diff --git a/README.md b/README.md index a82ea0f..d737f6a 100644 --- a/README.md +++ b/README.md @@ -3,211 +3,78 @@ 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. +* Bradley Crusco +* Tested on: Windows 10, i7-3770K @ 3.50GHz 16GB, 2 x GTX 980 4096MB (Personal Computer) + +### Description + +This project is a series of scan and stream compression algorithms. +Features: + * CPU Exclusive Prefix Sum Scan + * CPU Stream Compaction Without Scan + * CPU Stream Compaction using Exclusive Prefix Sum Scan + * Naive GPU Exclusive Preflix Sum Scan + * Work-Efficient GPU Exclusive Preflix Sum Scan + * GPU Stream Compaction using Work-Efficient GPU Exclusive Prefix Sum Scan + +### Performance Analysis + +**Scan Implementation Execution Time vs. Array Size** +![](images/Project 2 Analysis.png "Performance Analysis") +Unfortunately, the results from testing are not very impressive. The sequential CPU implementation easily out performs everything but the Thrust implementation, and the worst performer by far is the work-efficient implementation, which we'd expect to outperform the naive scan. So why is this? I am not 100% sure. However I had difficulty determining how to configure the grid and block size optimally, and as a result all the GPU implementations are using the same ratio, with 512 threads per block. A better understanding of how to configure this might result in performance more in line with what we'd expect to see. + +The other possible cause may be that our arrays are not very large, with the maximum array I tested with being 1024. It could be the case that this wasn't enough data for the GPU to take advantage of and counteract the overhead of the parallel algorithm vs. the sequential and is ultimately bottlenecked by memory I/O + +### Test Program Output (Array Size 256) + +``` +**************** +** SCAN TESTS ** +**************** + [ 3 29 33 19 0 16 10 40 39 50 44 30 9 ... 4 0 ] +==== cpu scan, power-of-two ==== +CPU execution time for scan: 0.00109ms + [ 0 3 32 65 84 84 100 110 150 189 239 283 313 ... 6684 6688 ] +==== cpu scan, non-power-of-two ==== +CPU execution time for scan: 0.00106ms + [ 0 3 32 65 84 84 100 110 150 189 239 283 313 ... 6613 6626 ] + passed +==== naive scan, power-of-two ==== +CUDA execution time for naive scan: 0.07440ms + passed +==== naive scan, non-power-of-two ==== +CUDA execution time for naive scan: 0.07222ms + passed +==== work-efficient scan, power-of-two ==== +CUDA execution time for work efficient scan: 0.21798ms + passed +==== work-efficient scan, non-power-of-two ==== +CUDA execution time for work efficient scan: 0.21632ms + passed +==== thrust scan, power-of-two ==== + passed +==== thrust scan, non-power-of-two ==== + passed + +***************************** +** STREAM COMPACTION TESTS ** +***************************** + [ 4 3 0 3 4 2 3 2 3 1 1 1 4 ... 3 0 ] +==== cpu compact without scan, power-of-two ==== +CPU execution time for compact without scan: 0.00106ms + [ 4 3 3 4 2 3 2 3 1 1 1 4 3 ... 3 3 ] + passed +==== cpu compact without scan, non-power-of-two ==== +CPU execution time for compact without scan: 0.00106ms + [ 4 3 3 4 2 3 2 3 1 1 1 4 3 ... 4 4 ] + passed +==== cpu compact with scan ==== +CPU execution time for compact with scan: 0.00109ms + [ 4 3 3 4 2 3 2 3 1 1 1 4 3 ... 