source: doc/papers/concurrency/Paper.tex @ e98c7ab

ADTarm-ehast-experimentalenumforall-pointer-decayjacob/cs343-translationjenkins-sandboxnew-astnew-ast-unique-exprpthread-emulationqualifiedEnum
Last change on this file since e98c7ab was e98c7ab, checked in by Thierry Delisle <tdelisle@…>, 5 years ago

Passed spell checker on the paper, it had a hard time with latex so it's not perfect

  • Property mode set to 100644
File size: 155.8 KB
Line 
1\documentclass[AMA,STIX1COL]{WileyNJD-v2}
2
3\articletype{RESEARCH ARTICLE}%
4
5\received{XXXXX}
6\revised{XXXXX}
7\accepted{XXXXX}
8
9\raggedbottom
10
11%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
12
13% Latex packages used in the document.
14
15\usepackage{epic,eepic}
16\usepackage{xspace}
17\usepackage{enumitem}
18\usepackage{comment}
19\usepackage{upquote}                                            % switch curled `'" to straight
20\usepackage{listings}                                           % format program code
21\usepackage[labelformat=simple,aboveskip=0pt,farskip=0pt]{subfig}
22\renewcommand{\thesubfigure}{(\Alph{subfigure})}
23\captionsetup{justification=raggedright,singlelinecheck=false}
24\usepackage{dcolumn}                                            % align decimal points in tables
25\usepackage{capt-of}
26\setlength{\multicolsep}{6.0pt plus 2.0pt minus 1.5pt}
27
28\hypersetup{breaklinks=true}
29\definecolor{OliveGreen}{cmyk}{0.64 0 0.95 0.40}
30\definecolor{Mahogany}{cmyk}{0 0.85 0.87 0.35}
31\definecolor{Plum}{cmyk}{0.50 1 0 0}
32
33\usepackage[pagewise]{lineno}
34\renewcommand{\linenumberfont}{\scriptsize\sffamily}
35
36\renewcommand{\topfraction}{0.8}                        % float must be greater than X of the page before it is forced onto its own page
37\renewcommand{\bottomfraction}{0.8}                     % float must be greater than X of the page before it is forced onto its own page
38\renewcommand{\floatpagefraction}{0.8}          % float must be greater than X of the page before it is forced onto its own page
39\renewcommand{\textfraction}{0.0}                       % the entire page maybe devoted to floats with no text on the page at all
40
41\lefthyphenmin=3                                                        % hyphen only after 4 characters
42\righthyphenmin=3
43
44%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
45
46% Names used in the document.
47
48\newcommand{\CFAIcon}{\textsf{C}\raisebox{\depth}{\rotatebox{180}{\textsf{A}}}\xspace} % Cforall symbolic name
49\newcommand{\CFA}{\protect\CFAIcon}             % safe for section/caption
50\newcommand{\CFL}{\textrm{Cforall}\xspace}      % Cforall symbolic name
51\newcommand{\Celeven}{\textrm{C11}\xspace}      % C11 symbolic name
52\newcommand{\CC}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}\xspace} % C++ symbolic name
53\newcommand{\CCeleven}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}11\xspace} % C++11 symbolic name
54\newcommand{\CCfourteen}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}14\xspace} % C++14 symbolic name
55\newcommand{\CCseventeen}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}17\xspace} % C++17 symbolic name
56\newcommand{\CCtwenty}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}20\xspace} % C++20 symbolic name
57\newcommand{\Csharp}{C\raisebox{-0.7ex}{\Large$^\sharp$}\xspace} % C# symbolic name
58
59%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
60
61\newcommand{\Textbf}[2][red]{{\color{#1}{\textbf{#2}}}}
62\newcommand{\Emph}[2][red]{{\color{#1}\textbf{\emph{#2}}}}
63\newcommand{\uC}{$\mu$\CC}
64\newcommand{\TODO}[1]{{\Textbf{#1}}}
65
66%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
67
68% Default underscore is too low and wide. Cannot use lstlisting "literate" as replacing underscore
69% removes it as a variable-name character so keywords in variables are highlighted. MUST APPEAR
70% AFTER HYPERREF.
71%\DeclareTextCommandDefault{\textunderscore}{\leavevmode\makebox[1.2ex][c]{\rule{1ex}{0.1ex}}}
72\renewcommand{\textunderscore}{\leavevmode\makebox[1.2ex][c]{\rule{1ex}{0.075ex}}}
73
74\renewcommand*{\thefootnote}{\Alph{footnote}} % hack because fnsymbol does not work
75%\renewcommand*{\thefootnote}{\fnsymbol{footnote}}
76
77\makeatletter
78% parindent is relative, i.e., toggled on/off in environments like itemize, so store the value for
79% use rather than use \parident directly.
80\newlength{\parindentlnth}
81\setlength{\parindentlnth}{\parindent}
82
83\newcommand{\LstBasicStyle}[1]{{\lst@basicstyle{\lst@basicstyle{#1}}}}
84\newcommand{\LstKeywordStyle}[1]{{\lst@basicstyle{\lst@keywordstyle{#1}}}}
85\newcommand{\LstCommentStyle}[1]{{\lst@basicstyle{\lst@commentstyle{#1}}}}
86
87\newlength{\gcolumnposn}                                        % temporary hack because lstlisting does not handle tabs correctly
88\newlength{\columnposn}
89\setlength{\gcolumnposn}{3.5in}
90\setlength{\columnposn}{\gcolumnposn}
91
92\newcommand{\C}[2][\@empty]{\ifx#1\@empty\else\global\setlength{\columnposn}{#1}\global\columnposn=\columnposn\fi\hfill\makebox[\textwidth-\columnposn][l]{\lst@basicstyle{\LstCommentStyle{#2}}}}
93\newcommand{\CRT}{\global\columnposn=\gcolumnposn}
94
95% Denote newterms in particular font and index them without particular font and in lowercase, e.g., \newterm{abc}.
96% The option parameter provides an index term different from the new term, e.g., \newterm[\texttt{abc}]{abc}
97% The star version does not lowercase the index information, e.g., \newterm*{IBM}.
98\newcommand{\newtermFontInline}{\emph}
99\newcommand{\newterm}{\@ifstar\@snewterm\@newterm}
100\newcommand{\@newterm}[2][\@empty]{\lowercase{\def\temp{#2}}{\newtermFontInline{#2}}\ifx#1\@empty\index{\temp}\else\index{#1@{\protect#2}}\fi}
101\newcommand{\@snewterm}[2][\@empty]{{\newtermFontInline{#2}}\ifx#1\@empty\index{#2}\else\index{#1@{\protect#2}}\fi}
102
103% Latin abbreviation
104\newcommand{\abbrevFont}{\textit}                       % set empty for no italics
105\@ifundefined{eg}{
106\newcommand{\EG}{\abbrevFont{e}\abbrevFont{g}}
107\newcommand*{\eg}{%
108        \@ifnextchar{,}{\EG}%
109                {\@ifnextchar{:}{\EG}%
110                        {\EG,\xspace}}%
111}}{}%
112\@ifundefined{ie}{
113\newcommand{\IE}{\abbrevFont{i}\abbrevFont{e}}
114\newcommand*{\ie}{%
115        \@ifnextchar{,}{\IE}%
116                {\@ifnextchar{:}{\IE}%
117                        {\IE,\xspace}}%
118}}{}%
119\@ifundefined{etc}{
120\newcommand{\ETC}{\abbrevFont{etc}}
121\newcommand*{\etc}{%
122        \@ifnextchar{.}{\ETC}%
123        {\ETC.\xspace}%
124}}{}%
125\@ifundefined{etal}{
126\newcommand{\ETAL}{\abbrevFont{et}~\abbrevFont{al}}
127\newcommand*{\etal}{%
128        \@ifnextchar{.}{\protect\ETAL}%
129                {\protect\ETAL.\xspace}%
130}}{}%
131\@ifundefined{viz}{
132\newcommand{\VIZ}{\abbrevFont{viz}}
133\newcommand*{\viz}{%
134        \@ifnextchar{.}{\VIZ}%
135                {\VIZ.\xspace}%
136}}{}%
137\makeatother
138
139\newenvironment{cquote}
140               {\list{}{\lstset{resetmargins=true,aboveskip=0pt,belowskip=0pt}\topsep=3pt\parsep=0pt\leftmargin=\parindentlnth\rightmargin\leftmargin}%
141                \item\relax}
142               {\endlist}
143
144%\newenvironment{cquote}{%
145%\list{}{\lstset{resetmargins=true,aboveskip=0pt,belowskip=0pt}\topsep=3pt\parsep=0pt\leftmargin=\parindentlnth\rightmargin\leftmargin}%
146%\item\relax%
147%}{%
148%\endlist%
149%}% cquote
150
151% CFA programming language, based on ANSI C (with some gcc additions)
152\lstdefinelanguage{CFA}[ANSI]{C}{
153        morekeywords={
154                _Alignas, _Alignof, __alignof, __alignof__, asm, __asm, __asm__, __attribute, __attribute__,
155                auto, _Bool, catch, catchResume, choose, _Complex, __complex, __complex__, __const, __const__,
156                coroutine, disable, dtype, enable, exception, __extension__, fallthrough, fallthru, finally,
157                __float80, float80, __float128, float128, forall, ftype, generator, _Generic, _Imaginary, __imag, __imag__,
158                inline, __inline, __inline__, __int128, int128, __label__, monitor, mutex, _Noreturn, one_t, or,
159                otype, restrict, __restrict, __restrict__, __signed, __signed__, _Static_assert, thread,
160                _Thread_local, throw, throwResume, timeout, trait, try, ttype, typeof, __typeof, __typeof__,
161                virtual, __volatile, __volatile__, waitfor, when, with, zero_t},
162        moredirectives={defined,include_next},
163        % replace/adjust listing characters that look bad in sanserif
164        literate={-}{\makebox[1ex][c]{\raisebox{0.4ex}{\rule{0.8ex}{0.1ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptstyle\land\,$}}1
165                {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
166                {<}{\textrm{\textless}}1 {>}{\textrm{\textgreater}}1
167                {<-}{$\leftarrow$}2 {=>}{$\Rightarrow$}2 {->}{\makebox[1ex][c]{\raisebox{0.5ex}{\rule{0.8ex}{0.075ex}}}\kern-0.2ex{\textrm{\textgreater}}}2,
168}
169
170\lstset{
171language=CFA,
172columns=fullflexible,
173basicstyle=\linespread{0.9}\sf,                                                 % reduce line spacing and use sanserif font
174stringstyle=\tt,                                                                                % use typewriter font
175tabsize=5,                                                                                              % N space tabbing
176xleftmargin=\parindentlnth,                                                             % indent code to paragraph indentation
177%mathescape=true,                                                                               % LaTeX math escape in CFA code $...$
178escapechar=\$,                                                                                  % LaTeX escape in CFA code
179keepspaces=true,                                                                                %
180showstringspaces=false,                                                                 % do not show spaces with cup
181showlines=true,                                                                                 % show blank lines at end of code
182aboveskip=4pt,                                                                                  % spacing above/below code block
183belowskip=3pt,
184moredelim=**[is][\color{red}]{`}{`},
185}% lstset
186
187% uC++ programming language, based on ANSI C++
188\lstdefinelanguage{uC++}[ANSI]{C++}{
189        morekeywords={
190                _Accept, _AcceptReturn, _AcceptWait, _Actor, _At, _CatchResume, _Cormonitor, _Coroutine, _Disable,
191                _Else, _Enable, _Event, _Finally, _Monitor, _Mutex, _Nomutex, _PeriodicTask, _RealTimeTask,
192                _Resume, _Select, _SporadicTask, _Task, _Timeout, _When, _With, _Throw},
193}
194\lstdefinelanguage{Golang}{
195        morekeywords=[1]{package,import,func,type,struct,return,defer,panic,recover,select,var,const,iota,},
196        morekeywords=[2]{string,uint,uint8,uint16,uint32,uint64,int,int8,int16,int32,int64,
197                bool,float32,float64,complex64,complex128,byte,rune,uintptr, error,interface},
198        morekeywords=[3]{map,slice,make,new,nil,len,cap,copy,close,true,false,delete,append,real,imag,complex,chan,},
199        morekeywords=[4]{for,break,continue,range,goto,switch,case,fallthrough,if,else,default,},
200        morekeywords=[5]{Println,Printf,Error,},
201        sensitive=true,
202        morecomment=[l]{//},
203        morecomment=[s]{/*}{*/},
204        morestring=[b]',
205        morestring=[b]",
206        morestring=[s]{`}{`},
207}
208
209% Go programming language: https://github.com/julienc91/listings-golang/blob/master/listings-golang.sty
210\lstdefinelanguage{Golang}{
211        morekeywords=[1]{package,import,func,type,struct,return,defer,panic,recover,select,var,const,iota,},
212        morekeywords=[2]{string,uint,uint8,uint16,uint32,uint64,int,int8,int16,int32,int64,
213                bool,float32,float64,complex64,complex128,byte,rune,uintptr, error,interface},
214        morekeywords=[3]{map,slice,make,new,nil,len,cap,copy,close,true,false,delete,append,real,imag,complex,chan,},
215        morekeywords=[4]{for,break,continue,range,goto,switch,case,fallthrough,if,else,default,},
216        morekeywords=[5]{Println,Printf,Error,},
217        sensitive=true,
218        morecomment=[l]{//},
219        morecomment=[s]{/*}{*/},
220        morestring=[b]',
221        morestring=[b]",
222        morestring=[s]{`}{`},
223        % replace/adjust listing characters that look bad in sanserif
224        literate={-}{\makebox[1ex][c]{\raisebox{0.4ex}{\rule{0.8ex}{0.1ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptstyle\land\,$}}1
225                {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
226                {<}{\textrm{\textless}}1 {>}{\textrm{\textgreater}}1
227                {<-}{\makebox[2ex][c]{\textrm{\textless}\raisebox{0.5ex}{\rule{0.8ex}{0.075ex}}}}2,
228}
229
230\lstnewenvironment{cfa}[1][]
231{\lstset{#1}}
232{}
233\lstnewenvironment{C++}[1][]                            % use C++ style
234{\lstset{language=C++,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
235{}
236\lstnewenvironment{uC++}[1][]
237{\lstset{#1}}
238{}
239\lstnewenvironment{Go}[1][]
240{\lstset{language=Golang,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
241{}
242\lstnewenvironment{python}[1][]
243{\lstset{language=python,moredelim=**[is][\protect\color{red}]{`}{`},#1}\lstset{#1}}
244{}
245
246% inline code @...@
247\lstMakeShortInline@%
248
249\let\OLDthebibliography\thebibliography
250\renewcommand\thebibliography[1]{
251  \OLDthebibliography{#1}
252  \setlength{\parskip}{0pt}
253  \setlength{\itemsep}{4pt plus 0.3ex}
254}
255
256\newbox\myboxA
257\newbox\myboxB
258\newbox\myboxC
259\newbox\myboxD
260
261\title{\texorpdfstring{Advanced Control-flow and Concurrency in \protect\CFA}{Advanced Control-flow in Cforall}}
262
263\author[1]{Thierry Delisle}
264\author[1]{Peter A. Buhr*}
265\authormark{DELISLE \textsc{et al.}}
266
267\address[1]{\orgdiv{Cheriton School of Computer Science}, \orgname{University of Waterloo}, \orgaddress{\state{Waterloo, ON}, \country{Canada}}}
268
269\corres{*Peter A. Buhr, Cheriton School of Computer Science, University of Waterloo, 200 University Avenue West, Waterloo, ON, N2L 3G1, Canada. \email{pabuhr{\char`\@}uwaterloo.ca}}
270
271% \fundingInfo{Natural Sciences and Engineering Research Council of Canada}
272
273\abstract[Summary]{
274\CFA is a polymorphic, non-object-oriented, concurrent, backwards-compatible extension of the C programming language.
275This paper discusses the design philosophy and implementation of its advanced control-flow and concurrent/parallel features, along with the supporting runtime written in \CFA.
276These features are created from scratch as ISO C has only low-level and/or unimplemented concurrency, so C programmers continue to rely on library features like pthreads.
277\CFA introduces modern language-level control-flow mechanisms, like generators, coroutines, user-level threading, and monitors for mutual exclusion and synchronization.
278% Library extension for executors, futures, and actors are built on these basic mechanisms.
279The runtime provides significant programmer simplification and safety by eliminating spurious wakeup and monitor barging.
280The runtime also ensures multiple monitors can be safely acquired \emph{simultaneously} (deadlock free), and this feature is fully integrated with all monitor synchronization mechanisms.
281All control-flow features integrate with the \CFA polymorphic type-system and exception handling, while respecting the expectations and style of C programmers.
282Experimental results show comparable performance of the new features with similar mechanisms in other concurrent programming languages.
283}%
284
285\keywords{generator, coroutine, concurrency, parallelism, thread, monitor, runtime, C, \CFA (Cforall)}
286
287
288\begin{document}
289\linenumbers                                            % comment out to turn off line numbering
290
291\maketitle
292
293
294\section{Introduction}
295
296This paper discusses the design philosophy and implementation of advanced language-level control-flow and concurrent/parallel features in \CFA~\cite{Moss18,Cforall} and its runtime, which is written entirely in \CFA.
297\CFA is a modern, polymorphic, non-object-oriented\footnote{
298\CFA has features often associated with object-oriented programming languages, such as constructors, destructors, virtuals and simple inheritance.
299However, functions \emph{cannot} be nested in structures, so there is no lexical binding between a structure and set of functions (member/method) implemented by an implicit \lstinline@this@ (receiver) parameter.},
300backwards-compatible extension of the C programming language.
301In many ways, \CFA is to C as Scala~\cite{Scala} is to Java, providing a \emph{research vehicle} for new typing and control-flow capabilities on top of a highly popular programming language allowing immediate dissemination.
302Within the \CFA framework, new control-flow features are created from scratch because ISO \Celeven defines only a subset of the \CFA extensions, where the overlapping features are concurrency~\cite[\S~7.26]{C11}.
303However, \Celeven concurrency is largely wrappers for a subset of the pthreads library~\cite{Butenhof97,Pthreads}, and \Celeven and pthreads concurrency is simple, based on thread fork/join in a function and a few locks, which is low-level and error-prone;
304no high-level language concurrency features are defined.
305Interestingly, almost a decade after publication of the \Celeven standard, neither gcc-8, clang-9 nor msvc-19 (most recent versions) support the \Celeven include @threads.h@, indicating little interest in the C11 concurrency approach.
306Finally, while the \Celeven standard does not state a threading model, the historical association with pthreads suggests implementations would adopt kernel-level threading (1:1)~\cite{ThreadModel}.
307
308In contrast, there has been a renewed interest during the past decade in user-level (M:N, green) threading in old and new programming languages.
309As multi-core hardware became available in the 1980/90s, both user and kernel threading were examined.
310Kernel threading was chosen, largely because of its simplicity and fit with the simpler operating systems and hardware architectures at the time, which gave it a performance advantage~\cite{Drepper03}.
311Libraries like pthreads were developed for C, and the Solaris operating-system switched from user (JDK 1.1~\cite{JDK1.1}) to kernel threads.
312As a result, languages like Java, Scala, Objective-C~\cite{obj-c-book}, \CCeleven~\cite{C11}, and C\#~\cite{Csharp} adopt the 1:1 kernel-threading model, with a variety of presentation mechanisms.
313From 2000 onwards, languages like Go~\cite{Go}, Erlang~\cite{Erlang}, Haskell~\cite{Haskell}, D~\cite{D}, and \uC~\cite{uC++,uC++book} have championed the M:N user-threading model, and many user-threading libraries have appeared~\cite{Qthreads,MPC,Marcel}, including putting green threads back into Java~\cite{Quasar}.
314The main argument for user-level threading is that they are lighter weight than kernel threads (locking and context switching do not cross the kernel boundary), so there is less restriction on programming styles that encourage large numbers of threads performing medium work units to facilitate load balancing by the runtime~\cite{Verch12}.
315As well, user-threading facilitates a simpler concurrency approach using thread objects that leverage sequential patterns versus events with call-backs~\cite{Adya02,vonBehren03}.
316Finally, performant user-threading implementations (both time and space) meet or exceed direct kernel-threading implementations, while achieving the programming advantages of high concurrency levels and safety.
317
318A further effort over the past two decades is the development of language memory models to deal with the conflict between language features and compiler/hardware optimizations, \ie, some language features are unsafe in the presence of aggressive sequential optimizations~\cite{Buhr95a,Boehm05}.
319The consequence is that a language must provide sufficient tools to program around safety issues, as inline and library code is all sequential to the compiler.
320One solution is low-level qualifiers and functions (\eg, @volatile@ and atomics) allowing \emph{programmers} to explicitly write safe (race-free~\cite{Boehm12}) programs.
321A safer solution is high-level language constructs so the \emph{compiler} knows the optimization boundaries, and hence, provides implicit safety.
322This problem is best known with respect to concurrency, but applies to other complex control-flow, like exceptions\footnote{
323\CFA exception handling will be presented in a separate paper.
324The key feature that dovetails with this paper is nonlocal exceptions allowing exceptions to be raised across stacks, with synchronous exceptions raised among coroutines and asynchronous exceptions raised among threads, similar to that in \uC~\cite[\S~5]{uC++}
325} and coroutines.
326Finally, language solutions allow matching constructs with language paradigm, \ie, imperative and functional languages often have different presentations of the same concept to fit their programming model.
