1d86ed7fbStbbdev /*
2*b15aabb3Stbbdev     Copyright (c) 2005-2021 Intel Corporation
3d86ed7fbStbbdev 
4d86ed7fbStbbdev     Licensed under the Apache License, Version 2.0 (the "License");
5d86ed7fbStbbdev     you may not use this file except in compliance with the License.
6d86ed7fbStbbdev     You may obtain a copy of the License at
7d86ed7fbStbbdev 
8d86ed7fbStbbdev         http://www.apache.org/licenses/LICENSE-2.0
9d86ed7fbStbbdev 
10d86ed7fbStbbdev     Unless required by applicable law or agreed to in writing, software
11d86ed7fbStbbdev     distributed under the License is distributed on an "AS IS" BASIS,
12d86ed7fbStbbdev     WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13d86ed7fbStbbdev     See the License for the specific language governing permissions and
14d86ed7fbStbbdev     limitations under the License.
15d86ed7fbStbbdev */
16d86ed7fbStbbdev 
17d86ed7fbStbbdev /*
18d86ed7fbStbbdev     The original source for this example is
19d86ed7fbStbbdev     Copyright (c) 1994-2008 John E. Stone
20d86ed7fbStbbdev     All rights reserved.
21d86ed7fbStbbdev 
22d86ed7fbStbbdev     Redistribution and use in source and binary forms, with or without
23d86ed7fbStbbdev     modification, are permitted provided that the following conditions
24d86ed7fbStbbdev     are met:
25d86ed7fbStbbdev     1. Redistributions of source code must retain the above copyright
26d86ed7fbStbbdev        notice, this list of conditions and the following disclaimer.
27d86ed7fbStbbdev     2. Redistributions in binary form must reproduce the above copyright
28d86ed7fbStbbdev        notice, this list of conditions and the following disclaimer in the
29d86ed7fbStbbdev        documentation and/or other materials provided with the distribution.
30d86ed7fbStbbdev     3. The name of the author may not be used to endorse or promote products
31d86ed7fbStbbdev        derived from this software without specific prior written permission.
32d86ed7fbStbbdev 
33d86ed7fbStbbdev     THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
34d86ed7fbStbbdev     OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
35d86ed7fbStbbdev     WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
36d86ed7fbStbbdev     ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
37d86ed7fbStbbdev     DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
38d86ed7fbStbbdev     DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
39d86ed7fbStbbdev     OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
40d86ed7fbStbbdev     HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
41d86ed7fbStbbdev     LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
42d86ed7fbStbbdev     OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43d86ed7fbStbbdev     SUCH DAMAGE.
44d86ed7fbStbbdev */
45d86ed7fbStbbdev 
46d86ed7fbStbbdev #include "machine.hpp"
47d86ed7fbStbbdev #include "types.hpp"
48d86ed7fbStbbdev #include "macros.hpp"
49d86ed7fbStbbdev #include "vector.hpp"
50d86ed7fbStbbdev #include "tgafile.hpp"
51d86ed7fbStbbdev #include "trace.hpp"
52d86ed7fbStbbdev #include "light.hpp"
53d86ed7fbStbbdev #include "shade.hpp"
54d86ed7fbStbbdev #include "camera.hpp"
55d86ed7fbStbbdev #include "util.hpp"
56d86ed7fbStbbdev #include "intersect.hpp"
57d86ed7fbStbbdev #include "global.hpp"
58d86ed7fbStbbdev #include "ui.hpp"
59d86ed7fbStbbdev #include "tachyon_video.hpp"
60d86ed7fbStbbdev 
61d86ed7fbStbbdev // shared but read-only so could be private too
62d86ed7fbStbbdev static thr_parms *all_parms;
63d86ed7fbStbbdev static scenedef scene;
64d86ed7fbStbbdev static int startx;
65d86ed7fbStbbdev static int stopx;
66d86ed7fbStbbdev static int starty;
67d86ed7fbStbbdev static int stopy;
68d86ed7fbStbbdev static flt jitterscale;
69d86ed7fbStbbdev static int totaly;
70d86ed7fbStbbdev 
render_one_pixel(int x,int y,unsigned int * local_mbox,unsigned int & serial,int startx,int stopx,int starty,int stopy)71d86ed7fbStbbdev static color_t render_one_pixel(int x,
72d86ed7fbStbbdev                                 int y,
73d86ed7fbStbbdev                                 unsigned int *local_mbox,
74d86ed7fbStbbdev                                 unsigned int &serial,
75d86ed7fbStbbdev                                 int startx,
76d86ed7fbStbbdev                                 int stopx,
77d86ed7fbStbbdev                                 int starty,
78d86ed7fbStbbdev                                 int stopy) {
79d86ed7fbStbbdev     /* private vars moved inside loop */
