])") |
| 652 |
__ASM_EMIT("movups %[x0], 0x20(%[out])") |
| 653 |
); |
| 654 |
++(*n_out); ++(*n_in); |
| 655 |
break; |
| 656 |
case 0x09: // 0 2 1 |
| 657 |
STR_SPLIT_1P("10", "20", "04", |
| 658 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 659 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 660 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 661 |
__ASM_EMIT("movups %[x2], 0x00(%[in])") |
| 662 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 663 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 664 |
__ASM_EMIT("movups %[x3], 0x00(%[out])") |
| 665 |
__ASM_EMIT("movups %[x1], 0x10(%[out])") |
| 666 |
__ASM_EMIT("movups %[x0], 0x20(%[out])") |
| 667 |
); |
| 668 |
++(*n_out); ++(*n_in); |
| 669 |
break; |
| 670 |
|
| 671 |
// 2 intersections, 1 triangle below, 2 triangles above |
| 672 |
case 0x02: // 0 0 2 |
| 673 |
STR_SPLIT_2P("00", "10", "20", "00", |
| 674 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 675 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 676 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 677 |
__ASM_EMIT("movups %[x2], 0x00(%[in])") |
| 678 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 679 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 680 |
__ASM_EMIT("movups %[x3], 0x00(%[out])") |
| 681 |
__ASM_EMIT("movups %[x1], 0x10(%[out])") |
| 682 |
__ASM_EMIT("movups %[x0], 0x20(%[out])") |
| 683 |
__ASM_EMIT("movups %[x4], 0x30(%[out])") |
| 684 |
__ASM_EMIT("movups %[x1], 0x40(%[out])") |
| 685 |
__ASM_EMIT("movups %[x3], 0x50(%[out])") |
| 686 |
); |
| 687 |
(*n_out) += 2; ++(*n_in); |
| 688 |
break; |
| 689 |
case 0x08: // 0 2 0 |
| 690 |
STR_SPLIT_2P("10", "00", "20", "04", |
| 691 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 692 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 693 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 694 |
__ASM_EMIT("movups %[x3], 0x00(%[in])") |
| 695 |
__ASM_EMIT("movups %[x1], 0x10(%[in])") |
| 696 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 697 |
__ASM_EMIT("movups %[x4], 0x00(%[out])") |
| 698 |
__ASM_EMIT("movups %[x0], 0x10(%[out])") |
| 699 |
__ASM_EMIT("movups %[x1], 0x20(%[out])") |
| 700 |
__ASM_EMIT("movups %[x2], 0x30(%[out])") |
| 701 |
__ASM_EMIT("movups %[x0], 0x40(%[out])") |
| 702 |
__ASM_EMIT("movups %[x4], 0x50(%[out])") |
| 703 |
); |
| 704 |
(*n_out) += 2; ++(*n_in); |
| 705 |
break; |
| 706 |
case 0x20: // 2 0 0 |
| 707 |
STR_SPLIT_2P("20", "00", "10", "08", |
| 708 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 709 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 710 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 711 |
__ASM_EMIT("movups %[x4], 0x00(%[in])") |
| 712 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 713 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 714 |
__ASM_EMIT("movups %[x2], 0x00(%[out])") |
| 715 |
