1 /*
2     Copyright (c) 2005-2021 Intel Corporation
3 
4     Licensed under the Apache License, Version 2.0 (the "License");
5     you may not use this file except in compliance with the License.
6     You may obtain a copy of the License at
7 
8         http://www.apache.org/licenses/LICENSE-2.0
9 
10     Unless required by applicable law or agreed to in writing, software
11     distributed under the License is distributed on an "AS IS" BASIS,
12     WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13     See the License for the specific language governing permissions and
14     limitations under the License.
15 */
16 
17 /*
18     The original source for this example is
19     Copyright (c) 1994-2008 John E. Stone
20     All rights reserved.
21 
22     Redistribution and use in source and binary forms, with or without
23     modification, are permitted provided that the following conditions
24     are met:
25     1. Redistributions of source code must retain the above copyright
26        notice, this list of conditions and the following disclaimer.
27     2. Redistributions in binary form must reproduce the above copyright
28        notice, this list of conditions and the following disclaimer in the
29        documentation and/or other materials provided with the distribution.
30     3. The name of the author may not be used to endorse or promote products
31        derived from this software without specific prior written permission.
32 
33     THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
34     OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
35     WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
36     ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
37     DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
38     DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
39     OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
40     HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
41     LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
42     OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43     SUCH DAMAGE.
44 */
45 
46 #include <omp.h>
47 #include <string.h>
48 
49 #include "machine.hpp"
50 #include "types.hpp"
51 #include "macros.hpp"
52 #include "vector.hpp"
53 #include "tgafile.hpp"
54 #include "trace.hpp"
55 #include "light.hpp"
56 #include "shade.hpp"
57 #include "camera.hpp"
58 #include "util.hpp"
59 #include "intersect.hpp"
60 #include "global.hpp"
61 #include "ui.hpp"
62 #include "tachyon_video.hpp"
63 
64 // shared but read-only so could be private too
65 static thr_parms *all_parms;
66 static scenedef scene;
67 static int startx;
68 static int stopx;
69 static int starty;
70 static int stopy;
71 static flt jitterscale;
72 static int totaly, totalx;
73 
74 static int grain_size = 50;
75 const int DIVFACTOR = 2;
76 
77 #define MIN(a, b) ((a) < (b) ? (a) : (b))
78 
render_one_pixel(int x,int y,unsigned int * local_mbox,unsigned int & serial,int startx,int stopx,int starty,int stopy)79 static color_t render_one_pixel(int x,
80                                 int y,
81                                 unsigned int *local_mbox,
82                                 unsigned int &serial,
83                                 int startx,
84                                 int stopx,
85                                 int starty,
86                                 int stopy) {
87     /* private vars moved inside loop */
88     ray primary, sample;
89     color col, avcol;
90     int R, G, B;
91     intersectstruct local_intersections;
92     int alias;
93     /* end private */
94 
95     primary = camray(&scene, x, y);
96     primary.intstruct = &local_intersections;
97     primary.flags = RT_RAY_REGULAR;
98 
99     serial++;
100     primary.serial = serial;
101     primary.mbox = local_mbox;
102     primary.maxdist = FHUGE;
103     primary.scene = &scene;
104     col = trace(&primary);
105 
106     serial = primary.serial;
107 
108     /* perform antialiasing if enabled.. */
109     if (scene.antialiasing > 0) {
110         for (alias = 0; alias < scene.antialiasing; alias++) {
111             serial++; /* increment serial number */
112             sample = primary; /* copy the regular primary ray to start with */
113             sample.serial = serial;
