diff --git a/README.md b/README.md index a82ea0f..8c4f0c7 100644 --- a/README.md +++ b/README.md @@ -3,211 +3,88 @@ CUDA Stream Compaction **University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 2** -* (TODO) YOUR NAME HERE -* Tested on: (TODO) Windows 22, i7-2222 @ 2.22GHz 22GB, GTX 222 222MB (Moore 2222 Lab) - -### (TODO: Your README) - -Include analysis, etc. (Remember, this is public, so don't put -anything here that you don't want to share with the world.) - -Instructions (delete me) -======================== - -This is due Sunday, September 13 at midnight. - -**Summary:** In this project, you'll implement GPU stream compaction in CUDA, -from scratch. This algorithm is widely used, and will be important for -accelerating your path tracer project. - -Your stream compaction implementations in this project will simply remove `0`s -from an array of `int`s. In the path tracer, you will remove terminated paths -from an array of rays. - -In addition to being useful for your path tracer, this project is meant to -reorient your algorithmic thinking to the way of the GPU. On GPUs, many -algorithms can benefit from massive parallelism and, in particular, data -parallelism: executing the same code many times simultaneously with different -data. - -You'll implement a few different versions of the *Scan* (*Prefix Sum*) -algorithm. First, you'll implement a CPU version of the algorithm to reinforce -your understanding. Then, you'll write a few GPU implementations: "naive" and -"work-efficient." Finally, you'll use some of these to implement GPU stream -compaction. - -**Algorithm overview & details:** There are two primary references for details -on the implementation of scan and stream compaction. - -* The [slides on Parallel Algorithms](https://github.com/CIS565-Fall-2015/cis565-fall-2015.github.io/raw/master/lectures/2-Parallel-Algorithms.pptx) - for Scan, Stream Compaction, and Work-Efficient Parallel Scan. -* GPU Gems 3, Chapter 39 - [Parallel Prefix Sum (Scan) with CUDA](http://http.developer.nvidia.com/GPUGems3/gpugems3_ch39.html). - -Your GPU stream compaction implementation will live inside of the -`stream_compaction` subproject. This way, you will be able to easily copy it -over for use in your GPU path tracer. - - -## Part 0: The Usual - -This project (and all other CUDA projects in this course) requires an NVIDIA -graphics card with CUDA capability. Any card with Compute Capability 2.0 -(`sm_20`) or greater will work. Check your GPU on this -[compatibility table](https://developer.nvidia.com/cuda-gpus). -If you do not have a personal machine with these specs, you may use those -computers in the Moore 100B/C which have supported GPUs. - -**HOWEVER**: If you need to use the lab computer for your development, you will -not presently be able to do GPU performance profiling. This will be very -important for debugging performance bottlenecks in your program. - -### Useful existing code - -* `stream_compaction/common.h` - * `checkCUDAError` macro: checks for CUDA errors and exits if there were any. - * `ilog2ceil(x)`: computes the ceiling of log2(x), as an integer. -* `main.cpp` - * Some testing code for your implementations. - - -## Part 1: CPU Scan & Stream Compaction - -This stream compaction method will remove `0`s from an array of `int`s. - -In `stream_compaction/cpu.cu`, implement: - -* `StreamCompaction::CPU::scan`: compute an exclusive prefix sum. -* `StreamCompaction::CPU::compactWithoutScan`: stream compaction without using - the `scan` function. -* `StreamCompaction::CPU::compactWithScan`: stream compaction using the `scan` - function. Map the input array to an array of 0s and 1s, scan it, and use - scatter to produce the output. You will need a **CPU** scatter implementation - for this (see slides or GPU Gems chapter for an explanation). - -These implementations should only be a few lines long. - - -## Part 2: Naive GPU Scan Algorithm - -In `stream_compaction/naive.cu`, implement `StreamCompaction::Naive::scan` - -This