CIS565-Fall-2016 · dengxianga · Oct 23, 2016 · Oct 23, 2016 · Oct 23, 2016 · Oct 23, 2016
diff --git a/README.md b/README.md
@@ -1,18 +1,106 @@
 CUDA Rasterizer
 ===============
 
-[CLICK ME FOR INSTRUCTION OF THIS PROJECT](./INSTRUCTION.md)
+**University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 4** 
 
-**University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 4**
+* Xiang Deng
+* Tested on:  Windows 10-Home, i7-6700U @ 2.6GHz 16GB, GTX 1060 6GB (Personal Computer)
 
-* (TODO) YOUR NAME HERE
-* Tested on: (TODO) Windows 22, i7-2222 @ 2.22GHz 22GB, GTX 222 222MB (Moore 2222 Lab)
+* Overview
+In this project we use CUDA to implement the raterizerized graphics pipeline. Major features
+include vertex shading, primitive assembly, raterization and fragment shading. Here is a list of features:
 
-### (TODO: Your README)
 
-*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:**
 
+* Basic pipeline
+  * Vertex assembly and vertex shader
+  * Primitive assembly
+  * Rasterization
+  * Depth test
+  * Race avoidance with atomic
+  * Fragment shader (with lambert lighting)
+  * Fragment-to-depth-buffer writing (with atomics for race avoidance).
+  * A depth buffer for storing and depth testing fragments. 
+  * Interpolation near primitive points
+  * Toon Shading
+* UV texture mapping with bilinear texture filtering and perspective correct texture coordinates
+* Support for rasterizing additional primitives: lines and points
+
+Milk truck | Duck
+:-------------------------:|:-------------------------: 
+![](imgs/milk1.gif) | ![](imgs/duck1.gif) 
+
+Demo of basic raterization pipeline, we used lambert lighting for the fragement shader.
+
+VC
+![](imgs/VC1.gif)
+
+Cow with Lines | VC with Lines |Truck with Lines 
+:-------------------------:|:-------------------------:  |:-------------------------:
+![](imgs/cow2.gif) | ![](imgs/VC2.gif) |![](imgs/truck2.gif)
+
+Here is a demo of rasterizing lines.
+
+
+
+Duck with Points (dense) | Duck with Points (sparse) |Truck with Points (sparse)
+:-------------------------:|:-------------------------: |:-------------------------: 
+![](imgs/duck3.gif) | ![](imgs/duck4.gif) | ![](imgs/truck3.gif)
+
+Here is a demo of rasterizing points;
+Bonus: I am not just rendering the primitive points, it's not beautiful. Sparse points are interpolated near the primitive points, but the
+user is allowed to control the sparsity of points though on the top of raterize.cu.. 
+
+Checkerboard with perspective correction | Checkerboard without perspective correction
+:-------------------------:|:-------------------------: 
+![](imgs/checkerboard.gif) | ![](imgs/checkerboard2.gif)
+
+Perspective correction: please see the reference https://en.wikipedia.org/wiki/Texture_mapping#Perspective_correctness. Without perspective correction,
+we can observe the 'distorted' image generated from the raterization pipeline.
+
+Checkerboard with bilinear filtering | Checkerboard without bilinear filtering
+:-------------------------:|:-------------------------: 
+![](imgs/checkwithbin.JPG) | ![](imgs/checkwithnobin.JPG)
+
+Bilinear filtering, reference https://en.wikipedia.org/wiki/Bilinear_filtering, we can observe the smoothing effect on the lines over the checkerboard.
+
+Duck toon | Di toon
+:-------------------------:|:-------------------------: 
+![](imgs/duck5.gif) | ![](imgs/di2.gif)
+
+Toon Shading, https://www.garagegames.com/community/forums/viewthread/24977
+
+Cow | Di|Engine | Buggy| Flower with Lines
+:-------------------------:|:-------------------------: |:-------------------------:|:-------------------------: |:-------------------------: 
+![](imgs/cow1.gif) | ![](imgs/di1.gif) |![](imgs/engine1.gif) | ![](imgs/buggy1.gif)  | ![](imgs/flower.gif)
+
+# Analysis
+
+Pipeline Timing(ms) - VC | Pipeline Timing(ms) - Cow |Pipeline Timing(ms) - Box
+:-------------------------:|:-------------------------:  |:-------------------------:
+![](imgs/ana1.JPG) | ![](imgs/ana2.JPG) |![](imgs/ana3.JPG)
+![](imgs/ana4.JPG) | ![](imgs/ana5.JPG) |![](imgs/ana6.JPG)
+triangle width-mean :20 |  triangle width-mean :2~3 |triangle width-mean :77
+
+Charts above shows that the number of primitives is not the dominant factor effecting the breakdown of time spent in each pipeline stage. 
+Indeed, the stage that consumes most of the time is rasterizing; however, it's not proportional to the number of primitives.
+In cases of cow, we have >5000 primitives in total, the rasterizing stage only consumes 36 percent of the total time;
+on the contrast, the box only has 12 primitives, the rasterizing consumes more than 93 percent of the total time. 
+A more careful examination of the rasterizing stage reveals another important factor in the rasterizing stage: the primitive size.
+As for each kernel we have to iterate through the entire boundary box to compute each fragments within a triangle, increasing the average triangle size 
+might significantly decrease the performance. If we can saturate the threads as much as possible while decrease the average size of the triangle, we
+can expect an improvment in the performance.
+
+On the other hand, in case our average triangle is small enough, we might need to concern about memory allocation and access. We
+could for example, use shared memomry to decrease the overhead of memomry allocation and access for initializing the depth buffer as well as the memory access in other kernel functions (we can find more evidence 
+as we observe the overhead of primitive assembly, rending and depth init in the left two charts).
+It would be cool to implement tile based rasterizing in the future.
+
+*PS: looking for models: just we & just tank : ]
+
+*CMakeLists change
+ 'common.h' file is added to the cmakelist file; sm20 --> sm61.
 
