ELI: Programming with Arrays Hanfeng Chen1 and Wai-Mee Ching2 1 [email protected] 2 [email protected] https://fastarray.appspot.com In this talk: ELI • What is ELI? • An array programming language^1 • Based on APL and influenced by KDB^2 • Support classic APL1 (flat arrays) and database components • Why is ELI? • No variable declarations • Fast programming, simple syntax, and parallel semantics • Easy to learn and easy to communicate! ^1 ELI homepage: https://fastarray.appspot.com ^2 KX company: https://www.kx.com Page 1 Programming with ELI 2 Introduction • ELI language • An ASCII version of APL • Monadic (unary) and dyadic (binary) primitive functions • Equal precedence of all primitive functions • Right-to-left execution order Page 3 ELI Types • Array types (Homogeneous, multi-dimensional) • Boolean (0 or 1), integer, floating-point number • Complex number (e.g. 2j3 for 2+3i) • Character (e.g. “tea”) • Symbol (e.g. `bubble) • List (Non-homogeneous data) • Database-related types • Date and time • Dictionary and tables Page 4 ELI Examples - I • Generating a vector ELI functions !10 // APL: �10 ! : iota 1 2 3 4 5 6 7 8 9 10 * : multiplication • Multiplying the vector by 100 # : reshape 100 * !10 // APL: 100 × �10 100 200 300 400 500 600 700 800 900 1000 • Generating a 3-by-4 matrix 3 4#!10 // APL: 3 4 � �10 1 2 3 4 5 6 7 8 9 10 1 2 Page 5 ELI Examples - II • Complex number ELI functions @1 // � ! : iota 3.141592654 * : multiplication 0j1*@1 // 0 + �� *.: f(x)= �) 0j3.141592654 @ : pi *.0j1*@1 // �-. _1 • Date and time 2020.11.24 + !7 // Next 7 days 2020.11.25 2020.11.26 2020.11.27 2020.11.28 2020.11.29 2020.11.30 2020.12.01 15:46 + 15 // Next 15 minutes 16:01 Page 6 ELI Examples - III • A table of temperature conversion (C to F) ELI functions • T = T × 9/5 + 32 (°F) (°C) ! : iota c,[1.5]32+1.8*c<-$_10+5*!10 * : multiplication 40 104 $ : rotate 35 95 <-: assign 30 86 ,[1.5]: laminate 25 77 20 68 15 59 10 50 5 41 0 32 _5 23 (C) (F) Page 7 ELI Examples - IV • List of numbers, symbols, and chars ELI functions (2 3#!6;`ny `ma `md;'ABCDE’) ! : iota <1 2 3 # : reshape 4 5 6 <`ny `ma `md <ABCDE • Dictionary and table D<-(`sym `price `hq:(`appl `ibm `hp `goog;449.1 108.2 24.5 890.3;4 2#'CANYCACA’)) D sym | appl ibm hp goog price| 449.1 108.2 24.5 890.3 hq | 'CANYCACA' Page 8 ELI Examples - V • Dictionary to table ELI functions []<-T<-&.D &.: flip sym price hq ------------- do: written in ELI appl 449.1 CA ibm 108.2 NY hp 24.5 CA goog 890.3 CA • Database query do 'SELECT sym,hq FROM T WHERE price>100’ sym hq ------- appl CA Ibm NY goog CA SELECT successful. Page 9 ELI Symbols • ELI use one or two characters to represent one APL symbol (Unicode) • E.g., using ? to represent the member function (∈ in APL) • The second character usually is dot (.) • E.g., using ?. to represent the random function (? in APL) • In other cases, the second character is intuitive • E.g., using <- to represent the assign function (← in APL) Page 10 APL ELI ELI Symbols (Overview) Source: ELI Homepage https://fastarray.appspot.com Page 11 ELI Tutorials • ELI Primer • A detailed document for learning ELI programming • ELI for kids • A novel way to learn math and coding • Introduction to Programming with Arrays using ELI • Paperback available on Amazon.com Source: https://fastarray.appspot.com/documents.html Page 12 ELI Studio IDE for ELI programming 13 ELI Studio • ELI Studio provides a GUI interface for programming ELI • Interactive programming environment • Write and load ELI scripts conveniently • Manage variables and libraries easily • Cross-platform support • Windows • Linux • Mac OS Source: https://fastarray.appspot.com/download.html Page 14 ELI Studio (v0.3) Page 15 Programming in ELI Studio • Standard input []<-’Please enter your name:’ Please enter your name: []<-’Welcome ‘,[] Welcome hanfeng • File handle h<-[]open ‘file1.txt’ @1 3.141592654 h<<@1 h<<3 5#!15 // 3-by-5 matrix []close h 0 Page 16 ELI Compiler Compile ELI programs 17 ECC: ELI-to-C Compiler • A compiler executable • Modeled on an APL-to-C compiler written in APL (operates under an APL interpreter) • Written in ELI and self-compiled into a binary executable^1 • Support ELI built-in functions derived from APL1 • Homogeneous array types Source Code Target Code Binary (ELI) ECC (C) GCC Library ^1 Hanfeng Chen, Wai-Mee Ching, and Laurie Hendren, An ELI-to-C Compiler: Design, Implementation, and Performance, Proceedings of the 4th ACM SIGPLAN International Workshop of Libraries, Languages and Compilers for Array Programming (ARRAY'17), pp. 9-16, June 2017. Page 18 Benchmark: morgan (I) Main function morgan (n is a scalar and a is a 2-M-N array) [0] @.r<-n morgan a;x;y;sx;sx2;sy;sy2;sxy // function: morgan(n,a) [1] x<-a[1;;] [2] y<-a[2;;] [3] sx<-n msum x // sx = msum(n, x) [4] sy<-n msum y // sy = msum(n, y) [5] sx2<-n msum x *. 2 // sx2 = msum(n, x^2) [6] sy2<-n msum y *. 2 // sy2 = msum(n, y^2) [7] sxy<-n msum x * y // sxy = msum(n, x*y) [8] r<-((sxy%n)-(sx*sy)%n*.2)%(|((sx2%n)+(sx%n)*.2)*.0.5)* (|(sy2%n)-(sy%n)*.2)*.0.5 [9] @. Page 19 Benchmark: morgan (II) Function msum (n is a scalar and a is a two-dim. array) [0] @.r<-n msum a // function: msum(n,a) [1] r<-((0,n-1)!.t)-0,(0,-n)!.t <- +\a // | Part3 | Part2 | Part1| [2] @. In steps, • Let the matrix a have P rows and Q columns 7 • Part 1: +\a computes � � � = ∑6 � � [�], given � ∈ [0, �) and � ∈ 0, � • Part 2: 0,(0,-n)!.t cuts n tailing columns, adds a leading column (all 0s) and returns the matrix M0 • Part 3: (0,n-1)!.t cuts the first n-1 columns and returns the matrix M1 • Finally, the return value r gets M1-M0 Page 20 Compilation • Parameter information for compiling the function morgan • Left parameter: morgan(integer, float) &LPARM C 1 9 ‘morgan IE’ & • Right parameter: starting from 0, morgan(salar, 3-dim) &RPARM I 1 6 0 0 3 _1 _1 _1 & • Save functions and parameters to an ELI script file, called cmain.esf • Run ./ecc cmain to generate a C file, morgan.c Page 21 Conclusions & Future Work 22 Conclusions • ELI is a succinct modern array language derived from APL • ELI inherits the classic APL1 and borrows database features from KDB • ELI provides an interpreter with a GUI-based IDE (ELI Studio) and an compiler executable for flat array programs Page 23 Future Plans Performance enhancement • Build GPU-based library functions for the ELI interpreter • Add optimizations to the ELI compiler • Generate parallel code for both CPUs and GPUs Feature improvement • Extend the current compiler to support database-related features • Add more features into ELI Studio to help ELI programming • Code highlighting / automatic completion / etc. Page 24 Publications • Hanfeng Chen, Wai-Mee Ching, and Laurie Hendren, An ELI-to-C Compiler: Design, Implementation, and Performance, Proceedings of the 4th ACM SIGPLAN International Workshop of Libraries, Languages and Compilers for Array Programming (ARRAY’17), pp. 9-16, June 2017. • Hanfeng Chen and Wai-Mee Ching, A Succinct Programming Language with a Native Database Component, Int'l Jour. of Prog. Lang. and Appl.(IJPLA) vol.7, No.1, Jan.2017. • Hanfeng Chen, Wai-Mee Ching and Da Zheng, A comparison study on execution performance of MATLAB and APL, accepted to Vector J. Br. APL Assoc. • Hanfeng Chen and Wai-Mee Ching, ELI: a simple system for array programming, Vector J. Br. APL Assoc. Vol.26, No.1, p94-102, 2013. • Wai-Mee Ching, The design and implementation of an APL dialect, ELI, Proceedings of the APL-Berlin-2000 Conference, p69-76, July 2000. Source: https://fastarray.appspot.com/documents.html Page 25 Thank you ELI: Programming with Arrays Hanfeng Chen and Wai-Mee Ching https://fastarray.appspot.com.