__ASM_EMIT("movups %[x1], 0x10(%[out])") |
| 716 |
__ASM_EMIT("movups %[x0], 0x20(%[out])") |
| 717 |
__ASM_EMIT("movups %[x3], 0x30(%[out])") |
| 718 |
__ASM_EMIT("movups %[x1], 0x40(%[out])") |
| 719 |
__ASM_EMIT("movups %[x2], 0x50(%[out])") |
| 720 |
); |
| 721 |
(*n_out) += 2; ++(*n_in); |
| 722 |
break; |
| 723 |
|
| 724 |
// 2 intersections, 1 triangle above, 2 triangles below |
| 725 |
case 0x28: // 2 2 0 |
| 726 |
STR_SPLIT_2P("00", "10", "20", "00", |
| 727 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 728 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 729 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 730 |
__ASM_EMIT("movups %[x3], 0x00(%[in])") |
| 731 |
__ASM_EMIT("movups %[x1], 0x10(%[in])") |
| 732 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 733 |
__ASM_EMIT("movups %[x4], 0x30(%[in])") |
| 734 |
__ASM_EMIT("movups %[x1], 0x40(%[in])") |
| 735 |
__ASM_EMIT("movups %[x3], 0x50(%[in])") |
| 736 |
__ASM_EMIT("movups %[x2], 0x00(%[out])") |
| 737 |
__ASM_EMIT("movups %[x0], 0x10(%[out])") |
| 738 |
__ASM_EMIT("movups %[x1], 0x20(%[out])") |
| 739 |
); |
| 740 |
++(*n_out); (*n_in) += 2; |
| 741 |
break; |
| 742 |
|
| 743 |
case 0x22: // 2 0 2 |
| 744 |
STR_SPLIT_2P("10", "00", "20", "04", |
| 745 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 746 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 747 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 748 |
__ASM_EMIT("movups %[x4], 0x00(%[in])") |
| 749 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 750 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 751 |
__ASM_EMIT("movups %[x2], 0x30(%[in])") |
| 752 |
__ASM_EMIT("movups %[x0], 0x40(%[in])") |
| 753 |
__ASM_EMIT("movups %[x4], 0x50(%[in])") |
| 754 |
__ASM_EMIT("movups %[x3], 0x00(%[out])") |
| 755 |
__ASM_EMIT("movups %[x1], 0x10(%[out])") |
| 756 |
__ASM_EMIT("movups %[x0], 0x20(%[out])") |
| 757 |
); |
| 758 |
++(*n_out); (*n_in) += 2; |
| 759 |
break; |
| 760 |
|
| 761 |
case 0x0a: // 0 2 2 |
| 762 |
STR_SPLIT_2P("20", "00", "10", "08", |
| 763 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 764 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 765 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 766 |
__ASM_EMIT("movups %[x2], 0x00(%[in])") |
| 767 |
__ASM_EMIT("movups %[x1], 0x10(%[in])") |
| 768 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 769 |
__ASM_EMIT("movups %[x3], 0x30(%[in])") |
| 770 |
__ASM_EMIT("movups %[x1], 0x40(%[in])") |
| 771 |
__ASM_EMIT("movups %[x2], 0x50(%[in])") |
| 772 |
__ASM_EMIT("movups %[x4], 0x00(%[out])") |
| 773 |
__ASM_EMIT("movups %[x0], 0x10(%[out])") |
| 774 |
__ASM_EMIT("movups %[x1], 0x20(%[out])") |
| 775 |
); |
| 776 |
++(*n_out); (*n_in) += 2; |
| 777 |
break; |
| 778 |
|
| 779 |
default: |
| 780 |
break; |
| 781 |
} |
| 782 |
|
| 783 |
#undef STR_COPY_TO |
| 784 |
#undef STR_SPLIT_1P |