 ### Credits
 

diff --git a/imgs/VC1.gif b/imgs/VC1.gif
diff --git a/imgs/VC2.gif b/imgs/VC2.gif
diff --git a/imgs/ana1.JPG b/imgs/ana1.JPG
diff --git a/imgs/ana2.JPG b/imgs/ana2.JPG
diff --git a/imgs/ana3.JPG b/imgs/ana3.JPG
diff --git a/imgs/ana4.JPG b/imgs/ana4.JPG
diff --git a/imgs/ana5.JPG b/imgs/ana5.JPG
diff --git a/imgs/ana6.JPG b/imgs/ana6.JPG
diff --git a/imgs/buggy1.gif b/imgs/buggy1.gif
diff --git a/imgs/checkerboard.gif b/imgs/checkerboard.gif
diff --git a/imgs/checkerboard2.gif b/imgs/checkerboard2.gif
diff --git a/imgs/checkwithbin.JPG b/imgs/checkwithbin.JPG
diff --git a/imgs/checkwithnobin.JPG b/imgs/checkwithnobin.JPG
diff --git a/imgs/cow1.gif b/imgs/cow1.gif
diff --git a/imgs/cow2.gif b/imgs/cow2.gif
diff --git a/imgs/cow3.gif b/imgs/cow3.gif
diff --git a/imgs/di1.gif b/imgs/di1.gif
diff --git a/imgs/di2.gif b/imgs/di2.gif
diff --git a/imgs/duck1.gif b/imgs/duck1.gif
diff --git a/imgs/duck3.gif b/imgs/duck3.gif
diff --git a/imgs/duck4.gif b/imgs/duck4.gif
diff --git a/imgs/duck5.gif b/imgs/duck5.gif
diff --git a/imgs/engine1.gif b/imgs/engine1.gif
diff --git a/imgs/flower.gif b/imgs/flower.gif
diff --git a/imgs/milk1.gif b/imgs/milk1.gif
diff --git a/imgs/truck2.gif b/imgs/truck2.gif
diff --git a/imgs/truck3.gif b/imgs/truck3.gif
diff --git a/imgs/truckwithbin.JPG b/imgs/truckwithbin.JPG
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
@@ -1,10 +1,11 @@
 set(SOURCE_FILES
     "rasterize.cu"
+	"common.h"
     "rasterize.h"
     "rasterizeTools.h"
     )
 