3 3 ] + passed +==== work-efficient compact, power-of-two ==== +CUDA execution time for stream compaction: 0.22755ms + passed +==== work-efficient compact, non-power-of-two ==== +CUDA execution time for stream compaction: 0.22557ms + passed diff --git a/images/Project 2 Analysis.png b/images/Project 2 Analysis.png new file mode 100644 index 0000000..5b27cf4 Binary files /dev/null and b/images/Project 2 Analysis.png differ diff --git a/src/main.cpp b/src/main.cpp index efc8c06..395bcbb 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -19,6 +19,7 @@ int main(int argc, char* argv[]) { const int SIZE = 1 << 8; const int NPOT = SIZE - 3; int a[SIZE], b[SIZE], c[SIZE]; + float ms_time = 0.0f; // Scan tests @@ -32,12 +33,14 @@ int main(int argc, char* argv[]) { zeroArray(SIZE, b); printDesc("cpu scan, power-of-two"); - StreamCompaction::CPU::scan(SIZE, b, a); + ms_time = StreamCompaction::CPU::scan(SIZE, b, a); + printf("CPU execution time for scan: %.5fms\n", ms_time); printArray(SIZE, b, true); zeroArray(SIZE, c); printDesc("cpu scan, non-power-of-two"); - StreamCompaction::CPU::scan(NPOT, c, a); + ms_time = StreamCompaction::CPU::scan(NPOT, c, a); + printf("CPU execution time for scan: %.5fms\n", ms_time); printArray(NPOT, b, true); printCmpResult(NPOT, b, c); @@ -52,16 +55,18 @@ int main(int argc, char* argv[]) { StreamCompaction::Naive::scan(NPOT, c, a); //printArray(SIZE, c, true); printCmpResult(NPOT, b, c); - + zeroArray(SIZE, c); printDesc("work-efficient scan, power-of-two"); - StreamCompaction::Efficient::scan(SIZE, c, a); + ms_time = StreamCompaction::Efficient::scan(SIZE, c, a); + printf("CUDA execution time for work efficient scan: %.5fms\n", ms_time); //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); + ms_time = StreamCompaction::Efficient::scan(NPOT, c, a); + printf("CUDA execution time for work efficient scan: %.5fms\n", ms_time); //printArray(NPOT, c, true); printCmpResult(NPOT, b, c); diff --git a/stream_compaction/common.cu b/stream_compaction/common.cu index fe872d4..41b537c 100644 --- a/stream_compaction/common.cu +++ b/stream_compaction/common.cu @@ -23,7 +23,11 @@ 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 k = threadIdx.x + (blockIdx.x * blockDim.x); + + if (k < n) { + bools[k] = !!idata[k]; + } } /** @@ -32,7 +36,13 @@ __global__ void kernMapToBoolean(int n, int *bools, const int *idata) { */ __global__ void kernScatter(int n, int *odata, const int *idata, const int *bools, const int *indices) { - // TODO + int k = threadIdx.x + (blockIdx.x * blockDim.x); + + if (k < n) { + if (bools[k] == 1) { + odata[indices[k]] = idata[k]; + } + } } } diff --git a/stream_compaction/common.h b/stream_compaction/common.h index 4f52663..23aafae 100644 --- a/stream_compaction/common.h +++ b/stream_compaction/common.h @@ -6,6 +6,7 @@ #define FILENAME (strrchr(__FILE__, '/') ? strrchr(__FILE__, '/') + 1 : __FILE__) #define checkCUDAError(msg) checkCUDAErrorFn(msg, FILENAME, __LINE__) +#define blockSize 512 /** * Check for CUDA errors; print and exit if there was a problem. diff --git a/stream_compaction/cpu.cu b/stream_compaction/cpu.cu index e600c29..a94d063 100644 --- a/stream_compaction/cpu.cu +++ b/stream_compaction/cpu.cu @@ -1,5 +1,9 @@ #include +#include #include "cpu.h" +#include +#include +#include namespace StreamCompaction { namespace CPU { @@ -7,9 +11,21 @@ namespace CPU { /** * CPU scan (prefix