| 785 |
#undef STR_SPLIT_2P |
| 786 |
} |
| 787 |
|
| 788 |
void cull_triangle_raw( |
| 789 |
raw_triangle_t *in, |
| 790 |
size_t *n_in, |
| 791 |
const vector3d_t *pl, |
| 792 |
const raw_triangle_t *pv |
| 793 |
) |
| 794 |
{ |
| 795 |
float x0, x1, x2, x3, x4, x5, x6, x7; |
| 796 |
float k[4] __lsp_aligned16; |
| 797 |
size_t t0; |
| 798 |
|
| 799 |
in += *n_in; |
| 800 |
|
| 801 |
// Estimate co-location of triangle and point |
| 802 |
ARCH_X86_ASM |
| 803 |
( |
| 804 |
__ASM_EMIT("movups (%[pl]), %[x3]") /* xmm3 = pl */ |
| 805 |
__ASM_EMIT("movups 0x00(%[pv]), %[x0]") /* xmm0 = p0 */ |
| 806 |
__ASM_EMIT("movups 0x10(%[pv]), %[x1]") /* xmm1 = p1 */ |
| 807 |
__ASM_EMIT("movups 0x20(%[pv]), %[x2]") /* xmm2 = p2 */ |
| 808 |
|
| 809 |
__ASM_EMIT("mulps %[x3], %[x0]") /* xmm0 = p0 * pl */ |
| 810 |
__ASM_EMIT("mulps %[x3], %[x1]") /* xmm1 = p1 * pl */ |
| 811 |
__ASM_EMIT("mulps %[x3], %[x2]") /* xmm2 = p2 * pl */ |
| 812 |
MAT4_TRANSPOSE("[x0]", "[x1]", "[x2]", "[x3]", "[x4]") |
| 813 |
|
| 814 |
__ASM_EMIT("addps %[x1], %[x0]") |
| 815 |
__ASM_EMIT("addps %[x3], %[x2]") |
| 816 |
__ASM_EMIT("addps %[x2], %[x0]") /* xmm0 = k0 k1 k2 ? */ |
| 817 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = k0 k1 k2 ? */ |
| 818 |
__ASM_EMIT("movaps %[x0], (%[k])") /* *k = xmm0 */ |
| 819 |
__ASM_EMIT("cmpps $2, %[PTOL], %[x0]") /* xmm0 = k0 <= +TOL k1 <= +TOL k2 <= +TOL ? */ |
| 820 |
__ASM_EMIT("cmpps $1, %[MTOL], %[x1]") /* xmm1 = k0 < -TOL k1 < -TOL k2 < -TOL ? */ |
| 821 |
__ASM_EMIT("pand %[ICULL3], %[x0]") /* xmm0 = 1*[k0 <= +TOL] 1*[k1 <= +TOL] 1*[k2 <= +TOL] ? */ |
| 822 |
__ASM_EMIT("pand %[ICULL3], %[x1]") /* xmm1 = 1*[k0 < -TOL] 1*[k1 < -TOL] 1*[k2 < -TOL] ? */ |
| 823 |
__ASM_EMIT("paddd %[x1], %[x0]") /* xmm0 = x0 x1 x2 0 */ |
| 824 |
__ASM_EMIT("movdqa %[x0], %[x2]") /* xmm2 = x0 x1 x2 0 */ |
| 825 |
__ASM_EMIT("movdqa %[x0], %[x1]") /* xmm1 = x0 x1 x2 0 */ |
| 826 |
__ASM_EMIT("psrldq $8, %[x2]") /* xmm2 = x2 0 0 0 */ |
| 827 |
__ASM_EMIT("psrldq $4, %[x1]") /* xmm1 = x1 x2 0 0 */ |
| 828 |
__ASM_EMIT("por %[x2], %[x0]") |
| 829 |
__ASM_EMIT("por %[x1], %[x0]") |
| 830 |
__ASM_EMIT32("movd %[x0], %[t0]") |
| 831 |
__ASM_EMIT64("movd %[x0], %k[t0]") |
| 832 |
: [t0] "=&r" (t0), |
| 833 |
[x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), |
| 834 |
[x4] "=&x" (x4) |
| 835 |
: [pl] "r" (pl), [pv] "r" (pv), |
| 836 |
[k] "r" (k), |
| 837 |
[PTOL] "m" (X_3D_TOLERANCE), |
| 838 |
[MTOL] "m" (X_3D_MTOLERANCE), |
| 839 |
[ICULL3] "m" (ICULL3) |
| 840 |
: "cc" |
| 841 |
); |
| 842 |
|
| 843 |
#define STR_COPY_TO(tgt) \ |
| 844 |
ARCH_X86_ASM( \ |
| 845 |
__ASM_EMIT("movups 0x00(%[src]), %[x0]") \ |
| 846 |
__ASM_EMIT("movups 0x10(%[src]), %[x1]") \ |
| 847 |
__ASM_EMIT("movups 0x20(%[src]), %[x2]") \ |
| 848 |
__ASM_EMIT("movups %[x0], 0x00(%[dst])") \ |
| 849 |
__ASM_EMIT("movups %[x1], 0x10(%[dst])") \ |
| 850 |