327
328Finally, it is important for a language to provide safety over performance \emph{as the default}, allowing careful reduction of safety for performance when necessary.
329Two concurrency violations of this philosophy are \emph{spurious wakeup} (random wakeup~\cite[\S~8]{Buhr05a}) and \emph{barging} (signals-as-hints~\cite[\S~8]{Buhr05a}), where one is a consequence of the other, \ie, once there is spurious wakeup, signals-as-hints follow.
330However, spurious wakeup is \emph{not} a foundational concurrency property~\cite[\S~8]{Buhr05a}, it is a performance design choice.
331Similarly, signals-as-hints are often a performance decision.
332We argue removing spurious wakeup and signals-as-hints make concurrent programming significantly safer because it removes local non-determinism and matches with programmer expectation.
333(Author experience teaching concurrency is that students are highly confused by these semantics.)
334Clawing back performance, when local non-determinism is unimportant, should be an option not the default.
335
336\begin{comment}
337Most augmented traditional (Fortran 18~\cite{Fortran18}, Cobol 14~\cite{Cobol14}, Ada 12~\cite{Ada12}, Java 11~\cite{Java11}) and new languages (Go~\cite{Go}, Rust~\cite{Rust}, and D~\cite{D}), except \CC, diverge from C with different syntax and semantics, only interoperate indirectly with C, and are not systems languages, for those with managed memory.
338As a result, there is a significant learning curve to move to these languages, and C legacy-code must be rewritten.
339While \CC, like \CFA, takes an evolutionary approach to extend C, \CC's constantly growing complex and interdependent features-set (\eg, objects, inheritance, templates, etc.) mean idiomatic \CC code is difficult to use from C, and C programmers must expend significant effort learning \CC.
340Hence, rewriting and retraining costs for these languages, even \CC, are prohibitive for companies with a large C software-base.
341\CFA with its orthogonal feature-set, its high-performance runtime, and direct access to all existing C libraries circumvents these problems.
342\end{comment}
343
344\CFA embraces user-level threading, language extensions for advanced control-flow, and safety as the default.
345We present comparative examples so the reader can judge if the \CFA control-flow extensions are better and safer than those in other concurrent, imperative programming languages, and perform experiments to show the \CFA runtime is competitive with other similar mechanisms.
346The main contributions of this work are:
347\begin{itemize}
348\item
349language-level generators, coroutines and user-level threading, which respect the expectations of C programmers.
350\item
351monitor synchronization without barging, and the ability to safely acquiring multiple monitors \emph{simultaneously} (deadlock free), while seamlessly integrating these capabilities with all monitor synchronization mechanisms.
352\item
353providing statically type-safe interfaces that integrate with the \CFA polymorphic type-system and other language features.
354% \item
355% library extensions for executors, futures, and actors built on the basic mechanisms.
356\item
357a runtime system with no spurious wakeup.
358\item
359a dynamic partitioning mechanism to segregate the execution environment for specialized requirements.
360% \item
361% a non-blocking I/O library
362\item
363experimental results showing comparable performance of the new features with similar mechanisms in other programming languages.
364\end{itemize}
365
366
367\section{Stateful Function}
368
369The stateful function is an old idea~\cite{Conway63,Marlin80} that is new again~\cite{C++20Coroutine19}, where execution is temporarily suspended and later resumed, \eg, plugin, device driver, finite-state machine.
370Hence, a stateful function may not end when it returns to its caller, allowing it to be restarted with the data and execution location present at the point of suspension.
371This capability is accomplished by retaining a data/execution \emph{closure} between invocations.
372If the closure is fixed size, we call it a \emph{generator} (or \emph{stackless}), and its control flow is restricted, \eg, suspending outside the generator is prohibited.
373If the closure is variably sized, we call it a \emph{coroutine} (or \emph{stackful}), and as the names implies, often implemented with a separate stack with no programming restrictions.
374Hence, refactoring a stackless coroutine may require changing it to stackful.
375A foundational property of all \emph{stateful functions} is that resume/suspend \emph{do not} cause incremental stack growth, \ie, resume/suspend operations are remembered through the closure not the stack.
376As well, activating a stateful function is \emph{asymmetric} or \emph{symmetric}, identified by resume/suspend (no cycles) and resume/resume (cycles).
377A fixed closure activated by modified call/return is faster than a variable closure activated by context switching.
378Additionally, any storage management for the closure (especially in unmanaged languages, \ie, no garbage collection) must also be factored into design and performance.
379Therefore, selecting between stackless and stackful semantics is a tradeoff between programming requirements and performance, where stackless is faster and stackful is more general.
380Note, creation cost is amortized across usage, so activation cost is usually the dominant factor.
381
382\begin{figure}
383\centering
384\begin{lrbox}{\myboxA}
385\begin{cfa}[aboveskip=0pt,belowskip=0pt]
386typedef struct {
387        int fn1, fn;
388} Fib;
389#define FibCtor { 1, 0 }
390int fib( Fib * f ) {
391
392
393
394        int fn = f->fn; f->fn = f->fn1;
395                f->fn1 = f->fn + fn;
396        return fn;
397
398}
399int main() {
400        Fib f1 = FibCtor, f2 = FibCtor;
401        for ( int i = 0; i < 10; i += 1 )
402                printf( "%d %d\n",
403                           fib( &f1 ), fib( &f2 ) );
404}
405\end{cfa}
406\end{lrbox}
407
408\begin{lrbox}{\myboxB}
409\begin{cfa}[aboveskip=0pt,belowskip=0pt]
410`generator` Fib {
411        int fn1, fn;
412};
413
414void `main(Fib & fib)` with(fib) {
415
416        [fn1, fn] = [1, 0];
417        for () {
418                `suspend;`
419                [fn1, fn] = [fn, fn + fn1];
420
421        }
422}
423int main() {
424        Fib f1, f2;
425        for ( 10 )
426                sout | `resume( f1 )`.fn
427                         | `resume( f2 )`.fn;
428}
429\end{cfa}
430\end{lrbox}
431
432\begin{lrbox}{\myboxC}
433\begin{cfa}[aboveskip=0pt,belowskip=0pt]
434typedef struct {
435        int fn1, fn;  void * `next`;
436} Fib;
437#define FibCtor { 1, 0, NULL }
438Fib * comain( Fib * f ) {
439        if ( f->next ) goto *f->next;
440        f->next = &&s1;
441        for ( ;; ) {
442                return f;
443          s1:; int fn = f->fn + f->fn1;
444                        f->fn1 = f->fn; f->fn = fn;
445        }
446}
447int main() {
448        Fib f1 = FibCtor, f2 = FibCtor;
449        for ( int i = 0; i < 10; i += 1 )
450                printf("%d %d\n",comain(&f1)->fn,
451                                 comain(&f2)->fn);
452}
453\end{cfa}
454\end{lrbox}
455
456\subfloat[C asymmetric generator]{\label{f:CFibonacci}\usebox\myboxA}
457\hspace{3pt}
458\vrule
459\hspace{3pt}
460\subfloat[\CFA asymmetric generator]{\label{f:CFAFibonacciGen}\usebox\myboxB}
461\hspace{3pt}
462\vrule
463\hspace{3pt}
464\subfloat[C generator implementation]{\label{f:CFibonacciSim}\usebox\myboxC}
465\caption{Fibonacci (output) asymmetric generator}
466\label{f:FibonacciAsymmetricGenerator}
467
468\bigskip
469
470\begin{lrbox}{\myboxA}
471\begin{cfa}[aboveskip=0pt,belowskip=0pt]
472`generator Fmt` {
473        char ch;
474        int g, b;
475};
476void ?{}( Fmt & fmt ) { `resume(fmt);` } // constructor
477void ^?{}( Fmt & f ) with(f) { $\C[1.75in]{// destructor}$
478        if ( g != 0 || b != 0 ) sout | nl; }
479void `main( Fmt & f )` with(f) {
480        for () { $\C{// until destructor call}$
481                for ( ; g < 5; g += 1 ) { $\C{// groups}$
482                        for ( ; b < 4; b += 1 ) { $\C{// blocks}$
483                                `suspend;` $\C{// wait for character}$
484                                while ( ch == '\n' ) `suspend;` // ignore
485                                sout | ch;                                              // newline
486                        } sout | " ";  // block spacer
487                } sout | nl; // group newline
488        }
489}
490int main() {
491        Fmt fmt; $\C{// fmt constructor called}$
492        for () {
493                sin | fmt.ch; $\C{// read into generator}$
494          if ( eof( sin ) ) break;
495                `resume( fmt );`
496        }
497
498} $\C{// fmt destructor called}\CRT$
499\end{cfa}
500\end{lrbox}
501
502\begin{lrbox}{\myboxB}
503\begin{cfa}[aboveskip=0pt,belowskip=0pt]
504typedef struct {
505        void * next;
506        char ch;
507        int g, b;
508} Fmt;
509void comain( Fmt * f ) {
510        if ( f->next ) goto *f->next;
511        f->next = &&s1;
512        for ( ;; ) {
513                for ( f->g = 0; f->g < 5; f->g += 1 ) {
514                        for ( f->b = 0; f->b < 4; f->b += 1 ) {
515                                return;
516                          s1:;  while ( f->ch == '\n' ) return;
517                                printf( "%c", f->ch );
518                        } printf( " " );
519                } printf( "\n" );
520        }
521}
522int main() {
523        Fmt fmt = { NULL };  comain( &fmt ); // prime
524        for ( ;; ) {
525                scanf( "%c", &fmt.ch );
526          if ( feof( stdin ) ) break;
527                comain( &fmt );
528        }
529        if ( fmt.g != 0 || fmt.b != 0 ) printf( "\n" );
530}
531\end{cfa}
532\end{lrbox}
533
534\subfloat[\CFA asymmetric generator]{\label{f:CFAFormatGen}\usebox\myboxA}
535\hspace{3pt}
536\vrule
537\hspace{3pt}
538\subfloat[C generator simulation]{\label{f:CFormatSim}\usebox\myboxB}
539\hspace{3pt}
540\caption{Formatter (input) asymmetric generator}
541\label{f:FormatterAsymmetricGenerator}
542\end{figure}
543
544Stateful functions appear as generators, coroutines, and threads, where presentations are based on function objects or pointers~\cite{Butenhof97, C++14, MS:VisualC++, BoostCoroutines15}.
545For example, Python presents generators as a function object:
546\begin{python}
547def Gen():
548        ... `yield val` ...
549gen = Gen()
550for i in range( 10 ):
551        print( next( gen ) )
552\end{python}
553Boost presents coroutines in terms of four functor object-types:
554\begin{cfa}
555asymmetric_coroutine<>::pull_type
556asymmetric_coroutine<>::push_type
557symmetric_coroutine<>::call_type
558symmetric_coroutine<>::yield_type
559\end{cfa}
560and many languages present threading using function pointers, @pthreads@~\cite{Butenhof97}, \Csharp~\cite{Csharp}, Go~\cite{Go}, and Scala~\cite{Scala}, \eg pthreads:
561\begin{cfa}
562void * rtn( void * arg ) { ... }
563int i = 3, rc;
564pthread_t t; $\C{// thread id}$
565`rc = pthread_create( &t, rtn, (void *)i );` $\C{// create and initialized task, type-unsafe input parameter}$
566\end{cfa}
567% void mycor( pthread_t cid, void * arg ) {
568%       int * value = (int *)arg;                               $\C{// type unsafe, pointer-size only}$
569%       // thread body
570% }
571% int main() {
572%       int input = 0, output;
573%       coroutine_t cid = coroutine_create( &mycor, (void *)&input ); $\C{// type unsafe, pointer-size only}$
574%       coroutine_resume( cid, (void *)input, (void **)&output ); $\C{// type unsafe, pointer-size only}$
575% }
576\CFA's preferred presentation model for generators/coroutines/threads is a hybrid of objects and functions, with an object-oriented flavour.
577Essentially, the generator/coroutine/thread function is semantically coupled with a generator/coroutine/thread custom type.
578The custom type solves several issues, while accessing the underlying mechanisms used by the custom types is still allowed.
579
580
581\subsection{Generator}
582
583Stackless generators have the potential to be very small and fast, \ie as small and fast as function call/return for both creation and execution.
584The \CFA goal is to achieve this performance target, possibly at the cost of some semantic complexity.
585A series of different kinds of generators and their implementation demonstrate how this goal is accomplished.
586
587Figure~\ref{f:FibonacciAsymmetricGenerator} shows an unbounded asymmetric generator for an infinite sequence of Fibonacci numbers written in C and \CFA, with a simple C implementation for the \CFA version.
588This generator is an \emph{output generator}, producing a new result on each resumption.
589To compute Fibonacci, the previous two values in the sequence are retained to generate the next value, \ie @fn1@ and @fn@, plus the execution location where control restarts when the generator is resumed, \ie top or middle.
590An additional requirement is the ability to create an arbitrary number of generators (of any kind), \ie retaining one state in global variables is insufficient;
591hence, state is retained in a closure between calls.
592Figure~\ref{f:CFibonacci} shows the C approach of manually creating the closure in structure @Fib@, and multiple instances of this closure provide multiple Fibonacci generators.
593The C version only has the middle execution state because the top execution state is declaration initialization.
594Figure~\ref{f:CFAFibonacciGen} shows the \CFA approach, which also has a manual closure, but replaces the structure with a custom \CFA @generator@ type.
595This generator type is then connected to a function that \emph{must be named \lstinline|main|},\footnote{
596The name \lstinline|main| has special meaning in C, specifically the function where a program starts execution.
597Hence, overloading this name for other starting points (generator/coroutine/thread) is a logical extension.}
598called a \emph{generator main},which takes as its only parameter a reference to the generator type.
599The generator main contains @suspend@ statements that suspend execution without ending the generator versus @return@.
600For the Fibonacci generator-main,\footnote{
601The \CFA \lstinline|with| opens an aggregate scope making its fields directly accessible, like Pascal \lstinline|with|, but using parallel semantics.
602Multiple aggregates may be opened.}
603the top initialization state appears at the start and the middle execution state is denoted by statement @suspend@.
604Any local variables in @main@ \emph{are not retained} between calls;
605hence local variables are only for temporary computations \emph{between} suspends.
606All retained state \emph{must} appear in the generator's type.
607As well, generator code containing a @suspend@ cannot be refactored into a helper function called by the generator, because @suspend@ is implemented via @return@, so a return from the helper function goes back to the current generator not the resumer.
608The generator is started by calling function @resume@ with a generator instance, which begins execution at the top of the generator main, and subsequent @resume@ calls restart the generator at its point of last suspension.
609Resuming an ended (returned) generator is undefined.
610Function @resume@ returns its argument generator so it can be cascaded in an expression, in this case to print the next Fibonacci value @fn@ computed in the generator instance.
611Figure~\ref{f:CFibonacciSim} shows the C implementation of the \CFA generator only needs one additional field, @next@, to handle retention of execution state.
612The computed @goto@ at the start of the generator main, which branches after the previous suspend, adds very little cost to the resume call.
613Finally, an explicit generator type provides both design and performance benefits, such as multiple type-safe interface functions taking and returning arbitrary types.
614\begin{cfa}
615int ?()( Fib & fib ) { return `resume( fib )`.fn; } $\C[3.9in]{// function-call interface}$
616int ?()( Fib & fib, int N ) { for ( N - 1 ) `fib()`; return `fib()`; } $\C{// use function-call interface to skip N values}$
617double ?()( Fib & fib ) { return (int)`fib()` / 3.14159; } $\C{// different return type, cast prevents recursive call}\CRT$
618sout | (int)f1() | (double)f1() | f2( 2 ); // alternative interface, cast selects call based on return type, step 2 values
619\end{cfa}
620Now, the generator can be a separately compiled opaque-type only accessed through its interface functions.
621For contrast, Figure~\ref{f:PythonFibonacci} shows the equivalent Python Fibonacci generator, which does not use a generator type, and hence only has a single interface, but an implicit closure.
622
623Having to manually create the generator closure by moving local-state variables into the generator type is an additional programmer burden.
624(This restriction is removed by the coroutine in Section~\ref{s:Coroutine}.)
625This requirement follows from the generality of variable-size local-state, \eg local state with a variable-length array requires dynamic allocation because the array size is unknown at compile time.
626However, dynamic allocation significantly increases the cost of generator creation/destruction and is a showstopper for embedded real-time programming.
627But more importantly, the size of the generator type is tied to the local state in the generator main, which precludes separate compilation of the generator main, \ie a generator must be inlined or local state must be dynamically allocated.
628With respect to safety, we believe static analysis can discriminate local state from temporary variables in a generator, \ie variable usage spanning @suspend@, and generate a compile-time error.
629Finally, our current experience is that most generator problems have simple data state, including local state, but complex execution state, so the burden of creating the generator type is small.
630As well, C programmers are not afraid of this kind of semantic programming requirement, if it results in very small, fast generators.
631
632Figure~\ref{f:CFAFormatGen} shows an asymmetric \newterm{input generator}, @Fmt@, for restructuring text into groups of characters of fixed-size blocks, \ie the input on the left is reformatted into the output on the right, where newlines are ignored.
633\begin{center}
634\tt
635\begin{tabular}{@{}l|l@{}}
636\multicolumn{1}{c|}{\textbf{\textrm{input}}} & \multicolumn{1}{c}{\textbf{\textrm{output}}} \\
637\begin{tabular}[t]{@{}ll@{}}
638abcdefghijklmnopqrstuvwxyz \\
639abcdefghijklmnopqrstuvwxyz
640\end{tabular}
641&
642\begin{tabular}[t]{@{}lllll@{}}
643abcd    & efgh  & ijkl  & mnop  & qrst  \\
644uvwx    & yzab  & cdef  & ghij  & klmn  \\
645opqr    & stuv  & wxyz  &               &
646\end{tabular}
647\end{tabular}
648\end{center}
649The example takes advantage of resuming a generator in the constructor to prime the loops so the first character sent for formatting appears inside the nested loops.
650The destructor provides a newline if formatted text ends with a full line.
651Figure~\ref{f:CFormatSim} shows the C implementation of the \CFA input generator with one additional field and the computed @goto@.
652For contrast, Figure~\ref{f:PythonFormatter} shows the equivalent Python format generator with the same properties as the Fibonacci generator.
653
654Figure~\ref{f:DeviceDriverGen} shows a \emph{killer} asymmetric generator, a device-driver, because device drivers caused 70\%-85\% of failures in Windows/Linux~\cite{Swift05}.
655Device drives follow the pattern of simple data state but complex execution state, \ie finite state-machine (FSM) parsing a protocol.
656For example, the following protocol:
657\begin{center}
658\ldots\, STX \ldots\, message \ldots\, ESC ETX \ldots\, message \ldots\, ETX 2-byte crc \ldots
659\end{center}
660is a network message beginning with the control character STX, ending with an ETX, and followed by a 2-byte cyclic-redundancy check.
661Control characters may appear in a message if preceded by an ESC.
662When a message byte arrives, it triggers an interrupt, and the operating system services the interrupt by calling the device driver with the byte read from a hardware register.
663The device driver returns a status code of its current state, and when a complete message is obtained, the operating system knows the message is in the message buffer.
664Hence, the device driver is an input/output generator.
665
666Note, the cost of creating and resuming the device-driver generator, @Driver@, is virtually identical to call/return, so performance in an operating-system kernel is excellent.
667As well, the data state is small, where variables @byte@ and @msg@ are communication variables for passing in message bytes and returning the message, and variables @lnth@, @crc@, and @sum@ are local variable that must be retained between calls and are manually hoisted into the generator type.
668% Manually, detecting and hoisting local-state variables is easy when the number is small.
669In contrast, the execution state is large, with one @resume@ and seven @suspend@s.
670Hence, the key benefits of the generator are correctness, safety, and maintenance because the execution states are transcribed directly into the programming language rather than using a table-driven approach.
671Because FSMs can be complex and frequently occur in important domains, direct support of the generator is crucial in a system programming language.
672
673\begin{figure}
674\centering
675\newbox\myboxA
676\begin{lrbox}{\myboxA}
677\begin{python}[aboveskip=0pt,belowskip=0pt]
678def Fib():
679        fn1, fn = 0, 1
680        while True:
681                `yield fn1`
682                fn1, fn = fn, fn1 + fn
683f1 = Fib()
684f2 = Fib()
685for i in range( 10 ):
686        print( next( f1 ), next( f2 ) )
687
688
689
690
691
692
693\end{python}
694\end{lrbox}
695
696\newbox\myboxB
697\begin{lrbox}{\myboxB}
698\begin{python}[aboveskip=0pt,belowskip=0pt]
699def Fmt():
700        try:
701                while True:
702                        for g in range( 5 ):
703                                for b in range( 4 ):
704                                        print( `(yield)`, end='' )
705                                print( '  ', end='' )
706                        print()
707        except GeneratorExit:
708                if g != 0 | b != 0:
709                        print()
710fmt = Fmt()
711`next( fmt )`                    # prime, next prewritten
712for i in range( 41 ):
713        `fmt.send( 'a' );`      # send to yield
714\end{python}
715\end{lrbox}
716\subfloat[Fibonacci]{\label{f:PythonFibonacci}\usebox\myboxA}
717\hspace{3pt}
718\vrule
719\hspace{3pt}
720\subfloat[Formatter]{\label{f:PythonFormatter}\usebox\myboxB}
721\caption{Python generator}
722\label{f:PythonGenerator}
723
724\bigskip
725
726\begin{tabular}{@{}l|l@{}}
727\begin{cfa}[aboveskip=0pt,belowskip=0pt]
728enum Status { CONT, MSG, ESTX,
729                                ELNTH, ECRC };
730`generator` Driver {
731        Status status;
732        unsigned char byte, * msg; // communication
733        unsigned int lnth, sum;      // local state
734        unsigned short int crc;
735};
736void ?{}( Driver & d, char * m ) { d.msg = m; }
737Status next( Driver & d, char b ) with( d ) {
738        byte = b; `resume( d );` return status;
739}
740void main( Driver & d ) with( d ) {
741        enum { STX = '\002', ESC = '\033',
742                        ETX = '\003', MaxMsg = 64 };
743  msg: for () { // parse message
744                status = CONT;
745                lnth = 0; sum = 0;
746                while ( byte != STX ) `suspend;`
747          emsg: for () {
748                        `suspend;` // process byte
749\end{cfa}
750&
751\begin{cfa}[aboveskip=0pt,belowskip=0pt]
752                        choose ( byte ) { // switch with implicit break
753                          case STX:
754                                status = ESTX; `suspend;` continue msg;
755                          case ETX:
756                                break emsg;
757                          case ESC:
758                                `suspend;`
759                        }
760                        if ( lnth >= MaxMsg ) { // buffer full ?
