A very short tour through the Julia community and how key features of the language interact to produce an expressive syntax that users like without sacrificing performance

1. compiler and community
Jiahao Chen
MIT Computer Science and Artificial Intelligence Laboratory
julialang.org
Alan EdelmanJeff Bezanson Stefan Karpinski Viral B. Shah

2. 433 code contributors as of 2014-06-30
The faces of Julia

3. as of 2014-06-30
The world of
181 package-only developers
252 contributors to julia repo
3,906 stargazers
321 watchers

4. What’s the big deal about Julia ?
julialang.org/benchmarks

5. What’s the big deal about Julia ?
Julia isn’t designed for performance using all
means necessary.
“FFTW (FFTs) and BLAS libraries (matrix multiplication)
take ~100,000 lines [of code] to solve problems that can
be implemented in ~15 lines of (slow) code… It usually
isn’t worth it to get the last factor of two in speed”
!
- Steven Johnson, author of FFTW (emphases his)

6. What’s the big deal about Julia ?
Julia isn’t designed for performance using all
means necessary. Instead, performance comes
from designing a language that is easy for both
humans and compilers to understand.
!
Multimethods (multiple dispatch) is one of the key
mechanisms.

7. Object-oriented programming with classes
What can I do with/to a thing?

8. Object-oriented programming with classes
What can I do with/to a thing?

9. Object-oriented programming with classes
What can I do with/to a thing?
top up
pay fare
lose
buy

10. Object-oriented programming with classes
What can I do with/to a thing?
top up
pay fare
lose
buy
top up
pay fare
lose
buy

11. Object-oriented programming with classes
What can I do with/to a thing?
top up
pay fare
lose
buy
top up
pay fare
lose
buy

12. Object-oriented programming with classes
What can I do with/to a thing?
top up
pay fare
lose
buy
top up
pay fare
lose
buy
pay fare
lose
buy

13. methods objects
Object-oriented programming with classes
What can I do with/to a thing?
top up
pay fare
lose
buy
top up
pay fare
lose
buy
pay fare
lose
buy

14. methods objects
Object-oriented programming with classes
What can I do with/to a thing?
top up
pay fare
lose
buy
top up
pay fare
lose
buy
pay fare
lose
buy
class-based OO
!
classes are more
fundamental
than methods

15. generic
function
objectsmethods
rechargeable
subway
pass
single-use
subway
ticket
is a subtype of
subway
ticket
top up
pay fare
lose
buy
abstract object
Multi-methods with type hierarchy

16. generic
function
objectsmethods
rechargeable
subway
pass
single-use
subway
ticket
is a subtype of
subway
ticket
top up
pay fare
lose
buy
abstract object
Multi-methods with type hierarchy

17. Multi-methods for linear algebra
What can I do with/to a thing?
find eigenvalues and eigenvectors
find singular values
find singular values and vectors
find eigenvalues
generic
function
objectsmethods
general matrix
symmetric tridiagonal matrix
bidiagonal matrix
Methods can take advantage of special matrix structures
eigvals
eigfact
svdvals
svdfact
Matrix
SymTridiagonal
Bidiagonal

18. So how does this help us with linear algebra?
Multi-method dispatch on special matrix types

19. So how does this help us with linear algebra?
Multi-method dispatch on special matrix types

20. So how does this help us with linear algebra?
Multi-method dispatch on special matrix types
stev!{T<:BlasFloat} calls sgestv
dgestv
cgestv
zgestv
and handles workspace
memory allocation

21. What’s the big deal about Julia ?
Julia isn’t designed for performance using all
means necessary. Instead, performance comes
from designing a language that is easy for both
humans and compilers to understand.
!
!

22. from the programmer’s point of view:
!
- dynamic language semantics: faster to prototype code
- dynamic types: compiler does not hinder experimentation
- multiple dispatch allows both methods for general
functionality and methods for specialized performance
- low-cost software abstractions allow code to retain
structure of mathematical abstractions
- metaprogramming (with macros) allows code generation:
takes software experimentation to the next level with less
lines of actual code
- flexible syntax, Unicode operators, great community…

23. Pure-­‐Julia
FFT
performance2
4
8
16
32
64
128
256
512
1024
2048
4096
8192
16384
32768
65536
131072
2621440
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
11000
12000
13000
14000
speed(mflops)
intel-mkl-dfti in-place
intel-mkl-dfti out-of-place
fftw3 out-of-place
fftw3 in-place
fftw3-no-simd out-of-place
fftw3-no-simd in-place
dfftpack
emayer
julia
bloodworth
cross
cwplib
esrfft
double-precision complex, 1d transforms
powers of two
already
comparable
to
FFTPACK
!
[
actually
was
even
a
bit
better;
some
recent
inlining
regressions
in
Julia
snapshots
]
Steven Johnson, 2014-06
Presented at NAIS Codegen workshop

24. from the compiler’s point of view:
!
- dynamic semantics with static analysis: when possible,
offload runtime checks to JIT compile stage
- multiple dispatch requires function signatures that expose
type information
- rich type system allows for detailed type inference
- inlining passes reduce abstraction cost
- automatic code specialization generates many performant
methods
- macros allow further fine-grained tradeoffs between
performance and correctness, e.g. @inbounds, @simd,
…