Experiments of ECC compiler

Content

• Introduction
• Configuration
• Benchmarks
• Feedback

Introduction

This repository contains all benchmarks used in our paper "An ELI-to-C Compiler: Design, Implementation and Performance". Each benchmark has at least three versions: C, ELI, and ECC. (See configuration)

There two experiments we conducted in our paper. The first experiment consists of three benchmarks: black-scholes, k-means and hotspot are from popular benchmarks, which are originally written in C. We manually translate the code to ELI code which is later compiled by the ECC.

The second experiment has another two benchmarks: morgan and rprime are inherited from APL code. APL and ELI are both array programming languages. Therefore, the two benchmarks are quite different from the above three benchmarks. We manually translate the code from ELI to C while maintain its consistency.

We used a language to implement its compiler within itself. Our compiler (ECCE) was first written in ELI. The ECCE was designed to emit C code based on the ELI interpreter. In order to provide an indepdent ELI compiler, we conducted bootstrapping to compile the ECCE to an functionally equivalent compiler in C code. Finally, an executable file was generated after compiled by the GNU GCC compiler. This compiler can compile ELI code to C code without the help of the ELI interpreter, called ECC.

Configuration

System information

• Name : Lynx
• OS : Ubuntu 16.04.2 LTS
• GCC : 5.4.0
• ELI : v0.1a download

Data download

• Experiment data [download, 138MB] (After unzipped, about 4GB)
• Source code [download]

Benchmark versions:

- [x] C       : C code
- [x] ECC     : ECC-generated C code
- [x] ELI     : ELI code
- [x] C-Opt   : Optimized C code (i.e. hotspot)
- [x] ECC-Opt : Optimized ECC-generated C code (i.e. rprime)

Benchmarks

1. black-scholes
2. k-means
3. hotspot
4. morgan
5. rprime

black-scholes

black-schole is a mathematical model of a financial market containing certain derivative investment instruments.

Source: Princeton PARSEC

k-means

k-means clustering is a method of vector quantization originally from signal processing, that is popular for cluster analysis in data mining.

Source: Rodinia 2.3

Hotspot

hotSpot is a widely used tool to estimate processor temperature based on an architectural floorplan and simulated power measurements.

Source: Rodinia 2.3

morgan

The morgan benchmark comes from a financial application. It has two functions: msum and morgan, where morgan is the main function.

@.r<-n msum a
r<-((0,n-1)!.t)-0,(0,-n)!.t<-+\a
@.

@.r<-n morgan a;x;y;sx;sx2;sy;sy2;sxy
  x<-a[1;;]
  y<-a[2;;]
  sx<-n msum x
  sy<-n msum y
  sx2<-n msum x *. 2
  sy2<-n msum y *. 2
  sxy<-n msum x *  y
  r<-((sxy%n)-(sx*sy)%n*.2)%(|((sx2%n)+(sx%n)*.2)*.0.5)*(|(sy2%n)-(sy%n)*.2)*.0.5
@.

rprime

Give an integer number n, rprime finds all prime numbers up to n.

Input scale:

100K 200K 400K 800K

Feedback

Please contact Hanfeng.