Project 3: Shenyue Chen

.gitignore

@@ -23,7 +23,8 @@ build
 .LSOverride
 
 # Icon must end with two \r
-Icon
+Icon
+
 
 # Thumbnails
 ._*
@@ -189,7 +190,6 @@ install_manifest.txt
 *.slo
 *.lo
 *.o
-*.obj
 
 # Precompiled Headers
 *.gch
@@ -276,7 +276,6 @@ artifacts/
 *_i.h
 *.ilk
 *.meta
-*.obj
 *.pch
 *.pdb
 *.pgc

.vscode/settings.json

@@ -0,0 +1,77 @@
+{
+  "files.associations": {
+    "*.js ": "javascript",
+    "string": "cpp",
+    "xstring": "cpp",
+    "algorithm": "cpp",
+    "atomic": "cpp",
+    "cctype": "cpp",
+    "chrono": "cpp",
+    "cmath": "cpp",
+    "compare": "cpp",
+    "concepts": "cpp",
+    "cstdarg": "cpp",
+    "cstddef": "cpp",
+    "cstdint": "cpp",
+    "cstdio": "cpp",
+    "cstdlib": "cpp",
+    "cstring": "cpp",
+    "ctime": "cpp",
+    "cwchar": "cpp",
+    "exception": "cpp",
+    "fstream": "cpp",
+    "initializer_list": "cpp",
+    "iomanip": "cpp",
+    "ios": "cpp",
+    "iosfwd": "cpp",
+    "iostream": "cpp",
+    "istream": "cpp",
+    "iterator": "cpp",
+    "limits": "cpp",
+    "locale": "cpp",
+    "map": "cpp",
+    "memory": "cpp",
+    "new": "cpp",
+    "ostream": "cpp",
+    "ratio": "cpp",
+    "set": "cpp",
+    "sstream": "cpp",
+    "stdexcept": "cpp",
+    "streambuf": "cpp",
+    "system_error": "cpp",
+    "tuple": "cpp",
+    "type_traits": "cpp",
+    "typeinfo": "cpp",
+    "utility": "cpp",
+    "vector": "cpp",
+    "xfacet": "cpp",
+    "xiosbase": "cpp",
+    "xlocale": "cpp",
+    "xlocbuf": "cpp",
+    "xlocinfo": "cpp",
+    "xlocmes": "cpp",
+    "xlocmon": "cpp",
+    "xlocnum": "cpp",
+    "xloctime": "cpp",
+    "xmemory": "cpp",
+    "xstddef": "cpp",
+    "xtr1common": "cpp",
+    "xtree": "cpp",
+    "xutility": "cpp",
+    "array": "cpp",
+    "bit": "cpp",
+    "*.tcc": "cpp",
+    "clocale": "cpp",
+    "complex": "cpp",
+    "cwctype": "cpp",
+    "deque": "cpp",
+    "unordered_map": "cpp",
+    "functional": "cpp",
+    "memory_resource": "cpp",
+    "numeric": "cpp",
+    "optional": "cpp",
+    "random": "cpp",
+    "string_view": "cpp",
+    "cinttypes": "cpp"
+  }
+}

CMakeLists.txt

@@ -73,6 +73,7 @@ set(headers
     src/sceneStructs.h
     src/preview.h
     src/utilities.h
+    src/tinyobj.h
     )
 
 set(sources
@@ -84,6 +85,7 @@ set(sources
     src/scene.cpp
     src/preview.cpp
     src/utilities.cpp
+    src/tinyobj.cc
     )
 
 list(SORT headers)

README.md

@@ -3,11 +3,61 @@ CUDA Path Tracer
 
 **University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 3**
 
-* (TODO) YOUR NAME HERE
-* Tested on: (TODO) Windows 22, i7-2222 @ 2.22GHz 22GB, GTX 222 222MB (Moore 2222 Lab)
+* Shenyue Chen
+  * [LinkedIn](https://www.linkedin.com/in/shenyue-chen-5b2728119/), [personal website](http://github.com/EvsChen)
+* Tested on: Windows 10, Intel Xeon Platinum 8259CL @ 2.50GHz 16GB, Tesla T4 (AWS g4dn-xlarge)
 
-### (TODO: Your README)
+### Output
+<p align="center">
+<img src="./img/custom_scene.png" />
+<span>Custom OBJ file loading</span>
+<img src="./img/materials.png" />
+<span>Scenes used for materials sorting test</span>
+<img src="./img/two-balls.png" />
+<span>Two balls</span>
+<img src="./img/two-balls-dof.png" />
+<span>Two balls with depth of field</span>
+<img src="./img/motion-blur.png" />
+<span>Motion blur</span>
+</p>
 
-*DO NOT* leave the README to the last minute! It is a crucial part of the
-project, and we will not be able to grade you without a good README.
+### Features
+Part 1
+* Specular and diffuse shading
+* Stream compaction to terminate rays
+* Materials sorting
+* First bounce cache
 
