Introduction

本文内容主要系摘录翻译自Ang Li的博士毕业论文。

1.Perfect

Power Efficiency Revolution for Embedded Computing

http://hpc.pnl.gov/PERFECT/

Application Domains	Kernels
PERFECT Application 1	Discrete Wavelet Transform
	2D Convolution
	Histogram Equalization
Space Time Adaptive Processing	System Solver
	Inner Product
	Outer Product
Synthetic Aperture Radar	Interpolation 1
	Interpolation 2
	Back Projection (Non-Fourier SAR)
Wide Area Motion Imaging	Debayer
	Image Registration
	Change Detection
Required Kernels	Sort
	FFT 1D
	FFT 2D

2. AxBench

A Multiplatform Benchmark Suite for Approximate Computing

One of the goals of AxBench is to provide a diverse set of applications to further facilitate research and development in approximate computing.

http://ieeexplore.ieee.org/abstract/document/7755728/

下载地址

http://axbench.org/

benchmark	platform	domain	Quality Metric
binarization	GPU	Image Processing	Image Diff
blackscholes	CPU, GPU	Finance	Avg. Relative Error
brent-kung	ASIC	Arithmetic Computation	Avg. Relative Error
canneal	CPU	Optimization	Avg. Relative Error
convolution	GPU	Machine Learning	Avg. Relative Error
fastwalsh	GPU	Signal Processing	Image Diff
fft	CPU	Signal Processing	Avg. Relative Error
fir	ASIC	Signal Processing	Avg. Relative Error
forwardk2j	CPU, ASIC	Robotics	Avg. Relative Error
inversek2j	CPU, GPU, ASIC	Robotics	Avg. Relative Error
jmeint	CPU, GPU	3D Gaming	Miss Rate
jpeg	CPU	Image Processing	Image Diff
kmeans	CPU, ASIC	Machine Learning	Image Diff
kogge-stone	ASIC	Arithmetic Computation	Avg. Relative Error
laplacian	GPU	Image Processing	Image Diff
meanfilter	GPU	Machine Vision	Image Diff
neural network	ASIC	Machine Learning	Avg. Relative Error
newton-raph	GPU	Numerical Analysis	Avg. Relative Error
sobel	CPU, GPU, ASIC	Image Processing	Image Diff
srad	GPU	Medical Imaging	Image Diff
wallace-tree	ASIC	Arithmetic Computation	Avg. Relative Error

3. Rodinia

http://rodinia.cs.virginia.edu/

下载页面：

http://lava.cs.virginia.edu/Rodinia/download_links.htm


Applications	Dwarves	Domains	Parallel Model	Incre. Ver.
Leukocyte	Structured Grid	Medical Imaging	CUDA, OMP, OCL	√
Heart Wall	Structured Grid	Medical Imaging	CUDA, OMP, OCL
MUMmerGPU	Graph Traversal	Bioinformatics	CUDA, OMP
CFD Solver1	Unstructured Grid	Fluid Dynamics	CUDA, OMP, OCL
LU Decomposition	Dense Linear Algebra	Linear Algebra	CUDA, OMP, OCL	√
HotSpot	Structured Grid	Physics Simulation	CUDA, OMP, OCL
Back Propagation	Unstructured Grid	Pattern Recognition	CUDA, OMP, OCL
Needleman-Wunsch	Dynamic Programming	Bioinformatics	CUDA, OMP, OCL	√
Kmeans	Dense Linear Algebra	Data Mining	CUDA, OMP, OCL
Breadth-First Search1	Graph Traversal	Graph Algorithms	CUDA, OMP, OCL
SRAD	Structured Grid	Image Processing	CUDA, OMP, OCL	√
Streamcluster1	Dense Linear Algebra	Data Mining	CUDA, OMP, OCL
Particle Filter	Structured Grid	Medical Imaging	CUDA, OMP, OCL
PathFinder	Dynamic Programming	Grid Traversal	CUDA, OMP, OCL
Gaussian Elimination	Dense Linear Algebra	Linear Algebra	CUDA, OCL
k-Nearest Neighbors	Dense Linear Algebra	Data Mining	CUDA, OMP, OCL
LavaMD2	N-Body	Molecular Dynamics	CUDA, OMP, OCL
Myocyte	Structured Grid	Biological Simulation	CUDA, OMP, OCL
B+ Tree	Graph Traversal	Search	CUDA, OMP, OCL
GPUDWT	Spectral Method	Image/Video Compression	CUDA, OCL
Hybrid Sort	Sorting	Sorting Algorithms	CUDA, OCL
Hotspot3D	Structured Grid	Physics Simulation	CUDA, OCL, OMP	Hotspot for 3D IC
Huffman	Finite State Machine	Lossless data compression	CUDA, OCL