__ASM_EMIT("movups %[x2], 0x20(%[dst])") \ |
| 851 |
: [x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2) \ |
| 852 |
: [dst] "r" (tgt), \ |
| 853 |
[src] "r" (pv) \ |
| 854 |
: "cc", "memory" \ |
| 855 |
); |
| 856 |
|
| 857 |
#define STR_SPLIT_1P(off0, off1, koff, store) \ |
| 858 |
ARCH_X86_ASM( \ |
| 859 |
__ASM_EMIT("movups 0x" off0 "(%[st]), %[x0]") /* xmm0 = p0 = lx0 ly0 lz0 1 */ \ |
| 860 |
__ASM_EMIT("movups 0x" off1 "(%[st]), %[x1]") /* xmm1 = p1 = lx1 ly1 lz1 1 */ \ |
| 861 |
__ASM_EMIT("movups (%[pl]), %[x2]") /* xmm2 = pl = nx ny nz nw */ \ |
| 862 |
\ |
| 863 |
__ASM_EMIT("subps %[x0], %[x1]") /* xmm1 = d = p1 - p0 = dx dy dz 0 */ \ |
| 864 |
__ASM_EMIT("movss 0x" koff "(%[k]), %[x3]") /* xmm3 = k 0 0 0 */ \ |
| 865 |
__ASM_EMIT("mulps %[x1], %[x2]") /* xmm2 = dx*nx dy*ny dz*nz 0 */ \ |
| 866 |
__ASM_EMIT("shufps $0x00, %[x3], %[x3]") /* xmm3 = k k k k */ \ |
| 867 |
__ASM_EMIT("movhlps %[x2], %[x4]") /* xmm4 = dz*nz 0 ? ? */ \ |
| 868 |
__ASM_EMIT("addps %[x4], %[x2]") /* xmm2 = dx*nx+dz*nz dy*ny ? ? */ \ |
| 869 |
__ASM_EMIT("unpcklps %[x2], %[x2]") /* xmm2 = dx*nx+dz*nz dx*nx+dz*nz dy*ny dy*ny */ \ |
| 870 |
__ASM_EMIT("mulps %[x3], %[x1]") /* xmm1 = k*dx k*dy k*dz 0 */ \ |
| 871 |
__ASM_EMIT("movhlps %[x2], %[x4]") /* xmm4 = dy*ny dy*ny ? ? */ \ |
| 872 |
__ASM_EMIT("addps %[x4], %[x2]") /* xmm2 = dx*nx+dy*ny+dw*nw dx*nx+dy*ny+dw*nw ? ? = b b ? ? */ \ |
| 873 |
__ASM_EMIT("unpcklps %[x2], %[x2]") /* xmm2 = b b b b */ \ |
| 874 |
__ASM_EMIT("divps %[x2], %[x1]") /* xmm1 = k*dx/b k*dy/b k*dz/b 0 */ \ |
| 875 |
__ASM_EMIT("subps %[x1], %[x0]") /* xmm0 = lx0-k*dx/b ly0-k*dy/b lz0-k*dz/b 1 */ \ |
| 876 |
\ |
| 877 |
store \ |
| 878 |
\ |
| 879 |
: [x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), [x4] "=&x" (x4) \ |
| 880 |
: [st] "r" (pv), [pl] "r" (pl), [in] "r" (in), \ |
| 881 |
[k] "r" (k) \ |
| 882 |
: "cc", "memory" \ |
| 883 |
); |
| 884 |
|
| 885 |
#define STR_SPLIT_2P(off0, off1, off2, koff, store) \ |
| 886 |
ARCH_X86_ASM( \ |
| 887 |
__ASM_EMIT("movups 0x" off0 "(%[st]), %[x0]") /* xmm0 = p0 = lx0 ly0 lz0 1 */ \ |
| 888 |
__ASM_EMIT("movups 0x" off1 "(%[st]), %[x2]") /* xmm2 = p1 = lx1 ly1 lz1 1 */ \ |
| 889 |
__ASM_EMIT("movups 0x" off2 "(%[st]), %[x3]") /* xmm3 = p2 = lx2 ly2 lz2 1 */ \ |
| 890 |
__ASM_EMIT("movups (%[pl]), %[x4]") /* xmm4 = pl = nx ny nz nw */ \ |
| 891 |
\ |
| 892 |
__ASM_EMIT("movaps %[x0], %[x1]") /* xmm1 = pl = nx ny nz nw */ \ |
| 893 |
__ASM_EMIT("movaps %[x4], %[x5]") /* xmm5 = pl = nx ny nz nw */ \ |
| 894 |
__ASM_EMIT("subps %[x0], %[x2]") /* xmm2 = d1 = p1 - p0 = dx1 dy1 dz1 0 */ \ |
| 895 |
__ASM_EMIT("subps %[x1], %[x3]") /* xmm3 = d2 = p2 - p0 = dx2 dy2 dz2 0 */ \ |
| 896 |
__ASM_EMIT("mulps %[x2], %[x4]") /* xmm4 = dx1*nx dy1*ny dz1*nz 0 */ \ |
| 897 |
__ASM_EMIT("mulps %[x3], %[x5]") /* xmm5 = dx2*nx dy2*ny dz2*nz 0 */ \ |
| 898 |