114 
115 #pragma omp critical
116             {
117                 sample.d.x += ((rand() % 100) - 50) / jitterscale;
118                 sample.d.y += ((rand() % 100) - 50) / jitterscale;
119                 sample.d.z += ((rand() % 100) - 50) / jitterscale;
120             }
121 
122             avcol = trace(&sample);
123 
124             serial = sample.serial; /* update our overall serial # */
125 
126             col.r += avcol.r;
127             col.g += avcol.g;
128             col.b += avcol.b;
129         }
130 
131         col.r /= (scene.antialiasing + 1.0);
132         col.g /= (scene.antialiasing + 1.0);
133         col.b /= (scene.antialiasing + 1.0);
134     }
135 
136     /* Handle overexposure and underexposure here... */
137     R = (int)(col.r * 255);
138     if (R > 255)
139         R = 255;
140     else if (R < 0)
141         R = 0;
142 
143     G = (int)(col.g * 255);
144     if (G > 255)
145         G = 255;
146     else if (G < 0)
147         G = 0;
148 
149     B = (int)(col.b * 255);
150     if (B > 255)
151         B = 255;
152     else if (B < 0)
153         B = 0;
154 
155     return video->get_color(R, G, B);
156 }
157 
parallel_thread(patch * pchin,int depth)158 static void parallel_thread(patch *pchin, int depth) {
159     unsigned char col[3];
160     col[0] = col[1] = col[2] = (32 * depth) % 256;
161     depth++;
162 #pragma intel omp taskq firstprivate(depth)
163     {
164         int startx, stopx, starty, stopy;
165         int xs, ys;
166 
167         startx = pchin->startx;
168         stopx = pchin->stopx;
169         starty = pchin->starty;
170         stopy = pchin->stopy;
171 
172         if (((stopx - startx) >= grain_size) || ((stopy - starty) >= grain_size)) {
173             int xpatchsize = (stopx - startx) / DIVFACTOR + 1;
174             int ypatchsize = (stopy - starty) / DIVFACTOR + 1;
175             for (ys = starty; ys <= stopy; ys += ypatchsize)
176                 for (xs = startx; xs <= stopx; xs += xpatchsize) {
177                     patch pch;
178                     pch.startx = xs;
179                     pch.starty = ys;
180                     pch.stopx = MIN(xs + xpatchsize, stopx);
181                     pch.stopy = MIN(ys + ypatchsize, stopy);
182 
183 #pragma intel omp task
184                     parallel_thread(&pch, depth);
185                 }
186         }
187         else {
188             /* just trace this patch */
189             unsigned int mboxsize = sizeof(unsigned int) * (max_objectid() + 20);
190             unsigned int *local_mbox = (unsigned int *)alloca(mboxsize);
191             memset(local_mbox, 0, mboxsize);
192 
193             drawing_area drawing(startx, totaly - stopy, stopx - startx, stopy - starty);
194             for (int i = 1, y = starty; y < stopy; ++y, i++) {
195                 if (!video->running)
196                     continue;
197                 drawing.set_pos(0, drawing.size_y - i);
198                 unsigned int serial = 5 * ((stopx - startx) + (stopy - starty) * totalx);
199                 for (int x = startx; x < stopx; x++) {
200                     color_t c =
201                         render_one_pixel(x, y, local_mbox, serial, startx, stopx, starty, stopy);
202                     drawing.put_pixel(c);
203                 }
204             }
205             video->next_frame();
206         }
207     }
208 }
209 
thread_trace(thr_parms * parms)210 void *thread_trace(thr_parms *parms) {
211     // shared but read-only so could be private too
212     all_parms = parms;
213     scene = parms->scene;
214     startx = parms->startx;
215     stopx = parms->stopx;
216     starty = parms->starty;
217     stopy = parms->stopy;
218     jitterscale = 40.0 * (scene.hres + scene.vres);
219     totalx = parms->stopx - parms->startx + 1;
220     totaly = parms->scene.vres;
221 
222     patch pch;
223     pch.startx = startx;
224     pch.stopx = stopx;
225     pch.starty = starty;
226     pch.stopy = stopy;
227     int g;
228     char *grain_str = getenv("TASKQ_GRAINSIZE");
229     if (grain_str && (sscanf(grain_str, "%d", &g) > 0) && (g > 0))
230         grain_size = g;
231 #pragma omp parallel
232     parallel_thread(&pch, 0);
233 
234     return (nullptr);
235 }
236