Ang Li的分类：

Application	Description	Domain	CUDA	OpenCL	OpenMP
backprop	Perceptron back propagation	Neural Network	Yes	Yes	Yes
bfs	Breadth first search	Graph Algorithm	Yes	Yes	Yes
b+tree	B+tree Operation	Searching Yes	Yes	Yes
leukocyte	Detect leukocytes in blood vessel video	Medical Imaging	Yes	Yes	Yes
heartwall	Tracks the mouse heart movement by stimulus	Medical Imaging	Yes	No	Yes
cfd	Finite volume solver for 3D Euler equations for flow	Fluid Dynamics	Yes	Yes	Yes
lud	Calculate the solutions of a set of linear equations	Linear Algebra	Yes	Yes	Yes
hotspot	Estimate processor temperature	Physical Simulation	Yes	Yes	Yes
nw	Optimization method for DNA sequence alignments	Bioinformatics	Yes	Yes	Yes
kmeans	Clustering algorithm	Data Mining	Yes	Yes	Yes
srad	Speckle reducing anisotropic diffusion	Image Processing	Yes	Yes	Yes
streamcluster	Finds medians to assign points to nearest centers	Data Mining	Yes	Yes	Yes
particlefilter	Locate object location based on Noise and path	Medical Imaging	Yes	Yes	Yes
pathfinder	Dynamic programming to find a path on a 2D grid Grid	Traversal	Yes	Yes	Yes
gaussian	Solving variables in a linear system	Linear Algebra	Yes	Yes	No
nn	Find k-nearest neighbors from an unstructured data set	Data Mining	Yes	Yes	Yes
lavaMD	Calculate particle potential and relocation in 3D	Molecular Dynamics	Yes	Yes	Yes
myocyte	Simulate the behavior of cardiac hear muscle cell	Biological Simulation	Yes	Yes	Yes

4. Parboil

Parboil强调面向吞吐量的流媒体应用。其中的每个应用都有原生的CUDA应用和优化过的应用。

Application	Description	Domain	CUDA	OpenCL	C
bfs	Breadth-first-search	Graph Algorithm	Yes	Yes	Yes
cutcp	Compute Coulombic potential for a 3D grid	Molecular Dynamics	Yes	Yes	Yes
histogram	Compute 2D saturating histogram with maximum 256 bins	Data Mining	Yes	Yes	Yes
lbm	Fluid dynamics simulation using Lattice-Bolzmann Method	Fluid Dynamics	Yes	Yes	Yes
mm	Dense matrix-matrix multiply	Linear Algebra	Yes	Yes	Yes
mri-gridding	Compute regular data grid via weighted interpolation	Medical Imaging	Yes	Yes	Yes
mir-q	Compute scanner configuration for calibration in 3D MRI	Medical Imaging	Yes	Yes	Yes
sad	Sum of absolute differences kernel in MPEG video encoders	Image Processing	Yes	Yes	Yes
spmv	Compute the product of a sparse matrix with a dense vector	Linear Algebra	Yes	Yes	Yes
stencil	An iterative Jacobi stencil operation on a regular 3D grid	Cellular Automation	Yes	Yes	Yes
tpacf	Analyze the spatial distribution of astronomical bodies	Data Mining	Yes	Yes	Yes