80d86ed7fbStbbdev     ray primary;
81d86ed7fbStbbdev     color col;
82d86ed7fbStbbdev     int R, G, B;
83d86ed7fbStbbdev     intersectstruct local_intersections;
84d86ed7fbStbbdev     /* end private */
85d86ed7fbStbbdev 
86d86ed7fbStbbdev     primary = camray(&scene, x, y);
87d86ed7fbStbbdev     primary.intstruct = &local_intersections;
88d86ed7fbStbbdev     primary.flags = RT_RAY_REGULAR;
89d86ed7fbStbbdev 
90d86ed7fbStbbdev     serial++;
91d86ed7fbStbbdev     primary.serial = serial;
92d86ed7fbStbbdev     primary.mbox = local_mbox;
93d86ed7fbStbbdev     primary.maxdist = FHUGE;
94d86ed7fbStbbdev     primary.scene = &scene;
95d86ed7fbStbbdev     col = trace(&primary);
96d86ed7fbStbbdev     serial = primary.serial;
97d86ed7fbStbbdev 
98d86ed7fbStbbdev     /* Handle overexposure and underexposure here... */
99d86ed7fbStbbdev     R = (int)(col.r * 255);
100d86ed7fbStbbdev     if (R > 255)
101d86ed7fbStbbdev         R = 255;
102d86ed7fbStbbdev     else if (R < 0)
103d86ed7fbStbbdev         R = 0;
104d86ed7fbStbbdev 
105d86ed7fbStbbdev     G = (int)(col.g * 255);
106d86ed7fbStbbdev     if (G > 255)
107d86ed7fbStbbdev         G = 255;
108d86ed7fbStbbdev     else if (G < 0)
109d86ed7fbStbbdev         G = 0;
110d86ed7fbStbbdev 
111d86ed7fbStbbdev     B = (int)(col.b * 255);
112d86ed7fbStbbdev     if (B > 255)
113d86ed7fbStbbdev         B = 255;
114d86ed7fbStbbdev     else if (B < 0)
115d86ed7fbStbbdev         B = 0;
116d86ed7fbStbbdev 
117d86ed7fbStbbdev     return video->get_color(R, G, B);
118d86ed7fbStbbdev }
119d86ed7fbStbbdev 
120d86ed7fbStbbdev #if DO_ITT_NOTIFY
121d86ed7fbStbbdev #include "ittnotify.h"
122d86ed7fbStbbdev #endif
123d86ed7fbStbbdev 
124d86ed7fbStbbdev #define RUNTIME_SERIAL 1
125d86ed7fbStbbdev #define RUNTIME_OPENMP 2
126d86ed7fbStbbdev #define RUNTIME_TBB    3
127d86ed7fbStbbdev 
128d86ed7fbStbbdev #ifndef RUNTIME
129d86ed7fbStbbdev #define RUNTIME RUNTIME_TBB
130d86ed7fbStbbdev #endif
131d86ed7fbStbbdev 
132d86ed7fbStbbdev #if RUNTIME == RUNTIME_OPENMP
133d86ed7fbStbbdev #include <omp.h>
134d86ed7fbStbbdev #elif RUNTIME == RUNTIME_TBB
135d86ed7fbStbbdev #include "oneapi/tbb.h"
136d86ed7fbStbbdev #endif
137d86ed7fbStbbdev 
parallel_thread(void)138d86ed7fbStbbdev static void parallel_thread(void) {
139d86ed7fbStbbdev     unsigned int mboxsize = sizeof(unsigned int) * (max_objectid() + 20);
140d86ed7fbStbbdev #if RUNTIME == RUNTIME_SERIAL
141d86ed7fbStbbdev     for (int y = starty; y < stopy; y++)
142d86ed7fbStbbdev #elif RUNTIME == RUNTIME_OPENMP
143d86ed7fbStbbdev #pragma omp parallel for
144d86ed7fbStbbdev     for (int y = starty; y < stopy; y++)
145d86ed7fbStbbdev #elif RUNTIME == RUNTIME_TBB
146d86ed7fbStbbdev     oneapi::tbb::parallel_for(starty, stopy, [mboxsize] (int y)
147d86ed7fbStbbdev #endif
148d86ed7fbStbbdev     {
149d86ed7fbStbbdev         unsigned int serial = 1;
150d86ed7fbStbbdev         unsigned int local_mbox[mboxsize];
151d86ed7fbStbbdev         memset(local_mbox, 0, mboxsize);
152d86ed7fbStbbdev         drawing_area drawing(startx, totaly - y, stopx - startx, 1);
153d86ed7fbStbbdev         for (int x = startx; x < stopx; x++) {
154d86ed7fbStbbdev             color_t c = render_one_pixel(x, y, local_mbox, serial, startx, stopx, starty, stopy);
155d86ed7fbStbbdev             drawing.put_pixel(c);
156d86ed7fbStbbdev         }
157d86ed7fbStbbdev         video->next_frame();
158d86ed7fbStbbdev     }
159d86ed7fbStbbdev #if RUNTIME == RUNTIME_TBB
160d86ed7fbStbbdev     );
161d86ed7fbStbbdev #endif
162d86ed7fbStbbdev }
163d86ed7fbStbbdev 
thread_trace(thr_parms * parms)164d86ed7fbStbbdev void *thread_trace(thr_parms *parms) {
165d86ed7fbStbbdev     // shared but read-only so could be private too
166d86ed7fbStbbdev     all_parms = parms;
167d86ed7fbStbbdev     scene = parms->scene;
168d86ed7fbStbbdev     startx = parms->startx;
169d86ed7fbStbbdev     stopx = parms->stopx;
170d86ed7fbStbbdev     starty = parms->starty;
171d86ed7fbStbbdev     stopy = parms->stopy;
172d86ed7fbStbbdev     jitterscale = 40.0 * (scene.hres + scene.vres);
173d86ed7fbStbbdev     totaly = parms->scene.vres - 1;
174d86ed7fbStbbdev 
175d86ed7fbStbbdev #if DO_ITT_NOTIFY
176d86ed7fbStbbdev     __itt_resume();
177d86ed7fbStbbdev #endif
178d86ed7fbStbbdev     parallel_thread();
179d86ed7fbStbbdev #if DO_ITT_NOTIFY
180d86ed7fbStbbdev     __itt_pause();
181d86ed7fbStbbdev #endif
182d86ed7fbStbbdev 
183d86ed7fbStbbdev     return (nullptr);
184d86ed7fbStbbdev }
185