761                                status = ELNTH; `suspend;` continue msg; }
762                        msg[lnth++] = byte;
763                        sum += byte;
764                }
765                msg[lnth] = '\0'; // terminate string
766                `suspend;`
767                crc = byte << 8;
768                `suspend;`
769                status = (crc | byte) == sum ? MSG : ECRC;
770                `suspend;`
771        }
772}
773\end{cfa}
774\end{tabular}
775\caption{Device-driver generator for communication protocol}
776\label{f:DeviceDriverGen}
777\end{figure}
778
779Figure~\ref{f:CFAPingPongGen} shows a symmetric generator, where the generator resumes another generator, forming a resume/resume cycle.
780(The trivial cycle is a generator resuming itself.)
781This control flow is similar to recursion for functions but without stack growth.
782The steps for symmetric control-flow are creating, executing, and terminating the cycle.
783Constructing the cycle must deal with definition-before-use to close the cycle, \ie, the first generator must know about the last generator, which is not within scope.
784(This issue occurs for any cyclic data structure.)
785% The example creates all the generators and then assigns the partners that form the cycle.
786% Alternatively, the constructor can assign the partners as they are declared, except the first, and the first-generator partner is set after the last generator declaration to close the cycle.
787Once the cycle is formed, the program main resumes one of the generators, and the generators can then traverse an arbitrary cycle using @resume@ to activate partner generator(s).
788Terminating the cycle is accomplished by @suspend@ or @return@, both of which go back to the stack frame that started the cycle (program main in the example).
789The starting stack-frame is below the last active generator because the resume/resume cycle does not grow the stack.
790Also, since local variables are not retained in the generator function, it does not contain any objects with destructors that must be called, so the  cost is the same as a function return.
791Destructor cost occurs when the generator instance is deallocated, which is easily controlled by the programmer.
792
793Figure~\ref{f:CPingPongSim} shows the implementation of the symmetric generator, where the complexity is the @resume@, which needs an extension to the calling convention to perform a forward rather than backward jump.
794This jump-starts at the top of the next generator main to re-execute the normal calling convention to make space on the stack for its local variables.
795However, before the jump, the caller must reset its stack (and any registers) equivalent to a @return@, but subsequently jump forward.
796This semantics is basically a tail-call optimization, which compilers already perform.
797The example shows the assembly code to undo the generator's entry code before the direct jump.
798This assembly code depends on what entry code is generated, specifically if there are local variables, and the level of optimization.
799To provide this new calling convention requires a mechanism built into the compiler, which is beyond the scope of \CFA at this time.
800Nevertheless, it is possible to hand generate any symmetric generators for proof of concept and performance testing.
801A compiler could also eliminate other artifacts in the generator simulation to further increase performance, \eg LLVM has various coroutine support~\cite{CoroutineTS}, and \CFA can leverage this support should it fork @clang@.
802
803\begin{figure}
804\centering
805\begin{lrbox}{\myboxA}
806\begin{cfa}[aboveskip=0pt,belowskip=0pt]
807`generator PingPong` {
808        const char * name;
809        int N;
810        int i;                          // local state
811        PingPong & partner; // rebindable reference
812};
813
814void `main( PingPong & pp )` with(pp) {
815        for ( ; i < N; i += 1 ) {
816                sout | name | i;
817                `resume( partner );`
818        }
819}
820int main() {
821        enum { N = 5 };
822        PingPong ping = {"ping",N,0}, pong = {"pong",N,0};
823        &ping.partner = &pong;  &pong.partner = &ping;
824        `resume( ping );`
825}
826\end{cfa}
827\end{lrbox}
828
829\begin{lrbox}{\myboxB}
830\begin{cfa}[escapechar={},aboveskip=0pt,belowskip=0pt]
831typedef struct PingPong {
832        const char * name;
833        int N, i;
834        struct PingPong * partner;
835        void * next;
836} PingPong;
837#define PPCtor(name, N) {name,N,0,NULL,NULL}
838void comain( PingPong * pp ) {
839        if ( pp->next ) goto *pp->next;
840        pp->next = &&cycle;
841        for ( ; pp->i < pp->N; pp->i += 1 ) {
842                printf( "%s %d\n", pp->name, pp->i );
843                asm( "mov  %0,%%rdi" : "=m" (pp->partner) );
844                asm( "mov  %rdi,%rax" );
845                asm( "popq %rbx" );
846                asm( "jmp  comain" );
847          cycle: ;
848        }
849}
850\end{cfa}
851\end{lrbox}
852
853\subfloat[\CFA symmetric generator]{\label{f:CFAPingPongGen}\usebox\myboxA}
854\hspace{3pt}
855\vrule
856\hspace{3pt}
857\subfloat[C generator simulation]{\label{f:CPingPongSim}\usebox\myboxB}
858\hspace{3pt}
859\caption{Ping-Pong symmetric generator}
860\label{f:PingPongSymmetricGenerator}
861\end{figure}
862
863Finally, part of this generator work was inspired by the recent \CCtwenty generator proposal~\cite{C++20Coroutine19} (which they call coroutines).
864Our work provides the same high-performance asymmetric generators as \CCtwenty, and extends their work with symmetric generators.
865An additional \CCtwenty generator feature allows @suspend@ and @resume@ to be followed by a restricted compound statement that is executed after the current generator has reset its stack but before calling the next generator, specified with \CFA syntax:
866\begin{cfa}
867... suspend`{ ... }`;
868... resume( C )`{ ... }` ...
869\end{cfa}
870Since the current generator's stack is released before calling the compound statement, the compound statement can only reference variables in the generator's type.
871This feature is useful when a generator is used in a concurrent context to ensure it is stopped before releasing a lock in the compound statement, which might immediately allow another thread to resume the generator.
872Hence, this mechanism provides a general and safe handoff of the generator among competing threads.
873
874
875\subsection{Coroutine}
876\label{s:Coroutine}
877
878Stackful coroutines extend generator semantics, \ie there is an implicit closure and @suspend@ may appear in a helper function called from the coroutine main.
879A coroutine is specified by replacing @generator@ with @coroutine@ for the type.
880Coroutine generality results in higher cost for creation, due to dynamic stack allocation, execution, due to context switching among stacks, and terminating, due to possible stack unwinding and dynamic stack deallocation.
881A series of different kinds of coroutines and their implementations demonstrate how coroutines extend generators.
882
883First, the previous generator examples are converted to their coroutine counterparts, allowing local-state variables to be moved from the generator type into the coroutine main.
884\begin{description}
885\item[Fibonacci]
886Move the declaration of @fn1@ to the start of coroutine main.
887\begin{cfa}[xleftmargin=0pt]
888void main( Fib & fib ) with(fib) {
889        `int fn1;`
890\end{cfa}
891\item[Formatter]
892Move the declaration of @g@ and @b@ to the for loops in the coroutine main.
893\begin{cfa}[xleftmargin=0pt]
894for ( `g`; 5 ) {
895        for ( `b`; 4 ) {
896\end{cfa}
897\item[Device Driver]
898Move the declaration of @lnth@ and @sum@ to their points of initialization.
899\begin{cfa}[xleftmargin=0pt]
900        status = CONT;
901        `unsigned int lnth = 0, sum = 0;`
902        ...
903        `unsigned short int crc = byte << 8;`
904\end{cfa}
905\item[PingPong]
906Move the declaration of @i@ to the for loop in the coroutine main.
907\begin{cfa}[xleftmargin=0pt]
908void main( PingPong & pp ) with(pp) {
909        for ( `i`; N ) {
910\end{cfa}
911\end{description}
912It is also possible to refactor code containing local-state and @suspend@ statements into a helper function, like the computation of the CRC for the device driver.
913\begin{cfa}
914unsigned int Crc() {
915        `suspend;`
916        unsigned short int crc = byte << 8;
917        `suspend;`
918        status = (crc | byte) == sum ? MSG : ECRC;
919        return crc;
920}
921\end{cfa}
922A call to this function is placed at the end of the driver's coroutine-main.
923For complex finite-state machines, refactoring is part of normal program abstraction, especially when code is used in multiple places.
924Again, this complexity is usually associated with execution state rather than data state.
925
926\begin{comment}
927Figure~\ref{f:Coroutine3States} creates a @coroutine@ type, @`coroutine` Fib { int fn; }@, which provides communication, @fn@, for the \newterm{coroutine main}, @main@, which runs on the coroutine stack, and possibly multiple interface functions, \eg @next@.
928Like the structure in Figure~\ref{f:ExternalState}, the coroutine type allows multiple instances, where instances of this type are passed to the (overloaded) coroutine main.
929The coroutine main's stack holds the state for the next generation, @f1@ and @f2@, and the code represents the three states in the Fibonacci formula via the three suspend points, to context switch back to the caller's @resume@.
930The interface function @next@, takes a Fibonacci instance and context switches to it using @resume@;
931on restart, the Fibonacci field, @fn@, contains the next value in the sequence, which is returned.
932The first @resume@ is special because it allocates the coroutine stack and cocalls its coroutine main on that stack;
933when the coroutine main returns, its stack is deallocated.
934Hence, @Fib@ is an object at creation, transitions to a coroutine on its first resume, and transitions back to an object when the coroutine main finishes.
935Figure~\ref{f:Coroutine1State} shows the coroutine version of the C version in Figure~\ref{f:ExternalState}.
936Coroutine generators are called \newterm{output coroutines} because values are only returned.
937
938\begin{figure}
939\centering
940\newbox\myboxA
941% \begin{lrbox}{\myboxA}
942% \begin{cfa}[aboveskip=0pt,belowskip=0pt]
943% `int fn1, fn2, state = 1;`   // single global variables
944% int fib() {
945%       int fn;
946%       `switch ( state )` {  // explicit execution state
947%         case 1: fn = 0;  fn1 = fn;  state = 2;  break;
948%         case 2: fn = 1;  fn2 = fn1;  fn1 = fn;  state = 3;  break;
949%         case 3: fn = fn1 + fn2;  fn2 = fn1;  fn1 = fn;  break;
950%       }
951%       return fn;
952% }
953% int main() {
954%
955%       for ( int i = 0; i < 10; i += 1 ) {
956%               printf( "%d\n", fib() );
957%       }
958% }
959% \end{cfa}
960% \end{lrbox}
961\begin{lrbox}{\myboxA}
962\begin{cfa}[aboveskip=0pt,belowskip=0pt]
963#define FibCtor { 0, 1 }
964typedef struct { int fn1, fn; } Fib;
965int fib( Fib * f ) {
966
967        int ret = f->fn1;
968        f->fn1 = f->fn;
969        f->fn = ret + f->fn;
970        return ret;
971}
972
973
974
975int main() {
976        Fib f1 = FibCtor, f2 = FibCtor;
977        for ( int i = 0; i < 10; i += 1 ) {
978                printf( "%d %d\n",
979                                fib( &f1 ), fib( &f2 ) );
980        }
981}
982\end{cfa}
983\end{lrbox}
984
985\newbox\myboxB
986\begin{lrbox}{\myboxB}
987\begin{cfa}[aboveskip=0pt,belowskip=0pt]
988`coroutine` Fib { int fn1; };
989void main( Fib & fib ) with( fib ) {
990        int fn;
991        [fn1, fn] = [0, 1];
992        for () {
993                `suspend;`
994                [fn1, fn] = [fn, fn1 + fn];
995        }
996}
997int ?()( Fib & fib ) with( fib ) {
998        return `resume( fib )`.fn1;
999}
1000int main() {
1001        Fib f1, f2;
1002        for ( 10 ) {
1003                sout | f1() | f2();
1004}
1005
1006
1007\end{cfa}
1008\end{lrbox}
1009
1010\newbox\myboxC
1011\begin{lrbox}{\myboxC}
1012\begin{python}[aboveskip=0pt,belowskip=0pt]
1013
1014def Fib():
1015
1016        fn1, fn = 0, 1
1017        while True:
1018                `yield fn1`
1019                fn1, fn = fn, fn1 + fn
1020
1021
1022// next prewritten
1023
1024
1025f1 = Fib()
1026f2 = Fib()
1027for i in range( 10 ):
1028        print( next( f1 ), next( f2 ) )
1029
1030
1031
1032\end{python}
1033\end{lrbox}
1034
1035\subfloat[C]{\label{f:GlobalVariables}\usebox\myboxA}
1036\hspace{3pt}
1037\vrule
1038\hspace{3pt}
1039\subfloat[\CFA]{\label{f:ExternalState}\usebox\myboxB}
1040\hspace{3pt}
1041\vrule
1042\hspace{3pt}
1043\subfloat[Python]{\label{f:ExternalState}\usebox\myboxC}
1044\caption{Fibonacci generator}
1045\label{f:C-fibonacci}
1046\end{figure}
1047
1048\bigskip
1049
1050\newbox\myboxA
1051\begin{lrbox}{\myboxA}
1052\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1053`coroutine` Fib { int fn; };
1054void main( Fib & fib ) with( fib ) {
1055        fn = 0;  int fn1 = fn; `suspend`;
1056        fn = 1;  int fn2 = fn1;  fn1 = fn; `suspend`;
1057        for () {
1058                fn = fn1 + fn2; fn2 = fn1; fn1 = fn; `suspend`; }
1059}
1060int next( Fib & fib ) with( fib ) { `resume( fib );` return fn; }
1061int main() {
1062        Fib f1, f2;
1063        for ( 10 )
1064                sout | next( f1 ) | next( f2 );
1065}
1066\end{cfa}
1067\end{lrbox}
1068\newbox\myboxB
1069\begin{lrbox}{\myboxB}
1070\begin{python}[aboveskip=0pt,belowskip=0pt]
1071
1072def Fibonacci():
1073        fn = 0; fn1 = fn; `yield fn`  # suspend
1074        fn = 1; fn2 = fn1; fn1 = fn; `yield fn`
1075        while True:
1076                fn = fn1 + fn2; fn2 = fn1; fn1 = fn; `yield fn`
1077
1078
1079f1 = Fibonacci()
1080f2 = Fibonacci()
1081for i in range( 10 ):
1082        print( `next( f1 )`, `next( f2 )` ) # resume
1083
1084\end{python}
1085\end{lrbox}
1086\subfloat[\CFA]{\label{f:Coroutine3States}\usebox\myboxA}
1087\qquad
1088\subfloat[Python]{\label{f:Coroutine1State}\usebox\myboxB}
1089\caption{Fibonacci input coroutine, 3 states, internal variables}
1090\label{f:cfa-fibonacci}
1091\end{figure}
1092\end{comment}
1093
1094\begin{figure}
1095\centering
1096\lstset{language=CFA,escapechar={},moredelim=**[is][\protect\color{red}]{`}{`}}% allow $
1097\begin{tabular}{@{}l@{\hspace{2\parindentlnth}}l@{}}
1098\begin{cfa}
1099`coroutine` Prod {
1100        Cons & c;                       // communication
1101        int N, money, receipt;
1102};
1103void main( Prod & prod ) with( prod ) {
1104        // 1st resume starts here
1105        for ( i; N ) {
1106                int p1 = random( 100 ), p2 = random( 100 );
1107                sout | p1 | " " | p2;
1108                int status = delivery( c, p1, p2 );
1109                sout | " $" | money | nl | status;
1110                receipt += 1;
1111        }
1112        stop( c );
1113        sout | "prod stops";
1114}
1115int payment( Prod & prod, int money ) {
1116        prod.money = money;
1117        `resume( prod );`
1118        return prod.receipt;
1119}
1120void start( Prod & prod, int N, Cons &c ) {
1121        &prod.c = &c;
1122        prod.[N, receipt] = [N, 0];
1123        `resume( prod );`
1124}
1125int main() {
1126        Prod prod;
1127        Cons cons = { prod };
1128        start( prod, 5, cons );
1129}
1130\end{cfa}
1131&
1132\begin{cfa}
1133`coroutine` Cons {
1134        Prod & p;                       // communication
1135        int p1, p2, status;
1136        bool done;
1137};
1138void ?{}( Cons & cons, Prod & p ) {
1139        &cons.p = &p; // reassignable reference
1140        cons.[status, done ] = [0, false];
1141}
1142void main( Cons & cons ) with( cons ) {
1143        // 1st resume starts here
1144        int money = 1, receipt;
1145        for ( ; ! done; ) {
1146                sout | p1 | " " | p2 | nl | " $" | money;
1147                status += 1;
1148                receipt = payment( p, money );
1149                sout | " #" | receipt;
1150                money += 1;
1151        }
1152        sout | "cons stops";
1153}
1154int delivery( Cons & cons, int p1, int p2 ) {
1155        cons.[p1, p2] = [p1, p2];
1156        `resume( cons );`
1157        return cons.status;
1158}
1159void stop( Cons & cons ) {
1160        cons.done = true;
1161        `resume( cons );`
1162}
1163
1164\end{cfa}
1165\end{tabular}
1166\caption{Producer / consumer: resume-resume cycle, bidirectional communication}
1167\label{f:ProdCons}
1168\end{figure}
1169
1170Figure~\ref{f:ProdCons} shows the ping-pong example in Figure~\ref{f:CFAPingPongGen} extended into a producer/consumer symmetric-coroutine performing bidirectional communication.
1171This example is illustrative because both producer/consumer have two interface functions with @resume@s that suspend execution in these interface (helper) functions.
1172The program main creates the producer coroutine, passes it to the consumer coroutine in its initialization, and closes the cycle at the call to @start@ along with the number of items to be produced.
1173The first @resume@ of @prod@ creates @prod@'s stack with a frame for @prod@'s coroutine main at the top, and context switches to it.
1174@prod@'s coroutine main starts, creates local-state variables that are retained between coroutine activations, and executes $N$ iterations, each generating two random values, calling the consumer to deliver the values, and printing the status returned from the consumer.
1175
1176The producer call to @delivery@ transfers values into the consumer's communication variables, resumes the consumer, and returns the consumer status.
1177On the first resume, @cons@'s stack is created and initialized, holding local-state variables retained between subsequent activations of the coroutine.
1178The consumer iterates until the @done@ flag is set, prints the values delivered by the producer, increments status, and calls back to the producer via @payment@, and on return from @payment@, prints the receipt from the producer and increments @money@ (inflation).
1179The call from the consumer to @payment@ introduces the cycle between producer and consumer.
1180When @payment@ is called, the consumer copies values into the producer's communication variable and a resume is executed.
1181The context switch restarts the producer at the point where it last context switched, so it continues in @delivery@ after the resume.
1182@delivery@ returns the status value in @prod@'s coroutine main, where the status is printed.
1183The loop then repeats calling @delivery@, where each call resumes the consumer coroutine.
1184The context switch to the consumer continues in @payment@.
1185The consumer increments and returns the receipt to the call in @cons@'s coroutine main.
1186The loop then repeats calling @payment@, where each call resumes the producer coroutine.
1187Figure~\ref{f:ProdConsRuntimeStacks} shows the runtime stacks of the program main, and the coroutine mains for @prod@ and @cons@ during the cycling.
1188
1189\begin{figure}
1190\begin{center}
1191\input{FullProdConsStack.pstex_t}
1192\end{center}
1193\vspace*{-10pt}
1194\caption{Producer / consumer runtime stacks}
1195\label{f:ProdConsRuntimeStacks}
1196
1197\medskip
1198
1199\begin{center}
1200\input{FullCoroutinePhases.pstex_t}
1201\end{center}
1202\vspace*{-10pt}
1203\caption{Ping / Pong coroutine steps}
1204\label{f:PingPongFullCoroutineSteps}
1205\end{figure}
1206
1207Terminating a coroutine cycle is more complex than a generator cycle, because it requires context switching to the program main's \emph{stack} to shutdown the program, whereas generators started by the program main run on its stack.
1208Furthermore, each deallocated coroutine must guarantee all destructors are run for object allocated in the coroutine type \emph{and} allocated on the coroutine's stack at the point of suspension, which can be arbitrarily deep.
1209When a coroutine's main ends, its stack is already unwound so any stack allocated objects with destructors have been finalized.
1210The na\"{i}ve semantics for coroutine-cycle termination is to context switch to the last resumer, like executing a @suspend@/@return@ in a generator.