__ASM_EMIT("movhlps %[x4], %[x6]") /* xmm6 = dz1*nz 0 ? ? */ \ |
| 899 |
__ASM_EMIT("movhlps %[x5], %[x7]") /* xmm7 = dz2*nz 0 ? ? */ \ |
| 900 |
__ASM_EMIT("addps %[x6], %[x4]") /* xmm4 = dx1*nx+dz1*nz dy1*ny ? ? */ \ |
| 901 |
__ASM_EMIT("addps %[x7], %[x5]") /* xmm5 = dx1*nx+dz1*nz dy1*ny ? ? */ \ |
| 902 |
__ASM_EMIT("unpcklps %[x4], %[x4]") /* xmm4 = dx1*nx+dz1*nz dx1*nx+dz1*nz dy1*ny dy1*ny */ \ |
| 903 |
__ASM_EMIT("unpcklps %[x5], %[x5]") /* xmm5 = dx2*nx+dz2*nz dx2*nx+dz2*nz dy2*ny dy2*ny */ \ |
| 904 |
__ASM_EMIT("movhlps %[x4], %[x6]") /* xmm6 = dy1*ny dy1*ny ? ? */ \ |
| 905 |
__ASM_EMIT("movhlps %[x5], %[x7]") /* xmm7 = dy2*ny dy2*ny ? ? */ \ |
| 906 |
__ASM_EMIT("addps %[x6], %[x4]") /* xmm4 = dx1*nx+dy1*ny+dz1*nz dx1*nx+dy1*ny+dz1*nz ? ? = b1 b1 ? ? */ \ |
| 907 |
__ASM_EMIT("addps %[x7], %[x5]") /* xmm5 = dx2*nx+dy2*ny+dz2*nz dx2*nx+dy2*ny+dz2*nz ? ? = b2 b2 ? ? */ \ |
| 908 |
__ASM_EMIT("unpcklps %[x4], %[x4]") /* xmm4 = b1 b1 b1 b1 */ \ |
| 909 |
__ASM_EMIT("unpcklps %[x5], %[x5]") /* xmm5 = b2 b2 b2 b2 */ \ |
| 910 |
__ASM_EMIT("movss 0x" koff "(%[k]), %[x6]") /* xmm6 = k */ \ |
| 911 |
__ASM_EMIT("divps %[x4], %[x2]") /* xmm2 = dx1/b1 dy1/b1 dz1/b1 0 */ \ |
| 912 |
__ASM_EMIT("shufps $0x00, %[x6], %[x6]") /* xmm6 = k k k k */ \ |
| 913 |
__ASM_EMIT("divps %[x5], %[x3]") /* xmm3 = dx2/b2 dy2/b2 dz2/b2 0 */ \ |
| 914 |
__ASM_EMIT("mulps %[x6], %[x2]") /* xmm2 = k*dx1/b1 k*dy1/b1 k*dz1/b1 0 */ \ |
| 915 |
__ASM_EMIT("mulps %[x6], %[x3]") /* xmm3 = k*dx2/b2 k*dy2/b2 k*dz2/b2 0 */ \ |
| 916 |
__ASM_EMIT("subps %[x2], %[x0]") /* xmm0 = lx0-k*dx1/b1 ly0-k*dy1/b1 lz0-k*dz1/b1 1 */ \ |
| 917 |
__ASM_EMIT("subps %[x3], %[x1]") /* xmm1 = lx1-k*dx2/b2 ly1-k*dy2/b2 lz1-k*dz2/b2 1 */ \ |
| 918 |
\ |
| 919 |
store \ |
| 920 |
\ |
| 921 |
: [x0] "=&x" (x0), [x1] "=&x" (x1), [x2] "=&x" (x2), [x3] "=&x" (x3), \ |
| 922 |
[x4] "=&x" (x4), [x5] "=&x" (x5), [x6] "=&x" (x6), [x7] "=&x" (x7) \ |
| 923 |
: [st] "r" (pv), [pl] "r" (pl), [in] "r" (in), \ |
| 924 |
[k] "r" (k) \ |
| 925 |
: "cc", "memory" \ |
| 926 |
); |
| 927 |
|
| 928 |
switch (t0) |
| 929 |
{ |
| 930 |
// 0 intersections, triangle is above |
| 931 |
case 0x00: // 0 0 0 |
| 932 |
case 0x01: // 0 0 1 |
| 933 |
case 0x04: // 0 1 0 |
| 934 |
case 0x05: // 0 1 1 |
| 935 |
case 0x10: // 1 0 0 |
| 936 |
case 0x11: // 1 0 1 |
| 937 |
case 0x14: // 1 1 0 |
| 938 |
case 0x15: // 1 1 1 |
| 939 |
break; |
| 940 |
|
| 941 |
// 0 intersections, triangle is below |
| 942 |
case 0x16: // 1 1 2 |
| 943 |
case 0x19: // 1 2 1 |
| 944 |
case 0x1a: // 1 2 2 |
| 945 |
case 0x25: // 2 1 1 |
| 946 |
case 0x26: // 2 1 2 |
| 947 |
case 0x29: // 2 2 1 |
| 948 |
case 0x2a: // 2 2 2 |
| 949 |
STR_COPY_TO(in); |
| 950 |
++(*n_in); |
| 951 |
break; |
| 952 |
|
| 953 |
// 1 intersection, 1 triangle above, 1 triangle below, counter-clockwise |
| 954 |
case 0x06: // 0 1 2 |
| 955 |
STR_SPLIT_1P("00", "20", "00", |
