Project 2: Yan Wu

README.md

@@ -3,12 +3,53 @@ CUDA Stream Compaction
 
 **University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 2**
 
-* (TODO) YOUR NAME HERE
-  * (TODO) [LinkedIn](), [personal website](), [twitter](), etc.
-* Tested on: (TODO) Windows 22, i7-2222 @ 2.22GHz 22GB, GTX 222 222MB (Moore 2222 Lab)
+* Yan Wu
+  * [LinkedIn](https://www.linkedin.com/in/yan-wu-a71270159/)
+* Tested on: Windows 10 Education, i7-8750H @ 2.2GHz 16GB, GTX 1060 6GB (Personal Laptop)
 
-### (TODO: Your README)
+### Project Description
 
-Include analysis, etc. (Remember, this is public, so don't put
-anything here that you don't want to share with the world.)
+This project let us implement a few different versions of the Scan (Prefix Sum) algorithm. 
+1. Implementing a CPU version of the algorithm. 
+2. Writing a few GPU implementations: "naive" and "work-efficient." 
+3. Using some of these to implement GPU stream compaction.
+
+* [Algorithm Slides:](https://docs.google.com/presentation/d/1ETVONA7QDM-WqsEj4qVOGD6Kura5I6E9yqH-7krnwZ0/edit#slide=id.p126)
+* Algorithm Demonstration:
+  * Naive Inclusive Scan:<br />
+  <img src="img/figure-39-2.jpg" width="70%"> <br />
+  * Work Efficient Scan (up-sweep): <br />
+  <img src="img/upSweep.PNG" width="70%"> <br />
+  * Work Efficient Scan (down-sweep): <br />
+  <img src="img/downSweep.PNG" width="70%">
+
+### Result and Performance Analysis
+
+* A sample of outcomes:
+  * Scan Test: <br />
+  <img src="img/scanTestsSample.PNG" width = "70%"> <br />
+  * Stream Compaction Test: <br />
+    <img src="img/CompactTestsSample.PNG" width = "70%">
+
+* Result Sheets:
+  * Scan Test Result:<br />
+  <img src="img/scanSheet.PNG" width = "90%"> <br />
+  * Compaction Test Result:<br />
+    <img src="img/compactSheet.PNG" width = "90%">
+
+* Analysis:
+  * Scan Test Charts:<br />
+  <img src="img/scanChart.PNG" width = "90%"> <br />
+  * Compaction Test Charts:<br />
+    <img src="img/compactChart.PNG" width = "90%"><br />
+    Just to mention, the two red lines regarding "cpu compact with scan" in both above charts are the same line.
+  * Analysis:<br />
+  First we can see that GPU isn't performing significantly better when the test array is short. But when array becomes very large, it also becomes a burden for the CPU and the executing time increases almost exponentially. Under this circumstance we should perceive a much greater performance outcome by GPU.<br />
+  Taking a closer look, my work efficient algorithm is slower than the naive method when the test array is relatively shorter. As array size increases, work efficient method has a trend of better performance. Thrust method is steady. It may seems slow at first, but when array size increases, its time remains almost the same. Comparing all these methods, thrust scan is clearly the best when array is incredibly large.
+
+* Q & A:
+  * Roughly optimize the block sizes of each of your implementations for minimal run time on your GPU.<br />
+  My block size was 128.
+  * Why does work efficien method slower than naive?<br />
+  When implementing algorithm from the course slides, we did reduced the calculation count since we only need to compute half of the elements involved each round. Although we did this, half of each warp didn't even work. Implementing a parallel reduction and reduce the working warp number could save a lot time on this issue.
 

src/main.cpp

@@ -13,7 +13,7 @@
 #include <stream_compaction/thrust.h>
 #include "testing_helpers.hpp"
 
-const int SIZE = 1 << 8; // feel free to change the size of array
+const int SIZE = 1 << 24; // feel free to change the size of array
 const int NPOT = SIZE - 3; // Non-Power-Of-Two
 int *a = new int[SIZE];
 int *b = new int[SIZE];
@@ -126,12 +126,13 @@ int main(int argc, char* argv[]) {
     printArray(count, c, true);
     printCmpLenResult(count, expectedNPOT, b, c);
 