1211However, for coroutines, the last resumer is \emph{not} implicitly below the current stack frame, as for generators, because each coroutine's stack is independent.
1212Unfortunately, it is impossible to determine statically if a coroutine is in a cycle and unrealistic to check dynamically (graph-cycle problem).
1213Hence, a compromise solution is necessary that works for asymmetric (acyclic) and symmetric (cyclic) coroutines.
1214
1215Our solution is to context switch back to the first resumer (starter) once the coroutine ends.
1216This semantics works well for the most common asymmetric and symmetric coroutine usage patterns.
1217For asymmetric coroutines, it is common for the first resumer (starter) coroutine to be the only resumer.
1218All previous generators converted to coroutines have this property.
1219For symmetric coroutines, it is common for the cycle creator to persist for the lifetime of the cycle.
1220Hence, the starter coroutine is remembered on the first resume and ending the coroutine resumes the starter.
1221Figure~\ref{f:ProdConsRuntimeStacks} shows this semantic by the dashed lines from the end of the coroutine mains: @prod@ starts @cons@ so @cons@ resumes @prod@ at the end, and the program main starts @prod@ so @prod@ resumes the program main at the end.
1222For other scenarios, it is always possible to devise a solution with additional programming effort, such as forcing the cycle forward (backward) to a safe point before starting termination.
1223
1224The producer/consumer example does not illustrate the full power of the starter semantics because @cons@ always ends first.
1225Assume generator @PingPong@ is converted to a coroutine.
1226Figure~\ref{f:PingPongFullCoroutineSteps} shows the creation, starter, and cyclic execution steps of the coroutine version.
1227The program main creates (declares) coroutine instances @ping@ and @pong@.
1228Next, program main resumes @ping@, making it @ping@'s starter, and @ping@'s main resumes @pong@'s main, making it @pong@'s starter.
1229Execution forms a cycle when @pong@ resumes @ping@, and cycles $N$ times.
1230By adjusting $N$ for either @ping@/@pong@, it is possible to have either one finish first, instead of @pong@ always ending first.
1231If @pong@ ends first, it resumes its starter @ping@ in its coroutine main, then @ping@ ends and resumes its starter the program main in function @start@.
1232If @ping@ ends first, it resumes its starter the program main in function @start@.
1233Regardless of the cycle complexity, the starter stack always leads back to the program main, but the stack can be entered at an arbitrary point.
1234Once back at the program main, coroutines @ping@ and @pong@ are deallocated.
1235For generators, deallocation runs the destructors for all objects in the generator type.
1236For coroutines, deallocation deals with objects in the coroutine type and must also run the destructors for any objects pending on the coroutine's stack for any unterminated coroutine.
1237Hence, if a coroutine's destructor detects the coroutine is not ended, it implicitly raises a cancellation exception (uncatchable exception) at the coroutine and resumes it so the cancellation exception can propagate to the root of the coroutine's stack destroying all local variable on the stack.
1238So the \CFA semantics for the generator and coroutine, ensure both can be safely deallocated at any time, regardless of their current state, like any other aggregate object.
1239Explicitly raising normal exceptions at another coroutine can replace flag variables, like @stop@, \eg @prod@ raises a @stop@ exception at @cons@ after it finishes generating values and resumes @cons@, which catches the @stop@ exception to terminate its loop.
1240
1241Finally, there is an interesting effect for @suspend@ with symmetric coroutines.
1242A coroutine must retain its last resumer to suspend back because the resumer is on a different stack.
1243These reverse pointers allow @suspend@ to cycle \emph{backwards}, which may be useful in certain cases.
1244However, there is an anomaly if a coroutine resumes itself, because it overwrites its last resumer with itself, losing the ability to resume the last external resumer.
1245To prevent losing this information, a self-resume does not overwrite the last resumer.
1246
1247
1248\subsection{Generator / Coroutine Implementation}
1249
1250A significant implementation challenge for generators/coroutines (and threads in Section~\ref{s:threads}) is adding extra fields to the custom types and related functions, \eg inserting code after/before the coroutine constructor/destructor and @main@ to create/initialize/de-initialize/destroy any extra fields, \eg stack.
1251There are several solutions to these problem, which follow from the object-oriented flavour of adopting custom types.
1252
1253For object-oriented languages, inheritance is used to provide extra fields and code via explicit inheritance:
1254\begin{cfa}[morekeywords={class,inherits}]
1255class myCoroutine inherits baseCoroutine { ... }
1256\end{cfa}
1257% The problem is that the programming language and its tool chain, \eg debugger, @valgrind@, need to understand @baseCoroutine@ because it infers special property, so type @baseCoroutine@ becomes a de facto keyword and all types inheriting from it are implicitly custom types.
1258The problem is that some special properties are not handled by existing language semantics, \eg the execution of constructors/destructors is in the wrong order to implicitly start threads because the thread must start \emph{after} all constructors as it relies on a completely initialized object, but the inherited constructor runs \emph{before} the derived.
1259Alternatives, such as explicitly starting threads as in Java, are repetitive and forgetting to call start is a common source of errors.
1260An alternative is composition:
1261\begin{cfa}
1262struct myCoroutine {
1263        ... // declaration/communication variables
1264        baseCoroutine dummy; // composition, last declaration
1265}
1266\end{cfa}
1267which also requires an explicit declaration that must be last to ensure correct initialization order.
1268However, there is nothing preventing wrong placement or multiple declarations.
1269
1270\CFA custom types make any special properties explicit to the language and its tool chain, \eg the language code-generator knows where to inject code
1271% and when it is unsafe to perform certain optimizations,
1272and IDEs using simple parsing can find and manipulate types with special properties.
1273The downside of this approach is that it makes custom types a special case in the language.
1274Users wanting to extend custom types or build their own can only do so in ways offered by the language.
1275Furthermore, implementing custom types without language support may display the power of a programming language.
1276\CFA blends the two approaches, providing custom type for idiomatic \CFA code, while extending and building new custom types is still possible, similar to Java concurrency with builtin and library.
1277
1278Part of the mechanism to generalize custom types is the \CFA trait~\cite[\S~2.3]{Moss18}, \eg the definition for custom-type @coroutine@ is anything satisfying the trait @is_coroutine@, and this trait both enforces and restricts the coroutine-interface functions.
1279\begin{cfa}
1280trait is_coroutine( `dtype` T ) {
1281        void main( T & );
1282        coroutine_desc * get_coroutine( T & );
1283};
1284forall( `dtype` T | is_coroutine(T) ) void $suspend$( T & ), resume( T & );
1285\end{cfa}
1286Note, copying generators/coroutines/threads is not meaningful.
1287For example, both the resumer and suspender descriptors can have bidirectional pointers;
1288copying these coroutines does not update the internal pointers so behaviour of both copies would be difficult to understand.
1289Furthermore, two coroutines cannot logically execute on the same stack.
1290A deep coroutine copy, which copies the stack, is also meaningless in an unmanaged language (no garbage collection), like C, because the stack may contain pointers to object within it that require updating for the copy.
1291The \CFA @dtype@ property provides no \emph{implicit} copying operations and the @is_coroutine@ trait provides no \emph{explicit} copying operations, so all coroutines must be passed by reference (pointer).
1292The function definitions ensure there is a statically typed @main@ function that is the starting point (first stack frame) of a coroutine, and a mechanism to get (read) the coroutine descriptor from its handle.
1293The @main@ function has no return value or additional parameters because the coroutine type allows an arbitrary number of interface functions with corresponding arbitrary typed input/output values versus fixed ones.
1294The advantage of this approach is that users can easily create different types of coroutines, \eg changing the memory layout of a coroutine is trivial when implementing the @get_coroutine@ function, and possibly redefining \textsf{suspend} and @resume@.
1295
1296The \CFA custom-type @coroutine@ implicitly implements the getter and forward declarations for the coroutine main.
1297\begin{cquote}
1298\begin{tabular}{@{}ccc@{}}
1299\begin{cfa}
1300coroutine MyCor {
1301        int value;
1302
1303};
1304\end{cfa}
1305&
1306{\Large $\Rightarrow$}
1307&
1308\begin{tabular}{@{}ccc@{}}
1309\begin{cfa}
1310struct MyCor {
1311        int value;
1312        coroutine_desc cor;
1313};
1314\end{cfa}
1315&
1316\begin{cfa}
1317static inline coroutine_desc *
1318get_coroutine( MyCor & this ) {
1319        return &this.cor;
1320}
1321\end{cfa}
1322&
1323\begin{cfa}
1324void main( MyCor * this );
1325
1326
1327
1328\end{cfa}
1329\end{tabular}
1330\end{tabular}
1331\end{cquote}
1332The combination of custom types and fundamental @trait@ description of these types allows a concise specification for programmers and tools, while more advanced programmers can have tighter control over memory layout and initialization.
1333
1334Figure~\ref{f:CoroutineMemoryLayout} shows different memory-layout options for a coroutine (where a task is similar).
1335The coroutine handle is the @coroutine@ instance containing programmer specified type global/communication variables across interface functions.
1336The coroutine descriptor contains all implicit declarations needed by the runtime, \eg @suspend@/@resume@, and can be part of the coroutine handle or separate.
1337The coroutine stack can appear in a number of locations and be fixed or variable sized.
1338Hence, the coroutine's stack could be a VLS\footnote{
1339We are examining variable-sized structures (VLS), where fields can be variable-sized structures or arrays.
1340Once allocated, a VLS is fixed sized.}
1341on the allocating stack, provided the allocating stack is large enough.
1342For a VLS stack allocation/deallocation is an inexpensive adjustment of the stack pointer, modulo any stack constructor costs (\eg initial frame setup).
1343For heap stack allocation, allocation/deallocation is an expensive heap allocation (where the heap can be a shared resource), modulo any stack constructor costs.
1344With heap stack allocation, it is also possible to use a split (segmented) stack calling convention, available with gcc and clang, so the stack is variable sized.
1345Currently, \CFA supports stack/heap allocated descriptors but only fixed-sized heap allocated stacks.
1346In \CFA debug-mode, the fixed-sized stack is terminated with a write-only page, which catches most stack overflows.
1347Experience teaching concurrency with \uC~\cite{CS343} shows fixed-sized stacks are rarely an issue for students.
1348Split-stack allocation is under development but requires recompilation of legacy code, which may be impossible.
1349
1350\begin{figure}
1351\centering
1352\input{corlayout.pstex_t}
1353\caption{Coroutine memory layout}
1354\label{f:CoroutineMemoryLayout}
1355\end{figure}
1356
1357
1358\section{Concurrency}
1359\label{s:Concurrency}
1360
1361Concurrency is nondeterministic scheduling of independent sequential execution paths (threads), where each thread has its own stack.
1362A single thread with multiple call stacks, \newterm{coroutining}~\cite{Conway63,Marlin80}, does \emph{not} imply concurrency~\cite[\S~2]{Buhr05a}.
1363In coroutining, coroutines self-schedule the thread across stacks so execution is deterministic.
1364(It is \emph{impossible} to generate a concurrency error when coroutining.)
1365However, coroutines are a stepping stone towards concurrency.
1366
1367The transition to concurrency, even for a single thread with multiple stacks, occurs when coroutines context switch to a \newterm{scheduling coroutine}, introducing non-determinism from the coroutine perspective~\cite[\S~3,]{Buhr05a}.
1368Therefore, a minimal concurrency system requires coroutines \emph{in conjunction with a nondeterministic scheduler}.
1369The resulting execution system now follows a cooperative threading model~\cite{Adya02,libdill}, called \newterm{non-preemptive scheduling}.
1370Adding \newterm{preemption} introduces non-cooperative scheduling, where context switching occurs randomly between any two instructions often based on a timer interrupt, called \newterm{preemptive scheduling}.
1371While a scheduler introduces uncertain execution among explicit context switches, preemption introduces uncertainty by introducing implicit context switches.
1372Uncertainty gives the illusion of parallelism on a single processor and provides a mechanism to access and increase performance on multiple processors.
1373The reason is that the scheduler/runtime have complete knowledge about resources and how to best utilized them.
1374However, the introduction of unrestricted nondeterminism results in the need for \newterm{mutual exclusion} and \newterm{synchronization}, which restrict nondeterminism for correctness;
1375otherwise, it is impossible to write meaningful concurrent programs.
1376Optimal concurrent performance is often obtained by having as much nondeterminism as mutual exclusion and synchronization correctness allow.
1377
1378A scheduler can either be a stackless or stackful.
1379For stackless, the scheduler performs scheduling on the stack of the current coroutine and switches directly to the next coroutine, so there is one context switch.
1380For stackful, the current coroutine switches to the scheduler, which performs scheduling, and it then switches to the next coroutine, so there are two context switches.
1381The \CFA runtime uses a stackful scheduler for uniformity and security.
1382
1383
1384\subsection{Thread}
1385\label{s:threads}
1386
1387Threading needs the ability to start a thread and wait for its completion.
1388A common API for this ability is @fork@ and @join@.
1389\begin{cquote}
1390\begin{tabular}{@{}lll@{}}
1391\multicolumn{1}{c}{\textbf{Java}} & \multicolumn{1}{c}{\textbf{\Celeven}} & \multicolumn{1}{c}{\textbf{pthreads}} \\
1392\begin{cfa}
1393class MyTask extends Thread {...}
1394mytask t = new MyTask(...);
1395`t.start();` // start
1396// concurrency
1397`t.join();` // wait
1398\end{cfa}
1399&
1400\begin{cfa}
1401class MyTask { ... } // functor
1402MyTask mytask;
1403`thread t( mytask, ... );` // start
1404// concurrency
1405`t.join();` // wait
1406\end{cfa}
1407&
1408\begin{cfa}
1409void * rtn( void * arg ) {...}
1410pthread_t t;  int i = 3;
1411`pthread_create( &t, rtn, (void *)i );` // start
1412// concurrency
1413`pthread_join( t, NULL );` // wait
1414\end{cfa}
1415\end{tabular}
1416\end{cquote}
1417\CFA has a simpler approach using a custom @thread@ type and leveraging declaration semantics (allocation/deallocation), where threads implicitly @fork@ after construction and @join@ before destruction.
1418\begin{cfa}
1419thread MyTask {};
1420void main( MyTask & this ) { ... }
1421int main() {
1422        MyTask team`[10]`; $\C[2.5in]{// allocate stack-based threads, implicit start after construction}$
1423        // concurrency
1424} $\C{// deallocate stack-based threads, implicit joins before destruction}$
1425\end{cfa}
1426This semantic ensures a thread is started and stopped exactly once, eliminating some programming error, and scales to multiple threads for basic (termination) synchronization.
1427For block allocation to arbitrary depth, including recursion, threads are created/destroyed in a lattice structure (tree with top and bottom).
1428Arbitrary topologies are possible using dynamic allocation, allowing threads to outlive their declaration scope, identical to normal dynamic allocation.
1429\begin{cfa}
1430MyTask * factory( int N ) { ... return `anew( N )`; } $\C{// allocate heap-based threads, implicit start after construction}$
1431int main() {
1432        MyTask * team = factory( 10 );
1433        // concurrency
1434        `delete( team );` $\C{// deallocate heap-based threads, implicit joins before destruction}\CRT$
1435}
1436\end{cfa}
1437
1438Figure~\ref{s:ConcurrentMatrixSummation} shows concurrently adding the rows of a matrix and then totalling the subtotals sequentially, after all the row threads have terminated.
1439The program uses heap-based threads because each thread needs different constructor values.
1440(Python provides a simple iteration mechanism to initialize array elements to different values allowing stack allocation.)
1441The allocation/deallocation pattern appears unusual because allocated objects are immediately deallocated without any intervening code.
1442However, for threads, the deletion provides implicit synchronization, which is the intervening code.
1443% While the subtotals are added in linear order rather than completion order, which slightly inhibits concurrency, the computation is restricted by the critical-path thread (\ie the thread that takes the longest), and so any inhibited concurrency is very small as totalling the subtotals is trivial.
1444
1445\begin{figure}
1446\begin{cfa}
1447`thread` Adder { int * row, cols, & subtotal; } $\C{// communication variables}$
1448void ?{}( Adder & adder, int row[], int cols, int & subtotal ) {
1449        adder.[ row, cols, &subtotal ] = [ row, cols, &subtotal ];
1450}
1451void main( Adder & adder ) with( adder ) {
1452        subtotal = 0;
1453        for ( c; cols ) { subtotal += row[c]; }
1454}
1455int main() {
1456        const int rows = 10, cols = 1000;
1457        int matrix[rows][cols], subtotals[rows], total = 0;
1458        // read matrix
1459        Adder * adders[rows];
1460        for ( r; rows; ) { $\C{// start threads to sum rows}$
1461                adders[r] = `new( matrix[r], cols, &subtotals[r] );`
1462        }
1463        for ( r; rows ) { $\C{// wait for threads to finish}$
1464                `delete( adders[r] );` $\C{// termination join}$
1465                total += subtotals[r]; $\C{// total subtotal}$
1466        }
1467        sout | total;
1468}
1469\end{cfa}
1470\caption{Concurrent matrix summation}
1471\label{s:ConcurrentMatrixSummation}
1472\end{figure}
1473
1474
1475\subsection{Thread Implementation}
1476
1477Threads in \CFA are user level run by runtime kernel threads (see Section~\ref{s:CFARuntimeStructure}), where user threads provide concurrency and kernel threads provide parallelism.
1478Like coroutines, and for the same design reasons, \CFA provides a custom @thread@ type and a @trait@ to enforce and restrict the task-interface functions.
1479\begin{cquote}
1480\begin{tabular}{@{}c@{\hspace{3\parindentlnth}}c@{}}
1481\begin{cfa}
1482thread myThread {
1483        ... // declaration/communication variables
1484};
1485
1486
1487\end{cfa}
1488&
1489\begin{cfa}
1490trait is_thread( `dtype` T ) {
1491        void main( T & );
1492        thread_desc * get_thread( T & );
1493        void ^?{}( T & `mutex` );
1494};
1495\end{cfa}
1496\end{tabular}
1497\end{cquote}
1498Like coroutines, the @dtype@ property prevents \emph{implicit} copy operations and the @is_thread@ trait provides no \emph{explicit} copy operations, so threads must be passed by reference (pointer).
1499Similarly, the function definitions ensure there is a statically typed @main@ function that is the thread starting point (first stack frame), a mechanism to get (read) the thread descriptor from its handle, and a special destructor to prevent deallocation while the thread is executing.
1500(The qualifier @mutex@ for the destructor parameter is discussed in Section~\ref{s:Monitor}.)
1501The difference between the coroutine and thread is that a coroutine borrows a thread from its caller, so the first thread resuming a coroutine creates the coroutine's stack and starts running the coroutine main on the stack;
1502whereas, a thread is scheduling for execution in @main@ immediately after its constructor is run.
1503No return value or additional parameters are necessary for this function because the @thread@ type allows an arbitrary number of interface functions with corresponding arbitrary typed input/output values.
1504
1505
1506\section{Mutual Exclusion / Synchronization}
1507
1508Unrestricted nondeterminism is meaningless as there is no way to know when the result is completed without synchronization.
1509To produce meaningful execution requires clawing back some determinism using mutual exclusion and synchronization, where mutual exclusion provides access control for threads using shared data, and synchronization is a timing relationship among threads~\cite[\S~4]{Buhr05a}.
1510Some concurrent systems eliminate mutable shared-state by switching to stateless communication like message passing~\cite{Thoth,Harmony,V-Kernel,MPI} (Erlang, MPI), channels~\cite{CSP} (CSP,Go), actors~\cite{Akka} (Akka, Scala), or functional techniques (Haskell).
1511However, these approaches introduce a new communication mechanism for concurrency different from the standard communication using function call/return.
1512Hence, a programmer must learn and manipulate two sets of design/programming patterns.
1513While this distinction can be hidden away in library code, effective use of the library still has to take both paradigms into account.
1514In contrast, approaches based on stateful models more closely resemble the standard call/return programming model, resulting in a single programming paradigm.
1515
1516At the lowest level, concurrent control is implemented by atomic operations, upon which different kinds of locking mechanisms are constructed, \eg semaphores~\cite{Dijkstra68b}, barriers, and path expressions~\cite{Campbell74}.
1517However, for productivity it is always desirable to use the highest-level construct that provides the necessary efficiency~\cite{Hochstein05}.
1518A newer approach for restricting non-determinism is transactional memory~\cite{Herlihy93}.
1519While this approach is pursued in hardware~\cite{Nakaike15} and system languages, like \CC~\cite{Cpp-Transactions}, the performance and feature set is still too restrictive to be the main concurrency paradigm for system languages, which is why it is rejected as the core paradigm for concurrency in \CFA.
1520
1521One of the most natural, elegant, and efficient mechanisms for mutual exclusion and synchronization for shared-memory systems is the \emph{monitor}.
1522First proposed by Brinch Hansen~\cite{Hansen73} and later described and extended by C.A.R.~Hoare~\cite{Hoare74}, many concurrent programming languages provide monitors as an explicit language construct: \eg Concurrent Pascal~\cite{ConcurrentPascal}, Mesa~\cite{Mesa}, Modula~\cite{Modula-2}, Turing~\cite{Turing:old}, Modula-3~\cite{Modula-3}, NeWS~\cite{NeWS}, Emerald~\cite{Emerald}, \uC~\cite{Buhr92a} and Java~\cite{Java}.