sum). */ -void scan(int n, int *odata, const int *idata) { - // TODO - printf("TODO\n"); +float scan(int n, int *odata, const int *idata) { + cudaEvent_t start, stop; + float ms_time = 0.0f; + cudaEventCreate(&start); + cudaEventCreate(&stop); + + cudaEventRecord(start); + odata[0] = 0; + for (int i = 1; i < n; i++) { + odata[i] = odata[i - 1] + idata[i - 1]; + } + cudaEventRecord(stop); + cudaEventSynchronize(stop); + cudaEventElapsedTime(&ms_time, start, stop); + return ms_time; } /** @@ -18,8 +34,31 @@ void scan(int n, int *odata, const int *idata) { * @returns the number of elements remaining after compaction. */ int compactWithoutScan(int n, int *odata, const int *idata) { - // TODO - return -1; + cudaEvent_t start, stop; + float ms_time = 0.0f; + cudaEventCreate(&start); + cudaEventCreate(&stop); + + cudaEventRecord(start); + int j = 0; + for (int i = 0; i < n; i++) { + if (idata[i] != 0) { + odata[j] = idata[i]; + j++; + } + } + cudaEventRecord(stop); + cudaEventSynchronize(stop); + cudaEventElapsedTime(&ms_time, start, stop); + printf("CPU execution time for compact without scan: %.5fms\n", ms_time); + + return j; +} + +void zeroArray(int n, int *a) { + for (int i = 0; i < n; i++) { + a[i] = 0; + } } /** @@ -28,8 +67,43 @@ int compactWithoutScan(int n, int *odata, const int *idata) { * @returns the number of elements remaining after compaction. */ int compactWithScan(int n, int *odata, const int *idata) { - // TODO - return -1; + int *temp = (int*)malloc(n * sizeof(int)); + zeroArray(n, temp); + int *scan_output = (int*)malloc(n * sizeof(int)); + zeroArray(n, scan_output); + + cudaEvent_t start, stop; + float ms_time = 0.0f; + float ms_total_time = 0.0f; + cudaEventCreate(&start); + cudaEventCreate(&stop); + + // Compute temporary array + for (int i = 0; i < n; i++) { + if (idata[i] != 0) { + temp[i] = 1; + } + } + + // Run exclusive scan on the temporary array + ms_time = scan(n, scan_output, temp); + ms_total_time += ms_time; + ms_time = 0.0f; + + // Scatter + cudaEventCreate(&start); + for (int i = 0; i < n; i++) { + if (temp[i] == 1) { + odata[scan_output[i]] = idata[i]; + } + } + cudaEventRecord(stop); + cudaEventSynchronize(stop); + cudaEventElapsedTime(&ms_time, start, stop); + ms_total_time += ms_time; + printf("CPU execution time for compact with scan: %.5fms\n", ms_total_time); + + return scan_output[n - 1] + temp[n - 1]; } } diff --git a/stream_compaction/cpu.h b/stream_compaction/cpu.h index 6348bf3..8f32b0d 100644 --- a/stream_compaction/cpu.h +++ b/stream_compaction/cpu.h @@ -2,7 +2,7 @@ namespace StreamCompaction { namespace CPU { - void scan(int n, int *odata, const int *idata); + float scan(int n, int *odata, const int *idata); int compactWithoutScan(int n, int *odata, const int *idata); diff --git a/stream_compaction/efficient.cu b/stream_compaction/efficient.cu index b2f739b..2cf8008 100644 --- a/stream_compaction/efficient.cu +++ b/stream_compaction/efficient.cu @@ -6,14 +6,92 @@ namespace StreamCompaction { namespace Efficient { -// TODO: __global__ +__global__ void up_sweep(int n, int d, int *data) { + int k = threadIdx.x + (blockIdx.x * blockDim.x); + if (k < n) { + int p2d = pow(2.0, (double)d); + int p2da1 = pow(2.0, (double)(d + 1)); + + if (k % p2da1 == 0) { + data[k + p2da1 - 1] += data[k + p2d - 1]; + } + } +} + +__global__ void