uses the "Naive" algorithm from GPU Gems 3, Section 39.2.1. We haven't yet -taught shared memory, and you **shouldn't use it yet**. Example 39-1 uses -shared memory, but is limited to operating on very small arrays! Instead, write -this using global memory only. As a result of this, you will have to do -`ilog2ceil(n)` separate kernel invocations. - -Beware of errors in Example 39-1 in the book; both the pseudocode and the CUDA -code in the online version of Chapter 39 are known to have a few small errors -(in superscripting, missing braces, bad indentation, etc.) - -Since the parallel scan algorithm operates on a binary tree structure, it works -best with arrays with power-of-two length. Make sure your implementation works -on non-power-of-two sized arrays (see `ilog2ceil`). This requires extra memory -- your intermediate array sizes will need to be rounded to the next power of -two. - - -## Part 3: Work-Efficient GPU Scan & Stream Compaction - -### 3.1. Scan - -In `stream_compaction/efficient.cu`, implement -`StreamCompaction::Efficient::scan` - -All of the text in Part 2 applies. - -* This uses the "Work-Efficient" algorithm from GPU Gems 3, Section 39.2.2. -* Beware of errors in Example 39-2. -* Test non-power-of-two sized arrays. - -### 3.2. Stream Compaction - -This stream compaction method will remove `0`s from an array of `int`s. - -In `stream_compaction/efficient.cu`, implement -`StreamCompaction::Efficient::compact` - -For compaction, you will also need to implement the scatter algorithm presented -in the slides and the GPU Gems chapter. - -In `stream_compaction/common.cu`, implement these for use in `compact`: - -* `StreamCompaction::Common::kernMapToBoolean` -* `StreamCompaction::Common::kernScatter` - - -## Part 4: Using Thrust's Implementation - -In `stream_compaction/thrust.cu`, implement: - -* `StreamCompaction::Thrust::scan` - -This should be a very short function which wraps a call to the Thrust library -function `thrust::exclusive_scan(first, last, result)`. - -To measure timing, be sure to exclude memory operations by passing -`exclusive_scan` a `thrust::device_vector` (which is already allocated on the -GPU). You can create a `thrust::device_vector` by creating a -`thrust::host_vector` from the given pointer, then casting it. - - -## Part 5: Radix Sort (Extra Credit) (+10) - -Add an additional module to the `stream_compaction` subproject. Implement radix -sort using one of your scan implementations. Add tests to check its correctness. - - -## Write-up - -1. Update all of the TODOs at the top of this README. -2. Add a description of this project including a list of its features. -3. Add your performance analysis (see below). - -All extra credit features must be documented in your README, explaining its -value (with performance comparison, if applicable!) and showing an example how -it works. For radix sort, show how it is called and an example of its output. - -Always profile with Release mode builds and run without debugging. - -### Questions - -* Roughly optimize the block sizes of each of your implementations for minimal - run time on your GPU. - * (You shouldn't compare unoptimized implementations to each other!) - -* Compare all of these GPU Scan implementations (Naive, Work-Efficient, and - Thrust) to the serial CPU version of Scan. Plot a graph of the comparison - (with array size on the independent axis). - * You should use CUDA events for timing. Be sure **not** to include any - explicit memory operations in your performance measurements, for - comparability. - * To guess at what might be happening inside the Thrust implementation, take - a look at the Nsight timeline for its execution. - -* Write a brief explanation of the phenomena you see here. - * Can you find the performance bottlenecks? Is it memory I/O? Computation? Is - it different for each implementation? - -* Paste the output of the test program into a triple-backtick block in your - README. - * If you add