1523In addition, operating-system kernels and device drivers have a monitor-like structure, although they often use lower-level primitives such as mutex locks or semaphores to simulate monitors.
1524For these reasons, \CFA selected monitors as the core high-level concurrency construct, upon which higher-level approaches can be easily constructed.
1525
1526
1527\subsection{Mutual Exclusion}
1528
1529A group of instructions manipulating a specific instance of shared data that must be performed atomically is called a \newterm{critical section}~\cite{Dijkstra65}, which is enforced by \newterm{simple mutual-exclusion}.
1530The generalization is called a \newterm{group critical-section}~\cite{Joung00}, where multiple tasks with the same session use the resource simultaneously and different sessions are segregated, which is enforced by \newterm{complex mutual-exclusion} providing the correct kind and number of threads using a group critical-section.
1531The readers/writer problem~\cite{Courtois71} is an instance of a group critical-section, where readers share a session but writers have a unique session.
1532
1533However, many solutions exist for mutual exclusion, which vary in terms of performance, flexibility and ease of use.
1534Methods range from low-level locks, which are fast and flexible but require significant attention for correctness, to higher-level concurrency techniques, which sacrifice some performance to improve ease of use.
1535Ease of use comes by either guaranteeing some problems cannot occur, \eg deadlock free, or by offering a more explicit coupling between shared data and critical section.
1536For example, the \CC @std::atomic<T>@ offers an easy way to express mutual-exclusion on a restricted set of operations, \eg reading/writing, for numerical types.
1537However, a significant challenge with locks is composability because it takes careful organization for multiple locks to be used while preventing deadlock.
1538Easing composability is another feature higher-level mutual-exclusion mechanisms can offer.
1539
1540
1541\subsection{Synchronization}
1542
1543Synchronization enforces relative ordering of execution, and synchronization tools provide numerous mechanisms to establish these timing relationships.
1544Low-level synchronization primitives offer good performance and flexibility at the cost of ease of use;
1545higher-level mechanisms often simplify usage by adding better coupling between synchronization and data, \eg receive-specific versus receive-any thread in message passing or offering specialized solutions, \eg barrier lock.
1546Often synchronization is used to order access to a critical section, \eg ensuring a waiting writer thread enters the critical section before a calling reader thread.
1547If the calling reader is scheduled before the waiting writer, the reader has \newterm{barged}.
1548Barging can result in staleness/freshness problems, where a reader barges ahead of a writer and reads temporally stale data, or a writer barges ahead of another writer overwriting data with a fresh value preventing the previous value from ever being read (lost computation).
1549Preventing or detecting barging is an involved challenge with low-level locks, which is made easier through higher-level constructs.
1550This challenge is often split into two different approaches: barging avoidance and prevention.
1551Algorithms that unconditionally releasing a lock for competing threads to acquire use barging avoidance during synchronization to force a barging thread to wait;
1552algorithms that conditionally hold locks during synchronization, \eg baton-passing~\cite{Andrews89}, prevent barging completely.
1553
1554
1555\section{Monitor}
1556\label{s:Monitor}
1557
1558A \textbf{monitor} is a set of functions that ensure mutual exclusion when accessing shared state.
1559More precisely, a monitor is a programming technique that implicitly binds mutual exclusion to static function scope, as opposed to locks, where mutual-exclusion is defined by acquire/release calls, independent of lexical context (analogous to block and heap storage allocation).
1560Restricting acquire/release points eases programming, comprehension, and maintenance, at a slight cost in flexibility and efficiency.
1561\CFA uses a custom @monitor@ type and leverages declaration semantics (deallocation) to protect active or waiting threads in a monitor.
1562
1563The following is a \CFA monitor implementation of an atomic counter.
1564\begin{cfa}[morekeywords=nomutex]
1565`monitor` Aint { int cnt; }; $\C[4.25in]{// atomic integer counter}$
1566int ++?( Aint & `mutex`$\(_{opt}\)$ this ) with( this ) { return ++cnt; } $\C{// increment}$
1567int ?=?( Aint & `mutex`$\(_{opt}\)$ lhs, int rhs ) with( lhs ) { cnt = rhs; } $\C{// conversions with int}\CRT$
1568int ?=?( int & lhs, Aint & `mutex`$\(_{opt}\)$ rhs ) with( rhs ) { lhs = cnt; }
1569\end{cfa}
1570% The @Aint@ constructor, @?{}@, uses the \lstinline[morekeywords=nomutex]@nomutex@ qualifier indicating mutual exclusion is unnecessary during construction because an object is inaccessible (private) until after it is initialized.
1571% (While a constructor may publish its address into a global variable, doing so generates a race-condition.)
1572The prefix increment operation, @++?@, is normally @mutex@, indicating mutual exclusion is necessary during function execution, to protect the incrementing from race conditions, unless there is an atomic increment instruction for the implementation type.
1573The assignment operators provide bidirectional conversion between an atomic and normal integer without accessing field @cnt@;
1574these operations only need @mutex@, if reading/writing the implementation type is not atomic.
1575The atomic counter is used without any explicit mutual-exclusion and provides thread-safe semantics, which is similar to the \CC template @std::atomic@.
1576\begin{cfa}
1577int i = 0, j = 0, k = 5;
1578Aint x = { 0 }, y = { 0 }, z = { 5 }; $\C{// no mutex required}$
1579++x; ++y; ++z; $\C{// safe increment by multiple threads}$
1580x = 2; y = i; z = k; $\C{// conversions}$
1581i = x; j = y; k = z;
1582\end{cfa}
1583
1584\CFA monitors have \newterm{multi-acquire} semantics so the thread in the monitor may acquire it multiple times without deadlock, allowing recursion and calling other interface functions.
1585\begin{cfa}
1586monitor M { ... } m;
1587void foo( M & mutex m ) { ... } $\C{// acquire mutual exclusion}$
1588void bar( M & mutex m ) { $\C{// acquire mutual exclusion}$
1589        ... `bar( m );` ... `foo( m );` ... $\C{// reacquire mutual exclusion}$
1590}
1591\end{cfa}
1592\CFA monitors also ensure the monitor lock is released regardless of how an acquiring function ends (normal or exceptional), and returning a shared variable is safe via copying before the lock is released.
1593Similar safety is offered by \emph{explicit} mechanisms like \CC RAII;
1594monitor \emph{implicit} safety ensures no programmer usage errors.
1595Furthermore, RAII mechanisms cannot handle complex synchronization within a monitor, where the monitor lock may not be released on function exit because it is passed to an unblocking thread;
1596RAII is purely a mutual-exclusion mechanism (see Section~\ref{s:Scheduling}).
1597
1598
1599\subsection{Monitor Implementation}
1600
1601For the same design reasons, \CFA provides a custom @monitor@ type and a @trait@ to enforce and restrict the monitor-interface functions.
1602\begin{cquote}
1603\begin{tabular}{@{}c@{\hspace{3\parindentlnth}}c@{}}
1604\begin{cfa}
1605monitor M {
1606        ... // shared data
1607};
1608
1609\end{cfa}
1610&
1611\begin{cfa}
1612trait is_monitor( `dtype` T ) {
1613        monitor_desc * get_monitor( T & );
1614        void ^?{}( T & mutex );
1615};
1616\end{cfa}
1617\end{tabular}
1618\end{cquote}
1619The @dtype@ property prevents \emph{implicit} copy operations and the @is_monitor@ trait provides no \emph{explicit} copy operations, so monitors must be passed by reference (pointer).
1620% Copying a lock is insecure because it is possible to copy an open lock and then use the open copy when the original lock is closed to simultaneously access the shared data.
1621% Copying a monitor is secure because both the lock and shared data are copies, but copying the shared data is meaningless because it no longer represents a unique entity.
1622Similarly, the function definitions ensures there is a mechanism to get (read) the monitor descriptor from its handle, and a special destructor to prevent deallocation if a thread using the shared data.
1623The custom monitor type also inserts any locks needed to implement the mutual exclusion semantics.
1624
1625
1626\subsection{Mutex Acquisition}
1627\label{s:MutexAcquisition}
1628
1629While the monitor lock provides mutual exclusion for shared data, there are implementation options for when and where the locking/unlocking occurs.
1630(Much of this discussion also applies to basic locks.)
1631For example, a monitor may be passed through multiple helper functions before it is necessary to acquire the monitor's mutual exclusion.
1632
1633The benefit of mandatory monitor qualifiers is self-documentation, but requiring both @mutex@ and \lstinline[morekeywords=nomutex]@nomutex@ for all monitor parameters is redundant.
1634Instead, the semantics has one qualifier as the default and the other required.
1635For example, make the safe @mutex@ qualifier the default because assuming \lstinline[morekeywords=nomutex]@nomutex@ may cause subtle errors.
1636Alternatively, make the unsafe \lstinline[morekeywords=nomutex]@nomutex@ qualifier the default because it is the \emph{normal} parameter semantics while @mutex@ parameters are rare.
1637Providing a default qualifier implies knowing whether a parameter is a monitor.
1638Since \CFA relies heavily on traits as an abstraction mechanism, types can coincidentally match the monitor trait but not be a monitor, similar to inheritance where a shape and playing card can both be drawable.
1639For this reason, \CFA requires programmers to identify the kind of parameter with the @mutex@ keyword and uses no keyword to mean \lstinline[morekeywords=nomutex]@nomutex@.
1640
1641The next semantic decision is establishing which parameter \emph{types} may be qualified with @mutex@.
1642The following has monitor parameter types that are composed of multiple objects.
1643\begin{cfa}
1644monitor M { ... }
1645int f1( M & mutex m ); $\C{// single parameter object}$
1646int f2( M * mutex m ); $\C{// single or multiple parameter object}$
1647int f3( M * mutex m[$\,$] ); $\C{// multiple parameter object}$
1648int f4( stack( M * ) & mutex m ); $\C{// multiple parameters object}$
1649\end{cfa}
1650Function @f1@ has a single parameter object, while @f2@'s indirection could be a single or multi-element array, where static array size is often unknown in C.
1651Function @f3@ has a multiple object matrix, and @f4@ a multiple object data structure.
1652While shown shortly, multiple object acquisition is possible, but the number of objects must be statically known.
1653Therefore, \CFA only acquires one monitor per parameter with at most one level of indirection, excluding pointers as it is impossible to statically determine the size.
1654
1655For object-oriented monitors, \eg Java, calling a mutex member \emph{implicitly} acquires mutual exclusion of the receiver object, @`rec`.foo(...)@.
1656\CFA has no receiver, and hence, the explicit @mutex@ qualifier is used to specify which objects acquire mutual exclusion.
1657A positive consequence of this design decision is the ability to support multi-monitor functions,\footnote{
1658While object-oriented monitors can be extended with a mutex qualifier for multiple-monitor members, no prior example of this feature could be found.}
1659called \newterm{bulk acquire}.
1660\CFA guarantees acquisition order is consistent across calls to @mutex@ functions using the same monitors as arguments, so acquiring multiple monitors is safe from deadlock.
1661Figure~\ref{f:BankTransfer} shows a trivial solution to the bank transfer problem~\cite{BankTransfer}, where two resources must be locked simultaneously, using \CFA monitors with implicit locking and \CC with explicit locking.
1662A \CFA programmer only has to manage when to acquire mutual exclusion;
1663a \CC programmer must select the correct lock and acquisition mechanism from a panoply of locking options.
1664Making good choices for common cases in \CFA simplifies the programming experience and enhances safety.
1665
1666\begin{figure}
1667\centering
1668\begin{lrbox}{\myboxA}
1669\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1670monitor BankAccount {
1671
1672        int balance;
1673} b1 = { 0 }, b2 = { 0 };
1674void deposit( BankAccount & `mutex` b,
1675                        int deposit ) with(b) {
1676        balance += deposit;
1677}
1678void transfer( BankAccount & `mutex` my,
1679        BankAccount & `mutex` your, int me2you ) {
1680
1681        deposit( my, -me2you ); // debit
1682        deposit( your, me2you ); // credit
1683}
1684`thread` Person { BankAccount & b1, & b2; };
1685void main( Person & person ) with(person) {
1686        for ( 10_000_000 ) {
1687                if ( random() % 3 ) deposit( b1, 3 );
1688                if ( random() % 3 ) transfer( b1, b2, 7 );
1689        }
1690}   
1691int main() {
1692        `Person p1 = { b1, b2 }, p2 = { b2, b1 };`
1693
1694} // wait for threads to complete
1695\end{cfa}
1696\end{lrbox}
1697
1698\begin{lrbox}{\myboxB}
1699\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1700struct BankAccount {
1701        `recursive_mutex m;`
1702        int balance = 0;
1703} b1, b2;
1704void deposit( BankAccount & b, int deposit ) {
1705        `scoped_lock lock( b.m );`
1706        b.balance += deposit;
1707}
1708void transfer( BankAccount & my,
1709                        BankAccount & your, int me2you ) {
1710        `scoped_lock lock( my.m, your.m );`
1711        deposit( my, -me2you ); // debit
1712        deposit( your, me2you ); // credit
1713}
1714
1715void person( BankAccount & b1, BankAccount & b2 ) {
1716        for ( int i = 0; i < 10$'$000$'$000; i += 1 ) {
1717                if ( random() % 3 ) deposit( b1, 3 );
1718                if ( random() % 3 ) transfer( b1, b2, 7 );
1719        }
1720}   
1721int main() {
1722        `thread p1(person, ref(b1), ref(b2)), p2(person, ref(b2), ref(b1));`
1723        `p1.join(); p2.join();`
1724}
1725\end{cfa}
1726\end{lrbox}
1727
1728\subfloat[\CFA]{\label{f:CFABank}\usebox\myboxA}
1729\hspace{3pt}
1730\vrule
1731\hspace{3pt}
1732\subfloat[\CC]{\label{f:C++Bank}\usebox\myboxB}
1733\hspace{3pt}
1734\caption{Bank transfer problem}
1735\label{f:BankTransfer}
1736\end{figure}
1737
1738Users can still force the acquiring order by using @mutex@/\lstinline[morekeywords=nomutex]@nomutex@.
1739\begin{cfa}
1740void foo( M & mutex m1, M & mutex m2 ); $\C{// acquire m1 and m2}$
1741void bar( M & mutex m1, M & /* nomutex */ m2 ) { $\C{// acquire m1}$
1742        ... foo( m1, m2 ); ... $\C{// acquire m2}$
1743}
1744void baz( M & /* nomutex */ m1, M & mutex m2 ) { $\C{// acquire m2}$
1745        ... foo( m1, m2 ); ... $\C{// acquire m1}$
1746}
1747\end{cfa}
1748The bulk-acquire semantics allow @bar@ or @baz@ to acquire a monitor lock and reacquire it in @foo@.
1749The calls to @bar@ and @baz@ acquired the monitors in opposite order, possibly resulting in deadlock.
1750However, this case is the simplest instance of the \emph{nested-monitor problem}~\cite{Lister77}, where monitors are acquired in sequence versus bulk.
1751Detecting the nested-monitor problem requires dynamic tracking of monitor calls, and dealing with it requires rollback semantics~\cite{Dice10}.
1752\CFA does not deal with this fundamental problem.
1753
1754Finally, like Java, \CFA offers an alternative @mutex@ statement to reduce refactoring and naming.
1755\begin{cquote}
1756\renewcommand{\arraystretch}{0.0}
1757\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
1758\multicolumn{1}{c}{\textbf{\lstinline@mutex@ call}} & \multicolumn{1}{c}{\lstinline@mutex@ \textbf{statement}} \\
1759\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1760monitor M { ... };
1761void foo( M & mutex m1, M & mutex m2 ) {
1762        // critical section
1763}
1764void bar( M & m1, M & m2 ) {
1765        foo( m1, m2 );
1766}
1767\end{cfa}
1768&
1769\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1770
1771void bar( M & m1, M & m2 ) {
1772        mutex( m1, m2 ) {       // remove refactoring and naming
1773                // critical section
1774        }
1775}
1776
1777\end{cfa}
1778\end{tabular}
1779\end{cquote}
1780
1781
1782\subsection{Scheduling}
1783\label{s:Scheduling}
1784
1785% There are many aspects of scheduling in a concurrency system, all related to resource utilization by waiting threads, \ie which thread gets the resource next.
1786% Different forms of scheduling include access to processors by threads (see Section~\ref{s:RuntimeStructureCluster}), another is access to a shared resource by a lock or monitor.
1787This section discusses monitor scheduling for waiting threads eligible for entry, \ie which thread gets the shared resource next. (See Section~\ref{s:RuntimeStructureCluster} for scheduling threads on virtual processors.)
1788While monitor mutual-exclusion provides safe access to shared data, the monitor data may indicate that a thread accessing it cannot proceed, \eg a bounded buffer may be full/empty so produce/consumer threads must block.
1789Leaving the monitor and trying again (busy waiting) is impractical for high-level programming.
1790Monitors eliminate busy waiting by providing synchronization to schedule threads needing access to the shared data, where threads block versus spinning.
1791Synchronization is generally achieved with internal~\cite{Hoare74} or external~\cite[\S~2.9.2]{uC++} scheduling.
1792\newterm{Internal scheduling} is characterized by each thread entering the monitor and making an individual decision about proceeding or blocking, while \newterm{external scheduling} is characterized by an entering thread making a decision about proceeding for itself and on behalf of other threads attempting entry.
1793Finally, \CFA monitors do not allow calling threads to barge ahead of signalled threads, which simplifies synchronization among threads in the monitor and increases correctness.
1794If barging is allowed, synchronization between a signaller and signallee is difficult, often requiring additional flags and multiple unblock/block cycles.
1795In fact, signals-as-hints is completely opposite from that proposed by Hoare in the seminal paper on monitors~\cite[p.~550]{Hoare74}.
1796% \begin{cquote}
1797% However, we decree that a signal operation be followed immediately by resumption of a waiting program, without possibility of an intervening procedure call from yet a third program.
1798% It is only in this way that a waiting program has an absolute guarantee that it can acquire the resource just released by the signalling program without any danger that a third program will interpose a monitor entry and seize the resource instead.~\cite[p.~550]{Hoare74}
1799% \end{cquote}
1800Furthermore, \CFA concurrency has no spurious wakeup~\cite[\S~9]{Buhr05a}, which eliminates an implicit form of self barging.
1801Hence, a \CFA @wait@ statement is not enclosed in a @while@ loop retesting a blocking predicate, which can cause thread starvation due to barging.
1802
1803Figure~\ref{f:MonitorScheduling} shows general internal/external scheduling (for the bounded-buffer example in Figure~\ref{f:InternalExternalScheduling}).
1804External calling threads block on the calling queue, if the monitor is occupied, otherwise they enter in FIFO order.
1805Internal threads block on condition queues via @wait@ and reenter from the condition in FIFO order.
1806Alternatively, internal threads block on urgent from the @signal_block@ or @waitfor@, and reenter implicitly when the monitor becomes empty, \ie, the thread in the monitor exits or waits.
1807
1808There are three signalling mechanisms to unblock waiting threads to enter the monitor.
1809Note, signalling cannot have the signaller and signalled thread in the monitor simultaneously because of the mutual exclusion, so either the signaller or signallee can proceed.
1810For internal scheduling, threads are unblocked from condition queues using @signal@, where the signallee is moved to urgent and the signaller continues (solid line).
1811Multiple signals move multiple signallees to urgent until the condition is empty.
1812When the signaller exits or waits, a thread blocked on urgent is processed before calling threads to prevent barging.
1813(Java conceptually moves the signalled thread to the calling queue, and hence, allows barging.)
1814The alternative unblock is in the opposite order using @signal_block@, where the signaller is moved to urgent and the signallee continues (dashed line), and is implicitly unblocked from urgent when the signallee exits or waits.
1815
1816For external scheduling, the condition queues are not used;
1817instead threads are unblocked directly from the calling queue using @waitfor@ based on function names requesting mutual exclusion.
1818(The linear search through the calling queue to locate a particular call can be reduced to $O(1)$.)
1819The @waitfor@ has the same semantics as @signal_block@, where the signalled thread executes before the signallee, which waits on urgent.
1820Executing multiple @waitfor@s from different signalled functions causes the calling threads to move to urgent.
1821External scheduling requires urgent to be a stack, because the signaller expects to execute immediately after the specified monitor call has exited or waited.
1822Internal scheduling behaves the same for an urgent stack or queue, except for multiple signalling, where the threads unblock from urgent in reverse order from signalling.
1823If the restart order is important, multiple signalling by a signal thread can be transformed into daisy-chain signalling among threads, where each thread signals the next thread.
1824We tried both a stack for @waitfor@ and queue for signalling, but that resulted in complex semantics about which thread enters next.
1825Hence, \CFA uses a single urgent stack to correctly handle @waitfor@ and adequately support both forms of signalling.
1826
1827\begin{figure}
1828\centering
1829% \subfloat[Scheduling Statements] {
1830% \label{fig:SchedulingStatements}
1831% {\resizebox{0.45\textwidth}{!}{\input{CondSigWait.pstex_t}}}
1832\input{CondSigWait.pstex_t}
1833% }% subfloat
1834% \quad
1835% \subfloat[Bulk acquire monitor] {
1836% \label{fig:BulkMonitor}
1837% {\resizebox{0.45\textwidth}{!}{\input{ext_monitor.pstex_t}}}
1838% }% subfloat
1839\caption{Monitor Scheduling}
1840\label{f:MonitorScheduling}
1841\end{figure}
1842
1843Figure~\ref{f:BBInt} shows a \CFA generic bounded-buffer with internal scheduling, where producers/consumers enter the monitor, detect the buffer is full/empty, and block on an appropriate condition variable, @full@/@empty@.