-    zeroArray(SIZE, c);
-    printDesc("cpu compact with scan");
-    count = StreamCompaction::CPU::compactWithScan(SIZE, c, a);
-    printElapsedTime(StreamCompaction::CPU::timer().getCpuElapsedTimeForPreviousOperation(), "(std::chrono Measured)");
-    printArray(count, c, true);
-    printCmpLenResult(count, expectedCount, b, c);
+	zeroArray(SIZE, c);
+	printDesc("cpu compact with scan");
+	count = StreamCompaction::CPU::compactWithScan(SIZE, c, a);
+	printElapsedTime(StreamCompaction::CPU::timer().getCpuElapsedTimeForPreviousOperation(), "(std::chrono Measured)");
+	//printArray(count, c, true);
+	expectedCount = count;
+	printCmpLenResult(count, expectedCount, b, c);
 
     zeroArray(SIZE, c);
     printDesc("work-efficient compact, power-of-two");

stream_compaction/CMakeLists.txt

@@ -13,5 +13,5 @@ set(SOURCE_FILES
 
 cuda_add_library(stream_compaction
     ${SOURCE_FILES}
-    OPTIONS -arch=sm_20
+    OPTIONS -arch=sm_61
     )

stream_compaction/common.cu

@@ -1,4 +1,5 @@
 #include "common.h"
+#include "device_launch_parameters.h"
 
 void checkCUDAErrorFn(const char *msg, const char *file, int line) {
     cudaError_t err = cudaGetLastError();
@@ -24,6 +25,9 @@ namespace StreamCompaction {
          */
         __global__ void kernMapToBoolean(int n, int *bools, const int *idata) {
             // TODO
+			int index = (blockIdx.x * blockDim.x) + threadIdx.x;
+			if (index >= n) return;
+			bools[index] = idata[index] == 0 ? 0 : 1;
         }
 
         /**
@@ -33,6 +37,11 @@ namespace StreamCompaction {
         __global__ void kernScatter(int n, int *odata,
                 const int *idata, const int *bools, const int *indices) {
             // TODO
+			int index = (blockIdx.x * blockDim.x) + threadIdx.x;
+			if (index >= n) return;
+			if (bools[index] == 1) {
+				odata[indices[index]] = idata[index];
+			}
         }
 
     }

stream_compaction/cpu.cu

@@ -1,15 +1,15 @@
 #include <cstdio>
 #include "cpu.h"
 
-#include "common.h"
+#include "common.h"
 
 namespace StreamCompaction {
     namespace CPU {
-        using StreamCompaction::Common::PerformanceTimer;
-        PerformanceTimer& timer()
-        {
-	        static PerformanceTimer timer;
-	        return timer;
+        using StreamCompaction::Common::PerformanceTimer;
+        PerformanceTimer& timer()
+        {
+	        static PerformanceTimer timer;
+	        return timer;
         }
 
         /**
@@ -18,9 +18,13 @@ namespace StreamCompaction {
          * (Optional) For better understanding before starting moving to GPU, you can simulate your GPU scan in this function first.
          */
         void scan(int n, int *odata, const int *idata) {
-	        timer().startCpuTimer();
-            // TODO
-	        timer().endCpuTimer();
+	        //timer().startCpuTimer();
+            // TODO	
+			odata[0] = 0;
+			for (int i = 1; i < n; ++i) {
+				odata[i] = odata[i - 1] + idata[i - 1];
+			}
+	        //timer().endCpuTimer();
         }
 
         /**
@@ -31,8 +35,14 @@ namespace StreamCompaction {
         int compactWithoutScan(int n, int *odata, const int *idata) {
 	        timer().startCpuTimer();
             // TODO
+			int index = 0;
+			for (int i = 0; i < n; i++) {
+				if (idata[i] != 0) {
+					odata[index++] = idata[i];
+				}
+			}
 	        timer().endCpuTimer();
-            return -1;
+            return index;
         }
 