5. Shoc

测量协处理的稳定性和性能，such as GPUs, Xeon-Phi, etc。

Application	Description	Domain	CUDA	OpenCL	C
qtclustering	Group genes into high quality clusters	Bioinformatics	Yes	No	No
s3d	Compute chemical reaction rate across a 3D grid	Simulation	Yes	Yes	No
scan	Parallel prefix sum of floating point numbers	Data Mining	Yes	Yes	No
reduction	Sum reduction operation of floating point numbers	Data Mining	Yes	Yes	No
md	Lennard-Jones potential computations	Molecular Dynamics	Yes	Yes	No
fft	Fast Fourier transform	Signal Processing	Yes	Yes	No
sgemm	Single precision general matrix multiply	Linear Algebra	Yes	Yes	No
sort	Fast radix sort program	Data Mining	Yes	Yes	No
stencil2d	Standard 2d 9 points stencil calculation	Cellular Automation	Yes	Yes	No
bfs	Breadth-first-search	Graph Algorithm	Yes	Yes	No
spmv	Sparse matrix vector multiplication	Linear Algebra	Yes	Yes	Yes

6. Polybench

包含从[非]结构循环嵌套转换的Kernel。这些循环以前用于评估基于多面体模型的优化工具。

Application	Description	Domain	CUDA	OpenCL	C
2dconv	2D convolution	Linear Algebra	Yes	Yes	Yes
2mm	2 matrix multiply	Linear Algebra	Yes	Yes	Yes
3dconv	3D convolution	Linear Algebra	Yes	Yes	Yes
3mm	3 matrix multiply	Linear Algebra	Yes	Yes	Yes
atax	Matrix transpose and vector multiplication	Linear Algebra	Yes	Yes	Yes
bicg	Bicg kernel for BiCGStab linear solver	Linear Algebra	Yes	Yes	Yes
corr	Correlation computation	Linear Algebra	Yes	Yes	Yes
covar	Covariance computation	Linear Algebra	Yes	Yes	Yes
fdtd2d	2D finite difference time domain kernel	Simulation	Yes	Yes	Yes
gemm	matrix multiply	Linear Algebra	Yes	Yes	Yes
gesummv	Scalar vector and matrix multiplication	Linear Algebra	Yes	Yes	Yes
gramschm	Gram-schmidt process	Linear Algebra	Yes	Yes	Yes
mvt	Matrix vector product and transpose	Linear Algebra	Yes	Yes	Yes
syr2k	Symmetric rank-2k operations	Linear Algebra	Yes	Yes	Yes
syrk	Symmetric rank-k operations	Linear Algebra	Yes	Yes	Yes

7. Mars

用map reduce实现的data-mining的benchmark。

Application	Description	Domain	CUDA	OpenCL	C
sm	Find the position of a string in a file	Data Mining	Yes	No	No
ii	Build inverted index for links in HTML files	Data Mining	Yes	No	No
ss	Compute pair-wise similarity score for docs	Data Mining	Yes	No	No
mm	Multiply two matrices	Linear Algebra	Yes	No	No
pvc	Count distinct page views from web logs	Data Mining	Yes	No	No
pvr	Find the top ten hottest pages in the web log	Data Mining	Yes	No	No

8. Longstar

关注于不规则的应用，主要是数据依赖和拓扑依赖。

Application	Description	Domain	CUDA	OpenCL	C
bfs	Breadth first search	Graph Algorithm	Yes	No	No
bh	Simulate the gravitational forces in Barnes-Hut	algorithm Simulation	Yes	No	No
dc	Lossless compression upon double-precision FP data	Signal Processing	Yes	No	No
dmr	Meshrefinement algorithm from computational geometry	Image Processing	Yes	No	No
pta	Andersen’s flow/context-insensitive points-to analysis	Graph Algorithm	Yes	No	No
sp	Heuristic SAT-solver based on BaYesian inference	Graph Algorithm	Yes	No	No
sssp	Shortest path in a directed graph with weighted edges	Graph Algorithm	Yes	No	No
tsp	Traveling salesman problem	Graph Algorithm	Yes	No	No