+Part 2
+* .OBJ file loading using [tinyobj](https://github.com/tinyobjloader/tinyobjloader)
+* Motion blur
+* Refraction and Schlick approximation
+
+Part 3
+* Depth of field
+* Anti-aliasing
+* [Small script](./scenes/gen_scene.py) for scene generation
+
+### Performance analysis
+#### Stream compaction for path termination
+Below is the change of paths in the first 10 iterations of the cornell box image, with the basic diffuse and specular shading.
+<p align="center">
+<img src="./img/path-termination.png" />
+</p>
+From the plot we can observe that the number of paths reduces greatly when the depth increases, because many rays terminated by not intersecting or intersecting with a light source. The number of paths is equal to the number of threads we launched in each kernel run. Therefore we increase the thread utilization by reducing the number of threads in kernel launch configuration.
+
+
+#### Different optimization techniques
+In this section, we measured the average time per iteration for the basic cornell box scene with different optimization techniques, for the first 1000 iterations. The result can be seen in the plot below.
+<p align="center">
+<img src="./img/diff_opti.png" />
+</p>
+We notice the the first bounce cache can improve the performance by a little bit. But the other parts seem to play a negative role. I assume this is because the scene is too simple and the sorting overhead is much over than the optimization is causes.
+
+
+#### Material sorting testing
+Specifically, I want to see whether more materials will affect the effect of optimization caused by the material sorting. So I created a scene of more than 30 materials and 30 objects, [materials.txt](./scenes/materials.txt). The rendered scene is on the top of the readme. And the time measures is shown below. 
+<p align="center">
+<img src="./img/mat_sort_perf.png" />
+</p>
+Unfortunately, the sorting still decreases the performance. But the overhead is not so serious compared to the simple scene.

analysis.py

@@ -0,0 +1,65 @@
+import csv
+import matplotlib.pyplot as plt
+
+# data = []
+# with open('path-termination.csv', 'r') as csvfile:
+#   spamreader = csv.reader(csvfile, delimiter=',')
+#   for row in spamreader:
+#     drow = []
+#     for data_str in row:
+#       drow.append(int(data_str))
+#     data.append(drow)
+
+# fig, ax = plt.subplots()
+
+# for i in range(0, len(data)):
+#   row = data[i]
+#   x = [j for j in range(0, len(row))]
+#   plt.plot(x, row)
+
+# plt.title("Change of paths in first 10 iterations")
+# plt.xlabel("Depth")
+# plt.ylabel("Path count")
+# plt.show()
+
+# base = [13.9153, 13.4523, 13.2604, 13.0256, 12.8251, 12.6814, 12.584, 12.467, 12.3808,
+#         12.3539, 12.3, 12.2559, 12.2064, 12.1793, 12.157, 12.1449, 12.1508, 12.1537, 12.147, 12.152]
+# first_cache = [13.994, 13.5278, 12.9571, 12.4803, 12.1568, 11.9892, 11.8137, 11.7125, 11.6294,
+#                11.5688, 11.602, 11.5473, 11.5048, 11.4699, 11.4974, 11.476, 11.4706, 11.4416, 11.4178, 11.411]
+# ray_term = [43.3666, 39.4445, 37.9706, 37.2061, 36.8238, 36.464, 36.3297, 36.1474, 36.0439,
+#             35.9822, 35.937, 35.907, 35.8626, 35.8245, 35.7628, 35.7196, 35.6834, 35.6314, 35.5985, 35.5634]
+# mat_sorting = [92.1103, 90.0405, 89.2113, 88.785, 88.5098, 88.3835, 88.2686, 88.1703, 88.1409,
+#                88.0665, 88.0275, 88.0738, 88.1269, 88.1731, 88.2425, 88.2707, 88.309, 88.3622, 88.4123, 88.4398]
+
+# x = [i * 50 for i in range(1, 21)]
+
+# fig, ax = plt.subplots()
+
+# plt.plot(x, base, label="base")
+# plt.plot(x, first_cache, label="first bounce cache")
+# plt.plot(x, ray_term, label="stream compaction")
+# plt.plot(x, mat_sorting, label="material sorting")
+
+# plt.title("Average time per iteration for different optimizations")
+# plt.xlabel("First # of iterations")
+# plt.ylabel("Average time (ms)")
+# plt.legend()
+# plt.show()
+
+mat_scene_with_sort = [77.1336, 75.3321, 74.7839, 74.586, 74.8918, 75.1525, 74.7427, 74.6345, 74.6205,
+                       74.6913, 74.7094, 74.7804, 74.789, 74.7278, 74.6984, 74.7119, 74.7549, 74.8404, 74.8833, 74.86]
+
+mat_scene = [42.1215, 40.9768, 40.5953, 40.4929, 40.7369, 40.6872, 40.7313, 40.7851, 40.8242, 40.67, 40.2865, 39.918, 39.6168, 39.3591, 39.1535, 38.9644, 38.8111, 38.6765, 38.569, 38.458]
+
+x = [i * 50 for i in range(1, 21)]
+
+fig, ax = plt.subplots()
+
+plt.plot(x, mat_scene, label="base")
+plt.plot(x, mat_scene_with_sort, label="with material sorting")
+
+plt.title("Average time per iteration for material sorting")
+plt.xlabel("First # of iterations")
+plt.ylabel("Average time (ms)")
+plt.legend()
+plt.show()

path-termination.csv

@@ -0,0 +1,10 @@
+640000,523159,362728,280463,223214,180495,146740,119710
+640000,523159,362288,280306,223398,180778,147049,120156
+640000,523159,362267,279749,223067,180340,146656,119784
+640000,523159,362483,280166,223311,180560,146641,119595
+640000,523159,362403,280050,223341,180564,146528,119454
+640000,523159,362148,280058,222828,180016,146613,119768
+640000,523159,362432,279908,222710,179843,146030,119223
+640000,523159,362638,279881,223265,180436,146775,119600
+640000,523159,362086,279934,223102,180527,147037,120059
+640000,523159,362323,279620,222546,179910,146255,119561