         /**
@@ -41,10 +51,24 @@ namespace StreamCompaction {
          * @returns the number of elements remaining after compaction.
          */
         int compactWithScan(int n, int *odata, const int *idata) {
-	        timer().startCpuTimer();
+			// allocate space for two middle status arrays
+			int *temp1 = (int*)malloc(n * sizeof(int));
+			int *temp2 = (int*)malloc(n * sizeof(int));
+			timer().startCpuTimer();
 	        // TODO
+			for (int i = 0; i < n; ++i) {
+				temp1[i] = idata[i] == 0 ? 0 : 1;
+			}
+			scan(n, temp2, temp1);
+			int index = 0;
+			for (int i = 0; i < n; ++i) {
+				if (temp1[i] == 1) {
+					odata[temp2[i]] = idata[i];
+					index++;
+				}
+			}
 	        timer().endCpuTimer();
-            return -1;
+            return index;
         }
     }
 }

stream_compaction/efficient.cu

@@ -2,23 +2,62 @@
 #include <cuda_runtime.h>
 #include "common.h"
 #include "efficient.h"
+#include "device_launch_parameters.h"
+#define blockSize 128
+
 
 namespace StreamCompaction {
     namespace Efficient {
-        using StreamCompaction::Common::PerformanceTimer;
-        PerformanceTimer& timer()
-        {
-            static PerformanceTimer timer;
-            return timer;
+        using StreamCompaction::Common::PerformanceTimer;
+        PerformanceTimer& timer()
+        {
+            static PerformanceTimer timer;
+            return timer;
         }
 
         /**
          * Performs prefix-sum (aka scan) on idata, storing the result into odata.
          */
+		__global__ void upSweep(int n, int k, int *dev) {
+			int index = (blockIdx.x * blockDim.x) + threadIdx.x;
+			if (index >= n) return;
+
+			if ((index % (2 * k) == 0) && (index + (2 * k) <= n))
+				dev[index + (2 * k) - 1] += dev[index + k - 1];
+		}
+
+		__global__ void downSweep(int n, int k, int *idata) {
+			int index = (blockIdx.x * blockDim.x) + threadIdx.x;
+			if (index >= n) return;
+			// need to check boundary
+			if ((index % (2 * k) == 0) && (index + (2 * k) <= n)) {
+				int temp = idata[index + k - 1];
+				idata[index + k - 1] = idata[index + (2 * k) - 1];
+				idata[index + (2 * k) - 1] += temp;
+			}
+		}
+
         void scan(int n, int *odata, const int *idata) {
-            timer().startGpuTimer();
-            // TODO
+			int *exclusive;
+			int length = pow(2, ilog2ceil(n));
+			cudaMalloc((void**)&exclusive, length * sizeof(int));
+			cudaMemset(exclusive, 0, length * sizeof(int));
+			cudaMemcpy(exclusive, idata, n * sizeof(int), cudaMemcpyHostToDevice);
+			dim3 fullBlocksPerGrid((length + blockSize - 1) / blockSize);
+			timer().startGpuTimer();
+			// TODO
+			// up-sweep
+			for (int d = 1; d < length; d *= 2) {
+				upSweep<<< fullBlocksPerGrid, blockSize >>>(length, d, exclusive);
+			}
+			cudaMemset(exclusive + length - 1, 0, sizeof(int));
+			// down-sweep
+			for (int d = length / 2; d >= 1; d /= 2) {
+				downSweep<<< fullBlocksPerGrid, blockSize >>>(length, d, exclusive);
+			}
             timer().endGpuTimer();
+			cudaMemcpy(odata, exclusive, n * sizeof(int), cudaMemcpyDeviceToHost);
+			cudaFree(exclusive);
         }
 