your own tests (e.g. for radix sort or to test additional corner - cases), be sure to mention it explicitly. - -These questions should help guide you in performance analysis on future -assignments, as well. - -## Submit - -If you have modified any of the `CMakeLists.txt` files at all (aside from the -list of `SOURCE_FILES`), you must test that your project can build in Moore -100B/C. Beware of any build issues discussed on the Google Group. - -1. Open a GitHub pull request so that we can see that you have finished. - The title should be "Submission: YOUR NAME". -2. Send an email to the TA (gmail: kainino1+cis565@) with: - * **Subject**: in the form of `[CIS565] Project 2: PENNKEY` - * Direct link to your pull request on GitHub - * In the form of a grade (0-100+) with comments, evaluate your own - performance on the project. - * Feedback on the project itself, if any. +* Guan Sun +* Tested on: Mac OS X Yosemite 10.10.5, Intel Core i7 @ 2.3GHz 8GB, GeForce GT 650M 1024MB (Personal Laptop) + +## Project Description: + +In this project, the widely used GPU stream compaction is implemented in CUDA from scratch. The code of this project implemented the following: +* CPU Scan & Stream Campaction +* Naive GPU Scan Algorithm +* Work-Efficient GPU Scan & Stream Compaction +* Tested Thrust's Implementation + +## Performance Analysis: + +After optimizing the block sizes of each of the implementations for minimal run time. The comparision of different scan implementations are shown in the folloing figure. The X axis is the log of the array length, and the Y axis is the run time in ms. + +![](images/image1.png) + +* For the GPU Naive implemention, it needs to invoke the kernal funciton for `ilog2ceil(x)` times and each time the input data memory and output data memory on the GPU need to be swapped. This should be the bottleneck. +* For the GPU Work-Efficient implementation, it need to sweep up first and then sweep down, that means it will need to invoke both of them `ilog2ceil(x)` times, and each time the input and output data memory also need to be swapped. Besides, after the up-sweep, the result need to be copied to host memory to insert 0 and then copy to device memory. All these factors limits the performance of the Work-Efficient implementation. +* For the Thrust's Implementation, since the run time is longer than all others, it is possible that there are time cosuming memory allocationa and memory copy operations inside it. While the time of these operations in GPU Naive and GPU Work-Efficient are not included in the run time. + + +## Test Output: + +``` +**************** +** SCAN TESTS ** +**************** + [ 30 41 15 22 11 41 10 37 48 41 44 30 26 ... 14 0 ] +==== cpu scan, power-of-two ==== +CPU scan time is 0.0110 ms + [ 0 30 71 86 108 119 160 170 207 255 296 340 370 ... 101214 101228 ] +==== cpu scan, non-power-of-two ==== +CPU scan time is 0.0114 ms + [ 0 30 71 86 108 119 160 170 207 255 296 340 370 ... 101145 101172 ] + passed +==== naive scan, power-of-two ==== +GPU naive scan time is 0.3148 ms + [ 0 30 71 86 108 119 160 170 207 255 296 340 370 ... 101214 101228 ] + passed +==== naive scan, non-power-of-two ==== +GPU naive scan time is 0.3000 ms + [ 0 30 71 86 108 119 160 170 207 255 296 340 370 ... 101172 101172 ] + passed +==== work-efficient scan, power-of-two ==== +GPU work-efficient scan time is 0.5871 ms + [ 0 30 71 86 108 119 160 170 207 255 296 340 370 ... 101214 101228 ] + passed +==== work-efficient scan, non-power-of-two ==== +GPU work-efficient scan time is 0.5866 ms + [ 0 30 71 86 108 119 160 170 207 255 296 340 370 ... 101145 101172 ] + passed +==== thrust scan, power-of-two ==== +Thrust scan time is 8.9855 ms + [ 0 30 71 86 108 119 160 170 207 255 296 340 370 ... 101214 101228 ] + passed +==== thrust scan, non-power-of-two ==== +Thrust scan time is 9.9013 ms + [ 0 30 71 86 108 119 160 170 207 255 296 340 370 ... 101145 101172 ] + passed + +***************************** +** STREAM COMPACTION TESTS ** +***************************** + [ 2 3 3 0 1 1 2 1 2 1 2 0 2 ... 