1844The @wait@ function atomically blocks the calling thread and implicitly releases the monitor lock(s) for all monitors in the function's parameter list.
1845The appropriate condition variable is signalled to unblock an opposite kind of thread after an element is inserted/removed from the buffer.
1846Signalling is unconditional, because signalling an empty condition variable does nothing.
1847It is common to declare condition variables as monitor fields to prevent shared access, hence no locking is required for access as the conditions are protected by the monitor lock.
1848In \CFA, a condition variable can be created/stored independently.
1849% To still prevent expensive locking on access, a condition variable is tied to a \emph{group} of monitors on first use, called \newterm{branding}, resulting in a low-cost boolean test to detect sharing from other monitors.
1850
1851% Signalling semantics cannot have the signaller and signalled thread in the monitor simultaneously, which means:
1852% \begin{enumerate}
1853% \item
1854% The signalling thread returns immediately and the signalled thread continues.
1855% \item
1856% The signalling thread continues and the signalled thread is marked for urgent unblocking at the next scheduling point (exit/wait).
1857% \item
1858% The signalling thread blocks but is marked for urgent unblocking at the next scheduling point and the signalled thread continues.
1859% \end{enumerate}
1860% The first approach is too restrictive, as it precludes solving a reasonable class of problems, \eg dating service (see Figure~\ref{f:DatingService}).
1861% \CFA supports the next two semantics as both are useful.
1862
1863\begin{figure}
1864\centering
1865\newbox\myboxA
1866\begin{lrbox}{\myboxA}
1867\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1868forall( otype T ) { // distribute forall
1869        monitor Buffer {
1870                `condition` full, empty;
1871                int front, back, count;
1872                T elements[10];
1873        };
1874        void ?{}( Buffer(T) & buffer ) with(buffer) {
1875                front = back = count = 0;
1876        }
1877        void insert( Buffer(T) & mutex buffer, T elem )
1878                                with(buffer) {
1879                if ( count == 10 ) `wait( empty )`;
1880                // insert elem into buffer
1881                `signal( full )`;
1882        }
1883        T remove( Buffer(T) & mutex buffer ) with(buffer) {
1884                if ( count == 0 ) `wait( full )`;
1885                // remove elem from buffer
1886                `signal( empty )`;
1887                return elem;
1888        }
1889}
1890\end{cfa}
1891\end{lrbox}
1892
1893% \newbox\myboxB
1894% \begin{lrbox}{\myboxB}
1895% \begin{cfa}[aboveskip=0pt,belowskip=0pt]
1896% forall( otype T ) { // distribute forall
1897%       monitor Buffer {
1898%
1899%               int front, back, count;
1900%               T elements[10];
1901%       };
1902%       void ?{}( Buffer(T) & buffer ) with(buffer) {
1903%               [front, back, count] = 0;
1904%       }
1905%       T remove( Buffer(T) & mutex buffer ); // forward
1906%       void insert( Buffer(T) & mutex buffer, T elem )
1907%                               with(buffer) {
1908%               if ( count == 10 ) `waitfor( remove, buffer )`;
1909%               // insert elem into buffer
1910%
1911%       }
1912%       T remove( Buffer(T) & mutex buffer ) with(buffer) {
1913%               if ( count == 0 ) `waitfor( insert, buffer )`;
1914%               // remove elem from buffer
1915%
1916%               return elem;
1917%       }
1918% }
1919% \end{cfa}
1920% \end{lrbox}
1921
1922\newbox\myboxB
1923\begin{lrbox}{\myboxB}
1924\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1925monitor ReadersWriter {
1926        int rcnt, wcnt; // readers/writer using resource
1927};
1928void ?{}( ReadersWriter & rw ) with(rw) {
1929        rcnt = wcnt = 0;
1930}
1931void EndRead( ReadersWriter & mutex rw ) with(rw) {
1932        rcnt -= 1;
1933}
1934void EndWrite( ReadersWriter & mutex rw ) with(rw) {
1935        wcnt = 0;
1936}
1937void StartRead( ReadersWriter & mutex rw ) with(rw) {
1938        if ( wcnt > 0 ) `waitfor( EndWrite, rw );`
1939        rcnt += 1;
1940}
1941void StartWrite( ReadersWriter & mutex rw ) with(rw) {
1942        if ( wcnt > 0 ) `waitfor( EndWrite, rw );`
1943        else while ( rcnt > 0 ) `waitfor( EndRead, rw );`
1944        wcnt = 1;
1945}
1946
1947\end{cfa}
1948\end{lrbox}
1949
1950\subfloat[Generic bounded buffer, internal scheduling]{\label{f:BBInt}\usebox\myboxA}
1951\hspace{3pt}
1952\vrule
1953\hspace{3pt}
1954\subfloat[Readers / writer lock, external scheduling]{\label{f:RWExt}\usebox\myboxB}
1955
1956\caption{Internal / external scheduling}
1957\label{f:InternalExternalScheduling}
1958\end{figure}
1959
1960Figure~\ref{f:BBInt} can be transformed into external scheduling by removing the condition variables and signals/waits, and adding the following lines at the locations of the current @wait@s in @insert@/@remove@, respectively.
1961\begin{cfa}[aboveskip=2pt,belowskip=1pt]
1962if ( count == 10 ) `waitfor( remove, buffer )`;       |      if ( count == 0 ) `waitfor( insert, buffer )`;
1963\end{cfa}
1964Here, the producers/consumers detects a full/\-empty buffer and prevents more producers/consumers from entering the monitor until there is a free/empty slot in the buffer.
1965External scheduling is controlled by the @waitfor@ statement, which atomically blocks the calling thread, releases the monitor lock, and restricts the function calls that can next acquire mutual exclusion.
1966If the buffer is full, only calls to @remove@ can acquire the buffer, and if the buffer is empty, only calls to @insert@ can acquire the buffer.
1967Threads calling excluded functions block outside of (external to) the monitor on the calling queue, versus blocking on condition queues inside of (internal to) the monitor.
1968Figure~\ref{f:RWExt} shows a readers/writer lock written using external scheduling, where a waiting reader detects a writer using the resource and restricts further calls until the writer exits by calling @EndWrite@.
1969The writer does a similar action for each reader or writer using the resource.
1970Note, no new calls to @StarRead@/@StartWrite@ may occur when waiting for the call to @EndRead@/@EndWrite@.
1971External scheduling allows waiting for events from other threads while restricting unrelated events, that would otherwise have to wait on conditions in the monitor.
1972The mechnaism can be done in terms of control flow, \eg Ada @accept@ or \uC @_Accept@, or in terms of data, \eg Go @select@ on channels.
1973While both mechanisms have strengths and weaknesses, this project uses the control-flow mechanism to be consistent with other language features.
1974% Two challenges specific to \CFA for external scheduling are loose object-definitions (see Section~\ref{s:LooseObjectDefinitions}) and multiple-monitor functions (see Section~\ref{s:Multi-MonitorScheduling}).
1975
1976Figure~\ref{f:DatingService} shows a dating service demonstrating non-blocking and blocking signalling.
1977The dating service matches girl and boy threads with matching compatibility codes so they can exchange phone numbers.
1978A thread blocks until an appropriate partner arrives.
1979The complexity is exchanging phone numbers in the monitor because of the mutual-exclusion property.
1980For signal scheduling, the @exchange@ condition is necessary to block the thread finding the match, while the matcher unblocks to take the opposite number, post its phone number, and unblock the partner.
1981For signal-block scheduling, the implicit urgent-queue replaces the explict @exchange@-condition and @signal_block@ puts the finding thread on the urgent condition and unblocks the matcher.
1982The dating service is an example of a monitor that cannot be written using external scheduling because it requires knowledge of calling parameters to make scheduling decisions, and parameters of waiting threads are unavailable;
1983as well, an arriving thread may not find a partner and must wait, which requires a condition variable, and condition variables imply internal scheduling.
1984Furthermore, barging corrupts the dating service during an exchange because a barger may also match and change the phone numbers, invalidating the previous exchange phone number.
1985Putting loops around the @wait@s does not correct the problem;
1986the simple solution must be restructured to account for barging.
1987
1988\begin{figure}
1989\centering
1990\newbox\myboxA
1991\begin{lrbox}{\myboxA}
1992\begin{cfa}[aboveskip=0pt,belowskip=0pt]
1993enum { CCodes = 20 };
1994monitor DS {
1995        int GirlPhNo, BoyPhNo;
1996        condition Girls[CCodes], Boys[CCodes];
1997        `condition exchange;`
1998};
1999int girl( DS & mutex ds, int phNo, int ccode ) {
2000        if ( is_empty( Boys[ccode] ) ) {
2001                wait( Girls[ccode] );
2002                GirlPhNo = phNo;
2003                `signal( exchange );`
2004        } else {
2005                GirlPhNo = phNo;
2006                `signal( Boys[ccode] );`
2007                `wait( exchange );`
2008        }
2009        return BoyPhNo;
2010}
2011int boy( DS & mutex ds, int phNo, int ccode ) {
2012        // as above with boy/girl interchanged
2013}
2014\end{cfa}
2015\end{lrbox}
2016
2017\newbox\myboxB
2018\begin{lrbox}{\myboxB}
2019\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2020
2021monitor DS {
2022        int GirlPhNo, BoyPhNo;
2023        condition Girls[CCodes], Boys[CCodes];
2024
2025};
2026int girl( DS & mutex ds, int phNo, int ccode ) {
2027        if ( is_empty( Boys[ccode] ) ) { // no compatible
2028                wait( Girls[ccode] ); // wait for boy
2029                GirlPhNo = phNo; // make phone number available
2030
2031        } else {
2032                GirlPhNo = phNo; // make phone number available
2033                `signal_block( Boys[ccode] );` // restart boy
2034
2035        } // if
2036        return BoyPhNo;
2037}
2038int boy( DS & mutex ds, int phNo, int ccode ) {
2039        // as above with boy/girl interchanged
2040}
2041\end{cfa}
2042\end{lrbox}
2043
2044\subfloat[\lstinline@signal@]{\label{f:DatingSignal}\usebox\myboxA}
2045\qquad
2046\subfloat[\lstinline@signal_block@]{\label{f:DatingSignalBlock}\usebox\myboxB}
2047\caption{Dating service}
2048\label{f:DatingService}
2049\end{figure}
2050
2051In summation, for internal scheduling, non-blocking signalling (as in the producer/consumer example) is used when the signaller is providing the cooperation for a waiting thread;
2052the signaller enters the monitor and changes state, detects a waiting threads that can use the state, performs a non-blocking signal on the condition queue for the waiting thread, and exits the monitor to run concurrently.
2053The waiter unblocks next from the urgent queue, uses/takes the state, and exits the monitor.
2054Blocking signal is the reverse, where the waiter is providing the cooperation for the signalling thread;
2055the signaller enters the monitor, detects a waiting thread providing the necessary state, performs a blocking signal to place it on the urgent queue and unblock the waiter.
2056The waiter changes state and exits the monitor, and the signaller unblocks next from the urgent queue to use/take the state.
2057
2058Both internal and external scheduling extend to multiple monitors in a natural way.
2059\begin{cquote}
2060\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
2061\begin{cfa}
2062monitor M { `condition e`; ... };
2063void foo( M & mutex m1, M & mutex m2 ) {
2064        ... wait( `e` ); ...   // wait( e, m1, m2 )
2065        ... wait( `e, m1` ); ...
2066        ... wait( `e, m2` ); ...
2067}
2068\end{cfa}
2069&
2070\begin{cfa}
2071void rtn$\(_1\)$( M & mutex m1, M & mutex m2 );
2072void rtn$\(_2\)$( M & mutex m1 );
2073void bar( M & mutex m1, M & mutex m2 ) {
2074        ... waitfor( `rtn` ); ...       // $\LstCommentStyle{waitfor( rtn\(_1\), m1, m2 )}$
2075        ... waitfor( `rtn, m1` ); ... // $\LstCommentStyle{waitfor( rtn\(_2\), m1 )}$
2076}
2077\end{cfa}
2078\end{tabular}
2079\end{cquote}
2080For @wait( e )@, the default semantics is to atomically block the signaller and release all acquired mutex parameters, \ie @wait( e, m1, m2 )@.
2081To override the implicit multi-monitor wait, specific mutex parameter(s) can be specified, \eg @wait( e, m1 )@.
2082Wait cannot statically verifies the released monitors are the acquired mutex-parameters without disallowing separately compiled helper functions calling @wait@.
2083While \CC supports bulk locking, @wait@ only accepts a single lock for a condition variable, so bulk locking with condition variables is asymmetric.
2084Finally, a signaller,
2085\begin{cfa}
2086void baz( M & mutex m1, M & mutex m2 ) {
2087        ... signal( e ); ...
2088}
2089\end{cfa}
2090must have acquired at least the same locks as the waiting thread signalled from a condition queue to allow the locks to be passed, and hence, prevent barging.
2091
2092Similarly, for @waitfor( rtn )@, the default semantics is to atomically block the acceptor and release all acquired mutex parameters, \ie @waitfor( rtn, m1, m2 )@.
2093To override the implicit multi-monitor wait, specific mutex parameter(s) can be specified, \eg @waitfor( rtn, m1 )@.
2094@waitfor@ does statically verify the monitor types passed are the same as the acquired mutex-parameters of the given function or function pointer, hence the function (pointer) prototype must be accessible.
2095% When an overloaded function appears in an @waitfor@ statement, calls to any function with that name are accepted.
2096% The rationale is that members with the same name should perform a similar function, and therefore, all should be eligible to accept a call.
2097Overloaded functions can be disambiguated using a cast
2098\begin{cfa}
2099void rtn( M & mutex m );
2100`int` rtn( M & mutex m );
2101waitfor( (`int` (*)( M & mutex ))rtn, m );
2102\end{cfa}
2103
2104The ability to release a subset of acquired monitors can result in a \newterm{nested monitor}~\cite{Lister77} deadlock.
2105\begin{cfa}
2106void foo( M & mutex m1, M & mutex m2 ) {
2107        ... wait( `e, m1` ); ...                                $\C{// release m1, keeping m2 acquired )}$
2108void bar( M & mutex m1, M & mutex m2 ) {        $\C{// must acquire m1 and m2 )}$
2109        ... signal( `e` ); ...
2110\end{cfa}
2111The @wait@ only releases @m1@ so the signalling thread cannot acquire @m1@ and @m2@ to enter @bar@ and @signal@ the condition.
2112While deadlock can occur with multiple/nesting acquisition, this is a consequence of locks, and by extension monitors, not being perfectly composable.
2113
2114
2115
2116\subsection{Extended \protect\lstinline@waitfor@}
2117
2118Figure~\ref{f:ExtendedWaitfor} show the extended form of the @waitfor@ statement to conditionally accept one of a group of mutex functions, with an optional statement to be performed \emph{after} the mutex function finishes.
2119For a @waitfor@ clause to be executed, its @when@ must be true and an outstanding call to its corresponding member(s) must exist.
2120The \emph{conditional-expression} of a @when@ may call a function, but the function must not block or context switch.
2121If there are multiple acceptable mutex calls, selection occurs top-to-bottom (prioritized) among the @waitfor@ clauses, whereas some programming languages with similar mechanisms accept nondeterministically for this case, \eg Go \lstinline[morekeywords=select]@select@.
2122If some accept guards are true and there are no outstanding calls to these members, the acceptor is blocked until a call to one of these members is made.
2123If there is a @timeout@ clause, it provides an upper bound on waiting.
2124If all the accept guards are false, the statement does nothing, unless there is a terminating @else@ clause with a true guard, which is executed instead.
2125Hence, the terminating @else@ clause allows a conditional attempt to accept a call without blocking.
2126If both @timeout@ and @else@ clause are present, the @else@ must be conditional, or the @timeout@ is never triggered.
2127
2128\begin{figure}
2129\centering
2130\begin{cfa}
2131`when` ( $\emph{conditional-expression}$ )      $\C{// optional guard}$
2132        waitfor( $\emph{mutex-member-name}$ ) $\emph{statement}$ $\C{// action after call}$
2133`or` `when` ( $\emph{conditional-expression}$ ) $\C{// any number of functions}$
2134        waitfor( $\emph{mutex-member-name}$ ) $\emph{statement}$
2135`or`    ...
2136`when` ( $\emph{conditional-expression}$ ) $\C{// optional guard}$
2137        `timeout` $\emph{statement}$ $\C{// optional terminating timeout clause}$
2138`when` ( $\emph{conditional-expression}$ ) $\C{// optional guard}$
2139        `else`  $\emph{statement}$ $\C{// optional terminating clause}$
2140\end{cfa}
2141\caption{Extended \protect\lstinline@waitfor@}
2142\label{f:ExtendedWaitfor}
2143\end{figure}
2144
2145Note, a group of conditional @waitfor@ clauses is \emph{not} the same as a group of @if@ statements, \eg:
2146\begin{cfa}
2147if ( C1 ) waitfor( mem1 );                       when ( C1 ) waitfor( mem1 );
2148else if ( C2 ) waitfor( mem2 );         or when ( C2 ) waitfor( mem2 );
2149\end{cfa}
2150The left example only accepts @mem1@ if @C1@ is true or only @mem2@ if @C2@ is true.
2151The right example accepts either @mem1@ or @mem2@ if @C1@ and @C2@ are true.
2152
2153An interesting use of @waitfor@ is accepting the @mutex@ destructor to know when an object is deallocated, \eg assume the bounded buffer is restructred from a monitor to a thread with the following @main@.
2154\begin{cfa}
2155void main( Buffer(T) & buffer ) with(buffer) {
2156        for () {
2157                `waitfor( ^?{}, buffer )` break;
2158                or when ( count != 20 ) waitfor( insert, buffer ) { ... }
2159                or when ( count != 0 ) waitfor( remove, buffer ) { ... }
2160        }
2161        // clean up
2162}
2163\end{cfa}
2164When the program main deallocates the buffer, it first calls the buffer's destructor, which is accepted, the destructor runs, and the buffer is deallocated.
2165However, the buffer thread cannot continue after the destructor call because the object is gone;
2166hence, clean up in @main@ cannot occur, which means destructors for local objects are not run.
2167To make this useful capability work, the semantics for accepting the destructor is the same as @signal@, \ie the destructor call is placed on urgent and the acceptor continues execution, which ends the loop, cleans up, and the thread terminates.
2168Then, the destructor caller unblocks from urgent to deallocate the object.
2169Accepting the destructor is the idiomatic way in \CFA to terminate a thread performing direct communication.
2170
2171
2172\subsection{Bulk Barging Prevention}
2173
2174Figure~\ref{f:BulkBargingPrevention} shows \CFA code where bulk acquire adds complexity to the internal-signalling semantics.
2175The complexity begins at the end of the inner @mutex@ statement, where the semantics of internal scheduling need to be extended for multiple monitors.
2176The problem is that bulk acquire is used in the inner @mutex@ statement where one of the monitors is already acquired.
2177When the signalling thread reaches the end of the inner @mutex@ statement, it should transfer ownership of @m1@ and @m2@ to the waiting threads to prevent barging into the outer @mutex@ statement by another thread.
2178However, both the signalling and waiting threads W1 and W2 need some subset of monitors @m1@ and @m2@.
2179\begin{cquote}
2180condition c: (order 1) W2(@m2@), W1(@m1@,@m2@)\ \ \ or\ \ \ (order 2) W1(@m1@,@m2@), W2(@m2@) \\
2181S: acq. @m1@ $\rightarrow$ acq. @m1,m2@ $\rightarrow$ @signal(c)@ $\rightarrow$ rel. @m2@ $\rightarrow$ pass @m2@ unblock W2 (order 2) $\rightarrow$ rel. @m1@ $\rightarrow$ pass @m1,m2@ unblock W1 \\
2182\hspace*{2.75in}$\rightarrow$ rel. @m1@ $\rightarrow$ pass @m1,m2@ unblock W1 (order 1)
2183\end{cquote}
2184
2185\begin{figure}
2186\newbox\myboxA
2187\begin{lrbox}{\myboxA}
2188\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2189monitor M m1, m2;
2190condition c;
2191mutex( m1 ) { // $\LstCommentStyle{\color{red}outer}$
2192        ...
2193        mutex( m1, m2 ) { // $\LstCommentStyle{\color{red}inner}$
2194                ... `signal( c )`; ...
2195                // m1, m2 still acquired
2196        } // $\LstCommentStyle{\color{red}release m2}$
2197        // m1 acquired
2198} // release m1
2199\end{cfa}
2200\end{lrbox}
2201
2202\newbox\myboxB
2203\begin{lrbox}{\myboxB}
2204\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2205
2206
2207mutex( m1 ) {
2208        ...