         /**
@@ -31,10 +70,42 @@ namespace StreamCompaction {
          * @returns      The number of elements remaining after compaction.
          */
         int compact(int n, int *odata, const int *idata) {
-            timer().startGpuTimer();
+			int *bools;
+			int *indices;
+			int *i_aug;
+			int *o_aug;
+			int length = pow(2, ilog2ceil(n));
+			cudaMalloc((void**)&bools, length * sizeof(int));
+			cudaMalloc((void**)&indices, length * sizeof(int));
+			cudaMalloc((void**)&i_aug, n * sizeof(int));
+			cudaMalloc((void**)&o_aug, n * sizeof(int));
+
+			cudaMemset(bools, 0, length * sizeof(int));
+			cudaMemset(indices, 0, length * sizeof(int));
+
+			cudaMemcpy(i_aug, idata, n * sizeof(int), cudaMemcpyHostToDevice);
+
+			dim3 fullBlocksPerGrid((length + blockSize - 1) / blockSize);
+
+			//timer().startGpuTimer();
             // TODO
-            timer().endGpuTimer();
-            return -1;
+			StreamCompaction::Common::kernMapToBoolean<<< fullBlocksPerGrid, blockSize >>>(n, bools, i_aug);
+			scan(n, indices, bools);
+			StreamCompaction::Common::kernScatter <<< fullBlocksPerGrid, blockSize >>>(n, o_aug, i_aug, bools, indices);
+            //timer().endGpuTimer();
+			cudaMemcpy(odata, o_aug, n * sizeof(int), cudaMemcpyDeviceToHost);
+
+			int num1 = 0;
+			cudaMemcpy(&num1, &bools[n - 1], sizeof(int), cudaMemcpyDeviceToHost);
+			int num2 = 0;
+			cudaMemcpy(&num2, &indices[n - 1], sizeof(int), cudaMemcpyDeviceToHost);
+
+			cudaFree(bools);
+			cudaFree(indices);
+			cudaFree(i_aug);
+			cudaFree(o_aug);
+
+            return num1 + num2;
         }
     }
 }

stream_compaction/naive.cu

@@ -2,24 +2,62 @@
 #include <cuda_runtime.h>
 #include "common.h"
 #include "naive.h"
+#include "device_launch_parameters.h"
+
+#define blockSize 128
+
 
 namespace StreamCompaction {
     namespace Naive {
-        using StreamCompaction::Common::PerformanceTimer;
-        PerformanceTimer& timer()
-        {
-            static PerformanceTimer timer;
-            return timer;
+        using StreamCompaction::Common::PerformanceTimer;
+        PerformanceTimer& timer()
+        {
+            static PerformanceTimer timer;
+            return timer;
         }
         // TODO: __global__
-
+		__global__ void naiveParallelScan(int n, int k, int *odata, int *idata) {
+			int index = (blockIdx.x * blockDim.x) + threadIdx.x;
+			if (index >= n) return;
+			if (index >= k) {
+				odata[index] = idata[index] + idata[index - k];
+			}
+			else {
+				odata[index] = idata[index];
+			}
+		}
+		__global__ void toExclusive(int n, int *odata, int *idata) {
+			int index = (blockIdx.x * blockDim.x) + threadIdx.x;
+			if (index >= n) return;
+			if (index > 0) {
+				odata[index] = idata[index - 1];
+			}
+			else {
+				odata[index] = 0;
+			}
+		}
         /**
          * Performs prefix-sum (aka scan) on idata, storing the result into odata.
          */
         void scan(int n, int *odata, const int *idata) {
+			int* inclusive;
+			int* exclusive;
+			cudaMalloc((int**)&inclusive, n * sizeof(int));
+			cudaMalloc((int**)&exclusive, n * sizeof(int));
+			cudaMemcpy(inclusive, idata, n * sizeof(int), cudaMemcpyHostToDevice);
+			dim3 fullBlocksPerGrid((n + blockSize - 1) / blockSize);
             timer().startGpuTimer();
             // TODO
+			for (int d = 0; d <= ilog2ceil(n); d++) {
+				naiveParallelScan<<< fullBlocksPerGrid, blockSize >>>(n, pow(2.0, d), exclusive, inclusive);
+				// ping-pong
+				std::swap(exclusive, inclusive);
+			}
+			toExclusive<<< fullBlocksPerGrid, blockSize >>>(n, exclusive, inclusive);
             timer().endGpuTimer();
+			cudaMemcpy(odata, exclusive, n * sizeof(int), cudaMemcpyDeviceToHost);
+			cudaFree(inclusive);
+			cudaFree(exclusive);
         }
     }
 }