down_sweep(int n, int d, int *data) { + int k = threadIdx.x + (blockIdx.x * blockDim.x); + + if (k < n) { + int p2d = pow(2.0, (double)d); + int p2da1 = pow(2.0, (double)(d + 1)); + + if (k % p2da1 == 0) { + int temp = data[k + p2d - 1]; + data[k + p2d - 1] = data[k + p2da1 - 1]; + data[k + p2da1 - 1] += temp; + } + } +} + +void padArrayRange(int start, int end, int *a) { + for (int i = start; i < end; i++) { + a[i] = 0; + } +} /** * 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"); +float scan(int n, int *odata, const int *idata) { + int m = pow(2, ilog2ceil(n)); + int *new_idata = (int*)malloc(m * sizeof(int)); + dim3 fullBlocksPerGrid((m + blockSize - 1) / blockSize); + dim3 threadsPerBlock(blockSize); + + cudaEvent_t start, stop; + float ms_time = 0.0f; + float ms_total_time = 0.0f; + cudaEventCreate(&start); + cudaEventCreate(&stop); + + // Expand array to next power of 2 size + for (int i = 0; i < n; i++) { + new_idata[i] = idata[i]; + } + padArrayRange(n, m, new_idata); + + // Can use one array for input and output in this implementation + int *dev_data; + cudaMalloc((void**)&dev_data, m * sizeof(int)); + cudaMemcpy(dev_data, new_idata, m * sizeof(int), cudaMemcpyHostToDevice); + + // Execute scan on device + cudaEventRecord(start); + for (int d = 0; d < ilog2ceil(n); d++) { + up_sweep<<>>(n, d, dev_data); + } + cudaEventRecord(stop); + cudaEventSynchronize(stop); + cudaEventElapsedTime(&ms_time, start, stop); + ms_total_time += ms_time; + ms_time = 0.0f; + + cudaMemset((void*)&dev_data[m - 1], 0, sizeof(int)); + cudaEventRecord(start); + for (int d = ilog2ceil(n) - 1; d >= 0; d--) { + down_sweep<<>>(n, d, dev_data); + } + cudaEventRecord(stop); + cudaEventSynchronize(stop); + cudaEventElapsedTime(&ms_time, start, stop); + ms_total_time += ms_time; + + cudaMemcpy(odata, dev_data, n * sizeof(int), cudaMemcpyDeviceToHost); + + cudaFree(dev_data); + free(new_idata); + + return ms_total_time; } /** @@ -26,8 +104,65 @@ 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 *bools = (int*)malloc(n * sizeof(int)); + int *scan_data = (int*)malloc(n * sizeof(int)); + int num_remaining = -1; + dim3 fullBlocksPerGrid((n + blockSize - 1) / blockSize); + dim3 threadsPerBlock(blockSize); + + cudaEvent_t start, stop; + float ms_time = 0.0f; + float ms_total_time = 0.0f; + cudaEventCreate(&start); + cudaEventCreate(&stop); + + int *dev_bools; + int *dev_idata; + int *dev_odata; + int *dev_scan_data; + + cudaMalloc((void**)&dev_bools, n * sizeof(int)); + cudaMalloc((void**)&dev_idata, n * sizeof(int)); + cudaMemcpy(dev_idata, idata, n * sizeof(int), cudaMemcpyHostToDevice); + + cudaMalloc((void**)&dev_odata, n * sizeof(int)); + cudaMalloc((void**)&dev_scan_data, n * sizeof(int)); + + // Map to boolean + cudaEventRecord(start); + StreamCompaction::Common::kernMapToBoolean<<>>(n, dev_bools, dev_idata); + cudaEventRecord(stop); + cudaEventSynchronize(stop); + cudaEventElapsedTime(&ms_time, start, stop); + ms_total_time += ms_time; + ms_time = 0.0f; + + cudaMemcpy(bools, dev_bools, n * sizeof(int), cudaMemcpyDeviceToHost); + + // Execute the scan + ms_total_time += scan(n, scan_data, bools); + num_remaining = scan_data[n - 1] + bools[n - 1]; + + // Execute the scatter + cudaMemcpy(dev_scan_data, scan_data, n * sizeof(int), cudaMemcpyHostToDevice); + cudaEventRecord(start); + StreamCompaction::Common::kernScatter<<>>(n, dev_odata, dev_idata, dev_bools, dev_scan_data); + cudaEventRecord(stop); + cudaEventSynchronize(stop); + cudaEventElapsedTime(&ms_time, start, stop); + ms_total_time += ms_time; + printf("CUDA execution time for stream compaction: %.5fms\n", ms_total_time); + + cudaMemcpy(odata, dev_odata, n * sizeof(int), cudaMemcpyDeviceToHost); + + cudaFree(dev_bools); + cudaFree(dev_idata); + cudaFree(dev_odata); + cudaFree(dev_scan_data); + free(bools); + free(scan_data); + + return num_remaining; } } diff --git a/stream_compaction/efficient.h b/stream_compaction/efficient.h index 395ba10..57afdf6 100644 --- a/stream_compaction/efficient.h +++ b/stream_compaction/efficient.h @@ -2,7 +2,7 @@ namespace StreamCompaction { namespace Efficient { - void scan(int n, int *odata, const int *idata); + float scan(int n, int *odata, const int *idata); int compact(int n, int *odata, const int *idata); } diff --git a/stream_compaction/naive.cu b/stream_compaction/naive.cu index 3d86b60..38bacab 100644 --- a/stream_compaction/naive.cu +++ b/stream_compaction/naive.cu @@ -6,15 +6,75 @@ namespace StreamCompaction { namespace Naive { -// TODO: __global__ +__global__ void kern_scan(int n, int d, int *idata, int *odata) { + int k = threadIdx.x + (blockIdx.x * blockDim.x); + + if (k < n) { + if (k >= (int)pow(2.0, (double)(d - 1))) { + odata[k] = idata[k - (int)pow(2.0, (double)(d - 1))] + idata[k]; + } + else { + odata[k] = idata[k]; + } + } +} + +void padArrayRange(int start, int end, int *a) { + for (int i = start; i < end; i++) { + a[i] = 0; + } +} /** * 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 m = pow(2, ilog2ceil(n)); + int *new_idata = (int*)malloc(m * sizeof(int)); + dim3 fullBlocksPerGrid((m + blockSize - 1) / blockSize); + dim3 threadsPerBlock(blockSize); + + cudaEvent_t start, stop; + float ms_time = 0.0f; + cudaEventCreate(&start); + cudaEventCreate(&stop); + + // Expand array to next power of 2 size + for (int i = 0; i < n; i++) { + new_idata[i] = idata[i]; + } + padArrayRange(n, m, new_idata); + + int *dev_idata; + int *dev_odata; + + cudaMalloc((void**)&dev_idata, m * sizeof(int)); + cudaMemcpy(dev_idata, new_idata, m * sizeof(int), cudaMemcpyHostToDevice); + + cudaMalloc((void**)&dev_odata, m * sizeof(int)); + + + // Execute scan on device + cudaEventRecord(start); + for (int d = 1; d <= ilog2ceil(n); d++) { + kern_scan<<>>(n, d, dev_idata, dev_odata); + dev_idata = dev_odata; + } + cudaEventRecord(stop); + cudaEventSynchronize(stop); + + cudaEventElapsedTime(&ms_time, start, stop); + printf("CUDA execution time for naive scan: %.5fms\n", ms_time); + + odata[0] = 0; + cudaMemcpy(odata + 1, dev_odata, (m * sizeof(int)) - sizeof(int), cudaMemcpyDeviceToHost); + + cudaFree(dev_idata); + cudaFree(dev_odata); + free(new_idata); } } } + + diff --git a/stream_compaction/thrust.cu b/stream_compaction/thrust.cu index d8dbb32..6d16f5f 100644 --- a/stream_compaction/thrust.cu +++ b/stream_compaction/thrust.cu @@ -13,9 +13,16 @@ 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 hst_in(idata, idata + n); + thrust::device_vector dev_in = hst_in; + thrust::device_vector dev_out(n); + + thrust::exclusive_scan(dev_in.begin(), dev_in.end(), dev_out.begin()); + thrust::host_vector hst_out = dev_out; + + for (int i = 0; i < n; i++) { + odata[i] = hst_out[i]; + } } }