0 0 ] +==== cpu compact without scan, power-of-two ==== + [ 2 3 3 1 1 2 1 2 1 2 2 2 3 ... 3 1 ] + passed +==== cpu compact without scan, non-power-of-two ==== + [ 2 3 3 1 1 2 1 2 1 2 2 2 3 ... 1 3 ] + passed +==== cpu compact with scan ==== +CPU scan time is 0.0162 ms + [ 2 3 3 1 1 2 1 2 1 2 2 2 3 ... 3 1 ] + passed +==== work-efficient compact, power-of-two ==== +GPU work-efficient scan time is 1.8664 ms + [ 2 3 3 1 1 2 1 2 1 2 2 2 3 ... 3 1 ] + passed +==== work-efficient compact, non-power-of-two ==== +GPU work-efficient scan time is 1.8314 ms + [ 2 3 3 1 1 2 1 2 1 2 2 2 3 ... 1 3 ] + passed + +``` \ No newline at end of file diff --git a/images/image1.png b/images/image1.png new file mode 100644 index 0000000..2a440b7 Binary files /dev/null and b/images/image1.png differ diff --git a/src/main.cpp b/src/main.cpp index 675da35..0eca358 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -14,7 +14,7 @@ #include "testing_helpers.hpp" int main(int argc, char* argv[]) { - const int SIZE = 1 << 8; + const int SIZE = 1 << 12; const int NPOT = SIZE - 3; int a[SIZE], b[SIZE], c[SIZE]; @@ -43,37 +43,37 @@ int main(int argc, char* argv[]) { zeroArray(SIZE, c); printDesc("naive scan, power-of-two"); StreamCompaction::Naive::scan(SIZE, c, a); - //printArray(SIZE, c, true); + printArray(SIZE, c, true); printCmpResult(SIZE, b, c); zeroArray(SIZE, c); printDesc("naive scan, non-power-of-two"); StreamCompaction::Naive::scan(NPOT, c, a); - //printArray(SIZE, c, true); + printArray(SIZE, c, true); printCmpResult(NPOT, b, c); zeroArray(SIZE, c); printDesc("work-efficient scan, power-of-two"); StreamCompaction::Efficient::scan(SIZE, c, a); - //printArray(SIZE, c, true); + printArray(SIZE, c, true); printCmpResult(SIZE, b, c); zeroArray(SIZE, c); printDesc("work-efficient scan, non-power-of-two"); StreamCompaction::Efficient::scan(NPOT, c, a); - //printArray(NPOT, c, true); + printArray(NPOT, c, true); printCmpResult(NPOT, b, c); zeroArray(SIZE, c); printDesc("thrust scan, power-of-two"); StreamCompaction::Thrust::scan(SIZE, c, a); - //printArray(SIZE, c, true); + printArray(SIZE, c, true); printCmpResult(SIZE, b, c); zeroArray(SIZE, c); printDesc("thrust scan, non-power-of-two"); StreamCompaction::Thrust::scan(NPOT, c, a); - //printArray(NPOT, c, true); + printArray(NPOT, c, true); printCmpResult(NPOT, b, c); printf("\n"); @@ -112,12 +112,12 @@ int main(int argc, char* argv[]) { zeroArray(SIZE, c); printDesc("work-efficient compact, power-of-two"); count = StreamCompaction::Efficient::compact(SIZE, c, a); - //printArray(count, c, true); + printArray(count, c, true); printCmpLenResult(count, expectedCount, b, c); zeroArray(SIZE, c); printDesc("work-efficient compact, non-power-of-two"); count = StreamCompaction::Efficient::compact(NPOT, c, a); - //printArray(count, c, true); + printArray(count, c, true); printCmpLenResult(count, expectedNPOT, b, c); } diff --git a/stream_compaction/common.cu b/stream_compaction/common.cu index fe872d4..e529766 100644 --- a/stream_compaction/common.cu +++ b/stream_compaction/common.cu @@ -23,7 +23,15 @@ namespace Common { * which map to 0 will be removed, and elements which map to 1 will be kept. */ __global__ void kernMapToBoolean(int n, int *bools, const int *idata) { - // TODO + int index = threadIdx.x + (blockIdx.x * blockDim.x); + + if( index0 && (index+1)%(1<>>(n_new, j, dev_odata, dev_idata); + //cudaMemcpy(dev_idata, dev_odata, n_new * sizeof(int), cudaMemcpyDeviceToDevice); + dev_idata = dev_odata; + } + cudaMemcpy(idata_new, dev_odata, n_new * sizeof(int), cudaMemcpyDeviceToHost); + + idata_new[n_new-1] = 0; + cudaMemcpy(dev_idata, idata_new, n_new * sizeof(int), cudaMemcpyHostToDevice); + for(int k=ilog2ceil(n)-1; k>=0; k--) { + gpu_scan_down<<>>(n_new, k, dev_odata, dev_idata); + //cudaMemcpy(dev_idata, dev_odata, n_new * sizeof(int), cudaMemcpyDeviceToDevice); + dev_idata = dev_odata; + } + cudaEventRecord(stop); + + cudaEventSynchronize(stop); + cudaEventElapsedTime(&time, start, stop); + printf("GPU work-efficient scan time is %.4f ms \n", time); + + cudaMemcpy(odata, dev_odata, n_new * sizeof(int), cudaMemcpyDeviceToHost); + cudaFree(dev_idata); + cudaFree(dev_odata); } /** @@ -26,8 +103,46 @@ void scan(int n, int *odata, const int *idata) { * @returns The number of elements remaining after compaction. */ int compact(int n, int *odata, const int *idata) { - // TODO - return -1; + + int *dev_idata; + int *dev_odata; + int *dev_bools; + int *dev_indices; + + int hst_bools[n]; + int hst_indices[n]; + + int blockSize = 1024; + + dim3 fullBlocksPerGrid((n + blockSize - 1) / blockSize); + + cudaMalloc((void**)&dev_idata, n * sizeof(int)); + cudaMalloc((void**)&dev_odata, n * sizeof(int)); + cudaMalloc((void**)&dev_bools, n * sizeof(int)); + cudaMalloc((void**)&dev_indices, n * sizeof(int)); + + cudaMemcpy(dev_idata, idata, n * sizeof(int), cudaMemcpyHostToDevice); + Common::kernMapToBoolean<<>>(n, dev_bools, dev_idata); + cudaMemcpy(hst_bools, dev_bools, n * sizeof(int), cudaMemcpyDeviceToHost); + + scan(n, hst_indices, hst_bools); + //printf("n is %d \n", n); + + cudaMemcpy(dev_indices, hst_indices, n * sizeof(int), cudaMemcpyHostToDevice); + Common::kernScatter<<>>(n, dev_odata, dev_idata, dev_bools, dev_indices); + cudaMemcpy(odata, dev_odata, n * sizeof(int), cudaMemcpyDeviceToHost); + + cudaFree(dev_idata); + cudaFree(dev_odata); + cudaFree(dev_bools); + cudaFree(dev_indices); + + if(idata[n-1] == 0) { + return hst_indices[n-1]; + } else { + return hst_indices[n-1] + 1; + } + } } diff --git a/stream_compaction/naive.cu b/stream_compaction/naive.cu index 3d86b60..c0056e7 100644 --- a/stream_compaction/naive.cu +++ b/stream_compaction/naive.cu @@ -3,17 +3,75 @@ #include "common.h" #include "naive.h" + namespace StreamCompaction { namespace Naive { -// TODO: __global__ +__global__ void gpu_scan(int n, int d, int *odata, int *idata) { + int index = threadIdx.x + (blockIdx.x * blockDim.x); + + if( index < n ) { + if( index >= (1 << (d-1)) ) { + odata[index] = idata[index] + idata[index-(1 << (d-1))]; + } else { + odata[index] = idata[index]; + } + } +} /** * Performs prefix-sum (aka scan) on idata, storing the result into odata. */ void scan(int n, int *odata, const int *idata) { - // TODO - printf("TODO\n"); + int *dev_idata; + int *dev_odata; + int blockSize=1024; + +// int idata_new[n]; +// idata_new[0] = 0; +// for(int k=0; k>>(n_new, i, dev_odata, dev_idata); + cudaMemcpy(dev_idata, dev_odata, n_new * sizeof(int), cudaMemcpyDeviceToDevice); + //dev_idata = dev_odata; + } + cudaEventRecord(stop); + + cudaEventSynchronize(stop); + cudaEventElapsedTime(&time, start, stop); + printf("GPU naive scan time is %.4f ms \n", time); + + cudaMemcpy(odata, dev_odata, n_new * sizeof(int), cudaMemcpyDeviceToHost); + cudaFree(dev_idata); + cudaFree(dev_odata); + } } diff --git a/stream_compaction/thrust.cu b/stream_compaction/thrust.cu index d8dbb32..51340a2 100644 --- a/stream_compaction/thrust.cu +++ b/stream_compaction/thrust.cu @@ -13,9 +13,25 @@ namespace Thrust { * Performs prefix-sum (aka scan) on idata, storing the result into odata. */ void scan(int n, int *odata, const int *idata) { - // TODO use `thrust::exclusive_scan` - // example: for device_vectors dv_in and dv_out: - // thrust::exclusive_scan(dv_in.begin(), dv_in.end(), dv_out.begin()); + + thrust::host_vector thrustHst_idata(idata, idata+n); + thrust::device_vector thrustDev_idata(thrustHst_idata); + thrust::device_vector thrustDev_odata(n); + + float time = 0; + cudaEvent_t start, stop; + cudaEventCreate(&start); + cudaEventCreate(&stop); + + cudaEventRecord(start); + thrust::exclusive_scan(thrustDev_idata.begin(), thrustDev_idata.end(), thrustDev_odata.begin()); + cudaEventRecord(stop); + + cudaEventSynchronize(stop); + cudaEventElapsedTime(&time, start, stop); + printf("Thrust scan time is %.4f ms \n", time); + + thrust::copy(thrustDev_odata.begin(), thrustDev_odata.end(), odata); } }