2209        mutex( m1, m2 ) {
2210                ... `wait( c )`; // release m1, m2
2211                // m1, m2 reacquired
2212        } // $\LstCommentStyle{\color{red}release m2}$
2213        // m1 acquired
2214} // release m1
2215\end{cfa}
2216\end{lrbox}
2217
2218\newbox\myboxC
2219\begin{lrbox}{\myboxC}
2220\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2221
2222
2223mutex( m2 ) {
2224        ... `wait( c )`; // release m2
2225        // m2 reacquired
2226} // $\LstCommentStyle{\color{red}release m2}$
2227
2228
2229
2230
2231\end{cfa}
2232\end{lrbox}
2233
2234\begin{cquote}
2235\subfloat[Signalling Thread (S)]{\label{f:SignallingThread}\usebox\myboxA}
2236\hspace{3\parindentlnth}
2237\subfloat[Waiting Thread (W1)]{\label{f:WaitingThread}\usebox\myboxB}
2238\hspace{2\parindentlnth}
2239\subfloat[Waiting Thread (W2)]{\label{f:OtherWaitingThread}\usebox\myboxC}
2240\end{cquote}
2241\caption{Bulk Barging Prevention}
2242\label{f:BulkBargingPrevention}
2243\end{figure}
2244
2245One scheduling solution is for the signaller S to keep ownership of all locks until the last lock is ready to be transferred, because this semantics fits most closely to the behaviour of single-monitor scheduling.
2246However, this solution is inefficient if W2 waited first and can be immediate passed @m2@ when released, while S retains @m1@ until completion of the outer mutex statement.
2247If W1 waited first, the signaller must retain @m1@ amd @m2@ until completion of the outer mutex statement and then pass both to W1.
2248% Furthermore, there is an execution sequence where the signaller always finds waiter W2, and hence, waiter W1 starves.
2249To support this efficient semantics (and prevent barging), the implementation maintains a list of monitors acquired for each blocked thread.
2250When a signaller exits or waits in a monitor function/statement, the front waiter on urgent is unblocked if all its monitors are released.
2251Implementing a fast subset check for the necessary released monitors is important.
2252% The benefit is encapsulating complexity into only two actions: passing monitors to the next owner when they should be released and conditionally waking threads if all conditions are met.
2253
2254
2255\subsection{Loose Object Definitions}
2256\label{s:LooseObjectDefinitions}
2257
2258In an object-oriented programming language, a class includes an exhaustive list of operations.
2259A new class can add members via static inheritance but the subclass still has an exhaustive list of operations.
2260(Dynamic member adding, \eg JavaScript~\cite{JavaScript}, is not considered.)
2261In the object-oriented scenario, the type and all its operators are always present at compilation (even separate compilation), so it is possible to number the operations in a bit mask and use an $O(1)$ compare with a similar bit mask created for the operations specified in a @waitfor@.
2262
2263However, in \CFA, monitor functions can be statically added/removed in translation units, making a fast subset check difficult.
2264\begin{cfa}
2265        monitor M { ... }; // common type, included in .h file
2266translation unit 1
2267        void `f`( M & mutex m );
2268        void g( M & mutex m ) { waitfor( `f`, m ); }
2269translation unit 2
2270        void `f`( M & mutex m ); $\C{// replacing f and g for type M in this translation unit}$
2271        void `g`( M & mutex m );
2272        void h( M & mutex m ) { waitfor( `f`, m ) or waitfor( `g`, m ); } $\C{// extending type M in this translation unit}$
2273\end{cfa}
2274The @waitfor@ statements in each translation unit cannot form a unique bit-mask because the monitor type does not carry that information.
2275Hence, function pointers are used to identify the functions listed in the @waitfor@ statement, stored in a variable-sized array.
2276Then, the same implementation approach used for the urgent stack is used for the calling queue.
2277Each caller has a list of monitors acquired, and the @waitfor@ statement performs a (usually short) linear search matching functions in the @waitfor@ list with called functions, and then verifying the associated mutex locks can be transfers.
2278(A possible way to construct a dense mapping is at link or load-time.)
2279
2280
2281\subsection{Multi-Monitor Scheduling}
2282\label{s:Multi-MonitorScheduling}
2283
2284External scheduling, like internal scheduling, becomes significantly more complex for multi-monitor semantics.
2285Even in the simplest case, new semantics need to be established.
2286\begin{cfa}
2287monitor M { ... };
2288void f( M & mutex m1 );
2289void g( M & mutex m1, M & mutex m2 ) { `waitfor( f );` } $\C{// pass m1 or m2 to f?}$
2290\end{cfa}
2291The solution is for the programmer to disambiguate:
2292\begin{cfa}
2293waitfor( f, `m2` ); $\C{// wait for call to f with argument m2}$
2294\end{cfa}
2295Both locks are acquired by function @g@, so when function @f@ is called, the lock for monitor @m2@ is passed from @g@ to @f@, while @g@ still holds lock @m1@.
2296This behaviour can be extended to the multi-monitor @waitfor@ statement.
2297\begin{cfa}
2298monitor M { ... };
2299void f( M & mutex m1, M & mutex m2 );
2300void g( M & mutex m1, M & mutex m2 ) { waitfor( f, `m1, m2` ); $\C{// wait for call to f with arguments m1 and m2}$
2301\end{cfa}
2302Again, the set of monitors passed to the @waitfor@ statement must be entirely contained in the set of monitors already acquired by the accepting function.
2303Also, the order of the monitors in a @waitfor@ statement is unimportant.
2304
2305Figure~\ref{f:UnmatchedMutexSets} shows an example where, for internal and external scheduling with multiple monitors, a signalling or accepting thread must match exactly, \ie partial matching results in waiting.
2306For both examples, the set of monitors is disjoint so unblocking is impossible.
2307
2308\begin{figure}
2309\centering
2310\begin{lrbox}{\myboxA}
2311\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2312monitor M1 {} m11, m12;
2313monitor M2 {} m2;
2314condition c;
2315void f( M1 & mutex m1, M2 & mutex m2 ) {
2316        signal( c );
2317}
2318void g( M1 & mutex m1, M2 & mutex m2 ) {
2319        wait( c );
2320}
2321g( `m11`, m2 ); // block on wait
2322f( `m12`, m2 ); // cannot fulfil
2323\end{cfa}
2324\end{lrbox}
2325
2326\begin{lrbox}{\myboxB}
2327\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2328monitor M1 {} m11, m12;
2329monitor M2 {} m2;
2330
2331void f( M1 & mutex m1, M2 & mutex m2 ) {
2332
2333}
2334void g( M1 & mutex m1, M2 & mutex m2 ) {
2335        waitfor( f, m1, m2 );
2336}
2337g( `m11`, m2 ); // block on accept
2338f( `m12`, m2 ); // cannot fulfil
2339\end{cfa}
2340\end{lrbox}
2341\subfloat[Internal scheduling]{\label{f:InternalScheduling}\usebox\myboxA}
2342\hspace{3pt}
2343\vrule
2344\hspace{3pt}
2345\subfloat[External scheduling]{\label{f:ExternalScheduling}\usebox\myboxB}
2346\caption{Unmatched \protect\lstinline@mutex@ sets}
2347\label{f:UnmatchedMutexSets}
2348\end{figure}
2349
2350
2351\subsection{\protect\lstinline@mutex@ Threads}
2352
2353Threads in \CFA can also be monitors to allow \emph{direct communication} among threads, \ie threads can have mutex functions that are called by other threads.
2354Hence, all monitor features are available when using threads.
2355Figure~\ref{f:DirectCommunication} shows a comparison of direct call communication in \CFA with direct channel communication in Go.
2356(Ada provides a similar mechanism to the \CFA direct communication.)
2357The program main in both programs communicates directly with the other thread versus indirect communication where two threads interact through a passive monitor.
2358Both direct and indirection thread communication are valuable tools in structuring concurrent programs.
2359
2360\begin{figure}
2361\centering
2362\begin{lrbox}{\myboxA}
2363\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2364
2365struct Msg { int i, j; };
2366thread GoRtn { int i;  float f;  Msg m; };
2367void mem1( GoRtn & mutex gortn, int i ) { gortn.i = i; }
2368void mem2( GoRtn & mutex gortn, float f ) { gortn.f = f; }
2369void mem3( GoRtn & mutex gortn, Msg m ) { gortn.m = m; }
2370void ^?{}( GoRtn & mutex ) {}
2371
2372void main( GoRtn & gortn ) with( gortn ) {  // thread starts
2373
2374        for () {
2375
2376                `waitfor( mem1, gortn )` sout | i;  // wait for calls
2377                or `waitfor( mem2, gortn )` sout | f;
2378                or `waitfor( mem3, gortn )` sout | m.i | m.j;
2379                or `waitfor( ^?{}, gortn )` break;
2380
2381        }
2382
2383}
2384int main() {
2385        GoRtn gortn; $\C[2.0in]{// start thread}$
2386        `mem1( gortn, 0 );` $\C{// different calls}\CRT$
2387        `mem2( gortn, 2.5 );`
2388        `mem3( gortn, (Msg){1, 2} );`
2389
2390
2391} // wait for completion
2392\end{cfa}
2393\end{lrbox}
2394
2395\begin{lrbox}{\myboxB}
2396\begin{Go}[aboveskip=0pt,belowskip=0pt]
2397func main() {
2398        type Msg struct{ i, j int }
2399
2400        ch1 := make( chan int )
2401        ch2 := make( chan float32 )
2402        ch3 := make( chan Msg )
2403        hand := make( chan string )
2404        shake := make( chan string )
2405        gortn := func() { $\C[1.5in]{// thread starts}$
2406                var i int;  var f float32;  var m Msg
2407                L: for {
2408                        select { $\C{// wait for messages}$
2409                          case `i = <- ch1`: fmt.Println( i )
2410                          case `f = <- ch2`: fmt.Println( f )
2411                          case `m = <- ch3`: fmt.Println( m )
2412                          case `<- hand`: break L $\C{// sentinel}$
2413                        }
2414                }
2415                `shake <- "SHAKE"` $\C{// completion}$
2416        }
2417
2418        go gortn() $\C{// start thread}$
2419        `ch1 <- 0` $\C{// different messages}$
2420        `ch2 <- 2.5`
2421        `ch3 <- Msg{1, 2}`
2422        `hand <- "HAND"` $\C{// sentinel value}$
2423        `<- shake` $\C{// wait for completion}\CRT$
2424}
2425\end{Go}
2426\end{lrbox}
2427
2428\subfloat[\CFA]{\label{f:CFAwaitfor}\usebox\myboxA}
2429\hspace{3pt}
2430\vrule
2431\hspace{3pt}
2432\subfloat[Go]{\label{f:Gochannel}\usebox\myboxB}
2433\caption{Direct communication}
2434\label{f:DirectCommunication}
2435\end{figure}
2436
2437\begin{comment}
2438The following shows an example of two threads directly calling each other and accepting calls from each other in a cycle.
2439\begin{cfa}
2440\end{cfa}
2441\vspace{-0.8\baselineskip}
2442\begin{cquote}
2443\begin{tabular}{@{}l@{\hspace{3\parindentlnth}}l@{}}
2444\begin{cfa}
2445thread Ping {} pi;
2446void ping( Ping & mutex ) {}
2447void main( Ping & pi ) {
2448        for ( 10 ) {
2449                `waitfor( ping, pi );`
2450                `pong( po );`
2451        }
2452}
2453int main() {}
2454\end{cfa}
2455&
2456\begin{cfa}
2457thread Pong {} po;
2458void pong( Pong & mutex ) {}
2459void main( Pong & po ) {
2460        for ( 10 ) {
2461                `ping( pi );`
2462                `waitfor( pong, po );`
2463        }
2464}
2465
2466\end{cfa}
2467\end{tabular}
2468\end{cquote}
2469% \lstMakeShortInline@%
2470% \caption{Threads ping/pong using external scheduling}
2471% \label{f:pingpong}
2472% \end{figure}
2473Note, the ping/pong threads are globally declared, @pi@/@po@, and hence, start (and possibly complete) before the program main starts.
2474\end{comment}
2475
2476
2477\subsection{Execution Properties}
2478
2479Table~\ref{t:ObjectPropertyComposition} shows how the \CFA high-level constructs cover 3 fundamental execution properties: thread, stateful function, and mutual exclusion.
2480Case 1 is a basic object, with none of the new execution properties.
2481Case 2 allows @mutex@ calls to Case 1 to protect shared data.
2482Case 3 allows stateful functions to suspend/resume but restricts operations because the state is stackless.
2483Case 4 allows @mutex@ calls to Case 3 to protect shared data.
2484Cases 5 and 6 are the same as 3 and 4 without restriction because the state is stackful.
2485Cases 7 and 8 are rejected because a thread cannot execute without a stackful state in a preemptive environment when context switching from the signal handler.
2486Cases 9 and 10 have a stackful thread without and with @mutex@ calls.
2487For situations where threads do not require direct communication, case 9 provides faster creation/destruction by eliminating @mutex@ setup.
2488
2489\begin{table}
2490\caption{Object property composition}
2491\centering
2492\label{t:ObjectPropertyComposition}
2493\renewcommand{\arraystretch}{1.25}
2494%\setlength{\tabcolsep}{5pt}
2495\begin{tabular}{c|c|l|l}
2496\multicolumn{2}{c|}{object properties} & \multicolumn{2}{c}{mutual exclusion} \\
2497\hline
2498thread  & stateful                              & \multicolumn{1}{c|}{No} & \multicolumn{1}{c}{Yes} \\
2499\hline
2500\hline
2501No              & No                                    & \textbf{1}\ \ \ aggregate type                & \textbf{2}\ \ \ @monitor@ aggregate type \\
2502\hline
2503No              & Yes (stackless)               & \textbf{3}\ \ \ @generator@                   & \textbf{4}\ \ \ @monitor@ @generator@ \\
2504\hline
2505No              & Yes (stackful)                & \textbf{5}\ \ \ @coroutine@                   & \textbf{6}\ \ \ @monitor@ @coroutine@ \\
2506\hline
2507Yes             & No / Yes (stackless)  & \textbf{7}\ \ \ {\color{red}rejected} & \textbf{8}\ \ \ {\color{red}rejected} \\
2508\hline
2509Yes             & Yes (stackful)                & \textbf{9}\ \ \ @thread@                              & \textbf{10}\ \ @monitor@ @thread@ \\
2510\end{tabular}
2511\end{table}
2512
2513
2514\subsection{Low-level Locks}
2515
2516For completeness and efficiency, \CFA provides a standard set of low-level locks: recursive mutex, condition, semaphore, barrier, \etc, and atomic instructions: @fetchAssign@, @fetchAdd@, @testSet@, @compareSet@, \etc.
2517Some of these low-level mechanism are used in the \CFA runtime, but we strongly advocate using high-level mechanisms whenever possible.
2518
2519
2520% \section{Parallelism}
2521% \label{s:Parallelism}
2522%
2523% Historically, computer performance was about processor speeds.
2524% However, with heat dissipation being a direct consequence of speed increase, parallelism is the new source for increased performance~\cite{Sutter05, Sutter05b}.
2525% Therefore, high-performance applications must care about parallelism, which requires concurrency.
2526% The lowest-level approach of parallelism is to use \newterm{kernel threads} in combination with semantics like @fork@, @join@, \etc.
2527% However, kernel threads are better as an implementation tool because of complexity and higher cost.
2528% Therefore, different abstractions are often layered onto kernel threads to simplify them, \eg pthreads.
2529%
2530%
2531% \subsection{User Threads}
2532%
2533% A direct improvement on kernel threads is user threads, \eg Erlang~\cite{Erlang} and \uC~\cite{uC++book}.
2534% This approach provides an interface that matches the language paradigms, gives more control over concurrency by the language runtime, and an abstract (and portable) interface to the underlying kernel threads across operating systems.
2535% In many cases, user threads can be used on a much larger scale (100,000 threads).
2536% Like kernel threads, user threads support preemption, which maximizes nondeterminism, but increases the potential for concurrency errors: race, livelock, starvation, and deadlock.
2537% \CFA adopts user-threads to provide more flexibility and a low-cost mechanism to build any other concurrency approach, \eg thread pools and actors~\cite{Actors}.
2538%
2539% A variant of user thread is \newterm{fibres}, which removes preemption, \eg Go~\cite{Go} @goroutine@s.
2540% Like functional programming, which removes mutation and its associated problems, removing preemption from concurrency reduces nondeterminism, making race and deadlock errors more difficult to generate.
2541% However, preemption is necessary for fairness and to reduce tail-latency.
2542% For concurrency that relies on spinning, if all cores spin the system is livelocked, whereas preemption breaks the livelock.
2543
2544
2545\begin{comment}
2546\subsection{Thread Pools}
2547
2548In contrast to direct threading is indirect \newterm{thread pools}, \eg Java @executor@, where small jobs (work units) are inserted into a work pool for execution.
2549If the jobs are dependent, \ie interact, there is an implicit/explicit dependency graph that ties them together.
2550While removing direct concurrency, and hence the amount of context switching, thread pools significantly limit the interaction that can occur among jobs.
2551Indeed, jobs should not block because that also blocks the underlying thread, which effectively means the CPU utilization, and therefore throughput, suffers.
2552While it is possible to tune the thread pool with sufficient threads, it becomes difficult to obtain high throughput and good core utilization as job interaction increases.
2553As well, concurrency errors return, which threads pools are suppose to mitigate.
2554
2555\begin{figure}
2556\centering
2557\begin{tabular}{@{}l|l@{}}
2558\begin{cfa}
2559struct Adder {
2560    int * row, cols;
2561};
2562int operator()() {
2563        subtotal = 0;
2564        for ( int c = 0; c < cols; c += 1 )
2565                subtotal += row[c];
2566        return subtotal;
2567}
2568void ?{}( Adder * adder, int row[$\,$], int cols, int & subtotal ) {
2569        adder.[rows, cols, subtotal] = [rows, cols, subtotal];
2570}
2571
2572
2573
2574
2575\end{cfa}
2576&
2577\begin{cfa}
2578int main() {
2579        const int rows = 10, cols = 10;
2580        int matrix[rows][cols], subtotals[rows], total = 0;
2581        // read matrix
2582        Executor executor( 4 ); // kernel threads
2583        Adder * adders[rows];
2584        for ( r; rows ) { // send off work for executor
2585                adders[r] = new( matrix[r], cols, &subtotal[r] );
2586                executor.send( *adders[r] );
2587        }
2588        for ( r; rows ) {       // wait for results
2589                delete( adders[r] );
2590                total += subtotals[r];
2591        }
2592        sout | total;
2593}
2594\end{cfa}
2595\end{tabular}
2596\caption{Executor}
2597\end{figure}
2598\end{comment}
2599
2600
2601\section{\protect\CFA Runtime Structure}
2602\label{s:CFARuntimeStructure}
2603
2604Figure~\ref{f:RunTimeStructure} illustrates the runtime structure of a \CFA program.
2605In addition to the new kinds of objects introduced by \CFA, there are two more runtime entities used to control parallel execution: cluster and (virtual) processor.
2606An executing thread is illustrated by its containment in a processor.
2607
2608\begin{figure}
2609\centering
2610\input{RunTimeStructure}
2611\caption{\CFA Runtime structure}
2612\label{f:RunTimeStructure}
2613\end{figure}
2614
2615
2616\subsection{Cluster}
2617\label{s:RuntimeStructureCluster}
2618
2619A \newterm{cluster} is a collection of threads and virtual processors (abstract kernel-thread) that execute the (user) threads from its own ready queue (like an OS executing kernel threads).
2620The purpose of a cluster is to control the amount of parallelism that is possible among threads, plus scheduling and other execution defaults.
2621The default cluster-scheduler is single-queue multi-server, which provides automatic load-balancing of threads on processors.
2622However, the design allows changing the scheduler, \eg multi-queue multi-server with work-stealing/sharing across the virtual processors.
2623If several clusters exist, both threads and virtual processors, can be explicitly migrated from one cluster to another.
2624No automatic load balancing among clusters is performed by \CFA.
2625
2626When a \CFA program begins execution, it creates a user cluster with a single processor and a special processor to handle preemption that does not execute user threads.
2627The user cluster is created to contain the application user-threads.
2628Having all threads execute on the one cluster often maximizes utilization of processors, which minimizes runtime.
2629However, because of limitations of scheduling requirements (real-time), NUMA architecture, heterogeneous hardware, or issues with the underlying operating system, multiple clusters are sometimes necessary.
2630
2631
2632\subsection{Virtual Processor}
2633\label{s:RuntimeStructureProcessor}
2634
2635A virtual processor is implemented by a kernel thread (\eg UNIX process), which are scheduled for execution on a hardware processor by the underlying operating system.
2636Programs may use more virtual processors than hardware processors.
2637On a multiprocessor, kernel threads are distributed across the hardware processors resulting in virtual processors executing in parallel.
2638(It is possible to use affinity to lock a virtual processor onto a particular hardware processor~\cite{affinityLinux, affinityWindows, affinityFreebsd, affinityNetbsd, affinityMacosx}, which is used when caching issues occur or for heterogeneous hardware processors.)
2639The \CFA runtime attempts to block unused processors and unblock processors as the system load increases;
2640balancing the workload with processors is difficult because it requires future knowledge, \ie what will the applicaton workload do next.
2641Preemption occurs on virtual processors rather than user threads, via operating-system interrupts.
2642Thus virtual processors execute user threads, where preemption frequency applies to a virtual processor, so preemption occurs randomly across the executed user threads.
2643Turning off preemption transforms user threads into fibres.
2644
2645
2646\begin{comment}
2647\section{Implementation}
2648\label{s:Implementation}
2649
2650A primary implementation challenge is avoiding contention from dynamically allocating memory because of bulk acquire, \eg the internal-scheduling design is (almost) free of allocations.
2651All blocking operations are made by parking threads onto queues, therefore all queues are designed with intrusive nodes, where each node has preallocated link fields for chaining.