 cuda_add_library(src
     ${SOURCE_FILES}
-    OPTIONS -arch=sm_20
+    OPTIONS -arch=sm_61
     )
diff --git a/src/common.h b/src/common.h
@@ -0,0 +1,126 @@
+#include <cmath>
+#include <cstdio>
+#include <cuda.h>
+#include <cuda_runtime.h>
+#include <thrust/random.h>
+#include <util/checkCUDAError.h>
+#include <util/tiny_gltf_loader.h>
+#include "rasterizeTools.h" 
+#include <glm/gtc/quaternion.hpp>
+#include <glm/gtc/matrix_transform.hpp>
+#include "util/utilityCore.hpp"
+namespace {
+
+	typedef unsigned short VertexIndex;
+	typedef glm::vec3 VertexAttributePosition;
+	typedef glm::vec3 VertexAttributeNormal;
+	typedef glm::vec2 VertexAttributeTexcoord;
+	typedef unsigned char TextureData;
+
+	typedef unsigned char BufferByte;
+
+	enum PrimitiveType{
+		Point = 1,
+		Line = 2,
+		Triangle = 3
+	};
+
+	struct VertexOut {
+		glm::vec4 pos;
+
+		// TODO: add new attributes to your VertexOut
+		// The attributes listed below might be useful, 
+		// but always feel free to modify on your own
+
+		glm::vec3 eyePos;	// eye space position used for shading
+		glm::vec3 eyeNor;	// eye space normal used for shading, cuz normal will go wrong after perspective transformation
+		glm::vec3 col;
+		glm::vec2 texcoord0;
+		TextureData* dev_diffuseTex = NULL;
+		int texWidth, texHeight;
+		int texture;
+		// ...
+	};
+
+	struct Primitive {
+		PrimitiveType primitiveType = Triangle;	// C++ 11 init
+		VertexOut v[3];
+
+		TextureData* dev_diffuseTex;
+		int texWidth, texHeight;
+		int texture;
+	};
+
+	struct Fragment {
+		glm::vec3 color;
+
+		// TODO: add new attributes to your Fragment
+		// The attributes listed below might be useful, 
+		// but always feel free to modify on your own
+
+		glm::vec3 eyePos;	// eye space position used for shading
+		glm::vec3 eyeNor;
+		VertexAttributeTexcoord texcoord0;
+		TextureData* dev_diffuseTex;
+		int texWidth, texHeight;
+		int texture;
+		// ...
+	};
+
+	struct PrimitiveDevBufPointers {
+		int primitiveMode;	//from tinygltfloader macro
+		PrimitiveType primitiveType;
+		int numPrimitives;
+		int numIndices;
+		int numVertices;
+
+		// Vertex In, const after loaded
+		VertexIndex* dev_indices;
+		VertexAttributePosition* dev_position;
+		VertexAttributeNormal* dev_normal;
+		VertexAttributeTexcoord* dev_texcoord0;
+
+		// Materials, add more attributes when needed
+
+		// TODO: add more attributes when needed
+		TextureData* dev_diffuseTex;
+		int texWidth, texHeight;
+		int texture;
+
+		// Vertex Out, vertex used for rasterization, this is changing every frame
+		VertexOut* dev_verticesOut;
+	};
+
+}
+
+static std::map<std::string, std::vector<PrimitiveDevBufPointers>> mesh2PrimitivesMap;
+
+
+static int width = 0;
+static int height = 0;
+
+static int totalNumPrimitives = 0;
+static Primitive *dev_primitives = NULL;
+static Fragment *dev_fragmentBuffer = NULL;
+static glm::vec3 *dev_framebuffer = NULL;
+
+static int * dev_depth = NULL;	// you might need this buffer when doing depth test
+#define blockSize 256
+
+//TILEs::::::::::::::::::::::
+
+#define tileSizeR2 256
+static int tileWidth;
+static int tileHeight;
+static int numTilesV;
+static int numTilesH;
+
+struct Tile{
+	float maxx;
+	float maxy;
+	float minx;
+	float miny;
+	int numPrims = 0;
+};
+//TILEs______________________
+static Tile * dev_tiles;