| 956 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 957 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 958 |
__ASM_EMIT("movups %[x1], 0x00(%[in])") |
| 959 |
__ASM_EMIT("movups %[x2], 0x10(%[in])") |
| 960 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 961 |
); |
| 962 |
++(*n_in); |
| 963 |
break; |
| 964 |
case 0x21: // 2 0 1 |
| 965 |
STR_SPLIT_1P("00", "10", "00", |
| 966 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 967 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 968 |
__ASM_EMIT("movups %[x3], 0x00(%[in])") |
| 969 |
__ASM_EMIT("movups %[x1], 0x10(%[in])") |
| 970 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 971 |
); |
| 972 |
++(*n_in); |
| 973 |
break; |
| 974 |
case 0x18: // 1 2 0 |
| 975 |
STR_SPLIT_1P("10", "20", "04", |
| 976 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 977 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 978 |
__ASM_EMIT("movups %[x2], 0x00(%[in])") |
| 979 |
__ASM_EMIT("movups %[x3], 0x10(%[in])") |
| 980 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 981 |
); |
| 982 |
++(*n_in); |
| 983 |
break; |
| 984 |
|
| 985 |
// 1 intersection, 1 triangle above, 1 triangle below, clockwise |
| 986 |
case 0x24: // 2 1 0 |
| 987 |
STR_SPLIT_1P("00", "20", "00", |
| 988 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 989 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 990 |
__ASM_EMIT("movups %[x3], 0x00(%[in])") |
| 991 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 992 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 993 |
); |
| 994 |
++(*n_in); |
| 995 |
break; |
| 996 |
case 0x12: // 1 0 2 |
| 997 |
STR_SPLIT_1P("00", "10", "00", |
| 998 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 999 |
__ASM_EMIT("movups 0x20(%[st]), %[x3]") |
| 1000 |
__ASM_EMIT("movups %[x1], 0x00(%[in])") |
| 1001 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 1002 |
__ASM_EMIT("movups %[x3], 0x20(%[in])") |
| 1003 |
); |
| 1004 |
++(*n_in); |
| 1005 |
break; |
| 1006 |
case 0x09: // 0 2 1 |
| 1007 |
STR_SPLIT_1P("10", "20", "04", |
| 1008 |
__ASM_EMIT("movups 0x00(%[st]), %[x1]") |
| 1009 |
__ASM_EMIT("movups 0x10(%[st]), %[x2]") |
| 1010 |
__ASM_EMIT("movups %[x2], 0x00(%[in])") |
| 1011 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 1012 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 1013 |
); |
| 1014 |
++(*n_in); |
| 1015 |
break; |
| 1016 |
|
| 1017 |
// 2 intersections, 1 triangle below, 2 triangles above |
| 1018 |
case 0x02: // 0 0 2 |
| 1019 |
STR_SPLIT_2P("00", "10", "20", "00", |
| 1020 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 1021 |
__ASM_EMIT("movups %[x2], 0x00(%[in])") |
| 1022 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 1023 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 1024 |
); |
| 1025 |
++(*n_in); |
| 1026 |
break; |
| 1027 |
case 0x08: // 0 2 0 |