9. CUDA SDK

Application	Description	Domain	CUDA	OpenCL	C
bilateralFilter	Edge-preserving non-linear smoothing filter	Image Processing	Yes	Yes	Yes
binomialOption	Evaluate option call price using binomial model	Computational Finance	Yes	Yes	Yes
BlackScholes	Evaluate option call price using Black-Scholes model	Computational Finance	Yes	Yes	Yes
convolutionFFT2D	2D convolutions using FFT	Image Processing	Yes	Yes	Yes
dct8x8	Discrete cosine transform for blocks of 8 by 8 pixels	Image Processing	Yes	Yes	Yes
dxtc	High quality DXT compression	Image Processing	Yes	Yes	Yes
dwtHaar1D	1D discrete Haar wavelet decomposition	Image Processing	Yes	Yes	Yes
eigenvalues	Eigenvalues of a tridiagonal symmetric matrix	Linear Algebra	Yes	Yes	Yes
fastWalshTransform	Hadamard-ordered Fast Walsh transform	Linear Algebra	Yes	Yes	Yes
FDTD3d	Finite differences	time domain progression stencil	Cellular Automation	Yes	Yes
grabcutNPP	GrabCut approach using the 8 neighborhood	Graph Algorithm	Yes	Yes	Yes
histogram	64/256 bin histogram	Data Mining	Yes	Yes	Yes
imageDenoising	Using KNN and NLM for image denoising	Image Processing	Yes	Yes	Yes
lineOfSight	A simple line-of-sight algorithm	Graphic Application	Yes	Yes	Yes
Mandelbrot	Mandelbrot or Julia sets interactively	Graphic Application	Yes	Yes	Yes
matrixMul	Matrix multiplication	Linear Algebra	Yes	Yes	Yes
mergeSortv	Merge Sort algorithm	Data Mining	Yes	Yes	No
MersenneTwister	The Mersenne Twister random number generator	Signal Processing	Yes	Yes	Yes
MonteCarlo	Evaluate option call price using Monte Carlo approach	Computational Finance	Yes	Yes	Yes
nbody	All-pairs gravitational n-body simulation	Simulation	Yes	Yes	Yes
oceanFFT	Simulate an Ocean height field	Simulation	Yes	Yes	Yes
reduction	Compute the sum of a large arrays of values	Data Mining	Yes	Yes	No
scalarProd	Calculate scalar products of input vector pairs	Linear Algebra	Yes	Yes	Yes
scan	Parallel prefix sum	Data Mining	Yes	Yes	Yes
SobelFilter	Sobel edge detection filter for 8-bit monochrome images	Image Processing	Yes	Yes	Yes
SobolQRNG	Sobol Quasirandom Sequence Generator	Computational Finance	Yes	Yes	Yes
transpose	Matrix transpose	Linear Algebra	Yes	Yes	Yes

10. GPGPU-Sim

Application	Description	Domain	CUDA	OpenCL	C
aes	AES algorithm in CUDA to encrypt and decrypt files	Cryptography	Yes	No	No
dc	A discontinuous Galerkin time-domain solver	Simulation	Yes	No	No
lps	3D Laplace Solver	Computational Finance	Yes	No	No
lib	Monte Carlo simulation in London-interbank-offered-rate Model	Computational Finance	Yes	No	No
mum	Pairwise local sequence alignment for DNA string	Bioinformatics	Yes	No	No
nn	Convolutional neural network to recognize handwritten digits	Machine Learning	Yes	No	No
nqu	The N-Queen solver	Simulation	Yes	No	No
ray	Ray-tracing (rendering graphics with near photo-realism)	Graphic Application	Yes	No	No
sto	Sliding-window implementation of the MD5 algorithm	Data Mining	Yes	No	No
wp	Accelerate part of the Weather Research and Forecast Model (WRF)	Simulation	Yes	No	No

Reference

https://www.findhao.net/easycoding/2304.html

GPU_benchmark说明（转）

Table of Contents