2652Furthermore, several bulk-acquire operations need a variable amount of memory.
2653This storage is allocated at the base of a thread's stack before blocking, which means programmers must add a small amount of extra space for stacks.
2654
2655In \CFA, ordering of monitor acquisition relies on memory ordering to prevent deadlock~\cite{Havender68}, because all objects have distinct non-overlapping memory layouts, and mutual-exclusion for a monitor is only defined for its lifetime.
2656When a mutex call is made, pointers to the concerned monitors are aggregated into a variable-length array and sorted.
2657This array persists for the entire duration of the mutual exclusion and is used extensively for synchronization operations.
2658
2659To improve performance and simplicity, context switching occurs inside a function call, so only callee-saved registers are copied onto the stack and then the stack register is switched;
2660the corresponding registers are then restored for the other context.
2661Note, the instruction pointer is untouched since the context switch is always inside the same function.
2662Experimental results (not presented) for a stackless or stackful scheduler (1 versus 2 context switches) (see Section~\ref{s:Concurrency}) show the performance is virtually equivalent, because both approaches are dominated by locking to prevent a race condition.
2663
2664All kernel threads (@pthreads@) created a stack.
2665Each \CFA virtual processor is implemented as a coroutine and these coroutines run directly on the kernel-thread stack, effectively stealing this stack.
2666The exception to this rule is the program main, \ie the initial kernel thread that is given to any program.
2667In order to respect C expectations, the stack of the initial kernel thread is used by program main rather than the main processor, allowing it to grow dynamically as in a normal C program.
2668\end{comment}
2669
2670
2671\subsection{Preemption}
2672
2673Nondeterministic preemption provides fairness from long-running threads, and forces concurrent programmers to write more robust programs, rather than relying on code between cooperative scheduling to be atomic.
2674This atomic reliance can fail on multi-core machines, because execution across cores is nondeterministic.
2675A different reason for not supporting preemption is that it significantly complicates the runtime system, \eg Microsoft runtime does not support interrupts and on Linux systems, interrupts are complex (see below).
2676Preemption is normally handled by setting a countdown timer on each virtual processor.
2677When the timer expires, an interrupt is delivered, and the interrupt handler resets the countdown timer, and if the virtual processor is executing in user code, the signal handler performs a user-level context-switch, or if executing in the language runtime kernel, the preemption is ignored or rolled forward to the point where the runtime kernel context switches back to user code.
2678Multiple signal handlers may be pending.
2679When control eventually switches back to the signal handler, it returns normally, and execution continues in the interrupted user thread, even though the return from the signal handler may be on a different kernel thread than the one where the signal is delivered.
2680The only issue with this approach is that signal masks from one kernel thread may be restored on another as part of returning from the signal handler;
2681therefore, the same signal mask is required for all virtual processors in a cluster.
2682Because preemption frequency is usually long (1 millisecond) performance cost is negligible.
2683
2684Linux switched a decade ago from specific to arbitrary process signal-delivery for applications with multiple kernel threads.
2685\begin{cquote}
2686A process-directed signal may be delivered to any one of the threads that does not currently have the signal blocked.
2687If more than one of the threads has the signal unblocked, then the kernel chooses an arbitrary thread to which it will deliver the signal.
2688SIGNAL(7) - Linux Programmer's Manual
2689\end{cquote}
2690Hence, the timer-expiry signal, which is generated \emph{externally} by the Linux kernel to an application, is delivered to any of its Linux subprocesses (kernel threads).
2691To ensure each virtual processor receives a preemption signal, a discrete-event simulation is run on a special virtual processor, and only it sets and receives timer events.
2692Virtual processors register an expiration time with the discrete-event simulator, which is inserted in sorted order.
2693The simulation sets the countdown timer to the value at the head of the event list, and when the timer expires, all events less than or equal to the current time are processed.
2694Processing a preemption event sends an \emph{internal} @SIGUSR1@ signal to the registered virtual processor, which is always delivered to that processor.
2695
2696
2697\subsection{Debug Kernel}
2698
2699There are two versions of the \CFA runtime kernel: debug and non-debug.
2700The debugging version has many runtime checks and internal assertions, \eg stack (non-writable) guard page, and checks for stack overflow whenever context switches occur among coroutines and threads, which catches most stack overflows.
2701After a program is debugged, the non-debugging version can be used to significantly decrease space and increase performance.
2702
2703
2704\section{Performance}
2705\label{results}
2706
2707To verify the implementation of the \CFA runtime, a series of microbenchmarks are performed comparing \CFA with pthreads, Java OpenJDK-9, Go 1.12.6 and \uC 7.0.0.
2708For comparison, the package must be multi-processor (M:N), which excludes libdill/libmil~\cite{libdill} (M:1)), and use a shared-memory programming model, \eg not message passing.
2709The benchmark computer is an AMD Opteron\texttrademark\ 6380 NUMA 64-core, 8 socket, 2.5 GHz processor, running Ubuntu 16.04.6 LTS, and \CFA/\uC are compiled with gcc 6.5.
2710
2711All benchmarks are run using the following harness.
2712\begin{cfa}
2713unsigned int N = 10_000_000;
2714#define BENCH( `run` ) Time before = getTimeNsec();  `run;`  Duration result = (getTimeNsec() - before) / N;
2715\end{cfa}
2716The method used to get time is @clock_gettime( CLOCK_REALTIME )@.
2717Each benchmark is performed @N@ times, where @N@ varies depending on the benchmark;
2718the total time is divided by @N@ to obtain the average time for a benchmark.
2719Each benchmark experiment is run 31 times.
2720All omitted tests for other languages are functionally identical to the \CFA tests and available online~\cite{CforallBenchMarks}.
2721
2722
2723\paragraph{Object Creation}
2724
2725Object creation is measured by creating/deleting the specific kind of concurrent object.
2726Figure~\ref{f:creation} shows the code for \CFA, with results in Table~\ref{tab:creation}.
2727The only note here is that the call stacks of \CFA coroutines are lazily created, therefore without priming the coroutine to force stack creation, the creation cost is artificially low.
2728
2729\begin{multicols}{2}
2730\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2731\begin{cfa}
2732@thread@ MyThread {};
2733void @main@( MyThread & ) {}
2734int main() {
2735        BENCH( for ( N ) { @MyThread m;@ } )
2736        sout | result`ns;
2737}
2738\end{cfa}
2739\captionof{figure}{\CFA object-creation benchmark}
2740\label{f:creation}
2741
2742\columnbreak
2743
2744\vspace*{-16pt}
2745\captionof{table}{Object creation comparison (nanoseconds)}
2746\label{tab:creation}
2747
2748\begin{tabular}[t]{@{}r*{3}{D{.}{.}{5.2}}@{}}
2749\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} & \multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2750\CFA Coroutine Lazy             & 14.3          & 14.3          & 0.32          \\
2751\CFA Coroutine Eager    & 2203.7        & 2205.6        & 26.03         \\
2752\CFA Thread                             & 1257.8        & 1291.2        & 86.19         \\
2753\uC Coroutine                   & 92.2          & 91.4          & 1.58          \\
2754\uC Thread                              & 499.5         & 500.1         & 5.67          \\
2755Goroutine                               & 4397.0        & 4362.8        & 390.77        \\
2756Java Thread                             & 107405.0      & 107794.8      & 1601.33       \\
2757% Qthreads                              & 159.9         & 159.6         & 0.73          \\
2758Pthreads                                & 32920.9       & 32882.7       & 213.55
2759\end{tabular}
2760\end{multicols}
2761
2762
2763\paragraph{Internal Scheduling}
2764
2765Internal scheduling is measured using a cycle of two threads signalling and waiting.
2766Figure~\ref{f:int-sched} shows the code for \CFA, with results in Table~\ref{tab:int-sched}.
2767Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2768Java scheduling is significantly greater because the benchmark explicitly creates multiple thread in order to prevent the JIT from making the program sequential, \ie removing all locking.
2769
2770\begin{multicols}{2}
2771\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2772\begin{cfa}
2773volatile int go = 0;
2774@monitor@ M { @condition c;@ } m;
2775void __attribute__((noinline))
2776do_call( M & @mutex@ a1 ) { @signal( c );@ }
2777thread T {};
2778void main( T & this ) {
2779        while ( go == 0 ) { yield(); }
2780        while ( go == 1 ) { do_call( m ); }
2781}
2782int  __attribute__((noinline))
2783do_wait( M & mutex m ) with(m) {
2784        go = 1; // continue other thread
2785        BENCH( for ( N ) { @wait( c );@ } );
2786        go = 0; // stop other thread
2787        sout | result`ns;
2788}
2789int main() {
2790        T t;
2791        do_wait( m );
2792}
2793\end{cfa}
2794\captionof{figure}{\CFA Internal-scheduling benchmark}
2795\label{f:int-sched}
2796
2797\columnbreak
2798
2799\vspace*{-16pt}
2800\captionof{table}{Internal-scheduling comparison (nanoseconds)}
2801\label{tab:int-sched}
2802\bigskip
2803
2804\begin{tabular}{@{}r*{3}{D{.}{.}{5.2}}@{}}
2805\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} & \multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2806\CFA @signal@, 1 @monitor@      & 367.0         & 371.5         & 17.34         \\
2807\CFA @signal@, 2 @monitor@      & 477.2         & 478.6         & 8.31          \\
2808\CFA @signal@, 4 @monitor@      & 725.8         & 734.0         & 17.98         \\
2809\uC @signal@                            & 322.8         & 323.0         & 3.64          \\
2810Java @notify@                           & 16520.0       & 20096.7       & 9378.53       \\
2811Pthreads Cond. Variable         & 4931.3        & 5057.0        & 326.80
2812\end{tabular}
2813\end{multicols}
2814
2815
2816\paragraph{External Scheduling}
2817
2818External scheduling is measured using a cycle of two threads calling and accepting the call using the @waitfor@ statement.
2819Figure~\ref{f:ext-sched} shows the code for \CFA, with results in Table~\ref{tab:ext-sched}.
2820Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2821
2822\begin{multicols}{2}
2823\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2824\vspace*{-16pt}
2825\begin{cfa}
2826volatile int go = 0;
2827@monitor@ M {} m;
2828thread T {};
2829void __attribute__((noinline))
2830do_call( M & @mutex@ ) {}
2831void main( T & ) {
2832        while ( go == 0 ) { yield(); }
2833        while ( go == 1 ) { do_call( m ); }
2834}
2835int __attribute__((noinline))
2836do_wait( M & @mutex@ m ) {
2837        go = 1; // continue other thread
2838        BENCH( for ( N ) { @waitfor( do_call, m );@ } )
2839        go = 0; // stop other thread
2840        sout | result`ns;
2841}
2842int main() {
2843        T t;
2844        do_wait( m );
2845}
2846\end{cfa}
2847\captionof{figure}{\CFA external-scheduling benchmark}
2848\label{f:ext-sched}
2849
2850\columnbreak
2851
2852\vspace*{-16pt}
2853\captionof{table}{External-scheduling comparison (nanoseconds)}
2854\label{tab:ext-sched}
2855\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2856\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2857\CFA @waitfor@, 1 @monitor@     & 366.7         & 369.5 & 7.52  \\
2858\CFA @waitfor@, 2 @monitor@     & 453.6         & 455.8 & 12.38 \\
2859\CFA @waitfor@, 4 @monitor@     & 671.6         & 672.4 & 14.16 \\
2860\uC @_Accept@                           & 336.0         & 335.8         & 3.22
2861\end{tabular}
2862\end{multicols}
2863
2864
2865\paragraph{Context-Switching}
2866
2867In procedural programming, the cost of a function call is important as modularization (refactoring) increases.
2868(In many cases, a compiler inlines function calls to eliminate this cost.)
2869Similarly, when modularization extends to coroutines/tasks, the time for a context switch becomes a relevant factor.
2870The coroutine test is from resumer to suspender and from suspender to resumer, which is two context switches.
2871The thread test is using yield to enter and return from the runtime kernel, which is two context switches.
2872The difference in performance between coroutine and thread context-switch is the cost of scheduling for threads, whereas coroutines are self-scheduling.
2873Figure~\ref{f:ctx-switch} only shows the \CFA code for coroutines/threads (other systems are similar) with all results in Table~\ref{tab:ctx-switch}.
2874
2875\begin{multicols}{2}
2876\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2877\begin{cfa}[aboveskip=0pt,belowskip=0pt]
2878@coroutine@ C {} c;
2879void main( C & ) { for ( ;; ) { @suspend;@ } }
2880int main() { // coroutine test
2881        BENCH( for ( N ) { @resume( c );@ } )
2882        sout | result`ns;
2883}
2884int main() { // task test
2885        BENCH( for ( N ) { @yield();@ } )
2886        sout | result`ns;
2887}
2888\end{cfa}
2889\captionof{figure}{\CFA context-switch benchmark}
2890\label{f:ctx-switch}
2891
2892\columnbreak
2893
2894\vspace*{-16pt}
2895\captionof{table}{Context switch comparison (nanoseconds)}
2896\label{tab:ctx-switch}
2897\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2898\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2899C function              & 1.8           & 1.8   & 0             \\
2900\CFA generator  & 2.7           & 2.4   & 0.27  \\
2901\CFA Coroutine  & 37.8          & 37.7  & 0.22  \\
2902\CFA Thread             & 93.6          & 93.8  & 1.46  \\
2903\uC Coroutine   & 52.7          & 52.8  & 0.28  \\
2904\uC Thread              & 93.4          & 93.7  & 1.04  \\
2905Goroutine               & 140.0         & 139.7 & 2.93  \\
2906Java Thread             & 374.0         & 375.8 & 10.38 \\
2907% Qthreads Thread       & 159.5         & 159.3 & 0.71  \\
2908Pthreads Thread & 334.4         & 335.0 & 1.95  \\
2909\end{tabular}
2910\end{multicols}
2911
2912
2913\paragraph{Mutual-Exclusion}
2914
2915Uncontented mutual exclusion, which frequently occurs, is measured by entering/leaving a critical section.
2916For monitors, entering and leaving a monitor function is measured.
2917To put the results in context, the cost of entering a non-inline function and the cost of acquiring and releasing a @pthread_mutex@ lock is also measured.
2918Figure~\ref{f:mutex} shows the code for \CFA with all results in Table~\ref{tab:mutex}.
2919Note, the incremental cost of bulk acquire for \CFA, which is largely a fixed cost for small numbers of mutex objects.
2920
2921\begin{multicols}{2}
2922\lstset{language=CFA,moredelim=**[is][\color{red}]{@}{@},deletedelim=**[is][]{`}{`}}
2923\begin{cfa}
2924@monitor@ M {} m1/*, m2, m3, m4*/;
2925void __attribute__((noinline))
2926do_call( M & @mutex m/*, m2, m3, m4*/@ ) {}
2927int main() {
2928        BENCH(
2929                for( N ) do_call( m1/*, m2, m3, m4*/ );
2930        )
2931        sout | result`ns;
2932}
2933\end{cfa}
2934\captionof{figure}{\CFA acquire/release mutex benchmark}
2935\label{f:mutex}
2936
2937\columnbreak
2938
2939\vspace*{-16pt}
2940\captionof{table}{Mutex comparison (nanoseconds)}
2941\label{tab:mutex}
2942\begin{tabular}{@{}r*{3}{D{.}{.}{3.2}}@{}}
2943\multicolumn{1}{@{}c}{} & \multicolumn{1}{c}{Median} &\multicolumn{1}{c}{Average} & \multicolumn{1}{c@{}}{Std Dev} \\
2944test and test-and-test lock             & 19.1  & 19.0  & 0.36  \\
2945\CFA @mutex@ function, 1 arg.   & 46.6  & 46.8  & 0.86  \\
2946\CFA @mutex@ function, 2 arg.   & 84.1  & 85.3  & 1.86  \\
2947\CFA @mutex@ function, 4 arg.   & 158.6 & 160.7 & 3.07  \\
2948\uC @monitor@ member rtn.               & 54.0  & 53.7  & 0.83  \\
2949Java synchronized method                & 27.0  & 27.1  & 0.25  \\
2950Pthreads Mutex Lock                             & 33.6  & 32.7  & 1.12
2951\end{tabular}
2952\end{multicols}
2953
2954
2955\section{Conclusion}
2956
2957Advanced control-flow will always be difficult, especially when there is temporal ordering and nondeterminism.
2958However, many systems exacerbate the difficulty through their presentation mechanisms.
2959This paper shows it is possible to present a hierarchy of control-flow features, generator, coroutine, thread, and monitor, providing an integrated set of high-level, efficient, and maintainable control-flow features.
2960Eliminated from \CFA are spurious wakeup and barging, which are nonintuitive and lead to errors, and having to work with a bewildering set of low-level locks and acquisition techniques.
2961\CFA high-level race-free monitors and tasks provide the core mechanisms for mutual exclusion and synchronization, without having to resort to magic qualifiers like @volatile@/@atomic@.
2962Extending these mechanisms to handle high-level deadlock-free bulk acquire across both mutual exclusion and synchronization is a unique contribution.
2963The \CFA runtime provides concurrency based on a preemptive M:N user-level threading-system, executing in clusters, which encapsulate scheduling of work on multiple kernel threads providing parallelism.
2964The M:N model is judged to be efficient and provide greater flexibility than a 1:1 threading model.
2965These concepts and the \CFA runtime-system are written in the \CFA language, extensively leveraging the \CFA type-system, which demonstrates the expressiveness of the \CFA language.
2966Performance comparisons with other concurrent systems/languages show the \CFA approach is competitive across all low-level operations, which translates directly into good performance in well-written concurrent applications.
2967C programmers should feel comfortable using these mechanisms for developing complex control-flow in applications, with the ability to obtain maximum available performance by selecting mechanisms at the appropriate level of need.
2968
2969
2970\section{Future Work}
2971
2972While control flow in \CFA has a strong start, development is still underway to complete a number of missing features.
2973
2974\paragraph{Flexible Scheduling}
2975\label{futur:sched}
2976
2977An important part of concurrency is scheduling.
2978Different scheduling algorithms can affect performance (both in terms of average and variation).
2979However, no single scheduler is optimal for all workloads and therefore there is value in being able to change the scheduler for given programs.
2980One solution is to offer various tuning options, allowing the scheduler to be adjusted to the requirements of the workload.
2981However, to be truly flexible, a pluggable scheduler is necessary.
2982Currently, the \CFA pluggable scheduler is too simple to handle complex scheduling, \eg quality of service and real-time, where the scheduler must interact with mutex objects to deal with issues like priority inversion~\cite{Buhr00b}.
2983
2984\paragraph{Non-Blocking I/O}
2985\label{futur:nbio}
2986
2987Many modern workloads are not bound by computation but IO operations, a common case being web servers and XaaS~\cite{XaaS} (anything as a service).
2988These types of workloads require significant engineering to amortizing costs of blocking IO-operations.
2989At its core, non-blocking I/O is an operating-system level feature queuing IO operations, \eg network operations, and registering for notifications instead of waiting for requests to complete.
2990Current trends use asynchronous programming like callbacks, futures, and/or promises, \eg Node.js~\cite{NodeJs} for JavaScript, Spring MVC~\cite{SpringMVC} for Java, and Django~\cite{Django} for Python.
2991However, these solutions lead to code that is hard to create, read and maintain.
2992A better approach is to tie non-blocking I/O into the concurrency system to provide ease of use with low overhead, \eg thread-per-connection web-services.
2993A non-blocking I/O library is currently under development for \CFA.
2994
2995\paragraph{Other Concurrency Tools}
2996\label{futur:tools}
2997
2998While monitors offer flexible and powerful concurrency for \CFA, other concurrency tools are also necessary for a complete multi-paradigm concurrency package.
2999Examples of such tools can include futures and promises~\cite{promises}, executors and actors.
3000These additional features are useful for applications that can be constructed without shared data and direct blocking.
3001As well, new \CFA extensions should make it possible to create a uniform interface for virtually all mutual exclusion, including monitors and low-level locks.
3002
3003\paragraph{Implicit Threading}
3004\label{futur:implcit}
3005
3006Basic concurrent (embarrassingly parallel) applications can benefit greatly from implicit concurrency, where sequential programs are converted to concurrent, possibly with some help from pragmas to guide the conversion.
3007This type of concurrency can be achieved both at the language level and at the library level.
3008The canonical example of implicit concurrency is concurrent nested @for@ loops, which are amenable to divide and conquer algorithms~\cite{uC++book}.
3009The \CFA language features should make it possible to develop a reasonable number of implicit concurrency mechanism to solve basic HPC data-concurrency problems.
3010However, implicit concurrency is a restrictive solution with significant limitations, so it can never replace explicit concurrent programming.
3011
3012
3013\section{Acknowledgements}
3014
3015The authors would like to recognize the design assistance of Aaron Moss, Rob Schluntz, Andrew Beach and Michael Brooks on the features described in this paper.
3016Funding for this project has been provided by Huawei Ltd.\ (\url{http://www.huawei.com}). %, and Peter Buhr is partially funded by the Natural Sciences and Engineering Research Council of Canada.
3017
3018{%
3019\fontsize{9bp}{12bp}\selectfont%
3020\bibliography{pl,local}
3021}%
3022
3023\end{document}
3024
3025% Local Variables: %
3026% tab-width: 4 %
3027% fill-column: 120 %
3028% compile-command: "make" %
3029% End: %
Note: See TracBrowser for help on using the repository browser.