| 1028 |
STR_SPLIT_2P("10", "00", "20", "04", |
| 1029 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 1030 |
__ASM_EMIT("movups %[x3], 0x00(%[in])") |
| 1031 |
__ASM_EMIT("movups %[x1], 0x10(%[in])") |
| 1032 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 1033 |
); |
| 1034 |
++(*n_in); |
| 1035 |
break; |
| 1036 |
case 0x20: // 2 0 0 |
| 1037 |
STR_SPLIT_2P("20", "00", "10", "08", |
| 1038 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 1039 |
__ASM_EMIT("movups %[x4], 0x00(%[in])") |
| 1040 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 1041 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 1042 |
); |
| 1043 |
++(*n_in); |
| 1044 |
break; |
| 1045 |
|
| 1046 |
// 2 intersections, 1 triangle above, 2 triangles below |
| 1047 |
case 0x28: // 2 2 0 |
| 1048 |
STR_SPLIT_2P("00", "10", "20", "00", |
| 1049 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 1050 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 1051 |
__ASM_EMIT("movups %[x3], 0x00(%[in])") |
| 1052 |
__ASM_EMIT("movups %[x1], 0x10(%[in])") |
| 1053 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 1054 |
__ASM_EMIT("movups %[x4], 0x30(%[in])") |
| 1055 |
__ASM_EMIT("movups %[x1], 0x40(%[in])") |
| 1056 |
__ASM_EMIT("movups %[x3], 0x50(%[in])") |
| 1057 |
); |
| 1058 |
(*n_in) += 2; |
| 1059 |
break; |
| 1060 |
|
| 1061 |
case 0x22: // 2 0 2 |
| 1062 |
STR_SPLIT_2P("10", "00", "20", "04", |
| 1063 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 1064 |
__ASM_EMIT("movups 0x20(%[st]), %[x4]") |
| 1065 |
__ASM_EMIT("movups %[x4], 0x00(%[in])") |
| 1066 |
__ASM_EMIT("movups %[x0], 0x10(%[in])") |
| 1067 |
__ASM_EMIT("movups %[x1], 0x20(%[in])") |
| 1068 |
__ASM_EMIT("movups %[x2], 0x30(%[in])") |
| 1069 |
__ASM_EMIT("movups %[x0], 0x40(%[in])") |
| 1070 |
__ASM_EMIT("movups %[x4], 0x50(%[in])") |
| 1071 |
); |
| 1072 |
(*n_in) += 2; |
| 1073 |
break; |
| 1074 |
|
| 1075 |
case 0x0a: // 0 2 2 |
| 1076 |
STR_SPLIT_2P("20", "00", "10", "08", |
| 1077 |
__ASM_EMIT("movups 0x00(%[st]), %[x2]") |
| 1078 |
__ASM_EMIT("movups 0x10(%[st]), %[x3]") |
| 1079 |
__ASM_EMIT("movups %[x2], 0x00(%[in])") |
| 1080 |
__ASM_EMIT("movups %[x1], 0x10(%[in])") |
| 1081 |
__ASM_EMIT("movups %[x0], 0x20(%[in])") |
| 1082 |
__ASM_EMIT("movups %[x3], 0x30(%[in])") |
| 1083 |
__ASM_EMIT("movups %[x1], 0x40(%[in])") |
| 1084 |
__ASM_EMIT("movups %[x2], 0x50(%[in])") |
| 1085 |
); |
| 1086 |
(*n_in) += 2; |
| 1087 |
break; |
| 1088 |
|
| 1089 |
default: |
| 1090 |
break; |
| 1091 |
} |
| 1092 |
|
| 1093 |
#undef STR_COPY_TO |
| 1094 |
#undef STR_SPLIT_1P |
| 1095 |
#undef STR_SPLIT_2P |
| 1096 |
} |
| 1097 |
|
| 1098 |
} /* namespace sse2 */ |
| 1099 |
} /* namespace lsp */ |
| 1100 |
|
| 1101 |
#undef MAT4_TRANSPOSE |
| 1102 |
#undef MAT4_LOAD |
| 1103 |
#undef MAT4_STORE |
| 1104 |
|
| 1105 |
#endif /* PRIVATE_DSP_ARCH_X86_SSE2_3DMATH_H_ */ |